diff options
Diffstat (limited to 'src/mongo/s')
252 files changed, 47941 insertions, 49494 deletions
diff --git a/src/mongo/s/balance.cpp b/src/mongo/s/balance.cpp index 0ef5d3c42ce..dc429e85d80 100644 --- a/src/mongo/s/balance.cpp +++ b/src/mongo/s/balance.cpp @@ -64,588 +64,559 @@ namespace mongo { - using std::map; - using std::set; - using std::shared_ptr; - using std::string; - using std::unique_ptr; - using std::vector; +using std::map; +using std::set; +using std::shared_ptr; +using std::string; +using std::unique_ptr; +using std::vector; - MONGO_FP_DECLARE(skipBalanceRound); +MONGO_FP_DECLARE(skipBalanceRound); - Balancer balancer; +Balancer balancer; - Balancer::Balancer() - : _balancedLastTime(0), - _policy(new BalancerPolicy()) { +Balancer::Balancer() : _balancedLastTime(0), _policy(new BalancerPolicy()) {} - } +Balancer::~Balancer() = default; - Balancer::~Balancer() = default; +int Balancer::_moveChunks(const vector<shared_ptr<MigrateInfo>>& candidateChunks, + const WriteConcernOptions* writeConcern, + bool waitForDelete) { + int movedCount = 0; - int Balancer::_moveChunks(const vector<shared_ptr<MigrateInfo>>& candidateChunks, - const WriteConcernOptions* writeConcern, - bool waitForDelete) - { - int movedCount = 0; + for (const auto& migrateInfo : candidateChunks) { + // If the balancer was disabled since we started this round, don't start new chunks + // moves. + const auto balSettingsResult = + grid.catalogManager()->getGlobalSettings(SettingsType::BalancerDocKey); - for (const auto& migrateInfo : candidateChunks) { - // If the balancer was disabled since we started this round, don't start new chunks - // moves. - const auto balSettingsResult = - grid.catalogManager()->getGlobalSettings(SettingsType::BalancerDocKey); + const bool isBalSettingsAbsent = + balSettingsResult.getStatus() == ErrorCodes::NoMatchingDocument; - const bool isBalSettingsAbsent = - balSettingsResult.getStatus() == ErrorCodes::NoMatchingDocument; + if (!balSettingsResult.isOK() && !isBalSettingsAbsent) { + warning() << balSettingsResult.getStatus(); + return movedCount; + } - if (!balSettingsResult.isOK() && !isBalSettingsAbsent) { - warning() << balSettingsResult.getStatus(); - return movedCount; - } + const SettingsType& balancerConfig = + isBalSettingsAbsent ? SettingsType{} : balSettingsResult.getValue(); - const SettingsType& balancerConfig = isBalSettingsAbsent ? - SettingsType{} : balSettingsResult.getValue(); + if ((!isBalSettingsAbsent && !grid.shouldBalance(balancerConfig)) || + MONGO_FAIL_POINT(skipBalanceRound)) { + LOG(1) << "Stopping balancing round early as balancing was disabled"; + return movedCount; + } - if ((!isBalSettingsAbsent && !grid.shouldBalance(balancerConfig)) || - MONGO_FAIL_POINT(skipBalanceRound)) { - LOG(1) << "Stopping balancing round early as balancing was disabled"; - return movedCount; - } + // Changes to metadata, borked metadata, and connectivity problems between shards + // should cause us to abort this chunk move, but shouldn't cause us to abort the entire + // round of chunks. + // + // TODO(spencer): We probably *should* abort the whole round on issues communicating + // with the config servers, but its impossible to distinguish those types of failures + // at the moment. + // + // TODO: Handle all these things more cleanly, since they're expected problems - // Changes to metadata, borked metadata, and connectivity problems between shards - // should cause us to abort this chunk move, but shouldn't cause us to abort the entire - // round of chunks. - // - // TODO(spencer): We probably *should* abort the whole round on issues communicating - // with the config servers, but its impossible to distinguish those types of failures - // at the moment. - // - // TODO: Handle all these things more cleanly, since they're expected problems + const NamespaceString nss(migrateInfo->ns); - const NamespaceString nss(migrateInfo->ns); + try { + auto status = grid.catalogCache()->getDatabase(nss.db().toString()); + fassert(28628, status.getStatus()); + + shared_ptr<DBConfig> cfg = status.getValue(); - try { - auto status = grid.catalogCache()->getDatabase(nss.db().toString()); - fassert(28628, status.getStatus()); + // NOTE: We purposely do not reload metadata here, since _doBalanceRound already + // tried to do so once. + shared_ptr<ChunkManager> cm = cfg->getChunkManager(migrateInfo->ns); + invariant(cm); - shared_ptr<DBConfig> cfg = status.getValue(); + ChunkPtr c = cm->findIntersectingChunk(migrateInfo->chunk.min); - // NOTE: We purposely do not reload metadata here, since _doBalanceRound already - // tried to do so once. - shared_ptr<ChunkManager> cm = cfg->getChunkManager(migrateInfo->ns); + if (c->getMin().woCompare(migrateInfo->chunk.min) || + c->getMax().woCompare(migrateInfo->chunk.max)) { + // Likely a split happened somewhere, so force reload the chunk manager + cm = cfg->getChunkManager(migrateInfo->ns, true); invariant(cm); - ChunkPtr c = cm->findIntersectingChunk(migrateInfo->chunk.min); + c = cm->findIntersectingChunk(migrateInfo->chunk.min); if (c->getMin().woCompare(migrateInfo->chunk.min) || - c->getMax().woCompare(migrateInfo->chunk.max)) { - - // Likely a split happened somewhere, so force reload the chunk manager - cm = cfg->getChunkManager(migrateInfo->ns, true); - invariant(cm); - - c = cm->findIntersectingChunk(migrateInfo->chunk.min); + c->getMax().woCompare(migrateInfo->chunk.max)) { + log() << "chunk mismatch after reload, ignoring will retry issue " + << migrateInfo->chunk.toString(); - if (c->getMin().woCompare(migrateInfo->chunk.min) || - c->getMax().woCompare(migrateInfo->chunk.max)) { - - log() << "chunk mismatch after reload, ignoring will retry issue " - << migrateInfo->chunk.toString(); - - continue; - } - } - - BSONObj res; - if (c->moveAndCommit(migrateInfo->to, - Chunk::MaxChunkSize, - writeConcern, - waitForDelete, - 0, /* maxTimeMS */ - res)) { - - movedCount++; continue; } + } - // The move requires acquiring the collection metadata's lock, which can fail. - log() << "balancer move failed: " << res - << " from: " << migrateInfo->from - << " to: " << migrateInfo->to - << " chunk: " << migrateInfo->chunk; + BSONObj res; + if (c->moveAndCommit(migrateInfo->to, + Chunk::MaxChunkSize, + writeConcern, + waitForDelete, + 0, /* maxTimeMS */ + res)) { + movedCount++; + continue; + } - if (res["chunkTooBig"].trueValue()) { - // Reload just to be safe - cm = cfg->getChunkManager(migrateInfo->ns); - invariant(cm); + // The move requires acquiring the collection metadata's lock, which can fail. + log() << "balancer move failed: " << res << " from: " << migrateInfo->from + << " to: " << migrateInfo->to << " chunk: " << migrateInfo->chunk; - c = cm->findIntersectingChunk(migrateInfo->chunk.min); + if (res["chunkTooBig"].trueValue()) { + // Reload just to be safe + cm = cfg->getChunkManager(migrateInfo->ns); + invariant(cm); + + c = cm->findIntersectingChunk(migrateInfo->chunk.min); - log() << "performing a split because migrate failed for size reasons"; + log() << "performing a split because migrate failed for size reasons"; - Status status = c->split(Chunk::normal, NULL, NULL); - log() << "split results: " << status; + Status status = c->split(Chunk::normal, NULL, NULL); + log() << "split results: " << status; - if (!status.isOK()) { - log() << "marking chunk as jumbo: " << c->toString(); + if (!status.isOK()) { + log() << "marking chunk as jumbo: " << c->toString(); - c->markAsJumbo(); + c->markAsJumbo(); - // We increment moveCount so we do another round right away - movedCount++; - } + // We increment moveCount so we do another round right away + movedCount++; } } - catch (const DBException& ex) { - warning() << "could not move chunk " << migrateInfo->chunk.toString() - << ", continuing balancing round" << causedBy(ex); - } + } catch (const DBException& ex) { + warning() << "could not move chunk " << migrateInfo->chunk.toString() + << ", continuing balancing round" << causedBy(ex); } - - return movedCount; } - void Balancer::_ping(bool waiting) { - grid.catalogManager()->update( - MongosType::ConfigNS, - BSON(MongosType::name(_myid)), - BSON("$set" << BSON(MongosType::ping(jsTime()) << - MongosType::up(static_cast<int>(time(0) - _started)) << - MongosType::waiting(waiting) << - MongosType::mongoVersion(versionString))), - true, - false, - NULL); - } + return movedCount; +} + +void Balancer::_ping(bool waiting) { + grid.catalogManager()->update( + MongosType::ConfigNS, + BSON(MongosType::name(_myid)), + BSON("$set" << BSON(MongosType::ping(jsTime()) + << MongosType::up(static_cast<int>(time(0) - _started)) + << MongosType::waiting(waiting) + << MongosType::mongoVersion(versionString))), + true, + false, + NULL); +} + +/* +* Builds the details object for the actionlog. +* Current formats for detail are: +* Success: { +* "candidateChunks" : , +* "chunksMoved" : , +* "executionTimeMillis" : , +* "errorOccured" : false +* } +* Failure: { +* "executionTimeMillis" : , +* "errmsg" : , +* "errorOccured" : true +* } +* @param didError, did this round end in an error? +* @param executionTime, the time this round took to run +* @param candidateChunks, the number of chunks identified to be moved +* @param chunksMoved, the number of chunks moved +* @param errmsg, the error message for this round +*/ - /* - * Builds the details object for the actionlog. - * Current formats for detail are: - * Success: { - * "candidateChunks" : , - * "chunksMoved" : , - * "executionTimeMillis" : , - * "errorOccured" : false - * } - * Failure: { - * "executionTimeMillis" : , - * "errmsg" : , - * "errorOccured" : true - * } - * @param didError, did this round end in an error? - * @param executionTime, the time this round took to run - * @param candidateChunks, the number of chunks identified to be moved - * @param chunksMoved, the number of chunks moved - * @param errmsg, the error message for this round - */ - - static BSONObj _buildDetails( bool didError, int executionTime, - int candidateChunks, int chunksMoved, const std::string& errmsg ) { - - BSONObjBuilder builder; - builder.append("executionTimeMillis", executionTime); - builder.append("errorOccured", didError); - - if ( didError ) { - builder.append("errmsg", errmsg); - } else { - builder.append("candidateChunks", candidateChunks); - builder.append("chunksMoved", chunksMoved); - } - return builder.obj(); +static BSONObj _buildDetails(bool didError, + int executionTime, + int candidateChunks, + int chunksMoved, + const std::string& errmsg) { + BSONObjBuilder builder; + builder.append("executionTimeMillis", executionTime); + builder.append("errorOccured", didError); + + if (didError) { + builder.append("errmsg", errmsg); + } else { + builder.append("candidateChunks", candidateChunks); + builder.append("chunksMoved", chunksMoved); } + return builder.obj(); +} - bool Balancer::_checkOIDs() { - vector<ShardId> all; - grid.shardRegistry()->getAllShardIds(&all); +bool Balancer::_checkOIDs() { + vector<ShardId> all; + grid.shardRegistry()->getAllShardIds(&all); - // map of OID machine ID => shardId - map<int, string> oids; + // map of OID machine ID => shardId + map<int, string> oids; - for (const ShardId& shardId : all) { - const auto s = grid.shardRegistry()->getShard(shardId); - if (!s) { - continue; - } + for (const ShardId& shardId : all) { + const auto s = grid.shardRegistry()->getShard(shardId); + if (!s) { + continue; + } - BSONObj f = s->runCommand("admin", "features"); - if ( f["oidMachine"].isNumber() ) { - int x = f["oidMachine"].numberInt(); - if (oids.count(x) == 0) { - oids[x] = shardId; + BSONObj f = s->runCommand("admin", "features"); + if (f["oidMachine"].isNumber()) { + int x = f["oidMachine"].numberInt(); + if (oids.count(x) == 0) { + oids[x] = shardId; + } else { + log() << "error: 2 machines have " << x << " as oid machine piece: " << shardId + << " and " << oids[x]; + s->runCommand("admin", BSON("features" << 1 << "oidReset" << 1)); + + const auto otherShard = grid.shardRegistry()->getShard(oids[x]); + if (otherShard) { + otherShard->runCommand("admin", BSON("features" << 1 << "oidReset" << 1)); } - else { - log() << "error: 2 machines have " << x - << " as oid machine piece: " << shardId - << " and " << oids[x]; - s->runCommand("admin", BSON("features" << 1 << "oidReset" << 1)); - - const auto otherShard = grid.shardRegistry()->getShard(oids[x]); - if (otherShard) { - otherShard->runCommand("admin", BSON("features" << 1 << "oidReset" << 1)); - } - return false; - } - } - else { - log() << "warning: oidMachine not set on: " << s->toString(); + return false; } + } else { + log() << "warning: oidMachine not set on: " << s->toString(); } - return true; } - - /** - * Occasionally prints a log message with shard versions if the versions are not the same - * in the cluster. - */ - void warnOnMultiVersion( const ShardInfoMap& shardInfo ) { - - bool isMultiVersion = false; - for ( ShardInfoMap::const_iterator i = shardInfo.begin(); i != shardInfo.end(); ++i ) { - if ( !isSameMajorVersion( i->second.getMongoVersion().c_str() ) ) { - isMultiVersion = true; - break; - } - } + return true; +} - // If we're all the same version, don't message - if ( !isMultiVersion ) return; - - warning() << "multiVersion cluster detected, my version is " << versionString; - for ( ShardInfoMap::const_iterator i = shardInfo.begin(); i != shardInfo.end(); ++i ) { - log() << i->first << " is at version " << i->second.getMongoVersion(); - } +/** + * Occasionally prints a log message with shard versions if the versions are not the same + * in the cluster. + */ +void warnOnMultiVersion(const ShardInfoMap& shardInfo) { + bool isMultiVersion = false; + for (ShardInfoMap::const_iterator i = shardInfo.begin(); i != shardInfo.end(); ++i) { + if (!isSameMajorVersion(i->second.getMongoVersion().c_str())) { + isMultiVersion = true; + break; + } } - void Balancer::_doBalanceRound(vector<shared_ptr<MigrateInfo>>* candidateChunks) { - invariant(candidateChunks); + // If we're all the same version, don't message + if (!isMultiVersion) + return; - vector<CollectionType> collections; - Status collsStatus = grid.catalogManager()->getCollections(nullptr, &collections); - if (!collsStatus.isOK()) { - warning() << "Failed to retrieve the set of collections during balancing round " - << collsStatus; - return; - } + warning() << "multiVersion cluster detected, my version is " << versionString; + for (ShardInfoMap::const_iterator i = shardInfo.begin(); i != shardInfo.end(); ++i) { + log() << i->first << " is at version " << i->second.getMongoVersion(); + } +} - if (collections.empty()) { - LOG(1) << "no collections to balance"; - return; - } +void Balancer::_doBalanceRound(vector<shared_ptr<MigrateInfo>>* candidateChunks) { + invariant(candidateChunks); - // Get a list of all the shards that are participating in this balance round along with any - // maximum allowed quotas and current utilization. We get the latter by issuing - // db.serverStatus() (mem.mapped) to all shards. - // - // TODO: skip unresponsive shards and mark information as stale. - ShardInfoMap shardInfo; - Status loadStatus = DistributionStatus::populateShardInfoMap(&shardInfo); - if (!loadStatus.isOK()) { - warning() << "failed to load shard metadata" << causedBy(loadStatus); - return; - } + vector<CollectionType> collections; + Status collsStatus = grid.catalogManager()->getCollections(nullptr, &collections); + if (!collsStatus.isOK()) { + warning() << "Failed to retrieve the set of collections during balancing round " + << collsStatus; + return; + } - if (shardInfo.size() < 2) { - LOG(1) << "can't balance without more active shards"; - return; - } - - OCCASIONALLY warnOnMultiVersion( shardInfo ); + if (collections.empty()) { + LOG(1) << "no collections to balance"; + return; + } - // For each collection, check if the balancing policy recommends moving anything around. - for (const auto& coll : collections) { - // Skip collections for which balancing is disabled - const NamespaceString& ns = coll.getNs(); + // Get a list of all the shards that are participating in this balance round along with any + // maximum allowed quotas and current utilization. We get the latter by issuing + // db.serverStatus() (mem.mapped) to all shards. + // + // TODO: skip unresponsive shards and mark information as stale. + ShardInfoMap shardInfo; + Status loadStatus = DistributionStatus::populateShardInfoMap(&shardInfo); + if (!loadStatus.isOK()) { + warning() << "failed to load shard metadata" << causedBy(loadStatus); + return; + } - if (!coll.getAllowBalance()) { - LOG(1) << "Not balancing collection " << ns << "; explicitly disabled."; - continue; - } + if (shardInfo.size() < 2) { + LOG(1) << "can't balance without more active shards"; + return; + } - std::vector<ChunkType> allNsChunks; - grid.catalogManager()->getChunks(Query(BSON(ChunkType::ns(ns))) - .sort(ChunkType::min()), - 0, // all chunks - &allNsChunks); + OCCASIONALLY warnOnMultiVersion(shardInfo); - set<BSONObj> allChunkMinimums; - map<string, vector<ChunkType>> shardToChunksMap; + // For each collection, check if the balancing policy recommends moving anything around. + for (const auto& coll : collections) { + // Skip collections for which balancing is disabled + const NamespaceString& ns = coll.getNs(); - for (const ChunkType& chunk : allNsChunks) { - allChunkMinimums.insert(chunk.getMin().getOwned()); + if (!coll.getAllowBalance()) { + LOG(1) << "Not balancing collection " << ns << "; explicitly disabled."; + continue; + } - vector<ChunkType>& chunksList = shardToChunksMap[chunk.getShard()]; - chunksList.push_back(chunk); - } + std::vector<ChunkType> allNsChunks; + grid.catalogManager()->getChunks(Query(BSON(ChunkType::ns(ns))).sort(ChunkType::min()), + 0, // all chunks + &allNsChunks); - if (shardToChunksMap.empty()) { - LOG(1) << "skipping empty collection (" << ns << ")"; - continue; - } + set<BSONObj> allChunkMinimums; + map<string, vector<ChunkType>> shardToChunksMap; - for (ShardInfoMap::const_iterator i = shardInfo.begin(); i != shardInfo.end(); ++i) { - // This loop just makes sure there is an entry in shardToChunksMap for every shard - shardToChunksMap[i->first]; - } + for (const ChunkType& chunk : allNsChunks) { + allChunkMinimums.insert(chunk.getMin().getOwned()); - DistributionStatus status(shardInfo, shardToChunksMap); + vector<ChunkType>& chunksList = shardToChunksMap[chunk.getShard()]; + chunksList.push_back(chunk); + } - // TODO: TagRange contains all the information from TagsType except for the namespace, - // so maybe the two can be merged at some point in order to avoid the - // transformation below. - vector<TagRange> ranges; + if (shardToChunksMap.empty()) { + LOG(1) << "skipping empty collection (" << ns << ")"; + continue; + } - { - vector<TagsType> collectionTags; - uassertStatusOK(grid.catalogManager()->getTagsForCollection(ns.toString(), - &collectionTags)); - for (const auto& tt : collectionTags) { - ranges.push_back(TagRange(tt.getMinKey().getOwned(), - tt.getMaxKey().getOwned(), - tt.getTag())); - uassert(16356, - str::stream() << "tag ranges not valid for: " << ns.toString(), - status.addTagRange(ranges.back())); - } - } + for (ShardInfoMap::const_iterator i = shardInfo.begin(); i != shardInfo.end(); ++i) { + // This loop just makes sure there is an entry in shardToChunksMap for every shard + shardToChunksMap[i->first]; + } - auto statusGetDb = grid.catalogCache()->getDatabase(ns.db().toString()); - if (!statusGetDb.isOK()) { - warning() << "could not load db config to balance collection [" << ns << "]: " - << statusGetDb.getStatus(); - continue; + DistributionStatus status(shardInfo, shardToChunksMap); + + // TODO: TagRange contains all the information from TagsType except for the namespace, + // so maybe the two can be merged at some point in order to avoid the + // transformation below. + vector<TagRange> ranges; + + { + vector<TagsType> collectionTags; + uassertStatusOK( + grid.catalogManager()->getTagsForCollection(ns.toString(), &collectionTags)); + for (const auto& tt : collectionTags) { + ranges.push_back( + TagRange(tt.getMinKey().getOwned(), tt.getMaxKey().getOwned(), tt.getTag())); + uassert(16356, + str::stream() << "tag ranges not valid for: " << ns.toString(), + status.addTagRange(ranges.back())); } + } - shared_ptr<DBConfig> cfg = statusGetDb.getValue(); + auto statusGetDb = grid.catalogCache()->getDatabase(ns.db().toString()); + if (!statusGetDb.isOK()) { + warning() << "could not load db config to balance collection [" << ns + << "]: " << statusGetDb.getStatus(); + continue; + } - // This line reloads the chunk manager once if this process doesn't know the collection - // is sharded yet. - shared_ptr<ChunkManager> cm = cfg->getChunkManagerIfExists(ns, true); - if (!cm) { - warning() << "could not load chunks to balance " << ns << " collection"; - continue; - } + shared_ptr<DBConfig> cfg = statusGetDb.getValue(); - // Loop through tags to make sure no chunk spans tags. Split on tag min for all chunks. - bool didAnySplits = false; + // This line reloads the chunk manager once if this process doesn't know the collection + // is sharded yet. + shared_ptr<ChunkManager> cm = cfg->getChunkManagerIfExists(ns, true); + if (!cm) { + warning() << "could not load chunks to balance " << ns << " collection"; + continue; + } - for (const TagRange& range : ranges) { - BSONObj min = - cm->getShardKeyPattern().getKeyPattern().extendRangeBound(range.min, false); + // Loop through tags to make sure no chunk spans tags. Split on tag min for all chunks. + bool didAnySplits = false; - if (allChunkMinimums.count(min) > 0) { - continue; - } + for (const TagRange& range : ranges) { + BSONObj min = + cm->getShardKeyPattern().getKeyPattern().extendRangeBound(range.min, false); - didAnySplits = true; + if (allChunkMinimums.count(min) > 0) { + continue; + } - log() << "ns: " << ns << " need to split on " << min - << " because there is a range there"; + didAnySplits = true; - ChunkPtr c = cm->findIntersectingChunk(min); + log() << "ns: " << ns << " need to split on " << min + << " because there is a range there"; - vector<BSONObj> splitPoints; - splitPoints.push_back( min ); + ChunkPtr c = cm->findIntersectingChunk(min); - Status status = c->multiSplit(splitPoints, NULL); - if (!status.isOK()) { - error() << "split failed: " << status; - } - else { - LOG(1) << "split worked"; - } + vector<BSONObj> splitPoints; + splitPoints.push_back(min); - break; + Status status = c->multiSplit(splitPoints, NULL); + if (!status.isOK()) { + error() << "split failed: " << status; + } else { + LOG(1) << "split worked"; } - if (didAnySplits) { - // State change, just wait till next round - continue; - } + break; + } - shared_ptr<MigrateInfo> migrateInfo(_policy->balance(ns, status, _balancedLastTime)); - if (migrateInfo) { - candidateChunks->push_back(migrateInfo); - } + if (didAnySplits) { + // State change, just wait till next round + continue; + } + + shared_ptr<MigrateInfo> migrateInfo(_policy->balance(ns, status, _balancedLastTime)); + if (migrateInfo) { + candidateChunks->push_back(migrateInfo); } } +} - bool Balancer::_init() { - try { - log() << "about to contact config servers and shards"; +bool Balancer::_init() { + try { + log() << "about to contact config servers and shards"; - // contact the config server and refresh shard information - // checks that each shard is indeed a different process (no hostname mixup) - // these checks are redundant in that they're redone at every new round but we want to do them initially here - // so to catch any problem soon - Shard::reloadShardInfo(); - _checkOIDs(); + // contact the config server and refresh shard information + // checks that each shard is indeed a different process (no hostname mixup) + // these checks are redundant in that they're redone at every new round but we want to do them initially here + // so to catch any problem soon + Shard::reloadShardInfo(); + _checkOIDs(); - log() << "config servers and shards contacted successfully"; + log() << "config servers and shards contacted successfully"; - StringBuilder buf; - buf << getHostNameCached() << ":" << serverGlobalParams.port; - _myid = buf.str(); - _started = time(0); + StringBuilder buf; + buf << getHostNameCached() << ":" << serverGlobalParams.port; + _myid = buf.str(); + _started = time(0); - log() << "balancer id: " << _myid << " started"; + log() << "balancer id: " << _myid << " started"; - return true; + return true; + } catch (std::exception& e) { + warning() << "could not initialize balancer, please check that all shards and config " + "servers are up: " << e.what(); + return false; + } +} + +void Balancer::run() { + Client::initThread("Balancer"); + + // This is the body of a BackgroundJob so if we throw here we're basically ending the + // balancer thread prematurely. + while (!inShutdown()) { + if (!_init()) { + log() << "will retry to initialize balancer in one minute"; + sleepsecs(60); + continue; } - catch ( std::exception& e ) { - warning() << "could not initialize balancer, please check that all shards and config servers are up: " << e.what(); - return false; - } + break; } - void Balancer::run() { - Client::initThread("Balancer"); + const int sleepTime = 10; - // This is the body of a BackgroundJob so if we throw here we're basically ending the - // balancer thread prematurely. - while (!inShutdown()) { - if (!_init()) { - log() << "will retry to initialize balancer in one minute"; - sleepsecs(60); - continue; - } + while (!inShutdown()) { + Timer balanceRoundTimer; + ActionLogType actionLog; - break; - } + actionLog.setServer(getHostNameCached()); + actionLog.setWhat("balancer.round"); + + try { + // ping has to be first so we keep things in the config server in sync + _ping(); - const int sleepTime = 10; + BSONObj balancerResult; - while (!inShutdown()) { - Timer balanceRoundTimer; - ActionLogType actionLog; + // use fresh shard state + Shard::reloadShardInfo(); - actionLog.setServer(getHostNameCached()); - actionLog.setWhat("balancer.round"); + // refresh chunk size (even though another balancer might be active) + Chunk::refreshChunkSize(); - try { - // ping has to be first so we keep things in the config server in sync - _ping(); + auto balSettingsResult = + grid.catalogManager()->getGlobalSettings(SettingsType::BalancerDocKey); + const bool isBalSettingsAbsent = + balSettingsResult.getStatus() == ErrorCodes::NoMatchingDocument; + if (!balSettingsResult.isOK() && !isBalSettingsAbsent) { + warning() << balSettingsResult.getStatus(); + return; + } + const SettingsType& balancerConfig = + isBalSettingsAbsent ? SettingsType{} : balSettingsResult.getValue(); - BSONObj balancerResult; + // now make sure we should even be running + if ((!isBalSettingsAbsent && !grid.shouldBalance(balancerConfig)) || + MONGO_FAIL_POINT(skipBalanceRound)) { + LOG(1) << "skipping balancing round because balancing is disabled"; - // use fresh shard state - Shard::reloadShardInfo(); + // Ping again so scripts can determine if we're active without waiting + _ping(true); - // refresh chunk size (even though another balancer might be active) - Chunk::refreshChunkSize(); + sleepsecs(sleepTime); + continue; + } - auto balSettingsResult = - grid.catalogManager()->getGlobalSettings(SettingsType::BalancerDocKey); - const bool isBalSettingsAbsent = - balSettingsResult.getStatus() == ErrorCodes::NoMatchingDocument; - if (!balSettingsResult.isOK() && !isBalSettingsAbsent) { - warning() << balSettingsResult.getStatus(); - return; - } - const SettingsType& balancerConfig = isBalSettingsAbsent ? - SettingsType{} : balSettingsResult.getValue(); + uassert(13258, "oids broken after resetting!", _checkOIDs()); - // now make sure we should even be running - if ((!isBalSettingsAbsent && !grid.shouldBalance(balancerConfig)) || - MONGO_FAIL_POINT(skipBalanceRound)) { + { + auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock( + "balancer", "doing balance round"); - LOG(1) << "skipping balancing round because balancing is disabled"; + if (!scopedDistLock.isOK()) { + LOG(1) << "skipping balancing round" << causedBy(scopedDistLock.getStatus()); // Ping again so scripts can determine if we're active without waiting - _ping( true ); + _ping(true); - sleepsecs( sleepTime ); + sleepsecs(sleepTime); // no need to wake up soon continue; } - uassert( 13258 , "oids broken after resetting!" , _checkOIDs() ); - - { - auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock( - "balancer", "doing balance round"); - - if (!scopedDistLock.isOK()) { - LOG(1) << "skipping balancing round" - << causedBy(scopedDistLock.getStatus()); - - // Ping again so scripts can determine if we're active without waiting - _ping( true ); - - sleepsecs( sleepTime ); // no need to wake up soon - continue; - } - - const bool waitForDelete = (balancerConfig.isWaitForDeleteSet() ? - balancerConfig.getWaitForDelete() : false); - - std::unique_ptr<WriteConcernOptions> writeConcern; - if (balancerConfig.isKeySet()) { // if balancer doc exists. - writeConcern = std::move(balancerConfig.getWriteConcern()); - } - - LOG(1) << "*** start balancing round. " - << "waitForDelete: " << waitForDelete - << ", secondaryThrottle: " - << (writeConcern.get() ? writeConcern->toBSON().toString() : "default") - ; - - vector<shared_ptr<MigrateInfo>> candidateChunks; - _doBalanceRound(&candidateChunks); - - if ( candidateChunks.size() == 0 ) { - LOG(1) << "no need to move any chunk"; - _balancedLastTime = 0; - } - else { - _balancedLastTime = _moveChunks(candidateChunks, - writeConcern.get(), - waitForDelete); - } - - actionLog.setDetails( - _buildDetails(false, - balanceRoundTimer.millis(), - static_cast<int>(candidateChunks.size()), - _balancedLastTime, - "")); - actionLog.setTime(jsTime()); - - grid.catalogManager()->logAction(actionLog); - - LOG(1) << "*** end of balancing round"; + const bool waitForDelete = + (balancerConfig.isWaitForDeleteSet() ? balancerConfig.getWaitForDelete() + : false); + + std::unique_ptr<WriteConcernOptions> writeConcern; + if (balancerConfig.isKeySet()) { // if balancer doc exists. + writeConcern = std::move(balancerConfig.getWriteConcern()); } - // Ping again so scripts can determine if we're active without waiting - _ping(true); + LOG(1) << "*** start balancing round. " + << "waitForDelete: " << waitForDelete << ", secondaryThrottle: " + << (writeConcern.get() ? writeConcern->toBSON().toString() : "default"); - sleepsecs(_balancedLastTime ? sleepTime / 10 : sleepTime); - } - catch ( std::exception& e ) { - log() << "caught exception while doing balance: " << e.what(); - - // Just to match the opening statement if in log level 1 - LOG(1) << "*** End of balancing round"; - - // This round failed, tell the world! - actionLog.setDetails( - _buildDetails(true, - balanceRoundTimer.millis(), - 0, - 0, - e.what())); + vector<shared_ptr<MigrateInfo>> candidateChunks; + _doBalanceRound(&candidateChunks); + + if (candidateChunks.size() == 0) { + LOG(1) << "no need to move any chunk"; + _balancedLastTime = 0; + } else { + _balancedLastTime = + _moveChunks(candidateChunks, writeConcern.get(), waitForDelete); + } + + actionLog.setDetails(_buildDetails(false, + balanceRoundTimer.millis(), + static_cast<int>(candidateChunks.size()), + _balancedLastTime, + "")); actionLog.setTime(jsTime()); grid.catalogManager()->logAction(actionLog); - // Sleep a fair amount before retrying because of the error - sleepsecs(sleepTime); - - continue; + LOG(1) << "*** end of balancing round"; } - } + // Ping again so scripts can determine if we're active without waiting + _ping(true); + + sleepsecs(_balancedLastTime ? sleepTime / 10 : sleepTime); + } catch (std::exception& e) { + log() << "caught exception while doing balance: " << e.what(); + + // Just to match the opening statement if in log level 1 + LOG(1) << "*** End of balancing round"; + + // This round failed, tell the world! + actionLog.setDetails(_buildDetails(true, balanceRoundTimer.millis(), 0, 0, e.what())); + actionLog.setTime(jsTime()); + + grid.catalogManager()->logAction(actionLog); + + // Sleep a fair amount before retrying because of the error + sleepsecs(sleepTime); + + continue; + } } +} } // namespace mongo diff --git a/src/mongo/s/balance.h b/src/mongo/s/balance.h index 1b7bd55aa29..c62e05c18f4 100644 --- a/src/mongo/s/balance.h +++ b/src/mongo/s/balance.h @@ -35,85 +35,86 @@ namespace mongo { - class BalancerPolicy; - struct MigrateInfo; - struct WriteConcernOptions; +class BalancerPolicy; +struct MigrateInfo; +struct WriteConcernOptions; + +/** + * The balancer is a background task that tries to keep the number of chunks across all + * servers of the cluster even. Although every mongos will have one balancer running, only one + * of them will be active at the any given point in time. The balancer uses a distributed lock + * for that coordination. + * + * The balancer does act continuously but in "rounds". At a given round, it would decide if + * there is an imbalance by checking the difference in chunks between the most and least + * loaded shards. It would issue a request for a chunk migration per round, if it found so. + */ +class Balancer : public BackgroundJob { +public: + Balancer(); + virtual ~Balancer(); + + // BackgroundJob methods + + virtual void run(); + + virtual std::string name() const { + return "Balancer"; + } + +private: + // hostname:port of my mongos + std::string _myid; + + // time the Balancer started running + time_t _started; + + // number of moved chunks in last round + int _balancedLastTime; + + // decide which chunks to move; owned here. + std::unique_ptr<BalancerPolicy> _policy; /** - * The balancer is a background task that tries to keep the number of chunks across all - * servers of the cluster even. Although every mongos will have one balancer running, only one - * of them will be active at the any given point in time. The balancer uses a distributed lock - * for that coordination. + * Checks that the balancer can connect to all servers it needs to do its job. + * + * @return true if balancing can be started * - * The balancer does act continuously but in "rounds". At a given round, it would decide if - * there is an imbalance by checking the difference in chunks between the most and least - * loaded shards. It would issue a request for a chunk migration per round, if it found so. + * This method throws on a network exception */ - class Balancer : public BackgroundJob { - public: - Balancer(); - virtual ~Balancer(); - - // BackgroundJob methods - - virtual void run(); - - virtual std::string name() const { return "Balancer"; } - - private: - // hostname:port of my mongos - std::string _myid; - - // time the Balancer started running - time_t _started; - - // number of moved chunks in last round - int _balancedLastTime; - - // decide which chunks to move; owned here. - std::unique_ptr<BalancerPolicy> _policy; - - /** - * Checks that the balancer can connect to all servers it needs to do its job. - * - * @return true if balancing can be started - * - * This method throws on a network exception - */ - bool _init(); - - /** - * Gathers all the necessary information about shards and chunks, and decides whether there are candidate chunks to - * be moved. - * - * @param conn is the connection with the config server(s) - * @param candidateChunks (IN/OUT) filled with candidate chunks, one per collection, that could possibly be moved - */ - void _doBalanceRound(std::vector<std::shared_ptr<MigrateInfo>>* candidateChunks); - - /** - * Issues chunk migration request, one at a time. - * - * @param candidateChunks possible chunks to move - * @param writeConcern detailed write concern. NULL means the default write concern. - * @param waitForDelete wait for deletes to complete after each chunk move - * @return number of chunks effectively moved - */ - int _moveChunks(const std::vector<std::shared_ptr<MigrateInfo>>& candidateChunks, - const WriteConcernOptions* writeConcern, - bool waitForDelete); - - /** - * Marks this balancer as being live on the config server(s). - */ - void _ping( bool waiting = false ); - - /** - * @return true if all the servers listed in configdb as being shards are reachable and are distinct processes - */ - bool _checkOIDs(); - - }; - - extern Balancer balancer; + bool _init(); + + /** + * Gathers all the necessary information about shards and chunks, and decides whether there are candidate chunks to + * be moved. + * + * @param conn is the connection with the config server(s) + * @param candidateChunks (IN/OUT) filled with candidate chunks, one per collection, that could possibly be moved + */ + void _doBalanceRound(std::vector<std::shared_ptr<MigrateInfo>>* candidateChunks); + + /** + * Issues chunk migration request, one at a time. + * + * @param candidateChunks possible chunks to move + * @param writeConcern detailed write concern. NULL means the default write concern. + * @param waitForDelete wait for deletes to complete after each chunk move + * @return number of chunks effectively moved + */ + int _moveChunks(const std::vector<std::shared_ptr<MigrateInfo>>& candidateChunks, + const WriteConcernOptions* writeConcern, + bool waitForDelete); + + /** + * Marks this balancer as being live on the config server(s). + */ + void _ping(bool waiting = false); + + /** + * @return true if all the servers listed in configdb as being shards are reachable and are distinct processes + */ + bool _checkOIDs(); +}; + +extern Balancer balancer; } diff --git a/src/mongo/s/balancer_policy.cpp b/src/mongo/s/balancer_policy.cpp index 7309261c902..cbfe1c8b9ab 100644 --- a/src/mongo/s/balancer_policy.cpp +++ b/src/mongo/s/balancer_policy.cpp @@ -46,511 +46,484 @@ namespace mongo { - using std::map; - using std::numeric_limits; - using std::set; - using std::string; - using std::vector; - - string TagRange::toString() const { - return str::stream() << min << " -->> " << max << " on " << tag; +using std::map; +using std::numeric_limits; +using std::set; +using std::string; +using std::vector; + +string TagRange::toString() const { + return str::stream() << min << " -->> " << max << " on " << tag; +} + +DistributionStatus::DistributionStatus(const ShardInfoMap& shardInfo, + const ShardToChunksMap& shardToChunksMap) + : _shardInfo(shardInfo), _shardChunks(shardToChunksMap) { + for (ShardInfoMap::const_iterator i = _shardInfo.begin(); i != _shardInfo.end(); ++i) { + _shardIds.insert(i->first); } +} - DistributionStatus::DistributionStatus(const ShardInfoMap& shardInfo, - const ShardToChunksMap& shardToChunksMap) - : _shardInfo(shardInfo), - _shardChunks(shardToChunksMap) { +const ShardInfo& DistributionStatus::shardInfo(const ShardId& shardId) const { + ShardInfoMap::const_iterator i = _shardInfo.find(shardId); + verify(i != _shardInfo.end()); + return i->second; +} - for (ShardInfoMap::const_iterator i = _shardInfo.begin(); i != _shardInfo.end(); ++i) { - _shardIds.insert(i->first); - } - } +unsigned DistributionStatus::totalChunks() const { + unsigned total = 0; - const ShardInfo& DistributionStatus::shardInfo(const ShardId& shardId) const { - ShardInfoMap::const_iterator i = _shardInfo.find(shardId); - verify( i != _shardInfo.end() ); - return i->second; + for (ShardToChunksMap::const_iterator i = _shardChunks.begin(); i != _shardChunks.end(); ++i) { + total += i->second.size(); } - unsigned DistributionStatus::totalChunks() const { - unsigned total = 0; + return total; +} - for (ShardToChunksMap::const_iterator i = _shardChunks.begin(); - i != _shardChunks.end(); - ++i) { - - total += i->second.size(); - } - - return total; +unsigned DistributionStatus::numberOfChunksInShard(const ShardId& shardId) const { + ShardToChunksMap::const_iterator i = _shardChunks.find(shardId); + if (i == _shardChunks.end()) { + return 0; } - unsigned DistributionStatus::numberOfChunksInShard(const ShardId& shardId) const { - ShardToChunksMap::const_iterator i = _shardChunks.find(shardId); - if (i == _shardChunks.end()) { - return 0; - } + return i->second.size(); +} - return i->second.size(); +unsigned DistributionStatus::numberOfChunksInShardWithTag(const ShardId& shardId, + const string& tag) const { + ShardToChunksMap::const_iterator i = _shardChunks.find(shardId); + if (i == _shardChunks.end()) { + return 0; } - unsigned DistributionStatus::numberOfChunksInShardWithTag(const ShardId& shardId, - const string& tag) const { - ShardToChunksMap::const_iterator i = _shardChunks.find(shardId); - if (i == _shardChunks.end()) { - return 0; - } - - unsigned total = 0; + unsigned total = 0; - const vector<ChunkType>& chunkList = i->second; - for (unsigned j = 0; j < i->second.size(); j++) { - if (tag == getTagForChunk(chunkList[j])) { - total++; - } + const vector<ChunkType>& chunkList = i->second; + for (unsigned j = 0; j < i->second.size(); j++) { + if (tag == getTagForChunk(chunkList[j])) { + total++; } - - return total; } - string DistributionStatus::getBestReceieverShard( const string& tag ) const { - string best; - unsigned minChunks = numeric_limits<unsigned>::max(); + return total; +} - for ( ShardInfoMap::const_iterator i = _shardInfo.begin(); i != _shardInfo.end(); ++i ) { - if ( i->second.isSizeMaxed() ) { - LOG(1) << i->first << " has already reached the maximum total chunk size."; - continue; - } +string DistributionStatus::getBestReceieverShard(const string& tag) const { + string best; + unsigned minChunks = numeric_limits<unsigned>::max(); - if ( i->second.isDraining() ) { - LOG(1) << i->first << " is currently draining."; - continue; - } + for (ShardInfoMap::const_iterator i = _shardInfo.begin(); i != _shardInfo.end(); ++i) { + if (i->second.isSizeMaxed()) { + LOG(1) << i->first << " has already reached the maximum total chunk size."; + continue; + } - if ( ! i->second.hasTag( tag ) ) { - LOG(1) << i->first << " doesn't have right tag"; - continue; - } + if (i->second.isDraining()) { + LOG(1) << i->first << " is currently draining."; + continue; + } - unsigned myChunks = numberOfChunksInShard( i->first ); - if ( myChunks >= minChunks ) { - LOG(1) << i->first << " has more chunks me:" << myChunks << " best: " << best << ":" << minChunks; - continue; - } + if (!i->second.hasTag(tag)) { + LOG(1) << i->first << " doesn't have right tag"; + continue; + } - best = i->first; - minChunks = myChunks; + unsigned myChunks = numberOfChunksInShard(i->first); + if (myChunks >= minChunks) { + LOG(1) << i->first << " has more chunks me:" << myChunks << " best: " << best << ":" + << minChunks; + continue; } - return best; + best = i->first; + minChunks = myChunks; } - string DistributionStatus::getMostOverloadedShard( const string& tag ) const { - string worst; - unsigned maxChunks = 0; + return best; +} - for ( ShardInfoMap::const_iterator i = _shardInfo.begin(); i != _shardInfo.end(); ++i ) { - unsigned myChunks = numberOfChunksInShardWithTag( i->first, tag ); - if ( myChunks <= maxChunks ) - continue; +string DistributionStatus::getMostOverloadedShard(const string& tag) const { + string worst; + unsigned maxChunks = 0; - worst = i->first; - maxChunks = myChunks; - } + for (ShardInfoMap::const_iterator i = _shardInfo.begin(); i != _shardInfo.end(); ++i) { + unsigned myChunks = numberOfChunksInShardWithTag(i->first, tag); + if (myChunks <= maxChunks) + continue; - return worst; + worst = i->first; + maxChunks = myChunks; } - const vector<ChunkType>& DistributionStatus::getChunks(const ShardId& shardId) const { - ShardToChunksMap::const_iterator i = _shardChunks.find(shardId); - invariant(i != _shardChunks.end()); + return worst; +} - return i->second; - } +const vector<ChunkType>& DistributionStatus::getChunks(const ShardId& shardId) const { + ShardToChunksMap::const_iterator i = _shardChunks.find(shardId); + invariant(i != _shardChunks.end()); - bool DistributionStatus::addTagRange( const TagRange& range ) { - // first check for overlaps - for ( map<BSONObj,TagRange>::const_iterator i = _tagRanges.begin(); - i != _tagRanges.end(); - ++i ) { - const TagRange& tocheck = i->second; + return i->second; +} - if ( range.min == tocheck.min ) { - LOG(1) << "have 2 ranges with the same min " << range << " " << tocheck; - return false; - } +bool DistributionStatus::addTagRange(const TagRange& range) { + // first check for overlaps + for (map<BSONObj, TagRange>::const_iterator i = _tagRanges.begin(); i != _tagRanges.end(); + ++i) { + const TagRange& tocheck = i->second; - if ( range.min < tocheck.min ) { - if ( range.max > tocheck.min ) { - LOG(1) << "have overlapping ranges " << range << " " << tocheck; - return false; - } + if (range.min == tocheck.min) { + LOG(1) << "have 2 ranges with the same min " << range << " " << tocheck; + return false; + } + + if (range.min < tocheck.min) { + if (range.max > tocheck.min) { + LOG(1) << "have overlapping ranges " << range << " " << tocheck; + return false; } - else { - // range.min > tocheck.min - if ( tocheck.max > range.min ) { - LOG(1) << "have overlapping ranges " << range << " " << tocheck; - return false; - } + } else { + // range.min > tocheck.min + if (tocheck.max > range.min) { + LOG(1) << "have overlapping ranges " << range << " " << tocheck; + return false; } - } - - _tagRanges[range.max.getOwned()] = range; - _allTags.insert( range.tag ); - - return true; } - string DistributionStatus::getTagForChunk( const ChunkType& chunk ) const { - if ( _tagRanges.size() == 0 ) - return ""; + _tagRanges[range.max.getOwned()] = range; + _allTags.insert(range.tag); - const BSONObj min(chunk.getMin()); + return true; +} - map<BSONObj,TagRange>::const_iterator i = _tagRanges.upper_bound( min ); - if ( i == _tagRanges.end() ) - return ""; +string DistributionStatus::getTagForChunk(const ChunkType& chunk) const { + if (_tagRanges.size() == 0) + return ""; - const TagRange& range = i->second; - if ( min < range.min ) - return ""; + const BSONObj min(chunk.getMin()); - return range.tag; - } + map<BSONObj, TagRange>::const_iterator i = _tagRanges.upper_bound(min); + if (i == _tagRanges.end()) + return ""; - void DistributionStatus::dump() const { - log() << "DistributionStatus"; - log() << " shards"; + const TagRange& range = i->second; + if (min < range.min) + return ""; - for ( ShardInfoMap::const_iterator i = _shardInfo.begin(); i != _shardInfo.end(); ++i ) { - log() << " " << i->first << "\t" << i->second.toString(); + return range.tag; +} - ShardToChunksMap::const_iterator j = _shardChunks.find(i->first); - verify(j != _shardChunks.end()); +void DistributionStatus::dump() const { + log() << "DistributionStatus"; + log() << " shards"; - const vector<ChunkType>& v = j->second; - for (unsigned x = 0; x < v.size(); x++) { - log() << " " << v[x]; - } - } + for (ShardInfoMap::const_iterator i = _shardInfo.begin(); i != _shardInfo.end(); ++i) { + log() << " " << i->first << "\t" << i->second.toString(); - if ( _tagRanges.size() > 0 ) { - log() << " tag ranges"; + ShardToChunksMap::const_iterator j = _shardChunks.find(i->first); + verify(j != _shardChunks.end()); - for ( map<BSONObj,TagRange>::const_iterator i = _tagRanges.begin(); - i != _tagRanges.end(); - ++i ) - log() << i->second.toString(); + const vector<ChunkType>& v = j->second; + for (unsigned x = 0; x < v.size(); x++) { + log() << " " << v[x]; } } - Status DistributionStatus::populateShardInfoMap(ShardInfoMap* shardInfo) { - try { - vector<ShardType> shards; - Status status = grid.catalogManager()->getAllShards(&shards); - if (!status.isOK()) { - return status; - } + if (_tagRanges.size() > 0) { + log() << " tag ranges"; - for (const ShardType& shardData : shards) { - std::shared_ptr<Shard> shard = grid.shardRegistry()->getShard(shardData.getName()); + for (map<BSONObj, TagRange>::const_iterator i = _tagRanges.begin(); i != _tagRanges.end(); + ++i) + log() << i->second.toString(); + } +} + +Status DistributionStatus::populateShardInfoMap(ShardInfoMap* shardInfo) { + try { + vector<ShardType> shards; + Status status = grid.catalogManager()->getAllShards(&shards); + if (!status.isOK()) { + return status; + } - // The shard must still exist in the registry. If it doesn't, which may happen in - // the very low proability case that it gets dropped between the call to - // getAllShards above and the call to getShard, just don't account for it since - // it is missing anyways. - if (!shard) { - warning() << "Shard [" << shardData.getName() << "] was not found. Skipping."; - continue; - } + for (const ShardType& shardData : shards) { + std::shared_ptr<Shard> shard = grid.shardRegistry()->getShard(shardData.getName()); - ShardStatus shardStatus = shard->getStatus(); + // The shard must still exist in the registry. If it doesn't, which may happen in + // the very low proability case that it gets dropped between the call to + // getAllShards above and the call to getShard, just don't account for it since + // it is missing anyways. + if (!shard) { + warning() << "Shard [" << shardData.getName() << "] was not found. Skipping."; + continue; + } - std::set<std::string> dummy; + ShardStatus shardStatus = shard->getStatus(); - ShardInfo newShardEntry(shardData.getMaxSizeMB(), - shardStatus.dataSizeBytes() / 1024 / 1024, - shardData.getDraining(), - dummy, - shardStatus.mongoVersion()); + std::set<std::string> dummy; - for (const string& shardTag : shardData.getTags()) { - newShardEntry.addTag(shardTag); - } + ShardInfo newShardEntry(shardData.getMaxSizeMB(), + shardStatus.dataSizeBytes() / 1024 / 1024, + shardData.getDraining(), + dummy, + shardStatus.mongoVersion()); - shardInfo->insert(make_pair(shardData.getName(), newShardEntry)); + for (const string& shardTag : shardData.getTags()) { + newShardEntry.addTag(shardTag); } - } - catch (const DBException& ex) { - return ex.toStatus(); - } - return Status::OK(); + shardInfo->insert(make_pair(shardData.getName(), newShardEntry)); + } + } catch (const DBException& ex) { + return ex.toStatus(); } - void DistributionStatus::populateShardToChunksMap(const ShardInfoMap& allShards, - const ChunkManager& chunkMgr, - ShardToChunksMap* shardToChunksMap) { + return Status::OK(); +} - // Makes sure there is an entry in shardToChunksMap for every shard. - for (ShardInfoMap::const_iterator it = allShards.begin(); it != allShards.end(); ++it) { - (*shardToChunksMap)[it->first]; - } +void DistributionStatus::populateShardToChunksMap(const ShardInfoMap& allShards, + const ChunkManager& chunkMgr, + ShardToChunksMap* shardToChunksMap) { + // Makes sure there is an entry in shardToChunksMap for every shard. + for (ShardInfoMap::const_iterator it = allShards.begin(); it != allShards.end(); ++it) { + (*shardToChunksMap)[it->first]; + } - const ChunkMap& chunkMap = chunkMgr.getChunkMap(); - for (ChunkMap::const_iterator it = chunkMap.begin(); it != chunkMap.end(); ++it) { - const ChunkPtr chunkPtr = it->second; + const ChunkMap& chunkMap = chunkMgr.getChunkMap(); + for (ChunkMap::const_iterator it = chunkMap.begin(); it != chunkMap.end(); ++it) { + const ChunkPtr chunkPtr = it->second; - ChunkType chunk; - chunk.setNS(chunkMgr.getns()); - chunk.setMin(chunkPtr->getMin().getOwned()); - chunk.setMax(chunkPtr->getMax().getOwned()); - chunk.setJumbo(chunkPtr->isJumbo()); // TODO: is this reliable? + ChunkType chunk; + chunk.setNS(chunkMgr.getns()); + chunk.setMin(chunkPtr->getMin().getOwned()); + chunk.setMax(chunkPtr->getMax().getOwned()); + chunk.setJumbo(chunkPtr->isJumbo()); // TODO: is this reliable? - const string shardName(chunkPtr->getShardId()); - chunk.setShard(shardName); + const string shardName(chunkPtr->getShardId()); + chunk.setShard(shardName); - (*shardToChunksMap)[shardName].push_back(chunk); - } + (*shardToChunksMap)[shardName].push_back(chunk); } +} - MigrateInfo* BalancerPolicy::balance( const string& ns, - const DistributionStatus& distribution, - int balancedLastTime ) { +MigrateInfo* BalancerPolicy::balance(const string& ns, + const DistributionStatus& distribution, + int balancedLastTime) { + // 1) check for shards that policy require to us to move off of: + // draining only + // 2) check tag policy violations + // 3) then we make sure chunks are balanced for each tag + // ---- - // 1) check for shards that policy require to us to move off of: - // draining only - // 2) check tag policy violations - // 3) then we make sure chunks are balanced for each tag + // 1) check things we have to move + { + for (const ShardId& shardId : distribution.shardIds()) { + const ShardInfo& info = distribution.shardInfo(shardId); - // ---- + if (!info.isDraining()) + continue; - // 1) check things we have to move - { - for (const ShardId& shardId : distribution.shardIds()) { - const ShardInfo& info = distribution.shardInfo(shardId); + if (distribution.numberOfChunksInShard(shardId) == 0) + continue; - if ( ! info.isDraining() ) - continue; + // now we know we need to move to chunks off this shard + // we will if we are allowed + const vector<ChunkType>& chunks = distribution.getChunks(shardId); + unsigned numJumboChunks = 0; - if (distribution.numberOfChunksInShard(shardId) == 0) + // since we have to move all chunks, lets just do in order + for (unsigned i = 0; i < chunks.size(); i++) { + const ChunkType& chunkToMove = chunks[i]; + if (chunkToMove.getJumbo()) { + numJumboChunks++; continue; - - // now we know we need to move to chunks off this shard - // we will if we are allowed - const vector<ChunkType>& chunks = distribution.getChunks(shardId); - unsigned numJumboChunks = 0; - - // since we have to move all chunks, lets just do in order - for ( unsigned i=0; i<chunks.size(); i++ ) { - const ChunkType& chunkToMove = chunks[i]; - if (chunkToMove.getJumbo()) { - numJumboChunks++; - continue; - } - - string tag = distribution.getTagForChunk( chunkToMove ); - const ShardId to = distribution.getBestReceieverShard( tag ); - - if ( to.size() == 0 ) { - warning() << "want to move chunk: " << chunkToMove - << "(" << tag << ") " - << "from " << shardId - << " but can't find anywhere to put it"; - continue; - } - - log() << "going to move " << chunkToMove - << " from " << shardId - << "(" << tag << ")" - << " to " << to; - - return new MigrateInfo(ns, to, shardId, chunkToMove.toBSON()); } - warning() << "can't find any chunk to move from: " << shardId - << " but we want to. " - << " numJumboChunks: " << numJumboChunks; - } - } + string tag = distribution.getTagForChunk(chunkToMove); + const ShardId to = distribution.getBestReceieverShard(tag); - // 2) tag violations - if ( distribution.tags().size() > 0 ) { - - for (const ShardId& shardId : distribution.shardIds()) { - const ShardInfo& info = distribution.shardInfo(shardId); - - const vector<ChunkType>& chunks = distribution.getChunks(shardId); - for ( unsigned j = 0; j < chunks.size(); j++ ) { - const ChunkType& chunk = chunks[j]; - string tag = distribution.getTagForChunk(chunk); - - if ( info.hasTag( tag ) ) - continue; - - // uh oh, this chunk is in the wrong place - log() << "chunk " << chunk - << " is not on a shard with the right tag: " - << tag; - - if (chunk.getJumbo()) { - warning() << "chunk " << chunk << " is jumbo, so cannot be moved"; - continue; - } - - const ShardId to = distribution.getBestReceieverShard( tag ); - if ( to.size() == 0 ) { - log() << "no where to put it :("; - continue; - } - verify(to != shardId); - log() << " going to move to: " << to; - return new MigrateInfo(ns, to, shardId, chunk.toBSON()); + if (to.size() == 0) { + warning() << "want to move chunk: " << chunkToMove << "(" << tag << ") " + << "from " << shardId << " but can't find anywhere to put it"; + continue; } + + log() << "going to move " << chunkToMove << " from " << shardId << "(" << tag << ")" + << " to " << to; + + return new MigrateInfo(ns, to, shardId, chunkToMove.toBSON()); } + + warning() << "can't find any chunk to move from: " << shardId << " but we want to. " + << " numJumboChunks: " << numJumboChunks; } + } - // 3) for each tag balance + // 2) tag violations + if (distribution.tags().size() > 0) { + for (const ShardId& shardId : distribution.shardIds()) { + const ShardInfo& info = distribution.shardInfo(shardId); - int threshold = 8; - if ( balancedLastTime || distribution.totalChunks() < 20 ) - threshold = 2; - else if ( distribution.totalChunks() < 80 ) - threshold = 4; + const vector<ChunkType>& chunks = distribution.getChunks(shardId); + for (unsigned j = 0; j < chunks.size(); j++) { + const ChunkType& chunk = chunks[j]; + string tag = distribution.getTagForChunk(chunk); - // randomize the order in which we balance the tags - // this is so that one bad tag doesn't prevent others from getting balanced - vector<string> tags; - { - set<string> t = distribution.tags(); - for ( set<string>::const_iterator i = t.begin(); i != t.end(); ++i ) - tags.push_back( *i ); - tags.push_back( "" ); + if (info.hasTag(tag)) + continue; + + // uh oh, this chunk is in the wrong place + log() << "chunk " << chunk << " is not on a shard with the right tag: " << tag; - std::random_shuffle( tags.begin(), tags.end() ); + if (chunk.getJumbo()) { + warning() << "chunk " << chunk << " is jumbo, so cannot be moved"; + continue; + } + + const ShardId to = distribution.getBestReceieverShard(tag); + if (to.size() == 0) { + log() << "no where to put it :("; + continue; + } + verify(to != shardId); + log() << " going to move to: " << to; + return new MigrateInfo(ns, to, shardId, chunk.toBSON()); + } } + } - for ( unsigned i=0; i<tags.size(); i++ ) { - string tag = tags[i]; + // 3) for each tag balance - const ShardId from = distribution.getMostOverloadedShard(tag); - if ( from.size() == 0 ) - continue; + int threshold = 8; + if (balancedLastTime || distribution.totalChunks() < 20) + threshold = 2; + else if (distribution.totalChunks() < 80) + threshold = 4; - unsigned max = distribution.numberOfChunksInShardWithTag( from, tag ); - if ( max == 0 ) - continue; + // randomize the order in which we balance the tags + // this is so that one bad tag doesn't prevent others from getting balanced + vector<string> tags; + { + set<string> t = distribution.tags(); + for (set<string>::const_iterator i = t.begin(); i != t.end(); ++i) + tags.push_back(*i); + tags.push_back(""); - string to = distribution.getBestReceieverShard( tag ); - if ( to.size() == 0 ) { - log() << "no available shards to take chunks for tag [" << tag << "]"; - return NULL; - } + std::random_shuffle(tags.begin(), tags.end()); + } - unsigned min = distribution.numberOfChunksInShardWithTag( to, tag ); + for (unsigned i = 0; i < tags.size(); i++) { + string tag = tags[i]; - const int imbalance = max - min; + const ShardId from = distribution.getMostOverloadedShard(tag); + if (from.size() == 0) + continue; - LOG(1) << "collection : " << ns; - LOG(1) << "donor : " << from << " chunks on " << max; - LOG(1) << "receiver : " << to << " chunks on " << min; - LOG(1) << "threshold : " << threshold; + unsigned max = distribution.numberOfChunksInShardWithTag(from, tag); + if (max == 0) + continue; - if ( imbalance < threshold ) - continue; + string to = distribution.getBestReceieverShard(tag); + if (to.size() == 0) { + log() << "no available shards to take chunks for tag [" << tag << "]"; + return NULL; + } - const vector<ChunkType>& chunks = distribution.getChunks(from); - unsigned numJumboChunks = 0; - for ( unsigned j = 0; j < chunks.size(); j++ ) { - const ChunkType& chunk = chunks[j]; - if (distribution.getTagForChunk(chunk) != tag) - continue; + unsigned min = distribution.numberOfChunksInShardWithTag(to, tag); - if (chunk.getJumbo()) { - numJumboChunks++; - continue; - } + const int imbalance = max - min; - log() << " ns: " << ns << " going to move " << chunk - << " from: " << from << " to: " << to << " tag [" << tag << "]" - ; - return new MigrateInfo(ns, to, from, chunk.toBSON()); - } + LOG(1) << "collection : " << ns; + LOG(1) << "donor : " << from << " chunks on " << max; + LOG(1) << "receiver : " << to << " chunks on " << min; + LOG(1) << "threshold : " << threshold; + + if (imbalance < threshold) + continue; + + const vector<ChunkType>& chunks = distribution.getChunks(from); + unsigned numJumboChunks = 0; + for (unsigned j = 0; j < chunks.size(); j++) { + const ChunkType& chunk = chunks[j]; + if (distribution.getTagForChunk(chunk) != tag) + continue; - if ( numJumboChunks ) { - error() << "shard: " << from << " ns: " << ns - << " has too many chunks, but they are all jumbo " - << " numJumboChunks: " << numJumboChunks - ; + if (chunk.getJumbo()) { + numJumboChunks++; continue; } - verify( false ); // should be impossible + log() << " ns: " << ns << " going to move " << chunk << " from: " << from + << " to: " << to << " tag [" << tag << "]"; + return new MigrateInfo(ns, to, from, chunk.toBSON()); } - // Everything is balanced here! - return NULL; - } - + if (numJumboChunks) { + error() << "shard: " << from << " ns: " << ns + << " has too many chunks, but they are all jumbo " + << " numJumboChunks: " << numJumboChunks; + continue; + } - ShardInfo::ShardInfo(long long maxSizeMB, - long long currSizeMB, - bool draining, - const set<string>& tags, - const string& mongoVersion): - _maxSizeMB(maxSizeMB), - _currSizeMB(currSizeMB), - _draining(draining), - _tags(tags), - _mongoVersion(mongoVersion) { + verify(false); // should be impossible } - ShardInfo::ShardInfo() - : _maxSizeMB(0), - _currSizeMB(0), - _draining(false) { - } + // Everything is balanced here! + return NULL; +} - void ShardInfo::addTag( const string& tag ) { - _tags.insert( tag ); - } +ShardInfo::ShardInfo(long long maxSizeMB, + long long currSizeMB, + bool draining, + const set<string>& tags, + const string& mongoVersion) + : _maxSizeMB(maxSizeMB), + _currSizeMB(currSizeMB), + _draining(draining), + _tags(tags), + _mongoVersion(mongoVersion) {} - bool ShardInfo::isSizeMaxed() const { - if (_maxSizeMB == 0 || _currSizeMB == 0) - return false; +ShardInfo::ShardInfo() : _maxSizeMB(0), _currSizeMB(0), _draining(false) {} - return _currSizeMB >= _maxSizeMB; - } +void ShardInfo::addTag(const string& tag) { + _tags.insert(tag); +} - bool ShardInfo::hasTag( const string& tag ) const { - if ( tag.size() == 0 ) - return true; - return _tags.count( tag ) > 0; - } - string ShardInfo::toString() const { - StringBuilder ss; - ss << " maxSizeMB: " << _maxSizeMB; - ss << " currSizeMB: " << _currSizeMB; - ss << " draining: " << _draining; - if ( _tags.size() > 0 ) { - ss << "tags : "; - for ( set<string>::const_iterator i = _tags.begin(); i != _tags.end(); ++i ) - ss << *i << ","; - } - ss << " version: " << _mongoVersion; - return ss.str(); - } +bool ShardInfo::isSizeMaxed() const { + if (_maxSizeMB == 0 || _currSizeMB == 0) + return false; + + return _currSizeMB >= _maxSizeMB; +} - string ChunkInfo::toString() const { - StringBuilder buf; - buf << " min: " << min; - buf << " max: " << max; - return buf.str(); +bool ShardInfo::hasTag(const string& tag) const { + if (tag.size() == 0) + return true; + return _tags.count(tag) > 0; +} + +string ShardInfo::toString() const { + StringBuilder ss; + ss << " maxSizeMB: " << _maxSizeMB; + ss << " currSizeMB: " << _currSizeMB; + ss << " draining: " << _draining; + if (_tags.size() > 0) { + ss << "tags : "; + for (set<string>::const_iterator i = _tags.begin(); i != _tags.end(); ++i) + ss << *i << ","; } + ss << " version: " << _mongoVersion; + return ss.str(); +} + +string ChunkInfo::toString() const { + StringBuilder buf; + buf << " min: " << min; + buf << " max: " << max; + return buf.str(); +} } // namespace mongo diff --git a/src/mongo/s/balancer_policy.h b/src/mongo/s/balancer_policy.h index b5203b37411..16614f424f1 100644 --- a/src/mongo/s/balancer_policy.h +++ b/src/mongo/s/balancer_policy.h @@ -37,198 +37,202 @@ namespace mongo { - class ChunkManager; - - struct ChunkInfo { - const BSONObj min; - const BSONObj max; - - ChunkInfo(const BSONObj& chunk) - : min(chunk[ChunkType::min()].Obj().getOwned()), - max(chunk[ChunkType::max()].Obj().getOwned()) { - - } - - std::string toString() const; - }; - - - struct TagRange { - BSONObj min; - BSONObj max; - std::string tag; - - TagRange(){} - - TagRange( const BSONObj& a_min, const BSONObj& a_max, const std::string& a_tag ) - : min( a_min.getOwned() ), max( a_max.getOwned() ), tag( a_tag ) {} - - std::string toString() const; - }; - - - class ShardInfo { - public: - ShardInfo(); - ShardInfo(long long maxSizeMB, - long long currSizeMB, - bool draining, - const std::set<std::string>& tags = std::set<std::string>(), - const std::string& _mongoVersion = std::string("")); - - void addTag( const std::string& tag ); - - /** @return true if we have the tag OR if the tag is "" */ - bool hasTag( const std::string& tag ) const; - - /** - * @return true if a shard cannot receive any new chunks because it reaches 'shardLimits'. - * Expects the optional fields "maxSize", can in size in MB, and "usedSize", currently used size - * in MB, on 'shardLimits'. - */ - bool isSizeMaxed() const; - - /** - * @return true if 'shardLimist' contains a field "draining". Expects the optional field - * "isDraining" on 'shrdLimits'. - */ - bool isDraining() const { return _draining; } - - long long getMaxSizeMB() const { return _maxSizeMB; } - - long long getCurrSizeMB() const { return _currSizeMB; } - - std::string getMongoVersion() const { return _mongoVersion; } - - std::string toString() const; - - private: - long long _maxSizeMB; - long long _currSizeMB; - bool _draining; - std::set<std::string> _tags; - std::string _mongoVersion; - }; - - - struct MigrateInfo { - MigrateInfo(const std::string& a_ns, - const ShardId& a_to, - const ShardId& a_from, - const BSONObj& a_chunk) - : ns(a_ns), - to(a_to), - from(a_from), - chunk(a_chunk) { - - } - - const std::string ns; - const ShardId to; - const ShardId from; - const ChunkInfo chunk; - }; - - typedef std::map<ShardId, ShardInfo> ShardInfoMap; - typedef std::map<ShardId, std::vector<ChunkType>> ShardToChunksMap; - - - class DistributionStatus { - MONGO_DISALLOW_COPYING(DistributionStatus); - public: - DistributionStatus(const ShardInfoMap& shardInfo, - const ShardToChunksMap& shardToChunksMap); - - // only used when building - - /** - * @return if range is valid - */ - bool addTagRange( const TagRange& range ); - - // ---- these methods might be better suiting in BalancerPolicy - - /** - * @param forTag "" if you don't care, or a tag - * @return shard best suited to receive a chunk - */ - std::string getBestReceieverShard( const std::string& forTag ) const; - - /** - * @return the shard with the most chunks - * based on # of chunks with the given tag - */ - std::string getMostOverloadedShard( const std::string& forTag ) const; - - - // ---- basic accessors, counters, etc... - - /** @return total number of chunks */ - unsigned totalChunks() const; - - /** @return number of chunks in this shard */ - unsigned numberOfChunksInShard(const ShardId& shardId) const; - - /** @return number of chunks in this shard with the given tag */ - unsigned numberOfChunksInShardWithTag(const ShardId& shardId, const std::string& tag ) const; - - /** @return chunks for the shard */ - const std::vector<ChunkType>& getChunks(const ShardId& shardId) const; - - /** @return all tags we know about, not include "" */ - const std::set<std::string>& tags() const { return _allTags; } - - /** @return the right tag for chunk, possibly "" */ - std::string getTagForChunk(const ChunkType& chunk) const; - - /** @return all shard ids we know about */ - const std::set<ShardId>& shardIds() const { return _shardIds; } - - /** @return the ShardInfo for the shard */ - const ShardInfo& shardInfo(const ShardId& shardId) const; - - /** writes all state to log() */ - void dump() const; - - /** - * Retrieves shard metadata information from the config server as well as some stats - * from the shards. - */ - static Status populateShardInfoMap(ShardInfoMap* shardInfo); - - /** - * Note: jumbo and versions are not set. - */ - static void populateShardToChunksMap(const ShardInfoMap& allShards, - const ChunkManager& chunkMgr, - ShardToChunksMap* shardToChunksMap); - - private: - const ShardInfoMap& _shardInfo; - const ShardToChunksMap& _shardChunks; - std::map<BSONObj,TagRange> _tagRanges; - std::set<std::string> _allTags; - std::set<ShardId> _shardIds; - }; - - - class BalancerPolicy { - public: - - /** - * Returns a suggested chunk to move whithin a collection's shards, given information about - * space usage and number of chunks for that collection. If the policy doesn't recommend - * moving, it returns NULL. - * - * @param ns is the collections namepace. - * @param DistributionStatus holds all the info about the current state of the cluster/namespace - * @param balancedLastTime is the number of chunks effectively moved in the last round. - * @returns NULL or MigrateInfo of the best move to make towards balacing the collection. - * caller owns the MigrateInfo instance - */ - static MigrateInfo* balance( const std::string& ns, - const DistributionStatus& distribution, - int balancedLastTime ); - }; +class ChunkManager; + +struct ChunkInfo { + const BSONObj min; + const BSONObj max; + + ChunkInfo(const BSONObj& chunk) + : min(chunk[ChunkType::min()].Obj().getOwned()), + max(chunk[ChunkType::max()].Obj().getOwned()) {} + + std::string toString() const; +}; + + +struct TagRange { + BSONObj min; + BSONObj max; + std::string tag; + + TagRange() {} + + TagRange(const BSONObj& a_min, const BSONObj& a_max, const std::string& a_tag) + : min(a_min.getOwned()), max(a_max.getOwned()), tag(a_tag) {} + + std::string toString() const; +}; + + +class ShardInfo { +public: + ShardInfo(); + ShardInfo(long long maxSizeMB, + long long currSizeMB, + bool draining, + const std::set<std::string>& tags = std::set<std::string>(), + const std::string& _mongoVersion = std::string("")); + + void addTag(const std::string& tag); + + /** @return true if we have the tag OR if the tag is "" */ + bool hasTag(const std::string& tag) const; + + /** + * @return true if a shard cannot receive any new chunks because it reaches 'shardLimits'. + * Expects the optional fields "maxSize", can in size in MB, and "usedSize", currently used size + * in MB, on 'shardLimits'. + */ + bool isSizeMaxed() const; + + /** + * @return true if 'shardLimist' contains a field "draining". Expects the optional field + * "isDraining" on 'shrdLimits'. + */ + bool isDraining() const { + return _draining; + } + + long long getMaxSizeMB() const { + return _maxSizeMB; + } + + long long getCurrSizeMB() const { + return _currSizeMB; + } + + std::string getMongoVersion() const { + return _mongoVersion; + } + + std::string toString() const; + +private: + long long _maxSizeMB; + long long _currSizeMB; + bool _draining; + std::set<std::string> _tags; + std::string _mongoVersion; +}; + + +struct MigrateInfo { + MigrateInfo(const std::string& a_ns, + const ShardId& a_to, + const ShardId& a_from, + const BSONObj& a_chunk) + : ns(a_ns), to(a_to), from(a_from), chunk(a_chunk) {} + + const std::string ns; + const ShardId to; + const ShardId from; + const ChunkInfo chunk; +}; + +typedef std::map<ShardId, ShardInfo> ShardInfoMap; +typedef std::map<ShardId, std::vector<ChunkType>> ShardToChunksMap; + + +class DistributionStatus { + MONGO_DISALLOW_COPYING(DistributionStatus); + +public: + DistributionStatus(const ShardInfoMap& shardInfo, const ShardToChunksMap& shardToChunksMap); + + // only used when building + + /** + * @return if range is valid + */ + bool addTagRange(const TagRange& range); + + // ---- these methods might be better suiting in BalancerPolicy + + /** + * @param forTag "" if you don't care, or a tag + * @return shard best suited to receive a chunk + */ + std::string getBestReceieverShard(const std::string& forTag) const; + + /** + * @return the shard with the most chunks + * based on # of chunks with the given tag + */ + std::string getMostOverloadedShard(const std::string& forTag) const; + + + // ---- basic accessors, counters, etc... + + /** @return total number of chunks */ + unsigned totalChunks() const; + + /** @return number of chunks in this shard */ + unsigned numberOfChunksInShard(const ShardId& shardId) const; + + /** @return number of chunks in this shard with the given tag */ + unsigned numberOfChunksInShardWithTag(const ShardId& shardId, const std::string& tag) const; + + /** @return chunks for the shard */ + const std::vector<ChunkType>& getChunks(const ShardId& shardId) const; + + /** @return all tags we know about, not include "" */ + const std::set<std::string>& tags() const { + return _allTags; + } + + /** @return the right tag for chunk, possibly "" */ + std::string getTagForChunk(const ChunkType& chunk) const; + + /** @return all shard ids we know about */ + const std::set<ShardId>& shardIds() const { + return _shardIds; + } + + /** @return the ShardInfo for the shard */ + const ShardInfo& shardInfo(const ShardId& shardId) const; + + /** writes all state to log() */ + void dump() const; + + /** + * Retrieves shard metadata information from the config server as well as some stats + * from the shards. + */ + static Status populateShardInfoMap(ShardInfoMap* shardInfo); + + /** + * Note: jumbo and versions are not set. + */ + static void populateShardToChunksMap(const ShardInfoMap& allShards, + const ChunkManager& chunkMgr, + ShardToChunksMap* shardToChunksMap); + +private: + const ShardInfoMap& _shardInfo; + const ShardToChunksMap& _shardChunks; + std::map<BSONObj, TagRange> _tagRanges; + std::set<std::string> _allTags; + std::set<ShardId> _shardIds; +}; + + +class BalancerPolicy { +public: + /** + * Returns a suggested chunk to move whithin a collection's shards, given information about + * space usage and number of chunks for that collection. If the policy doesn't recommend + * moving, it returns NULL. + * + * @param ns is the collections namepace. + * @param DistributionStatus holds all the info about the current state of the cluster/namespace + * @param balancedLastTime is the number of chunks effectively moved in the last round. + * @returns NULL or MigrateInfo of the best move to make towards balacing the collection. + * caller owns the MigrateInfo instance + */ + static MigrateInfo* balance(const std::string& ns, + const DistributionStatus& distribution, + int balancedLastTime); +}; } // namespace mongo diff --git a/src/mongo/s/balancer_policy_tests.cpp b/src/mongo/s/balancer_policy_tests.cpp index 03ee8a8f6e0..b200576ee7b 100644 --- a/src/mongo/s/balancer_policy_tests.cpp +++ b/src/mongo/s/balancer_policy_tests.cpp @@ -37,594 +37,587 @@ namespace { - using namespace mongo; - - using std::map; - using std::string; - using std::stringstream; - using std::vector; - - - TEST(BalancerPolicyTests, SizeMaxedShardTest) { - ASSERT(!ShardInfo(0, 0, false).isSizeMaxed()); - ASSERT(!ShardInfo(100LL, 80LL, false).isSizeMaxed()); - ASSERT(ShardInfo(100LL, 110LL, false).isSizeMaxed()); - } +using namespace mongo; - TEST(BalancerPolicyTests, BalanceNormalTest) { - ShardToChunksMap chunkMap; - vector<ChunkType> chunks; +using std::map; +using std::string; +using std::stringstream; +using std::vector; - { - ChunkType chunk; - chunk.setMin(BSON("x" << BSON("$minKey" << 1))); - chunk.setMax(BSON("x" << 49)); - chunks.push_back(chunk); - } - { - ChunkType chunk; - chunk.setMin(BSON("x" << 49)); - chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); - chunks.push_back(chunk); - } +TEST(BalancerPolicyTests, SizeMaxedShardTest) { + ASSERT(!ShardInfo(0, 0, false).isSizeMaxed()); + ASSERT(!ShardInfo(100LL, 80LL, false).isSizeMaxed()); + ASSERT(ShardInfo(100LL, 110LL, false).isSizeMaxed()); +} - chunkMap["shard0"] = chunks; - chunkMap["shard1"] = vector<ChunkType>(); +TEST(BalancerPolicyTests, BalanceNormalTest) { + ShardToChunksMap chunkMap; + vector<ChunkType> chunks; - // no limits - ShardInfoMap info; - info["shard0"] = ShardInfo(0, 2, false); - info["shard1"] = ShardInfo(0, 0, false); - - DistributionStatus status(info, chunkMap); - std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 1)); + { + ChunkType chunk; + chunk.setMin(BSON("x" << BSON("$minKey" << 1))); + chunk.setMax(BSON("x" << 49)); + chunks.push_back(chunk); + } - ASSERT(c); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 49)); + chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); + chunks.push_back(chunk); } + chunkMap["shard0"] = chunks; + chunkMap["shard1"] = vector<ChunkType>(); - TEST(BalancerPolicyTests, BalanceJumbo) { - ShardToChunksMap chunkMap; - vector<ChunkType> chunks; + // no limits + ShardInfoMap info; + info["shard0"] = ShardInfo(0, 2, false); + info["shard1"] = ShardInfo(0, 0, false); - { - ChunkType chunk; - chunk.setMin(BSON("x" << BSON("$minKey" << 1))); - chunk.setMax(BSON("x" << 10)); - chunk.setJumbo(true); - chunks.push_back(chunk); - } + DistributionStatus status(info, chunkMap); + std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 1)); - { - ChunkType chunk; - chunk.setMin(BSON("x" << 10)); - chunk.setMax(BSON("x" << 20)); - chunk.setJumbo(true); - chunks.push_back(chunk); - } + ASSERT(c); +} - { - ChunkType chunk; - chunk.setMin(BSON("x" << 20)); - chunk.setMax(BSON("x" << 30)); - chunks.push_back(chunk); - } - { - ChunkType chunk; - chunk.setMin(BSON("x" << 30)); - chunk.setMax(BSON("x" << 40)); - chunk.setJumbo(true); - chunks.push_back(chunk); - } +TEST(BalancerPolicyTests, BalanceJumbo) { + ShardToChunksMap chunkMap; + vector<ChunkType> chunks; - { - ChunkType chunk; - chunk.setMin(BSON("x" << 40)); - chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); - chunks.push_back(chunk); - } + { + ChunkType chunk; + chunk.setMin(BSON("x" << BSON("$minKey" << 1))); + chunk.setMax(BSON("x" << 10)); + chunk.setJumbo(true); + chunks.push_back(chunk); + } - chunkMap["shard0"] = chunks; - chunkMap["shard1"] = vector<ChunkType>(); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 10)); + chunk.setMax(BSON("x" << 20)); + chunk.setJumbo(true); + chunks.push_back(chunk); + } - // no limits - ShardInfoMap info; - info["shard0"] = ShardInfo(0, 2, false); - info["shard1"] = ShardInfo(0, 0, false); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 20)); + chunk.setMax(BSON("x" << 30)); + chunks.push_back(chunk); + } - DistributionStatus status(info, chunkMap); - std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 1)); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 30)); + chunk.setMax(BSON("x" << 40)); + chunk.setJumbo(true); + chunks.push_back(chunk); + } - ASSERT(c); - ASSERT_EQUALS(30, c->chunk.max["x"].numberInt()); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 40)); + chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); + chunks.push_back(chunk); } - TEST(BalanceNormalTests, BalanceDrainingTest) { - ShardToChunksMap chunkMap; - vector<ChunkType> chunks; + chunkMap["shard0"] = chunks; + chunkMap["shard1"] = vector<ChunkType>(); - { - ChunkType chunk; - chunk.setMin(BSON("x" << BSON("$minKey" << 1))); - chunk.setMax(BSON("x" << 49)); - chunks.push_back(chunk); - } + // no limits + ShardInfoMap info; + info["shard0"] = ShardInfo(0, 2, false); + info["shard1"] = ShardInfo(0, 0, false); - { - ChunkType chunk; - chunk.setMin(BSON("x" << 49)); - chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); - chunks.push_back(chunk); - } + DistributionStatus status(info, chunkMap); + std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 1)); - chunkMap["shard0"] = chunks; - chunkMap["shard1"] = vector<ChunkType>(); + ASSERT(c); + ASSERT_EQUALS(30, c->chunk.max["x"].numberInt()); +} - // shard0 is draining - ShardInfoMap limitsMap; - limitsMap["shard0"] = ShardInfo(0LL, 2LL, true); - limitsMap["shard1"] = ShardInfo(0LL, 0LL, false); +TEST(BalanceNormalTests, BalanceDrainingTest) { + ShardToChunksMap chunkMap; + vector<ChunkType> chunks; - DistributionStatus status(limitsMap, chunkMap); - std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 0)); + { + ChunkType chunk; + chunk.setMin(BSON("x" << BSON("$minKey" << 1))); + chunk.setMax(BSON("x" << 49)); + chunks.push_back(chunk); + } - ASSERT(c); - ASSERT_EQUALS(c->to, "shard1"); - ASSERT_EQUALS(c->from, "shard0"); - ASSERT(!c->chunk.min.isEmpty()); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 49)); + chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); + chunks.push_back(chunk); } - TEST(BalancerPolicyTests, BalanceEndedDrainingTest) { - ShardToChunksMap chunkMap; - vector<ChunkType> chunks; + chunkMap["shard0"] = chunks; + chunkMap["shard1"] = vector<ChunkType>(); - { - ChunkType chunk; - chunk.setMin(BSON("x" << BSON("$minKey" << 1))); - chunk.setMax(BSON("x" << 49)); - chunks.push_back(chunk); - } + // shard0 is draining + ShardInfoMap limitsMap; + limitsMap["shard0"] = ShardInfo(0LL, 2LL, true); + limitsMap["shard1"] = ShardInfo(0LL, 0LL, false); - { - ChunkType chunk; - chunk.setMin(BSON("x" << 49)); - chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); - chunks.push_back(chunk); - } + DistributionStatus status(limitsMap, chunkMap); + std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 0)); - chunkMap["shard0"] = chunks; - chunkMap["shard1"] = vector<ChunkType>(); + ASSERT(c); + ASSERT_EQUALS(c->to, "shard1"); + ASSERT_EQUALS(c->from, "shard0"); + ASSERT(!c->chunk.min.isEmpty()); +} - // no limits - ShardInfoMap limitsMap; - limitsMap["shard0"] = ShardInfo(0, 2, false); - limitsMap["shard1"] = ShardInfo(0, 0, true); +TEST(BalancerPolicyTests, BalanceEndedDrainingTest) { + ShardToChunksMap chunkMap; + vector<ChunkType> chunks; - DistributionStatus status(limitsMap, chunkMap); - std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 0)); + { + ChunkType chunk; + chunk.setMin(BSON("x" << BSON("$minKey" << 1))); + chunk.setMax(BSON("x" << 49)); + chunks.push_back(chunk); + } - ASSERT(!c); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 49)); + chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); + chunks.push_back(chunk); } - TEST(BalancerPolicyTests, BalanceImpasseTest) { - ShardToChunksMap chunkMap; - vector<ChunkType> chunks; + chunkMap["shard0"] = chunks; + chunkMap["shard1"] = vector<ChunkType>(); - { - ChunkType chunk; - chunk.setMin(BSON("x" << BSON("$minKey" << 1))); - chunk.setMax(BSON("x" << 49)); - chunks.push_back(chunk); - } + // no limits + ShardInfoMap limitsMap; + limitsMap["shard0"] = ShardInfo(0, 2, false); + limitsMap["shard1"] = ShardInfo(0, 0, true); - { - ChunkType chunk; - chunk.setMin(BSON("x" << 49)); - chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); - chunks.push_back(chunk); - } + DistributionStatus status(limitsMap, chunkMap); + std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 0)); - chunkMap["shard0"] = vector<ChunkType>(); - chunkMap["shard1"] = chunks; - chunkMap["shard2"] = vector<ChunkType>(); + ASSERT(!c); +} - // shard0 is draining, shard1 is maxed out, shard2 has writebacks pending - ShardInfoMap limitsMap; - limitsMap["shard0"] = ShardInfo(0, 2, true); - limitsMap["shard1"] = ShardInfo(1, 1, false); - limitsMap["shard2"] = ShardInfo(0, 1, true); +TEST(BalancerPolicyTests, BalanceImpasseTest) { + ShardToChunksMap chunkMap; + vector<ChunkType> chunks; - DistributionStatus status(limitsMap, chunkMap); - std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 0)); + { + ChunkType chunk; + chunk.setMin(BSON("x" << BSON("$minKey" << 1))); + chunk.setMax(BSON("x" << 49)); + chunks.push_back(chunk); + } - ASSERT(!c); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 49)); + chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); + chunks.push_back(chunk); } + chunkMap["shard0"] = vector<ChunkType>(); + chunkMap["shard1"] = chunks; + chunkMap["shard2"] = vector<ChunkType>(); - void addShard(ShardToChunksMap& shardToChunks, unsigned numChunks, bool last) { - unsigned total = 0; - for (const auto& chunk : shardToChunks) { - total += chunk.second.size(); - } + // shard0 is draining, shard1 is maxed out, shard2 has writebacks pending + ShardInfoMap limitsMap; + limitsMap["shard0"] = ShardInfo(0, 2, true); + limitsMap["shard1"] = ShardInfo(1, 1, false); + limitsMap["shard2"] = ShardInfo(0, 1, true); - stringstream ss; - ss << "shard" << shardToChunks.size(); - string myName = ss.str(); + DistributionStatus status(limitsMap, chunkMap); + std::unique_ptr<MigrateInfo> c(BalancerPolicy::balance("ns", status, 0)); - vector<ChunkType> chunksList; + ASSERT(!c); +} - for (unsigned i = 0; i < numChunks; i++) { - ChunkType chunk; - if (i == 0 && total == 0) { - chunk.setMin(BSON("x" << BSON("$minKey" << 1))); - } - else { - chunk.setMin(BSON("x" << total + i)); - } +void addShard(ShardToChunksMap& shardToChunks, unsigned numChunks, bool last) { + unsigned total = 0; + for (const auto& chunk : shardToChunks) { + total += chunk.second.size(); + } - if (last && i == (numChunks - 1)) { - chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); - } - else { - chunk.setMax(BSON("x" << 1 + total + i)); - } + stringstream ss; + ss << "shard" << shardToChunks.size(); + string myName = ss.str(); + + vector<ChunkType> chunksList; - chunksList.push_back(chunk); + for (unsigned i = 0; i < numChunks; i++) { + ChunkType chunk; + + if (i == 0 && total == 0) { + chunk.setMin(BSON("x" << BSON("$minKey" << 1))); + } else { + chunk.setMin(BSON("x" << total + i)); } - shardToChunks[myName] = chunksList; + if (last && i == (numChunks - 1)) { + chunk.setMax(BSON("x" << BSON("$maxKey" << 1))); + } else { + chunk.setMax(BSON("x" << 1 + total + i)); + } + + chunksList.push_back(chunk); } - void moveChunk(ShardToChunksMap& shardToChunks, MigrateInfo* m) { - vector<ChunkType>& chunks = shardToChunks[m->from]; + shardToChunks[myName] = chunksList; +} - for (vector<ChunkType>::iterator i = chunks.begin(); i != chunks.end(); ++i) { - if (i->getMin() == m->chunk.min) { - shardToChunks[m->to].push_back(*i); - chunks.erase(i); - return; - } - } +void moveChunk(ShardToChunksMap& shardToChunks, MigrateInfo* m) { + vector<ChunkType>& chunks = shardToChunks[m->from]; - invariant(false); + for (vector<ChunkType>::iterator i = chunks.begin(); i != chunks.end(); ++i) { + if (i->getMin() == m->chunk.min) { + shardToChunks[m->to].push_back(*i); + chunks.erase(i); + return; + } } + invariant(false); +} - TEST(BalancerPolicyTests, MultipleDraining) { - ShardToChunksMap chunks; - addShard(chunks, 5, false); - addShard(chunks, 10, false); - addShard(chunks, 5, true); - ShardInfoMap shards; - shards["shard0"] = ShardInfo(0, 5, true); - shards["shard1"] = ShardInfo(0, 5, true); - shards["shard2"] = ShardInfo(0, 5, false); +TEST(BalancerPolicyTests, MultipleDraining) { + ShardToChunksMap chunks; + addShard(chunks, 5, false); + addShard(chunks, 10, false); + addShard(chunks, 5, true); - DistributionStatus d(shards, chunks); - std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); + ShardInfoMap shards; + shards["shard0"] = ShardInfo(0, 5, true); + shards["shard1"] = ShardInfo(0, 5, true); + shards["shard2"] = ShardInfo(0, 5, false); - ASSERT(m); - ASSERT_EQUALS("shard2", m->to); - } + DistributionStatus d(shards, chunks); + std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); + ASSERT(m); + ASSERT_EQUALS("shard2", m->to); +} - TEST(BalancerPolicyTests, TagsDraining) { - ShardToChunksMap chunks; - addShard(chunks, 5, false); - addShard(chunks, 5, false); - addShard(chunks, 5, true); - ShardInfoMap shards; - shards["shard0"] = ShardInfo(0, 5, false); - shards["shard1"] = ShardInfo(0, 5, true); - shards["shard2"] = ShardInfo(0, 5, false); +TEST(BalancerPolicyTests, TagsDraining) { + ShardToChunksMap chunks; + addShard(chunks, 5, false); + addShard(chunks, 5, false); + addShard(chunks, 5, true); - shards["shard0"].addTag("a"); - shards["shard1"].addTag("a"); - shards["shard1"].addTag("b"); - shards["shard2"].addTag("b"); + ShardInfoMap shards; + shards["shard0"] = ShardInfo(0, 5, false); + shards["shard1"] = ShardInfo(0, 5, true); + shards["shard2"] = ShardInfo(0, 5, false); - while (true) { - DistributionStatus d(shards, chunks); - d.addTagRange(TagRange(BSON("x" << -1), BSON("x" << 7), "a")); - d.addTagRange(TagRange(BSON("x" << 7), BSON("x" << 1000), "b")); + shards["shard0"].addTag("a"); + shards["shard1"].addTag("a"); + shards["shard1"].addTag("b"); + shards["shard2"].addTag("b"); - std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); - if (!m) { - break; - } + while (true) { + DistributionStatus d(shards, chunks); + d.addTagRange(TagRange(BSON("x" << -1), BSON("x" << 7), "a")); + d.addTagRange(TagRange(BSON("x" << 7), BSON("x" << 1000), "b")); - if (m->chunk.min["x"].numberInt() < 7) { - ASSERT_EQUALS("shard0", m->to); - } - else { - ASSERT_EQUALS("shard2", m->to); - } + std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); + if (!m) { + break; + } - moveChunk(chunks, m.get()); + if (m->chunk.min["x"].numberInt() < 7) { + ASSERT_EQUALS("shard0", m->to); + } else { + ASSERT_EQUALS("shard2", m->to); } - ASSERT_EQUALS(7U, chunks["shard0"].size()); - ASSERT_EQUALS(0U, chunks["shard1"].size()); - ASSERT_EQUALS(8U, chunks["shard2"].size()); + moveChunk(chunks, m.get()); } + ASSERT_EQUALS(7U, chunks["shard0"].size()); + ASSERT_EQUALS(0U, chunks["shard1"].size()); + ASSERT_EQUALS(8U, chunks["shard2"].size()); +} - TEST(BalancerPolicyTests, TagsPolicyChange) { - ShardToChunksMap chunks; - addShard(chunks, 5, false); - addShard(chunks, 5, false); - addShard(chunks, 5, true); - ShardInfoMap shards; - shards["shard0"] = ShardInfo(0, 5, false); - shards["shard1"] = ShardInfo(0, 5, false); - shards["shard2"] = ShardInfo(0, 5, false); +TEST(BalancerPolicyTests, TagsPolicyChange) { + ShardToChunksMap chunks; + addShard(chunks, 5, false); + addShard(chunks, 5, false); + addShard(chunks, 5, true); - shards["shard0"].addTag("a"); - shards["shard1"].addTag("a"); + ShardInfoMap shards; + shards["shard0"] = ShardInfo(0, 5, false); + shards["shard1"] = ShardInfo(0, 5, false); + shards["shard2"] = ShardInfo(0, 5, false); - while (true) { - DistributionStatus d(shards, chunks); - d.addTagRange(TagRange(BSON("x" << -1), BSON("x" << 1000), "a")); + shards["shard0"].addTag("a"); + shards["shard1"].addTag("a"); - std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); - if (!m) { - break; - } + while (true) { + DistributionStatus d(shards, chunks); + d.addTagRange(TagRange(BSON("x" << -1), BSON("x" << 1000), "a")); - moveChunk(chunks, m.get()); + std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); + if (!m) { + break; } - const size_t shard0Size = chunks["shard0"].size(); - const size_t shard1Size = chunks["shard1"].size(); - - ASSERT_EQUALS(15U, shard0Size + shard1Size); - ASSERT(shard0Size == 7U || shard0Size == 8U); - ASSERT_EQUALS(0U, chunks["shard2"].size()); + moveChunk(chunks, m.get()); } + const size_t shard0Size = chunks["shard0"].size(); + const size_t shard1Size = chunks["shard1"].size(); - TEST(BalancerPolicyTests, TagsSelector) { - ShardToChunksMap chunks; - ShardInfoMap shards; - DistributionStatus d(shards, chunks); + ASSERT_EQUALS(15U, shard0Size + shard1Size); + ASSERT(shard0Size == 7U || shard0Size == 8U); + ASSERT_EQUALS(0U, chunks["shard2"].size()); +} - ASSERT(d.addTagRange(TagRange(BSON("x" << 1), BSON("x" << 10), "a"))); - ASSERT(d.addTagRange(TagRange(BSON("x" << 10), BSON("x" << 20), "b"))); - ASSERT(d.addTagRange(TagRange(BSON("x" << 20), BSON("x" << 30), "c"))); - ASSERT(!d.addTagRange(TagRange(BSON("x" << 20), BSON("x" << 30), "c"))); - ASSERT(!d.addTagRange(TagRange(BSON("x" << 22), BSON("x" << 28), "c"))); - ASSERT(!d.addTagRange(TagRange(BSON("x" << 28), BSON("x" << 33), "c"))); +TEST(BalancerPolicyTests, TagsSelector) { + ShardToChunksMap chunks; + ShardInfoMap shards; + DistributionStatus d(shards, chunks); - { - ChunkType chunk; - chunk.setMin(BSON("x" << -4)); - ASSERT_EQUALS("", d.getTagForChunk(chunk)); - } + ASSERT(d.addTagRange(TagRange(BSON("x" << 1), BSON("x" << 10), "a"))); + ASSERT(d.addTagRange(TagRange(BSON("x" << 10), BSON("x" << 20), "b"))); + ASSERT(d.addTagRange(TagRange(BSON("x" << 20), BSON("x" << 30), "c"))); - { - ChunkType chunk; - chunk.setMin(BSON("x" << 0)); - ASSERT_EQUALS("", d.getTagForChunk(chunk)); - } + ASSERT(!d.addTagRange(TagRange(BSON("x" << 20), BSON("x" << 30), "c"))); + ASSERT(!d.addTagRange(TagRange(BSON("x" << 22), BSON("x" << 28), "c"))); + ASSERT(!d.addTagRange(TagRange(BSON("x" << 28), BSON("x" << 33), "c"))); - { - ChunkType chunk; - chunk.setMin(BSON("x" << 1)); - ASSERT_EQUALS("a", d.getTagForChunk(chunk)); - } + { + ChunkType chunk; + chunk.setMin(BSON("x" << -4)); + ASSERT_EQUALS("", d.getTagForChunk(chunk)); + } - { - ChunkType chunk; - chunk.setMin(BSON("x" << 10)); - ASSERT_EQUALS("b", d.getTagForChunk(chunk)); - } + { + ChunkType chunk; + chunk.setMin(BSON("x" << 0)); + ASSERT_EQUALS("", d.getTagForChunk(chunk)); + } - { - ChunkType chunk; - chunk.setMin(BSON("x" << 15)); - ASSERT_EQUALS("b", d.getTagForChunk(chunk)); - } + { + ChunkType chunk; + chunk.setMin(BSON("x" << 1)); + ASSERT_EQUALS("a", d.getTagForChunk(chunk)); + } - { - ChunkType chunk; - chunk.setMin(BSON("x" << 25)); - ASSERT_EQUALS("c", d.getTagForChunk(chunk)); - } + { + ChunkType chunk; + chunk.setMin(BSON("x" << 10)); + ASSERT_EQUALS("b", d.getTagForChunk(chunk)); + } - { - ChunkType chunk; - chunk.setMin(BSON("x" << 35)); - ASSERT_EQUALS("", d.getTagForChunk(chunk)); - } + { + ChunkType chunk; + chunk.setMin(BSON("x" << 15)); + ASSERT_EQUALS("b", d.getTagForChunk(chunk)); } - /** - * Idea for this test is to set up three shards, one of which is overloaded (too much data). - * - * Even though the overloaded shard has less chunks, we shouldn't move chunks to that shard. - */ - TEST(BalancerPolicyTests, MaxSizeRespect) { - ShardToChunksMap chunks; - addShard(chunks, 3, false); - addShard(chunks, 4, false); - addShard(chunks, 6, true); + { + ChunkType chunk; + chunk.setMin(BSON("x" << 25)); + ASSERT_EQUALS("c", d.getTagForChunk(chunk)); + } - // Note that maxSize of shard0 is 1, and it is therefore overloaded with currSize = 3. - // Other shards have maxSize = 0 = unset. + { + ChunkType chunk; + chunk.setMin(BSON("x" << 35)); + ASSERT_EQUALS("", d.getTagForChunk(chunk)); + } +} - ShardInfoMap shards; - shards["shard0"] = ShardInfo(1, 3, false); - shards["shard1"] = ShardInfo(0, 4, false); - shards["shard2"] = ShardInfo(0, 6, false); +/** + * Idea for this test is to set up three shards, one of which is overloaded (too much data). + * + * Even though the overloaded shard has less chunks, we shouldn't move chunks to that shard. + */ +TEST(BalancerPolicyTests, MaxSizeRespect) { + ShardToChunksMap chunks; + addShard(chunks, 3, false); + addShard(chunks, 4, false); + addShard(chunks, 6, true); - DistributionStatus d(shards, chunks); - std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); + // Note that maxSize of shard0 is 1, and it is therefore overloaded with currSize = 3. + // Other shards have maxSize = 0 = unset. - ASSERT(m); - ASSERT_EQUALS("shard2", m->from); - ASSERT_EQUALS("shard1", m->to); - } + ShardInfoMap shards; + shards["shard0"] = ShardInfo(1, 3, false); + shards["shard1"] = ShardInfo(0, 4, false); + shards["shard2"] = ShardInfo(0, 6, false); - /** - * Here we check that being over the maxSize is *not* equivalent to draining, we don't want - * to empty shards for no other reason than they are over this limit. - */ - TEST(BalancerPolicyTests, MaxSizeNoDrain) { - ShardToChunksMap chunks; + DistributionStatus d(shards, chunks); + std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); - // Shard0 will be overloaded - addShard(chunks, 4, false); - addShard(chunks, 4, false); - addShard(chunks, 4, true); + ASSERT(m); + ASSERT_EQUALS("shard2", m->from); + ASSERT_EQUALS("shard1", m->to); +} - // Note that maxSize of shard0 is 1, and it is therefore overloaded with currSize = 4. - // Other shards have maxSize = 0 = unset. +/** + * Here we check that being over the maxSize is *not* equivalent to draining, we don't want + * to empty shards for no other reason than they are over this limit. + */ +TEST(BalancerPolicyTests, MaxSizeNoDrain) { + ShardToChunksMap chunks; - ShardInfoMap shards; - // ShardInfo(maxSize, currSize, draining, opsQueued) - shards["shard0"] = ShardInfo(1, 4, false); - shards["shard1"] = ShardInfo(0, 4, false); - shards["shard2"] = ShardInfo(0, 4, false); + // Shard0 will be overloaded + addShard(chunks, 4, false); + addShard(chunks, 4, false); + addShard(chunks, 4, true); - DistributionStatus d(shards, chunks); - std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); + // Note that maxSize of shard0 is 1, and it is therefore overloaded with currSize = 4. + // Other shards have maxSize = 0 = unset. - ASSERT(!m); - } + ShardInfoMap shards; + // ShardInfo(maxSize, currSize, draining, opsQueued) + shards["shard0"] = ShardInfo(1, 4, false); + shards["shard1"] = ShardInfo(0, 4, false); + shards["shard2"] = ShardInfo(0, 4, false); - /** - * Idea behind this test is that we set up several shards, the first two of which are - * draining and the second two of which have a data size limit. We also simulate a random - * number of chunks on each shard. - * - * Once the shards are setup, we virtually migrate numChunks times, or until there are no - * more migrations to run. Each chunk is assumed to have a size of 1 unit, and we increment - * our currSize for each shard as the chunks move. - * - * Finally, we ensure that the drained shards are drained, the data-limited shards aren't - * overloaded, and that all shards (including the data limited shard if the baseline isn't - * over the limit are balanced to within 1 unit of some baseline. - * - */ - TEST(BalancerPolicyTests, Simulation) { - // Hardcode seed here, make test deterministic. - int64_t seed = 1337; - PseudoRandom rng(seed); - - // Run test 10 times - for (int test = 0; test < 10; test++) { - // Setup our shards as draining, with maxSize, and normal - int numShards = 7; - int numChunks = 0; - - ShardToChunksMap chunks; - ShardInfoMap shards; - - map<string, int> expected; - - for (int i = 0; i < numShards; i++) { - int numShardChunks = rng.nextInt32(100); - bool draining = i < 2; - bool maxed = i >= 2 && i < 4; - - if (draining) { - expected[str::stream() << "shard" << i] = 0; - } + DistributionStatus d(shards, chunks); + std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, 0)); - if (maxed) { - expected[str::stream() << "shard" << i] = numShardChunks + 1; - } + ASSERT(!m); +} - addShard(chunks, numShardChunks, false); - numChunks += numShardChunks; +/** + * Idea behind this test is that we set up several shards, the first two of which are + * draining and the second two of which have a data size limit. We also simulate a random + * number of chunks on each shard. + * + * Once the shards are setup, we virtually migrate numChunks times, or until there are no + * more migrations to run. Each chunk is assumed to have a size of 1 unit, and we increment + * our currSize for each shard as the chunks move. + * + * Finally, we ensure that the drained shards are drained, the data-limited shards aren't + * overloaded, and that all shards (including the data limited shard if the baseline isn't + * over the limit are balanced to within 1 unit of some baseline. + * + */ +TEST(BalancerPolicyTests, Simulation) { + // Hardcode seed here, make test deterministic. + int64_t seed = 1337; + PseudoRandom rng(seed); - shards[str::stream() << "shard" << i] = - ShardInfo(maxed ? numShardChunks + 1 : 0, numShardChunks, draining); + // Run test 10 times + for (int test = 0; test < 10; test++) { + // Setup our shards as draining, with maxSize, and normal + int numShards = 7; + int numChunks = 0; + + ShardToChunksMap chunks; + ShardInfoMap shards; + + map<string, int> expected; + + for (int i = 0; i < numShards; i++) { + int numShardChunks = rng.nextInt32(100); + bool draining = i < 2; + bool maxed = i >= 2 && i < 4; + + if (draining) { + expected[str::stream() << "shard" << i] = 0; } - for (ShardInfoMap::iterator it = shards.begin(); it != shards.end(); ++it) { - log() << it->first << " : " << it->second.toString(); + if (maxed) { + expected[str::stream() << "shard" << i] = numShardChunks + 1; } - // Perform migrations and increment data size as chunks move - for (int i = 0; i < numChunks; i++) { - DistributionStatus d(shards, chunks); - std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, i != 0)); - if (!m) { - log() << "Finished with test moves."; - break; - } + addShard(chunks, numShardChunks, false); + numChunks += numShardChunks; - moveChunk(chunks, m.get()); + shards[str::stream() << "shard" << i] = + ShardInfo(maxed ? numShardChunks + 1 : 0, numShardChunks, draining); + } - { - ShardInfo& info = shards[m->from]; - shards[m->from] = ShardInfo(info.getMaxSizeMB(), - info.getCurrSizeMB() - 1, - info.isDraining()); - } + for (ShardInfoMap::iterator it = shards.begin(); it != shards.end(); ++it) { + log() << it->first << " : " << it->second.toString(); + } - { - ShardInfo& info = shards[m->to]; - shards[m->to] = ShardInfo(info.getMaxSizeMB(), - info.getCurrSizeMB() + 1, - info.isDraining()); - } + // Perform migrations and increment data size as chunks move + for (int i = 0; i < numChunks; i++) { + DistributionStatus d(shards, chunks); + std::unique_ptr<MigrateInfo> m(BalancerPolicy::balance("ns", d, i != 0)); + if (!m) { + log() << "Finished with test moves."; + break; } - // Make sure our balance is correct and our data size is low. + moveChunk(chunks, m.get()); - // The balanced value is the count on the last shard, since it's not draining or - // limited. - int balancedSize = (--shards.end())->second.getCurrSizeMB(); + { + ShardInfo& info = shards[m->from]; + shards[m->from] = + ShardInfo(info.getMaxSizeMB(), info.getCurrSizeMB() - 1, info.isDraining()); + } - for (ShardInfoMap::iterator it = shards.begin(); it != shards.end(); ++it) { - log() << it->first << " : " << it->second.toString(); + { + ShardInfo& info = shards[m->to]; + shards[m->to] = + ShardInfo(info.getMaxSizeMB(), info.getCurrSizeMB() + 1, info.isDraining()); } + } - for (ShardInfoMap::iterator it = shards.begin(); it != shards.end(); ++it) { - log() << it->first << " : " << it->second.toString(); + // Make sure our balance is correct and our data size is low. - map<string, int>::iterator expectedIt = expected.find(it->first); + // The balanced value is the count on the last shard, since it's not draining or + // limited. + int balancedSize = (--shards.end())->second.getCurrSizeMB(); - if (expectedIt == expected.end()) { - bool isInRange = it->second.getCurrSizeMB() >= balancedSize - 1 - && it->second.getCurrSizeMB() <= balancedSize + 1; - if (!isInRange) { - warning() << "non-limited and non-draining shard had " - << it->second.getCurrSizeMB() << " chunks, expected near " - << balancedSize; - } + for (ShardInfoMap::iterator it = shards.begin(); it != shards.end(); ++it) { + log() << it->first << " : " << it->second.toString(); + } + + for (ShardInfoMap::iterator it = shards.begin(); it != shards.end(); ++it) { + log() << it->first << " : " << it->second.toString(); + + map<string, int>::iterator expectedIt = expected.find(it->first); + + if (expectedIt == expected.end()) { + bool isInRange = it->second.getCurrSizeMB() >= balancedSize - 1 && + it->second.getCurrSizeMB() <= balancedSize + 1; + if (!isInRange) { + warning() << "non-limited and non-draining shard had " + << it->second.getCurrSizeMB() << " chunks, expected near " + << balancedSize; + } + + ASSERT(isInRange); + } else { + int expectedSize = expectedIt->second; + bool isInRange = it->second.getCurrSizeMB() <= expectedSize; - ASSERT(isInRange); + if (isInRange && expectedSize >= balancedSize) { + isInRange = it->second.getCurrSizeMB() >= balancedSize - 1 && + it->second.getCurrSizeMB() <= balancedSize + 1; } - else { - int expectedSize = expectedIt->second; - bool isInRange = it->second.getCurrSizeMB() <= expectedSize; - - if (isInRange && expectedSize >= balancedSize) { - isInRange = it->second.getCurrSizeMB() >= balancedSize - 1 - && it->second.getCurrSizeMB() <= balancedSize + 1; - } - - if (!isInRange) { - warning() << "limited or draining shard had " - << it->second.getCurrSizeMB() << " chunks, expected less than " - << expectedSize << " and (if less than expected) near " - << balancedSize; - } - - ASSERT(isInRange); + + if (!isInRange) { + warning() << "limited or draining shard had " << it->second.getCurrSizeMB() + << " chunks, expected less than " << expectedSize + << " and (if less than expected) near " << balancedSize; } + + ASSERT(isInRange); } } } +} -} // namespace +} // namespace diff --git a/src/mongo/s/bson_serializable.h b/src/mongo/s/bson_serializable.h index e957701df1c..a1d0695ee5c 100644 --- a/src/mongo/s/bson_serializable.h +++ b/src/mongo/s/bson_serializable.h @@ -34,81 +34,74 @@ namespace mongo { +/** + * "Types" are the interface to a known data structure that will be serialized to and + * deserialized from BSON. + */ +class BSONSerializable { +public: + virtual ~BSONSerializable() {} + /** - * "Types" are the interface to a known data structure that will be serialized to and - * deserialized from BSON. + * Returns true if all the mandatory fields are present and have valid + * representations. Otherwise returns false and fills in the optional 'errMsg' string. */ - class BSONSerializable { - public: - - virtual ~BSONSerializable() {} - - /** - * Returns true if all the mandatory fields are present and have valid - * representations. Otherwise returns false and fills in the optional 'errMsg' string. - */ - virtual bool isValid( std::string* errMsg ) const = 0; - - /** Returns the BSON representation of the entry. */ - virtual BSONObj toBSON() const = 0; + virtual bool isValid(std::string* errMsg) const = 0; - /** - * Clears and populates the internal state using the 'source' BSON object if the - * latter contains valid values. Otherwise sets errMsg and returns false. - */ - virtual bool parseBSON( const BSONObj& source, std::string* errMsg ) = 0; - - /** Clears the internal state. */ - virtual void clear() = 0; - - /** Returns a std::string representation of the current internal state. */ - virtual std::string toString() const = 0; - }; + /** Returns the BSON representation of the entry. */ + virtual BSONObj toBSON() const = 0; /** - * Generic implementation which accepts and stores any BSON object - * - * Generally this should only be used for compatibility reasons - newer requests should be - * fully typed. + * Clears and populates the internal state using the 'source' BSON object if the + * latter contains valid values. Otherwise sets errMsg and returns false. */ - class RawBSONSerializable : public BSONSerializable { - public: + virtual bool parseBSON(const BSONObj& source, std::string* errMsg) = 0; + + /** Clears the internal state. */ + virtual void clear() = 0; - RawBSONSerializable() { - } + /** Returns a std::string representation of the current internal state. */ + virtual std::string toString() const = 0; +}; - explicit RawBSONSerializable(const BSONObj& raw) - : _raw(raw) { - } +/** + * Generic implementation which accepts and stores any BSON object + * + * Generally this should only be used for compatibility reasons - newer requests should be + * fully typed. + */ +class RawBSONSerializable : public BSONSerializable { +public: + RawBSONSerializable() {} - virtual ~RawBSONSerializable() { - } + explicit RawBSONSerializable(const BSONObj& raw) : _raw(raw) {} - virtual bool isValid(std::string* errMsg) const { - return true; - } + virtual ~RawBSONSerializable() {} - virtual BSONObj toBSON() const { - return _raw; - } + virtual bool isValid(std::string* errMsg) const { + return true; + } - virtual bool parseBSON(const BSONObj& source, std::string* errMsg) { - _raw = source.getOwned(); - return true; - } + virtual BSONObj toBSON() const { + return _raw; + } - virtual void clear() { - _raw = BSONObj(); - } + virtual bool parseBSON(const BSONObj& source, std::string* errMsg) { + _raw = source.getOwned(); + return true; + } - virtual std::string toString() const { - return toBSON().toString(); - } + virtual void clear() { + _raw = BSONObj(); + } - private: + virtual std::string toString() const { + return toBSON().toString(); + } - BSONObj _raw; - }; +private: + BSONObj _raw; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/catalog_cache.cpp b/src/mongo/s/catalog/catalog_cache.cpp index db27e5b7778..416a0a7368a 100644 --- a/src/mongo/s/catalog/catalog_cache.cpp +++ b/src/mongo/s/catalog/catalog_cache.cpp @@ -38,51 +38,49 @@ namespace mongo { - using std::shared_ptr; - using std::string; +using std::shared_ptr; +using std::string; - CatalogCache::CatalogCache(CatalogManager* catalogManager) - : _catalogManager(catalogManager) { +CatalogCache::CatalogCache(CatalogManager* catalogManager) : _catalogManager(catalogManager) { + invariant(_catalogManager); +} - invariant(_catalogManager); - } - - StatusWith<shared_ptr<DBConfig>> CatalogCache::getDatabase(const string& dbName) { - stdx::lock_guard<stdx::mutex> guard(_mutex); +StatusWith<shared_ptr<DBConfig>> CatalogCache::getDatabase(const string& dbName) { + stdx::lock_guard<stdx::mutex> guard(_mutex); - ShardedDatabasesMap::iterator it = _databases.find(dbName); - if (it != _databases.end()) { - return it->second; - } + ShardedDatabasesMap::iterator it = _databases.find(dbName); + if (it != _databases.end()) { + return it->second; + } - // Need to load from the store - StatusWith<DatabaseType> status = _catalogManager->getDatabase(dbName); - if (!status.isOK()) { - return status.getStatus(); - } + // Need to load from the store + StatusWith<DatabaseType> status = _catalogManager->getDatabase(dbName); + if (!status.isOK()) { + return status.getStatus(); + } - shared_ptr<DBConfig> db = std::make_shared<DBConfig>(dbName, status.getValue()); - db->load(); + shared_ptr<DBConfig> db = std::make_shared<DBConfig>(dbName, status.getValue()); + db->load(); - invariant(_databases.insert(std::make_pair(dbName, db)).second); + invariant(_databases.insert(std::make_pair(dbName, db)).second); - return db; - } + return db; +} - void CatalogCache::invalidate(const string& dbName) { - stdx::lock_guard<stdx::mutex> guard(_mutex); +void CatalogCache::invalidate(const string& dbName) { + stdx::lock_guard<stdx::mutex> guard(_mutex); - ShardedDatabasesMap::iterator it = _databases.find(dbName); - if (it != _databases.end()) { - _databases.erase(it); - } + ShardedDatabasesMap::iterator it = _databases.find(dbName); + if (it != _databases.end()) { + _databases.erase(it); } +} - void CatalogCache::invalidateAll() { - stdx::lock_guard<stdx::mutex> guard(_mutex); +void CatalogCache::invalidateAll() { + stdx::lock_guard<stdx::mutex> guard(_mutex); - _databases.clear(); - } + _databases.clear(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/catalog_cache.h b/src/mongo/s/catalog/catalog_cache.h index 9d7c18cad76..a138c793c64 100644 --- a/src/mongo/s/catalog/catalog_cache.h +++ b/src/mongo/s/catalog/catalog_cache.h @@ -36,53 +36,55 @@ namespace mongo { - class CatalogManager; - class DBConfig; - template<typename T> class StatusWith; +class CatalogManager; +class DBConfig; +template <typename T> +class StatusWith; +/** + * This is the root of the "read-only" hierarchy of cached catalog metadata. It is read only + * in the sense that it only reads from the persistent store, but never writes to it. Instead + * writes happen thorugh the CatalogManager and the cache hierarchy needs to be invalidated. + */ +class CatalogCache { + MONGO_DISALLOW_COPYING(CatalogCache); + +public: + explicit CatalogCache(CatalogManager* catalogManager); + /** - * This is the root of the "read-only" hierarchy of cached catalog metadata. It is read only - * in the sense that it only reads from the persistent store, but never writes to it. Instead - * writes happen thorugh the CatalogManager and the cache hierarchy needs to be invalidated. + * Retrieves the cached metadata for the specified database. The returned value is still + * owned by the cache and it should not be cached elsewhere, but instead only used as a + * local variable. The reason for this is so that if the cache gets invalidated, the caller + * does not miss getting the most up-to-date value. + * + * @param dbname The name of the database (must not contain dots, etc). + * @return The database if it exists, NULL otherwise. */ - class CatalogCache { - MONGO_DISALLOW_COPYING(CatalogCache); - public: - explicit CatalogCache(CatalogManager* catalogManager); - - /** - * Retrieves the cached metadata for the specified database. The returned value is still - * owned by the cache and it should not be cached elsewhere, but instead only used as a - * local variable. The reason for this is so that if the cache gets invalidated, the caller - * does not miss getting the most up-to-date value. - * - * @param dbname The name of the database (must not contain dots, etc). - * @return The database if it exists, NULL otherwise. - */ - StatusWith<std::shared_ptr<DBConfig>> getDatabase(const std::string& dbName); - - /** - * Removes the database information for the specified name from the cache, so that the - * next time getDatabase is called, it will be reloaded. - */ - void invalidate(const std::string& dbName); - - /** - * Purges all cached database information, which will cause the data to be reloaded again. - */ - void invalidateAll(); - - private: - typedef std::map<std::string, std::shared_ptr<DBConfig>> ShardedDatabasesMap; - - - // Reference to the catalog manager. Not owned. - CatalogManager* const _catalogManager; - - // Databases catalog map and mutex to protect it - stdx::mutex _mutex; - ShardedDatabasesMap _databases; - }; - -} // namespace mongo + StatusWith<std::shared_ptr<DBConfig>> getDatabase(const std::string& dbName); + + /** + * Removes the database information for the specified name from the cache, so that the + * next time getDatabase is called, it will be reloaded. + */ + void invalidate(const std::string& dbName); + + /** + * Purges all cached database information, which will cause the data to be reloaded again. + */ + void invalidateAll(); + +private: + typedef std::map<std::string, std::shared_ptr<DBConfig>> ShardedDatabasesMap; + + + // Reference to the catalog manager. Not owned. + CatalogManager* const _catalogManager; + + // Databases catalog map and mutex to protect it + stdx::mutex _mutex; + ShardedDatabasesMap _databases; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/catalog_manager.cpp b/src/mongo/s/catalog/catalog_manager.cpp index 02da9c79c58..46acce91140 100644 --- a/src/mongo/s/catalog/catalog_manager.cpp +++ b/src/mongo/s/catalog/catalog_manager.cpp @@ -44,114 +44,109 @@ namespace mongo { - using std::unique_ptr; - using std::string; +using std::unique_ptr; +using std::string; namespace { - Status getStatus(const BatchedCommandResponse& response) { - if (response.getOk() == 0) { - return Status(static_cast<ErrorCodes::Error>(response.getErrCode()), - response.getErrMessage()); - } - - if (response.isErrDetailsSet()) { - const WriteErrorDetail* errDetail = response.getErrDetails().front(); - - return Status(static_cast<ErrorCodes::Error>(errDetail->getErrCode()), - errDetail->getErrMessage()); - } - - if (response.isWriteConcernErrorSet()) { - const WCErrorDetail* errDetail = response.getWriteConcernError(); +Status getStatus(const BatchedCommandResponse& response) { + if (response.getOk() == 0) { + return Status(static_cast<ErrorCodes::Error>(response.getErrCode()), + response.getErrMessage()); + } - return Status(static_cast<ErrorCodes::Error>(errDetail->getErrCode()), - errDetail->getErrMessage()); - } + if (response.isErrDetailsSet()) { + const WriteErrorDetail* errDetail = response.getErrDetails().front(); - return Status::OK(); + return Status(static_cast<ErrorCodes::Error>(errDetail->getErrCode()), + errDetail->getErrMessage()); } -} // namespace + if (response.isWriteConcernErrorSet()) { + const WCErrorDetail* errDetail = response.getWriteConcernError(); - Status CatalogManager::insert(const string& ns, - const BSONObj& doc, - BatchedCommandResponse* response) { + return Status(static_cast<ErrorCodes::Error>(errDetail->getErrCode()), + errDetail->getErrMessage()); + } - unique_ptr<BatchedInsertRequest> insert(new BatchedInsertRequest()); - insert->addToDocuments(doc); + return Status::OK(); +} - BatchedCommandRequest request(insert.release()); - request.setNS(ns); - request.setWriteConcern(WriteConcernOptions::Majority); +} // namespace - BatchedCommandResponse dummyResponse; - if (response == NULL) { - response = &dummyResponse; - } +Status CatalogManager::insert(const string& ns, + const BSONObj& doc, + BatchedCommandResponse* response) { + unique_ptr<BatchedInsertRequest> insert(new BatchedInsertRequest()); + insert->addToDocuments(doc); - // Make sure to add ids to the request, since this is an insert operation - unique_ptr<BatchedCommandRequest> requestWithIds( - BatchedCommandRequest::cloneWithIds(request)); - const BatchedCommandRequest& requestToSend = - requestWithIds.get() ? *requestWithIds : request; + BatchedCommandRequest request(insert.release()); + request.setNS(ns); + request.setWriteConcern(WriteConcernOptions::Majority); - writeConfigServerDirect(requestToSend, response); - return getStatus(*response); + BatchedCommandResponse dummyResponse; + if (response == NULL) { + response = &dummyResponse; } - Status CatalogManager::update(const string& ns, - const BSONObj& query, - const BSONObj& update, - bool upsert, - bool multi, - BatchedCommandResponse* response) { - - unique_ptr<BatchedUpdateDocument> updateDoc(new BatchedUpdateDocument()); - updateDoc->setQuery(query); - updateDoc->setUpdateExpr(update); - updateDoc->setUpsert(upsert); - updateDoc->setMulti(multi); - - unique_ptr<BatchedUpdateRequest> updateRequest(new BatchedUpdateRequest()); - updateRequest->addToUpdates(updateDoc.release()); - updateRequest->setWriteConcern(WriteConcernOptions::Majority); - - BatchedCommandRequest request(updateRequest.release()); - request.setNS(ns); - - BatchedCommandResponse dummyResponse; - if (response == NULL) { - response = &dummyResponse; - } - - writeConfigServerDirect(request, response); - return getStatus(*response); + // Make sure to add ids to the request, since this is an insert operation + unique_ptr<BatchedCommandRequest> requestWithIds(BatchedCommandRequest::cloneWithIds(request)); + const BatchedCommandRequest& requestToSend = requestWithIds.get() ? *requestWithIds : request; + + writeConfigServerDirect(requestToSend, response); + return getStatus(*response); +} + +Status CatalogManager::update(const string& ns, + const BSONObj& query, + const BSONObj& update, + bool upsert, + bool multi, + BatchedCommandResponse* response) { + unique_ptr<BatchedUpdateDocument> updateDoc(new BatchedUpdateDocument()); + updateDoc->setQuery(query); + updateDoc->setUpdateExpr(update); + updateDoc->setUpsert(upsert); + updateDoc->setMulti(multi); + + unique_ptr<BatchedUpdateRequest> updateRequest(new BatchedUpdateRequest()); + updateRequest->addToUpdates(updateDoc.release()); + updateRequest->setWriteConcern(WriteConcernOptions::Majority); + + BatchedCommandRequest request(updateRequest.release()); + request.setNS(ns); + + BatchedCommandResponse dummyResponse; + if (response == NULL) { + response = &dummyResponse; } - Status CatalogManager::remove(const string& ns, - const BSONObj& query, - int limit, - BatchedCommandResponse* response) { - - unique_ptr<BatchedDeleteDocument> deleteDoc(new BatchedDeleteDocument); - deleteDoc->setQuery(query); - deleteDoc->setLimit(limit); + writeConfigServerDirect(request, response); + return getStatus(*response); +} - unique_ptr<BatchedDeleteRequest> deleteRequest(new BatchedDeleteRequest()); - deleteRequest->addToDeletes(deleteDoc.release()); - deleteRequest->setWriteConcern(WriteConcernOptions::Majority); +Status CatalogManager::remove(const string& ns, + const BSONObj& query, + int limit, + BatchedCommandResponse* response) { + unique_ptr<BatchedDeleteDocument> deleteDoc(new BatchedDeleteDocument); + deleteDoc->setQuery(query); + deleteDoc->setLimit(limit); - BatchedCommandRequest request(deleteRequest.release()); - request.setNS(ns); + unique_ptr<BatchedDeleteRequest> deleteRequest(new BatchedDeleteRequest()); + deleteRequest->addToDeletes(deleteDoc.release()); + deleteRequest->setWriteConcern(WriteConcernOptions::Majority); - BatchedCommandResponse dummyResponse; - if (response == NULL) { - response = &dummyResponse; - } + BatchedCommandRequest request(deleteRequest.release()); + request.setNS(ns); - writeConfigServerDirect(request, response); - return getStatus(*response); + BatchedCommandResponse dummyResponse; + if (response == NULL) { + response = &dummyResponse; } -} // namespace mongo + writeConfigServerDirect(request, response); + return getStatus(*response); +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/catalog_manager.h b/src/mongo/s/catalog/catalog_manager.h index ca3dd44dffa..bda4a1b79bd 100644 --- a/src/mongo/s/catalog/catalog_manager.h +++ b/src/mongo/s/catalog/catalog_manager.h @@ -37,354 +37,352 @@ namespace mongo { - class ActionLogType; - class BatchedCommandRequest; - class BatchedCommandResponse; - struct BSONArray; - class BSONObj; - class BSONObjBuilder; - class ChunkType; - class CollectionType; - class ConnectionString; - class DatabaseType; - class DistLockManager; - class OperationContext; - class Query; - class SettingsType; - class ShardKeyPattern; - class ShardType; - class Status; - template<typename T> class StatusWith; - class TagsType; +class ActionLogType; +class BatchedCommandRequest; +class BatchedCommandResponse; +struct BSONArray; +class BSONObj; +class BSONObjBuilder; +class ChunkType; +class CollectionType; +class ConnectionString; +class DatabaseType; +class DistLockManager; +class OperationContext; +class Query; +class SettingsType; +class ShardKeyPattern; +class ShardType; +class Status; +template <typename T> +class StatusWith; +class TagsType; + +/** + * Used to indicate to the caller of the removeShard method whether draining of chunks for + * a particular shard has started, is ongoing, or has been completed. + */ +enum ShardDrainingStatus { + STARTED, + ONGOING, + COMPLETED, +}; + +/** + * Abstracts reads and writes of the sharding catalog metadata. + * + * All implementations of this interface should go directly to the persistent backing store + * and should avoid doing any caching of their own. The caching is delegated to a parallel + * read-only view of the catalog, which is maintained by a higher level code. + */ +class CatalogManager { + MONGO_DISALLOW_COPYING(CatalogManager); + +public: + virtual ~CatalogManager() = default; + + /** + * Retrieves the connection string for the catalog manager's backing server. + */ + virtual ConnectionString connectionString() const = 0; + + /** + * Performs implementation-specific startup tasks including but not limited to doing any + * necessary config schema upgrade work. Must be run after the catalog manager + * has been installed into the global 'grid' object. + */ + virtual Status startup(bool upgrade) = 0; + + /** + * Performs necessary cleanup when shutting down cleanly. + */ + virtual void shutDown() = 0; + + /** + * Creates a new database or updates the sharding status for an existing one. Cannot be + * used for the admin/config/local DBs, which should not be created or sharded manually + * anyways. + * + * Returns Status::OK on success or any error code indicating the failure. These are some + * of the known failures: + * - DatabaseDifferCase - database already exists, but with a different case + * - ShardNotFound - could not find a shard to place the DB on + */ + virtual Status enableSharding(const std::string& dbName) = 0; + + /** + * Shards a collection. Assumes that the database is enabled for sharding. + * + * @param ns: namespace of collection to shard + * @param fieldsAndOrder: shardKey pattern + * @param unique: if true, ensure underlying index enforces a unique constraint. + * @param initPoints: create chunks based on a set of specified split points. + * @param initShardIds: if nullptr, use primary shard as lone shard for DB. + * + * WARNING: It's not completely safe to place initial chunks onto non-primary + * shards using this method because a conflict may result if multiple map-reduce + * operations are writing to the same output collection, for instance. + * + */ + virtual Status shardCollection(const std::string& ns, + const ShardKeyPattern& fieldsAndOrder, + bool unique, + std::vector<BSONObj>* initPoints, + std::set<ShardId>* initShardIds = nullptr) = 0; + + /** + * + * Adds a new shard. It expects a standalone mongod process or replica set to be running + * on the provided address. + * + * @param name is an optional string with the name of the shard. + * If empty, a name will be automatically generated. + * @param shardConnectionString is the connection string of the shard being added. + * @param maxSize is the optional space quota in bytes. Zeros means there's + * no limitation to space usage. + * @return either an !OK status or the name of the newly added shard. + */ + virtual StatusWith<std::string> addShard(const std::string& name, + const ConnectionString& shardConnectionString, + const long long maxSize) = 0; + + /** + * Tries to remove a shard. To completely remove a shard from a sharded cluster, + * the data residing in that shard must be moved to the remaining shards in the + * cluster by "draining" chunks from that shard. + * + * Because of the asynchronous nature of the draining mechanism, this method returns + * the current draining status. See ShardDrainingStatus enum definition for more details. + */ + virtual StatusWith<ShardDrainingStatus> removeShard(OperationContext* txn, + const std::string& name) = 0; + + /** + * Creates a new database entry for the specified database name in the configuration + * metadata and sets the specified shard as primary. + * + * @param dbName name of the database (case sensitive) + * + * Returns Status::OK on success or any error code indicating the failure. These are some + * of the known failures: + * - NamespaceExists - database already exists + * - DatabaseDifferCase - database already exists, but with a different case + * - ShardNotFound - could not find a shard to place the DB on + */ + virtual Status createDatabase(const std::string& dbName) = 0; + + /** + * Updates or creates the metadata for a given database. + */ + virtual Status updateDatabase(const std::string& dbName, const DatabaseType& db) = 0; + + /** + * Retrieves the metadata for a given database, if it exists. + * + * @param dbName name of the database (case sensitive) + * + * Returns Status::OK along with the database information or any error code indicating the + * failure. These are some of the known failures: + * - DatabaseNotFound - database does not exist + */ + virtual StatusWith<DatabaseType> getDatabase(const std::string& dbName) = 0; + + /** + * Updates or creates the metadata for a given collection. + */ + virtual Status updateCollection(const std::string& collNs, const CollectionType& coll) = 0; + + /** + * Retrieves the metadata for a given collection, if it exists. + * + * @param collectionNs fully qualified name of the collection (case sensitive) + * + * Returns Status::OK along with the collection information or any error code indicating + * the failure. These are some of the known failures: + * - NamespaceNotFound - collection does not exist + */ + virtual StatusWith<CollectionType> getCollection(const std::string& collNs) = 0; + + /** + * Retrieves all collections undera specified database (or in the system). + * + * @param dbName an optional database name. Must be nullptr or non-empty. If nullptr is + * specified, all collections on the system are returned. + * @param collections variable to receive the set of collections. + * + * Returns a !OK status if an error occurs. + */ + virtual Status getCollections(const std::string* dbName, + std::vector<CollectionType>* collections) = 0; + + /** + * Drops the specified collection from the collection metadata store. + * + * Returns Status::OK if successful or any error code indicating the failure. These are + * some of the known failures: + * - NamespaceNotFound - collection does not exist + */ + virtual Status dropCollection(const std::string& collectionNs) = 0; /** - * Used to indicate to the caller of the removeShard method whether draining of chunks for - * a particular shard has started, is ongoing, or has been completed. + * Retrieves all databases for a shard. + * + * Returns a !OK status if an error occurs. */ - enum ShardDrainingStatus { - STARTED, - ONGOING, - COMPLETED, - }; + virtual Status getDatabasesForShard(const std::string& shardName, + std::vector<std::string>* dbs) = 0; /** - * Abstracts reads and writes of the sharding catalog metadata. + * Gets the requested number of chunks (of type ChunkType) that satisfy a query. * - * All implementations of this interface should go directly to the persistent backing store - * and should avoid doing any caching of their own. The caching is delegated to a parallel - * read-only view of the catalog, which is maintained by a higher level code. + * @param query The query to filter out the results. + * @param nToReturn The number of chunk entries to return. 0 means all. + * @param chunks Vector entry to receive the results + * + * Returns a !OK status if an error occurs. + */ + virtual Status getChunks(const Query& query, int nToReturn, std::vector<ChunkType>* chunks) = 0; + + /** + * Retrieves all tags for the specified collection. + */ + virtual Status getTagsForCollection(const std::string& collectionNs, + std::vector<TagsType>* tags) = 0; + + /** + * Retrieves the most appropriate tag, which overlaps with the specified chunk. If no tags + * overlap, returns an empty string. + */ + virtual StatusWith<std::string> getTagForChunk(const std::string& collectionNs, + const ChunkType& chunk) = 0; + + /** + * Retrieves all shards in this sharded cluster. + * Returns a !OK status if an error occurs. */ - class CatalogManager { - MONGO_DISALLOW_COPYING(CatalogManager); - public: - virtual ~CatalogManager() = default; - - /** - * Retrieves the connection string for the catalog manager's backing server. - */ - virtual ConnectionString connectionString() const = 0; - - /** - * Performs implementation-specific startup tasks including but not limited to doing any - * necessary config schema upgrade work. Must be run after the catalog manager - * has been installed into the global 'grid' object. - */ - virtual Status startup(bool upgrade) = 0; - - /** - * Performs necessary cleanup when shutting down cleanly. - */ - virtual void shutDown() = 0; - - /** - * Creates a new database or updates the sharding status for an existing one. Cannot be - * used for the admin/config/local DBs, which should not be created or sharded manually - * anyways. - * - * Returns Status::OK on success or any error code indicating the failure. These are some - * of the known failures: - * - DatabaseDifferCase - database already exists, but with a different case - * - ShardNotFound - could not find a shard to place the DB on - */ - virtual Status enableSharding(const std::string& dbName) = 0; - - /** - * Shards a collection. Assumes that the database is enabled for sharding. - * - * @param ns: namespace of collection to shard - * @param fieldsAndOrder: shardKey pattern - * @param unique: if true, ensure underlying index enforces a unique constraint. - * @param initPoints: create chunks based on a set of specified split points. - * @param initShardIds: if nullptr, use primary shard as lone shard for DB. - * - * WARNING: It's not completely safe to place initial chunks onto non-primary - * shards using this method because a conflict may result if multiple map-reduce - * operations are writing to the same output collection, for instance. - * - */ - virtual Status shardCollection(const std::string& ns, - const ShardKeyPattern& fieldsAndOrder, - bool unique, - std::vector<BSONObj>* initPoints, - std::set<ShardId>* initShardIds = nullptr) = 0; - - /** - * - * Adds a new shard. It expects a standalone mongod process or replica set to be running - * on the provided address. - * - * @param name is an optional string with the name of the shard. - * If empty, a name will be automatically generated. - * @param shardConnectionString is the connection string of the shard being added. - * @param maxSize is the optional space quota in bytes. Zeros means there's - * no limitation to space usage. - * @return either an !OK status or the name of the newly added shard. - */ - virtual StatusWith<std::string> addShard(const std::string& name, - const ConnectionString& shardConnectionString, - const long long maxSize) = 0; - - /** - * Tries to remove a shard. To completely remove a shard from a sharded cluster, - * the data residing in that shard must be moved to the remaining shards in the - * cluster by "draining" chunks from that shard. - * - * Because of the asynchronous nature of the draining mechanism, this method returns - * the current draining status. See ShardDrainingStatus enum definition for more details. - */ - virtual StatusWith<ShardDrainingStatus> removeShard(OperationContext* txn, - const std::string& name) = 0; - - /** - * Creates a new database entry for the specified database name in the configuration - * metadata and sets the specified shard as primary. - * - * @param dbName name of the database (case sensitive) - * - * Returns Status::OK on success or any error code indicating the failure. These are some - * of the known failures: - * - NamespaceExists - database already exists - * - DatabaseDifferCase - database already exists, but with a different case - * - ShardNotFound - could not find a shard to place the DB on - */ - virtual Status createDatabase(const std::string& dbName) = 0; - - /** - * Updates or creates the metadata for a given database. - */ - virtual Status updateDatabase(const std::string& dbName, const DatabaseType& db) = 0; - - /** - * Retrieves the metadata for a given database, if it exists. - * - * @param dbName name of the database (case sensitive) - * - * Returns Status::OK along with the database information or any error code indicating the - * failure. These are some of the known failures: - * - DatabaseNotFound - database does not exist - */ - virtual StatusWith<DatabaseType> getDatabase(const std::string& dbName) = 0; - - /** - * Updates or creates the metadata for a given collection. - */ - virtual Status updateCollection(const std::string& collNs, const CollectionType& coll) = 0; - - /** - * Retrieves the metadata for a given collection, if it exists. - * - * @param collectionNs fully qualified name of the collection (case sensitive) - * - * Returns Status::OK along with the collection information or any error code indicating - * the failure. These are some of the known failures: - * - NamespaceNotFound - collection does not exist - */ - virtual StatusWith<CollectionType> getCollection(const std::string& collNs) = 0; - - /** - * Retrieves all collections undera specified database (or in the system). - * - * @param dbName an optional database name. Must be nullptr or non-empty. If nullptr is - * specified, all collections on the system are returned. - * @param collections variable to receive the set of collections. - * - * Returns a !OK status if an error occurs. - */ - virtual Status getCollections(const std::string* dbName, - std::vector<CollectionType>* collections) = 0; - - /** - * Drops the specified collection from the collection metadata store. - * - * Returns Status::OK if successful or any error code indicating the failure. These are - * some of the known failures: - * - NamespaceNotFound - collection does not exist - */ - virtual Status dropCollection(const std::string& collectionNs) = 0; - - /** - * Retrieves all databases for a shard. - * - * Returns a !OK status if an error occurs. - */ - virtual Status getDatabasesForShard(const std::string& shardName, - std::vector<std::string>* dbs) = 0; - - /** - * Gets the requested number of chunks (of type ChunkType) that satisfy a query. - * - * @param query The query to filter out the results. - * @param nToReturn The number of chunk entries to return. 0 means all. - * @param chunks Vector entry to receive the results - * - * Returns a !OK status if an error occurs. - */ - virtual Status getChunks(const Query& query, - int nToReturn, - std::vector<ChunkType>* chunks) = 0; - - /** - * Retrieves all tags for the specified collection. - */ - virtual Status getTagsForCollection(const std::string& collectionNs, - std::vector<TagsType>* tags) = 0; - - /** - * Retrieves the most appropriate tag, which overlaps with the specified chunk. If no tags - * overlap, returns an empty string. - */ - virtual StatusWith<std::string> getTagForChunk(const std::string& collectionNs, - const ChunkType& chunk) = 0; - - /** - * Retrieves all shards in this sharded cluster. - * Returns a !OK status if an error occurs. - */ - virtual Status getAllShards(std::vector<ShardType>* shards) = 0; - - /** - * Returns true if host is being used as a shard. - * Otherwise, returns false. - */ - virtual bool isShardHost(const ConnectionString& shardConnectionString) = 0; - - /** - * Runs a user management command on the config servers. - * @param commandName: name of command - * @param dbname: database for which the user management command is invoked - * @param cmdObj: command obj - * @param result: contains data returned from config servers - * Returns true on success. - */ - virtual bool runUserManagementWriteCommand(const std::string& commandName, - const std::string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) = 0; - - /** - * Runs a read-only user management command on a single config server. - */ - virtual bool runUserManagementReadCommand(const std::string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) = 0; - - /** - * Applies oplog entries to the config servers. - * Used by mergeChunk, splitChunk, and moveChunk commands. - * - * @param updateOps: oplog entries to apply - * @param preCondition: preconditions for applying oplog entries - */ - virtual Status applyChunkOpsDeprecated(const BSONArray& updateOps, - const BSONArray& preCondition) = 0; - - /** - * Logs to the actionlog. - * Used by the balancer to report the result of a balancing round. - */ - virtual void logAction(const ActionLogType& actionLog) = 0; - - /** - * Logs a diagnostic event locally and on the config server. - * - * NOTE: This method is best effort so it should never throw. - * - * @param opCtx The operation context of the call doing the logging - * @param what E.g. "split", "migrate" - * @param ns To which collection the metadata change is being applied - * @param detail Additional info about the metadata change (not interpreted) - */ - virtual void logChange(OperationContext* opCtx, - const std::string& what, - const std::string& ns, - const BSONObj& detail) = 0; - - /** - * Returns global settings for a certain key. - * @param key: key for SettingsType::ConfigNS document. - * - * Returns ErrorCodes::NoMatchingDocument if no SettingsType::ConfigNS document - * with such key exists. - * Returns ErrorCodes::FailedToParse if we encountered an error while parsing - * the settings document. - */ - virtual StatusWith<SettingsType> getGlobalSettings(const std::string& key) = 0; - - /** - * Directly sends the specified command to the config server and returns the response. - * - * NOTE: Usage of this function is disallowed in new code, which should instead go through - * the regular catalog management calls. It is currently only used privately by this - * class and externally for writes to the admin/config namespaces. - * - * @param request Request to be sent to the config server. - * @param response Out parameter to receive the response. Can be nullptr. - */ - virtual void writeConfigServerDirect(const BatchedCommandRequest& request, - BatchedCommandResponse* response) = 0; - - virtual DistLockManager* getDistLockManager() = 0; - - /** - * Directly inserts a document in the specified namespace on the config server (only the - * config or admin databases). If the document does not have _id field, the field will be - * added. - * - * NOTE: Should not be used in new code. Instead add a new metadata operation to the - * interface. - */ - Status insert(const std::string& ns, - const BSONObj& doc, - BatchedCommandResponse* response); - - /** - * Updates a document in the specified namespace on the config server (only the config or - * admin databases). - * - * NOTE: Should not be used in new code. Instead add a new metadata operation to the - * interface. - */ - Status update(const std::string& ns, - const BSONObj& query, - const BSONObj& update, - bool upsert, - bool multi, - BatchedCommandResponse* response); - - /** - * Removes a document from the specified namespace on the config server (only the config - * or admin databases). - * - * NOTE: Should not be used in new code. Instead add a new metadata operation to the - * interface. - */ - Status remove(const std::string& ns, - const BSONObj& query, - int limit, - BatchedCommandResponse* response); - - protected: - CatalogManager() = default; - }; - -} // namespace mongo + virtual Status getAllShards(std::vector<ShardType>* shards) = 0; + + /** + * Returns true if host is being used as a shard. + * Otherwise, returns false. + */ + virtual bool isShardHost(const ConnectionString& shardConnectionString) = 0; + + /** + * Runs a user management command on the config servers. + * @param commandName: name of command + * @param dbname: database for which the user management command is invoked + * @param cmdObj: command obj + * @param result: contains data returned from config servers + * Returns true on success. + */ + virtual bool runUserManagementWriteCommand(const std::string& commandName, + const std::string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) = 0; + + /** + * Runs a read-only user management command on a single config server. + */ + virtual bool runUserManagementReadCommand(const std::string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) = 0; + + /** + * Applies oplog entries to the config servers. + * Used by mergeChunk, splitChunk, and moveChunk commands. + * + * @param updateOps: oplog entries to apply + * @param preCondition: preconditions for applying oplog entries + */ + virtual Status applyChunkOpsDeprecated(const BSONArray& updateOps, + const BSONArray& preCondition) = 0; + + /** + * Logs to the actionlog. + * Used by the balancer to report the result of a balancing round. + */ + virtual void logAction(const ActionLogType& actionLog) = 0; + + /** + * Logs a diagnostic event locally and on the config server. + * + * NOTE: This method is best effort so it should never throw. + * + * @param opCtx The operation context of the call doing the logging + * @param what E.g. "split", "migrate" + * @param ns To which collection the metadata change is being applied + * @param detail Additional info about the metadata change (not interpreted) + */ + virtual void logChange(OperationContext* opCtx, + const std::string& what, + const std::string& ns, + const BSONObj& detail) = 0; + + /** + * Returns global settings for a certain key. + * @param key: key for SettingsType::ConfigNS document. + * + * Returns ErrorCodes::NoMatchingDocument if no SettingsType::ConfigNS document + * with such key exists. + * Returns ErrorCodes::FailedToParse if we encountered an error while parsing + * the settings document. + */ + virtual StatusWith<SettingsType> getGlobalSettings(const std::string& key) = 0; + + /** + * Directly sends the specified command to the config server and returns the response. + * + * NOTE: Usage of this function is disallowed in new code, which should instead go through + * the regular catalog management calls. It is currently only used privately by this + * class and externally for writes to the admin/config namespaces. + * + * @param request Request to be sent to the config server. + * @param response Out parameter to receive the response. Can be nullptr. + */ + virtual void writeConfigServerDirect(const BatchedCommandRequest& request, + BatchedCommandResponse* response) = 0; + + virtual DistLockManager* getDistLockManager() = 0; + + /** + * Directly inserts a document in the specified namespace on the config server (only the + * config or admin databases). If the document does not have _id field, the field will be + * added. + * + * NOTE: Should not be used in new code. Instead add a new metadata operation to the + * interface. + */ + Status insert(const std::string& ns, const BSONObj& doc, BatchedCommandResponse* response); + + /** + * Updates a document in the specified namespace on the config server (only the config or + * admin databases). + * + * NOTE: Should not be used in new code. Instead add a new metadata operation to the + * interface. + */ + Status update(const std::string& ns, + const BSONObj& query, + const BSONObj& update, + bool upsert, + bool multi, + BatchedCommandResponse* response); + + /** + * Removes a document from the specified namespace on the config server (only the config + * or admin databases). + * + * NOTE: Should not be used in new code. Instead add a new metadata operation to the + * interface. + */ + Status remove(const std::string& ns, + const BSONObj& query, + int limit, + BatchedCommandResponse* response); + +protected: + CatalogManager() = default; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/dist_lock_catalog.cpp b/src/mongo/s/catalog/dist_lock_catalog.cpp index 44f024e2747..33d14a590e0 100644 --- a/src/mongo/s/catalog/dist_lock_catalog.cpp +++ b/src/mongo/s/catalog/dist_lock_catalog.cpp @@ -32,8 +32,7 @@ namespace mongo { - DistLockCatalog::ServerInfo::ServerInfo(Date_t time, OID _electionId): - serverTime(std::move(time)), electionId(std::move(_electionId)) { - } +DistLockCatalog::ServerInfo::ServerInfo(Date_t time, OID _electionId) + : serverTime(std::move(time)), electionId(std::move(_electionId)) {} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/dist_lock_catalog.h b/src/mongo/s/catalog/dist_lock_catalog.h index 4bde917e763..3c83e0433a7 100644 --- a/src/mongo/s/catalog/dist_lock_catalog.h +++ b/src/mongo/s/catalog/dist_lock_catalog.h @@ -34,121 +34,121 @@ namespace mongo { - class LockpingsType; - class LocksType; - class Status; - template <typename T> class StatusWith; +class LockpingsType; +class LocksType; +class Status; +template <typename T> +class StatusWith; +/** + * Interface for the distributed lock operations. + */ +class DistLockCatalog { +public: /** - * Interface for the distributed lock operations. + * Simple data structure for storing server local time and election id. */ - class DistLockCatalog { + struct ServerInfo { public: - /** - * Simple data structure for storing server local time and election id. - */ - struct ServerInfo { - public: - ServerInfo(); // TODO: SERVER-18007 - ServerInfo(Date_t time, OID electionId); - - // The local time of the server at the time this was created. - Date_t serverTime; - - // The election id of the replica set member at the time this was created. - OID electionId; - }; - - virtual ~DistLockCatalog() = default; - - /** - * Returns the ping document of the specified processID. - * Common status errors include socket errors. - */ - virtual StatusWith<LockpingsType> getPing(StringData processID) = 0; - - /** - * Updates the ping document. Creates a new entry if it does not exists. - * Common status errors include socket errors. - */ - virtual Status ping(StringData processID, Date_t ping) = 0; - - /** - * Attempts to update the owner of a lock identified by lockID to lockSessionID. - * Will only be successful if lock is not held. - * - * The other parameters are for diagnostic purposes: - * - who: unique string for the caller trying to grab the lock. - * - processId: unique string for the process trying to grab the lock. - * - time: the time when this is attempted. - * - why: reason for taking the lock. - * - * Returns the result of the operation. - * Returns LockStateChangeFailed if the lock acquisition cannot be done because lock - * is already held elsewhere. - * - * Common status errors include socket and duplicate key errors. - */ - virtual StatusWith<LocksType> grabLock(StringData lockID, + ServerInfo(); // TODO: SERVER-18007 + ServerInfo(Date_t time, OID electionId); + + // The local time of the server at the time this was created. + Date_t serverTime; + + // The election id of the replica set member at the time this was created. + OID electionId; + }; + + virtual ~DistLockCatalog() = default; + + /** + * Returns the ping document of the specified processID. + * Common status errors include socket errors. + */ + virtual StatusWith<LockpingsType> getPing(StringData processID) = 0; + + /** + * Updates the ping document. Creates a new entry if it does not exists. + * Common status errors include socket errors. + */ + virtual Status ping(StringData processID, Date_t ping) = 0; + + /** + * Attempts to update the owner of a lock identified by lockID to lockSessionID. + * Will only be successful if lock is not held. + * + * The other parameters are for diagnostic purposes: + * - who: unique string for the caller trying to grab the lock. + * - processId: unique string for the process trying to grab the lock. + * - time: the time when this is attempted. + * - why: reason for taking the lock. + * + * Returns the result of the operation. + * Returns LockStateChangeFailed if the lock acquisition cannot be done because lock + * is already held elsewhere. + * + * Common status errors include socket and duplicate key errors. + */ + virtual StatusWith<LocksType> grabLock(StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) = 0; + + /** + * Attempts to forcefully transfer the ownership of a lock from currentHolderTS + * to lockSessionID. + * + * The other parameters are for diagnostic purposes: + * - who: unique string for the caller trying to grab the lock. + * - processId: unique string for the process trying to grab the lock. + * - time: the time when this is attempted. + * - why: reason for taking the lock. + * + * Returns the result of the operation. + * Returns LockStateChangeFailed if the lock acquisition fails. + * + * Common status errors include socket errors. + */ + virtual StatusWith<LocksType> overtakeLock(StringData lockID, const OID& lockSessionID, + const OID& currentHolderTS, StringData who, StringData processId, Date_t time, StringData why) = 0; - /** - * Attempts to forcefully transfer the ownership of a lock from currentHolderTS - * to lockSessionID. - * - * The other parameters are for diagnostic purposes: - * - who: unique string for the caller trying to grab the lock. - * - processId: unique string for the process trying to grab the lock. - * - time: the time when this is attempted. - * - why: reason for taking the lock. - * - * Returns the result of the operation. - * Returns LockStateChangeFailed if the lock acquisition fails. - * - * Common status errors include socket errors. - */ - virtual StatusWith<LocksType> overtakeLock(StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) = 0; - - /** - * Attempts to set the state of the lock document with lockSessionID to unlocked. - * Common status errors include socket errors. - */ - virtual Status unlock(const OID& lockSessionID) = 0; - - /** - * Get some information from the config server primary. - * Common status errors include socket errors. - */ - virtual StatusWith<ServerInfo> getServerInfo() = 0; - - /** - * Returns the lock document. - * Returns LockNotFound if lock document doesn't exist. - * Common status errors include socket errors. - */ - virtual StatusWith<LocksType> getLockByTS(const OID& lockSessionID) = 0; - - /** - * Returns the lock document. - * Common status errors include socket errors. - */ - virtual StatusWith<LocksType> getLockByName(StringData name) = 0; - - /** - * Attempts to delete the ping document corresponding to the given processId. - * Common status errors include socket errors. - */ - virtual Status stopPing(StringData processId) = 0; + /** + * Attempts to set the state of the lock document with lockSessionID to unlocked. + * Common status errors include socket errors. + */ + virtual Status unlock(const OID& lockSessionID) = 0; + + /** + * Get some information from the config server primary. + * Common status errors include socket errors. + */ + virtual StatusWith<ServerInfo> getServerInfo() = 0; - }; + /** + * Returns the lock document. + * Returns LockNotFound if lock document doesn't exist. + * Common status errors include socket errors. + */ + virtual StatusWith<LocksType> getLockByTS(const OID& lockSessionID) = 0; + + /** + * Returns the lock document. + * Common status errors include socket errors. + */ + virtual StatusWith<LocksType> getLockByName(StringData name) = 0; + + /** + * Attempts to delete the ping document corresponding to the given processId. + * Common status errors include socket errors. + */ + virtual Status stopPing(StringData processId) = 0; +}; } diff --git a/src/mongo/s/catalog/dist_lock_catalog_impl.cpp b/src/mongo/s/catalog/dist_lock_catalog_impl.cpp index 57aab2dd2fc..e7a2830e546 100644 --- a/src/mongo/s/catalog/dist_lock_catalog_impl.cpp +++ b/src/mongo/s/catalog/dist_lock_catalog_impl.cpp @@ -49,397 +49,371 @@ namespace mongo { - using std::string; +using std::string; namespace { - const char kCmdResponseWriteConcernField[] = "writeConcernError"; - const char kFindAndModifyResponseResultDocField[] = "value"; - const char kLocalTimeField[] = "localTime"; - const ReadPreferenceSetting kReadPref(ReadPreference::PrimaryOnly, TagSet()); - - /** - * Returns the resulting new object from the findAndModify response object. - * Returns LockStateChangeFailed if value field was null, which indicates that - * the findAndModify command did not modify any document. - * This also checks for errors in the response object. - */ - StatusWith<BSONObj> extractFindAndModifyNewObj(const BSONObj& responseObj) { - auto cmdStatus = getStatusFromCommandResult(responseObj); - - if (!cmdStatus.isOK()) { - return cmdStatus; - } +const char kCmdResponseWriteConcernField[] = "writeConcernError"; +const char kFindAndModifyResponseResultDocField[] = "value"; +const char kLocalTimeField[] = "localTime"; +const ReadPreferenceSetting kReadPref(ReadPreference::PrimaryOnly, TagSet()); - BSONElement wcErrorElem; - auto wcErrStatus = bsonExtractTypedField(responseObj, - kCmdResponseWriteConcernField, - Object, - &wcErrorElem); +/** + * Returns the resulting new object from the findAndModify response object. + * Returns LockStateChangeFailed if value field was null, which indicates that + * the findAndModify command did not modify any document. + * This also checks for errors in the response object. + */ +StatusWith<BSONObj> extractFindAndModifyNewObj(const BSONObj& responseObj) { + auto cmdStatus = getStatusFromCommandResult(responseObj); - if (wcErrStatus.isOK()) { - BSONObj wcErrObj(wcErrorElem.Obj()); - WCErrorDetail wcError; + if (!cmdStatus.isOK()) { + return cmdStatus; + } - string wcErrorParseMsg; - if (!wcError.parseBSON(wcErrObj, &wcErrorParseMsg)) { - return Status(ErrorCodes::UnsupportedFormat, wcErrorParseMsg); - } + BSONElement wcErrorElem; + auto wcErrStatus = + bsonExtractTypedField(responseObj, kCmdResponseWriteConcernField, Object, &wcErrorElem); - return {ErrorCodes::WriteConcernFailed, wcError.getErrMessage()}; - } + if (wcErrStatus.isOK()) { + BSONObj wcErrObj(wcErrorElem.Obj()); + WCErrorDetail wcError; - if (wcErrStatus != ErrorCodes::NoSuchKey) { - return wcErrStatus; + string wcErrorParseMsg; + if (!wcError.parseBSON(wcErrObj, &wcErrorParseMsg)) { + return Status(ErrorCodes::UnsupportedFormat, wcErrorParseMsg); } - if (const auto& newDocElem = responseObj[kFindAndModifyResponseResultDocField]) { - if (newDocElem.isNull()) { - return {ErrorCodes::LockStateChangeFailed, - "findAndModify query predicate didn't match any lock document"}; - } + return {ErrorCodes::WriteConcernFailed, wcError.getErrMessage()}; + } - if (!newDocElem.isABSONObj()) { - return {ErrorCodes::UnsupportedFormat, - str::stream() << "expected an object from the findAndModify response '" - << kFindAndModifyResponseResultDocField << "'field, got: " - << newDocElem}; - } + if (wcErrStatus != ErrorCodes::NoSuchKey) { + return wcErrStatus; + } - return newDocElem.Obj(); + if (const auto& newDocElem = responseObj[kFindAndModifyResponseResultDocField]) { + if (newDocElem.isNull()) { + return {ErrorCodes::LockStateChangeFailed, + "findAndModify query predicate didn't match any lock document"}; } - return {ErrorCodes::LockStateChangeFailed, - str::stream() << "no '" << kFindAndModifyResponseResultDocField - << "' in findAndModify response"}; - } - - /** - * Extract the electionId from a command response. - * - * TODO: this needs to support OP_COMMAND metadata. - */ - StatusWith<OID> extractElectionId(const BSONObj& responseObj) { - - BSONElement gleStatsElem; - auto gleStatus = bsonExtractTypedField(responseObj, - "$gleStats", - Object, - &gleStatsElem); - - if (!gleStatus.isOK()) { - return {ErrorCodes::UnsupportedFormat, gleStatus.reason()}; + if (!newDocElem.isABSONObj()) { + return {ErrorCodes::UnsupportedFormat, + str::stream() << "expected an object from the findAndModify response '" + << kFindAndModifyResponseResultDocField + << "'field, got: " << newDocElem}; } - OID electionId; + return newDocElem.Obj(); + } - auto electionIdStatus = bsonExtractOIDField(gleStatsElem.Obj(), - "electionId", - &electionId); + return {ErrorCodes::LockStateChangeFailed, + str::stream() << "no '" << kFindAndModifyResponseResultDocField + << "' in findAndModify response"}; +} - if (!electionIdStatus.isOK()) { - return {ErrorCodes::UnsupportedFormat, electionIdStatus.reason()}; - } +/** + * Extract the electionId from a command response. + * + * TODO: this needs to support OP_COMMAND metadata. + */ +StatusWith<OID> extractElectionId(const BSONObj& responseObj) { + BSONElement gleStatsElem; + auto gleStatus = bsonExtractTypedField(responseObj, "$gleStats", Object, &gleStatsElem); - return electionId; + if (!gleStatus.isOK()) { + return {ErrorCodes::UnsupportedFormat, gleStatus.reason()}; } -} // unnamed namespace - - DistLockCatalogImpl::DistLockCatalogImpl(RemoteCommandTargeter* targeter, - RemoteCommandRunner* executor, - Milliseconds writeConcernTimeout): - _cmdRunner(executor), - _targeter(targeter), - _writeConcern(WriteConcernOptions(WriteConcernOptions::kMajority, - WriteConcernOptions::JOURNAL, - writeConcernTimeout.count())), - _lockPingNS(LockpingsType::ConfigNS), - _locksNS(LocksType::ConfigNS) { + OID electionId; + + auto electionIdStatus = bsonExtractOIDField(gleStatsElem.Obj(), "electionId", &electionId); + + if (!electionIdStatus.isOK()) { + return {ErrorCodes::UnsupportedFormat, electionIdStatus.reason()}; } - DistLockCatalogImpl::~DistLockCatalogImpl() = default; + return electionId; +} + +} // unnamed namespace + +DistLockCatalogImpl::DistLockCatalogImpl(RemoteCommandTargeter* targeter, + RemoteCommandRunner* executor, + Milliseconds writeConcernTimeout) + : _cmdRunner(executor), + _targeter(targeter), + _writeConcern(WriteConcernOptions(WriteConcernOptions::kMajority, + WriteConcernOptions::JOURNAL, + writeConcernTimeout.count())), + _lockPingNS(LockpingsType::ConfigNS), + _locksNS(LocksType::ConfigNS) {} + +DistLockCatalogImpl::~DistLockCatalogImpl() = default; - StatusWith<LockpingsType> DistLockCatalogImpl::getPing(StringData processID) { - invariant(false); // TODO +StatusWith<LockpingsType> DistLockCatalogImpl::getPing(StringData processID) { + invariant(false); // TODO +} + +Status DistLockCatalogImpl::ping(StringData processID, Date_t ping) { + auto targetStatus = _targeter->findHost(kReadPref); + + if (!targetStatus.isOK()) { + return targetStatus.getStatus(); } - Status DistLockCatalogImpl::ping(StringData processID, Date_t ping) { - auto targetStatus = _targeter->findHost(kReadPref); + auto request = + FindAndModifyRequest::makeUpdate(_lockPingNS, + BSON(LockpingsType::process(processID.toString())), + BSON("$set" << BSON(LockpingsType::ping(ping)))); + request.setUpsert(true); + request.setWriteConcern(_writeConcern); - if (!targetStatus.isOK()) { - return targetStatus.getStatus(); - } + auto resultStatus = _cmdRunner->runCommand( + RemoteCommandRequest(targetStatus.getValue(), _locksNS.db().toString(), request.toBSON())); - auto request = FindAndModifyRequest::makeUpdate(_lockPingNS, - BSON(LockpingsType::process(processID.toString())), - BSON("$set" << BSON(LockpingsType::ping(ping)))); - request.setUpsert(true); - request.setWriteConcern(_writeConcern); + if (!resultStatus.isOK()) { + return resultStatus.getStatus(); + } - auto resultStatus = _cmdRunner->runCommand( - RemoteCommandRequest(targetStatus.getValue(), - _locksNS.db().toString(), - request.toBSON())); + const RemoteCommandResponse& response = resultStatus.getValue(); + BSONObj responseObj(response.data); - if (!resultStatus.isOK()) { - return resultStatus.getStatus(); - } + auto cmdStatus = getStatusFromCommandResult(responseObj); - const RemoteCommandResponse& response = resultStatus.getValue(); - BSONObj responseObj(response.data); + if (!cmdStatus.isOK()) { + return cmdStatus; + } - auto cmdStatus = getStatusFromCommandResult(responseObj); + auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj); + return findAndModifyStatus.getStatus(); +} - if (!cmdStatus.isOK()) { - return cmdStatus; - } +StatusWith<LocksType> DistLockCatalogImpl::grabLock(StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { + auto targetStatus = _targeter->findHost(kReadPref); - auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj); - return findAndModifyStatus.getStatus(); + if (!targetStatus.isOK()) { + return targetStatus.getStatus(); } - StatusWith<LocksType> DistLockCatalogImpl::grabLock(StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why) { - auto targetStatus = _targeter->findHost(kReadPref); + BSONObj newLockDetails(BSON(LocksType::lockID(lockSessionID) + << LocksType::state(LocksType::LOCKED) << LocksType::who() << who + << LocksType::process() << processId << LocksType::when(time) + << LocksType::why() << why)); - if (!targetStatus.isOK()) { - return targetStatus.getStatus(); - } + auto request = FindAndModifyRequest::makeUpdate( + _locksNS, + BSON(LocksType::name() << lockID << LocksType::state(LocksType::UNLOCKED)), + BSON("$set" << newLockDetails)); + request.setUpsert(true); + request.setShouldReturnNew(true); + request.setWriteConcern(_writeConcern); - BSONObj newLockDetails(BSON(LocksType::lockID(lockSessionID) - << LocksType::state(LocksType::LOCKED) - << LocksType::who() << who - << LocksType::process() << processId - << LocksType::when(time) - << LocksType::why() << why)); - - auto request = FindAndModifyRequest::makeUpdate(_locksNS, - BSON(LocksType::name() << lockID << LocksType::state(LocksType::UNLOCKED)), - BSON("$set" << newLockDetails)); - request.setUpsert(true); - request.setShouldReturnNew(true); - request.setWriteConcern(_writeConcern); - - auto resultStatus = _cmdRunner->runCommand( - RemoteCommandRequest(targetStatus.getValue(), - _locksNS.db().toString(), - request.toBSON())); - - if (!resultStatus.isOK()) { - return resultStatus.getStatus(); - } + auto resultStatus = _cmdRunner->runCommand( + RemoteCommandRequest(targetStatus.getValue(), _locksNS.db().toString(), request.toBSON())); - const RemoteCommandResponse& response = resultStatus.getValue(); - BSONObj responseObj(response.data); + if (!resultStatus.isOK()) { + return resultStatus.getStatus(); + } - auto cmdStatus = getStatusFromCommandResult(responseObj); + const RemoteCommandResponse& response = resultStatus.getValue(); + BSONObj responseObj(response.data); - if (!cmdStatus.isOK()) { - return cmdStatus; - } + auto cmdStatus = getStatusFromCommandResult(responseObj); - auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj); - if (!findAndModifyStatus.isOK()) { - return findAndModifyStatus.getStatus(); - } + if (!cmdStatus.isOK()) { + return cmdStatus; + } - BSONObj newDoc(findAndModifyStatus.getValue()); + auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj); + if (!findAndModifyStatus.isOK()) { + return findAndModifyStatus.getStatus(); + } - if (newDoc.isEmpty()) { - return LocksType(); - } + BSONObj newDoc(findAndModifyStatus.getValue()); - LocksType lockDoc; - string errMsg; - if (!lockDoc.parseBSON(newDoc, &errMsg)) { - return {ErrorCodes::FailedToParse, errMsg}; - } + if (newDoc.isEmpty()) { + return LocksType(); + } - return lockDoc; + LocksType lockDoc; + string errMsg; + if (!lockDoc.parseBSON(newDoc, &errMsg)) { + return {ErrorCodes::FailedToParse, errMsg}; } - StatusWith<LocksType> DistLockCatalogImpl::overtakeLock(StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { - auto targetStatus = _targeter->findHost(kReadPref); - - if (!targetStatus.isOK()) { - return targetStatus.getStatus(); - } + return lockDoc; +} - BSONArrayBuilder orQueryBuilder; - orQueryBuilder.append(BSON(LocksType::name() << lockID - << LocksType::state(LocksType::UNLOCKED))); - orQueryBuilder.append(BSON(LocksType::name() << lockID - << LocksType::lockID(currentHolderTS))); - - BSONObj newLockDetails(BSON(LocksType::lockID(lockSessionID) - << LocksType::state(LocksType::LOCKED) - << LocksType::who() << who - << LocksType::process() << processId - << LocksType::when(time) - << LocksType::why() << why)); - - auto request = FindAndModifyRequest::makeUpdate(_locksNS, - BSON("$or" << orQueryBuilder.arr()), - BSON("$set" << newLockDetails)); - request.setShouldReturnNew(true); - request.setWriteConcern(_writeConcern); - - auto resultStatus = _cmdRunner->runCommand( - RemoteCommandRequest(targetStatus.getValue(), - _locksNS.db().toString(), - request.toBSON())); - - if (!resultStatus.isOK()) { - return resultStatus.getStatus(); - } +StatusWith<LocksType> DistLockCatalogImpl::overtakeLock(StringData lockID, + const OID& lockSessionID, + const OID& currentHolderTS, + StringData who, + StringData processId, + Date_t time, + StringData why) { + auto targetStatus = _targeter->findHost(kReadPref); - const RemoteCommandResponse& response = resultStatus.getValue(); - BSONObj responseObj(response.data); + if (!targetStatus.isOK()) { + return targetStatus.getStatus(); + } - auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj); - if (!findAndModifyStatus.isOK()) { - return findAndModifyStatus.getStatus(); - } + BSONArrayBuilder orQueryBuilder; + orQueryBuilder.append( + BSON(LocksType::name() << lockID << LocksType::state(LocksType::UNLOCKED))); + orQueryBuilder.append(BSON(LocksType::name() << lockID << LocksType::lockID(currentHolderTS))); - BSONObj newDoc(findAndModifyStatus.getValue()); + BSONObj newLockDetails(BSON(LocksType::lockID(lockSessionID) + << LocksType::state(LocksType::LOCKED) << LocksType::who() << who + << LocksType::process() << processId << LocksType::when(time) + << LocksType::why() << why)); - if (newDoc.isEmpty()) { - return LocksType(); - } + auto request = FindAndModifyRequest::makeUpdate( + _locksNS, BSON("$or" << orQueryBuilder.arr()), BSON("$set" << newLockDetails)); + request.setShouldReturnNew(true); + request.setWriteConcern(_writeConcern); - LocksType lockDoc; - string errMsg; - if (!lockDoc.parseBSON(newDoc, &errMsg)) { - return {ErrorCodes::FailedToParse, errMsg}; - } + auto resultStatus = _cmdRunner->runCommand( + RemoteCommandRequest(targetStatus.getValue(), _locksNS.db().toString(), request.toBSON())); - return lockDoc; + if (!resultStatus.isOK()) { + return resultStatus.getStatus(); } - Status DistLockCatalogImpl::unlock(const OID& lockSessionID) { - auto targetStatus = _targeter->findHost(kReadPref); - - if (!targetStatus.isOK()) { - return targetStatus.getStatus(); - } + const RemoteCommandResponse& response = resultStatus.getValue(); + BSONObj responseObj(response.data); - auto request = FindAndModifyRequest::makeUpdate(_locksNS, - BSON(LocksType::lockID(lockSessionID)), - BSON("$set" << BSON(LocksType::state(LocksType::UNLOCKED)))); - request.setWriteConcern(_writeConcern); + auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj); + if (!findAndModifyStatus.isOK()) { + return findAndModifyStatus.getStatus(); + } - auto resultStatus = _cmdRunner->runCommand( - RemoteCommandRequest(targetStatus.getValue(), - _locksNS.db().toString(), - request.toBSON())); + BSONObj newDoc(findAndModifyStatus.getValue()); - if (!resultStatus.isOK()) { - return resultStatus.getStatus(); - } + if (newDoc.isEmpty()) { + return LocksType(); + } - const RemoteCommandResponse& response = resultStatus.getValue(); - BSONObj responseObj(response.data); + LocksType lockDoc; + string errMsg; + if (!lockDoc.parseBSON(newDoc, &errMsg)) { + return {ErrorCodes::FailedToParse, errMsg}; + } - auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj).getStatus(); + return lockDoc; +} - if (findAndModifyStatus == ErrorCodes::LockStateChangeFailed) { - // Did not modify any document, which implies that the lock already has a - // a different owner. This is ok since it means that the objective of - // releasing ownership of the lock has already been accomplished. - return Status::OK(); - } +Status DistLockCatalogImpl::unlock(const OID& lockSessionID) { + auto targetStatus = _targeter->findHost(kReadPref); - return findAndModifyStatus; + if (!targetStatus.isOK()) { + return targetStatus.getStatus(); } - StatusWith<DistLockCatalog::ServerInfo> DistLockCatalogImpl::getServerInfo() { - auto targetStatus = _targeter->findHost(kReadPref); + auto request = FindAndModifyRequest::makeUpdate( + _locksNS, + BSON(LocksType::lockID(lockSessionID)), + BSON("$set" << BSON(LocksType::state(LocksType::UNLOCKED)))); + request.setWriteConcern(_writeConcern); - if (!targetStatus.isOK()) { - return targetStatus.getStatus(); - } + auto resultStatus = _cmdRunner->runCommand( + RemoteCommandRequest(targetStatus.getValue(), _locksNS.db().toString(), request.toBSON())); - auto resultStatus = _cmdRunner->runCommand(RemoteCommandRequest( - targetStatus.getValue(), - "admin", - BSON("serverStatus" << 1))); + if (!resultStatus.isOK()) { + return resultStatus.getStatus(); + } - if (!resultStatus.isOK()) { - return resultStatus.getStatus(); - } + const RemoteCommandResponse& response = resultStatus.getValue(); + BSONObj responseObj(response.data); - const RemoteCommandResponse& response = resultStatus.getValue(); - BSONObj responseObj(response.data); + auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj).getStatus(); - auto cmdStatus = getStatusFromCommandResult(responseObj); + if (findAndModifyStatus == ErrorCodes::LockStateChangeFailed) { + // Did not modify any document, which implies that the lock already has a + // a different owner. This is ok since it means that the objective of + // releasing ownership of the lock has already been accomplished. + return Status::OK(); + } - if (!cmdStatus.isOK()) { - return cmdStatus; - } + return findAndModifyStatus; +} - BSONElement localTimeElem; - auto localTimeStatus = bsonExtractTypedField(responseObj, - kLocalTimeField, - Date, - &localTimeElem); +StatusWith<DistLockCatalog::ServerInfo> DistLockCatalogImpl::getServerInfo() { + auto targetStatus = _targeter->findHost(kReadPref); - if (!localTimeStatus.isOK()) { - return {ErrorCodes::UnsupportedFormat, localTimeStatus.reason()}; - } + if (!targetStatus.isOK()) { + return targetStatus.getStatus(); + } - auto electionIdStatus = extractElectionId(responseObj); + auto resultStatus = _cmdRunner->runCommand( + RemoteCommandRequest(targetStatus.getValue(), "admin", BSON("serverStatus" << 1))); - if (!electionIdStatus.isOK()) { - return {ErrorCodes::UnsupportedFormat, electionIdStatus.getStatus().reason()}; - } + if (!resultStatus.isOK()) { + return resultStatus.getStatus(); + } + + const RemoteCommandResponse& response = resultStatus.getValue(); + BSONObj responseObj(response.data); - return DistLockCatalog::ServerInfo(localTimeElem.date(), electionIdStatus.getValue()); + auto cmdStatus = getStatusFromCommandResult(responseObj); + + if (!cmdStatus.isOK()) { + return cmdStatus; } - StatusWith<LocksType> DistLockCatalogImpl::getLockByTS(const OID& lockSessionID) { - invariant(false); // TODO + BSONElement localTimeElem; + auto localTimeStatus = + bsonExtractTypedField(responseObj, kLocalTimeField, Date, &localTimeElem); + + if (!localTimeStatus.isOK()) { + return {ErrorCodes::UnsupportedFormat, localTimeStatus.reason()}; } - StatusWith<LocksType> DistLockCatalogImpl::getLockByName(StringData name) { - invariant(false); // TODO + auto electionIdStatus = extractElectionId(responseObj); + + if (!electionIdStatus.isOK()) { + return {ErrorCodes::UnsupportedFormat, electionIdStatus.getStatus().reason()}; } - Status DistLockCatalogImpl::stopPing(StringData processId) { - auto targetStatus = _targeter->findHost(kReadPref); + return DistLockCatalog::ServerInfo(localTimeElem.date(), electionIdStatus.getValue()); +} - if (!targetStatus.isOK()) { - return targetStatus.getStatus(); - } +StatusWith<LocksType> DistLockCatalogImpl::getLockByTS(const OID& lockSessionID) { + invariant(false); // TODO +} - auto request = FindAndModifyRequest::makeRemove(_lockPingNS, - BSON(LockpingsType::process() << processId)); - request.setWriteConcern(_writeConcern); +StatusWith<LocksType> DistLockCatalogImpl::getLockByName(StringData name) { + invariant(false); // TODO +} - auto resultStatus = _cmdRunner->runCommand( - RemoteCommandRequest(targetStatus.getValue(), - _locksNS.db().toString(), - request.toBSON())); +Status DistLockCatalogImpl::stopPing(StringData processId) { + auto targetStatus = _targeter->findHost(kReadPref); - if (!resultStatus.isOK()) { - return resultStatus.getStatus(); - } + if (!targetStatus.isOK()) { + return targetStatus.getStatus(); + } - const RemoteCommandResponse& response = resultStatus.getValue(); - BSONObj responseObj(response.data); + auto request = + FindAndModifyRequest::makeRemove(_lockPingNS, BSON(LockpingsType::process() << processId)); + request.setWriteConcern(_writeConcern); - auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj); - return findAndModifyStatus.getStatus(); + auto resultStatus = _cmdRunner->runCommand( + RemoteCommandRequest(targetStatus.getValue(), _locksNS.db().toString(), request.toBSON())); + + if (!resultStatus.isOK()) { + return resultStatus.getStatus(); } -} // namespace mongo + const RemoteCommandResponse& response = resultStatus.getValue(); + BSONObj responseObj(response.data); + + auto findAndModifyStatus = extractFindAndModifyNewObj(responseObj); + return findAndModifyStatus.getStatus(); +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/dist_lock_catalog_impl.h b/src/mongo/s/catalog/dist_lock_catalog_impl.h index 4ec349cbfb8..c75d4abbace 100644 --- a/src/mongo/s/catalog/dist_lock_catalog_impl.h +++ b/src/mongo/s/catalog/dist_lock_catalog_impl.h @@ -37,53 +37,53 @@ namespace mongo { - class RemoteCommandRunner; - class RemoteCommandTargeter; +class RemoteCommandRunner; +class RemoteCommandTargeter; - class DistLockCatalogImpl final : public DistLockCatalog { - public: - DistLockCatalogImpl(RemoteCommandTargeter* targeter, - RemoteCommandRunner* executor, - Milliseconds writeConcernTimeout); +class DistLockCatalogImpl final : public DistLockCatalog { +public: + DistLockCatalogImpl(RemoteCommandTargeter* targeter, + RemoteCommandRunner* executor, + Milliseconds writeConcernTimeout); - virtual ~DistLockCatalogImpl(); + virtual ~DistLockCatalogImpl(); - virtual StatusWith<LockpingsType> getPing(StringData processID) override; + virtual StatusWith<LockpingsType> getPing(StringData processID) override; - virtual Status ping(StringData processID, Date_t ping) override; + virtual Status ping(StringData processID, Date_t ping) override; - virtual StatusWith<LocksType> grabLock(StringData lockID, + virtual StatusWith<LocksType> grabLock(StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) override; + + virtual StatusWith<LocksType> overtakeLock(StringData lockID, const OID& lockSessionID, + const OID& currentHolderTS, StringData who, StringData processId, Date_t time, StringData why) override; - virtual StatusWith<LocksType> overtakeLock(StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) override; - - virtual Status unlock(const OID& lockSessionID) override; + virtual Status unlock(const OID& lockSessionID) override; - virtual StatusWith<ServerInfo> getServerInfo() override; + virtual StatusWith<ServerInfo> getServerInfo() override; - virtual StatusWith<LocksType> getLockByTS(const OID& lockSessionID) override; + virtual StatusWith<LocksType> getLockByTS(const OID& lockSessionID) override; - virtual StatusWith<LocksType> getLockByName(StringData name) override; + virtual StatusWith<LocksType> getLockByName(StringData name) override; - virtual Status stopPing(StringData processId) override; + virtual Status stopPing(StringData processId) override; - private: - RemoteCommandRunner* _cmdRunner; - RemoteCommandTargeter* _targeter; +private: + RemoteCommandRunner* _cmdRunner; + RemoteCommandTargeter* _targeter; - // These are not static to avoid initialization order fiasco. - const WriteConcernOptions _writeConcern; - const NamespaceString _lockPingNS; - const NamespaceString _locksNS; - }; + // These are not static to avoid initialization order fiasco. + const WriteConcernOptions _writeConcern; + const NamespaceString _lockPingNS; + const NamespaceString _locksNS; +}; } diff --git a/src/mongo/s/catalog/dist_lock_catalog_impl_test.cpp b/src/mongo/s/catalog/dist_lock_catalog_impl_test.cpp index e10a582e4a1..bdbe5bd9059 100644 --- a/src/mongo/s/catalog/dist_lock_catalog_impl_test.cpp +++ b/src/mongo/s/catalog/dist_lock_catalog_impl_test.cpp @@ -44,20 +44,21 @@ f * Copyright (C) 2015 MongoDB Inc. namespace mongo { namespace { - const HostAndPort dummyHost("dummy", 123); - const Milliseconds kWTimeout(100); +const HostAndPort dummyHost("dummy", 123); +const Milliseconds kWTimeout(100); - const auto noTest = [](const RemoteCommandRequest& request) {}; +const auto noTest = [](const RemoteCommandRequest& request) {}; - TEST(DistLockCatalogImpl, BasicPing) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, BasicPing) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + executor.setNextExpectedCommand( + [](const RemoteCommandRequest& request) { ASSERT_EQUALS(dummyHost, request.target); ASSERT_EQUALS("config", request.dbname); @@ -81,92 +82,93 @@ namespace { _id: "abcd", ping: { $date: "2014-03-11T09:17:18.098Z" } } - })"), Milliseconds(0))); + })"), + Milliseconds(0))); - Date_t ping(dateFromISOString("2014-03-11T09:17:18.098Z").getValue()); - auto status = catalog.ping("abcd", ping); - ASSERT_OK(status); - } + Date_t ping(dateFromISOString("2014-03-11T09:17:18.098Z").getValue()); + auto status = catalog.ping("abcd", ping); + ASSERT_OK(status); +} - TEST(DistLockCatalogImpl, PingTargetError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, PingTargetError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); + targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - auto status = catalog.ping("abcd", Date_t::now()); - ASSERT_NOT_OK(status); - } + auto status = catalog.ping("abcd", Date_t::now()); + ASSERT_NOT_OK(status); +} - TEST(DistLockCatalogImpl, PingRunnerError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, PingRunnerError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); + executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); - auto status = catalog.ping("abcd", Date_t::now()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.ping("abcd", Date_t::now()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, PingCommandError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, PingCommandError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, errmsg: "bad", code: 9 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.ping("abcd", Date_t::now()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.ping("abcd", Date_t::now()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, PingWriteError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, PingWriteError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, code: 11000, errmsg: "E11000 duplicate key error" })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.ping("abcd", Date_t::now()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::DuplicateKey, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.ping("abcd", Date_t::now()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::DuplicateKey, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, PingWriteConcernError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, PingWriteConcernError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -174,23 +176,23 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.ping("abcd", Date_t::now()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.ping("abcd", Date_t::now()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, PingUnsupportedWriteConcernResponse) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, PingUnsupportedWriteConcernResponse) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -198,45 +200,47 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.ping("abcd", Date_t::now()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.ping("abcd", Date_t::now()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, PingUnsupportedResponseFormat) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, PingUnsupportedResponseFormat) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, - RemoteCommandResponse(BSON("ok" << 1 << "value" << "NaN"), Milliseconds(0))); + executor.setNextExpectedCommand(noTest, + RemoteCommandResponse(BSON("ok" << 1 << "value" + << "NaN"), + Milliseconds(0))); - auto status = catalog.ping("abcd", Date_t::now()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - } + auto status = catalog.ping("abcd", Date_t::now()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); +} - TEST(DistLockCatalogImpl, GrabLockNoOp) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockNoOp) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj responseObj(fromjson("{ ok: 1, value: null }")); + BSONObj responseObj(fromjson("{ ok: 1, value: null }")); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { - ASSERT_EQUALS(dummyHost, request.target); - ASSERT_EQUALS("config", request.dbname); + executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + ASSERT_EQUALS(dummyHost, request.target); + ASSERT_EQUALS("config", request.dbname); - BSONObj expectedCmd(fromjson(R"({ + BSONObj expectedCmd(fromjson(R"({ findAndModify: "locks", query: { _id: "test", state: 0 }, update: { @@ -254,29 +258,26 @@ namespace { writeConcern: { w: "majority", j: true, wtimeout: 100 } })")); - ASSERT_EQUALS(expectedCmd, request.cmdObj); - }, - RemoteCommandResponse(responseObj, Milliseconds(0))); - - OID myID("555f80be366c194b13fb0372"); - Date_t now(dateFromISOString("2015-05-22T19:17:18.098Z").getValue()); - auto resultStatus = catalog.grabLock("test", myID, "me", - "mongos", now, "because").getStatus(); + ASSERT_EQUALS(expectedCmd, request.cmdObj); + }, RemoteCommandResponse(responseObj, Milliseconds(0))); - ASSERT_NOT_OK(resultStatus); - ASSERT_EQUALS(ErrorCodes::LockStateChangeFailed, resultStatus.code()); + OID myID("555f80be366c194b13fb0372"); + Date_t now(dateFromISOString("2015-05-22T19:17:18.098Z").getValue()); + auto resultStatus = catalog.grabLock("test", myID, "me", "mongos", now, "because").getStatus(); - } + ASSERT_NOT_OK(resultStatus); + ASSERT_EQUALS(ErrorCodes::LockStateChangeFailed, resultStatus.code()); +} - TEST(DistLockCatalogImpl, GrabLockWithNewDoc) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockWithNewDoc) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj responseObj(fromjson(R"({ + BSONObj responseObj(fromjson(R"({ lastErrorObject: { updatedExisting: false, n: 1, @@ -294,11 +295,11 @@ namespace { ok: 1 })")); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { - ASSERT_EQUALS(dummyHost, request.target); - ASSERT_EQUALS("config", request.dbname); + executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + ASSERT_EQUALS(dummyHost, request.target); + ASSERT_EQUALS("config", request.dbname); - BSONObj expectedCmd(fromjson(R"({ + BSONObj expectedCmd(fromjson(R"({ findAndModify: "locks", query: { _id: "test", state: 0 }, update: { @@ -316,103 +317,102 @@ namespace { writeConcern: { w: "majority", j: true, wtimeout: 100 } })")); - ASSERT_EQUALS(expectedCmd, request.cmdObj); - }, - RemoteCommandResponse(responseObj, Milliseconds(0))); + ASSERT_EQUALS(expectedCmd, request.cmdObj); + }, RemoteCommandResponse(responseObj, Milliseconds(0))); - OID myID("555f80be366c194b13fb0372"); - Date_t now(dateFromISOString("2015-05-22T19:17:18.098Z").getValue()); - auto resultStatus = catalog.grabLock("test", myID, "me", "mongos", now, "because"); - ASSERT_OK(resultStatus.getStatus()); + OID myID("555f80be366c194b13fb0372"); + Date_t now(dateFromISOString("2015-05-22T19:17:18.098Z").getValue()); + auto resultStatus = catalog.grabLock("test", myID, "me", "mongos", now, "because"); + ASSERT_OK(resultStatus.getStatus()); - const auto& lockDoc = resultStatus.getValue(); - ASSERT_TRUE(lockDoc.isValid(nullptr)); - ASSERT_EQUALS("test", lockDoc.getName()); - ASSERT_EQUALS(myID, lockDoc.getLockID()); - ASSERT_EQUALS("me", lockDoc.getWho()); - ASSERT_EQUALS("mongos", lockDoc.getProcess()); - ASSERT_EQUALS("because", lockDoc.getWhy()); - } + const auto& lockDoc = resultStatus.getValue(); + ASSERT_TRUE(lockDoc.isValid(nullptr)); + ASSERT_EQUALS("test", lockDoc.getName()); + ASSERT_EQUALS(myID, lockDoc.getLockID()); + ASSERT_EQUALS("me", lockDoc.getWho()); + ASSERT_EQUALS("mongos", lockDoc.getProcess()); + ASSERT_EQUALS("because", lockDoc.getWhy()); +} - TEST(DistLockCatalogImpl, GrabLockTargetError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockTargetError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); + targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - } + auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); +} - TEST(DistLockCatalogImpl, GrabLockRunnerError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockRunnerError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); + executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); - auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GrabLockCommandError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockCommandError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, errmsg: "bad", code: 9 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GrabLockWriteError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockWriteError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, code: 11000, errmsg: "E11000 duplicate key error" })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::DuplicateKey, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::DuplicateKey, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GrabLockWriteConcernError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockWriteConcernError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -420,23 +420,23 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GrabLockUnsupportedWriteConcernResponse) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockUnsupportedWriteConcernResponse) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -444,46 +444,47 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, - RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GrabLockUnsupportedResponseFormat) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GrabLockUnsupportedResponseFormat) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, - RemoteCommandResponse(BSON("ok" << 1 << "value" << "NaN"), Milliseconds(0))); + executor.setNextExpectedCommand(noTest, + RemoteCommandResponse(BSON("ok" << 1 << "value" + << "NaN"), + Milliseconds(0))); - auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - } + auto status = catalog.grabLock("", OID::gen(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); +} - TEST(DistLockCatalogImpl, OvertakeLockNoOp) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockNoOp) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj responseObj(fromjson("{ ok: 1, value: null }")); + BSONObj responseObj(fromjson("{ ok: 1, value: null }")); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { - ASSERT_EQUALS(dummyHost, request.target); - ASSERT_EQUALS("config", request.dbname); + executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + ASSERT_EQUALS(dummyHost, request.target); + ASSERT_EQUALS("config", request.dbname); - BSONObj expectedCmd(fromjson(R"({ + BSONObj expectedCmd(fromjson(R"({ findAndModify: "locks", query: { $or: [ @@ -505,29 +506,29 @@ namespace { writeConcern: { w: "majority", j: true, wtimeout: 100 } })")); - ASSERT_EQUALS(expectedCmd, request.cmdObj); - }, - RemoteCommandResponse(responseObj, Milliseconds(0))); + ASSERT_EQUALS(expectedCmd, request.cmdObj); + }, RemoteCommandResponse(responseObj, Milliseconds(0))); - OID myID("555f80be366c194b13fb0372"); - OID currentOwner("555f99712c99a78c5b083358"); - Date_t now(dateFromISOString("2015-05-22T19:17:18.098Z").getValue()); - auto resultStatus = catalog.overtakeLock("test", myID, currentOwner, - "me", "mongos", now, "because").getStatus(); + OID myID("555f80be366c194b13fb0372"); + OID currentOwner("555f99712c99a78c5b083358"); + Date_t now(dateFromISOString("2015-05-22T19:17:18.098Z").getValue()); + auto resultStatus = + catalog.overtakeLock("test", myID, currentOwner, "me", "mongos", now, "because") + .getStatus(); - ASSERT_NOT_OK(resultStatus); - ASSERT_EQUALS(ErrorCodes::LockStateChangeFailed, resultStatus.code()); - } + ASSERT_NOT_OK(resultStatus); + ASSERT_EQUALS(ErrorCodes::LockStateChangeFailed, resultStatus.code()); +} - TEST(DistLockCatalogImpl, OvertakeLockWithNewDoc) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockWithNewDoc) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj responseObj(fromjson(R"({ + BSONObj responseObj(fromjson(R"({ lastErrorObject: { updatedExisting: false, n: 1, @@ -545,11 +546,11 @@ namespace { ok: 1 })")); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { - ASSERT_EQUALS(dummyHost, request.target); - ASSERT_EQUALS("config", request.dbname); + executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + ASSERT_EQUALS(dummyHost, request.target); + ASSERT_EQUALS("config", request.dbname); - BSONObj expectedCmd(fromjson(R"({ + BSONObj expectedCmd(fromjson(R"({ findAndModify: "locks", query: { $or: [ @@ -571,109 +572,104 @@ namespace { writeConcern: { w: "majority", j: true, wtimeout: 100 } })")); - ASSERT_EQUALS(expectedCmd, request.cmdObj); - }, - RemoteCommandResponse(responseObj, Milliseconds(0))); + ASSERT_EQUALS(expectedCmd, request.cmdObj); + }, RemoteCommandResponse(responseObj, Milliseconds(0))); - OID myID("555f80be366c194b13fb0372"); - OID currentOwner("555f99712c99a78c5b083358"); - Date_t now(dateFromISOString("2015-05-22T19:17:18.098Z").getValue()); - auto resultStatus = catalog.overtakeLock("test", myID, currentOwner, - "me", "mongos", now, "because"); - ASSERT_OK(resultStatus.getStatus()); + OID myID("555f80be366c194b13fb0372"); + OID currentOwner("555f99712c99a78c5b083358"); + Date_t now(dateFromISOString("2015-05-22T19:17:18.098Z").getValue()); + auto resultStatus = + catalog.overtakeLock("test", myID, currentOwner, "me", "mongos", now, "because"); + ASSERT_OK(resultStatus.getStatus()); - const auto& lockDoc = resultStatus.getValue(); - ASSERT_TRUE(lockDoc.isValid(nullptr)); - ASSERT_EQUALS("test", lockDoc.getName()); - ASSERT_EQUALS(myID, lockDoc.getLockID()); - ASSERT_EQUALS("me", lockDoc.getWho()); - ASSERT_EQUALS("mongos", lockDoc.getProcess()); - ASSERT_EQUALS("because", lockDoc.getWhy()); - } + const auto& lockDoc = resultStatus.getValue(); + ASSERT_TRUE(lockDoc.isValid(nullptr)); + ASSERT_EQUALS("test", lockDoc.getName()); + ASSERT_EQUALS(myID, lockDoc.getLockID()); + ASSERT_EQUALS("me", lockDoc.getWho()); + ASSERT_EQUALS("mongos", lockDoc.getProcess()); + ASSERT_EQUALS("because", lockDoc.getWhy()); +} - TEST(DistLockCatalogImpl, OvertakeLockTargetError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockTargetError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); + targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - auto status = catalog.overtakeLock( - "", OID(), OID(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - } + auto status = catalog.overtakeLock("", OID(), OID(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); +} - TEST(DistLockCatalogImpl, OvertakeLockRunnerError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockRunnerError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); + executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); - auto status = catalog.overtakeLock( - "", OID(), OID(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.overtakeLock("", OID(), OID(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, OvertakeLockCommandError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockCommandError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, errmsg: "bad", code: 9 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.overtakeLock( - "", OID(), OID(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.overtakeLock("", OID(), OID(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, OvertakeLockWriteError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockWriteError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, code: 11000, errmsg: "E11000 duplicate key error" })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.overtakeLock( - "", OID(), OID(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::DuplicateKey, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.overtakeLock("", OID(), OID(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::DuplicateKey, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, OvertakeLockWriteConcernError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockWriteConcernError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -681,24 +677,23 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.overtakeLock( - "", OID(), OID(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.overtakeLock("", OID(), OID(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, OvertakeLockUnsupportedWriteConcernResponse) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockUnsupportedWriteConcernResponse) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -706,42 +701,42 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, - RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.overtakeLock( - "", OID(), OID(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.overtakeLock("", OID(), OID(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, OvertakeLockUnsupportedResponseFormat) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, OvertakeLockUnsupportedResponseFormat) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, - RemoteCommandResponse(BSON("ok" << 1 << "value" << "NaN"), Milliseconds(0))); + executor.setNextExpectedCommand(noTest, + RemoteCommandResponse(BSON("ok" << 1 << "value" + << "NaN"), + Milliseconds(0))); - auto status = catalog.overtakeLock( - "", OID(), OID(), "", "", Date_t::now(), "").getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - } + auto status = catalog.overtakeLock("", OID(), OID(), "", "", Date_t::now(), "").getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); +} - TEST(DistLockCatalogImpl, BasicUnlock) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, BasicUnlock) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + executor.setNextExpectedCommand( + [](const RemoteCommandRequest& request) { ASSERT_EQUALS(dummyHost, request.target); ASSERT_EQUALS("config", request.dbname); @@ -761,21 +756,23 @@ namespace { ts: ObjectId("555f99712c99a78c5b083358"), state: 0 } - })"), Milliseconds(0))); + })"), + Milliseconds(0))); - auto status = catalog.unlock(OID("555f99712c99a78c5b083358")); - ASSERT_OK(status); - } + auto status = catalog.unlock(OID("555f99712c99a78c5b083358")); + ASSERT_OK(status); +} - TEST(DistLockCatalogImpl, UnlockWithNoNewDoc) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, UnlockWithNoNewDoc) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + executor.setNextExpectedCommand( + [](const RemoteCommandRequest& request) { ASSERT_EQUALS(dummyHost, request.target); ASSERT_EQUALS("config", request.dbname); @@ -791,91 +788,92 @@ namespace { RemoteCommandResponse(fromjson(R"({ ok: 1, value: null - })"), Milliseconds(0))); + })"), + Milliseconds(0))); - auto status = catalog.unlock(OID("555f99712c99a78c5b083358")); - ASSERT_OK(status); - } + auto status = catalog.unlock(OID("555f99712c99a78c5b083358")); + ASSERT_OK(status); +} - TEST(DistLockCatalogImpl, UnlockTargetError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, UnlockTargetError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); + targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - auto status = catalog.unlock(OID()); - ASSERT_NOT_OK(status); - } + auto status = catalog.unlock(OID()); + ASSERT_NOT_OK(status); +} - TEST(DistLockCatalogImpl, UnlockRunnerError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, UnlockRunnerError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); + executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); - auto status = catalog.unlock(OID()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.unlock(OID()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, UnlockCommandError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, UnlockCommandError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, errmsg: "bad", code: 9 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.unlock(OID()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.unlock(OID()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, UnlockWriteError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, UnlockWriteError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, code: 11000, errmsg: "E11000 duplicate key error" })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.unlock(OID()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::DuplicateKey, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.unlock(OID()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::DuplicateKey, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, UnlockWriteConcernError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, UnlockWriteConcernError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -883,23 +881,23 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.unlock(OID()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.unlock(OID()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, UnlockUnsupportedWriteConcernResponse) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, UnlockUnsupportedWriteConcernResponse) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -907,39 +905,41 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.unlock(OID()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.unlock(OID()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, UnlockUnsupportedResponseFormat) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, UnlockUnsupportedResponseFormat) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, - RemoteCommandResponse(BSON("ok" << 1 << "value" << "NaN"), Milliseconds(0))); + executor.setNextExpectedCommand(noTest, + RemoteCommandResponse(BSON("ok" << 1 << "value" + << "NaN"), + Milliseconds(0))); - auto status = catalog.unlock(OID()); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - } + auto status = catalog.unlock(OID()); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); +} - TEST(DistLockCatalogImpl, BasicGetServerInfo) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, BasicGetServerInfo) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj responseObj(fromjson(R"({ + BSONObj responseObj(fromjson(R"({ localTime: { $date: "2015-05-26T13:06:27.293Z" }, $gleStats: { lastOpTime: { $timestamp: { t: 0, i: 0 }}, @@ -948,81 +948,80 @@ namespace { ok: 1 })")); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { - ASSERT_EQUALS(dummyHost, request.target); - ASSERT_EQUALS("admin", request.dbname); - ASSERT_EQUALS(BSON("serverStatus" << 1), request.cmdObj); - }, - RemoteCommandResponse(responseObj, Milliseconds(0))); + executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + ASSERT_EQUALS(dummyHost, request.target); + ASSERT_EQUALS("admin", request.dbname); + ASSERT_EQUALS(BSON("serverStatus" << 1), request.cmdObj); + }, RemoteCommandResponse(responseObj, Milliseconds(0))); - Date_t localTime(dateFromISOString("2015-05-26T13:06:27.293Z").getValue()); - OID electionID("555fa85d4d8640862a0fc79b"); - auto resultStatus = catalog.getServerInfo(); - ASSERT_OK(resultStatus.getStatus()); + Date_t localTime(dateFromISOString("2015-05-26T13:06:27.293Z").getValue()); + OID electionID("555fa85d4d8640862a0fc79b"); + auto resultStatus = catalog.getServerInfo(); + ASSERT_OK(resultStatus.getStatus()); - const auto& serverInfo = resultStatus.getValue(); - ASSERT_EQUALS(electionID, serverInfo.electionId); - ASSERT_EQUALS(localTime, serverInfo.serverTime); - } + const auto& serverInfo = resultStatus.getValue(); + ASSERT_EQUALS(electionID, serverInfo.electionId); + ASSERT_EQUALS(localTime, serverInfo.serverTime); +} - TEST(DistLockCatalogImpl, GetServerTargetError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GetServerTargetError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); + targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - auto status = catalog.getServerInfo().getStatus(); - ASSERT_NOT_OK(status); - } + auto status = catalog.getServerInfo().getStatus(); + ASSERT_NOT_OK(status); +} - TEST(DistLockCatalogImpl, GetServerRunnerError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GetServerRunnerError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); + executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); - auto status = catalog.getServerInfo().getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.getServerInfo().getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GetServerCommandError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GetServerCommandError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, errmsg: "bad", code: 9 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.getServerInfo().getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.getServerInfo().getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GetServerBadElectionId) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GetServerBadElectionId) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ localTime: { $date: "2015-05-26T13:06:27.293Z" }, $gleStats: { lastOpTime: { $timestamp: { t: 0, i: 0 }}, @@ -1030,23 +1029,23 @@ namespace { }, ok: 1 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.getServerInfo().getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.getServerInfo().getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GetServerBadLocalTime) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GetServerBadLocalTime) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ localTime: "2015-05-26T13:06:27.293Z", $gleStats: { lastOpTime: { $timestamp: { t: 0, i: 0 }}, @@ -1054,43 +1053,43 @@ namespace { }, ok: 1 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.getServerInfo().getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.getServerInfo().getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GetServerNoGLEStats) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GetServerNoGLEStats) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ localTime: { $date: "2015-05-26T13:06:27.293Z" }, ok: 1 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.getServerInfo().getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.getServerInfo().getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, GetServerNoElectionId) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, GetServerNoElectionId) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ localTime: { $date: "2015-05-26T13:06:27.293Z" }, $gleStats: { lastOpTime: { $timestamp: { t: 0, i: 0 }}, @@ -1098,23 +1097,24 @@ namespace { }, ok: 1 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.getServerInfo().getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.getServerInfo().getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, BasicStopPing) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, BasicStopPing) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand([](const RemoteCommandRequest& request) { + executor.setNextExpectedCommand( + [](const RemoteCommandRequest& request) { ASSERT_EQUALS(dummyHost, request.target); ASSERT_EQUALS("config", request.dbname); @@ -1133,91 +1133,92 @@ namespace { _id: "test", ping: { $date: "2014-03-11T09:17:18.098Z" } } - })"), Milliseconds(0))); + })"), + Milliseconds(0))); - auto status = catalog.stopPing("test"); - ASSERT_OK(status); - } + auto status = catalog.stopPing("test"); + ASSERT_OK(status); +} - TEST(DistLockCatalogImpl, StopPingTargetError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, StopPingTargetError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); + targeter.setFindHostReturnValue({ErrorCodes::InternalError, "can't target"}); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - auto status = catalog.stopPing(""); - ASSERT_NOT_OK(status); - } + auto status = catalog.stopPing(""); + ASSERT_NOT_OK(status); +} - TEST(DistLockCatalogImpl, StopPingRunnerError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, StopPingRunnerError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); + executor.setNextExpectedCommand(noTest, {ErrorCodes::InternalError, "Bad"}); - auto status = catalog.stopPing(""); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.stopPing(""); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::InternalError, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, StopPingCommandError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, StopPingCommandError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, errmsg: "bad", code: 9 })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.stopPing(""); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.stopPing(""); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::FailedToParse, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, StopPingWriteError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, StopPingWriteError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 0, code: 13, errmsg: "Unauthorized" })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.stopPing(""); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::Unauthorized, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.stopPing(""); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::Unauthorized, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, StopPingWriteConcernError) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, StopPingWriteConcernError) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -1225,23 +1226,23 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.stopPing(""); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.stopPing(""); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::WriteConcernFailed, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, StopPingUnsupportedWriteConcernResponse) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, StopPingUnsupportedWriteConcernResponse) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - BSONObj returnObj(fromjson(R"({ + BSONObj returnObj(fromjson(R"({ ok: 1, value: null, writeConcernError: { @@ -1249,29 +1250,31 @@ namespace { errmsg: "waiting for replication timed out" } })")); - executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); + executor.setNextExpectedCommand(noTest, RemoteCommandResponse(returnObj, Milliseconds(0))); - auto status = catalog.stopPing(""); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - ASSERT_FALSE(status.reason().empty()); - } + auto status = catalog.stopPing(""); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_FALSE(status.reason().empty()); +} - TEST(DistLockCatalogImpl, StopPingUnsupportedResponseFormat) { - RemoteCommandTargeterMock targeter; - RemoteCommandRunnerMock executor; +TEST(DistLockCatalogImpl, StopPingUnsupportedResponseFormat) { + RemoteCommandTargeterMock targeter; + RemoteCommandRunnerMock executor; - targeter.setFindHostReturnValue(dummyHost); + targeter.setFindHostReturnValue(dummyHost); - DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); + DistLockCatalogImpl catalog(&targeter, &executor, kWTimeout); - executor.setNextExpectedCommand(noTest, - RemoteCommandResponse(BSON("ok" << 1 << "value" << "NaN"), Milliseconds(0))); + executor.setNextExpectedCommand(noTest, + RemoteCommandResponse(BSON("ok" << 1 << "value" + << "NaN"), + Milliseconds(0))); - auto status = catalog.stopPing(""); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); - } + auto status = catalog.stopPing(""); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); +} -} // unnamed namespace -} // namespace mongo +} // unnamed namespace +} // namespace mongo diff --git a/src/mongo/s/catalog/dist_lock_catalog_mock.cpp b/src/mongo/s/catalog/dist_lock_catalog_mock.cpp index 9b8fc3f8f5a..260d4dce6f3 100644 --- a/src/mongo/s/catalog/dist_lock_catalog_mock.cpp +++ b/src/mongo/s/catalog/dist_lock_catalog_mock.cpp @@ -40,302 +40,292 @@ namespace mongo { namespace { - Status kBadRetValue(ErrorCodes::InternalError, "no return value"); - StatusWith<LocksType> kLocksTypeBadRetValue(kBadRetValue); - StatusWith<LockpingsType> kLockpingsTypeBadRetValue(kBadRetValue); - StatusWith<DistLockCatalog::ServerInfo> kServerInfoBadRetValue(kBadRetValue); +Status kBadRetValue(ErrorCodes::InternalError, "no return value"); +StatusWith<LocksType> kLocksTypeBadRetValue(kBadRetValue); +StatusWith<LockpingsType> kLockpingsTypeBadRetValue(kBadRetValue); +StatusWith<DistLockCatalog::ServerInfo> kServerInfoBadRetValue(kBadRetValue); + +void noGrabLockFuncSet(StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { + FAIL(str::stream() << "grabLock not expected to be called. " + << "lockID: " << lockID << ", who: " << who << ", processId: " << processId + << ", why: " << why); +} - void noGrabLockFuncSet(StringData lockID, +void noOvertakeLockFuncSet(StringData lockID, const OID& lockSessionID, + const OID& currentHolderTS, StringData who, StringData processId, Date_t time, StringData why) { - FAIL(str::stream() << "grabLock not expected to be called. " - << "lockID: " << lockID - << ", who: " << who - << ", processId: " << processId - << ", why: " << why); - } - - void noOvertakeLockFuncSet(StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { - FAIL(str::stream() << "overtakeLock not expected to be called. " - << "lockID: " << lockID - << ", currentHolderTS: " << currentHolderTS - << ", who: " << who - << ", processId: " << processId - << ", why: " << why); - } - - void noUnLockFuncSet(const OID& lockSessionID) { - FAIL(str::stream() << "unlock not expected to be called. " - << "lockSessionID: " << lockSessionID); - } + FAIL(str::stream() << "overtakeLock not expected to be called. " + << "lockID: " << lockID << ", currentHolderTS: " << currentHolderTS + << ", who: " << who << ", processId: " << processId << ", why: " << why); +} - void noPingFuncSet(StringData processID, Date_t ping) { - // Ping is expected to be called all the time, so default behavior is do nothing. - } +void noUnLockFuncSet(const OID& lockSessionID) { + FAIL(str::stream() << "unlock not expected to be called. " + << "lockSessionID: " << lockSessionID); +} - void noStopPingFuncSet(StringData processID) { - FAIL(str::stream() << "stopPing not expected to be called. " - << "processID: " << processID); - } +void noPingFuncSet(StringData processID, Date_t ping) { + // Ping is expected to be called all the time, so default behavior is do nothing. +} - void noGetLockByTSSet(const OID& lockSessionID) { - FAIL(str::stream() << "getLockByTS not expected to be called. " - << "lockSessionID: " << lockSessionID); - } +void noStopPingFuncSet(StringData processID) { + FAIL(str::stream() << "stopPing not expected to be called. " + << "processID: " << processID); +} - void noGetLockByNameSet(StringData name) { - FAIL(str::stream() << "getLockByName not expected to be called. " - << "lockName: " << name); - } +void noGetLockByTSSet(const OID& lockSessionID) { + FAIL(str::stream() << "getLockByTS not expected to be called. " + << "lockSessionID: " << lockSessionID); +} - void noGetPingSet(StringData processId) { - FAIL(str::stream() << "getPing not expected to be called. " - << "lockName: " << processId); - } +void noGetLockByNameSet(StringData name) { + FAIL(str::stream() << "getLockByName not expected to be called. " + << "lockName: " << name); +} - void noGetServerInfoSet() { - FAIL("getServerInfo not expected to be called"); - } +void noGetPingSet(StringData processId) { + FAIL(str::stream() << "getPing not expected to be called. " + << "lockName: " << processId); +} -} // unnamed namespace - - DistLockCatalogMock::DistLockCatalogMock(): - _grabLockChecker(noGrabLockFuncSet), - _grabLockReturnValue(kLocksTypeBadRetValue), - _unlockChecker(noUnLockFuncSet), - _unlockReturnValue(kBadRetValue), - _pingChecker(noPingFuncSet), - _pingReturnValue(kBadRetValue), - _stopPingChecker(noStopPingFuncSet), - _stopPingReturnValue(kBadRetValue), - _getLockByTSChecker(noGetLockByTSSet), - _getLockByTSReturnValue(kLocksTypeBadRetValue), - _getLockByNameChecker(noGetLockByNameSet), - _getLockByNameReturnValue(kLocksTypeBadRetValue), - _overtakeLockChecker(noOvertakeLockFuncSet), - _overtakeLockReturnValue(kLocksTypeBadRetValue), - _getPingChecker(noGetPingSet), - _getPingReturnValue(kLockpingsTypeBadRetValue), - _getServerInfoChecker(noGetServerInfoSet), - _getServerInfoReturnValue(kServerInfoBadRetValue) { - } +void noGetServerInfoSet() { + FAIL("getServerInfo not expected to be called"); +} - DistLockCatalogMock::~DistLockCatalogMock() { +} // unnamed namespace + +DistLockCatalogMock::DistLockCatalogMock() + : _grabLockChecker(noGrabLockFuncSet), + _grabLockReturnValue(kLocksTypeBadRetValue), + _unlockChecker(noUnLockFuncSet), + _unlockReturnValue(kBadRetValue), + _pingChecker(noPingFuncSet), + _pingReturnValue(kBadRetValue), + _stopPingChecker(noStopPingFuncSet), + _stopPingReturnValue(kBadRetValue), + _getLockByTSChecker(noGetLockByTSSet), + _getLockByTSReturnValue(kLocksTypeBadRetValue), + _getLockByNameChecker(noGetLockByNameSet), + _getLockByNameReturnValue(kLocksTypeBadRetValue), + _overtakeLockChecker(noOvertakeLockFuncSet), + _overtakeLockReturnValue(kLocksTypeBadRetValue), + _getPingChecker(noGetPingSet), + _getPingReturnValue(kLockpingsTypeBadRetValue), + _getServerInfoChecker(noGetServerInfoSet), + _getServerInfoReturnValue(kServerInfoBadRetValue) {} + +DistLockCatalogMock::~DistLockCatalogMock() {} + +StatusWith<LockpingsType> DistLockCatalogMock::getPing(StringData processID) { + auto ret = kLockpingsTypeBadRetValue; + GetPingFunc checkerFunc = noGetPingSet; + + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + ret = _getPingReturnValue; + checkerFunc = _getPingChecker; } - StatusWith<LockpingsType> DistLockCatalogMock::getPing(StringData processID) { - auto ret = kLockpingsTypeBadRetValue; - GetPingFunc checkerFunc = noGetPingSet; + checkerFunc(processID); + return ret; +} - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _getPingReturnValue; - checkerFunc = _getPingChecker; - } +Status DistLockCatalogMock::ping(StringData processID, Date_t ping) { + auto ret = kBadRetValue; + PingFunc checkerFunc = noPingFuncSet; - checkerFunc(processID); - return ret; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + ret = _pingReturnValue; + checkerFunc = _pingChecker; } - Status DistLockCatalogMock::ping(StringData processID, Date_t ping) { - auto ret = kBadRetValue; - PingFunc checkerFunc = noPingFuncSet; + checkerFunc(processID, ping); + return ret; +} - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _pingReturnValue; - checkerFunc = _pingChecker; - } +StatusWith<LocksType> DistLockCatalogMock::grabLock(StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { + auto ret = kLocksTypeBadRetValue; + GrabLockFunc checkerFunc = noGrabLockFuncSet; - checkerFunc(processID, ping); - return ret; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + ret = _grabLockReturnValue; + checkerFunc = _grabLockChecker; } - StatusWith<LocksType> DistLockCatalogMock::grabLock(StringData lockID, + checkerFunc(lockID, lockSessionID, who, processId, time, why); + return ret; +} + +StatusWith<LocksType> DistLockCatalogMock::overtakeLock(StringData lockID, const OID& lockSessionID, + const OID& currentHolderTS, StringData who, StringData processId, Date_t time, StringData why) { - auto ret = kLocksTypeBadRetValue; - GrabLockFunc checkerFunc = noGrabLockFuncSet; - - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _grabLockReturnValue; - checkerFunc = _grabLockChecker; - } + auto ret = kLocksTypeBadRetValue; + OvertakeLockFunc checkerFunc = noOvertakeLockFuncSet; - checkerFunc(lockID, lockSessionID, who, processId, time, why); - return ret; - } - - StatusWith<LocksType> DistLockCatalogMock::overtakeLock(StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { - auto ret = kLocksTypeBadRetValue; - OvertakeLockFunc checkerFunc = noOvertakeLockFuncSet; - - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _overtakeLockReturnValue; - checkerFunc = _overtakeLockChecker; - } - - checkerFunc(lockID, lockSessionID, currentHolderTS, who, processId, time, why); - return ret; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + ret = _overtakeLockReturnValue; + checkerFunc = _overtakeLockChecker; } - Status DistLockCatalogMock::unlock(const OID& lockSessionID) { - auto ret = kBadRetValue; - UnlockFunc checkerFunc = noUnLockFuncSet; + checkerFunc(lockID, lockSessionID, currentHolderTS, who, processId, time, why); + return ret; +} - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _unlockReturnValue; - checkerFunc = _unlockChecker; - } +Status DistLockCatalogMock::unlock(const OID& lockSessionID) { + auto ret = kBadRetValue; + UnlockFunc checkerFunc = noUnLockFuncSet; - checkerFunc(lockSessionID); - return ret; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + ret = _unlockReturnValue; + checkerFunc = _unlockChecker; } - StatusWith<DistLockCatalog::ServerInfo> DistLockCatalogMock::getServerInfo() { - auto ret = kServerInfoBadRetValue; - GetServerInfoFunc checkerFunc = noGetServerInfoSet; + checkerFunc(lockSessionID); + return ret; +} - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _getServerInfoReturnValue; - checkerFunc = _getServerInfoChecker; - } +StatusWith<DistLockCatalog::ServerInfo> DistLockCatalogMock::getServerInfo() { + auto ret = kServerInfoBadRetValue; + GetServerInfoFunc checkerFunc = noGetServerInfoSet; - checkerFunc(); - return ret; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + ret = _getServerInfoReturnValue; + checkerFunc = _getServerInfoChecker; } - StatusWith<LocksType> DistLockCatalogMock::getLockByTS(const OID& lockSessionID) { - auto ret = kLocksTypeBadRetValue; - GetLockByTSFunc checkerFunc = noGetLockByTSSet; + checkerFunc(); + return ret; +} - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _getLockByTSReturnValue; - checkerFunc = _getLockByTSChecker; - } +StatusWith<LocksType> DistLockCatalogMock::getLockByTS(const OID& lockSessionID) { + auto ret = kLocksTypeBadRetValue; + GetLockByTSFunc checkerFunc = noGetLockByTSSet; - checkerFunc(lockSessionID); - return ret; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + ret = _getLockByTSReturnValue; + checkerFunc = _getLockByTSChecker; } - StatusWith<LocksType> DistLockCatalogMock::getLockByName(StringData name) { - auto ret = kLocksTypeBadRetValue; - GetLockByNameFunc checkerFunc = noGetLockByNameSet; + checkerFunc(lockSessionID); + return ret; +} - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _getLockByNameReturnValue; - checkerFunc = _getLockByNameChecker; - } +StatusWith<LocksType> DistLockCatalogMock::getLockByName(StringData name) { + auto ret = kLocksTypeBadRetValue; + GetLockByNameFunc checkerFunc = noGetLockByNameSet; - checkerFunc(name); - return ret; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + ret = _getLockByNameReturnValue; + checkerFunc = _getLockByNameChecker; } - Status DistLockCatalogMock::stopPing(StringData processId) { - auto ret = kBadRetValue; - StopPingFunc checkerFunc = noStopPingFuncSet; - - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - ret = _stopPingReturnValue; - checkerFunc = _stopPingChecker; - } + checkerFunc(name); + return ret; +} - checkerFunc(processId); - return ret; - } +Status DistLockCatalogMock::stopPing(StringData processId) { + auto ret = kBadRetValue; + StopPingFunc checkerFunc = noStopPingFuncSet; - void DistLockCatalogMock::expectGrabLock(DistLockCatalogMock::GrabLockFunc checkerFunc, - StatusWith<LocksType> returnThis) { + { stdx::lock_guard<stdx::mutex> lk(_mutex); - _grabLockChecker = checkerFunc; - _grabLockReturnValue = returnThis; + ret = _stopPingReturnValue; + checkerFunc = _stopPingChecker; } - void DistLockCatalogMock::expectNoGrabLock() { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _grabLockChecker = noGrabLockFuncSet; - _grabLockReturnValue = kLocksTypeBadRetValue; - } + checkerFunc(processId); + return ret; +} - void DistLockCatalogMock::expectUnLock(DistLockCatalogMock::UnlockFunc checkerFunc, - Status returnThis) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _unlockChecker = checkerFunc; - _unlockReturnValue = returnThis; - } +void DistLockCatalogMock::expectGrabLock(DistLockCatalogMock::GrabLockFunc checkerFunc, + StatusWith<LocksType> returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _grabLockChecker = checkerFunc; + _grabLockReturnValue = returnThis; +} - void DistLockCatalogMock::expectPing(DistLockCatalogMock::PingFunc checkerFunc, - Status returnThis) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _pingChecker = checkerFunc; - _pingReturnValue = returnThis; - } +void DistLockCatalogMock::expectNoGrabLock() { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _grabLockChecker = noGrabLockFuncSet; + _grabLockReturnValue = kLocksTypeBadRetValue; +} - void DistLockCatalogMock::expectStopPing(StopPingFunc checkerFunc, Status returnThis) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _stopPingChecker = checkerFunc; - _stopPingReturnValue = returnThis; - } +void DistLockCatalogMock::expectUnLock(DistLockCatalogMock::UnlockFunc checkerFunc, + Status returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _unlockChecker = checkerFunc; + _unlockReturnValue = returnThis; +} - void DistLockCatalogMock::expectGetLockByTS(GetLockByTSFunc checkerFunc, - StatusWith<LocksType> returnThis) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _getLockByTSChecker = checkerFunc; - _getLockByTSReturnValue = returnThis; - } +void DistLockCatalogMock::expectPing(DistLockCatalogMock::PingFunc checkerFunc, Status returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _pingChecker = checkerFunc; + _pingReturnValue = returnThis; +} - void DistLockCatalogMock::expectGetLockByName(GetLockByNameFunc checkerFunc, - StatusWith<LocksType> returnThis) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _getLockByNameChecker = checkerFunc; - _getLockByNameReturnValue = returnThis; - } +void DistLockCatalogMock::expectStopPing(StopPingFunc checkerFunc, Status returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _stopPingChecker = checkerFunc; + _stopPingReturnValue = returnThis; +} - void DistLockCatalogMock::expectOvertakeLock(OvertakeLockFunc checkerFunc, - StatusWith<LocksType> returnThis) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _overtakeLockChecker = checkerFunc; - _overtakeLockReturnValue = returnThis; - } +void DistLockCatalogMock::expectGetLockByTS(GetLockByTSFunc checkerFunc, + StatusWith<LocksType> returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _getLockByTSChecker = checkerFunc; + _getLockByTSReturnValue = returnThis; +} - void DistLockCatalogMock::expectGetPing(GetPingFunc checkerFunc, - StatusWith<LockpingsType> returnThis) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _getPingChecker = checkerFunc; - _getPingReturnValue = returnThis; - } +void DistLockCatalogMock::expectGetLockByName(GetLockByNameFunc checkerFunc, + StatusWith<LocksType> returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _getLockByNameChecker = checkerFunc; + _getLockByNameReturnValue = returnThis; +} - void DistLockCatalogMock::expectGetServerInfo( - GetServerInfoFunc checkerFunc, - StatusWith<DistLockCatalog::ServerInfo> returnThis) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _getServerInfoChecker = checkerFunc; - _getServerInfoReturnValue = returnThis; - } +void DistLockCatalogMock::expectOvertakeLock(OvertakeLockFunc checkerFunc, + StatusWith<LocksType> returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _overtakeLockChecker = checkerFunc; + _overtakeLockReturnValue = returnThis; +} + +void DistLockCatalogMock::expectGetPing(GetPingFunc checkerFunc, + StatusWith<LockpingsType> returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _getPingChecker = checkerFunc; + _getPingReturnValue = returnThis; +} +void DistLockCatalogMock::expectGetServerInfo(GetServerInfoFunc checkerFunc, + StatusWith<DistLockCatalog::ServerInfo> returnThis) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _getServerInfoChecker = checkerFunc; + _getServerInfoReturnValue = returnThis; +} } diff --git a/src/mongo/s/catalog/dist_lock_catalog_mock.h b/src/mongo/s/catalog/dist_lock_catalog_mock.h index 1b8282f4704..6a8ffa01f33 100644 --- a/src/mongo/s/catalog/dist_lock_catalog_mock.h +++ b/src/mongo/s/catalog/dist_lock_catalog_mock.h @@ -37,174 +37,174 @@ namespace mongo { - /** - * Mock implementation of DistLockCatalog for testing. - * - * Example usage: - * - * DistLockCatalogMock mock; - * LocksType badLock; - * mock.expectGrabLock([](StringData lockID, - * const OID& lockSessionID, - * StringData who, - * StringData processId, - * Date_t time, - * StringData why) { - * ASSERT_EQUALS("test", lockID); - * }, badLock); - * - * mock.grabLock("test", OID(), "me", "x", Date_t::now(), "end"); - * - * It is also possible to chain the callbacks. For example, if we want to set the test - * such that grabLock can only be called once, you can do this: - * - * DistLockCatalogMock mock; - * mock.expectGrabLock([&mock](...) { - * mock.expectNoGrabLock(); - * }, Status::OK()); - */ - class DistLockCatalogMock : public DistLockCatalog { - public: - DistLockCatalogMock(); - virtual ~DistLockCatalogMock(); - - using GrabLockFunc = stdx::function<void (StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why)>; - using OvertakeLockFunc = stdx::function<void (StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why)>; - using UnlockFunc = stdx::function<void (const OID& lockSessionID)>; - using PingFunc = stdx::function<void (StringData processID, Date_t ping)>; - using StopPingFunc = stdx::function<void (StringData processID)>; - using GetPingFunc = StopPingFunc; - using GetLockByTSFunc = stdx::function<void (const OID& ts)>; - using GetLockByNameFunc = stdx::function<void (StringData name)>; - using GetServerInfoFunc = stdx::function<void ()>; - - virtual StatusWith<LockpingsType> getPing(StringData processID) override; - - virtual Status ping(StringData processID, Date_t ping) override; - - virtual StatusWith<LocksType> grabLock(StringData lockID, +/** + * Mock implementation of DistLockCatalog for testing. + * + * Example usage: + * + * DistLockCatalogMock mock; + * LocksType badLock; + * mock.expectGrabLock([](StringData lockID, + * const OID& lockSessionID, + * StringData who, + * StringData processId, + * Date_t time, + * StringData why) { + * ASSERT_EQUALS("test", lockID); + * }, badLock); + * + * mock.grabLock("test", OID(), "me", "x", Date_t::now(), "end"); + * + * It is also possible to chain the callbacks. For example, if we want to set the test + * such that grabLock can only be called once, you can do this: + * + * DistLockCatalogMock mock; + * mock.expectGrabLock([&mock](...) { + * mock.expectNoGrabLock(); + * }, Status::OK()); + */ +class DistLockCatalogMock : public DistLockCatalog { +public: + DistLockCatalogMock(); + virtual ~DistLockCatalogMock(); + + using GrabLockFunc = stdx::function<void(StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why)>; + using OvertakeLockFunc = stdx::function<void(StringData lockID, + const OID& lockSessionID, + const OID& currentHolderTS, + StringData who, + StringData processId, + Date_t time, + StringData why)>; + using UnlockFunc = stdx::function<void(const OID& lockSessionID)>; + using PingFunc = stdx::function<void(StringData processID, Date_t ping)>; + using StopPingFunc = stdx::function<void(StringData processID)>; + using GetPingFunc = StopPingFunc; + using GetLockByTSFunc = stdx::function<void(const OID& ts)>; + using GetLockByNameFunc = stdx::function<void(StringData name)>; + using GetServerInfoFunc = stdx::function<void()>; + + virtual StatusWith<LockpingsType> getPing(StringData processID) override; + + virtual Status ping(StringData processID, Date_t ping) override; + + virtual StatusWith<LocksType> grabLock(StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) override; + + virtual StatusWith<LocksType> overtakeLock(StringData lockID, const OID& lockSessionID, + const OID& currentHolderTS, StringData who, StringData processId, Date_t time, StringData why) override; - virtual StatusWith<LocksType> overtakeLock(StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) override; - - virtual Status unlock(const OID& lockSessionID) override; - - virtual StatusWith<ServerInfo> getServerInfo() override; - - virtual StatusWith<LocksType> getLockByTS(const OID& lockSessionID) override; - - virtual StatusWith<LocksType> getLockByName(StringData name) override; - - virtual Status stopPing(StringData processId) override; - - /** - * Sets the checker method to use and the return value for grabLock to return every - * time it is called. - */ - void expectGrabLock(GrabLockFunc checkerFunc, StatusWith<LocksType> returnThis); - - /** - * Expect grabLock to never be called after this is called. - */ - void expectNoGrabLock(); - - /** - * Sets the checker method to use and the return value for unlock to return every - * time it is called. - */ - void expectUnLock(UnlockFunc checkerFunc, Status returnThis); - - /** - * Sets the checker method to use and its return value the every time ping is called. - */ - void expectPing(PingFunc checkerFunc, Status returnThis); - - /** - * Sets the checker method to use and its return value the every time stopPing is called. - */ - void expectStopPing(StopPingFunc checkerFunc, Status returnThis); - - /** - * Sets the checker method to use and its return value the every time - * getLockByTS is called. - */ - void expectGetLockByTS(GetLockByTSFunc checkerFunc, StatusWith<LocksType> returnThis); - - /** - * Sets the checker method to use and its return value the every time - * getLockByName is called. - */ - void expectGetLockByName(GetLockByNameFunc checkerFunc, StatusWith<LocksType> returnThis); - - /** - * Sets the checker method to use and its return value the every time - * overtakeLock is called. - */ - void expectOvertakeLock(OvertakeLockFunc checkerFunc, StatusWith<LocksType> returnThis); - - /** - * Sets the checker method to use and its return value the every time - * getPing is called. - */ - void expectGetPing(GetPingFunc checkerFunc, StatusWith<LockpingsType> returnThis); - - /** - * Sets the checker method to use and its return value the every time - * getServerInfo is called. - */ - void expectGetServerInfo(GetServerInfoFunc checkerFunc, - StatusWith<DistLockCatalog::ServerInfo> returnThis); - - private: - // Protects all the member variables. - stdx::mutex _mutex; - - GrabLockFunc _grabLockChecker; - StatusWith<LocksType> _grabLockReturnValue; - - UnlockFunc _unlockChecker; - Status _unlockReturnValue; - - PingFunc _pingChecker; - Status _pingReturnValue; - - StopPingFunc _stopPingChecker; - Status _stopPingReturnValue; - - GetLockByTSFunc _getLockByTSChecker; - StatusWith<LocksType> _getLockByTSReturnValue; - - GetLockByNameFunc _getLockByNameChecker; - StatusWith<LocksType> _getLockByNameReturnValue; - - OvertakeLockFunc _overtakeLockChecker; - StatusWith<LocksType> _overtakeLockReturnValue; - - GetPingFunc _getPingChecker; - StatusWith<LockpingsType> _getPingReturnValue; - - GetServerInfoFunc _getServerInfoChecker; - StatusWith<DistLockCatalog::ServerInfo> _getServerInfoReturnValue; - }; + virtual Status unlock(const OID& lockSessionID) override; + + virtual StatusWith<ServerInfo> getServerInfo() override; + + virtual StatusWith<LocksType> getLockByTS(const OID& lockSessionID) override; + + virtual StatusWith<LocksType> getLockByName(StringData name) override; + + virtual Status stopPing(StringData processId) override; + + /** + * Sets the checker method to use and the return value for grabLock to return every + * time it is called. + */ + void expectGrabLock(GrabLockFunc checkerFunc, StatusWith<LocksType> returnThis); + + /** + * Expect grabLock to never be called after this is called. + */ + void expectNoGrabLock(); + + /** + * Sets the checker method to use and the return value for unlock to return every + * time it is called. + */ + void expectUnLock(UnlockFunc checkerFunc, Status returnThis); + + /** + * Sets the checker method to use and its return value the every time ping is called. + */ + void expectPing(PingFunc checkerFunc, Status returnThis); + + /** + * Sets the checker method to use and its return value the every time stopPing is called. + */ + void expectStopPing(StopPingFunc checkerFunc, Status returnThis); + + /** + * Sets the checker method to use and its return value the every time + * getLockByTS is called. + */ + void expectGetLockByTS(GetLockByTSFunc checkerFunc, StatusWith<LocksType> returnThis); + + /** + * Sets the checker method to use and its return value the every time + * getLockByName is called. + */ + void expectGetLockByName(GetLockByNameFunc checkerFunc, StatusWith<LocksType> returnThis); + + /** + * Sets the checker method to use and its return value the every time + * overtakeLock is called. + */ + void expectOvertakeLock(OvertakeLockFunc checkerFunc, StatusWith<LocksType> returnThis); + + /** + * Sets the checker method to use and its return value the every time + * getPing is called. + */ + void expectGetPing(GetPingFunc checkerFunc, StatusWith<LockpingsType> returnThis); + + /** + * Sets the checker method to use and its return value the every time + * getServerInfo is called. + */ + void expectGetServerInfo(GetServerInfoFunc checkerFunc, + StatusWith<DistLockCatalog::ServerInfo> returnThis); + +private: + // Protects all the member variables. + stdx::mutex _mutex; + + GrabLockFunc _grabLockChecker; + StatusWith<LocksType> _grabLockReturnValue; + + UnlockFunc _unlockChecker; + Status _unlockReturnValue; + + PingFunc _pingChecker; + Status _pingReturnValue; + + StopPingFunc _stopPingChecker; + Status _stopPingReturnValue; + + GetLockByTSFunc _getLockByTSChecker; + StatusWith<LocksType> _getLockByTSReturnValue; + + GetLockByNameFunc _getLockByNameChecker; + StatusWith<LocksType> _getLockByNameReturnValue; + + OvertakeLockFunc _overtakeLockChecker; + StatusWith<LocksType> _overtakeLockReturnValue; + + GetPingFunc _getPingChecker; + StatusWith<LockpingsType> _getPingReturnValue; + + GetServerInfoFunc _getServerInfoChecker; + StatusWith<DistLockCatalog::ServerInfo> _getServerInfoReturnValue; +}; } diff --git a/src/mongo/s/catalog/dist_lock_manager.cpp b/src/mongo/s/catalog/dist_lock_manager.cpp index e40504e2ff9..834e235b1e9 100644 --- a/src/mongo/s/catalog/dist_lock_manager.cpp +++ b/src/mongo/s/catalog/dist_lock_manager.cpp @@ -36,45 +36,41 @@ namespace mongo { - const stdx::chrono::milliseconds DistLockManager::kDefaultSingleLockAttemptTimeout(0); - const stdx::chrono::milliseconds DistLockManager::kDefaultLockRetryInterval(1000); +const stdx::chrono::milliseconds DistLockManager::kDefaultSingleLockAttemptTimeout(0); +const stdx::chrono::milliseconds DistLockManager::kDefaultLockRetryInterval(1000); - DistLockManager::ScopedDistLock::ScopedDistLock(): _lockManager(nullptr) { - } - - DistLockManager::ScopedDistLock::ScopedDistLock(DistLockHandle lockHandle, - DistLockManager* lockManager): - _lockID(std::move(lockHandle)), _lockManager(lockManager) { - } +DistLockManager::ScopedDistLock::ScopedDistLock() : _lockManager(nullptr) {} - DistLockManager::ScopedDistLock::~ScopedDistLock() { - if (_lockManager) { - _lockManager->unlock(_lockID); - } - } +DistLockManager::ScopedDistLock::ScopedDistLock(DistLockHandle lockHandle, + DistLockManager* lockManager) + : _lockID(std::move(lockHandle)), _lockManager(lockManager) {} - DistLockManager::ScopedDistLock::ScopedDistLock(ScopedDistLock&& other) { - *this = std::move(other); +DistLockManager::ScopedDistLock::~ScopedDistLock() { + if (_lockManager) { + _lockManager->unlock(_lockID); } +} - DistLockManager::ScopedDistLock& - DistLockManager::ScopedDistLock::operator=(ScopedDistLock&& other) { - if (this != &other) { - _lockID = std::move(other._lockID); - _lockManager = other._lockManager; - other._lockManager = nullptr; - } +DistLockManager::ScopedDistLock::ScopedDistLock(ScopedDistLock&& other) { + *this = std::move(other); +} - return *this; +DistLockManager::ScopedDistLock& DistLockManager::ScopedDistLock::operator=( + ScopedDistLock&& other) { + if (this != &other) { + _lockID = std::move(other._lockID); + _lockManager = other._lockManager; + other._lockManager = nullptr; } - Status DistLockManager::ScopedDistLock::checkStatus() { - if (!_lockManager) { - return Status(ErrorCodes::IllegalOperation, - "no lock manager, lock was not acquired"); - } + return *this; +} - return _lockManager->checkStatus(_lockID); +Status DistLockManager::ScopedDistLock::checkStatus() { + if (!_lockManager) { + return Status(ErrorCodes::IllegalOperation, "no lock manager, lock was not acquired"); } + return _lockManager->checkStatus(_lockID); +} } diff --git a/src/mongo/s/catalog/dist_lock_manager.h b/src/mongo/s/catalog/dist_lock_manager.h index 17fc7ca0ad3..c313447ae11 100644 --- a/src/mongo/s/catalog/dist_lock_manager.h +++ b/src/mongo/s/catalog/dist_lock_manager.h @@ -34,94 +34,93 @@ namespace mongo { - using DistLockHandle = OID; - class Status; - template <typename T> class StatusWith; +using DistLockHandle = OID; +class Status; +template <typename T> +class StatusWith; + +/** + * Interface for handling distributed locks. + * + * Usage: + * + * auto scopedDistLock = mgr->lock(...); + * + * if (!scopedDistLock.isOK()) { + * // Did not get lock. scopedLockStatus destructor will not call unlock. + * } + * + * // To check if lock is still owned: + * auto status = scopedDistLock.getValue().checkStatus(); + * + * if (!status.isOK()) { + * // Someone took over the lock! Unlock will still be called at destructor, but will + * // practically be a no-op since it doesn't own the lock anymore. + * } + */ +class DistLockManager { +public: + static const stdx::chrono::milliseconds kDefaultSingleLockAttemptTimeout; + static const stdx::chrono::milliseconds kDefaultLockRetryInterval; /** - * Interface for handling distributed locks. - * - * Usage: - * - * auto scopedDistLock = mgr->lock(...); - * - * if (!scopedDistLock.isOK()) { - * // Did not get lock. scopedLockStatus destructor will not call unlock. - * } - * - * // To check if lock is still owned: - * auto status = scopedDistLock.getValue().checkStatus(); - * - * if (!status.isOK()) { - * // Someone took over the lock! Unlock will still be called at destructor, but will - * // practically be a no-op since it doesn't own the lock anymore. - * } + * RAII type for distributed lock. Not meant to be shared across multiple threads. */ - class DistLockManager { + class ScopedDistLock { + MONGO_DISALLOW_COPYING(ScopedDistLock); + public: + ScopedDistLock(); // TODO: SERVER-18007 + ScopedDistLock(DistLockHandle lockHandle, DistLockManager* lockManager); + ~ScopedDistLock(); - static const stdx::chrono::milliseconds kDefaultSingleLockAttemptTimeout; - static const stdx::chrono::milliseconds kDefaultLockRetryInterval; + ScopedDistLock(ScopedDistLock&& other); + ScopedDistLock& operator=(ScopedDistLock&& other); /** - * RAII type for distributed lock. Not meant to be shared across multiple threads. + * Checks whether the lock is still being held by querying the config server. */ - class ScopedDistLock { - MONGO_DISALLOW_COPYING(ScopedDistLock); - - public: - ScopedDistLock(); // TODO: SERVER-18007 - ScopedDistLock(DistLockHandle lockHandle, DistLockManager* lockManager); - ~ScopedDistLock(); - - ScopedDistLock(ScopedDistLock&& other); - ScopedDistLock& operator=(ScopedDistLock&& other); - - /** - * Checks whether the lock is still being held by querying the config server. - */ - Status checkStatus(); + Status checkStatus(); - private: - DistLockHandle _lockID; - DistLockManager* _lockManager; // Not owned here. - }; + private: + DistLockHandle _lockID; + DistLockManager* _lockManager; // Not owned here. + }; - virtual ~DistLockManager() = default; + virtual ~DistLockManager() = default; - virtual void startUp() = 0; - virtual void shutDown() = 0; + virtual void startUp() = 0; + virtual void shutDown() = 0; - /** - * Tries multiple times to lock, using the specified lock try interval, until - * a certain amount of time has passed or when any error that is not LockBusy - * occurred. - * - * waitFor = 0 indicates there should only be one attempt to acquire the lock, and - * no waiting. - * waitFor = -1 indicates we should retry indefinitely. - * - * Returns OK if the lock was successfully acquired. - * Returns ErrorCodes::DistributedClockSkewed when a clock skew is detected. - * Returns ErrorCodes::LockBusy if the lock is being held. - */ - virtual StatusWith<ScopedDistLock> lock( - StringData name, - StringData whyMessage, - stdx::chrono::milliseconds waitFor = kDefaultSingleLockAttemptTimeout, - stdx::chrono::milliseconds lockTryInterval = kDefaultLockRetryInterval) = 0; - - protected: + /** + * Tries multiple times to lock, using the specified lock try interval, until + * a certain amount of time has passed or when any error that is not LockBusy + * occurred. + * + * waitFor = 0 indicates there should only be one attempt to acquire the lock, and + * no waiting. + * waitFor = -1 indicates we should retry indefinitely. + * + * Returns OK if the lock was successfully acquired. + * Returns ErrorCodes::DistributedClockSkewed when a clock skew is detected. + * Returns ErrorCodes::LockBusy if the lock is being held. + */ + virtual StatusWith<ScopedDistLock> lock( + StringData name, + StringData whyMessage, + stdx::chrono::milliseconds waitFor = kDefaultSingleLockAttemptTimeout, + stdx::chrono::milliseconds lockTryInterval = kDefaultLockRetryInterval) = 0; - /** - * Unlocks the given lockHandle. Will attempt to retry again later if the config - * server is not reachable. - */ - virtual void unlock(const DistLockHandle& lockHandle) = 0; +protected: + /** + * Unlocks the given lockHandle. Will attempt to retry again later if the config + * server is not reachable. + */ + virtual void unlock(const DistLockHandle& lockHandle) = 0; - /** - * Checks if the lockHandle still exists in the config server. - */ - virtual Status checkStatus(const DistLockHandle& lockHandle) = 0; - }; + /** + * Checks if the lockHandle still exists in the config server. + */ + virtual Status checkStatus(const DistLockHandle& lockHandle) = 0; +}; } diff --git a/src/mongo/s/catalog/dist_lock_manager_mock.cpp b/src/mongo/s/catalog/dist_lock_manager_mock.cpp index 1231edfea26..2c28aea9a0e 100644 --- a/src/mongo/s/catalog/dist_lock_manager_mock.cpp +++ b/src/mongo/s/catalog/dist_lock_manager_mock.cpp @@ -39,82 +39,71 @@ namespace mongo { - using stdx::chrono::milliseconds; +using stdx::chrono::milliseconds; namespace { - void NoLockFuncSet(StringData name, - StringData whyMessage, - milliseconds waitFor, - milliseconds lockTryInterval) { - FAIL(str::stream() << "Lock not expected to be called. " - << "Name: " << name - << ", whyMessage: " << whyMessage - << ", waitFor: " << waitFor - << ", lockTryInterval: " << lockTryInterval); - } +void NoLockFuncSet(StringData name, + StringData whyMessage, + milliseconds waitFor, + milliseconds lockTryInterval) { + FAIL(str::stream() << "Lock not expected to be called. " + << "Name: " << name << ", whyMessage: " << whyMessage + << ", waitFor: " << waitFor << ", lockTryInterval: " << lockTryInterval); +} -} // namespace +} // namespace - DistLockManagerMock::DistLockManagerMock() : _lockReturnStatus{Status::OK()}, - _lockChecker{NoLockFuncSet} {} +DistLockManagerMock::DistLockManagerMock() + : _lockReturnStatus{Status::OK()}, _lockChecker{NoLockFuncSet} {} - void DistLockManagerMock::startUp() {} +void DistLockManagerMock::startUp() {} - void DistLockManagerMock::shutDown() { - uassert(28659, - "DistLockManagerMock shut down with outstanding locks present", - _locks.empty()); - } +void DistLockManagerMock::shutDown() { + uassert(28659, "DistLockManagerMock shut down with outstanding locks present", _locks.empty()); +} - StatusWith<DistLockManager::ScopedDistLock> DistLockManagerMock::lock( - StringData name, - StringData whyMessage, - milliseconds waitFor, - milliseconds lockTryInterval) { - - _lockChecker(name, whyMessage, waitFor, lockTryInterval); - _lockChecker = NoLockFuncSet; - - if (!_lockReturnStatus.isOK()) { - return _lockReturnStatus; - } - - if (_locks.end() != std::find_if( - _locks.begin(), - _locks.end(), - [name](LockInfo info)->bool { return info.name == name; })) { - return Status(ErrorCodes::LockBusy, - str::stream() << "Lock \"" << name << "\" is already taken"); - } - - LockInfo info; - info.name = name.toString(); - info.lockID = DistLockHandle::gen(); - _locks.push_back(info); - - return DistLockManager::ScopedDistLock(info.lockID, this); - } +StatusWith<DistLockManager::ScopedDistLock> DistLockManagerMock::lock( + StringData name, StringData whyMessage, milliseconds waitFor, milliseconds lockTryInterval) { + _lockChecker(name, whyMessage, waitFor, lockTryInterval); + _lockChecker = NoLockFuncSet; - void DistLockManagerMock::unlock(const DistLockHandle& lockHandle) { - std::vector<LockInfo>::iterator it = - std::find_if( - _locks.begin(), - _locks.end(), - [&lockHandle](LockInfo info)->bool { return info.lockID == lockHandle; }); - if (it == _locks.end()) { - return; - } - _locks.erase(it); + if (!_lockReturnStatus.isOK()) { + return _lockReturnStatus; } - Status DistLockManagerMock::checkStatus(const DistLockHandle& lockHandle) { - return Status::OK(); + if (_locks.end() != std::find_if(_locks.begin(), + _locks.end(), + [name](LockInfo info) -> bool { return info.name == name; })) { + return Status(ErrorCodes::LockBusy, + str::stream() << "Lock \"" << name << "\" is already taken"); } - void DistLockManagerMock::expectLock(LockFunc checker, Status status) { - _lockReturnStatus = std::move(status); - _lockChecker = checker; + LockInfo info; + info.name = name.toString(); + info.lockID = DistLockHandle::gen(); + _locks.push_back(info); + + return DistLockManager::ScopedDistLock(info.lockID, this); +} + +void DistLockManagerMock::unlock(const DistLockHandle& lockHandle) { + std::vector<LockInfo>::iterator it = + std::find_if(_locks.begin(), + _locks.end(), + [&lockHandle](LockInfo info) -> bool { return info.lockID == lockHandle; }); + if (it == _locks.end()) { + return; } + _locks.erase(it); +} +Status DistLockManagerMock::checkStatus(const DistLockHandle& lockHandle) { + return Status::OK(); +} + +void DistLockManagerMock::expectLock(LockFunc checker, Status status) { + _lockReturnStatus = std::move(status); + _lockChecker = checker; +} } diff --git a/src/mongo/s/catalog/dist_lock_manager_mock.h b/src/mongo/s/catalog/dist_lock_manager_mock.h index bc8b1e47004..6b9f9e87801 100644 --- a/src/mongo/s/catalog/dist_lock_manager_mock.h +++ b/src/mongo/s/catalog/dist_lock_manager_mock.h @@ -36,44 +36,41 @@ namespace mongo { - class DistLockManagerMock: public DistLockManager { - public: - DistLockManagerMock(); +class DistLockManagerMock : public DistLockManager { +public: + DistLockManagerMock(); - virtual ~DistLockManagerMock() = default; + virtual ~DistLockManagerMock() = default; - virtual void startUp() override; - virtual void shutDown() override; + virtual void startUp() override; + virtual void shutDown() override; - virtual StatusWith<DistLockManager::ScopedDistLock> lock( - StringData name, - StringData whyMessage, - stdx::chrono::milliseconds waitFor, - stdx::chrono::milliseconds lockTryInterval) override; + virtual StatusWith<DistLockManager::ScopedDistLock> lock( + StringData name, + StringData whyMessage, + stdx::chrono::milliseconds waitFor, + stdx::chrono::milliseconds lockTryInterval) override; - using LockFunc = stdx::function<void (StringData name, - StringData whyMessage, - stdx::chrono::milliseconds waitFor, - stdx::chrono::milliseconds lockTryInterval)>; + using LockFunc = stdx::function<void(StringData name, + StringData whyMessage, + stdx::chrono::milliseconds waitFor, + stdx::chrono::milliseconds lockTryInterval)>; - void expectLock(LockFunc checkerFunc, Status lockStatus); + void expectLock(LockFunc checkerFunc, Status lockStatus); - protected: +protected: + virtual void unlock(const DistLockHandle& lockHandle) override; - virtual void unlock(const DistLockHandle& lockHandle) override; + virtual Status checkStatus(const DistLockHandle& lockHandle) override; - virtual Status checkStatus(const DistLockHandle& lockHandle) override; - - private: - - struct LockInfo { - DistLockHandle lockID; - std::string name; - }; - - std::vector<LockInfo> _locks; - Status _lockReturnStatus; - LockFunc _lockChecker; +private: + struct LockInfo { + DistLockHandle lockID; + std::string name; }; + std::vector<LockInfo> _locks; + Status _lockReturnStatus; + LockFunc _lockChecker; +}; } diff --git a/src/mongo/s/catalog/dist_lock_ping_info.cpp b/src/mongo/s/catalog/dist_lock_ping_info.cpp index 070ae1ed77f..a81cc1c749c 100644 --- a/src/mongo/s/catalog/dist_lock_ping_info.cpp +++ b/src/mongo/s/catalog/dist_lock_ping_info.cpp @@ -32,18 +32,13 @@ namespace mongo { - DistLockPingInfo::DistLockPingInfo() = default; - - DistLockPingInfo::DistLockPingInfo(StringData idArg, - Date_t lastPingArg, - Date_t remoteArg, - OID tsArg, - OID electionIdArg) : - processId(idArg.toString()), - lastPing(lastPingArg), - configLocalTime(remoteArg), - lockSessionId(std::move(tsArg)), - electionId(std::move(electionIdArg)) { - } +DistLockPingInfo::DistLockPingInfo() = default; +DistLockPingInfo::DistLockPingInfo( + StringData idArg, Date_t lastPingArg, Date_t remoteArg, OID tsArg, OID electionIdArg) + : processId(idArg.toString()), + lastPing(lastPingArg), + configLocalTime(remoteArg), + lockSessionId(std::move(tsArg)), + electionId(std::move(electionIdArg)) {} } diff --git a/src/mongo/s/catalog/dist_lock_ping_info.h b/src/mongo/s/catalog/dist_lock_ping_info.h index 58e7cc0bc64..34f4483bd9c 100644 --- a/src/mongo/s/catalog/dist_lock_ping_info.h +++ b/src/mongo/s/catalog/dist_lock_ping_info.h @@ -36,31 +36,31 @@ namespace mongo { - /** - * Data structure for storing information about distributed lock pings. - */ - struct DistLockPingInfo { - DistLockPingInfo(); - DistLockPingInfo(StringData processId, - Date_t lastPing, - Date_t configLocalTime, - OID lockSessionId, - OID electionId); +/** + * Data structure for storing information about distributed lock pings. + */ +struct DistLockPingInfo { + DistLockPingInfo(); + DistLockPingInfo(StringData processId, + Date_t lastPing, + Date_t configLocalTime, + OID lockSessionId, + OID electionId); - // the process processId of the last known owner of the lock. - std::string processId; + // the process processId of the last known owner of the lock. + std::string processId; - // the ping value from the last owner of the lock. - Date_t lastPing; + // the ping value from the last owner of the lock. + Date_t lastPing; - // the config server local time when this object was updated. - Date_t configLocalTime; + // the config server local time when this object was updated. + Date_t configLocalTime; - // last known owner of the lock. - OID lockSessionId; + // last known owner of the lock. + OID lockSessionId; - // the election id of the config server when this object was updated. - // Note: unused by legacy dist lock. - OID electionId; - }; + // the election id of the config server when this object was updated. + // Note: unused by legacy dist lock. + OID electionId; +}; } diff --git a/src/mongo/s/catalog/legacy/catalog_manager_legacy.cpp b/src/mongo/s/catalog/legacy/catalog_manager_legacy.cpp index ab274c579fb..c25618e31e8 100644 --- a/src/mongo/s/catalog/legacy/catalog_manager_legacy.cpp +++ b/src/mongo/s/catalog/legacy/catalog_manager_legacy.cpp @@ -80,1657 +80,1570 @@ namespace mongo { - using std::map; - using std::pair; - using std::set; - using std::string; - using std::vector; - using str::stream; +using std::map; +using std::pair; +using std::set; +using std::string; +using std::vector; +using str::stream; namespace { - bool validConfigWC(const BSONObj& writeConcern) { - BSONElement elem(writeConcern["w"]); - if (elem.eoo()) { - return true; - } +bool validConfigWC(const BSONObj& writeConcern) { + BSONElement elem(writeConcern["w"]); + if (elem.eoo()) { + return true; + } - if (elem.isNumber() && elem.numberInt() <= 1) { - return true; - } + if (elem.isNumber() && elem.numberInt() <= 1) { + return true; + } - if (elem.type() == String && elem.str() == "majority") { - return true; - } + if (elem.type() == String && elem.str() == "majority") { + return true; + } - return false; + return false; +} + +void toBatchError(const Status& status, BatchedCommandResponse* response) { + response->clear(); + response->setErrCode(status.code()); + response->setErrMessage(status.reason()); + response->setOk(false); + + dassert(response->isValid(NULL)); +} + +StatusWith<string> isValidShard(const string& name, + const ConnectionString& shardConnectionString, + ScopedDbConnection& conn) { + if (conn->type() == ConnectionString::SYNC) { + return Status(ErrorCodes::BadValue, + "can't use sync cluster as a shard; for a replica set, " + "you have to use <setname>/<server1>,<server2>,..."); } - void toBatchError(const Status& status, BatchedCommandResponse* response) { - response->clear(); - response->setErrCode(status.code()); - response->setErrMessage(status.reason()); - response->setOk(false); + BSONObj resIsMongos; + // (ok == 0) implies that it is a mongos + if (conn->runCommand("admin", BSON("isdbgrid" << 1), resIsMongos)) { + return Status(ErrorCodes::BadValue, "can't add a mongos process as a shard"); + } - dassert(response->isValid(NULL)); + BSONObj resIsMaster; + if (!conn->runCommand("admin", BSON("isMaster" << 1), resIsMaster)) { + return Status(ErrorCodes::OperationFailed, + str::stream() << "failed running isMaster: " << resIsMaster); } - StatusWith<string> isValidShard(const string& name, - const ConnectionString& shardConnectionString, - ScopedDbConnection& conn) { - if (conn->type() == ConnectionString::SYNC) { - return Status(ErrorCodes::BadValue, - "can't use sync cluster as a shard; for a replica set, " - "you have to use <setname>/<server1>,<server2>,..."); - } + // if the shard has only one host, make sure it is not part of a replica set + string setName = resIsMaster["setName"].str(); + string commandSetName = shardConnectionString.getSetName(); + if (commandSetName.empty() && !setName.empty()) { + return Status(ErrorCodes::BadValue, + str::stream() << "host is part of set " << setName << "; " + << "use replica set url format " + << "<setname>/<server1>,<server2>, ..."); + } - BSONObj resIsMongos; - // (ok == 0) implies that it is a mongos - if (conn->runCommand("admin", BSON("isdbgrid" << 1), resIsMongos)) { - return Status(ErrorCodes::BadValue, - "can't add a mongos process as a shard"); - } + if (!commandSetName.empty() && setName.empty()) { + return Status(ErrorCodes::OperationFailed, + str::stream() << "host did not return a set name; " + << "is the replica set still initializing? " << resIsMaster); + } - BSONObj resIsMaster; - if (!conn->runCommand("admin", BSON("isMaster" << 1), resIsMaster)) { - return Status(ErrorCodes::OperationFailed, - str::stream() << "failed running isMaster: " << resIsMaster); - } + // if the shard is part of replica set, make sure it is the right one + if (!commandSetName.empty() && (commandSetName != setName)) { + return Status(ErrorCodes::OperationFailed, + str::stream() << "host is part of a different set: " << setName); + } - // if the shard has only one host, make sure it is not part of a replica set - string setName = resIsMaster["setName"].str(); - string commandSetName = shardConnectionString.getSetName(); - if (commandSetName.empty() && !setName.empty()) { + if (setName.empty()) { + // check this isn't a --configsvr + BSONObj res; + bool ok = conn->runCommand("admin", BSON("replSetGetStatus" << 1), res); + if (!ok && res["info"].type() == String && res["info"].String() == "configsvr") { return Status(ErrorCodes::BadValue, - str::stream() << "host is part of set " << setName << "; " - << "use replica set url format " - << "<setname>/<server1>,<server2>, ..."); - } - - if (!commandSetName.empty() && setName.empty()) { - return Status(ErrorCodes::OperationFailed, - str::stream() << "host did not return a set name; " - << "is the replica set still initializing? " - << resIsMaster); + "the specified mongod is a --configsvr and " + "should thus not be a shard server"); } + } - // if the shard is part of replica set, make sure it is the right one - if (!commandSetName.empty() && (commandSetName != setName)) { - return Status(ErrorCodes::OperationFailed, - str::stream() << "host is part of a different set: " << setName); + // if the shard is part of a replica set, + // make sure all the hosts mentioned in 'shardConnectionString' are part of + // the set. It is fine if not all members of the set are present in 'shardConnectionString'. + bool foundAll = true; + string offendingHost; + if (!commandSetName.empty()) { + set<string> hostSet; + BSONObjIterator iter(resIsMaster["hosts"].Obj()); + while (iter.more()) { + hostSet.insert(iter.next().String()); // host:port + } + if (resIsMaster["passives"].isABSONObj()) { + BSONObjIterator piter(resIsMaster["passives"].Obj()); + while (piter.more()) { + hostSet.insert(piter.next().String()); // host:port + } } - - if (setName.empty()) { - // check this isn't a --configsvr - BSONObj res; - bool ok = conn->runCommand("admin", - BSON("replSetGetStatus" << 1), - res); - if(!ok && res["info"].type() == String && res["info"].String() == "configsvr") { - return Status(ErrorCodes::BadValue, - "the specified mongod is a --configsvr and " - "should thus not be a shard server"); + if (resIsMaster["arbiters"].isABSONObj()) { + BSONObjIterator piter(resIsMaster["arbiters"].Obj()); + while (piter.more()) { + hostSet.insert(piter.next().String()); // host:port } } - // if the shard is part of a replica set, - // make sure all the hosts mentioned in 'shardConnectionString' are part of - // the set. It is fine if not all members of the set are present in 'shardConnectionString'. - bool foundAll = true; - string offendingHost; - if (!commandSetName.empty()) { - set<string> hostSet; - BSONObjIterator iter(resIsMaster["hosts"].Obj()); - while (iter.more()) { - hostSet.insert(iter.next().String()); // host:port - } - if (resIsMaster["passives"].isABSONObj()) { - BSONObjIterator piter(resIsMaster["passives"].Obj()); - while (piter.more()) { - hostSet.insert(piter.next().String()); // host:port - } - } - if (resIsMaster["arbiters"].isABSONObj()) { - BSONObjIterator piter(resIsMaster["arbiters"].Obj()); - while (piter.more()) { - hostSet.insert(piter.next().String()); // host:port - } + vector<HostAndPort> hosts = shardConnectionString.getServers(); + for (size_t i = 0; i < hosts.size(); i++) { + if (!hosts[i].hasPort()) { + hosts[i] = HostAndPort(hosts[i].host(), hosts[i].port()); } - - vector<HostAndPort> hosts = shardConnectionString.getServers(); - for (size_t i = 0; i < hosts.size(); i++) { - if (!hosts[i].hasPort()) { - hosts[i] = HostAndPort(hosts[i].host(), hosts[i].port()); - } - string host = hosts[i].toString(); // host:port - if (hostSet.find(host) == hostSet.end()) { - offendingHost = host; - foundAll = false; - break; - } + string host = hosts[i].toString(); // host:port + if (hostSet.find(host) == hostSet.end()) { + offendingHost = host; + foundAll = false; + break; } } - if (!foundAll) { - return Status(ErrorCodes::OperationFailed, - str::stream() << "in seed list " << shardConnectionString.toString() - << ", host " << offendingHost - << " does not belong to replica set " << setName); - } - - string shardName(name); - // shard name defaults to the name of the replica set - if (name.empty() && !setName.empty()) { - shardName = setName; - } - - // disallow adding shard replica set with name 'config' - if (shardName == "config") { - return Status(ErrorCodes::BadValue, - "use of shard replica set with name 'config' is not allowed"); - } + } + if (!foundAll) { + return Status(ErrorCodes::OperationFailed, + str::stream() << "in seed list " << shardConnectionString.toString() + << ", host " << offendingHost + << " does not belong to replica set " << setName); + } - return shardName; + string shardName(name); + // shard name defaults to the name of the replica set + if (name.empty() && !setName.empty()) { + shardName = setName; } - // In order to be accepted as a new shard, that mongod must not have - // any database name that exists already in any other shards. - // If that test passes, the new shard's databases are going to be entered as - // non-sharded db's whose primary is the newly added shard. - StatusWith<vector<string>> getDBNames(const ConnectionString& shardConnectionString, - ScopedDbConnection& conn) { - vector<string> dbNames; + // disallow adding shard replica set with name 'config' + if (shardName == "config") { + return Status(ErrorCodes::BadValue, + "use of shard replica set with name 'config' is not allowed"); + } - BSONObj resListDB; - if (!conn->runCommand("admin", BSON("listDatabases" << 1), resListDB)) { - return Status(ErrorCodes::OperationFailed, - str::stream() << "failed listing " - << shardConnectionString.toString() - << "'s databases:" << resListDB); - } + return shardName; +} - BSONObjIterator i(resListDB["databases"].Obj()); - while (i.more()) { - BSONObj dbEntry = i.next().Obj(); - const string& dbName = dbEntry["name"].String(); - if (!(dbName == "local" || dbName == "admin" || dbName == "config")) { - dbNames.push_back(dbName); - } - } +// In order to be accepted as a new shard, that mongod must not have +// any database name that exists already in any other shards. +// If that test passes, the new shard's databases are going to be entered as +// non-sharded db's whose primary is the newly added shard. +StatusWith<vector<string>> getDBNames(const ConnectionString& shardConnectionString, + ScopedDbConnection& conn) { + vector<string> dbNames; - return dbNames; + BSONObj resListDB; + if (!conn->runCommand("admin", BSON("listDatabases" << 1), resListDB)) { + return Status(ErrorCodes::OperationFailed, + str::stream() << "failed listing " << shardConnectionString.toString() + << "'s databases:" << resListDB); } - BSONObj buildRemoveLogEntry(const string& shardName, bool isDraining) { - BSONObjBuilder details; - details.append("shard", shardName); - details.append("isDraining", isDraining); - - return details.obj(); + BSONObjIterator i(resListDB["databases"].Obj()); + while (i.more()) { + BSONObj dbEntry = i.next().Obj(); + const string& dbName = dbEntry["name"].String(); + if (!(dbName == "local" || dbName == "admin" || dbName == "config")) { + dbNames.push_back(dbName); + } } - // Whether the logChange call should attempt to create the changelog collection - AtomicInt32 changeLogCollectionCreated(0); + return dbNames; +} - // Whether the logAction call should attempt to create the actionlog collection - AtomicInt32 actionLogCollectionCreated(0); +BSONObj buildRemoveLogEntry(const string& shardName, bool isDraining) { + BSONObjBuilder details; + details.append("shard", shardName); + details.append("isDraining", isDraining); -} // namespace + return details.obj(); +} +// Whether the logChange call should attempt to create the changelog collection +AtomicInt32 changeLogCollectionCreated(0); - CatalogManagerLegacy::CatalogManagerLegacy() = default; +// Whether the logAction call should attempt to create the actionlog collection +AtomicInt32 actionLogCollectionCreated(0); - CatalogManagerLegacy::~CatalogManagerLegacy() = default; +} // namespace - Status CatalogManagerLegacy::init(const ConnectionString& configDBCS) { - // Initialization should not happen more than once - invariant(!_configServerConnectionString.isValid()); - invariant(_configServers.empty()); - invariant(configDBCS.isValid()); - - // Extract the hosts in HOST:PORT format - set<HostAndPort> configHostsAndPortsSet; - set<string> configHostsOnly; - std::vector<HostAndPort> configHostAndPorts = configDBCS.getServers(); - for (size_t i = 0; i < configHostAndPorts.size(); i++) { - // Append the default port, if not specified - HostAndPort configHost = configHostAndPorts[i]; - if (!configHost.hasPort()) { - configHost = HostAndPort(configHost.host(), ServerGlobalParams::ConfigServerPort); - } - - // Make sure there are no duplicates - if (!configHostsAndPortsSet.insert(configHost).second) { - StringBuilder sb; - sb << "Host " << configHost.toString() - << " exists twice in the config servers listing."; - return Status(ErrorCodes::InvalidOptions, sb.str()); - } +CatalogManagerLegacy::CatalogManagerLegacy() = default; - configHostsOnly.insert(configHost.host()); - } +CatalogManagerLegacy::~CatalogManagerLegacy() = default; - // Make sure the hosts are reachable - for (set<string>::const_iterator i = configHostsOnly.begin(); - i != configHostsOnly.end(); - i++) { +Status CatalogManagerLegacy::init(const ConnectionString& configDBCS) { + // Initialization should not happen more than once + invariant(!_configServerConnectionString.isValid()); + invariant(_configServers.empty()); + invariant(configDBCS.isValid()); - const string host = *i; + // Extract the hosts in HOST:PORT format + set<HostAndPort> configHostsAndPortsSet; + set<string> configHostsOnly; + std::vector<HostAndPort> configHostAndPorts = configDBCS.getServers(); + for (size_t i = 0; i < configHostAndPorts.size(); i++) { + // Append the default port, if not specified + HostAndPort configHost = configHostAndPorts[i]; + if (!configHost.hasPort()) { + configHost = HostAndPort(configHost.host(), ServerGlobalParams::ConfigServerPort); + } - // If this is a CUSTOM connection string (for testing) don't do DNS resolution - string errMsg; - if (ConnectionString::parse(host, errMsg).type() == ConnectionString::CUSTOM) { - continue; - } + // Make sure there are no duplicates + if (!configHostsAndPortsSet.insert(configHost).second) { + StringBuilder sb; + sb << "Host " << configHost.toString() + << " exists twice in the config servers listing."; - bool ok = false; + return Status(ErrorCodes::InvalidOptions, sb.str()); + } - for (int x = 10; x > 0; x--) { - if (!hostbyname(host.c_str()).empty()) { - ok = true; - break; - } + configHostsOnly.insert(configHost.host()); + } - log() << "can't resolve DNS for [" << host << "] sleeping and trying " - << x << " more times"; - sleepsecs(10); - } + // Make sure the hosts are reachable + for (set<string>::const_iterator i = configHostsOnly.begin(); i != configHostsOnly.end(); i++) { + const string host = *i; - if (!ok) { - return Status(ErrorCodes::HostNotFound, - stream() << "unable to resolve DNS for host " << host); - } + // If this is a CUSTOM connection string (for testing) don't do DNS resolution + string errMsg; + if (ConnectionString::parse(host, errMsg).type() == ConnectionString::CUSTOM) { + continue; } - LOG(1) << " config string : " << configDBCS.toString(); - - // Now that the config hosts are verified, initialize the catalog manager. The code below - // should never fail. + bool ok = false; - _configServerConnectionString = configDBCS; - - if (_configServerConnectionString.type() == ConnectionString::MASTER) { - _configServers.push_back(_configServerConnectionString); - } - else if (_configServerConnectionString.type() == ConnectionString::SYNC || - (_configServerConnectionString.type() == ConnectionString::SET && - _configServerConnectionString.getServers().size() == 1)) { - // TODO(spencer): Remove second part of the above or statement that allows replset - // config server strings once we've separated the legacy catalog manager from the - // CSRS version. - const vector<HostAndPort> configHPs = _configServerConnectionString.getServers(); - for (vector<HostAndPort>::const_iterator it = configHPs.begin(); - it != configHPs.end(); - ++it) { - - _configServers.push_back(ConnectionString(*it)); + for (int x = 10; x > 0; x--) { + if (!hostbyname(host.c_str()).empty()) { + ok = true; + break; } - } - else { - // This is only for tests. - invariant(_configServerConnectionString.type() == ConnectionString::CUSTOM); - _configServers.push_back(_configServerConnectionString); - } - _distLockManager = stdx::make_unique<LegacyDistLockManager>(_configServerConnectionString); - _distLockManager->startUp(); + log() << "can't resolve DNS for [" << host << "] sleeping and trying " << x + << " more times"; + sleepsecs(10); + } - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _inShutdown = false; - _consistentFromLastCheck = true; + if (!ok) { + return Status(ErrorCodes::HostNotFound, + stream() << "unable to resolve DNS for host " << host); } + } - return Status::OK(); + LOG(1) << " config string : " << configDBCS.toString(); + + // Now that the config hosts are verified, initialize the catalog manager. The code below + // should never fail. + + _configServerConnectionString = configDBCS; + + if (_configServerConnectionString.type() == ConnectionString::MASTER) { + _configServers.push_back(_configServerConnectionString); + } else if (_configServerConnectionString.type() == ConnectionString::SYNC || + (_configServerConnectionString.type() == ConnectionString::SET && + _configServerConnectionString.getServers().size() == 1)) { + // TODO(spencer): Remove second part of the above or statement that allows replset + // config server strings once we've separated the legacy catalog manager from the + // CSRS version. + const vector<HostAndPort> configHPs = _configServerConnectionString.getServers(); + for (vector<HostAndPort>::const_iterator it = configHPs.begin(); it != configHPs.end(); + ++it) { + _configServers.push_back(ConnectionString(*it)); + } + } else { + // This is only for tests. + invariant(_configServerConnectionString.type() == ConnectionString::CUSTOM); + _configServers.push_back(_configServerConnectionString); } - Status CatalogManagerLegacy::startup(bool upgrade) { - Status status = _startConfigServerChecker(); - if (!status.isOK()) { - return status; - } + _distLockManager = stdx::make_unique<LegacyDistLockManager>(_configServerConnectionString); + _distLockManager->startUp(); - status = _checkAndUpgradeConfigMetadata(upgrade); - return status; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _inShutdown = false; + _consistentFromLastCheck = true; } - Status CatalogManagerLegacy::_checkAndUpgradeConfigMetadata(bool doUpgrade) { - VersionType initVersionInfo; - VersionType versionInfo; - string errMsg; + return Status::OK(); +} - bool upgraded = checkAndUpgradeConfigVersion(this, - doUpgrade, - &initVersionInfo, - &versionInfo, - &errMsg); - if (!upgraded) { - return Status(ErrorCodes::IncompatibleShardingMetadata, - str::stream() << "error upgrading config database to v" - << CURRENT_CONFIG_VERSION << causedBy(errMsg)); - } - - return Status::OK(); +Status CatalogManagerLegacy::startup(bool upgrade) { + Status status = _startConfigServerChecker(); + if (!status.isOK()) { + return status; } - Status CatalogManagerLegacy::_startConfigServerChecker() { - if (!_checkConfigServersConsistent()) { - return Status(ErrorCodes::ConfigServersInconsistent, - "Data inconsistency detected amongst config servers"); - } + status = _checkAndUpgradeConfigMetadata(upgrade); + return status; +} + +Status CatalogManagerLegacy::_checkAndUpgradeConfigMetadata(bool doUpgrade) { + VersionType initVersionInfo; + VersionType versionInfo; + string errMsg; + + bool upgraded = + checkAndUpgradeConfigVersion(this, doUpgrade, &initVersionInfo, &versionInfo, &errMsg); + if (!upgraded) { + return Status(ErrorCodes::IncompatibleShardingMetadata, + str::stream() << "error upgrading config database to v" + << CURRENT_CONFIG_VERSION << causedBy(errMsg)); + } - stdx::thread t(stdx::bind(&CatalogManagerLegacy::_consistencyChecker, this)); - _consistencyCheckerThread.swap(t); + return Status::OK(); +} - return Status::OK(); +Status CatalogManagerLegacy::_startConfigServerChecker() { + if (!_checkConfigServersConsistent()) { + return Status(ErrorCodes::ConfigServersInconsistent, + "Data inconsistency detected amongst config servers"); } - ConnectionString CatalogManagerLegacy::connectionString() const { - return _configServerConnectionString; - } + stdx::thread t(stdx::bind(&CatalogManagerLegacy::_consistencyChecker, this)); + _consistencyCheckerThread.swap(t); - void CatalogManagerLegacy::shutDown() { - LOG(1) << "CatalogManagerLegacy::shutDown() called."; - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _inShutdown = true; - _consistencyCheckerCV.notify_one(); - } - _consistencyCheckerThread.join(); + return Status::OK(); +} + +ConnectionString CatalogManagerLegacy::connectionString() const { + return _configServerConnectionString; +} - invariant(_distLockManager); - _distLockManager->shutDown(); +void CatalogManagerLegacy::shutDown() { + LOG(1) << "CatalogManagerLegacy::shutDown() called."; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _inShutdown = true; + _consistencyCheckerCV.notify_one(); } + _consistencyCheckerThread.join(); - Status CatalogManagerLegacy::enableSharding(const std::string& dbName) { - invariant(nsIsDbOnly(dbName)); + invariant(_distLockManager); + _distLockManager->shutDown(); +} - DatabaseType db; +Status CatalogManagerLegacy::enableSharding(const std::string& dbName) { + invariant(nsIsDbOnly(dbName)); - // Check for case sensitivity violations - Status status = _checkDbDoesNotExist(dbName); - if (status.isOK()) { - // Database does not exist, create a new entry - const ShardPtr primary = Shard::pick(); - if (primary) { - log() << "Placing [" << dbName << "] on: " << primary->toString(); + DatabaseType db; - db.setName(dbName); - db.setPrimary(primary->getId()); - db.setSharded(true); - } - else { - return Status(ErrorCodes::ShardNotFound, "can't find a shard to put new db on"); - } - } - else if (status.code() == ErrorCodes::NamespaceExists) { - // Database exists, so just update it - StatusWith<DatabaseType> dbStatus = getDatabase(dbName); - if (!dbStatus.isOK()) { - return dbStatus.getStatus(); - } + // Check for case sensitivity violations + Status status = _checkDbDoesNotExist(dbName); + if (status.isOK()) { + // Database does not exist, create a new entry + const ShardPtr primary = Shard::pick(); + if (primary) { + log() << "Placing [" << dbName << "] on: " << primary->toString(); - db = dbStatus.getValue(); + db.setName(dbName); + db.setPrimary(primary->getId()); db.setSharded(true); + } else { + return Status(ErrorCodes::ShardNotFound, "can't find a shard to put new db on"); } - else { - // Some fatal error - return status; + } else if (status.code() == ErrorCodes::NamespaceExists) { + // Database exists, so just update it + StatusWith<DatabaseType> dbStatus = getDatabase(dbName); + if (!dbStatus.isOK()) { + return dbStatus.getStatus(); } - log() << "Enabling sharding for database [" << dbName << "] in config db"; - - return updateDatabase(dbName, db); + db = dbStatus.getValue(); + db.setSharded(true); + } else { + // Some fatal error + return status; } - Status CatalogManagerLegacy::shardCollection(const string& ns, - const ShardKeyPattern& fieldsAndOrder, - bool unique, - vector<BSONObj>* initPoints, - set<ShardId>* initShardIds) { + log() << "Enabling sharding for database [" << dbName << "] in config db"; - StatusWith<DatabaseType> status = getDatabase(nsToDatabase(ns)); - if (!status.isOK()) { - return status.getStatus(); - } + return updateDatabase(dbName, db); +} - DatabaseType dbt = status.getValue(); - ShardId dbPrimaryShardId = dbt.getPrimary(); - - // This is an extra safety check that the collection is not getting sharded concurrently by - // two different mongos instances. It is not 100%-proof, but it reduces the chance that two - // invocations of shard collection will step on each other's toes. - { - ScopedDbConnection conn(_configServerConnectionString, 30); - unsigned long long existingChunks = conn->count(ChunkType::ConfigNS, - BSON(ChunkType::ns(ns))); - if (existingChunks > 0) { - conn.done(); - return Status(ErrorCodes::AlreadyInitialized, - str::stream() << "collection " << ns << " already sharded with " - << existingChunks << " chunks."); - } +Status CatalogManagerLegacy::shardCollection(const string& ns, + const ShardKeyPattern& fieldsAndOrder, + bool unique, + vector<BSONObj>* initPoints, + set<ShardId>* initShardIds) { + StatusWith<DatabaseType> status = getDatabase(nsToDatabase(ns)); + if (!status.isOK()) { + return status.getStatus(); + } + + DatabaseType dbt = status.getValue(); + ShardId dbPrimaryShardId = dbt.getPrimary(); + // This is an extra safety check that the collection is not getting sharded concurrently by + // two different mongos instances. It is not 100%-proof, but it reduces the chance that two + // invocations of shard collection will step on each other's toes. + { + ScopedDbConnection conn(_configServerConnectionString, 30); + unsigned long long existingChunks = + conn->count(ChunkType::ConfigNS, BSON(ChunkType::ns(ns))); + if (existingChunks > 0) { conn.done(); + return Status(ErrorCodes::AlreadyInitialized, + str::stream() << "collection " << ns << " already sharded with " + << existingChunks << " chunks."); } - log() << "enable sharding on: " << ns << " with shard key: " << fieldsAndOrder; + conn.done(); + } - // Record start in changelog - BSONObjBuilder collectionDetail; - collectionDetail.append("shardKey", fieldsAndOrder.toBSON()); - collectionDetail.append("collection", ns); - string dbPrimaryShardStr; - { - const auto shard = grid.shardRegistry()->getShard(dbPrimaryShardId); - dbPrimaryShardStr = shard->toString(); - } - collectionDetail.append("primary", dbPrimaryShardStr); - - BSONArray initialShards; - if (initShardIds == NULL) - initialShards = BSONArray(); - else { - BSONArrayBuilder b; - for (const ShardId& shardId : *initShardIds) { - b.append(shardId); - } - initialShards = b.arr(); + log() << "enable sharding on: " << ns << " with shard key: " << fieldsAndOrder; + + // Record start in changelog + BSONObjBuilder collectionDetail; + collectionDetail.append("shardKey", fieldsAndOrder.toBSON()); + collectionDetail.append("collection", ns); + string dbPrimaryShardStr; + { + const auto shard = grid.shardRegistry()->getShard(dbPrimaryShardId); + dbPrimaryShardStr = shard->toString(); + } + collectionDetail.append("primary", dbPrimaryShardStr); + + BSONArray initialShards; + if (initShardIds == NULL) + initialShards = BSONArray(); + else { + BSONArrayBuilder b; + for (const ShardId& shardId : *initShardIds) { + b.append(shardId); } + initialShards = b.arr(); + } - collectionDetail.append("initShards", initialShards); - collectionDetail.append("numChunks", static_cast<int>(initPoints->size() + 1)); - - logChange(NULL, "shardCollection.start", ns, collectionDetail.obj()); - - ChunkManagerPtr manager(new ChunkManager(ns, fieldsAndOrder, unique)); - manager->createFirstChunks(dbPrimaryShardId, - initPoints, - initShardIds); - manager->loadExistingRanges(nullptr); - - CollectionInfo collInfo; - collInfo.useChunkManager(manager); - collInfo.save(ns); - manager->reload(true); - - // Tell the primary mongod to refresh its data - // TODO: Think the real fix here is for mongos to just - // assume that all collections are sharded, when we get there - for (int i = 0;i < 4;i++) { - if (i == 3) { - warning() << "too many tries updating initial version of " << ns - << " on shard primary " << dbPrimaryShardStr - << ", other mongoses may not see the collection as sharded immediately"; - break; - } + collectionDetail.append("initShards", initialShards); + collectionDetail.append("numChunks", static_cast<int>(initPoints->size() + 1)); - try { - const auto shard = grid.shardRegistry()->getShard(dbPrimaryShardId); - ShardConnection conn(shard->getConnString(), ns); - bool isVersionSet = conn.setVersion(); - conn.done(); - if (!isVersionSet) { - warning() << "could not update initial version of " - << ns << " on shard primary " << dbPrimaryShardStr; - } else { - break; - } - } - catch (const DBException& e) { - warning() << "could not update initial version of " << ns - << " on shard primary " << dbPrimaryShardStr - << causedBy(e); - } + logChange(NULL, "shardCollection.start", ns, collectionDetail.obj()); - sleepsecs(i); - } + ChunkManagerPtr manager(new ChunkManager(ns, fieldsAndOrder, unique)); + manager->createFirstChunks(dbPrimaryShardId, initPoints, initShardIds); + manager->loadExistingRanges(nullptr); - // Record finish in changelog - BSONObjBuilder finishDetail; + CollectionInfo collInfo; + collInfo.useChunkManager(manager); + collInfo.save(ns); + manager->reload(true); - finishDetail.append("version", manager->getVersion().toString()); + // Tell the primary mongod to refresh its data + // TODO: Think the real fix here is for mongos to just + // assume that all collections are sharded, when we get there + for (int i = 0; i < 4; i++) { + if (i == 3) { + warning() << "too many tries updating initial version of " << ns << " on shard primary " + << dbPrimaryShardStr + << ", other mongoses may not see the collection as sharded immediately"; + break; + } - logChange(NULL, "shardCollection", ns, finishDetail.obj()); + try { + const auto shard = grid.shardRegistry()->getShard(dbPrimaryShardId); + ShardConnection conn(shard->getConnString(), ns); + bool isVersionSet = conn.setVersion(); + conn.done(); + if (!isVersionSet) { + warning() << "could not update initial version of " << ns << " on shard primary " + << dbPrimaryShardStr; + } else { + break; + } + } catch (const DBException& e) { + warning() << "could not update initial version of " << ns << " on shard primary " + << dbPrimaryShardStr << causedBy(e); + } - return Status::OK(); + sleepsecs(i); } - Status CatalogManagerLegacy::createDatabase(const std::string& dbName) { - invariant(nsIsDbOnly(dbName)); + // Record finish in changelog + BSONObjBuilder finishDetail; - // The admin and config databases should never be explicitly created. They "just exist", - // i.e. getDatabase will always return an entry for them. - invariant(dbName != "admin"); - invariant(dbName != "config"); + finishDetail.append("version", manager->getVersion().toString()); - // Lock the database globally to prevent conflicts with simultaneous database creation. - auto scopedDistLock = getDistLockManager()->lock(dbName, - "createDatabase", - Seconds{5}, - Milliseconds{500}); - if (!scopedDistLock.isOK()) { - return scopedDistLock.getStatus(); - } - - // Check for case sensitivity violations - auto status = _checkDbDoesNotExist(dbName); - if (!status.isOK()) { - return status; - } + logChange(NULL, "shardCollection", ns, finishDetail.obj()); - // Database does not exist, pick a shard and create a new entry - const ShardPtr primaryShard = Shard::pick(); - if (!primaryShard) { - return Status(ErrorCodes::ShardNotFound, "can't find a shard to put new db on"); - } + return Status::OK(); +} - log() << "Placing [" << dbName << "] on: " << primaryShard->toString(); +Status CatalogManagerLegacy::createDatabase(const std::string& dbName) { + invariant(nsIsDbOnly(dbName)); - DatabaseType db; - db.setName(dbName); - db.setPrimary(primaryShard->getId()); - db.setSharded(false); + // The admin and config databases should never be explicitly created. They "just exist", + // i.e. getDatabase will always return an entry for them. + invariant(dbName != "admin"); + invariant(dbName != "config"); - BatchedCommandResponse response; - status = insert(DatabaseType::ConfigNS, db.toBSON(), &response); - if (status.isOK()) { - return status; - } + // Lock the database globally to prevent conflicts with simultaneous database creation. + auto scopedDistLock = + getDistLockManager()->lock(dbName, "createDatabase", Seconds{5}, Milliseconds{500}); + if (!scopedDistLock.isOK()) { + return scopedDistLock.getStatus(); + } - if (status.code() == ErrorCodes::DuplicateKey) { - return Status(ErrorCodes::NamespaceExists, "database " + dbName + " already exists"); - } + // Check for case sensitivity violations + auto status = _checkDbDoesNotExist(dbName); + if (!status.isOK()) { + return status; + } - return Status(status.code(), str::stream() << "database metadata write failed for " - << dbName << ". Error: " << response.toBSON()); + // Database does not exist, pick a shard and create a new entry + const ShardPtr primaryShard = Shard::pick(); + if (!primaryShard) { + return Status(ErrorCodes::ShardNotFound, "can't find a shard to put new db on"); } - StatusWith<string> CatalogManagerLegacy::addShard(const string& name, - const ConnectionString& shardConnectionString, - const long long maxSize) { + log() << "Placing [" << dbName << "] on: " << primaryShard->toString(); - string shardName; - ReplicaSetMonitorPtr rsMonitor; - vector<string> dbNames; + DatabaseType db; + db.setName(dbName); + db.setPrimary(primaryShard->getId()); + db.setSharded(false); - try { - ScopedDbConnection newShardConn(shardConnectionString); - newShardConn->getLastError(); - - StatusWith<string> validShard = isValidShard(name, - shardConnectionString, - newShardConn); - if (!validShard.isOK()) { - newShardConn.done(); - return validShard.getStatus(); - } - shardName = validShard.getValue(); + BatchedCommandResponse response; + status = insert(DatabaseType::ConfigNS, db.toBSON(), &response); + if (status.isOK()) { + return status; + } - StatusWith<vector<string>> shardDBNames = getDBNames(shardConnectionString, - newShardConn); - if (!shardDBNames.isOK()) { - newShardConn.done(); - return shardDBNames.getStatus(); - } - dbNames = shardDBNames.getValue(); + if (status.code() == ErrorCodes::DuplicateKey) { + return Status(ErrorCodes::NamespaceExists, "database " + dbName + " already exists"); + } - if (newShardConn->type() == ConnectionString::SET) { - rsMonitor = ReplicaSetMonitor::get(shardConnectionString.getSetName()); - } + return Status(status.code(), + str::stream() << "database metadata write failed for " << dbName + << ". Error: " << response.toBSON()); +} + +StatusWith<string> CatalogManagerLegacy::addShard(const string& name, + const ConnectionString& shardConnectionString, + const long long maxSize) { + string shardName; + ReplicaSetMonitorPtr rsMonitor; + vector<string> dbNames; + try { + ScopedDbConnection newShardConn(shardConnectionString); + newShardConn->getLastError(); + + StatusWith<string> validShard = isValidShard(name, shardConnectionString, newShardConn); + if (!validShard.isOK()) { newShardConn.done(); + return validShard.getStatus(); } - catch (const DBException& e) { - if (shardConnectionString.type() == ConnectionString::SET) { - shardConnectionPool.removeHost(shardConnectionString.getSetName()); - ReplicaSetMonitor::remove(shardConnectionString.getSetName()); - } + shardName = validShard.getValue(); - return Status(ErrorCodes::OperationFailed, - str::stream() << "couldn't connect to new shard " - << e.what()); + StatusWith<vector<string>> shardDBNames = getDBNames(shardConnectionString, newShardConn); + if (!shardDBNames.isOK()) { + newShardConn.done(); + return shardDBNames.getStatus(); } + dbNames = shardDBNames.getValue(); - // check that none of the existing shard candidate's db's exist elsewhere - for (vector<string>::const_iterator it = dbNames.begin(); it != dbNames.end(); ++it) { - StatusWith<DatabaseType> dbt = getDatabase(*it); - if (dbt.isOK()) { - return Status(ErrorCodes::OperationFailed, - str::stream() << "can't add shard " - << "'" << shardConnectionString.toString() << "'" - << " because a local database '" << *it - << "' exists in another " - << dbt.getValue().getPrimary()); - } + if (newShardConn->type() == ConnectionString::SET) { + rsMonitor = ReplicaSetMonitor::get(shardConnectionString.getSetName()); } - // if a name for a shard wasn't provided, pick one. - if (shardName.empty()) { - StatusWith<string> result = _getNewShardName(); - if (!result.isOK()) { - return Status(ErrorCodes::OperationFailed, - "error generating new shard name"); - } - shardName = result.getValue(); + newShardConn.done(); + } catch (const DBException& e) { + if (shardConnectionString.type() == ConnectionString::SET) { + shardConnectionPool.removeHost(shardConnectionString.getSetName()); + ReplicaSetMonitor::remove(shardConnectionString.getSetName()); } - // build the ConfigDB shard document - BSONObjBuilder b; - b.append(ShardType::name(), shardName); - b.append(ShardType::host(), - rsMonitor ? rsMonitor->getServerAddress() : shardConnectionString.toString()); - if (maxSize > 0) { - b.append(ShardType::maxSizeMB(), maxSize); - } - BSONObj shardDoc = b.obj(); + return Status(ErrorCodes::OperationFailed, + str::stream() << "couldn't connect to new shard " << e.what()); + } - if (isShardHost(shardConnectionString)) { - return Status(ErrorCodes::OperationFailed, "host already used"); + // check that none of the existing shard candidate's db's exist elsewhere + for (vector<string>::const_iterator it = dbNames.begin(); it != dbNames.end(); ++it) { + StatusWith<DatabaseType> dbt = getDatabase(*it); + if (dbt.isOK()) { + return Status(ErrorCodes::OperationFailed, + str::stream() << "can't add shard " + << "'" << shardConnectionString.toString() << "'" + << " because a local database '" << *it + << "' exists in another " << dbt.getValue().getPrimary()); } + } - log() << "going to add shard: " << shardDoc; - - Status result = insert(ShardType::ConfigNS, shardDoc, NULL); + // if a name for a shard wasn't provided, pick one. + if (shardName.empty()) { + StatusWith<string> result = _getNewShardName(); if (!result.isOK()) { - log() << "error adding shard: " << shardDoc << " err: " << result.reason(); - return result; + return Status(ErrorCodes::OperationFailed, "error generating new shard name"); } + shardName = result.getValue(); + } - Shard::reloadShardInfo(); - - // add all databases of the new shard - for (vector<string>::const_iterator it = dbNames.begin(); it != dbNames.end(); ++it) { - DatabaseType dbt; - dbt.setName(*it); - dbt.setPrimary(shardName); - dbt.setSharded(false); - Status status = updateDatabase(*it, dbt); - if (!status.isOK()) { - log() << "adding shard " << shardConnectionString.toString() - << " even though could not add database " << *it; - } - } + // build the ConfigDB shard document + BSONObjBuilder b; + b.append(ShardType::name(), shardName); + b.append(ShardType::host(), + rsMonitor ? rsMonitor->getServerAddress() : shardConnectionString.toString()); + if (maxSize > 0) { + b.append(ShardType::maxSizeMB(), maxSize); + } + BSONObj shardDoc = b.obj(); - // Record in changelog - BSONObjBuilder shardDetails; - shardDetails.append("name", shardName); - shardDetails.append("host", shardConnectionString.toString()); + if (isShardHost(shardConnectionString)) { + return Status(ErrorCodes::OperationFailed, "host already used"); + } - logChange(NULL, "addShard", "", shardDetails.obj()); + log() << "going to add shard: " << shardDoc; - return shardName; + Status result = insert(ShardType::ConfigNS, shardDoc, NULL); + if (!result.isOK()) { + log() << "error adding shard: " << shardDoc << " err: " << result.reason(); + return result; } - StatusWith<ShardDrainingStatus> CatalogManagerLegacy::removeShard(OperationContext* txn, - const std::string& name) { - ScopedDbConnection conn(_configServerConnectionString, 30); + Shard::reloadShardInfo(); - if (conn->count(ShardType::ConfigNS, - BSON(ShardType::name() << NE << name - << ShardType::draining(true)))) { - conn.done(); - return Status(ErrorCodes::ConflictingOperationInProgress, - "Can't have more than one draining shard at a time"); + // add all databases of the new shard + for (vector<string>::const_iterator it = dbNames.begin(); it != dbNames.end(); ++it) { + DatabaseType dbt; + dbt.setName(*it); + dbt.setPrimary(shardName); + dbt.setSharded(false); + Status status = updateDatabase(*it, dbt); + if (!status.isOK()) { + log() << "adding shard " << shardConnectionString.toString() + << " even though could not add database " << *it; } + } - if (conn->count(ShardType::ConfigNS, - BSON(ShardType::name() << NE << name)) == 0) { - conn.done(); - return Status(ErrorCodes::IllegalOperation, - "Can't remove last shard"); - } + // Record in changelog + BSONObjBuilder shardDetails; + shardDetails.append("name", shardName); + shardDetails.append("host", shardConnectionString.toString()); - BSONObj searchDoc = BSON(ShardType::name() << name); + logChange(NULL, "addShard", "", shardDetails.obj()); - // Case 1: start draining chunks - BSONObj drainingDoc = BSON(ShardType::name() << name << ShardType::draining(true)); - BSONObj shardDoc = conn->findOne(ShardType::ConfigNS, drainingDoc); - if (shardDoc.isEmpty()) { - log() << "going to start draining shard: " << name; - BSONObj newStatus = BSON("$set" << BSON(ShardType::draining(true))); + return shardName; +} - Status status = update(ShardType::ConfigNS, searchDoc, newStatus, false, false, NULL); - if (!status.isOK()) { - log() << "error starting removeShard: " << name - << "; err: " << status.reason(); - return status; - } +StatusWith<ShardDrainingStatus> CatalogManagerLegacy::removeShard(OperationContext* txn, + const std::string& name) { + ScopedDbConnection conn(_configServerConnectionString, 30); - BSONObj primaryLocalDoc = BSON(DatabaseType::name("local") << - DatabaseType::primary(name)); - log() << "primaryLocalDoc: " << primaryLocalDoc; - if (conn->count(DatabaseType::ConfigNS, primaryLocalDoc)) { - log() << "This shard is listed as primary of local db. Removing entry."; - - Status status = remove(DatabaseType::ConfigNS, - BSON(DatabaseType::name("local")), - 0, - NULL); - if (!status.isOK()) { - log() << "error removing local db: " - << status.reason(); - return status; - } - } + if (conn->count(ShardType::ConfigNS, + BSON(ShardType::name() << NE << name << ShardType::draining(true)))) { + conn.done(); + return Status(ErrorCodes::ConflictingOperationInProgress, + "Can't have more than one draining shard at a time"); + } - Shard::reloadShardInfo(); - conn.done(); + if (conn->count(ShardType::ConfigNS, BSON(ShardType::name() << NE << name)) == 0) { + conn.done(); + return Status(ErrorCodes::IllegalOperation, "Can't remove last shard"); + } + + BSONObj searchDoc = BSON(ShardType::name() << name); + + // Case 1: start draining chunks + BSONObj drainingDoc = BSON(ShardType::name() << name << ShardType::draining(true)); + BSONObj shardDoc = conn->findOne(ShardType::ConfigNS, drainingDoc); + if (shardDoc.isEmpty()) { + log() << "going to start draining shard: " << name; + BSONObj newStatus = BSON("$set" << BSON(ShardType::draining(true))); - // Record start in changelog - logChange(txn, "removeShard.start", "", buildRemoveLogEntry(name, true)); - return ShardDrainingStatus::STARTED; + Status status = update(ShardType::ConfigNS, searchDoc, newStatus, false, false, NULL); + if (!status.isOK()) { + log() << "error starting removeShard: " << name << "; err: " << status.reason(); + return status; } - // Case 2: all chunks drained - BSONObj shardIDDoc = BSON(ChunkType::shard(shardDoc[ShardType::name()].str())); - long long chunkCount = conn->count(ChunkType::ConfigNS, shardIDDoc); - long long dbCount = conn->count(DatabaseType::ConfigNS, - BSON(DatabaseType::name.ne("local") - << DatabaseType::primary(name))); - if (chunkCount == 0 && dbCount == 0) { - log() << "going to remove shard: " << name; - audit::logRemoveShard(ClientBasic::getCurrent(), name); - - Status status = remove(ShardType::ConfigNS, searchDoc, 0, NULL); + BSONObj primaryLocalDoc = BSON(DatabaseType::name("local") << DatabaseType::primary(name)); + log() << "primaryLocalDoc: " << primaryLocalDoc; + if (conn->count(DatabaseType::ConfigNS, primaryLocalDoc)) { + log() << "This shard is listed as primary of local db. Removing entry."; + + Status status = + remove(DatabaseType::ConfigNS, BSON(DatabaseType::name("local")), 0, NULL); if (!status.isOK()) { - log() << "Error concluding removeShard operation on: " << name - << "; err: " << status.reason(); + log() << "error removing local db: " << status.reason(); return status; } + } - grid.shardRegistry()->remove(name); - - shardConnectionPool.removeHost(name); - ReplicaSetMonitor::remove(name); + Shard::reloadShardInfo(); + conn.done(); - Shard::reloadShardInfo(); - conn.done(); + // Record start in changelog + logChange(txn, "removeShard.start", "", buildRemoveLogEntry(name, true)); + return ShardDrainingStatus::STARTED; + } - // Record finish in changelog - logChange(txn, "removeShard", "", buildRemoveLogEntry(name, false)); - return ShardDrainingStatus::COMPLETED; + // Case 2: all chunks drained + BSONObj shardIDDoc = BSON(ChunkType::shard(shardDoc[ShardType::name()].str())); + long long chunkCount = conn->count(ChunkType::ConfigNS, shardIDDoc); + long long dbCount = + conn->count(DatabaseType::ConfigNS, + BSON(DatabaseType::name.ne("local") << DatabaseType::primary(name))); + if (chunkCount == 0 && dbCount == 0) { + log() << "going to remove shard: " << name; + audit::logRemoveShard(ClientBasic::getCurrent(), name); + + Status status = remove(ShardType::ConfigNS, searchDoc, 0, NULL); + if (!status.isOK()) { + log() << "Error concluding removeShard operation on: " << name + << "; err: " << status.reason(); + return status; } - // case 3: draining ongoing - return ShardDrainingStatus::ONGOING; - } + grid.shardRegistry()->remove(name); - Status CatalogManagerLegacy::updateDatabase(const std::string& dbName, const DatabaseType& db) { - fassert(28616, db.validate()); + shardConnectionPool.removeHost(name); + ReplicaSetMonitor::remove(name); - BatchedCommandResponse response; - Status status = update(DatabaseType::ConfigNS, - BSON(DatabaseType::name(dbName)), - db.toBSON(), - true, // upsert - false, // multi - &response); - if (!status.isOK()) { - return Status(status.code(), - str::stream() << "database metadata write failed: " - << response.toBSON() << "; status: " << status.toString()); - } + Shard::reloadShardInfo(); + conn.done(); - return Status::OK(); + // Record finish in changelog + logChange(txn, "removeShard", "", buildRemoveLogEntry(name, false)); + return ShardDrainingStatus::COMPLETED; } - StatusWith<DatabaseType> CatalogManagerLegacy::getDatabase(const std::string& dbName) { - invariant(nsIsDbOnly(dbName)); + // case 3: draining ongoing + return ShardDrainingStatus::ONGOING; +} + +Status CatalogManagerLegacy::updateDatabase(const std::string& dbName, const DatabaseType& db) { + fassert(28616, db.validate()); + + BatchedCommandResponse response; + Status status = update(DatabaseType::ConfigNS, + BSON(DatabaseType::name(dbName)), + db.toBSON(), + true, // upsert + false, // multi + &response); + if (!status.isOK()) { + return Status(status.code(), + str::stream() << "database metadata write failed: " << response.toBSON() + << "; status: " << status.toString()); + } - // The two databases that are hosted on the config server are config and admin - if (dbName == "config" || dbName == "admin") { - DatabaseType dbt; - dbt.setName(dbName); - dbt.setSharded(false); - dbt.setPrimary("config"); + return Status::OK(); +} - return dbt; - } +StatusWith<DatabaseType> CatalogManagerLegacy::getDatabase(const std::string& dbName) { + invariant(nsIsDbOnly(dbName)); - ScopedDbConnection conn(_configServerConnectionString, 30.0); - - BSONObj dbObj = conn->findOne(DatabaseType::ConfigNS, BSON(DatabaseType::name(dbName))); - if (dbObj.isEmpty()) { - conn.done(); - return Status(ErrorCodes::DatabaseNotFound, - stream() << "database " << dbName << " not found"); - } + // The two databases that are hosted on the config server are config and admin + if (dbName == "config" || dbName == "admin") { + DatabaseType dbt; + dbt.setName(dbName); + dbt.setSharded(false); + dbt.setPrimary("config"); - conn.done(); - return DatabaseType::fromBSON(dbObj); + return dbt; } - Status CatalogManagerLegacy::updateCollection(const std::string& collNs, - const CollectionType& coll) { - fassert(28634, coll.validate()); + ScopedDbConnection conn(_configServerConnectionString, 30.0); - BatchedCommandResponse response; - Status status = update(CollectionType::ConfigNS, - BSON(CollectionType::fullNs(collNs)), - coll.toBSON(), - true, // upsert - false, // multi - &response); - if (!status.isOK()) { - return Status(status.code(), - str::stream() << "collection metadata write failed: " - << response.toBSON() << "; status: " << status.toString()); - } + BSONObj dbObj = conn->findOne(DatabaseType::ConfigNS, BSON(DatabaseType::name(dbName))); + if (dbObj.isEmpty()) { + conn.done(); + return Status(ErrorCodes::DatabaseNotFound, + stream() << "database " << dbName << " not found"); + } - return Status::OK(); + conn.done(); + return DatabaseType::fromBSON(dbObj); +} + +Status CatalogManagerLegacy::updateCollection(const std::string& collNs, + const CollectionType& coll) { + fassert(28634, coll.validate()); + + BatchedCommandResponse response; + Status status = update(CollectionType::ConfigNS, + BSON(CollectionType::fullNs(collNs)), + coll.toBSON(), + true, // upsert + false, // multi + &response); + if (!status.isOK()) { + return Status(status.code(), + str::stream() << "collection metadata write failed: " << response.toBSON() + << "; status: " << status.toString()); } - StatusWith<CollectionType> CatalogManagerLegacy::getCollection(const std::string& collNs) { - ScopedDbConnection conn(_configServerConnectionString, 30.0); + return Status::OK(); +} - BSONObj collObj = conn->findOne(CollectionType::ConfigNS, - BSON(CollectionType::fullNs(collNs))); - if (collObj.isEmpty()) { - conn.done(); - return Status(ErrorCodes::NamespaceNotFound, - stream() << "collection " << collNs << " not found"); - } +StatusWith<CollectionType> CatalogManagerLegacy::getCollection(const std::string& collNs) { + ScopedDbConnection conn(_configServerConnectionString, 30.0); + BSONObj collObj = conn->findOne(CollectionType::ConfigNS, BSON(CollectionType::fullNs(collNs))); + if (collObj.isEmpty()) { conn.done(); - return CollectionType::fromBSON(collObj); + return Status(ErrorCodes::NamespaceNotFound, + stream() << "collection " << collNs << " not found"); } - Status CatalogManagerLegacy::getCollections(const std::string* dbName, - std::vector<CollectionType>* collections) { - collections->clear(); + conn.done(); + return CollectionType::fromBSON(collObj); +} - BSONObjBuilder b; - if (dbName) { - invariant(!dbName->empty()); - b.appendRegex(CollectionType::fullNs(), - (string)"^" + pcrecpp::RE::QuoteMeta(*dbName) + "\\."); - } +Status CatalogManagerLegacy::getCollections(const std::string* dbName, + std::vector<CollectionType>* collections) { + collections->clear(); - ScopedDbConnection conn(_configServerConnectionString, 30.0); + BSONObjBuilder b; + if (dbName) { + invariant(!dbName->empty()); + b.appendRegex(CollectionType::fullNs(), + (string) "^" + pcrecpp::RE::QuoteMeta(*dbName) + "\\."); + } - std::unique_ptr<DBClientCursor> cursor(_safeCursor(conn->query(CollectionType::ConfigNS, - b.obj()))); + ScopedDbConnection conn(_configServerConnectionString, 30.0); - while (cursor->more()) { - const BSONObj collObj = cursor->next(); + std::unique_ptr<DBClientCursor> cursor( + _safeCursor(conn->query(CollectionType::ConfigNS, b.obj()))); - auto status = CollectionType::fromBSON(collObj); - if (!status.isOK()) { - conn.done(); - return status.getStatus(); - } + while (cursor->more()) { + const BSONObj collObj = cursor->next(); - collections->push_back(status.getValue()); + auto status = CollectionType::fromBSON(collObj); + if (!status.isOK()) { + conn.done(); + return status.getStatus(); } - conn.done(); - return Status::OK(); + collections->push_back(status.getValue()); } - Status CatalogManagerLegacy::dropCollection(const std::string& collectionNs) { - logChange(NULL, "dropCollection.start", collectionNs, BSONObj()); + conn.done(); + return Status::OK(); +} - // Lock the collection globally so that split/migrate cannot run - auto scopedDistLock = getDistLockManager()->lock(collectionNs, "drop"); - if (!scopedDistLock.isOK()) { - return scopedDistLock.getStatus(); - } +Status CatalogManagerLegacy::dropCollection(const std::string& collectionNs) { + logChange(NULL, "dropCollection.start", collectionNs, BSONObj()); - LOG(1) << "dropCollection " << collectionNs << " started"; + // Lock the collection globally so that split/migrate cannot run + auto scopedDistLock = getDistLockManager()->lock(collectionNs, "drop"); + if (!scopedDistLock.isOK()) { + return scopedDistLock.getStatus(); + } - // This cleans up the collection on all shards - vector<ShardType> allShards; - Status status = getAllShards(&allShards); - if (!status.isOK()) { - return status; - } + LOG(1) << "dropCollection " << collectionNs << " started"; - LOG(1) << "dropCollection " << collectionNs << " locked"; + // This cleans up the collection on all shards + vector<ShardType> allShards; + Status status = getAllShards(&allShards); + if (!status.isOK()) { + return status; + } - map<string, BSONObj> errors; + LOG(1) << "dropCollection " << collectionNs << " locked"; - // Delete data from all mongods - for (vector<ShardType>::const_iterator i = allShards.begin(); i != allShards.end(); i++) { - const auto shard = grid.shardRegistry()->getShard(i->getName()); - ScopedDbConnection conn(shard->getConnString()); + map<string, BSONObj> errors; - BSONObj info; - if (!conn->dropCollection(collectionNs, &info)) { - // Ignore the database not found errors - if (info["code"].isNumber() && - (info["code"].Int() == ErrorCodes::NamespaceNotFound)) { - conn.done(); - continue; - } + // Delete data from all mongods + for (vector<ShardType>::const_iterator i = allShards.begin(); i != allShards.end(); i++) { + const auto shard = grid.shardRegistry()->getShard(i->getName()); + ScopedDbConnection conn(shard->getConnString()); - errors[shard->getConnString().toString()] = info; + BSONObj info; + if (!conn->dropCollection(collectionNs, &info)) { + // Ignore the database not found errors + if (info["code"].isNumber() && (info["code"].Int() == ErrorCodes::NamespaceNotFound)) { + conn.done(); + continue; } - conn.done(); + errors[shard->getConnString().toString()] = info; } - if (!errors.empty()) { - StringBuilder sb; - sb << "Dropping collection failed on the following hosts: "; - - for (map<string, BSONObj>::const_iterator it = errors.begin(); - it != errors.end(); - ++it) { + conn.done(); + } - if (it != errors.begin()) { - sb << ", "; - } + if (!errors.empty()) { + StringBuilder sb; + sb << "Dropping collection failed on the following hosts: "; - sb << it->first << ": " << it->second; + for (map<string, BSONObj>::const_iterator it = errors.begin(); it != errors.end(); ++it) { + if (it != errors.begin()) { + sb << ", "; } - return Status(ErrorCodes::OperationFailed, sb.str()); + sb << it->first << ": " << it->second; } - LOG(1) << "dropCollection " << collectionNs << " shard data deleted"; - - // remove chunk data - Status result = remove(ChunkType::ConfigNS, - BSON(ChunkType::ns(collectionNs)), - 0, - NULL); - if (!result.isOK()) { - return result; - } + return Status(ErrorCodes::OperationFailed, sb.str()); + } - LOG(1) << "dropCollection " << collectionNs << " chunk data deleted"; + LOG(1) << "dropCollection " << collectionNs << " shard data deleted"; - for (vector<ShardType>::const_iterator i = allShards.begin(); i != allShards.end(); i++) { - const auto shard = grid.shardRegistry()->getShard(i->getName()); - ScopedDbConnection conn(shard->getConnString()); + // remove chunk data + Status result = remove(ChunkType::ConfigNS, BSON(ChunkType::ns(collectionNs)), 0, NULL); + if (!result.isOK()) { + return result; + } - BSONObj res; + LOG(1) << "dropCollection " << collectionNs << " chunk data deleted"; - // this is horrible - // we need a special command for dropping on the d side - // this hack works for the moment + for (vector<ShardType>::const_iterator i = allShards.begin(); i != allShards.end(); i++) { + const auto shard = grid.shardRegistry()->getShard(i->getName()); + ScopedDbConnection conn(shard->getConnString()); - if (!setShardVersion(conn.conn(), - collectionNs, - _configServerConnectionString.toString(), - ChunkVersion(0, 0, OID()), - NULL, - true, - res)) { + BSONObj res; - return Status(static_cast<ErrorCodes::Error>(8071), - str::stream() << "cleaning up after drop failed: " << res); - } + // this is horrible + // we need a special command for dropping on the d side + // this hack works for the moment - conn->simpleCommand("admin", 0, "unsetSharding"); - conn.done(); + if (!setShardVersion(conn.conn(), + collectionNs, + _configServerConnectionString.toString(), + ChunkVersion(0, 0, OID()), + NULL, + true, + res)) { + return Status(static_cast<ErrorCodes::Error>(8071), + str::stream() << "cleaning up after drop failed: " << res); } - LOG(1) << "dropCollection " << collectionNs << " completed"; + conn->simpleCommand("admin", 0, "unsetSharding"); + conn.done(); + } - logChange(NULL, "dropCollection", collectionNs, BSONObj()); + LOG(1) << "dropCollection " << collectionNs << " completed"; - return Status::OK(); - } + logChange(NULL, "dropCollection", collectionNs, BSONObj()); - void CatalogManagerLegacy::logAction(const ActionLogType& actionLog) { - // Create the action log collection and ensure that it is capped. Wrap in try/catch, - // because creating an existing collection throws. - if (actionLogCollectionCreated.load() == 0) { - try { - ScopedDbConnection conn(_configServerConnectionString, 30.0); - conn->createCollection(ActionLogType::ConfigNS, 1024 * 1024 * 2, true); - conn.done(); + return Status::OK(); +} - actionLogCollectionCreated.store(1); - } - catch (const DBException& e) { - // It's ok to ignore this exception - LOG(1) << "couldn't create actionlog collection: " << e; - } - } +void CatalogManagerLegacy::logAction(const ActionLogType& actionLog) { + // Create the action log collection and ensure that it is capped. Wrap in try/catch, + // because creating an existing collection throws. + if (actionLogCollectionCreated.load() == 0) { + try { + ScopedDbConnection conn(_configServerConnectionString, 30.0); + conn->createCollection(ActionLogType::ConfigNS, 1024 * 1024 * 2, true); + conn.done(); - Status result = insert(ActionLogType::ConfigNS, actionLog.toBSON(), NULL); - if (!result.isOK()) { - log() << "error encountered while logging action: " << result; + actionLogCollectionCreated.store(1); + } catch (const DBException& e) { + // It's ok to ignore this exception + LOG(1) << "couldn't create actionlog collection: " << e; } } - void CatalogManagerLegacy::logChange(OperationContext* opCtx, - const string& what, - const string& ns, - const BSONObj& detail) { - - // Create the change log collection and ensure that it is capped. Wrap in try/catch, - // because creating an existing collection throws. - if (changeLogCollectionCreated.load() == 0) { - try { - ScopedDbConnection conn(_configServerConnectionString, 30.0); - conn->createCollection(ChangelogType::ConfigNS, 1024 * 1024 * 10, true); - conn.done(); + Status result = insert(ActionLogType::ConfigNS, actionLog.toBSON(), NULL); + if (!result.isOK()) { + log() << "error encountered while logging action: " << result; + } +} + +void CatalogManagerLegacy::logChange(OperationContext* opCtx, + const string& what, + const string& ns, + const BSONObj& detail) { + // Create the change log collection and ensure that it is capped. Wrap in try/catch, + // because creating an existing collection throws. + if (changeLogCollectionCreated.load() == 0) { + try { + ScopedDbConnection conn(_configServerConnectionString, 30.0); + conn->createCollection(ChangelogType::ConfigNS, 1024 * 1024 * 10, true); + conn.done(); - changeLogCollectionCreated.store(1); - } - catch (const UserException& e) { - // It's ok to ignore this exception - LOG(1) << "couldn't create changelog collection: " << e; - } + changeLogCollectionCreated.store(1); + } catch (const UserException& e) { + // It's ok to ignore this exception + LOG(1) << "couldn't create changelog collection: " << e; } + } - // Store this entry's ID so we can use on the exception code path too - StringBuilder changeIdBuilder; - changeIdBuilder << getHostNameCached() << "-" << terseCurrentTime() - << "-" << OID::gen(); + // Store this entry's ID so we can use on the exception code path too + StringBuilder changeIdBuilder; + changeIdBuilder << getHostNameCached() << "-" << terseCurrentTime() << "-" << OID::gen(); - const string changeID = changeIdBuilder.str(); + const string changeID = changeIdBuilder.str(); - Client* client; - if (opCtx) { - client = opCtx->getClient(); - } - else if (haveClient()) { - client = &cc(); - } - else { - client = nullptr; - } + Client* client; + if (opCtx) { + client = opCtx->getClient(); + } else if (haveClient()) { + client = &cc(); + } else { + client = nullptr; + } - // Send a copy of the message to the local log in case it doesn't manage to reach - // config.changelog - BSONObj msg = BSON(ChangelogType::changeID(changeID) << - ChangelogType::server(getHostNameCached()) << - ChangelogType::clientAddr((client ? - client->clientAddress(true) : "")) << - ChangelogType::time(jsTime()) << - ChangelogType::what(what) << - ChangelogType::ns(ns) << - ChangelogType::details(detail)); + // Send a copy of the message to the local log in case it doesn't manage to reach + // config.changelog + BSONObj msg = BSON(ChangelogType::changeID(changeID) + << ChangelogType::server(getHostNameCached()) + << ChangelogType::clientAddr((client ? client->clientAddress(true) : "")) + << ChangelogType::time(jsTime()) << ChangelogType::what(what) + << ChangelogType::ns(ns) << ChangelogType::details(detail)); - log() << "about to log metadata event: " << msg; + log() << "about to log metadata event: " << msg; - Status result = insert(ChangelogType::ConfigNS, msg, NULL); - if (!result.isOK()) { - warning() << "Error encountered while logging config change with ID " - << changeID << ": " << result; - } + Status result = insert(ChangelogType::ConfigNS, msg, NULL); + if (!result.isOK()) { + warning() << "Error encountered while logging config change with ID " << changeID << ": " + << result; } +} - StatusWith<SettingsType> CatalogManagerLegacy::getGlobalSettings(const string& key) { - try { - ScopedDbConnection conn(_configServerConnectionString, 30); - BSONObj settingsDoc = conn->findOne(SettingsType::ConfigNS, - BSON(SettingsType::key(key))); - conn.done(); - - if (settingsDoc.isEmpty()) { - return Status(ErrorCodes::NoMatchingDocument, - str::stream() << "can't find settings document with key: " << key); - } +StatusWith<SettingsType> CatalogManagerLegacy::getGlobalSettings(const string& key) { + try { + ScopedDbConnection conn(_configServerConnectionString, 30); + BSONObj settingsDoc = conn->findOne(SettingsType::ConfigNS, BSON(SettingsType::key(key))); + conn.done(); - StatusWith<SettingsType> settingsResult = SettingsType::fromBSON(settingsDoc); - if (!settingsResult.isOK()) { - return Status(ErrorCodes::FailedToParse, - str::stream() << "error while parsing settings document: " - << settingsDoc - << " : " << settingsResult.getStatus().toString()); - } + if (settingsDoc.isEmpty()) { + return Status(ErrorCodes::NoMatchingDocument, + str::stream() << "can't find settings document with key: " << key); + } - const SettingsType& settings = settingsResult.getValue(); + StatusWith<SettingsType> settingsResult = SettingsType::fromBSON(settingsDoc); + if (!settingsResult.isOK()) { + return Status(ErrorCodes::FailedToParse, + str::stream() << "error while parsing settings document: " << settingsDoc + << " : " << settingsResult.getStatus().toString()); + } - Status validationStatus = settings.validate(); - if (!validationStatus.isOK()) { - return validationStatus; - } + const SettingsType& settings = settingsResult.getValue(); - return settingsResult; - } - catch (const DBException& ex) { - return Status(ErrorCodes::OperationFailed, - str::stream() << "unable to successfully obtain " - << "config.settings document: " << causedBy(ex)); + Status validationStatus = settings.validate(); + if (!validationStatus.isOK()) { + return validationStatus; } - } - Status CatalogManagerLegacy::getDatabasesForShard(const string& shardName, - vector<string>* dbs) { - dbs->clear(); - - try { - ScopedDbConnection conn(_configServerConnectionString, 30.0); - std::unique_ptr<DBClientCursor> cursor(_safeCursor( - conn->query(DatabaseType::ConfigNS, - Query(BSON(DatabaseType::primary(shardName)))))); - if (!cursor.get()) { - conn.done(); - return Status(ErrorCodes::HostUnreachable, "unable to open chunk cursor"); - } + return settingsResult; + } catch (const DBException& ex) { + return Status(ErrorCodes::OperationFailed, + str::stream() << "unable to successfully obtain " + << "config.settings document: " << causedBy(ex)); + } +} - while (cursor->more()) { - BSONObj shard = cursor->nextSafe(); - dbs->push_back(shard[DatabaseType::name()].str()); - } +Status CatalogManagerLegacy::getDatabasesForShard(const string& shardName, vector<string>* dbs) { + dbs->clear(); + try { + ScopedDbConnection conn(_configServerConnectionString, 30.0); + std::unique_ptr<DBClientCursor> cursor(_safeCursor( + conn->query(DatabaseType::ConfigNS, Query(BSON(DatabaseType::primary(shardName)))))); + if (!cursor.get()) { conn.done(); + return Status(ErrorCodes::HostUnreachable, "unable to open chunk cursor"); } - catch (const DBException& ex) { - return ex.toStatus(); + + while (cursor->more()) { + BSONObj shard = cursor->nextSafe(); + dbs->push_back(shard[DatabaseType::name()].str()); } - return Status::OK(); + conn.done(); + } catch (const DBException& ex) { + return ex.toStatus(); } - Status CatalogManagerLegacy::getChunks(const Query& query, - int nToReturn, - vector<ChunkType>* chunks) { - chunks->clear(); + return Status::OK(); +} - try { - ScopedDbConnection conn(_configServerConnectionString, 30.0); - std::unique_ptr<DBClientCursor> cursor(_safeCursor(conn->query(ChunkType::ConfigNS, - query, - nToReturn))); - if (!cursor.get()) { - conn.done(); - return Status(ErrorCodes::HostUnreachable, "unable to open chunk cursor"); - } +Status CatalogManagerLegacy::getChunks(const Query& query, + int nToReturn, + vector<ChunkType>* chunks) { + chunks->clear(); - while (cursor->more()) { - BSONObj chunkObj = cursor->nextSafe(); + try { + ScopedDbConnection conn(_configServerConnectionString, 30.0); + std::unique_ptr<DBClientCursor> cursor( + _safeCursor(conn->query(ChunkType::ConfigNS, query, nToReturn))); + if (!cursor.get()) { + conn.done(); + return Status(ErrorCodes::HostUnreachable, "unable to open chunk cursor"); + } - StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(chunkObj); - if (!chunkRes.isOK()) { - conn.done(); - chunks->clear(); - return {ErrorCodes::FailedToParse, - stream() << "Failed to parse chunk with id (" - << chunkObj[ChunkType::name()].toString() << "): " - << chunkRes.getStatus().reason()}; - } + while (cursor->more()) { + BSONObj chunkObj = cursor->nextSafe(); - chunks->push_back(chunkRes.getValue()); + StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(chunkObj); + if (!chunkRes.isOK()) { + conn.done(); + chunks->clear(); + return {ErrorCodes::FailedToParse, + stream() << "Failed to parse chunk with id (" + << chunkObj[ChunkType::name()].toString() + << "): " << chunkRes.getStatus().reason()}; } - conn.done(); - } - catch (const DBException& ex) { - return ex.toStatus(); + chunks->push_back(chunkRes.getValue()); } - return Status::OK(); + conn.done(); + } catch (const DBException& ex) { + return ex.toStatus(); } - Status CatalogManagerLegacy::getTagsForCollection(const std::string& collectionNs, - std::vector<TagsType>* tags) { - tags->clear(); + return Status::OK(); +} - try { - ScopedDbConnection conn(_configServerConnectionString, 30); - std::unique_ptr<DBClientCursor> cursor(_safeCursor( - conn->query(TagsType::ConfigNS, - Query(BSON(TagsType::ns(collectionNs))) - .sort(TagsType::min())))); - if (!cursor.get()) { - conn.done(); - return Status(ErrorCodes::HostUnreachable, "unable to open tags cursor"); - } +Status CatalogManagerLegacy::getTagsForCollection(const std::string& collectionNs, + std::vector<TagsType>* tags) { + tags->clear(); - while (cursor->more()) { - BSONObj tagObj = cursor->nextSafe(); + try { + ScopedDbConnection conn(_configServerConnectionString, 30); + std::unique_ptr<DBClientCursor> cursor(_safeCursor(conn->query( + TagsType::ConfigNS, Query(BSON(TagsType::ns(collectionNs))).sort(TagsType::min())))); + if (!cursor.get()) { + conn.done(); + return Status(ErrorCodes::HostUnreachable, "unable to open tags cursor"); + } - StatusWith<TagsType> tagRes = TagsType::fromBSON(tagObj); - if (!tagRes.isOK()) { - conn.done(); - return Status(ErrorCodes::FailedToParse, - str::stream() << "Failed to parse tag BSONObj: " - << tagRes.getStatus().reason()); - } + while (cursor->more()) { + BSONObj tagObj = cursor->nextSafe(); - tags->push_back(tagRes.getValue()); + StatusWith<TagsType> tagRes = TagsType::fromBSON(tagObj); + if (!tagRes.isOK()) { + conn.done(); + return Status(ErrorCodes::FailedToParse, + str::stream() << "Failed to parse tag BSONObj: " + << tagRes.getStatus().reason()); } - conn.done(); - } - catch (const DBException& ex) { - return ex.toStatus(); + tags->push_back(tagRes.getValue()); } - return Status::OK(); + conn.done(); + } catch (const DBException& ex) { + return ex.toStatus(); } - StatusWith<string> CatalogManagerLegacy::getTagForChunk(const std::string& collectionNs, - const ChunkType& chunk) { - BSONObj tagDoc; + return Status::OK(); +} - try { - ScopedDbConnection conn(_configServerConnectionString, 30); +StatusWith<string> CatalogManagerLegacy::getTagForChunk(const std::string& collectionNs, + const ChunkType& chunk) { + BSONObj tagDoc; - Query query(BSON(TagsType::ns(collectionNs) << - TagsType::min() << BSON("$lte" << chunk.getMin()) << - TagsType::max() << BSON("$gte" << chunk.getMax()))); + try { + ScopedDbConnection conn(_configServerConnectionString, 30); - tagDoc = conn->findOne(TagsType::ConfigNS, query); - conn.done(); - } - catch (const DBException& ex) { - return ex.toStatus(); - } + Query query(BSON(TagsType::ns(collectionNs) + << TagsType::min() << BSON("$lte" << chunk.getMin()) << TagsType::max() + << BSON("$gte" << chunk.getMax()))); - if (tagDoc.isEmpty()) { - return std::string(""); - } + tagDoc = conn->findOne(TagsType::ConfigNS, query); + conn.done(); + } catch (const DBException& ex) { + return ex.toStatus(); + } - auto status = TagsType::fromBSON(tagDoc); - if (status.isOK()) { - return status.getValue().getTag(); - } + if (tagDoc.isEmpty()) { + return std::string(""); + } - return status.getStatus(); + auto status = TagsType::fromBSON(tagDoc); + if (status.isOK()) { + return status.getValue().getTag(); } - Status CatalogManagerLegacy::getAllShards(vector<ShardType>* shards) { - ScopedDbConnection conn(_configServerConnectionString, 30.0); - std::unique_ptr<DBClientCursor> cursor(_safeCursor(conn->query(ShardType::ConfigNS, - BSONObj()))); - while (cursor->more()) { - BSONObj shardObj = cursor->nextSafe(); + return status.getStatus(); +} - StatusWith<ShardType> shardRes = ShardType::fromBSON(shardObj); - if (!shardRes.isOK()) { - shards->clear(); - conn.done(); - return Status(ErrorCodes::FailedToParse, - str::stream() << "Failed to parse shard with id (" - << shardObj[ShardType::name()].toString() << "): " - << shardRes.getStatus().reason()); - } +Status CatalogManagerLegacy::getAllShards(vector<ShardType>* shards) { + ScopedDbConnection conn(_configServerConnectionString, 30.0); + std::unique_ptr<DBClientCursor> cursor( + _safeCursor(conn->query(ShardType::ConfigNS, BSONObj()))); + while (cursor->more()) { + BSONObj shardObj = cursor->nextSafe(); - shards->push_back(shardRes.getValue()); + StatusWith<ShardType> shardRes = ShardType::fromBSON(shardObj); + if (!shardRes.isOK()) { + shards->clear(); + conn.done(); + return Status(ErrorCodes::FailedToParse, + str::stream() << "Failed to parse shard with id (" + << shardObj[ShardType::name()].toString() + << "): " << shardRes.getStatus().reason()); } - conn.done(); - return Status::OK(); + shards->push_back(shardRes.getValue()); } - - bool CatalogManagerLegacy::isShardHost(const ConnectionString& connectionString) { - return _getShardCount(BSON(ShardType::host(connectionString.toString()))); + conn.done(); + + return Status::OK(); +} + +bool CatalogManagerLegacy::isShardHost(const ConnectionString& connectionString) { + return _getShardCount(BSON(ShardType::host(connectionString.toString()))); +} + +bool CatalogManagerLegacy::runUserManagementWriteCommand(const string& commandName, + const string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) { + DBClientMultiCommand dispatcher; + RawBSONSerializable requestCmdSerial(cmdObj); + for (const ConnectionString& configServer : _configServers) { + dispatcher.addCommand(configServer, dbname, requestCmdSerial); } - bool CatalogManagerLegacy::runUserManagementWriteCommand(const string& commandName, - const string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) { - DBClientMultiCommand dispatcher; - RawBSONSerializable requestCmdSerial(cmdObj); - for (const ConnectionString& configServer : _configServers) { - dispatcher.addCommand(configServer, dbname, requestCmdSerial); - } - - auto scopedDistLock = getDistLockManager()->lock("authorizationData", - commandName, - Seconds{5}); - if (!scopedDistLock.isOK()) { - return Command::appendCommandStatus(*result, scopedDistLock.getStatus()); - } - - dispatcher.sendAll(); + auto scopedDistLock = getDistLockManager()->lock("authorizationData", commandName, Seconds{5}); + if (!scopedDistLock.isOK()) { + return Command::appendCommandStatus(*result, scopedDistLock.getStatus()); + } - BSONObj responseObj; + dispatcher.sendAll(); - Status prevStatus{Status::OK()}; - Status currStatus{Status::OK()}; + BSONObj responseObj; - BSONObjBuilder responses; - unsigned failedCount = 0; - bool sameError = true; - while (dispatcher.numPending() > 0) { - ConnectionString host; - RawBSONSerializable responseCmdSerial; + Status prevStatus{Status::OK()}; + Status currStatus{Status::OK()}; - Status dispatchStatus = dispatcher.recvAny(&host, - &responseCmdSerial); + BSONObjBuilder responses; + unsigned failedCount = 0; + bool sameError = true; + while (dispatcher.numPending() > 0) { + ConnectionString host; + RawBSONSerializable responseCmdSerial; - if (!dispatchStatus.isOK()) { - return Command::appendCommandStatus(*result, dispatchStatus); - } + Status dispatchStatus = dispatcher.recvAny(&host, &responseCmdSerial); - responseObj = responseCmdSerial.toBSON(); - responses.append(host.toString(), responseObj); - - currStatus = Command::getStatusFromCommandResult(responseObj); - if (!currStatus.isOK()) { - // same error <=> adjacent error statuses are the same - if (failedCount > 0 && prevStatus != currStatus) { - sameError = false; - } - failedCount++; - prevStatus = currStatus; - } + if (!dispatchStatus.isOK()) { + return Command::appendCommandStatus(*result, dispatchStatus); } - if (failedCount == 0) { - result->appendElements(responseObj); - return true; - } + responseObj = responseCmdSerial.toBSON(); + responses.append(host.toString(), responseObj); - // if the command succeeds on at least one config server and fails on at least one, - // manual intervention is required - if (failedCount < _configServers.size()) { - Status status(ErrorCodes::ManualInterventionRequired, - str::stream() << "Config write was not consistent - " - << "user management command failed on at least one " - << "config server but passed on at least one other. " - << "Manual intervention may be required. " - << "Config responses: " << responses.obj().toString()); - return Command::appendCommandStatus(*result, status); + currStatus = Command::getStatusFromCommandResult(responseObj); + if (!currStatus.isOK()) { + // same error <=> adjacent error statuses are the same + if (failedCount > 0 && prevStatus != currStatus) { + sameError = false; + } + failedCount++; + prevStatus = currStatus; } + } - if (sameError) { - result->appendElements(responseObj); - return false; - } + if (failedCount == 0) { + result->appendElements(responseObj); + return true; + } + // if the command succeeds on at least one config server and fails on at least one, + // manual intervention is required + if (failedCount < _configServers.size()) { Status status(ErrorCodes::ManualInterventionRequired, str::stream() << "Config write was not consistent - " - << "user management command produced inconsistent results. " + << "user management command failed on at least one " + << "config server but passed on at least one other. " << "Manual intervention may be required. " << "Config responses: " << responses.obj().toString()); return Command::appendCommandStatus(*result, status); } - bool CatalogManagerLegacy::runUserManagementReadCommand(const string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) { - try { - // let SyncClusterConnection handle connecting to the first config server - // that is reachable and returns data - ScopedDbConnection conn(_configServerConnectionString, 30); - - BSONObj cmdResult; - const bool ok = conn->runCommand(dbname, cmdObj, cmdResult); - result->appendElements(cmdResult); - conn.done(); - return ok; - } - catch (const DBException& ex) { - return Command::appendCommandStatus(*result, ex.toStatus()); - } + if (sameError) { + result->appendElements(responseObj); + return false; } - Status CatalogManagerLegacy::applyChunkOpsDeprecated(const BSONArray& updateOps, - const BSONArray& preCondition) { - BSONObj cmd = BSON("applyOps" << updateOps << - "preCondition" << preCondition); - bool ok; + Status status(ErrorCodes::ManualInterventionRequired, + str::stream() << "Config write was not consistent - " + << "user management command produced inconsistent results. " + << "Manual intervention may be required. " + << "Config responses: " << responses.obj().toString()); + return Command::appendCommandStatus(*result, status); +} + +bool CatalogManagerLegacy::runUserManagementReadCommand(const string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) { + try { + // let SyncClusterConnection handle connecting to the first config server + // that is reachable and returns data + ScopedDbConnection conn(_configServerConnectionString, 30); + BSONObj cmdResult; - try { - ScopedDbConnection conn(_configServerConnectionString, 30); - ok = conn->runCommand("config", cmd, cmdResult); - conn.done(); - } - catch (const DBException& ex) { - return ex.toStatus(); - } + const bool ok = conn->runCommand(dbname, cmdObj, cmdResult); + result->appendElements(cmdResult); + conn.done(); + return ok; + } catch (const DBException& ex) { + return Command::appendCommandStatus(*result, ex.toStatus()); + } +} + +Status CatalogManagerLegacy::applyChunkOpsDeprecated(const BSONArray& updateOps, + const BSONArray& preCondition) { + BSONObj cmd = BSON("applyOps" << updateOps << "preCondition" << preCondition); + bool ok; + BSONObj cmdResult; + try { + ScopedDbConnection conn(_configServerConnectionString, 30); + ok = conn->runCommand("config", cmd, cmdResult); + conn.done(); + } catch (const DBException& ex) { + return ex.toStatus(); + } - if (!ok) { - string errMsg(str::stream() << "Unable to save chunk ops. Command: " - << cmd << ". Result: " << cmdResult); + if (!ok) { + string errMsg(str::stream() << "Unable to save chunk ops. Command: " << cmd + << ". Result: " << cmdResult); - return Status(ErrorCodes::OperationFailed, errMsg); - } + return Status(ErrorCodes::OperationFailed, errMsg); + } - return Status::OK(); + return Status::OK(); +} + +void CatalogManagerLegacy::writeConfigServerDirect(const BatchedCommandRequest& request, + BatchedCommandResponse* response) { + // check if config servers are consistent + if (!_isConsistentFromLastCheck()) { + toBatchError(Status(ErrorCodes::ConfigServersInconsistent, + "Data inconsistency detected amongst config servers"), + response); + return; } - void CatalogManagerLegacy::writeConfigServerDirect(const BatchedCommandRequest& request, - BatchedCommandResponse* response) { + // We only support batch sizes of one for config writes + if (request.sizeWriteOps() != 1) { + toBatchError(Status(ErrorCodes::InvalidOptions, + str::stream() << "Writes to config servers must have batch size of 1, " + << "found " << request.sizeWriteOps()), + response); - // check if config servers are consistent - if (!_isConsistentFromLastCheck()) { - toBatchError( - Status(ErrorCodes::ConfigServersInconsistent, - "Data inconsistency detected amongst config servers"), - response); - return; - } + return; + } - // We only support batch sizes of one for config writes - if (request.sizeWriteOps() != 1) { - toBatchError( - Status(ErrorCodes::InvalidOptions, - str::stream() << "Writes to config servers must have batch size of 1, " - << "found " << request.sizeWriteOps()), - response); + // We only support {w: 0}, {w: 1}, and {w: 'majority'} write concern for config writes + if (request.isWriteConcernSet() && !validConfigWC(request.getWriteConcern())) { + toBatchError(Status(ErrorCodes::InvalidOptions, + str::stream() << "Invalid write concern for write to " + << "config servers: " << request.getWriteConcern()), + response); - return; - } + return; + } - // We only support {w: 0}, {w: 1}, and {w: 'majority'} write concern for config writes - if (request.isWriteConcernSet() && !validConfigWC(request.getWriteConcern())) { + DBClientMultiCommand dispatcher; + if (_configServers.size() > 1) { + // We can't support no-_id upserts to multiple config servers - the _ids will differ + if (BatchedCommandRequest::containsNoIDUpsert(request)) { toBatchError( Status(ErrorCodes::InvalidOptions, - str::stream() << "Invalid write concern for write to " - << "config servers: " << request.getWriteConcern()), + str::stream() << "upserts to multiple config servers must include _id"), response); - return; } - - DBClientMultiCommand dispatcher; - if (_configServers.size() > 1) { - // We can't support no-_id upserts to multiple config servers - the _ids will differ - if (BatchedCommandRequest::containsNoIDUpsert(request)) { - toBatchError( - Status(ErrorCodes::InvalidOptions, - str::stream() << "upserts to multiple config servers must include _id"), - response); - return; - } - } - - ConfigCoordinator exec(&dispatcher, _configServerConnectionString); - exec.executeBatch(request, response); } - Status CatalogManagerLegacy::_checkDbDoesNotExist(const std::string& dbName) const { - ScopedDbConnection conn(_configServerConnectionString, 30); + ConfigCoordinator exec(&dispatcher, _configServerConnectionString); + exec.executeBatch(request, response); +} - BSONObjBuilder b; - b.appendRegex(DatabaseType::name(), - (string)"^" + pcrecpp::RE::QuoteMeta(dbName) + "$", "i"); +Status CatalogManagerLegacy::_checkDbDoesNotExist(const std::string& dbName) const { + ScopedDbConnection conn(_configServerConnectionString, 30); - BSONObj dbObj = conn->findOne(DatabaseType::ConfigNS, b.obj()); - conn.done(); + BSONObjBuilder b; + b.appendRegex(DatabaseType::name(), (string) "^" + pcrecpp::RE::QuoteMeta(dbName) + "$", "i"); - // If our name is exactly the same as the name we want, try loading - // the database again. - if (!dbObj.isEmpty() && dbObj[DatabaseType::name()].String() == dbName) { - return Status(ErrorCodes::NamespaceExists, - str::stream() << "database " << dbName << " already exists"); - } + BSONObj dbObj = conn->findOne(DatabaseType::ConfigNS, b.obj()); + conn.done(); - if (!dbObj.isEmpty()) { - return Status(ErrorCodes::DatabaseDifferCase, - str::stream() << "can't have 2 databases that just differ on case " - << " have: " << dbObj[DatabaseType::name()].String() - << " want to add: " << dbName); - } + // If our name is exactly the same as the name we want, try loading + // the database again. + if (!dbObj.isEmpty() && dbObj[DatabaseType::name()].String() == dbName) { + return Status(ErrorCodes::NamespaceExists, + str::stream() << "database " << dbName << " already exists"); + } - return Status::OK(); + if (!dbObj.isEmpty()) { + return Status(ErrorCodes::DatabaseDifferCase, + str::stream() << "can't have 2 databases that just differ on case " + << " have: " << dbObj[DatabaseType::name()].String() + << " want to add: " << dbName); } - StatusWith<string> CatalogManagerLegacy::_getNewShardName() const { - BSONObj o; - { - ScopedDbConnection conn(_configServerConnectionString, 30); - o = conn->findOne(ShardType::ConfigNS, - Query(fromjson("{" + ShardType::name() + ": /^shard/}")) - .sort(BSON(ShardType::name() << -1 ))); - conn.done(); - } + return Status::OK(); +} - int count = 0; - if (!o.isEmpty()) { - string last = o[ShardType::name()].String(); - std::istringstream is(last.substr(5)); - is >> count; - count++; - } +StatusWith<string> CatalogManagerLegacy::_getNewShardName() const { + BSONObj o; + { + ScopedDbConnection conn(_configServerConnectionString, 30); + o = conn->findOne(ShardType::ConfigNS, + Query(fromjson("{" + ShardType::name() + ": /^shard/}")) + .sort(BSON(ShardType::name() << -1))); + conn.done(); + } - // TODO fix so that we can have more than 10000 automatically generated shard names - if (count < 9999) { - std::stringstream ss; - ss << "shard" << std::setfill('0') << std::setw(4) << count; - return ss.str(); - } + int count = 0; + if (!o.isEmpty()) { + string last = o[ShardType::name()].String(); + std::istringstream is(last.substr(5)); + is >> count; + count++; + } - return Status(ErrorCodes::OperationFailed, - "unable to generate new shard name"); + // TODO fix so that we can have more than 10000 automatically generated shard names + if (count < 9999) { + std::stringstream ss; + ss << "shard" << std::setfill('0') << std::setw(4) << count; + return ss.str(); } - size_t CatalogManagerLegacy::_getShardCount(const BSONObj& query) const { - ScopedDbConnection conn(_configServerConnectionString, 30.0); - long long shardCount = conn->count(ShardType::ConfigNS, query); - conn.done(); + return Status(ErrorCodes::OperationFailed, "unable to generate new shard name"); +} - return shardCount; - } +size_t CatalogManagerLegacy::_getShardCount(const BSONObj& query) const { + ScopedDbConnection conn(_configServerConnectionString, 30.0); + long long shardCount = conn->count(ShardType::ConfigNS, query); + conn.done(); - DistLockManager* CatalogManagerLegacy::getDistLockManager() { - invariant(_distLockManager); - return _distLockManager.get(); - } + return shardCount; +} - bool CatalogManagerLegacy::_checkConfigServersConsistent(const unsigned tries) const { - if (tries <= 0) - return false; +DistLockManager* CatalogManagerLegacy::getDistLockManager() { + invariant(_distLockManager); + return _distLockManager.get(); +} - unsigned firstGood = 0; - int up = 0; - vector<BSONObj> res; +bool CatalogManagerLegacy::_checkConfigServersConsistent(const unsigned tries) const { + if (tries <= 0) + return false; - // The last error we saw on a config server - string errMsg; + unsigned firstGood = 0; + int up = 0; + vector<BSONObj> res; - for (unsigned i = 0; i < _configServers.size(); i++) { - BSONObj result; - std::unique_ptr<ScopedDbConnection> conn; - - try { - conn.reset(new ScopedDbConnection(_configServers[i], 30.0)); - - if (!conn->get()->runCommand("config", - BSON("dbhash" << 1 << - "collections" << BSON_ARRAY("chunks" << - "databases" << - "collections" << - "shards" << - "version")), - result)) { - - errMsg = result["errmsg"].eoo() ? "" : result["errmsg"].String(); - if (!result["assertion"].eoo()) errMsg = result["assertion"].String(); - - warning() << "couldn't check dbhash on config server " << _configServers[i] - << causedBy(result.toString()); - - result = BSONObj(); - } - else { - result = result.getOwned(); - if (up == 0) - firstGood = i; - up++; - } - conn->done(); - } - catch (const DBException& e) { - if (conn) { - conn->kill(); - } + // The last error we saw on a config server + string errMsg; - // We need to catch DBExceptions b/c sometimes we throw them - // instead of socket exceptions when findN fails + for (unsigned i = 0; i < _configServers.size(); i++) { + BSONObj result; + std::unique_ptr<ScopedDbConnection> conn; - errMsg = e.toString(); - warning() << " couldn't check dbhash on config server " - << _configServers[i] << causedBy(e); + try { + conn.reset(new ScopedDbConnection(_configServers[i], 30.0)); + + if (!conn->get()->runCommand( + "config", + BSON("dbhash" << 1 << "collections" << BSON_ARRAY("chunks" + << "databases" + << "collections" + << "shards" + << "version")), + result)) { + errMsg = result["errmsg"].eoo() ? "" : result["errmsg"].String(); + if (!result["assertion"].eoo()) + errMsg = result["assertion"].String(); + + warning() << "couldn't check dbhash on config server " << _configServers[i] + << causedBy(result.toString()); + + result = BSONObj(); + } else { + result = result.getOwned(); + if (up == 0) + firstGood = i; + up++; + } + conn->done(); + } catch (const DBException& e) { + if (conn) { + conn->kill(); } - res.push_back(result); - } - if (_configServers.size() == 1) - return true; + // We need to catch DBExceptions b/c sometimes we throw them + // instead of socket exceptions when findN fails - if (up == 0) { - // Use a ptr to error so if empty we won't add causedby - error() << "no config servers successfully contacted" << causedBy(&errMsg); - return false; - } - else if (up == 1) { - warning() << "only 1 config server reachable, continuing"; - return true; + errMsg = e.toString(); + warning() << " couldn't check dbhash on config server " << _configServers[i] + << causedBy(e); } + res.push_back(result); + } - BSONObj base = res[firstGood]; - for (unsigned i = firstGood+1; i < res.size(); i++) { - if (res[i].isEmpty()) - continue; + if (_configServers.size() == 1) + return true; - string chunksHash1 = base.getFieldDotted("collections.chunks"); - string chunksHash2 = res[i].getFieldDotted("collections.chunks"); + if (up == 0) { + // Use a ptr to error so if empty we won't add causedby + error() << "no config servers successfully contacted" << causedBy(&errMsg); + return false; + } else if (up == 1) { + warning() << "only 1 config server reachable, continuing"; + return true; + } - string databaseHash1 = base.getFieldDotted("collections.databases"); - string databaseHash2 = res[i].getFieldDotted("collections.databases"); + BSONObj base = res[firstGood]; + for (unsigned i = firstGood + 1; i < res.size(); i++) { + if (res[i].isEmpty()) + continue; - string collectionsHash1 = base.getFieldDotted("collections.collections"); - string collectionsHash2 = res[i].getFieldDotted("collections.collections"); + string chunksHash1 = base.getFieldDotted("collections.chunks"); + string chunksHash2 = res[i].getFieldDotted("collections.chunks"); - string shardHash1 = base.getFieldDotted("collections.shards"); - string shardHash2 = res[i].getFieldDotted("collections.shards"); + string databaseHash1 = base.getFieldDotted("collections.databases"); + string databaseHash2 = res[i].getFieldDotted("collections.databases"); - string versionHash1 = base.getFieldDotted("collections.version") ; - string versionHash2 = res[i].getFieldDotted("collections.version"); + string collectionsHash1 = base.getFieldDotted("collections.collections"); + string collectionsHash2 = res[i].getFieldDotted("collections.collections"); - if (chunksHash1 == chunksHash2 && - databaseHash1 == databaseHash2 && - collectionsHash1 == collectionsHash2 && - shardHash1 == shardHash2 && - versionHash1 == versionHash2) { - continue; - } + string shardHash1 = base.getFieldDotted("collections.shards"); + string shardHash2 = res[i].getFieldDotted("collections.shards"); - warning() << "config servers " << _configServers[firstGood].toString() - << " and " << _configServers[i].toString() << " differ"; - if (tries <= 1) { - error() << ": " << base["collections"].Obj() - << " vs " << res[i]["collections"].Obj(); - return false; - } + string versionHash1 = base.getFieldDotted("collections.version"); + string versionHash2 = res[i].getFieldDotted("collections.version"); - return _checkConfigServersConsistent(tries - 1); + if (chunksHash1 == chunksHash2 && databaseHash1 == databaseHash2 && + collectionsHash1 == collectionsHash2 && shardHash1 == shardHash2 && + versionHash1 == versionHash2) { + continue; } - return true; + warning() << "config servers " << _configServers[firstGood].toString() << " and " + << _configServers[i].toString() << " differ"; + if (tries <= 1) { + error() << ": " << base["collections"].Obj() << " vs " << res[i]["collections"].Obj(); + return false; + } + + return _checkConfigServersConsistent(tries - 1); } - void CatalogManagerLegacy::_consistencyChecker() { - stdx::unique_lock<stdx::mutex> lk(_mutex); - while (!_inShutdown) { - lk.unlock(); - const bool isConsistent = _checkConfigServersConsistent(); + return true; +} - lk.lock(); - _consistentFromLastCheck = isConsistent; - if (_inShutdown) break; - _consistencyCheckerCV.wait_for(lk, Seconds(60)); - } - LOG(1) << "Consistency checker thread shutting down"; - } +void CatalogManagerLegacy::_consistencyChecker() { + stdx::unique_lock<stdx::mutex> lk(_mutex); + while (!_inShutdown) { + lk.unlock(); + const bool isConsistent = _checkConfigServersConsistent(); - bool CatalogManagerLegacy::_isConsistentFromLastCheck() { - stdx::unique_lock<stdx::mutex> lk(_mutex); - return _consistentFromLastCheck; + lk.lock(); + _consistentFromLastCheck = isConsistent; + if (_inShutdown) + break; + _consistencyCheckerCV.wait_for(lk, Seconds(60)); } + LOG(1) << "Consistency checker thread shutting down"; +} + +bool CatalogManagerLegacy::_isConsistentFromLastCheck() { + stdx::unique_lock<stdx::mutex> lk(_mutex); + return _consistentFromLastCheck; +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/legacy/catalog_manager_legacy.h b/src/mongo/s/catalog/legacy/catalog_manager_legacy.h index aff0ce6d185..5a762a1aebc 100644 --- a/src/mongo/s/catalog/legacy/catalog_manager_legacy.h +++ b/src/mongo/s/catalog/legacy/catalog_manager_legacy.h @@ -39,173 +39,170 @@ namespace mongo { +/** + * Implements the catalog manager using the legacy 3-config server protocol. + */ +class CatalogManagerLegacy final : public CatalogManager { +public: + CatalogManagerLegacy(); + ~CatalogManagerLegacy(); + /** - * Implements the catalog manager using the legacy 3-config server protocol. + * Initializes the catalog manager with the hosts, which will be used as a configuration + * server. Can only be called once for the lifetime. */ - class CatalogManagerLegacy final : public CatalogManager { - public: - CatalogManagerLegacy(); - ~CatalogManagerLegacy(); + Status init(const ConnectionString& configCS); - /** - * Initializes the catalog manager with the hosts, which will be used as a configuration - * server. Can only be called once for the lifetime. - */ - Status init(const ConnectionString& configCS); + Status startup(bool upgrade) override; - Status startup(bool upgrade) override; + ConnectionString connectionString() const override; - ConnectionString connectionString() const override; + void shutDown() override; - void shutDown() override; + Status enableSharding(const std::string& dbName) override; - Status enableSharding(const std::string& dbName) override; + Status shardCollection(const std::string& ns, + const ShardKeyPattern& fieldsAndOrder, + bool unique, + std::vector<BSONObj>* initPoints, + std::set<ShardId>* initShardIds) override; - Status shardCollection(const std::string& ns, - const ShardKeyPattern& fieldsAndOrder, - bool unique, - std::vector<BSONObj>* initPoints, - std::set<ShardId>* initShardIds) override; + StatusWith<std::string> addShard(const std::string& name, + const ConnectionString& shardConnectionString, + const long long maxSize) override; - StatusWith<std::string> addShard(const std::string& name, - const ConnectionString& shardConnectionString, - const long long maxSize) override; + StatusWith<ShardDrainingStatus> removeShard(OperationContext* txn, + const std::string& name) override; - StatusWith<ShardDrainingStatus> removeShard(OperationContext* txn, - const std::string& name) override; + Status createDatabase(const std::string& dbName) override; - Status createDatabase(const std::string& dbName) override; + Status updateDatabase(const std::string& dbName, const DatabaseType& db) override; - Status updateDatabase(const std::string& dbName, const DatabaseType& db) override; + StatusWith<DatabaseType> getDatabase(const std::string& dbName) override; - StatusWith<DatabaseType> getDatabase(const std::string& dbName) override; + Status updateCollection(const std::string& collNs, const CollectionType& coll) override; - Status updateCollection(const std::string& collNs, const CollectionType& coll) override; + StatusWith<CollectionType> getCollection(const std::string& collNs) override; - StatusWith<CollectionType> getCollection(const std::string& collNs) override; + Status getCollections(const std::string* dbName, std::vector<CollectionType>* collections); - Status getCollections(const std::string* dbName, std::vector<CollectionType>* collections); + Status dropCollection(const std::string& collectionNs); - Status dropCollection(const std::string& collectionNs); + Status getDatabasesForShard(const std::string& shardName, + std::vector<std::string>* dbs) override; - Status getDatabasesForShard(const std::string& shardName, - std::vector<std::string>* dbs) override; + Status getChunks(const Query& query, int nToReturn, std::vector<ChunkType>* chunks) override; - Status getChunks(const Query& query, - int nToReturn, - std::vector<ChunkType>* chunks) override; + Status getTagsForCollection(const std::string& collectionNs, + std::vector<TagsType>* tags) override; - Status getTagsForCollection(const std::string& collectionNs, - std::vector<TagsType>* tags) override; + StatusWith<std::string> getTagForChunk(const std::string& collectionNs, + const ChunkType& chunk) override; - StatusWith<std::string> getTagForChunk(const std::string& collectionNs, - const ChunkType& chunk) override; + Status getAllShards(std::vector<ShardType>* shards) override; - Status getAllShards(std::vector<ShardType>* shards) override; + bool isShardHost(const ConnectionString& shardConnectionString) override; - bool isShardHost(const ConnectionString& shardConnectionString) override; - - /** - * Grabs a distributed lock and runs the command on all config servers. - */ - bool runUserManagementWriteCommand(const std::string& commandName, - const std::string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) override; + /** + * Grabs a distributed lock and runs the command on all config servers. + */ + bool runUserManagementWriteCommand(const std::string& commandName, + const std::string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) override; - bool runUserManagementReadCommand(const std::string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) override; + bool runUserManagementReadCommand(const std::string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) override; - Status applyChunkOpsDeprecated(const BSONArray& updateOps, - const BSONArray& preCondition) override; + Status applyChunkOpsDeprecated(const BSONArray& updateOps, + const BSONArray& preCondition) override; - void logAction(const ActionLogType& actionLog); + void logAction(const ActionLogType& actionLog); - void logChange(OperationContext* txn, - const std::string& what, - const std::string& ns, - const BSONObj& detail) override; + void logChange(OperationContext* txn, + const std::string& what, + const std::string& ns, + const BSONObj& detail) override; - StatusWith<SettingsType> getGlobalSettings(const std::string& key) override; + StatusWith<SettingsType> getGlobalSettings(const std::string& key) override; - void writeConfigServerDirect(const BatchedCommandRequest& request, - BatchedCommandResponse* response) override; + void writeConfigServerDirect(const BatchedCommandRequest& request, + BatchedCommandResponse* response) override; - DistLockManager* getDistLockManager() override; + DistLockManager* getDistLockManager() override; - private: +private: + /** + * Updates the config server's metadata to the current version. + */ + Status _checkAndUpgradeConfigMetadata(bool doUpgrade); - /** - * Updates the config server's metadata to the current version. - */ - Status _checkAndUpgradeConfigMetadata(bool doUpgrade); + /** + * Starts the thread that periodically checks data consistency amongst the config servers. + * Note: this is not thread safe and can only be called once for the lifetime. + */ + Status _startConfigServerChecker(); - /** - * Starts the thread that periodically checks data consistency amongst the config servers. - * Note: this is not thread safe and can only be called once for the lifetime. - */ - Status _startConfigServerChecker(); + /** + * Direct network check to see if a particular database does not already exist with the + * same name or different case. + */ + Status _checkDbDoesNotExist(const std::string& dbName) const; - /** - * Direct network check to see if a particular database does not already exist with the - * same name or different case. - */ - Status _checkDbDoesNotExist(const std::string& dbName) const; + /** + * Generates a new shard name "shard<xxxx>" + * where <xxxx> is an autoincrementing value and <xxxx> < 10000 + */ + StatusWith<std::string> _getNewShardName() const; - /** - * Generates a new shard name "shard<xxxx>" - * where <xxxx> is an autoincrementing value and <xxxx> < 10000 - */ - StatusWith<std::string> _getNewShardName() const; + /** + * Returns the number of shards recognized by the config servers + * in this sharded cluster. + * Optional: use query parameter to filter shard count. + */ + size_t _getShardCount(const BSONObj& query = {}) const; - /** - * Returns the number of shards recognized by the config servers - * in this sharded cluster. - * Optional: use query parameter to filter shard count. - */ - size_t _getShardCount(const BSONObj& query = {}) const; + /** + * Returns true if all config servers have the same state. + * If inconsistency detected on first attempt, checks at most 3 more times. + */ + bool _checkConfigServersConsistent(const unsigned tries = 4) const; - /** - * Returns true if all config servers have the same state. - * If inconsistency detected on first attempt, checks at most 3 more times. - */ - bool _checkConfigServersConsistent(const unsigned tries = 4) const; + /** + * Checks data consistency amongst config servers every 60 seconds. + */ + void _consistencyChecker(); - /** - * Checks data consistency amongst config servers every 60 seconds. - */ - void _consistencyChecker(); + /** + * Returns true if the config servers have the same contents since the last + * check was performed. + */ + bool _isConsistentFromLastCheck(); - /** - * Returns true if the config servers have the same contents since the last - * check was performed. - */ - bool _isConsistentFromLastCheck(); + // Parsed config server hosts, as specified on the command line. + ConnectionString _configServerConnectionString; + std::vector<ConnectionString> _configServers; - // Parsed config server hosts, as specified on the command line. - ConnectionString _configServerConnectionString; - std::vector<ConnectionString> _configServers; + // Distribted lock manager singleton. + std::unique_ptr<DistLockManager> _distLockManager; - // Distribted lock manager singleton. - std::unique_ptr<DistLockManager> _distLockManager; + // protects _inShutdown, _consistentFromLastCheck; used by _consistencyCheckerCV + stdx::mutex _mutex; - // protects _inShutdown, _consistentFromLastCheck; used by _consistencyCheckerCV - stdx::mutex _mutex; + // True if CatalogManagerLegacy::shutDown has been called. False, otherwise. + bool _inShutdown = false; - // True if CatalogManagerLegacy::shutDown has been called. False, otherwise. - bool _inShutdown = false; + // used by consistency checker thread to check if config + // servers are consistent + bool _consistentFromLastCheck = false; - // used by consistency checker thread to check if config - // servers are consistent - bool _consistentFromLastCheck = false; + // Thread that runs dbHash on config servers for checking data consistency. + stdx::thread _consistencyCheckerThread; - // Thread that runs dbHash on config servers for checking data consistency. - stdx::thread _consistencyCheckerThread; + // condition variable used by the consistency checker thread to wait + // for <= 60s, on every iteration, until shutDown is called + stdx::condition_variable _consistencyCheckerCV; +}; - // condition variable used by the consistency checker thread to wait - // for <= 60s, on every iteration, until shutDown is called - stdx::condition_variable _consistencyCheckerCV; - }; - -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/legacy/cluster_client_internal.cpp b/src/mongo/s/catalog/legacy/cluster_client_internal.cpp index 5af263b5e56..c8b60c9c9dc 100644 --- a/src/mongo/s/catalog/legacy/cluster_client_internal.cpp +++ b/src/mongo/s/catalog/legacy/cluster_client_internal.cpp @@ -44,188 +44,175 @@ namespace mongo { - using std::endl; - using std::string; - using std::unique_ptr; - using std::vector; - using mongoutils::str::stream; - - Status checkClusterMongoVersions(CatalogManager* catalogManager, - const string& minMongoVersion) { - - unique_ptr<ScopedDbConnection> connPtr; - - // - // Find mongos pings in config server - // - - try { - connPtr.reset(new ScopedDbConnection(catalogManager->connectionString(), 30)); - ScopedDbConnection& conn = *connPtr; - unique_ptr<DBClientCursor> cursor(_safeCursor(conn->query(MongosType::ConfigNS, - Query()))); - - while (cursor->more()) { - - BSONObj pingDoc = cursor->next(); - - MongosType ping; - string errMsg; - // NOTE: We don't care if the ping is invalid, legacy stuff will be - if (!ping.parseBSON(pingDoc, &errMsg)) { - warning() << "could not parse ping document: " << pingDoc << causedBy(errMsg) - << endl; - continue; - } +using std::endl; +using std::string; +using std::unique_ptr; +using std::vector; +using mongoutils::str::stream; + +Status checkClusterMongoVersions(CatalogManager* catalogManager, const string& minMongoVersion) { + unique_ptr<ScopedDbConnection> connPtr; + + // + // Find mongos pings in config server + // + + try { + connPtr.reset(new ScopedDbConnection(catalogManager->connectionString(), 30)); + ScopedDbConnection& conn = *connPtr; + unique_ptr<DBClientCursor> cursor(_safeCursor(conn->query(MongosType::ConfigNS, Query()))); + + while (cursor->more()) { + BSONObj pingDoc = cursor->next(); + + MongosType ping; + string errMsg; + // NOTE: We don't care if the ping is invalid, legacy stuff will be + if (!ping.parseBSON(pingDoc, &errMsg)) { + warning() << "could not parse ping document: " << pingDoc << causedBy(errMsg) + << endl; + continue; + } - string mongoVersion = "2.0"; - // Hack to determine older mongos versions from ping format - if (ping.isWaitingSet()) mongoVersion = "2.2"; - if (ping.isMongoVersionSet() && ping.getMongoVersion() != "") { - mongoVersion = ping.getMongoVersion(); - } + string mongoVersion = "2.0"; + // Hack to determine older mongos versions from ping format + if (ping.isWaitingSet()) + mongoVersion = "2.2"; + if (ping.isMongoVersionSet() && ping.getMongoVersion() != "") { + mongoVersion = ping.getMongoVersion(); + } - Date_t lastPing = ping.getPing(); + Date_t lastPing = ping.getPing(); - Minutes quietIntervalMins{0}; - Date_t currentJsTime = jsTime(); - if (currentJsTime >= lastPing) { - quietIntervalMins = duration_cast<Minutes>(currentJsTime - lastPing); - } + Minutes quietIntervalMins{0}; + Date_t currentJsTime = jsTime(); + if (currentJsTime >= lastPing) { + quietIntervalMins = duration_cast<Minutes>(currentJsTime - lastPing); + } - // We assume that anything that hasn't pinged in 5 minutes is probably down - if (quietIntervalMins >= Minutes{5}) { - log() << "stale mongos detected " << quietIntervalMins.count() - << " minutes ago, network location is " << pingDoc["_id"].String() - << ", not checking version"; - } - else { - if (versionCmp(mongoVersion, minMongoVersion) < 0) { - return Status(ErrorCodes::RemoteValidationError, - stream() << "version " << mongoVersion - << " detected on mongos at " - << ping.getName() - << ", but version >= " << minMongoVersion - << " required; you must wait 5 minutes " - << "after shutting down a pre-" << minMongoVersion - << " mongos"); - } + // We assume that anything that hasn't pinged in 5 minutes is probably down + if (quietIntervalMins >= Minutes{5}) { + log() << "stale mongos detected " << quietIntervalMins.count() + << " minutes ago, network location is " << pingDoc["_id"].String() + << ", not checking version"; + } else { + if (versionCmp(mongoVersion, minMongoVersion) < 0) { + return Status( + ErrorCodes::RemoteValidationError, + stream() << "version " << mongoVersion << " detected on mongos at " + << ping.getName() << ", but version >= " << minMongoVersion + << " required; you must wait 5 minutes " + << "after shutting down a pre-" << minMongoVersion << " mongos"); } } } - catch (const DBException& e) { - return e.toStatus("could not read mongos pings collection"); - } + } catch (const DBException& e) { + return e.toStatus("could not read mongos pings collection"); + } - connPtr->done(); + connPtr->done(); - // - // Load shards from config server - // + // + // Load shards from config server + // - vector<HostAndPort> servers; + vector<HostAndPort> servers; - try { - vector<ShardType> shards; - Status status = catalogManager->getAllShards(&shards); + try { + vector<ShardType> shards; + Status status = catalogManager->getAllShards(&shards); + if (!status.isOK()) { + return status; + } + + for (const ShardType& shard : shards) { + Status status = shard.validate(); if (!status.isOK()) { - return status; + return Status(ErrorCodes::UnsupportedFormat, + stream() << "shard " << shard.toBSON() + << " failed validation: " << causedBy(status)); } - for (const ShardType& shard : shards) { - Status status = shard.validate(); - if (!status.isOK()) { - return Status(ErrorCodes::UnsupportedFormat, - stream() << "shard " << shard.toBSON() - << " failed validation: " << causedBy(status)); - } - - const auto shardConnStatus = ConnectionString::parse(shard.getHost()); - if (!shardConnStatus.isOK()) { - return Status(ErrorCodes::UnsupportedFormat, - stream() << "invalid shard host " << shard.getHost() - << " read from the config server" - << shardConnStatus.getStatus().toString()); - } - - vector<HostAndPort> shardServers = shardConnStatus.getValue().getServers(); - servers.insert(servers.end(), shardServers.begin(), shardServers.end()); + const auto shardConnStatus = ConnectionString::parse(shard.getHost()); + if (!shardConnStatus.isOK()) { + return Status(ErrorCodes::UnsupportedFormat, + stream() << "invalid shard host " << shard.getHost() + << " read from the config server" + << shardConnStatus.getStatus().toString()); } - } - catch (const DBException& e) { - return e.toStatus("could not read shards collection"); - } - - // Add config servers to list of servers to check version against - vector<HostAndPort> configServers = catalogManager->connectionString().getServers(); - servers.insert(servers.end(), configServers.begin(), configServers.end()); - // - // We've now got all the shard info from the config server, start contacting the shards - // and config servers and verifying their versions. - // + vector<HostAndPort> shardServers = shardConnStatus.getValue().getServers(); + servers.insert(servers.end(), shardServers.begin(), shardServers.end()); + } + } catch (const DBException& e) { + return e.toStatus("could not read shards collection"); + } - for (vector<HostAndPort>::iterator serverIt = servers.begin(); - serverIt != servers.end(); ++serverIt) { + // Add config servers to list of servers to check version against + vector<HostAndPort> configServers = catalogManager->connectionString().getServers(); + servers.insert(servers.end(), configServers.begin(), configServers.end()); - // Note: This will *always* be a single-host connection - ConnectionString serverLoc(*serverIt); - dassert(serverLoc.type() == ConnectionString::MASTER || - serverLoc.type() == ConnectionString::CUSTOM); // for dbtests + // + // We've now got all the shard info from the config server, start contacting the shards + // and config servers and verifying their versions. + // - log() << "checking that version of host " << serverLoc << " is compatible with " - << minMongoVersion << endl; + for (vector<HostAndPort>::iterator serverIt = servers.begin(); serverIt != servers.end(); + ++serverIt) { + // Note: This will *always* be a single-host connection + ConnectionString serverLoc(*serverIt); + dassert(serverLoc.type() == ConnectionString::MASTER || + serverLoc.type() == ConnectionString::CUSTOM); // for dbtests - unique_ptr<ScopedDbConnection> serverConnPtr; + log() << "checking that version of host " << serverLoc << " is compatible with " + << minMongoVersion << endl; - bool resultOk; - BSONObj buildInfo; + unique_ptr<ScopedDbConnection> serverConnPtr; - try { - serverConnPtr.reset(new ScopedDbConnection(serverLoc, 30)); - ScopedDbConnection& serverConn = *serverConnPtr; + bool resultOk; + BSONObj buildInfo; - resultOk = serverConn->runCommand("admin", - BSON("buildInfo" << 1), - buildInfo); - } - catch (const DBException& e) { - warning() << "could not run buildInfo command on " << serverLoc.toString() << " " - << causedBy(e) << ". Please ensure that this server is up and at a " - "version >= " - << minMongoVersion; - continue; - } + try { + serverConnPtr.reset(new ScopedDbConnection(serverLoc, 30)); + ScopedDbConnection& serverConn = *serverConnPtr; + + resultOk = serverConn->runCommand("admin", BSON("buildInfo" << 1), buildInfo); + } catch (const DBException& e) { + warning() << "could not run buildInfo command on " << serverLoc.toString() << " " + << causedBy(e) << ". Please ensure that this server is up and at a " + "version >= " << minMongoVersion; + continue; + } - // TODO: Make running commands saner such that we can consolidate error handling - if (!resultOk) { - return Status(ErrorCodes::UnknownError, - stream() << DBClientConnection::getLastErrorString(buildInfo) - << causedBy(buildInfo.toString())); - } + // TODO: Make running commands saner such that we can consolidate error handling + if (!resultOk) { + return Status(ErrorCodes::UnknownError, + stream() << DBClientConnection::getLastErrorString(buildInfo) + << causedBy(buildInfo.toString())); + } - serverConnPtr->done(); + serverConnPtr->done(); - verify(buildInfo["version"].type() == String); - string mongoVersion = buildInfo["version"].String(); + verify(buildInfo["version"].type() == String); + string mongoVersion = buildInfo["version"].String(); - if (versionCmp(mongoVersion, minMongoVersion) < 0) { - return Status(ErrorCodes::RemoteValidationError, - stream() << "version " << mongoVersion << " detected on mongo " - "server at " << serverLoc.toString() << - ", but version >= " << minMongoVersion << " required"); - } + if (versionCmp(mongoVersion, minMongoVersion) < 0) { + return Status(ErrorCodes::RemoteValidationError, + stream() << "version " << mongoVersion << " detected on mongo " + "server at " + << serverLoc.toString() << ", but version >= " << minMongoVersion + << " required"); } - - return Status::OK(); } - // Helper function for safe cursors - DBClientCursor* _safeCursor(unique_ptr<DBClientCursor> cursor) { - // TODO: Make error handling more consistent, it's annoying that cursors error out by - // throwing exceptions *and* being empty - uassert(16625, str::stream() << "cursor not found, transport error", cursor.get()); - return cursor.release(); - } + return Status::OK(); +} +// Helper function for safe cursors +DBClientCursor* _safeCursor(unique_ptr<DBClientCursor> cursor) { + // TODO: Make error handling more consistent, it's annoying that cursors error out by + // throwing exceptions *and* being empty + uassert(16625, str::stream() << "cursor not found, transport error", cursor.get()); + return cursor.release(); +} } diff --git a/src/mongo/s/catalog/legacy/cluster_client_internal.h b/src/mongo/s/catalog/legacy/cluster_client_internal.h index 8a140e171cd..ae60e519fbd 100644 --- a/src/mongo/s/catalog/legacy/cluster_client_internal.h +++ b/src/mongo/s/catalog/legacy/cluster_client_internal.h @@ -34,27 +34,26 @@ namespace mongo { - class CatalogManager; - class DBClientCursor; - class Status; - - /** - * Tries to check the versions of all active hosts in a cluster. Not 100% accurate, but pretty - * effective if hosts are reachable. - * - * Returns OK if hosts are compatible as far as we know, RemoteValidationError if hosts are not - * compatible, and an error Status if anything else goes wrong. - */ - Status checkClusterMongoVersions(CatalogManager* catalogManager, - const std::string& minMongoVersion); - - // - // Needed to normalize exception behavior of connections and cursors - // TODO: Remove when we refactor the client connection interface to something more consistent. - // - - // Helper function which throws for invalid cursor initialization. - // Note: cursor ownership will be passed to this function. - DBClientCursor* _safeCursor(std::unique_ptr<DBClientCursor> cursor); +class CatalogManager; +class DBClientCursor; +class Status; +/** + * Tries to check the versions of all active hosts in a cluster. Not 100% accurate, but pretty + * effective if hosts are reachable. + * + * Returns OK if hosts are compatible as far as we know, RemoteValidationError if hosts are not + * compatible, and an error Status if anything else goes wrong. + */ +Status checkClusterMongoVersions(CatalogManager* catalogManager, + const std::string& minMongoVersion); + +// +// Needed to normalize exception behavior of connections and cursors +// TODO: Remove when we refactor the client connection interface to something more consistent. +// + +// Helper function which throws for invalid cursor initialization. +// Note: cursor ownership will be passed to this function. +DBClientCursor* _safeCursor(std::unique_ptr<DBClientCursor> cursor); } diff --git a/src/mongo/s/catalog/legacy/config_coordinator.cpp b/src/mongo/s/catalog/legacy/config_coordinator.cpp index e7f33095be9..f4d02c5433a 100644 --- a/src/mongo/s/catalog/legacy/config_coordinator.cpp +++ b/src/mongo/s/catalog/legacy/config_coordinator.cpp @@ -43,431 +43,424 @@ namespace mongo { - using std::string; - using std::vector; +using std::string; +using std::vector; namespace { - /** - * A BSON serializable object representing a setShardVersion command request. - */ - class SSVRequest : public BSONSerializable { - MONGO_DISALLOW_COPYING(SSVRequest); - public: - SSVRequest(const std::string& configDBString) : _configDBString(configDBString) { - - } - - bool isValid(std::string* errMsg) const { - return true; - } +/** + * A BSON serializable object representing a setShardVersion command request. + */ +class SSVRequest : public BSONSerializable { + MONGO_DISALLOW_COPYING(SSVRequest); - /** Returns the BSON representation of the entry. */ - BSONObj toBSON() const { - BSONObjBuilder builder; - builder.append("setShardVersion", ""); // empty ns for init - builder.append("configdb", _configDBString); - builder.append("init", true); - builder.append("authoritative", true); - return builder.obj(); - } +public: + SSVRequest(const std::string& configDBString) : _configDBString(configDBString) {} - bool parseBSON(const BSONObj& source, std::string* errMsg) { - // Not implemented - invariant(false); - return false; - } + bool isValid(std::string* errMsg) const { + return true; + } - void clear() { - // Not implemented - invariant(false); - } + /** Returns the BSON representation of the entry. */ + BSONObj toBSON() const { + BSONObjBuilder builder; + builder.append("setShardVersion", ""); // empty ns for init + builder.append("configdb", _configDBString); + builder.append("init", true); + builder.append("authoritative", true); + return builder.obj(); + } - string toString() const { - return toBSON().toString(); - } + bool parseBSON(const BSONObj& source, std::string* errMsg) { + // Not implemented + invariant(false); + return false; + } - private: - const std::string _configDBString; - }; + void clear() { + // Not implemented + invariant(false); + } - /** - * A BSON serializable object representing a setShardVersion command response. - */ - class SSVResponse : public BSONSerializable { - MONGO_DISALLOW_COPYING(SSVResponse); - public: - static const BSONField<int> ok; - static const BSONField<int> errCode; - static const BSONField<string> errMessage; + string toString() const { + return toBSON().toString(); + } +private: + const std::string _configDBString; +}; - SSVResponse() { - clear(); - } +/** + * A BSON serializable object representing a setShardVersion command response. + */ +class SSVResponse : public BSONSerializable { + MONGO_DISALLOW_COPYING(SSVResponse); - bool isValid(std::string* errMsg) const { - return _isOkSet; - } +public: + static const BSONField<int> ok; + static const BSONField<int> errCode; + static const BSONField<string> errMessage; - BSONObj toBSON() const { - BSONObjBuilder builder; - if (_isOkSet) builder << ok(_ok); - if (_isErrCodeSet) builder << errCode(_errCode); - if (_isErrMessageSet) builder << errMessage(_errMessage); + SSVResponse() { + clear(); + } - return builder.obj(); - } + bool isValid(std::string* errMsg) const { + return _isOkSet; + } - bool parseBSON(const BSONObj& source, std::string* errMsg) { - FieldParser::FieldState result; + BSONObj toBSON() const { + BSONObjBuilder builder; - result = FieldParser::extractNumber(source, ok, &_ok, errMsg); - if (result == FieldParser::FIELD_INVALID) { - return false; - } - _isOkSet = result != FieldParser::FIELD_NONE; + if (_isOkSet) + builder << ok(_ok); + if (_isErrCodeSet) + builder << errCode(_errCode); + if (_isErrMessageSet) + builder << errMessage(_errMessage); - result = FieldParser::extract(source, errCode, &_errCode, errMsg); - if (result == FieldParser::FIELD_INVALID) { - return false; - } - _isErrCodeSet = result != FieldParser::FIELD_NONE; + return builder.obj(); + } - result = FieldParser::extract(source, errMessage, &_errMessage, errMsg); - if (result == FieldParser::FIELD_INVALID) { - return false; - } - _isErrMessageSet = result != FieldParser::FIELD_NONE; + bool parseBSON(const BSONObj& source, std::string* errMsg) { + FieldParser::FieldState result; - return true; + result = FieldParser::extractNumber(source, ok, &_ok, errMsg); + if (result == FieldParser::FIELD_INVALID) { + return false; } + _isOkSet = result != FieldParser::FIELD_NONE; - void clear() { - _ok = false; - _isOkSet = false; - - _errCode = 0; - _isErrCodeSet = false; - - _errMessage = ""; - _isErrMessageSet = false; + result = FieldParser::extract(source, errCode, &_errCode, errMsg); + if (result == FieldParser::FIELD_INVALID) { + return false; } + _isErrCodeSet = result != FieldParser::FIELD_NONE; - string toString() const { - return toBSON().toString(); + result = FieldParser::extract(source, errMessage, &_errMessage, errMsg); + if (result == FieldParser::FIELD_INVALID) { + return false; } + _isErrMessageSet = result != FieldParser::FIELD_NONE; - int getOk() { - dassert(_isOkSet); - return _ok; - } + return true; + } - void setOk(int ok) { - _ok = ok; - _isOkSet = true; - } + void clear() { + _ok = false; + _isOkSet = false; - int getErrCode() { - if (_isErrCodeSet) { - return _errCode; - } - else { - return errCode.getDefault(); - } - } + _errCode = 0; + _isErrCodeSet = false; - void setErrCode(int errCode) { - _errCode = errCode; - _isErrCodeSet = true; - } + _errMessage = ""; + _isErrMessageSet = false; + } - bool isErrCodeSet() const { - return _isErrCodeSet; - } + string toString() const { + return toBSON().toString(); + } - const string& getErrMessage() { - dassert(_isErrMessageSet); - return _errMessage; - } + int getOk() { + dassert(_isOkSet); + return _ok; + } - void setErrMessage(StringData errMsg) { - _errMessage = errMsg.toString(); - _isErrMessageSet = true; + void setOk(int ok) { + _ok = ok; + _isOkSet = true; + } + + int getErrCode() { + if (_isErrCodeSet) { + return _errCode; + } else { + return errCode.getDefault(); } + } - private: - int _ok; - bool _isOkSet; + void setErrCode(int errCode) { + _errCode = errCode; + _isErrCodeSet = true; + } - int _errCode; - bool _isErrCodeSet; + bool isErrCodeSet() const { + return _isErrCodeSet; + } - string _errMessage; - bool _isErrMessageSet; - }; + const string& getErrMessage() { + dassert(_isErrMessageSet); + return _errMessage; + } - const BSONField<int> SSVResponse::ok("ok"); - const BSONField<int> SSVResponse::errCode("code"); - const BSONField<string> SSVResponse::errMessage("errmsg"); + void setErrMessage(StringData errMsg) { + _errMessage = errMsg.toString(); + _isErrMessageSet = true; + } +private: + int _ok; + bool _isOkSet; - struct ConfigResponse { - ConnectionString configHost; - BatchedCommandResponse response; - }; + int _errCode; + bool _isErrCodeSet; - void buildErrorFrom(const Status& status, BatchedCommandResponse* response) { - response->setOk(false); - response->setErrCode(static_cast<int>(status.code())); - response->setErrMessage(status.reason()); + string _errMessage; + bool _isErrMessageSet; +}; - dassert(response->isValid(NULL)); - } +const BSONField<int> SSVResponse::ok("ok"); +const BSONField<int> SSVResponse::errCode("code"); +const BSONField<string> SSVResponse::errMessage("errmsg"); - bool areResponsesEqual(const BatchedCommandResponse& responseA, - const BatchedCommandResponse& responseB) { - // Note: This needs to also take into account comparing responses from legacy writes - // and write commands. +struct ConfigResponse { + ConnectionString configHost; + BatchedCommandResponse response; +}; - // TODO: Better reporting of why not equal - if (responseA.getOk() != responseB.getOk()) { - return false; - } +void buildErrorFrom(const Status& status, BatchedCommandResponse* response) { + response->setOk(false); + response->setErrCode(static_cast<int>(status.code())); + response->setErrMessage(status.reason()); - if (responseA.getN() != responseB.getN()) { - return false; - } + dassert(response->isValid(NULL)); +} - if (responseA.isUpsertDetailsSet()) { - // TODO: - } +bool areResponsesEqual(const BatchedCommandResponse& responseA, + const BatchedCommandResponse& responseB) { + // Note: This needs to also take into account comparing responses from legacy writes + // and write commands. - if (responseA.getOk()) { - return true; - } + // TODO: Better reporting of why not equal + if (responseA.getOk() != responseB.getOk()) { + return false; + } + + if (responseA.getN() != responseB.getN()) { + return false; + } - // TODO: Compare errors here + if (responseA.isUpsertDetailsSet()) { + // TODO: + } + if (responseA.getOk()) { return true; } - bool areAllResponsesEqual(const vector<ConfigResponse*>& responses) { - BatchedCommandResponse* lastResponse = NULL; + // TODO: Compare errors here - for (vector<ConfigResponse*>::const_iterator it = responses.begin(); - it != responses.end(); - ++it) { + return true; +} - BatchedCommandResponse* response = &(*it)->response; +bool areAllResponsesEqual(const vector<ConfigResponse*>& responses) { + BatchedCommandResponse* lastResponse = NULL; - if (lastResponse != NULL) { - if (!areResponsesEqual(*lastResponse, *response)) { - return false; - } - } + for (vector<ConfigResponse*>::const_iterator it = responses.begin(); it != responses.end(); + ++it) { + BatchedCommandResponse* response = &(*it)->response; - lastResponse = response; + if (lastResponse != NULL) { + if (!areResponsesEqual(*lastResponse, *response)) { + return false; + } } - return true; + lastResponse = response; } - void combineResponses(const vector<ConfigResponse*>& responses, - BatchedCommandResponse* clientResponse) { - - if (areAllResponsesEqual(responses)) { - responses.front()->response.cloneTo(clientResponse); - return; - } - - BSONObjBuilder builder; - for (vector<ConfigResponse*>::const_iterator it = responses.begin(); - it != responses.end(); - ++it) { + return true; +} - builder.append((*it)->configHost.toString(), (*it)->response.toBSON()); - } +void combineResponses(const vector<ConfigResponse*>& responses, + BatchedCommandResponse* clientResponse) { + if (areAllResponsesEqual(responses)) { + responses.front()->response.cloneTo(clientResponse); + return; + } - clientResponse->setOk(false); - clientResponse->setErrCode(ErrorCodes::ManualInterventionRequired); - clientResponse->setErrMessage("config write was not consistent, " - "manual intervention may be required. " - "config responses: " + builder.obj().toString()); + BSONObjBuilder builder; + for (vector<ConfigResponse*>::const_iterator it = responses.begin(); it != responses.end(); + ++it) { + builder.append((*it)->configHost.toString(), (*it)->response.toBSON()); } -} // namespace + clientResponse->setOk(false); + clientResponse->setErrCode(ErrorCodes::ManualInterventionRequired); + clientResponse->setErrMessage( + "config write was not consistent, " + "manual intervention may be required. " + "config responses: " + + builder.obj().toString()); +} +} // namespace - ConfigCoordinator::ConfigCoordinator(MultiCommandDispatch* dispatcher, - const ConnectionString& configServerConnectionString) - : _dispatcher(dispatcher), - _configServerConnectionString(configServerConnectionString) { - } +ConfigCoordinator::ConfigCoordinator(MultiCommandDispatch* dispatcher, + const ConnectionString& configServerConnectionString) + : _dispatcher(dispatcher), _configServerConnectionString(configServerConnectionString) {} - bool ConfigCoordinator::_checkConfigString(BatchedCommandResponse* clientResponse) { - // - // Send side - // +bool ConfigCoordinator::_checkConfigString(BatchedCommandResponse* clientResponse) { + // + // Send side + // - for (const HostAndPort& server : _configServerConnectionString.getServers()) { - SSVRequest ssvRequest(_configServerConnectionString.toString()); - _dispatcher->addCommand(ConnectionString(server), "admin", ssvRequest); - } + for (const HostAndPort& server : _configServerConnectionString.getServers()) { + SSVRequest ssvRequest(_configServerConnectionString.toString()); + _dispatcher->addCommand(ConnectionString(server), "admin", ssvRequest); + } - _dispatcher->sendAll(); + _dispatcher->sendAll(); - // - // Recv side - // + // + // Recv side + // - bool ssvError = false; - while (_dispatcher->numPending() > 0) { - ConnectionString configHost; - SSVResponse response; + bool ssvError = false; + while (_dispatcher->numPending() > 0) { + ConnectionString configHost; + SSVResponse response; - // We've got to recv everything, no matter what - even if some failed. - Status dispatchStatus = _dispatcher->recvAny(&configHost, &response); + // We've got to recv everything, no matter what - even if some failed. + Status dispatchStatus = _dispatcher->recvAny(&configHost, &response); - if (ssvError) { - // record only the first failure. - continue; - } + if (ssvError) { + // record only the first failure. + continue; + } - if (!dispatchStatus.isOK()) { - ssvError = true; - clientResponse->setOk(false); - clientResponse->setErrCode(static_cast<int>(dispatchStatus.code())); - clientResponse->setErrMessage(dispatchStatus.reason()); - } - else if (!response.getOk()) { - ssvError = true; - clientResponse->setOk(false); - clientResponse->setErrMessage(response.getErrMessage()); + if (!dispatchStatus.isOK()) { + ssvError = true; + clientResponse->setOk(false); + clientResponse->setErrCode(static_cast<int>(dispatchStatus.code())); + clientResponse->setErrMessage(dispatchStatus.reason()); + } else if (!response.getOk()) { + ssvError = true; + clientResponse->setOk(false); + clientResponse->setErrMessage(response.getErrMessage()); - if (response.isErrCodeSet()) { - clientResponse->setErrCode(response.getErrCode()); - } + if (response.isErrCodeSet()) { + clientResponse->setErrCode(response.getErrCode()); } } - - return !ssvError; } - /** - * The core config write functionality. - * - * Config writes run in two passes - the first is a quick check to ensure the config servers - * are all reachable, the second runs the actual write. - * - * TODO: Upgrade and move this logic to the config servers, a state machine implementation - * is probably the next step. - */ - void ConfigCoordinator::executeBatch(const BatchedCommandRequest& clientRequest, - BatchedCommandResponse* clientResponse) { - - const NamespaceString nss(clientRequest.getNS()); - - // Should never use it for anything other than DBs residing on the config server - dassert(nss.db() == "config" || nss.db() == "admin"); - dassert(clientRequest.sizeWriteOps() == 1u); - - // This is an opportunistic check that all config servers look healthy by calling - // getLastError on each one of them. If there was some form of write/journaling error, get - // last error would fail. - { - for (const HostAndPort& server : _configServerConnectionString.getServers()) { - _dispatcher->addCommand(ConnectionString(server), - "admin", - RawBSONSerializable(BSON("getLastError" << true << - "fsync" << true))); - } + return !ssvError; +} + +/** + * The core config write functionality. + * + * Config writes run in two passes - the first is a quick check to ensure the config servers + * are all reachable, the second runs the actual write. + * + * TODO: Upgrade and move this logic to the config servers, a state machine implementation + * is probably the next step. + */ +void ConfigCoordinator::executeBatch(const BatchedCommandRequest& clientRequest, + BatchedCommandResponse* clientResponse) { + const NamespaceString nss(clientRequest.getNS()); + + // Should never use it for anything other than DBs residing on the config server + dassert(nss.db() == "config" || nss.db() == "admin"); + dassert(clientRequest.sizeWriteOps() == 1u); + + // This is an opportunistic check that all config servers look healthy by calling + // getLastError on each one of them. If there was some form of write/journaling error, get + // last error would fail. + { + for (const HostAndPort& server : _configServerConnectionString.getServers()) { + _dispatcher->addCommand( + ConnectionString(server), + "admin", + RawBSONSerializable(BSON("getLastError" << true << "fsync" << true))); + } - _dispatcher->sendAll(); + _dispatcher->sendAll(); - bool error = false; - while (_dispatcher->numPending()) { - ConnectionString host; - RawBSONSerializable response; + bool error = false; + while (_dispatcher->numPending()) { + ConnectionString host; + RawBSONSerializable response; - Status status = _dispatcher->recvAny(&host, &response); - if (status.isOK()) { - BSONObj obj = response.toBSON(); + Status status = _dispatcher->recvAny(&host, &response); + if (status.isOK()) { + BSONObj obj = response.toBSON(); - LOG(3) << "Response " << obj.toString(); + LOG(3) << "Response " << obj.toString(); - // If the ok field is anything other than 1, count it as error - if (!obj["ok"].trueValue()) { - error = true; - log() << "Config server check for host " << host - << " returned error: " << response; - } - } - else { + // If the ok field is anything other than 1, count it as error + if (!obj["ok"].trueValue()) { error = true; log() << "Config server check for host " << host - << " failed with status: " << status; + << " returned error: " << response; } - } - - // All responses should have been gathered by this point - if (error) { - clientResponse->setOk(false); - clientResponse->setErrCode(ErrorCodes::RemoteValidationError); - clientResponse->setErrMessage("Could not verify that config servers were active" - " and reachable before write"); - return; + } else { + error = true; + log() << "Config server check for host " << host + << " failed with status: " << status; } } - if (!_checkConfigString(clientResponse)) { + // All responses should have been gathered by this point + if (error) { + clientResponse->setOk(false); + clientResponse->setErrCode(ErrorCodes::RemoteValidationError); + clientResponse->setErrMessage( + "Could not verify that config servers were active" + " and reachable before write"); return; } + } - // - // Do the actual writes - // + if (!_checkConfigString(clientResponse)) { + return; + } - BatchedCommandRequest configRequest( clientRequest.getBatchType() ); - clientRequest.cloneTo( &configRequest ); - configRequest.setNS( nss.coll() ); + // + // Do the actual writes + // - OwnedPointerVector<ConfigResponse> responsesOwned; - vector<ConfigResponse*>& responses = responsesOwned.mutableVector(); + BatchedCommandRequest configRequest(clientRequest.getBatchType()); + clientRequest.cloneTo(&configRequest); + configRequest.setNS(nss.coll()); - // - // Send the actual config writes - // + OwnedPointerVector<ConfigResponse> responsesOwned; + vector<ConfigResponse*>& responses = responsesOwned.mutableVector(); - // Get as many batches as we can at once - for (const HostAndPort& server : _configServerConnectionString.getServers()) { - _dispatcher->addCommand(ConnectionString(server), nss.db(), configRequest); - } + // + // Send the actual config writes + // - // Send them all out - _dispatcher->sendAll(); + // Get as many batches as we can at once + for (const HostAndPort& server : _configServerConnectionString.getServers()) { + _dispatcher->addCommand(ConnectionString(server), nss.db(), configRequest); + } - // - // Recv side - // + // Send them all out + _dispatcher->sendAll(); - while (_dispatcher->numPending() > 0) { - // Get the response - responses.push_back(new ConfigResponse()); + // + // Recv side + // - ConfigResponse& configResponse = *responses.back(); - Status dispatchStatus = _dispatcher->recvAny(&configResponse.configHost, - &configResponse.response); + while (_dispatcher->numPending() > 0) { + // Get the response + responses.push_back(new ConfigResponse()); - if (!dispatchStatus.isOK()) { - buildErrorFrom(dispatchStatus, &configResponse.response); - } - } + ConfigResponse& configResponse = *responses.back(); + Status dispatchStatus = + _dispatcher->recvAny(&configResponse.configHost, &configResponse.response); - combineResponses(responses, clientResponse); + if (!dispatchStatus.isOK()) { + buildErrorFrom(dispatchStatus, &configResponse.response); + } } -} // namespace mongo + combineResponses(responses, clientResponse); +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/legacy/config_coordinator.h b/src/mongo/s/catalog/legacy/config_coordinator.h index 3cd0301f23f..b1a3e18ca88 100644 --- a/src/mongo/s/catalog/legacy/config_coordinator.h +++ b/src/mongo/s/catalog/legacy/config_coordinator.h @@ -36,27 +36,26 @@ namespace mongo { - class MultiCommandDispatch; +class MultiCommandDispatch; - class ConfigCoordinator { - public: - ConfigCoordinator(MultiCommandDispatch* dispatcher, - const ConnectionString& configServerConnectionString); +class ConfigCoordinator { +public: + ConfigCoordinator(MultiCommandDispatch* dispatcher, + const ConnectionString& configServerConnectionString); - void executeBatch(const BatchedCommandRequest& request, BatchedCommandResponse* response); + void executeBatch(const BatchedCommandRequest& request, BatchedCommandResponse* response); - private: - /** - * Initialize configDB string in config server or if already initialized, - * check that it matches. Returns false if an error occured. - */ - bool _checkConfigString(BatchedCommandResponse* clientResponse); +private: + /** + * Initialize configDB string in config server or if already initialized, + * check that it matches. Returns false if an error occured. + */ + bool _checkConfigString(BatchedCommandResponse* clientResponse); - // Not owned here - MultiCommandDispatch* const _dispatcher; - - const ConnectionString _configServerConnectionString; - }; + // Not owned here + MultiCommandDispatch* const _dispatcher; + const ConnectionString _configServerConnectionString; +}; } diff --git a/src/mongo/s/catalog/legacy/config_upgrade.cpp b/src/mongo/s/catalog/legacy/config_upgrade.cpp index 2ed119d914f..3592d109e2b 100644 --- a/src/mongo/s/catalog/legacy/config_upgrade.cpp +++ b/src/mongo/s/catalog/legacy/config_upgrade.cpp @@ -52,519 +52,492 @@ namespace mongo { - using std::unique_ptr; - using std::make_pair; - using std::map; - using std::string; - using std::vector; - using str::stream; - - // Implemented in the respective steps' .cpp file - bool doUpgradeV0ToV7(CatalogManager* catalogManager, - const VersionType& lastVersionInfo, - std::string* errMsg); - - bool doUpgradeV6ToV7(CatalogManager* catalogManager, - const VersionType& lastVersionInfo, - std::string* errMsg); +using std::unique_ptr; +using std::make_pair; +using std::map; +using std::string; +using std::vector; +using str::stream; + +// Implemented in the respective steps' .cpp file +bool doUpgradeV0ToV7(CatalogManager* catalogManager, + const VersionType& lastVersionInfo, + std::string* errMsg); + +bool doUpgradeV6ToV7(CatalogManager* catalogManager, + const VersionType& lastVersionInfo, + std::string* errMsg); namespace { - struct VersionRange { - VersionRange(int _minCompatibleVersion, int _currentVersion) - : minCompatibleVersion(_minCompatibleVersion), - currentVersion(_currentVersion) { +struct VersionRange { + VersionRange(int _minCompatibleVersion, int _currentVersion) + : minCompatibleVersion(_minCompatibleVersion), currentVersion(_currentVersion) {} - } - - bool operator==(const VersionRange& other) const { - return (other.minCompatibleVersion == minCompatibleVersion) - && (other.currentVersion == currentVersion); - } - - bool operator!=(const VersionRange& other) const { - return !(*this == other); - } - - int minCompatibleVersion; - int currentVersion; - }; - - enum VersionStatus { - // No way to upgrade the test version to be compatible with current version - VersionStatus_Incompatible, + bool operator==(const VersionRange& other) const { + return (other.minCompatibleVersion == minCompatibleVersion) && + (other.currentVersion == currentVersion); + } - // Current version is compatible with test version - VersionStatus_Compatible, + bool operator!=(const VersionRange& other) const { + return !(*this == other); + } - // Test version must be upgraded to be compatible with current version - VersionStatus_NeedUpgrade - }; + int minCompatibleVersion; + int currentVersion; +}; - /** - * Encapsulates the information needed to register a config upgrade. - */ - struct UpgradeStep { - typedef stdx::function<bool(CatalogManager*, const VersionType&, string*)> UpgradeCallback; +enum VersionStatus { + // No way to upgrade the test version to be compatible with current version + VersionStatus_Incompatible, - UpgradeStep(int _fromVersion, - const VersionRange& _toVersionRange, - UpgradeCallback _upgradeCallback) - : fromVersion(_fromVersion), - toVersionRange(_toVersionRange), - upgradeCallback(_upgradeCallback) { + // Current version is compatible with test version + VersionStatus_Compatible, - } + // Test version must be upgraded to be compatible with current version + VersionStatus_NeedUpgrade +}; - // The config version we're upgrading from - int fromVersion; +/** + * Encapsulates the information needed to register a config upgrade. + */ +struct UpgradeStep { + typedef stdx::function<bool(CatalogManager*, const VersionType&, string*)> UpgradeCallback; - // The config version we're upgrading to and the min compatible config version (min, to) - VersionRange toVersionRange; + UpgradeStep(int _fromVersion, + const VersionRange& _toVersionRange, + UpgradeCallback _upgradeCallback) + : fromVersion(_fromVersion), + toVersionRange(_toVersionRange), + upgradeCallback(_upgradeCallback) {} - // The upgrade callback which performs the actual upgrade - UpgradeCallback upgradeCallback; - }; + // The config version we're upgrading from + int fromVersion; - typedef map<int, UpgradeStep> ConfigUpgradeRegistry; + // The config version we're upgrading to and the min compatible config version (min, to) + VersionRange toVersionRange; - /** - * Does a sanity-check validation of the registry ensuring three things: - * 1. All upgrade paths lead to the same minCompatible/currentVersion - * 2. Our constants match this final version pair - * 3. There is a zero-version upgrade path - */ - void validateRegistry(const ConfigUpgradeRegistry& registry) { - VersionRange maxCompatibleConfigVersionRange(-1, -1); - bool hasZeroVersionUpgrade = false; + // The upgrade callback which performs the actual upgrade + UpgradeCallback upgradeCallback; +}; - for (const auto& upgradeStep : registry) { - const UpgradeStep& upgrade = upgradeStep.second; +typedef map<int, UpgradeStep> ConfigUpgradeRegistry; - if (upgrade.fromVersion == 0) { - hasZeroVersionUpgrade = true; - } +/** + * Does a sanity-check validation of the registry ensuring three things: + * 1. All upgrade paths lead to the same minCompatible/currentVersion + * 2. Our constants match this final version pair + * 3. There is a zero-version upgrade path + */ +void validateRegistry(const ConfigUpgradeRegistry& registry) { + VersionRange maxCompatibleConfigVersionRange(-1, -1); + bool hasZeroVersionUpgrade = false; - if (maxCompatibleConfigVersionRange.currentVersion - < upgrade.toVersionRange.currentVersion) { + for (const auto& upgradeStep : registry) { + const UpgradeStep& upgrade = upgradeStep.second; - maxCompatibleConfigVersionRange = upgrade.toVersionRange; - } - else if (maxCompatibleConfigVersionRange.currentVersion - == upgrade.toVersionRange.currentVersion) { + if (upgrade.fromVersion == 0) { + hasZeroVersionUpgrade = true; + } - // Make sure all max upgrade paths end up with same version and compatibility - fassert(16621, maxCompatibleConfigVersionRange == upgrade.toVersionRange); - } + if (maxCompatibleConfigVersionRange.currentVersion < + upgrade.toVersionRange.currentVersion) { + maxCompatibleConfigVersionRange = upgrade.toVersionRange; + } else if (maxCompatibleConfigVersionRange.currentVersion == + upgrade.toVersionRange.currentVersion) { + // Make sure all max upgrade paths end up with same version and compatibility + fassert(16621, maxCompatibleConfigVersionRange == upgrade.toVersionRange); } + } - // Make sure we have a zero-version upgrade - fassert(16622, hasZeroVersionUpgrade); + // Make sure we have a zero-version upgrade + fassert(16622, hasZeroVersionUpgrade); - // Make sure our max registered range is the same as our constants - fassert(16623, - maxCompatibleConfigVersionRange - == VersionRange(MIN_COMPATIBLE_CONFIG_VERSION, CURRENT_CONFIG_VERSION)); - } + // Make sure our max registered range is the same as our constants + fassert(16623, + maxCompatibleConfigVersionRange == + VersionRange(MIN_COMPATIBLE_CONFIG_VERSION, CURRENT_CONFIG_VERSION)); +} - /** - * Creates a registry of config upgrades used by the code below. - * - * MODIFY THIS CODE HERE TO CREATE A NEW UPGRADE PATH FROM X to Y - * YOU MUST ALSO MODIFY THE VERSION DECLARATIONS IN config_upgrade.h - * - * Caveats: - * - All upgrade paths must eventually lead to the exact same version range of - * min and max compatible versions. - * - This resulting version range must be equal to: - * make_pair(MIN_COMPATIBLE_CONFIG_VERSION, CURRENT_CONFIG_VERSION) - * - There must always be an upgrade path from the empty version (0) to the latest - * config version. - * - * If any of the above is false, we fassert and fail to start. - */ - ConfigUpgradeRegistry createRegistry() { - ConfigUpgradeRegistry registry; - - // v0 to v7 - UpgradeStep v0ToV7(0, VersionRange(6, 7), doUpgradeV0ToV7); - registry.insert(make_pair(v0ToV7.fromVersion, v0ToV7)); - - // v6 to v7 - UpgradeStep v6ToV7(6, VersionRange(6, 7), doUpgradeV6ToV7); - registry.insert(make_pair(v6ToV7.fromVersion, v6ToV7)); - - validateRegistry(registry); - - return registry; - } +/** + * Creates a registry of config upgrades used by the code below. + * + * MODIFY THIS CODE HERE TO CREATE A NEW UPGRADE PATH FROM X to Y + * YOU MUST ALSO MODIFY THE VERSION DECLARATIONS IN config_upgrade.h + * + * Caveats: + * - All upgrade paths must eventually lead to the exact same version range of + * min and max compatible versions. + * - This resulting version range must be equal to: + * make_pair(MIN_COMPATIBLE_CONFIG_VERSION, CURRENT_CONFIG_VERSION) + * - There must always be an upgrade path from the empty version (0) to the latest + * config version. + * + * If any of the above is false, we fassert and fail to start. + */ +ConfigUpgradeRegistry createRegistry() { + ConfigUpgradeRegistry registry; - /** - * Checks whether or not a particular cluster version is compatible with our current - * version and mongodb version. The version is compatible if it falls between the - * MIN_COMPATIBLE_CONFIG_VERSION and CURRENT_CONFIG_VERSION and is not explicitly excluded. - * - * @return a VersionStatus enum indicating compatibility - */ - VersionStatus isConfigVersionCompatible(const VersionType& versionInfo, string* whyNot) { - string dummy; - if (!whyNot) { - whyNot = &dummy; - } + // v0 to v7 + UpgradeStep v0ToV7(0, VersionRange(6, 7), doUpgradeV0ToV7); + registry.insert(make_pair(v0ToV7.fromVersion, v0ToV7)); - // Check if we're empty - if (versionInfo.getCurrentVersion() == UpgradeHistory_EmptyVersion) { - return VersionStatus_NeedUpgrade; - } + // v6 to v7 + UpgradeStep v6ToV7(6, VersionRange(6, 7), doUpgradeV6ToV7); + registry.insert(make_pair(v6ToV7.fromVersion, v6ToV7)); - // Check that we aren't too old - if (CURRENT_CONFIG_VERSION < versionInfo.getMinCompatibleVersion()) { + validateRegistry(registry); - *whyNot = stream() << "the config version " << CURRENT_CONFIG_VERSION - << " of our process is too old " - << "for the detected config version " - << versionInfo.getMinCompatibleVersion(); + return registry; +} - return VersionStatus_Incompatible; - } +/** + * Checks whether or not a particular cluster version is compatible with our current + * version and mongodb version. The version is compatible if it falls between the + * MIN_COMPATIBLE_CONFIG_VERSION and CURRENT_CONFIG_VERSION and is not explicitly excluded. + * + * @return a VersionStatus enum indicating compatibility + */ +VersionStatus isConfigVersionCompatible(const VersionType& versionInfo, string* whyNot) { + string dummy; + if (!whyNot) { + whyNot = &dummy; + } - // Check that the mongo version of this process hasn't been excluded from the cluster - vector<MongoVersionRange> excludedRanges; - if (versionInfo.isExcludingMongoVersionsSet() && - !MongoVersionRange::parseBSONArray(versionInfo.getExcludingMongoVersions(), - &excludedRanges, - whyNot)) - { + // Check if we're empty + if (versionInfo.getCurrentVersion() == UpgradeHistory_EmptyVersion) { + return VersionStatus_NeedUpgrade; + } - *whyNot = stream() << "could not understand excluded version ranges" - << causedBy(whyNot); + // Check that we aren't too old + if (CURRENT_CONFIG_VERSION < versionInfo.getMinCompatibleVersion()) { + *whyNot = stream() << "the config version " << CURRENT_CONFIG_VERSION + << " of our process is too old " + << "for the detected config version " + << versionInfo.getMinCompatibleVersion(); - return VersionStatus_Incompatible; - } + return VersionStatus_Incompatible; + } - // versionString is the global version of this process - if (isInMongoVersionRanges(versionString, excludedRanges)) { + // Check that the mongo version of this process hasn't been excluded from the cluster + vector<MongoVersionRange> excludedRanges; + if (versionInfo.isExcludingMongoVersionsSet() && + !MongoVersionRange::parseBSONArray( + versionInfo.getExcludingMongoVersions(), &excludedRanges, whyNot)) { + *whyNot = stream() << "could not understand excluded version ranges" << causedBy(whyNot); - // Cast needed here for MSVC compiler issue - *whyNot = stream() << "not compatible with current config version, version " - << reinterpret_cast<const char*>(versionString) - << "has been excluded."; + return VersionStatus_Incompatible; + } - return VersionStatus_Incompatible; - } + // versionString is the global version of this process + if (isInMongoVersionRanges(versionString, excludedRanges)) { + // Cast needed here for MSVC compiler issue + *whyNot = stream() << "not compatible with current config version, version " + << reinterpret_cast<const char*>(versionString) << "has been excluded."; - // Check if we need to upgrade - if (versionInfo.getCurrentVersion() >= CURRENT_CONFIG_VERSION) { - return VersionStatus_Compatible; - } + return VersionStatus_Incompatible; + } - return VersionStatus_NeedUpgrade; + // Check if we need to upgrade + if (versionInfo.getCurrentVersion() >= CURRENT_CONFIG_VERSION) { + return VersionStatus_Compatible; } - // Checks that all config servers are online - bool _checkConfigServersAlive(const ConnectionString& configLoc, string* errMsg) { - bool resultOk; - BSONObj result; - try { - ScopedDbConnection conn(configLoc, 30); - if (conn->type() == ConnectionString::SYNC) { - // TODO: Dynamic cast is bad, we need a better way of managing this op - // via the heirarchy (or not) - SyncClusterConnection* scc = dynamic_cast<SyncClusterConnection*>(conn.get()); - fassert(16729, scc != NULL); - return scc->prepare(*errMsg); - } - else { - resultOk = conn->runCommand("admin", BSON( "fsync" << 1 ), result); - } - conn.done(); - } - catch (const DBException& e) { - *errMsg = e.toString(); - return false; + return VersionStatus_NeedUpgrade; +} + +// Checks that all config servers are online +bool _checkConfigServersAlive(const ConnectionString& configLoc, string* errMsg) { + bool resultOk; + BSONObj result; + try { + ScopedDbConnection conn(configLoc, 30); + if (conn->type() == ConnectionString::SYNC) { + // TODO: Dynamic cast is bad, we need a better way of managing this op + // via the heirarchy (or not) + SyncClusterConnection* scc = dynamic_cast<SyncClusterConnection*>(conn.get()); + fassert(16729, scc != NULL); + return scc->prepare(*errMsg); + } else { + resultOk = conn->runCommand("admin", BSON("fsync" << 1), result); } - - if (!resultOk) { - *errMsg = DBClientWithCommands::getLastErrorString(result); - return false; - } - - return true; + conn.done(); + } catch (const DBException& e) { + *errMsg = e.toString(); + return false; } - // Dispatches upgrades based on version to the upgrades registered in the upgrade registry - bool _nextUpgrade(CatalogManager* catalogManager, - const ConfigUpgradeRegistry& registry, - const VersionType& lastVersionInfo, - VersionType* upgradedVersionInfo, - string* errMsg) { + if (!resultOk) { + *errMsg = DBClientWithCommands::getLastErrorString(result); + return false; + } - int fromVersion = lastVersionInfo.getCurrentVersion(); + return true; +} - ConfigUpgradeRegistry::const_iterator foundIt = registry.find(fromVersion); +// Dispatches upgrades based on version to the upgrades registered in the upgrade registry +bool _nextUpgrade(CatalogManager* catalogManager, + const ConfigUpgradeRegistry& registry, + const VersionType& lastVersionInfo, + VersionType* upgradedVersionInfo, + string* errMsg) { + int fromVersion = lastVersionInfo.getCurrentVersion(); - if (foundIt == registry.end()) { + ConfigUpgradeRegistry::const_iterator foundIt = registry.find(fromVersion); - *errMsg = stream() << "newer version " << CURRENT_CONFIG_VERSION - << " of mongo config metadata is required, " << "current version is " - << fromVersion << ", " - << "don't know how to upgrade from this version"; + if (foundIt == registry.end()) { + *errMsg = stream() << "newer version " << CURRENT_CONFIG_VERSION + << " of mongo config metadata is required, " + << "current version is " << fromVersion << ", " + << "don't know how to upgrade from this version"; - return false; - } + return false; + } - const UpgradeStep& upgrade = foundIt->second; - int toVersion = upgrade.toVersionRange.currentVersion; + const UpgradeStep& upgrade = foundIt->second; + int toVersion = upgrade.toVersionRange.currentVersion; - log() << "starting next upgrade step from v" << fromVersion << " to v" << toVersion; + log() << "starting next upgrade step from v" << fromVersion << " to v" << toVersion; - // Log begin to config.changelog - catalogManager->logChange(NULL, - "starting upgrade of config database", - VersionType::ConfigNS, - BSON("from" << fromVersion << "to" << toVersion)); + // Log begin to config.changelog + catalogManager->logChange(NULL, + "starting upgrade of config database", + VersionType::ConfigNS, + BSON("from" << fromVersion << "to" << toVersion)); - if (!upgrade.upgradeCallback(catalogManager, lastVersionInfo, errMsg)) { - *errMsg = stream() << "error upgrading config database from v" - << fromVersion << " to v" << toVersion << causedBy(errMsg); - return false; - } + if (!upgrade.upgradeCallback(catalogManager, lastVersionInfo, errMsg)) { + *errMsg = stream() << "error upgrading config database from v" << fromVersion << " to v" + << toVersion << causedBy(errMsg); + return false; + } - // Get the config version we've upgraded to and make sure it's sane - Status verifyConfigStatus = getConfigVersion(catalogManager, upgradedVersionInfo); + // Get the config version we've upgraded to and make sure it's sane + Status verifyConfigStatus = getConfigVersion(catalogManager, upgradedVersionInfo); - if (!verifyConfigStatus.isOK()) { - *errMsg = stream() << "failed to validate v" << fromVersion << " config version upgrade" - << causedBy(verifyConfigStatus); + if (!verifyConfigStatus.isOK()) { + *errMsg = stream() << "failed to validate v" << fromVersion << " config version upgrade" + << causedBy(verifyConfigStatus); - return false; - } - - catalogManager->logChange(NULL, - "finished upgrade of config database", - VersionType::ConfigNS, - BSON("from" << fromVersion << "to" << toVersion)); - return true; + return false; } -} // namespace + catalogManager->logChange(NULL, + "finished upgrade of config database", + VersionType::ConfigNS, + BSON("from" << fromVersion << "to" << toVersion)); + return true; +} +} // namespace - /** - * Returns the config version of the cluster pointed at by the connection string. - * - * @return OK if version found successfully, error status if something bad happened. - */ - Status getConfigVersion(CatalogManager* catalogManager, VersionType* versionInfo) { - try { - versionInfo->clear(); - ScopedDbConnection conn(catalogManager->connectionString(), 30); - - unique_ptr<DBClientCursor> cursor(_safeCursor(conn->query("config.version", - BSONObj()))); +/** + * Returns the config version of the cluster pointed at by the connection string. + * + * @return OK if version found successfully, error status if something bad happened. + */ +Status getConfigVersion(CatalogManager* catalogManager, VersionType* versionInfo) { + try { + versionInfo->clear(); - bool hasConfigData = conn->count(ShardType::ConfigNS) - || conn->count(DatabaseType::ConfigNS) - || conn->count(CollectionType::ConfigNS); + ScopedDbConnection conn(catalogManager->connectionString(), 30); - if (!cursor->more()) { + unique_ptr<DBClientCursor> cursor(_safeCursor(conn->query("config.version", BSONObj()))); - // Version is 1 if we have data, 0 if we're completely empty - if (hasConfigData) { - versionInfo->setMinCompatibleVersion(UpgradeHistory_UnreportedVersion); - versionInfo->setCurrentVersion(UpgradeHistory_UnreportedVersion); - } - else { - versionInfo->setMinCompatibleVersion(UpgradeHistory_EmptyVersion); - versionInfo->setCurrentVersion(UpgradeHistory_EmptyVersion); - } + bool hasConfigData = conn->count(ShardType::ConfigNS) || + conn->count(DatabaseType::ConfigNS) || conn->count(CollectionType::ConfigNS); - conn.done(); - return Status::OK(); + if (!cursor->more()) { + // Version is 1 if we have data, 0 if we're completely empty + if (hasConfigData) { + versionInfo->setMinCompatibleVersion(UpgradeHistory_UnreportedVersion); + versionInfo->setCurrentVersion(UpgradeHistory_UnreportedVersion); + } else { + versionInfo->setMinCompatibleVersion(UpgradeHistory_EmptyVersion); + versionInfo->setCurrentVersion(UpgradeHistory_EmptyVersion); } - BSONObj versionDoc = cursor->next(); - string errMsg; + conn.done(); + return Status::OK(); + } - if (!versionInfo->parseBSON(versionDoc, &errMsg) || !versionInfo->isValid(&errMsg)) { - conn.done(); + BSONObj versionDoc = cursor->next(); + string errMsg; - return Status(ErrorCodes::UnsupportedFormat, - stream() << "invalid config version document " << versionDoc - << causedBy(errMsg)); - } + if (!versionInfo->parseBSON(versionDoc, &errMsg) || !versionInfo->isValid(&errMsg)) { + conn.done(); - if (cursor->more()) { - conn.done(); + return Status(ErrorCodes::UnsupportedFormat, + stream() << "invalid config version document " << versionDoc + << causedBy(errMsg)); + } - return Status(ErrorCodes::RemoteValidationError, - stream() << "should only have 1 document " - << "in config.version collection"); - } + if (cursor->more()) { conn.done(); - } - catch (const DBException& e) { - return e.toStatus(); - } - return Status::OK(); + return Status(ErrorCodes::RemoteValidationError, + stream() << "should only have 1 document " + << "in config.version collection"); + } + conn.done(); + } catch (const DBException& e) { + return e.toStatus(); } - bool checkAndUpgradeConfigVersion(CatalogManager* catalogManager, - bool upgrade, - VersionType* initialVersionInfo, - VersionType* versionInfo, - string* errMsg) { + return Status::OK(); +} + +bool checkAndUpgradeConfigVersion(CatalogManager* catalogManager, + bool upgrade, + VersionType* initialVersionInfo, + VersionType* versionInfo, + string* errMsg) { + string dummy; + if (!errMsg) { + errMsg = &dummy; + } - string dummy; - if (!errMsg) { - errMsg = &dummy; - } + Status getConfigStatus = getConfigVersion(catalogManager, versionInfo); + if (!getConfigStatus.isOK()) { + *errMsg = stream() << "could not load config version for upgrade" + << causedBy(getConfigStatus); + return false; + } - Status getConfigStatus = getConfigVersion(catalogManager, versionInfo); - if (!getConfigStatus.isOK()) { - *errMsg = stream() << "could not load config version for upgrade" - << causedBy(getConfigStatus); - return false; - } + versionInfo->cloneTo(initialVersionInfo); - versionInfo->cloneTo(initialVersionInfo); + VersionStatus comp = isConfigVersionCompatible(*versionInfo, errMsg); - VersionStatus comp = isConfigVersionCompatible(*versionInfo, errMsg); + if (comp == VersionStatus_Incompatible) + return false; + if (comp == VersionStatus_Compatible) + return true; - if (comp == VersionStatus_Incompatible) return false; - if (comp == VersionStatus_Compatible) return true; + invariant(comp == VersionStatus_NeedUpgrade); - invariant(comp == VersionStatus_NeedUpgrade); + // + // Our current config version is now greater than the current version, so we should upgrade + // if possible. + // - // - // Our current config version is now greater than the current version, so we should upgrade - // if possible. - // + // The first empty version is technically an upgrade, but has special semantics + bool isEmptyVersion = versionInfo->getCurrentVersion() == UpgradeHistory_EmptyVersion; - // The first empty version is technically an upgrade, but has special semantics - bool isEmptyVersion = versionInfo->getCurrentVersion() == UpgradeHistory_EmptyVersion; + // First check for the upgrade flag (but no flag is needed if we're upgrading from empty) + if (!isEmptyVersion && !upgrade) { + *errMsg = stream() << "newer version " << CURRENT_CONFIG_VERSION + << " of mongo config metadata is required, " + << "current version is " << versionInfo->getCurrentVersion() << ", " + << "need to run mongos with --upgrade"; - // First check for the upgrade flag (but no flag is needed if we're upgrading from empty) - if (!isEmptyVersion && !upgrade) { - *errMsg = stream() << "newer version " << CURRENT_CONFIG_VERSION - << " of mongo config metadata is required, " << "current version is " - << versionInfo->getCurrentVersion() << ", " - << "need to run mongos with --upgrade"; + return false; + } - return false; + // Contact the config servers to make sure all are online - otherwise we wait a long time + // for locks. + if (!_checkConfigServersAlive(catalogManager->connectionString(), errMsg)) { + if (isEmptyVersion) { + *errMsg = stream() << "all config servers must be reachable for initial" + << " config database creation" << causedBy(errMsg); + } else { + *errMsg = stream() << "all config servers must be reachable for config upgrade" + << causedBy(errMsg); } - // Contact the config servers to make sure all are online - otherwise we wait a long time - // for locks. - if (!_checkConfigServersAlive(catalogManager->connectionString(), errMsg)) { + return false; + } - if (isEmptyVersion) { - *errMsg = stream() << "all config servers must be reachable for initial" - << " config database creation" << causedBy(errMsg); - } - else { - *errMsg = stream() << "all config servers must be reachable for config upgrade" + // Check whether or not the balancer is online, if it is online we will not upgrade + // (but we will initialize the config server) + if (!isEmptyVersion) { + auto balSettingsResult = catalogManager->getGlobalSettings(SettingsType::BalancerDocKey); + if (balSettingsResult.isOK()) { + SettingsType balSettings = balSettingsResult.getValue(); + if (!balSettings.getBalancerStopped()) { + *errMsg = stream() << "balancer must be stopped for config upgrade" << causedBy(errMsg); } - - return false; } + } - // Check whether or not the balancer is online, if it is online we will not upgrade - // (but we will initialize the config server) - if (!isEmptyVersion) { - auto balSettingsResult = - catalogManager->getGlobalSettings(SettingsType::BalancerDocKey); - if (balSettingsResult.isOK()) { - SettingsType balSettings = balSettingsResult.getValue(); - if (!balSettings.getBalancerStopped()) { - *errMsg = stream() << "balancer must be stopped for config upgrade" - << causedBy(errMsg); - } - } - } + // + // Acquire a lock for the upgrade process. + // + // We want to ensure that only a single mongo process is upgrading the config server at a + // time. + // + + string whyMessage(stream() << "upgrading config database to new format v" + << CURRENT_CONFIG_VERSION); + auto lockTimeout = stdx::chrono::milliseconds(20 * 60 * 1000); + auto scopedDistLock = + catalogManager->getDistLockManager()->lock("configUpgrade", whyMessage, lockTimeout); + if (!scopedDistLock.isOK()) { + *errMsg = scopedDistLock.getStatus().toString(); + return false; + } - // - // Acquire a lock for the upgrade process. - // - // We want to ensure that only a single mongo process is upgrading the config server at a - // time. - // + // + // Double-check compatibility inside the upgrade lock + // Another process may have won the lock earlier and done the upgrade for us, check + // if this is the case. + // + + getConfigStatus = getConfigVersion(catalogManager, versionInfo); + if (!getConfigStatus.isOK()) { + *errMsg = stream() << "could not reload config version for upgrade" + << causedBy(getConfigStatus); + return false; + } - string whyMessage(stream() << "upgrading config database to new format v" - << CURRENT_CONFIG_VERSION); - auto lockTimeout = stdx::chrono::milliseconds(20 * 60 * 1000); - auto scopedDistLock = catalogManager->getDistLockManager()->lock("configUpgrade", - whyMessage, - lockTimeout); - if (!scopedDistLock.isOK()) { - *errMsg = scopedDistLock.getStatus().toString(); - return false; - } + versionInfo->cloneTo(initialVersionInfo); - // - // Double-check compatibility inside the upgrade lock - // Another process may have won the lock earlier and done the upgrade for us, check - // if this is the case. - // + comp = isConfigVersionCompatible(*versionInfo, errMsg); - getConfigStatus = getConfigVersion(catalogManager, versionInfo); - if (!getConfigStatus.isOK()) { - *errMsg = stream() << "could not reload config version for upgrade" - << causedBy(getConfigStatus); - return false; - } + if (comp == VersionStatus_Incompatible) + return false; + if (comp == VersionStatus_Compatible) + return true; - versionInfo->cloneTo(initialVersionInfo); + invariant(comp == VersionStatus_NeedUpgrade); - comp = isConfigVersionCompatible(*versionInfo, errMsg); + // + // Run through the upgrade steps necessary to bring our config version to the current + // version + // - if (comp == VersionStatus_Incompatible) return false; - if (comp == VersionStatus_Compatible) return true; + log() << "starting upgrade of config server from v" << versionInfo->getCurrentVersion() + << " to v" << CURRENT_CONFIG_VERSION; - invariant(comp == VersionStatus_NeedUpgrade); + ConfigUpgradeRegistry registry(createRegistry()); + + while (versionInfo->getCurrentVersion() < CURRENT_CONFIG_VERSION) { + int fromVersion = versionInfo->getCurrentVersion(); // - // Run through the upgrade steps necessary to bring our config version to the current - // version + // Run the next upgrade process and replace versionInfo with the result of the + // upgrade. // - log() << "starting upgrade of config server from v" << versionInfo->getCurrentVersion() - << " to v" << CURRENT_CONFIG_VERSION; - - ConfigUpgradeRegistry registry(createRegistry()); - - while (versionInfo->getCurrentVersion() < CURRENT_CONFIG_VERSION) { - int fromVersion = versionInfo->getCurrentVersion(); - - // - // Run the next upgrade process and replace versionInfo with the result of the - // upgrade. - // - - if (!_nextUpgrade(catalogManager, registry, *versionInfo, versionInfo, errMsg)) { - return false; - } - - // Ensure we're making progress here - if (versionInfo->getCurrentVersion() <= fromVersion) { + if (!_nextUpgrade(catalogManager, registry, *versionInfo, versionInfo, errMsg)) { + return false; + } - *errMsg = stream() << "bad v" << fromVersion << " config version upgrade, " - << "version did not increment and is now " - << versionInfo->getCurrentVersion(); + // Ensure we're making progress here + if (versionInfo->getCurrentVersion() <= fromVersion) { + *errMsg = stream() << "bad v" << fromVersion << " config version upgrade, " + << "version did not increment and is now " + << versionInfo->getCurrentVersion(); - return false; - } + return false; } + } - invariant(versionInfo->getCurrentVersion() == CURRENT_CONFIG_VERSION); + invariant(versionInfo->getCurrentVersion() == CURRENT_CONFIG_VERSION); - log() << "upgrade of config server to v" << versionInfo->getCurrentVersion() - << " successful"; + log() << "upgrade of config server to v" << versionInfo->getCurrentVersion() << " successful"; - return true; - } + return true; +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/legacy/config_upgrade.h b/src/mongo/s/catalog/legacy/config_upgrade.h index cd37a9cb634..7caf9d5a177 100644 --- a/src/mongo/s/catalog/legacy/config_upgrade.h +++ b/src/mongo/s/catalog/legacy/config_upgrade.h @@ -32,118 +32,118 @@ namespace mongo { - class CatalogManager; - class Status; - class VersionType; +class CatalogManager; +class Status; +class VersionType; + +/** + * UPGRADE HISTORY + * + * The enum below documents the version changes to *both* the config server data layout + * and the versioning protocol between clients (i.e. the set of calls between mongos and + * mongod). + * + * Friendly notice: + * + * EVERY CHANGE EITHER IN CONFIG LAYOUT AND IN S/D PROTOCOL MUST BE RECORDED HERE BY AN INCREASE + * IN THE VERSION AND BY TAKING THE FOLLOWING STEPS. (IF YOU DON'T UNDERSTAND THESE STEPS, YOU + * SHOULD PROBABLY NOT BE UPGRADING THE VERSIONS BY YOURSELF.) + * + * + A new entry in the UpgradeHistory enum is created + * + The CURRENT_CONFIG_VERSION below is incremented to that version + * + There should be a determination if the MIN_COMPATIBLE_CONFIG_VERSION should be increased or + * not. This means determining if, by introducing the changes to layout and/or protocol, the + * new mongos/d can co-exist in a cluster with the old ones. + * + If layout changes are involved, there should be a corresponding layout upgrade routine. See + * for instance config_upgrade_vX_to_vY.cpp. + * + Again, if a layout change occurs, the base upgrade method, config_upgrade_v0_to_vX.cpp must + * be upgraded. This means that all new clusters will start at the newest versions. + * + */ +enum UpgradeHistory { /** - * UPGRADE HISTORY - * - * The enum below documents the version changes to *both* the config server data layout - * and the versioning protocol between clients (i.e. the set of calls between mongos and - * mongod). - * - * Friendly notice: + * The empty version, reported when there is no config server data + */ + UpgradeHistory_EmptyVersion = 0, + + /** + * The unreported version older mongoses used before config.version collection existed * - * EVERY CHANGE EITHER IN CONFIG LAYOUT AND IN S/D PROTOCOL MUST BE RECORDED HERE BY AN INCREASE - * IN THE VERSION AND BY TAKING THE FOLLOWING STEPS. (IF YOU DON'T UNDERSTAND THESE STEPS, YOU - * SHOULD PROBABLY NOT BE UPGRADING THE VERSIONS BY YOURSELF.) + * If there is a config.shards/databases/collections collection but no config.version + * collection, version 1 is assumed + */ + UpgradeHistory_UnreportedVersion = 1, + + /** + * NOTE: We skip version 2 here since it is very old and we shouldn't see it in the wild. * - * + A new entry in the UpgradeHistory enum is created - * + The CURRENT_CONFIG_VERSION below is incremented to that version - * + There should be a determination if the MIN_COMPATIBLE_CONFIG_VERSION should be increased or - * not. This means determining if, by introducing the changes to layout and/or protocol, the - * new mongos/d can co-exist in a cluster with the old ones. - * + If layout changes are involved, there should be a corresponding layout upgrade routine. See - * for instance config_upgrade_vX_to_vY.cpp. - * + Again, if a layout change occurs, the base upgrade method, config_upgrade_v0_to_vX.cpp must - * be upgraded. This means that all new clusters will start at the newest versions. + * Do not skip upgrade versions in the future. + */ + + /** + * Base version used by pre-2.4 mongoses with no collection epochs. + */ + UpgradeHistory_NoEpochVersion = 3, + + /** + * Version upgrade which added collection epochs to all sharded collections and + * chunks. * + * Also: + * + Version document in config.version now of the form: + * { minVersion : X, currentVersion : Y, clusterId : OID(...) } + * + Mongos pings include a "mongoVersion" field indicating the mongos version + * + Mongos pings include a "configVersion" field indicating the current config version + * + Mongos explicitly ignores any collection with a "primary" field */ - enum UpgradeHistory { - - /** - * The empty version, reported when there is no config server data - */ - UpgradeHistory_EmptyVersion = 0, - - /** - * The unreported version older mongoses used before config.version collection existed - * - * If there is a config.shards/databases/collections collection but no config.version - * collection, version 1 is assumed - */ - UpgradeHistory_UnreportedVersion = 1, - - /** - * NOTE: We skip version 2 here since it is very old and we shouldn't see it in the wild. - * - * Do not skip upgrade versions in the future. - */ - - /** - * Base version used by pre-2.4 mongoses with no collection epochs. - */ - UpgradeHistory_NoEpochVersion = 3, - - /** - * Version upgrade which added collection epochs to all sharded collections and - * chunks. - * - * Also: - * + Version document in config.version now of the form: - * { minVersion : X, currentVersion : Y, clusterId : OID(...) } - * + Mongos pings include a "mongoVersion" field indicating the mongos version - * + Mongos pings include a "configVersion" field indicating the current config version - * + Mongos explicitly ignores any collection with a "primary" field - */ - UpgradeHistory_MandatoryEpochVersion = 4, - - /** - * Version upgrade with the following changes: - * - * + Dropping a collection from mongos now waits for the chunks to be removed from the - * config server before contacting each shard. Because of this, mongos should be - * upgraded first before mongod or never drop collections during upgrade. - */ - UpgradeHistory_DummyBumpPre2_6 = 5, - - /** - * Version upgrade with the following changes: - * - * + "_secondaryThrottle" field for config.settings now accepts write concern - * specifications. - * + config.locks { ts: 1 } index is no longer unique. - */ - UpgradeHistory_DummyBumpPre2_8 = 6, // Note: 2.8 is also known as 3.0. - - UpgradeHistory_DummyBumpPre3_0 = 7, - }; - - // Earliest version we're compatible with - const int MIN_COMPATIBLE_CONFIG_VERSION = UpgradeHistory_DummyBumpPre2_8; - - // Latest version we know how to communicate with - const int CURRENT_CONFIG_VERSION = UpgradeHistory_DummyBumpPre3_0; + UpgradeHistory_MandatoryEpochVersion = 4, /** - * Returns the config version of the cluster pointed at by the connection string. + * Version upgrade with the following changes: * - * @return OK if version found successfully, error status if something bad happened. + * + Dropping a collection from mongos now waits for the chunks to be removed from the + * config server before contacting each shard. Because of this, mongos should be + * upgraded first before mongod or never drop collections during upgrade. */ - Status getConfigVersion(CatalogManager* catalogManager, VersionType* versionInfo); + UpgradeHistory_DummyBumpPre2_6 = 5, /** - * Checks the config version and ensures it's the latest version, otherwise tries to update. + * Version upgrade with the following changes: * - * @return true if the config version is now compatible. - * @return initial and finalVersionInfo indicating the start and end versions of the upgrade. - * These are the same if no upgrade occurred. + * + "_secondaryThrottle" field for config.settings now accepts write concern + * specifications. + * + config.locks { ts: 1 } index is no longer unique. */ - bool checkAndUpgradeConfigVersion(CatalogManager* catalogManager, - bool upgrade, - VersionType* initialVersionInfo, - VersionType* finalVersionInfo, - std::string* errMsg); + UpgradeHistory_DummyBumpPre2_8 = 6, // Note: 2.8 is also known as 3.0. + + UpgradeHistory_DummyBumpPre3_0 = 7, +}; + +// Earliest version we're compatible with +const int MIN_COMPATIBLE_CONFIG_VERSION = UpgradeHistory_DummyBumpPre2_8; + +// Latest version we know how to communicate with +const int CURRENT_CONFIG_VERSION = UpgradeHistory_DummyBumpPre3_0; + +/** + * Returns the config version of the cluster pointed at by the connection string. + * + * @return OK if version found successfully, error status if something bad happened. + */ +Status getConfigVersion(CatalogManager* catalogManager, VersionType* versionInfo); + +/** + * Checks the config version and ensures it's the latest version, otherwise tries to update. + * + * @return true if the config version is now compatible. + * @return initial and finalVersionInfo indicating the start and end versions of the upgrade. + * These are the same if no upgrade occurred. + */ +bool checkAndUpgradeConfigVersion(CatalogManager* catalogManager, + bool upgrade, + VersionType* initialVersionInfo, + VersionType* finalVersionInfo, + std::string* errMsg); -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/legacy/config_upgrade_helpers.cpp b/src/mongo/s/catalog/legacy/config_upgrade_helpers.cpp index dfde57d6774..41bc88eeb41 100644 --- a/src/mongo/s/catalog/legacy/config_upgrade_helpers.cpp +++ b/src/mongo/s/catalog/legacy/config_upgrade_helpers.cpp @@ -45,81 +45,76 @@ namespace mongo { - using std::endl; - using std::string; - using std::unique_ptr; +using std::endl; +using std::string; +using std::unique_ptr; - using mongoutils::str::stream; +using mongoutils::str::stream; - // Custom field used in upgrade state to determine if/where we failed on last upgrade - const BSONField<bool> inCriticalSectionField("inCriticalSection", false); +// Custom field used in upgrade state to determine if/where we failed on last upgrade +const BSONField<bool> inCriticalSectionField("inCriticalSection", false); - Status preUpgradeCheck(CatalogManager* catalogManager, - const VersionType& lastVersionInfo, - string minMongosVersion) { - - if (lastVersionInfo.isUpgradeIdSet() && lastVersionInfo.getUpgradeId().isSet()) { - // - // Another upgrade failed, so cleanup may be necessary - // - - BSONObj lastUpgradeState = lastVersionInfo.getUpgradeState(); +Status preUpgradeCheck(CatalogManager* catalogManager, + const VersionType& lastVersionInfo, + string minMongosVersion) { + if (lastVersionInfo.isUpgradeIdSet() && lastVersionInfo.getUpgradeId().isSet()) { + // + // Another upgrade failed, so cleanup may be necessary + // - bool inCriticalSection; - string errMsg; - if (!FieldParser::extract(lastUpgradeState, - inCriticalSectionField, - &inCriticalSection, - &errMsg)) { - return Status(ErrorCodes::FailedToParse, causedBy(errMsg)); - } + BSONObj lastUpgradeState = lastVersionInfo.getUpgradeState(); - if (inCriticalSection) { - // Note: custom message must be supplied by caller - return Status(ErrorCodes::ManualInterventionRequired, ""); - } + bool inCriticalSection; + string errMsg; + if (!FieldParser::extract( + lastUpgradeState, inCriticalSectionField, &inCriticalSection, &errMsg)) { + return Status(ErrorCodes::FailedToParse, causedBy(errMsg)); } - // - // Check the versions of other mongo processes in the cluster before upgrade. - // We can't upgrade if there are active pre-v2.4 processes in the cluster - // - return checkClusterMongoVersions(catalogManager, string(minMongosVersion)); - } - - Status commitConfigUpgrade(CatalogManager* catalogManager, - int currentVersion, - int minCompatibleVersion, - int newVersion) { - - // Note: DO NOT CLEAR the config version unless bumping the minCompatibleVersion, - // we want to save the excludes that were set. - - BSONObjBuilder setObj; - setObj << VersionType::minCompatibleVersion(minCompatibleVersion); - setObj << VersionType::currentVersion(newVersion); - - BSONObjBuilder unsetObj; - unsetObj.append(VersionType::upgradeId(), 1); - unsetObj.append(VersionType::upgradeState(), 1); - unsetObj.append("version", 1); // remove deprecated field, no longer supported >= v3.0. - - Status result = catalogManager->update( - VersionType::ConfigNS, - BSON("_id" << 1 << VersionType::currentVersion(currentVersion)), - BSON("$set" << setObj.done() << "$unset" << unsetObj.done()), - false, - false, - NULL); - if (!result.isOK()) { - return Status(result.code(), - str::stream() << "could not write new version info " - << " and exit critical upgrade section: " - << result.reason()); + if (inCriticalSection) { + // Note: custom message must be supplied by caller + return Status(ErrorCodes::ManualInterventionRequired, ""); } + } - return result; + // + // Check the versions of other mongo processes in the cluster before upgrade. + // We can't upgrade if there are active pre-v2.4 processes in the cluster + // + return checkClusterMongoVersions(catalogManager, string(minMongosVersion)); +} + +Status commitConfigUpgrade(CatalogManager* catalogManager, + int currentVersion, + int minCompatibleVersion, + int newVersion) { + // Note: DO NOT CLEAR the config version unless bumping the minCompatibleVersion, + // we want to save the excludes that were set. + + BSONObjBuilder setObj; + setObj << VersionType::minCompatibleVersion(minCompatibleVersion); + setObj << VersionType::currentVersion(newVersion); + + BSONObjBuilder unsetObj; + unsetObj.append(VersionType::upgradeId(), 1); + unsetObj.append(VersionType::upgradeState(), 1); + unsetObj.append("version", 1); // remove deprecated field, no longer supported >= v3.0. + + Status result = + catalogManager->update(VersionType::ConfigNS, + BSON("_id" << 1 << VersionType::currentVersion(currentVersion)), + BSON("$set" << setObj.done() << "$unset" << unsetObj.done()), + false, + false, + NULL); + if (!result.isOK()) { + return Status(result.code(), + str::stream() << "could not write new version info " + << " and exit critical upgrade section: " << result.reason()); } -} // namespace mongo + return result; +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/legacy/config_upgrade_helpers.h b/src/mongo/s/catalog/legacy/config_upgrade_helpers.h index 10d03676ea8..7411737e262 100644 --- a/src/mongo/s/catalog/legacy/config_upgrade_helpers.h +++ b/src/mongo/s/catalog/legacy/config_upgrade_helpers.h @@ -39,32 +39,32 @@ namespace mongo { - class CatalogManager; - class ConnectionString; - class OID; - class Status; - class VersionType; +class CatalogManager; +class ConnectionString; +class OID; +class Status; +class VersionType; - /** - * Checks whether an unsuccessful upgrade was performed last time and also checks whether - * the mongos in the current cluster have the mimimum version required. Returns not ok if - * the check failed and the upgrade should not proceed. - * - * Note: There is also a special case for ManualInterventionRequired error where the - * message will be empty. - */ - Status preUpgradeCheck(CatalogManager* catalogManager, - const VersionType& lastVersionInfo, - std::string minMongosVersion); +/** + * Checks whether an unsuccessful upgrade was performed last time and also checks whether + * the mongos in the current cluster have the mimimum version required. Returns not ok if + * the check failed and the upgrade should not proceed. + * + * Note: There is also a special case for ManualInterventionRequired error where the + * message will be empty. + */ +Status preUpgradeCheck(CatalogManager* catalogManager, + const VersionType& lastVersionInfo, + std::string minMongosVersion); - /** - * Informs the config server that the upgrade task was completed by bumping the version. - * This also clears all upgrade state effectively leaving the critical section if the - * upgrade process did enter it. - */ - Status commitConfigUpgrade(CatalogManager* catalogManager, - int currentVersion, - int minCompatibleVersion, - int newVersion); +/** + * Informs the config server that the upgrade task was completed by bumping the version. + * This also clears all upgrade state effectively leaving the critical section if the + * upgrade process did enter it. + */ +Status commitConfigUpgrade(CatalogManager* catalogManager, + int currentVersion, + int minCompatibleVersion, + int newVersion); -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/legacy/config_upgrade_v0_to_v7.cpp b/src/mongo/s/catalog/legacy/config_upgrade_v0_to_v7.cpp index 787d0114e73..f086fabc521 100644 --- a/src/mongo/s/catalog/legacy/config_upgrade_v0_to_v7.cpp +++ b/src/mongo/s/catalog/legacy/config_upgrade_v0_to_v7.cpp @@ -39,59 +39,57 @@ namespace mongo { - using std::string; - using mongo::str::stream; - - - /** - * Upgrade v0 to v7 described here - * - * This upgrade takes the config server from empty to an initial version. - */ - bool doUpgradeV0ToV7(CatalogManager* catalogManager, - const VersionType& lastVersionInfo, - string* errMsg) { - - string dummy; - if (!errMsg) errMsg = &dummy; - - verify(lastVersionInfo.getCurrentVersion() == UpgradeHistory_EmptyVersion); - - // - // Even though the initial config write is a single-document update, that single document - // is on multiple config servers and requests can interleave. The upgrade lock prevents - // this. - // - - log() << "writing initial config version at v" << CURRENT_CONFIG_VERSION; - - OID newClusterId = OID::gen(); - - VersionType versionInfo; - - // Upgrade to new version - versionInfo.setMinCompatibleVersion(MIN_COMPATIBLE_CONFIG_VERSION); - versionInfo.setCurrentVersion(CURRENT_CONFIG_VERSION); - versionInfo.setClusterId(newClusterId); - - verify(versionInfo.isValid(NULL)); - - // If the cluster has not previously been initialized, we need to set the version before - // using so subsequent mongoses use the config data the same way. This requires all three - // config servers online initially. - Status result = catalogManager->update(VersionType::ConfigNS, - BSON("_id" << 1), - versionInfo.toBSON(), - true, // upsert - false, // multi - NULL); - if (!result.isOK()) { - *errMsg = stream() << "error writing initial config version: " - << result.reason(); - return false; - } - - return true; +using std::string; +using mongo::str::stream; + + +/** + * Upgrade v0 to v7 described here + * + * This upgrade takes the config server from empty to an initial version. + */ +bool doUpgradeV0ToV7(CatalogManager* catalogManager, + const VersionType& lastVersionInfo, + string* errMsg) { + string dummy; + if (!errMsg) + errMsg = &dummy; + + verify(lastVersionInfo.getCurrentVersion() == UpgradeHistory_EmptyVersion); + + // + // Even though the initial config write is a single-document update, that single document + // is on multiple config servers and requests can interleave. The upgrade lock prevents + // this. + // + + log() << "writing initial config version at v" << CURRENT_CONFIG_VERSION; + + OID newClusterId = OID::gen(); + + VersionType versionInfo; + + // Upgrade to new version + versionInfo.setMinCompatibleVersion(MIN_COMPATIBLE_CONFIG_VERSION); + versionInfo.setCurrentVersion(CURRENT_CONFIG_VERSION); + versionInfo.setClusterId(newClusterId); + + verify(versionInfo.isValid(NULL)); + + // If the cluster has not previously been initialized, we need to set the version before + // using so subsequent mongoses use the config data the same way. This requires all three + // config servers online initially. + Status result = catalogManager->update(VersionType::ConfigNS, + BSON("_id" << 1), + versionInfo.toBSON(), + true, // upsert + false, // multi + NULL); + if (!result.isOK()) { + *errMsg = stream() << "error writing initial config version: " << result.reason(); + return false; } + return true; +} } diff --git a/src/mongo/s/catalog/legacy/config_upgrade_v6_to_v7.cpp b/src/mongo/s/catalog/legacy/config_upgrade_v6_to_v7.cpp index 6f841b0dc87..430d2eedb6e 100644 --- a/src/mongo/s/catalog/legacy/config_upgrade_v6_to_v7.cpp +++ b/src/mongo/s/catalog/legacy/config_upgrade_v6_to_v7.cpp @@ -40,74 +40,73 @@ namespace mongo { - using std::list; - using std::string; - using std::vector; - - static const char* minMongoProcessVersion = "3.0"; - - static const char* cannotCleanupMessage = - "\n\n" - "******\n" - "Cannot upgrade config database from v6 to v7 because a previous upgrade\n" - "failed in the critical section. Manual intervention is required to re-sync\n" - "the config servers.\n" - "******\n"; - - /** - * Upgrades v6 to v7. - */ - bool doUpgradeV6ToV7(CatalogManager* catalogManager, - const VersionType& lastVersionInfo, - string* errMsg) { - - string dummy; - if (!errMsg) errMsg = &dummy; - - invariant(lastVersionInfo.getCurrentVersion() == UpgradeHistory_DummyBumpPre2_8); - - Status result = preUpgradeCheck(catalogManager, lastVersionInfo, minMongoProcessVersion); - if (!result.isOK()) { - if (result.code() == ErrorCodes::ManualInterventionRequired) { - *errMsg = cannotCleanupMessage; - } - else { - *errMsg = result.toString(); - } - - return false; - } +using std::list; +using std::string; +using std::vector; + +static const char* minMongoProcessVersion = "3.0"; + +static const char* cannotCleanupMessage = + "\n\n" + "******\n" + "Cannot upgrade config database from v6 to v7 because a previous upgrade\n" + "failed in the critical section. Manual intervention is required to re-sync\n" + "the config servers.\n" + "******\n"; - // This is not needed because we are not actually going to make any modifications - // on the other collections in the config server for this particular upgrade. - // startConfigUpgrade(configLoc.toString(), - // lastVersionInfo.getCurrentVersion(), - // OID::gen()); - - // If we actually need to modify something in the config servers these need to follow - // after calling startConfigUpgrade(...): - // - // 1. Acquire necessary locks. - // 2. Make a backup of the collections we are about to modify. - // 3. Perform the upgrade process on the backup collection. - // 4. Verify that no changes were made to the collections since the backup was performed. - // 5. Call enterConfigUpgradeCriticalSection(configLoc.toString(), - // lastVersionInfo.getCurrentVersion()). - // 6. Rename the backup collection to the name of the original collection with - // dropTarget set to true. - - // We're only after the version bump in commitConfigUpgrade here since we never - // get into the critical section. - Status commitStatus = commitConfigUpgrade(catalogManager, - lastVersionInfo.getCurrentVersion(), - MIN_COMPATIBLE_CONFIG_VERSION, - CURRENT_CONFIG_VERSION); - - if (!commitStatus.isOK()) { - *errMsg = commitStatus.toString(); - return false; +/** + * Upgrades v6 to v7. + */ +bool doUpgradeV6ToV7(CatalogManager* catalogManager, + const VersionType& lastVersionInfo, + string* errMsg) { + string dummy; + if (!errMsg) + errMsg = &dummy; + + invariant(lastVersionInfo.getCurrentVersion() == UpgradeHistory_DummyBumpPre2_8); + + Status result = preUpgradeCheck(catalogManager, lastVersionInfo, minMongoProcessVersion); + if (!result.isOK()) { + if (result.code() == ErrorCodes::ManualInterventionRequired) { + *errMsg = cannotCleanupMessage; + } else { + *errMsg = result.toString(); } - return true; + return false; } + + // This is not needed because we are not actually going to make any modifications + // on the other collections in the config server for this particular upgrade. + // startConfigUpgrade(configLoc.toString(), + // lastVersionInfo.getCurrentVersion(), + // OID::gen()); + + // If we actually need to modify something in the config servers these need to follow + // after calling startConfigUpgrade(...): + // + // 1. Acquire necessary locks. + // 2. Make a backup of the collections we are about to modify. + // 3. Perform the upgrade process on the backup collection. + // 4. Verify that no changes were made to the collections since the backup was performed. + // 5. Call enterConfigUpgradeCriticalSection(configLoc.toString(), + // lastVersionInfo.getCurrentVersion()). + // 6. Rename the backup collection to the name of the original collection with + // dropTarget set to true. + + // We're only after the version bump in commitConfigUpgrade here since we never + // get into the critical section. + Status commitStatus = commitConfigUpgrade(catalogManager, + lastVersionInfo.getCurrentVersion(), + MIN_COMPATIBLE_CONFIG_VERSION, + CURRENT_CONFIG_VERSION); + + if (!commitStatus.isOK()) { + *errMsg = commitStatus.toString(); + return false; + } + + return true; +} } diff --git a/src/mongo/s/catalog/legacy/distlock.cpp b/src/mongo/s/catalog/legacy/distlock.cpp index 22c0a4e6128..b0d30b28967 100644 --- a/src/mongo/s/catalog/legacy/distlock.cpp +++ b/src/mongo/s/catalog/legacy/distlock.cpp @@ -43,796 +43,760 @@ namespace mongo { - using std::endl; - using std::list; - using std::set; - using std::string; - using std::stringstream; - using std::unique_ptr; - using std::vector; - - LabeledLevel DistributedLock::logLvl( 1 ); - DistributedLock::LastPings DistributedLock::lastPings; - - ThreadLocalValue<string> distLockIds(""); - - /* ================== - * Module initialization - */ - - static SimpleMutex _cachedProcessMutex; - static string* _cachedProcessString = NULL; - - static void initModule() { - stdx::lock_guard<SimpleMutex> lk(_cachedProcessMutex); - if (_cachedProcessString) { - // someone got the lock before us - return; - } - - // cache process string - stringstream ss; - ss << getHostName() << ":" << serverGlobalParams.port << ":" << time(0) << ":" << rand(); - _cachedProcessString = new string( ss.str() ); - } - - /* =================== */ +using std::endl; +using std::list; +using std::set; +using std::string; +using std::stringstream; +using std::unique_ptr; +using std::vector; - string getDistLockProcess() { - if (!_cachedProcessString) - initModule(); - verify( _cachedProcessString ); - return *_cachedProcessString; - } +LabeledLevel DistributedLock::logLvl(1); +DistributedLock::LastPings DistributedLock::lastPings; - string getDistLockId() { - string s = distLockIds.get(); - if ( s.empty() ) { - stringstream ss; - ss << getDistLockProcess() << ":" << getThreadName() << ":" << rand(); - s = ss.str(); - distLockIds.set( s ); - } - return s; - } +ThreadLocalValue<string> distLockIds(""); - LockException::LockException(StringData msg, int code): LockException(msg, code, OID()) { - } +/* ================== + * Module initialization + */ - LockException::LockException(StringData msg, int code, DistLockHandle lockID): - DBException(msg.toString(), code), - _mustUnlockID(lockID) { - } +static SimpleMutex _cachedProcessMutex; +static string* _cachedProcessString = NULL; - DistLockHandle LockException::getMustUnlockID() const { - return _mustUnlockID; +static void initModule() { + stdx::lock_guard<SimpleMutex> lk(_cachedProcessMutex); + if (_cachedProcessString) { + // someone got the lock before us + return; } - /** - * Create a new distributed lock, potentially with a custom sleep and takeover time. If a custom sleep time is - * specified (time between pings) - */ - DistributedLock::DistributedLock( const ConnectionString& conn , const string& name , unsigned long long lockTimeout, bool asProcess ) - : _conn(conn), _name(name), - _processId( asProcess ? getDistLockId() : getDistLockProcess() ), - _lockTimeout( lockTimeout == 0 ? LOCK_TIMEOUT : lockTimeout ), - _maxClockSkew( _lockTimeout / LOCK_SKEW_FACTOR ), _maxNetSkew( _maxClockSkew ), - _lockPing( _maxClockSkew ) - { - LOG( logLvl ) << "created new distributed lock for " << name << " on " << conn - << " ( lock timeout : " << _lockTimeout - << ", ping interval : " << _lockPing << ", process : " << asProcess << " )" << endl; - - - } + // cache process string + stringstream ss; + ss << getHostName() << ":" << serverGlobalParams.port << ":" << time(0) << ":" << rand(); + _cachedProcessString = new string(ss.str()); +} - DistLockPingInfo DistributedLock::LastPings::getLastPing(const ConnectionString& conn, - const string& lockName) { - stdx::lock_guard<stdx::mutex> lock(_mutex); - return _lastPings[std::make_pair(conn.toString(), lockName)]; - } +/* =================== */ - void DistributedLock::LastPings::setLastPing(const ConnectionString& conn, - const string& lockName, - const DistLockPingInfo& pd) { - stdx::lock_guard<stdx::mutex> lock(_mutex); - _lastPings[std::make_pair(conn.toString(), lockName)] = pd; - } +string getDistLockProcess() { + if (!_cachedProcessString) + initModule(); + verify(_cachedProcessString); + return *_cachedProcessString; +} - Date_t DistributedLock::getRemoteTime() const { - return DistributedLock::remoteTime(_conn, _maxNetSkew); +string getDistLockId() { + string s = distLockIds.get(); + if (s.empty()) { + stringstream ss; + ss << getDistLockProcess() << ":" << getThreadName() << ":" << rand(); + s = ss.str(); + distLockIds.set(s); } + return s; +} - bool DistributedLock::isRemoteTimeSkewed() const { - return !DistributedLock::checkSkew(_conn, - NUM_LOCK_SKEW_CHECKS, - _maxClockSkew, - _maxNetSkew); - } +LockException::LockException(StringData msg, int code) : LockException(msg, code, OID()) {} - const ConnectionString& DistributedLock::getRemoteConnection() const { - return _conn; - } +LockException::LockException(StringData msg, int code, DistLockHandle lockID) + : DBException(msg.toString(), code), _mustUnlockID(lockID) {} - const string& DistributedLock::getProcessId() const { - return _processId; - } - - /** - * Returns the remote time as reported by the cluster or server. The maximum difference between the reported time - * and the actual time on the remote server (at the completion of the function) is the maxNetSkew - */ - Date_t DistributedLock::remoteTime( const ConnectionString& cluster, unsigned long long maxNetSkew ) { +DistLockHandle LockException::getMustUnlockID() const { + return _mustUnlockID; +} - ConnectionString server( *cluster.getServers().begin() ); +/** + * Create a new distributed lock, potentially with a custom sleep and takeover time. If a custom sleep time is + * specified (time between pings) + */ +DistributedLock::DistributedLock(const ConnectionString& conn, + const string& name, + unsigned long long lockTimeout, + bool asProcess) + : _conn(conn), + _name(name), + _processId(asProcess ? getDistLockId() : getDistLockProcess()), + _lockTimeout(lockTimeout == 0 ? LOCK_TIMEOUT : lockTimeout), + _maxClockSkew(_lockTimeout / LOCK_SKEW_FACTOR), + _maxNetSkew(_maxClockSkew), + _lockPing(_maxClockSkew) { + LOG(logLvl) << "created new distributed lock for " << name << " on " << conn + << " ( lock timeout : " << _lockTimeout << ", ping interval : " << _lockPing + << ", process : " << asProcess << " )" << endl; +} - // Get result and delay if successful, errMsg if not - bool success = false; - BSONObj result; - string errMsg; - Milliseconds delay{0}; +DistLockPingInfo DistributedLock::LastPings::getLastPing(const ConnectionString& conn, + const string& lockName) { + stdx::lock_guard<stdx::mutex> lock(_mutex); + return _lastPings[std::make_pair(conn.toString(), lockName)]; +} - unique_ptr<ScopedDbConnection> connPtr; - try { - connPtr.reset( new ScopedDbConnection( server.toString() ) ); - ScopedDbConnection& conn = *connPtr; +void DistributedLock::LastPings::setLastPing(const ConnectionString& conn, + const string& lockName, + const DistLockPingInfo& pd) { + stdx::lock_guard<stdx::mutex> lock(_mutex); + _lastPings[std::make_pair(conn.toString(), lockName)] = pd; +} - Date_t then = jsTime(); - success = conn->runCommand( string( "admin" ), BSON( "serverStatus" << 1 ), result ); - delay = jsTime() - then; +Date_t DistributedLock::getRemoteTime() const { + return DistributedLock::remoteTime(_conn, _maxNetSkew); +} - if ( !success ) errMsg = result.toString(); - conn.done(); - } - catch ( const DBException& ex ) { +bool DistributedLock::isRemoteTimeSkewed() const { + return !DistributedLock::checkSkew(_conn, NUM_LOCK_SKEW_CHECKS, _maxClockSkew, _maxNetSkew); +} - if ( connPtr && connPtr->get()->isFailed() ) { - // Return to the pool so the pool knows about the failure - connPtr->done(); - } +const ConnectionString& DistributedLock::getRemoteConnection() const { + return _conn; +} - success = false; - errMsg = ex.toString(); - } - - if( !success ) { - throw TimeNotFoundException( str::stream() << "could not get status from server " - << server.toString() << " in cluster " - << cluster.toString() << " to check time" - << causedBy( errMsg ), - 13647 ); - } +const string& DistributedLock::getProcessId() const { + return _processId; +} - // Make sure that our delay is not more than 2x our maximum network skew, since this is the max our remote - // time value can be off by if we assume a response in the middle of the delay. - if (delay > Milliseconds(maxNetSkew * 2)) { - throw TimeNotFoundException( str::stream() - << "server " << server.toString() << " in cluster " - << cluster.toString() - << " did not respond within max network delay of " - << maxNetSkew << "ms", - 13648 ); +/** + * Returns the remote time as reported by the cluster or server. The maximum difference between the reported time + * and the actual time on the remote server (at the completion of the function) is the maxNetSkew + */ +Date_t DistributedLock::remoteTime(const ConnectionString& cluster, unsigned long long maxNetSkew) { + ConnectionString server(*cluster.getServers().begin()); + + // Get result and delay if successful, errMsg if not + bool success = false; + BSONObj result; + string errMsg; + Milliseconds delay{0}; + + unique_ptr<ScopedDbConnection> connPtr; + try { + connPtr.reset(new ScopedDbConnection(server.toString())); + ScopedDbConnection& conn = *connPtr; + + Date_t then = jsTime(); + success = conn->runCommand(string("admin"), BSON("serverStatus" << 1), result); + delay = jsTime() - then; + + if (!success) + errMsg = result.toString(); + conn.done(); + } catch (const DBException& ex) { + if (connPtr && connPtr->get()->isFailed()) { + // Return to the pool so the pool knows about the failure + connPtr->done(); } - return result["localTime"].Date() - (delay / 2); + success = false; + errMsg = ex.toString(); } - bool DistributedLock::checkSkew( const ConnectionString& cluster, unsigned skewChecks, unsigned long long maxClockSkew, unsigned long long maxNetSkew ) { - - vector<HostAndPort> servers = cluster.getServers(); + if (!success) { + throw TimeNotFoundException(str::stream() << "could not get status from server " + << server.toString() << " in cluster " + << cluster.toString() << " to check time" + << causedBy(errMsg), + 13647); + } - if(servers.size() < 1) return true; + // Make sure that our delay is not more than 2x our maximum network skew, since this is the max our remote + // time value can be off by if we assume a response in the middle of the delay. + if (delay > Milliseconds(maxNetSkew * 2)) { + throw TimeNotFoundException( + str::stream() << "server " << server.toString() << " in cluster " << cluster.toString() + << " did not respond within max network delay of " << maxNetSkew << "ms", + 13648); + } - vector<long long> avgSkews; + return result["localTime"].Date() - (delay / 2); +} - for(unsigned i = 0; i < skewChecks; i++) { +bool DistributedLock::checkSkew(const ConnectionString& cluster, + unsigned skewChecks, + unsigned long long maxClockSkew, + unsigned long long maxNetSkew) { + vector<HostAndPort> servers = cluster.getServers(); - // Find the average skew for each server - unsigned s = 0; - for(vector<HostAndPort>::iterator si = servers.begin(); si != servers.end(); ++si,s++) { + if (servers.size() < 1) + return true; - if(i == 0) avgSkews.push_back(0); + vector<long long> avgSkews; - // Could check if this is self, but shouldn't matter since local network connection should be fast. - ConnectionString server( *si ); + for (unsigned i = 0; i < skewChecks; i++) { + // Find the average skew for each server + unsigned s = 0; + for (vector<HostAndPort>::iterator si = servers.begin(); si != servers.end(); ++si, s++) { + if (i == 0) + avgSkews.push_back(0); - vector<long long> skew; + // Could check if this is self, but shouldn't matter since local network connection should be fast. + ConnectionString server(*si); - BSONObj result; + vector<long long> skew; - Date_t remote = remoteTime( server, maxNetSkew ); - Date_t local = jsTime(); + BSONObj result; - // Remote time can be delayed by at most MAX_NET_SKEW + Date_t remote = remoteTime(server, maxNetSkew); + Date_t local = jsTime(); - // Skew is how much time we'd have to add to local to get to remote - avgSkews[s] += (remote - local).count(); + // Remote time can be delayed by at most MAX_NET_SKEW - LOG( logLvl + 1 ) << "skew from remote server " << server << " found: " - << (remote - local).count(); + // Skew is how much time we'd have to add to local to get to remote + avgSkews[s] += (remote - local).count(); - } + LOG(logLvl + 1) << "skew from remote server " << server + << " found: " << (remote - local).count(); } + } - // Analyze skews - - long long serverMaxSkew = 0; - long long serverMinSkew = 0; + // Analyze skews - for(unsigned s = 0; s < avgSkews.size(); s++) { + long long serverMaxSkew = 0; + long long serverMinSkew = 0; - long long avgSkew = (avgSkews[s] /= skewChecks); + for (unsigned s = 0; s < avgSkews.size(); s++) { + long long avgSkew = (avgSkews[s] /= skewChecks); - // Keep track of max and min skews - if(s == 0) { + // Keep track of max and min skews + if (s == 0) { + serverMaxSkew = avgSkew; + serverMinSkew = avgSkew; + } else { + if (avgSkew > serverMaxSkew) serverMaxSkew = avgSkew; + if (avgSkew < serverMinSkew) serverMinSkew = avgSkew; - } - else { - if(avgSkew > serverMaxSkew) - serverMaxSkew = avgSkew; - if(avgSkew < serverMinSkew) - serverMinSkew = avgSkew; - } - } + } - long long totalSkew = serverMaxSkew - serverMinSkew; - - // Make sure our max skew is not more than our pre-set limit - if(totalSkew > (long long) maxClockSkew) { - LOG( logLvl + 1 ) << "total clock skew of " << totalSkew << "ms for servers " << cluster << " is out of " << maxClockSkew << "ms bounds." << endl; - return false; - } + long long totalSkew = serverMaxSkew - serverMinSkew; - LOG( logLvl + 1 ) << "total clock skew of " << totalSkew << "ms for servers " << cluster << " is in " << maxClockSkew << "ms bounds." << endl; - return true; + // Make sure our max skew is not more than our pre-set limit + if (totalSkew > (long long)maxClockSkew) { + LOG(logLvl + 1) << "total clock skew of " << totalSkew << "ms for servers " << cluster + << " is out of " << maxClockSkew << "ms bounds." << endl; + return false; } - Status DistributedLock::checkStatus(double timeout) { - - BSONObj lockObj; - try { - ScopedDbConnection conn(_conn.toString(), timeout ); - lockObj = conn->findOne( LocksType::ConfigNS, - BSON( LocksType::name(_name) ) ).getOwned(); - conn.done(); - } - catch ( DBException& e ) { - return e.toStatus(); - } + LOG(logLvl + 1) << "total clock skew of " << totalSkew << "ms for servers " << cluster + << " is in " << maxClockSkew << "ms bounds." << endl; + return true; +} - if ( lockObj.isEmpty() ) { - return Status(ErrorCodes::LockFailed, - str::stream() << "no lock for " << _name << " exists in the locks collection"); - } +Status DistributedLock::checkStatus(double timeout) { + BSONObj lockObj; + try { + ScopedDbConnection conn(_conn.toString(), timeout); + lockObj = conn->findOne(LocksType::ConfigNS, BSON(LocksType::name(_name))).getOwned(); + conn.done(); + } catch (DBException& e) { + return e.toStatus(); + } - if ( lockObj[LocksType::state()].numberInt() < 2 ) { - return Status(ErrorCodes::LockFailed, - str::stream() << "lock " << _name << " current state is not held (" - << lockObj[LocksType::state()].numberInt() << ")"); - } + if (lockObj.isEmpty()) { + return Status(ErrorCodes::LockFailed, + str::stream() << "no lock for " << _name + << " exists in the locks collection"); + } - if ( lockObj[LocksType::process()].String() != _processId ) { - return Status(ErrorCodes::LockFailed, - str::stream() << "lock " << _name << " is currently being held by " - << "another process (" - << lockObj[LocksType::process()].String() << ")"); - } + if (lockObj[LocksType::state()].numberInt() < 2) { + return Status(ErrorCodes::LockFailed, + str::stream() << "lock " << _name << " current state is not held (" + << lockObj[LocksType::state()].numberInt() << ")"); + } - return Status::OK(); + if (lockObj[LocksType::process()].String() != _processId) { + return Status(ErrorCodes::LockFailed, + str::stream() << "lock " << _name << " is currently being held by " + << "another process (" << lockObj[LocksType::process()].String() + << ")"); } - static void logErrMsgOrWarn(StringData messagePrefix, - StringData lockName, - StringData errMsg, - StringData altErrMsg) { + return Status::OK(); +} - if (errMsg.empty()) { - LOG(DistributedLock::logLvl - 1) << messagePrefix << " '" << lockName << "' " << - altErrMsg << std::endl; - } - else { - warning() << messagePrefix << " '" << lockName << "' " << causedBy(errMsg.toString()); - } +static void logErrMsgOrWarn(StringData messagePrefix, + StringData lockName, + StringData errMsg, + StringData altErrMsg) { + if (errMsg.empty()) { + LOG(DistributedLock::logLvl - 1) << messagePrefix << " '" << lockName << "' " << altErrMsg + << std::endl; + } else { + warning() << messagePrefix << " '" << lockName << "' " << causedBy(errMsg.toString()); } +} - // Semantics of this method are basically that if the lock cannot be acquired, returns false, - // can be retried. If the lock should not be tried again (some unexpected error), - // a LockException is thrown. - bool DistributedLock::lock_try(const string& why, BSONObj* other, double timeout) { - // This should always be true, if not, we are using the lock incorrectly. - verify( _name != "" ); +// Semantics of this method are basically that if the lock cannot be acquired, returns false, +// can be retried. If the lock should not be tried again (some unexpected error), +// a LockException is thrown. +bool DistributedLock::lock_try(const string& why, BSONObj* other, double timeout) { + // This should always be true, if not, we are using the lock incorrectly. + verify(_name != ""); - LOG( logLvl ) << "trying to acquire new distributed lock for " << _name << " on " << _conn - << " ( lock timeout : " << _lockTimeout - << ", ping interval : " << _lockPing << ", process : " << _processId << " )" - << endl; + LOG(logLvl) << "trying to acquire new distributed lock for " << _name << " on " << _conn + << " ( lock timeout : " << _lockTimeout << ", ping interval : " << _lockPing + << ", process : " << _processId << " )" << endl; - // write to dummy if 'other' is null - BSONObj dummyOther; - if ( other == NULL ) - other = &dummyOther; + // write to dummy if 'other' is null + BSONObj dummyOther; + if (other == NULL) + other = &dummyOther; - ScopedDbConnection conn(_conn.toString(), timeout ); + ScopedDbConnection conn(_conn.toString(), timeout); - BSONObjBuilder queryBuilder; - queryBuilder.append( LocksType::name() , _name ); - queryBuilder.append( LocksType::state() , 0 ); + BSONObjBuilder queryBuilder; + queryBuilder.append(LocksType::name(), _name); + queryBuilder.append(LocksType::state(), 0); - { - // make sure its there so we can use simple update logic below - BSONObj o = conn->findOne( LocksType::ConfigNS , BSON( LocksType::name(_name) ) ).getOwned(); + { + // make sure its there so we can use simple update logic below + BSONObj o = conn->findOne(LocksType::ConfigNS, BSON(LocksType::name(_name))).getOwned(); - // Case 1: No locks - if ( o.isEmpty() ) { - try { - LOG( logLvl ) << "inserting initial doc in " << LocksType::ConfigNS << " for lock " << _name << endl; - conn->insert( LocksType::ConfigNS, - BSON( LocksType::name(_name) - << LocksType::state(0) - << LocksType::who("") - << LocksType::lockID(OID()) )); - } - catch ( UserException& e ) { - warning() << "could not insert initial doc for distributed lock " << _name << causedBy( e ) << endl; - } + // Case 1: No locks + if (o.isEmpty()) { + try { + LOG(logLvl) << "inserting initial doc in " << LocksType::ConfigNS << " for lock " + << _name << endl; + conn->insert(LocksType::ConfigNS, + BSON(LocksType::name(_name) << LocksType::state(0) + << LocksType::who("") + << LocksType::lockID(OID()))); + } catch (UserException& e) { + warning() << "could not insert initial doc for distributed lock " << _name + << causedBy(e) << endl; } + } - // Case 2: A set lock that we might be able to force - else if ( o[LocksType::state()].numberInt() > 0 ) { - - string lockName = o[LocksType::name()].String() + string("/") + o[LocksType::process()].String(); - - BSONObj lastPing = conn->findOne( LockpingsType::ConfigNS, o[LocksType::process()].wrap( LockpingsType::process() ) ); - if ( lastPing.isEmpty() ) { - LOG( logLvl ) << "empty ping found for process in lock '" << lockName << "'" << endl; - // TODO: Using 0 as a "no time found" value Will fail if dates roll over, but then, so will a lot. - lastPing = BSON( LockpingsType::process(o[LocksType::process()].String()) << - LockpingsType::ping(Date_t()) ); - } - - unsigned long long elapsed = 0; - unsigned long long takeover = _lockTimeout; - DistLockPingInfo lastPingEntry = getLastPing(); + // Case 2: A set lock that we might be able to force + else if (o[LocksType::state()].numberInt() > 0) { + string lockName = + o[LocksType::name()].String() + string("/") + o[LocksType::process()].String(); + + BSONObj lastPing = conn->findOne( + LockpingsType::ConfigNS, o[LocksType::process()].wrap(LockpingsType::process())); + if (lastPing.isEmpty()) { + LOG(logLvl) << "empty ping found for process in lock '" << lockName << "'" << endl; + // TODO: Using 0 as a "no time found" value Will fail if dates roll over, but then, so will a lot. + lastPing = BSON(LockpingsType::process(o[LocksType::process()].String()) + << LockpingsType::ping(Date_t())); + } - LOG(logLvl) << "checking last ping for lock '" << lockName - << "' against process " << lastPingEntry.processId - << " and ping " << lastPingEntry.lastPing; + unsigned long long elapsed = 0; + unsigned long long takeover = _lockTimeout; + DistLockPingInfo lastPingEntry = getLastPing(); - try { + LOG(logLvl) << "checking last ping for lock '" << lockName << "' against process " + << lastPingEntry.processId << " and ping " << lastPingEntry.lastPing; - Date_t remote = remoteTime( _conn ); - - auto pingDocProcessId = lastPing[LockpingsType::process()].String(); - auto pingDocPingValue = lastPing[LockpingsType::ping()].Date(); - - // Timeout the elapsed time using comparisons of remote clock - // For non-finalized locks, timeout 15 minutes since last seen (ts) - // For finalized locks, timeout 15 minutes since last ping - bool recPingChange = - o[LocksType::state()].numberInt() == 2 && - (lastPingEntry.processId != pingDocProcessId || - lastPingEntry.lastPing != pingDocPingValue); - bool recTSChange = lastPingEntry.lockSessionId != o[LocksType::lockID()].OID(); - - if (recPingChange || recTSChange) { - // If the ping has changed since we last checked, mark the current date and time - setLastPing(DistLockPingInfo(pingDocProcessId, - pingDocPingValue, - remote, - o[LocksType::lockID()].OID(), - OID())); - } - else { - - // GOTCHA! Due to network issues, it is possible that the current time - // is less than the remote time. We *have* to check this here, otherwise - // we overflow and our lock breaks. - if (lastPingEntry.configLocalTime >= remote) - elapsed = 0; - else - elapsed = (remote - lastPingEntry.configLocalTime).count(); - } + try { + Date_t remote = remoteTime(_conn); + + auto pingDocProcessId = lastPing[LockpingsType::process()].String(); + auto pingDocPingValue = lastPing[LockpingsType::ping()].Date(); + + // Timeout the elapsed time using comparisons of remote clock + // For non-finalized locks, timeout 15 minutes since last seen (ts) + // For finalized locks, timeout 15 minutes since last ping + bool recPingChange = o[LocksType::state()].numberInt() == 2 && + (lastPingEntry.processId != pingDocProcessId || + lastPingEntry.lastPing != pingDocPingValue); + bool recTSChange = lastPingEntry.lockSessionId != o[LocksType::lockID()].OID(); + + if (recPingChange || recTSChange) { + // If the ping has changed since we last checked, mark the current date and time + setLastPing(DistLockPingInfo(pingDocProcessId, + pingDocPingValue, + remote, + o[LocksType::lockID()].OID(), + OID())); + } else { + // GOTCHA! Due to network issues, it is possible that the current time + // is less than the remote time. We *have* to check this here, otherwise + // we overflow and our lock breaks. + if (lastPingEntry.configLocalTime >= remote) + elapsed = 0; + else + elapsed = (remote - lastPingEntry.configLocalTime).count(); } - catch( LockException& e ) { - - // Remote server cannot be found / is not responsive - warning() << "Could not get remote time from " << _conn << causedBy( e ); - // If our config server is having issues, forget all the pings until we can see it again - resetLastPing(); + } catch (LockException& e) { + // Remote server cannot be found / is not responsive + warning() << "Could not get remote time from " << _conn << causedBy(e); + // If our config server is having issues, forget all the pings until we can see it again + resetLastPing(); + } - } + if (elapsed <= takeover) { + LOG(1) << "could not force lock '" << lockName << "' because elapsed time " + << elapsed << " <= takeover time " << takeover; + *other = o; + other->getOwned(); + conn.done(); + return false; + } - if (elapsed <= takeover) { - LOG(1) << "could not force lock '" << lockName - << "' because elapsed time " << elapsed - << " <= takeover time " << takeover; - *other = o; other->getOwned(); conn.done(); - return false; - } + LOG(0) << "forcing lock '" << lockName << "' because elapsed time " << elapsed + << " > takeover time " << takeover; - LOG(0) << "forcing lock '" << lockName - << "' because elapsed time " << elapsed - << " > takeover time " << takeover; - - if( elapsed > takeover ) { - - // Lock may forced, reset our timer if succeeds or fails - // Ensures that another timeout must happen if something borks up here, and resets our pristine - // ping state if acquired. - resetLastPing(); - - try { - - // Check the clock skew again. If we check this before we get a lock - // and after the lock times out, we can be pretty sure the time is - // increasing at the same rate on all servers and therefore our - // timeout is accurate - if (isRemoteTimeSkewed()) { - string msg(str::stream() << "remote time in cluster " - << _conn.toString() - << " is now skewed, cannot force lock."); - throw LockException(msg, ErrorCodes::DistributedClockSkewed); - } - - // Make sure we break the lock with the correct "ts" (OID) value, otherwise - // we can overwrite a new lock inserted in the meantime. - conn->update( LocksType::ConfigNS, - BSON( LocksType::name(_name) << - LocksType::state(o[LocksType::state()].numberInt()) << - LocksType::lockID(o[LocksType::lockID()].OID()) ), - BSON( "$set" << BSON( LocksType::state(0) ) ) ); - - BSONObj err = conn->getLastErrorDetailed(); - string errMsg = DBClientWithCommands::getLastErrorString(err); - - // TODO: Clean up all the extra code to exit this method, probably with a refactor - if ( !errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1 ) { - logErrMsgOrWarn("Could not force lock", lockName, errMsg, "(another force won"); - *other = o; other->getOwned(); conn.done(); - return false; - } + if (elapsed > takeover) { + // Lock may forced, reset our timer if succeeds or fails + // Ensures that another timeout must happen if something borks up here, and resets our pristine + // ping state if acquired. + resetLastPing(); + try { + // Check the clock skew again. If we check this before we get a lock + // and after the lock times out, we can be pretty sure the time is + // increasing at the same rate on all servers and therefore our + // timeout is accurate + if (isRemoteTimeSkewed()) { + string msg(str::stream() << "remote time in cluster " << _conn.toString() + << " is now skewed, cannot force lock."); + throw LockException(msg, ErrorCodes::DistributedClockSkewed); } - catch( UpdateNotTheSame& ) { - // Ok to continue since we know we forced at least one lock document, and all lock docs - // are required for a lock to be held. - warning() << "lock forcing " << lockName << " inconsistent" << endl; - } - catch (const LockException& ) { - // Let the exception go up and don't repackage the exception. - throw; - } - catch( std::exception& e ) { + + // Make sure we break the lock with the correct "ts" (OID) value, otherwise + // we can overwrite a new lock inserted in the meantime. + conn->update(LocksType::ConfigNS, + BSON(LocksType::name(_name) + << LocksType::state(o[LocksType::state()].numberInt()) + << LocksType::lockID(o[LocksType::lockID()].OID())), + BSON("$set" << BSON(LocksType::state(0)))); + + BSONObj err = conn->getLastErrorDetailed(); + string errMsg = DBClientWithCommands::getLastErrorString(err); + + // TODO: Clean up all the extra code to exit this method, probably with a refactor + if (!errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1) { + logErrMsgOrWarn( + "Could not force lock", lockName, errMsg, "(another force won"); + *other = o; + other->getOwned(); conn.done(); - string msg(str::stream() << "exception forcing distributed lock " - << lockName << causedBy(e)); - throw LockException(msg, 13660); + return false; } + } catch (UpdateNotTheSame&) { + // Ok to continue since we know we forced at least one lock document, and all lock docs + // are required for a lock to be held. + warning() << "lock forcing " << lockName << " inconsistent" << endl; + } catch (const LockException&) { + // Let the exception go up and don't repackage the exception. + throw; + } catch (std::exception& e) { + conn.done(); + string msg(str::stream() << "exception forcing distributed lock " << lockName + << causedBy(e)); + throw LockException(msg, 13660); } - else { - // Not strictly necessary, but helpful for small timeouts where thread - // scheduling is significant. This ensures that two attempts are still - // required for a force if not acquired, and resets our state if we - // are acquired. - resetLastPing(); - - // Test that the lock is held by trying to update the finalized state of the lock to the same state - // if it does not update or does not update on all servers, we can't re-enter. - try { - - // Test the lock with the correct "ts" (OID) value - conn->update( LocksType::ConfigNS, - BSON( LocksType::name(_name) << - LocksType::state(2) << - LocksType::lockID(o[LocksType::lockID()].OID()) ), - BSON( "$set" << BSON( LocksType::state(2) ) ) ); - - BSONObj err = conn->getLastErrorDetailed(); - string errMsg = DBClientWithCommands::getLastErrorString(err); - - // TODO: Clean up all the extra code to exit this method, probably with a refactor - if ( ! errMsg.empty() || ! err["n"].type() || err["n"].numberInt() < 1 ) { - logErrMsgOrWarn("Could not re-enter lock", lockName, errMsg, "(not sure lock is held"); - *other = o; other->getOwned(); conn.done(); - return false; - } - } - catch( UpdateNotTheSame& ) { - // NOT ok to continue since our lock isn't held by all servers, so isn't valid. - warning() << "inconsistent state re-entering lock, lock " << lockName << " not held" << endl; - *other = o; other->getOwned(); conn.done(); - return false; - } - catch( std::exception& e ) { + } else { + // Not strictly necessary, but helpful for small timeouts where thread + // scheduling is significant. This ensures that two attempts are still + // required for a force if not acquired, and resets our state if we + // are acquired. + resetLastPing(); + + // Test that the lock is held by trying to update the finalized state of the lock to the same state + // if it does not update or does not update on all servers, we can't re-enter. + try { + // Test the lock with the correct "ts" (OID) value + conn->update(LocksType::ConfigNS, + BSON(LocksType::name(_name) + << LocksType::state(2) + << LocksType::lockID(o[LocksType::lockID()].OID())), + BSON("$set" << BSON(LocksType::state(2)))); + + BSONObj err = conn->getLastErrorDetailed(); + string errMsg = DBClientWithCommands::getLastErrorString(err); + + // TODO: Clean up all the extra code to exit this method, probably with a refactor + if (!errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1) { + logErrMsgOrWarn( + "Could not re-enter lock", lockName, errMsg, "(not sure lock is held"); + *other = o; + other->getOwned(); conn.done(); - string msg(str::stream() << "exception re-entering distributed lock " - << lockName << causedBy(e)); - throw LockException(msg, 13660); + return false; } - LOG( logLvl - 1 ) << "re-entered distributed lock '" << lockName << "'" << endl; - *other = o.getOwned(); + } catch (UpdateNotTheSame&) { + // NOT ok to continue since our lock isn't held by all servers, so isn't valid. + warning() << "inconsistent state re-entering lock, lock " << lockName + << " not held" << endl; + *other = o; + other->getOwned(); conn.done(); - return true; - + return false; + } catch (std::exception& e) { + conn.done(); + string msg(str::stream() << "exception re-entering distributed lock " + << lockName << causedBy(e)); + throw LockException(msg, 13660); } - LOG( logLvl - 1 ) << "lock '" << lockName << "' successfully forced" << endl; - - // We don't need the ts value in the query, since we will only ever replace locks with state=0. - } - // Case 3: We have an expired lock - else if ( o[LocksType::lockID()].type() ) { - queryBuilder.append( o[LocksType::lockID()] ); + LOG(logLvl - 1) << "re-entered distributed lock '" << lockName << "'" << endl; + *other = o.getOwned(); + conn.done(); + return true; } - } - // Always reset our ping if we're trying to get a lock, since getting a lock implies the lock state is open - // and no locks need to be forced. If anything goes wrong, we don't want to remember an old lock. - resetLastPing(); + LOG(logLvl - 1) << "lock '" << lockName << "' successfully forced" << endl; - bool gotLock = false; - BSONObj currLock; + // We don't need the ts value in the query, since we will only ever replace locks with state=0. + } + // Case 3: We have an expired lock + else if (o[LocksType::lockID()].type()) { + queryBuilder.append(o[LocksType::lockID()]); + } + } - BSONObj lockDetails = BSON( LocksType::state(1) - << LocksType::who(getDistLockId()) - << LocksType::process(_processId) - << LocksType::when(jsTime()) - << LocksType::why(why) - << LocksType::lockID(OID::gen()) ); - BSONObj whatIWant = BSON( "$set" << lockDetails ); + // Always reset our ping if we're trying to get a lock, since getting a lock implies the lock state is open + // and no locks need to be forced. If anything goes wrong, we don't want to remember an old lock. + resetLastPing(); - BSONObj query = queryBuilder.obj(); + bool gotLock = false; + BSONObj currLock; - string lockName = _name + string("/") + _processId; + BSONObj lockDetails = + BSON(LocksType::state(1) << LocksType::who(getDistLockId()) + << LocksType::process(_processId) << LocksType::when(jsTime()) + << LocksType::why(why) << LocksType::lockID(OID::gen())); + BSONObj whatIWant = BSON("$set" << lockDetails); - try { + BSONObj query = queryBuilder.obj(); - // Main codepath to acquire lock + string lockName = _name + string("/") + _processId; - LOG( logLvl ) << "about to acquire distributed lock '" << lockName << "'"; + try { + // Main codepath to acquire lock - LOG( logLvl + 1 ) << "trying to acquire lock " << query.toString( false, true ) - << " with details " << lockDetails.toString( false, true ) << endl; + LOG(logLvl) << "about to acquire distributed lock '" << lockName << "'"; - conn->update( LocksType::ConfigNS , query , whatIWant ); + LOG(logLvl + 1) << "trying to acquire lock " << query.toString(false, true) + << " with details " << lockDetails.toString(false, true) << endl; - BSONObj err = conn->getLastErrorDetailed(); - string errMsg = DBClientWithCommands::getLastErrorString(err); + conn->update(LocksType::ConfigNS, query, whatIWant); - currLock = conn->findOne( LocksType::ConfigNS , BSON( LocksType::name(_name) ) ); + BSONObj err = conn->getLastErrorDetailed(); + string errMsg = DBClientWithCommands::getLastErrorString(err); - if ( !errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1 ) { - logErrMsgOrWarn("could not acquire lock", lockName, errMsg, "(another update won)"); - *other = currLock; - other->getOwned(); - gotLock = false; - } - else { - gotLock = true; - } + currLock = conn->findOne(LocksType::ConfigNS, BSON(LocksType::name(_name))); + if (!errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1) { + logErrMsgOrWarn("could not acquire lock", lockName, errMsg, "(another update won)"); + *other = currLock; + other->getOwned(); + gotLock = false; + } else { + gotLock = true; } - catch ( UpdateNotTheSame& up ) { - - // this means our update got through on some, but not others - warning() << "distributed lock '" << lockName << " did not propagate properly." << causedBy( up ) << endl; - - // Overall protection derives from: - // All unlocking updates use the ts value when setting state to 0 - // This ensures that during locking, we can override all smaller ts locks with - // our own safe ts value and not be unlocked afterward. - for ( unsigned i = 0; i < up.size(); i++ ) { - ScopedDbConnection indDB(up[i].first); - BSONObj indUpdate; + } catch (UpdateNotTheSame& up) { + // this means our update got through on some, but not others + warning() << "distributed lock '" << lockName << " did not propagate properly." + << causedBy(up) << endl; - try { + // Overall protection derives from: + // All unlocking updates use the ts value when setting state to 0 + // This ensures that during locking, we can override all smaller ts locks with + // our own safe ts value and not be unlocked afterward. + for (unsigned i = 0; i < up.size(); i++) { + ScopedDbConnection indDB(up[i].first); + BSONObj indUpdate; - indUpdate = indDB->findOne( LocksType::ConfigNS , BSON( LocksType::name(_name) ) ); - - // If we override this lock in any way, grab and protect it. - // We assume/ensure that if a process does not have all lock documents, it is no longer - // holding the lock. - // Note - finalized locks may compete too, but we know they've won already if competing - // in this round. Cleanup of crashes during finalizing may take a few tries. - if( indUpdate[LocksType::lockID()] < lockDetails[LocksType::lockID()] || indUpdate[LocksType::state()].numberInt() == 0 ) { - - BSONObj grabQuery = BSON( LocksType::name(_name) - << LocksType::lockID(indUpdate[LocksType::lockID()].OID()) ); - - // Change ts so we won't be forced, state so we won't be relocked - BSONObj grabChanges = BSON( LocksType::lockID(lockDetails[LocksType::lockID()].OID()) - << LocksType::state(1) ); - - // Either our update will succeed, and we'll grab the lock, or it will fail b/c some other - // process grabbed the lock (which will change the ts), but the lock will be set until forcing - indDB->update( LocksType::ConfigNS, grabQuery, BSON( "$set" << grabChanges ) ); - - indUpdate = indDB->findOne( LocksType::ConfigNS, BSON( LocksType::name(_name) ) ); - - // The tournament was interfered and it is not safe to proceed further. - // One case this could happen is when the LockPinger processes old - // entries from addUnlockOID. See SERVER-10688 for more detailed - // description of race. - if ( indUpdate[LocksType::state()].numberInt() <= 0 ) { - LOG( logLvl - 1 ) << "lock tournament interrupted, " - << "so no lock was taken; " - << "new state of lock: " << indUpdate << endl; - - // We now break and set our currLock lockID value to zero, so that - // we know that we did not acquire the lock below. Later code will - // cleanup failed entries. - currLock = BSON(LocksType::lockID(OID())); - indDB.done(); - break; - } + try { + indUpdate = indDB->findOne(LocksType::ConfigNS, BSON(LocksType::name(_name))); + + // If we override this lock in any way, grab and protect it. + // We assume/ensure that if a process does not have all lock documents, it is no longer + // holding the lock. + // Note - finalized locks may compete too, but we know they've won already if competing + // in this round. Cleanup of crashes during finalizing may take a few tries. + if (indUpdate[LocksType::lockID()] < lockDetails[LocksType::lockID()] || + indUpdate[LocksType::state()].numberInt() == 0) { + BSONObj grabQuery = + BSON(LocksType::name(_name) + << LocksType::lockID(indUpdate[LocksType::lockID()].OID())); + + // Change ts so we won't be forced, state so we won't be relocked + BSONObj grabChanges = + BSON(LocksType::lockID(lockDetails[LocksType::lockID()].OID()) + << LocksType::state(1)); + + // Either our update will succeed, and we'll grab the lock, or it will fail b/c some other + // process grabbed the lock (which will change the ts), but the lock will be set until forcing + indDB->update(LocksType::ConfigNS, grabQuery, BSON("$set" << grabChanges)); + + indUpdate = indDB->findOne(LocksType::ConfigNS, BSON(LocksType::name(_name))); + + // The tournament was interfered and it is not safe to proceed further. + // One case this could happen is when the LockPinger processes old + // entries from addUnlockOID. See SERVER-10688 for more detailed + // description of race. + if (indUpdate[LocksType::state()].numberInt() <= 0) { + LOG(logLvl - 1) << "lock tournament interrupted, " + << "so no lock was taken; " + << "new state of lock: " << indUpdate << endl; + + // We now break and set our currLock lockID value to zero, so that + // we know that we did not acquire the lock below. Later code will + // cleanup failed entries. + currLock = BSON(LocksType::lockID(OID())); + indDB.done(); + break; } - // else our lock is the same, in which case we're safe, or it's a bigger lock, - // in which case we won't need to protect anything since we won't have the lock. - - } - catch( std::exception& e ) { - conn.done(); - string msg(str::stream() << "distributed lock " << lockName - << " had errors communicating with individual server " - << up[1].first << causedBy(e)); - throw LockException(msg, 13661); - } - - verify( !indUpdate.isEmpty() ); - - // Find max TS value - if ( currLock.isEmpty() || currLock[LocksType::lockID()] < indUpdate[LocksType::lockID()] ) { - currLock = indUpdate.getOwned(); } + // else our lock is the same, in which case we're safe, or it's a bigger lock, + // in which case we won't need to protect anything since we won't have the lock. - indDB.done(); - + } catch (std::exception& e) { + conn.done(); + string msg(str::stream() << "distributed lock " << lockName + << " had errors communicating with individual server " + << up[1].first << causedBy(e)); + throw LockException(msg, 13661); } - // Locks on all servers are now set and safe until forcing + verify(!indUpdate.isEmpty()); - if ( currLock[LocksType::lockID()] == lockDetails[LocksType::lockID()] ) { - LOG( logLvl - 1 ) << "lock update won, completing lock propagation for '" << lockName << "'" << endl; - gotLock = true; + // Find max TS value + if (currLock.isEmpty() || + currLock[LocksType::lockID()] < indUpdate[LocksType::lockID()]) { + currLock = indUpdate.getOwned(); } - else { - LOG( logLvl - 1 ) << "lock update lost, lock '" << lockName << "' not propagated." << endl; - gotLock = false; - } - } - catch( std::exception& e ) { - conn.done(); - string msg(str::stream() << "exception creating distributed lock " - << lockName << causedBy(e)); - throw LockException(msg, 13663 ); - } - // Complete lock propagation - if( gotLock ) { + indDB.done(); + } - // This is now safe, since we know that no new locks will be placed on top of the ones we've checked for at - // least 15 minutes. Sets the state = 2, so that future clients can determine that the lock is truly set. - // The invariant for rollbacks is that we will never force locks with state = 2 and active pings, since that - // indicates the lock is active, but this means the process creating/destroying them must explicitly poll - // when something goes wrong. - try { + // Locks on all servers are now set and safe until forcing - BSONObjBuilder finalLockDetails; - BSONObjIterator bi( lockDetails ); - while( bi.more() ) { - BSONElement el = bi.next(); - if( (string) ( el.fieldName() ) == LocksType::state() ) - finalLockDetails.append( LocksType::state(), 2 ); - else finalLockDetails.append( el ); - } + if (currLock[LocksType::lockID()] == lockDetails[LocksType::lockID()]) { + LOG(logLvl - 1) << "lock update won, completing lock propagation for '" << lockName + << "'" << endl; + gotLock = true; + } else { + LOG(logLvl - 1) << "lock update lost, lock '" << lockName << "' not propagated." + << endl; + gotLock = false; + } + } catch (std::exception& e) { + conn.done(); + string msg(str::stream() << "exception creating distributed lock " << lockName + << causedBy(e)); + throw LockException(msg, 13663); + } - conn->update( LocksType::ConfigNS , BSON( LocksType::name(_name) ) , BSON( "$set" << finalLockDetails.obj() ) ); + // Complete lock propagation + if (gotLock) { + // This is now safe, since we know that no new locks will be placed on top of the ones we've checked for at + // least 15 minutes. Sets the state = 2, so that future clients can determine that the lock is truly set. + // The invariant for rollbacks is that we will never force locks with state = 2 and active pings, since that + // indicates the lock is active, but this means the process creating/destroying them must explicitly poll + // when something goes wrong. + try { + BSONObjBuilder finalLockDetails; + BSONObjIterator bi(lockDetails); + while (bi.more()) { + BSONElement el = bi.next(); + if ((string)(el.fieldName()) == LocksType::state()) + finalLockDetails.append(LocksType::state(), 2); + else + finalLockDetails.append(el); + } - BSONObj err = conn->getLastErrorDetailed(); - string errMsg = DBClientWithCommands::getLastErrorString(err); + conn->update(LocksType::ConfigNS, + BSON(LocksType::name(_name)), + BSON("$set" << finalLockDetails.obj())); - currLock = conn->findOne( LocksType::ConfigNS , BSON( LocksType::name(_name) ) ); + BSONObj err = conn->getLastErrorDetailed(); + string errMsg = DBClientWithCommands::getLastErrorString(err); - if ( !errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1 ) { - warning() << "could not finalize winning lock " << lockName - << ( !errMsg.empty() ? causedBy( errMsg ) : " (did not update lock) " ) << endl; - gotLock = false; - } - else { - // SUCCESS! - gotLock = true; - } + currLock = conn->findOne(LocksType::ConfigNS, BSON(LocksType::name(_name))); - } - catch( std::exception& e ) { - conn.done(); - string msg(str::stream() << "exception finalizing winning lock" << causedBy(e)); - // Inform caller about the potential orphan lock. - throw LockException(msg, - 13662, - lockDetails[LocksType::lockID()].OID()); + if (!errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1) { + warning() << "could not finalize winning lock " << lockName + << (!errMsg.empty() ? causedBy(errMsg) : " (did not update lock) ") + << endl; + gotLock = false; + } else { + // SUCCESS! + gotLock = true; } + } catch (std::exception& e) { + conn.done(); + string msg(str::stream() << "exception finalizing winning lock" << causedBy(e)); + // Inform caller about the potential orphan lock. + throw LockException(msg, 13662, lockDetails[LocksType::lockID()].OID()); } - - *other = currLock; - other->getOwned(); - - // Log our lock results - if(gotLock) - LOG( logLvl - 1 ) << "distributed lock '" << lockName << - "' acquired, ts : " << currLock[LocksType::lockID()].OID() << endl; - else - LOG( logLvl - 1 ) << "distributed lock '" << lockName << "' was not acquired." << endl; - - conn.done(); - - return gotLock; } - // This function *must not* throw exceptions, since it can be used in destructors - failure - // results in queuing and trying again later. - bool DistributedLock::unlock(const DistLockHandle& lockID) { + *other = currLock; + other->getOwned(); - verify(_name != ""); - string lockName = _name + string("/") + _processId; + // Log our lock results + if (gotLock) + LOG(logLvl - 1) << "distributed lock '" << lockName + << "' acquired, ts : " << currLock[LocksType::lockID()].OID() << endl; + else + LOG(logLvl - 1) << "distributed lock '" << lockName << "' was not acquired." << endl; - const int maxAttempts = 3; - int attempted = 0; + conn.done(); - while ( ++attempted <= maxAttempts ) { + return gotLock; +} - // Awkward, but necessary since the constructor itself throws exceptions - unique_ptr<ScopedDbConnection> connPtr; +// This function *must not* throw exceptions, since it can be used in destructors - failure +// results in queuing and trying again later. +bool DistributedLock::unlock(const DistLockHandle& lockID) { + verify(_name != ""); + string lockName = _name + string("/") + _processId; - try { + const int maxAttempts = 3; + int attempted = 0; - connPtr.reset( new ScopedDbConnection( _conn.toString() ) ); - ScopedDbConnection& conn = *connPtr; + while (++attempted <= maxAttempts) { + // Awkward, but necessary since the constructor itself throws exceptions + unique_ptr<ScopedDbConnection> connPtr; - // Use ts when updating lock, so that new locks can be sure they won't get trampled. - conn->update(LocksType::ConfigNS, - BSON(LocksType::name(_name) - << LocksType::lockID(lockID)), - BSON("$set" << BSON(LocksType::state(0)))); + try { + connPtr.reset(new ScopedDbConnection(_conn.toString())); + ScopedDbConnection& conn = *connPtr; - // Check that the lock was actually unlocked... if not, try again - BSONObj err = conn->getLastErrorDetailed(); - string errMsg = DBClientWithCommands::getLastErrorString(err); + // Use ts when updating lock, so that new locks can be sure they won't get trampled. + conn->update(LocksType::ConfigNS, + BSON(LocksType::name(_name) << LocksType::lockID(lockID)), + BSON("$set" << BSON(LocksType::state(0)))); - if ( !errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1 ){ - warning() << "distributed lock unlock update failed, retrying " - << ( errMsg.empty() ? causedBy( "( update not registered )" ) : causedBy( errMsg ) ) << endl; - conn.done(); - continue; - } + // Check that the lock was actually unlocked... if not, try again + BSONObj err = conn->getLastErrorDetailed(); + string errMsg = DBClientWithCommands::getLastErrorString(err); - LOG(0) << "distributed lock '" << lockName << "' unlocked. "; + if (!errMsg.empty() || !err["n"].type() || err["n"].numberInt() < 1) { + warning() << "distributed lock unlock update failed, retrying " + << (errMsg.empty() ? causedBy("( update not registered )") + : causedBy(errMsg)) << endl; conn.done(); - return true; - } - catch (const UpdateNotTheSame&) { - LOG(0) << "distributed lock '" << lockName << "' unlocked (messily). "; - // This isn't a connection problem, so don't throw away the conn - connPtr->done(); - return true; + continue; } - catch ( std::exception& e) { - warning() << "distributed lock '" << lockName << "' failed unlock attempt." - << causedBy( e ) << endl; - // TODO: If our lock timeout is small, sleeping this long may be unsafe. - if( attempted != maxAttempts) sleepsecs(1 << attempted); - } + LOG(0) << "distributed lock '" << lockName << "' unlocked. "; + conn.done(); + return true; + } catch (const UpdateNotTheSame&) { + LOG(0) << "distributed lock '" << lockName << "' unlocked (messily). "; + // This isn't a connection problem, so don't throw away the conn + connPtr->done(); + return true; + } catch (std::exception& e) { + warning() << "distributed lock '" << lockName << "' failed unlock attempt." + << causedBy(e) << endl; + + // TODO: If our lock timeout is small, sleeping this long may be unsafe. + if (attempted != maxAttempts) + sleepsecs(1 << attempted); } - - return false; } + return false; +} } diff --git a/src/mongo/s/catalog/legacy/distlock.h b/src/mongo/s/catalog/legacy/distlock.h index b569080147e..b267be42a04 100644 --- a/src/mongo/s/catalog/legacy/distlock.h +++ b/src/mongo/s/catalog/legacy/distlock.h @@ -37,199 +37,198 @@ namespace mongo { - namespace { - - enum TimeConstants { - LOCK_TIMEOUT = 15 * 60 * 1000, - LOCK_SKEW_FACTOR = 30, - LOCK_PING = LOCK_TIMEOUT / LOCK_SKEW_FACTOR, - MAX_LOCK_NET_SKEW = LOCK_TIMEOUT / LOCK_SKEW_FACTOR, - MAX_LOCK_CLOCK_SKEW = LOCK_TIMEOUT / LOCK_SKEW_FACTOR, - NUM_LOCK_SKEW_CHECKS = 3, - }; - - // The maximum clock skew we need to handle between config servers is - // 2 * MAX_LOCK_NET_SKEW + MAX_LOCK_CLOCK_SKEW. - - // Net effect of *this* clock being slow is effectively a multiplier on the max net skew - // and a linear increase or decrease of the max clock skew. - } +namespace { + +enum TimeConstants { + LOCK_TIMEOUT = 15 * 60 * 1000, + LOCK_SKEW_FACTOR = 30, + LOCK_PING = LOCK_TIMEOUT / LOCK_SKEW_FACTOR, + MAX_LOCK_NET_SKEW = LOCK_TIMEOUT / LOCK_SKEW_FACTOR, + MAX_LOCK_CLOCK_SKEW = LOCK_TIMEOUT / LOCK_SKEW_FACTOR, + NUM_LOCK_SKEW_CHECKS = 3, +}; + +// The maximum clock skew we need to handle between config servers is +// 2 * MAX_LOCK_NET_SKEW + MAX_LOCK_CLOCK_SKEW. + +// Net effect of *this* clock being slow is effectively a multiplier on the max net skew +// and a linear increase or decrease of the max clock skew. +} + +/** + * Exception class to encapsulate exceptions while managing distributed locks + */ +class LockException : public DBException { +public: + LockException(StringData msg, int code); /** - * Exception class to encapsulate exceptions while managing distributed locks + * Use this to signal that a lock with lockID needs to be unlocked. For example, in cases + * where the final lock acquisition was not propagated properly to all config servers. */ - class LockException : public DBException { - public: - LockException(StringData msg, int code); + LockException(StringData msg, int code, DistLockHandle lockID); - /** - * Use this to signal that a lock with lockID needs to be unlocked. For example, in cases - * where the final lock acquisition was not propagated properly to all config servers. - */ - LockException(StringData msg, int code, DistLockHandle lockID); + /** + * Returns the OID of the lock that needs to be unlocked. + */ + DistLockHandle getMustUnlockID() const; - /** - * Returns the OID of the lock that needs to be unlocked. - */ - DistLockHandle getMustUnlockID() const; + virtual ~LockException() = default; - virtual ~LockException() = default; +private: + // The identifier of a lock that needs to be unlocked. + DistLockHandle _mustUnlockID; +}; - private: - // The identifier of a lock that needs to be unlocked. - DistLockHandle _mustUnlockID; - }; +/** + * Indicates an error in retrieving time values from remote servers. + */ +class TimeNotFoundException : public LockException { +public: + TimeNotFoundException(const char* msg, int code) : LockException(msg, code) {} + TimeNotFoundException(const std::string& msg, int code) : LockException(msg, code) {} + virtual ~TimeNotFoundException() = default; +}; + +/** + * The distributed lock is a configdb backed way of synchronizing system-wide tasks. A task + * must be identified by a unique name across the system (e.g., "balancer"). A lock is taken + * by writing a document in the configdb's locks collection with that name. + * + * To be maintained, each taken lock needs to be revalidated ("pinged") within a + * pre-established amount of time. This class does this maintenance automatically once a + * DistributedLock object was constructed. The ping procedure records the local time to + * the ping document, but that time is untrusted and is only used as a point of reference + * of whether the ping was refreshed or not. Ultimately, the clock a configdb is the source + * of truth when determining whether a ping is still fresh or not. This is achieved by + * (1) remembering the ping document time along with config server time when unable to + * take a lock, and (2) ensuring all config servers report similar times and have similar + * time rates (the difference in times must start and stay small). + * + * Lock states include: + * 0: unlocked + * 1: about to be locked + * 2: locked + * + * Valid state transitions: + * 0 -> 1 + * 1 -> 2 + * 2 -> 0 + * + * Note that at any point in time, a lock can be force unlocked if the ping for the lock + * becomes too stale. + */ +class DistributedLock { +public: + static logger::LabeledLevel logLvl; - /** - * Indicates an error in retrieving time values from remote servers. - */ - class TimeNotFoundException : public LockException { + class LastPings { public: - TimeNotFoundException( const char * msg , int code ) : LockException( msg, code ) {} - TimeNotFoundException( const std::string& msg, int code ) : LockException( msg, code ) {} - virtual ~TimeNotFoundException() = default; + DistLockPingInfo getLastPing(const ConnectionString& conn, const std::string& lockName); + void setLastPing(const ConnectionString& conn, + const std::string& lockName, + const DistLockPingInfo& pd); + + stdx::mutex _mutex; + std::map<std::pair<std::string, std::string>, DistLockPingInfo> _lastPings; }; + static LastPings lastPings; + /** - * The distributed lock is a configdb backed way of synchronizing system-wide tasks. A task - * must be identified by a unique name across the system (e.g., "balancer"). A lock is taken - * by writing a document in the configdb's locks collection with that name. + * The constructor does not connect to the configdb yet and constructing does not mean the lock was acquired. + * Construction does trigger a lock "pinging" mechanism, though. * - * To be maintained, each taken lock needs to be revalidated ("pinged") within a - * pre-established amount of time. This class does this maintenance automatically once a - * DistributedLock object was constructed. The ping procedure records the local time to - * the ping document, but that time is untrusted and is only used as a point of reference - * of whether the ping was refreshed or not. Ultimately, the clock a configdb is the source - * of truth when determining whether a ping is still fresh or not. This is achieved by - * (1) remembering the ping document time along with config server time when unable to - * take a lock, and (2) ensuring all config servers report similar times and have similar - * time rates (the difference in times must start and stay small). + * @param conn address of config(s) server(s) + * @param name identifier for the lock + * @param lockTimeout how long can the log go "unpinged" before a new attempt to lock steals it (in minutes). + * @param lockPing how long to wait between lock pings + * @param legacy use legacy logic * - * Lock states include: - * 0: unlocked - * 1: about to be locked - * 2: locked - * - * Valid state transitions: - * 0 -> 1 - * 1 -> 2 - * 2 -> 0 + */ + DistributedLock(const ConnectionString& conn, + const std::string& name, + unsigned long long lockTimeout = 0, + bool asProcess = false); + ~DistributedLock(){}; + + /** + * Attempts to acquire 'this' lock, checking if it could or should be stolen from the previous holder. Please + * consider using the dist_lock_try construct to acquire this lock in an exception safe way. * - * Note that at any point in time, a lock can be force unlocked if the ping for the lock - * becomes too stale. + * @param why human readable description of why the lock is being taken (used to log) + * @param other configdb's lock document that is currently holding the lock, if lock is taken, or our own lock + * details if not + * @return true if it managed to grab the lock */ - class DistributedLock { - public: + bool lock_try(const std::string& why, BSONObj* other = 0, double timeout = 0.0); - static logger::LabeledLevel logLvl; - - class LastPings { - public: - DistLockPingInfo getLastPing(const ConnectionString& conn, - const std::string& lockName); - void setLastPing(const ConnectionString& conn, - const std::string& lockName, - const DistLockPingInfo& pd); - - stdx::mutex _mutex; - std::map< std::pair<std::string, std::string>, DistLockPingInfo > _lastPings; - }; - - static LastPings lastPings; - - /** - * The constructor does not connect to the configdb yet and constructing does not mean the lock was acquired. - * Construction does trigger a lock "pinging" mechanism, though. - * - * @param conn address of config(s) server(s) - * @param name identifier for the lock - * @param lockTimeout how long can the log go "unpinged" before a new attempt to lock steals it (in minutes). - * @param lockPing how long to wait between lock pings - * @param legacy use legacy logic - * - */ - DistributedLock( const ConnectionString& conn , const std::string& name , unsigned long long lockTimeout = 0, bool asProcess = false ); - ~DistributedLock(){}; - - /** - * Attempts to acquire 'this' lock, checking if it could or should be stolen from the previous holder. Please - * consider using the dist_lock_try construct to acquire this lock in an exception safe way. - * - * @param why human readable description of why the lock is being taken (used to log) - * @param other configdb's lock document that is currently holding the lock, if lock is taken, or our own lock - * details if not - * @return true if it managed to grab the lock - */ - bool lock_try(const std::string& why, BSONObj* other = 0, double timeout = 0.0); - - /** - * Returns OK if this lock is held (but does not guarantee that this owns it) and - * it was possible to confirm that, within 'timeout' seconds, if provided, with the - * config servers. - */ - Status checkStatus(double timeout); - - /** - * Releases a previously taken lock. Returns true on success. - */ - bool unlock(const OID& lockID); - - Date_t getRemoteTime() const; - - bool isRemoteTimeSkewed() const; - - const std::string& getProcessId() const; - - const ConnectionString& getRemoteConnection() const; - - /** - * Checks the skew among a cluster of servers and returns true if the min and max clock - * times among the servers are within maxClockSkew. - */ - static bool checkSkew( const ConnectionString& cluster, - unsigned skewChecks = NUM_LOCK_SKEW_CHECKS, - unsigned long long maxClockSkew = MAX_LOCK_CLOCK_SKEW, - unsigned long long maxNetSkew = MAX_LOCK_NET_SKEW ); - - /** - * Get the remote time from a server or cluster - */ - static Date_t remoteTime( const ConnectionString& cluster, unsigned long long maxNetSkew = MAX_LOCK_NET_SKEW ); - - /** - * Namespace for lock pings - */ - static const std::string lockPingNS; - - /** - * Namespace for locks - */ - static const std::string locksNS; - - const ConnectionString _conn; - const std::string _name; - const std::string _processId; - - // Timeout for lock, usually LOCK_TIMEOUT - const unsigned long long _lockTimeout; - const unsigned long long _maxClockSkew; - const unsigned long long _maxNetSkew; - const unsigned long long _lockPing; - - private: - - void resetLastPing() { - lastPings.setLastPing(_conn, _name, DistLockPingInfo()); - } - - void setLastPing(const DistLockPingInfo& pd) { - lastPings.setLastPing(_conn, _name, pd); - } - - DistLockPingInfo getLastPing() { - return lastPings.getLastPing(_conn, _name); - } - }; + /** + * Returns OK if this lock is held (but does not guarantee that this owns it) and + * it was possible to confirm that, within 'timeout' seconds, if provided, with the + * config servers. + */ + Status checkStatus(double timeout); -} + /** + * Releases a previously taken lock. Returns true on success. + */ + bool unlock(const OID& lockID); + + Date_t getRemoteTime() const; + + bool isRemoteTimeSkewed() const; + + const std::string& getProcessId() const; + + const ConnectionString& getRemoteConnection() const; + + /** + * Checks the skew among a cluster of servers and returns true if the min and max clock + * times among the servers are within maxClockSkew. + */ + static bool checkSkew(const ConnectionString& cluster, + unsigned skewChecks = NUM_LOCK_SKEW_CHECKS, + unsigned long long maxClockSkew = MAX_LOCK_CLOCK_SKEW, + unsigned long long maxNetSkew = MAX_LOCK_NET_SKEW); + + /** + * Get the remote time from a server or cluster + */ + static Date_t remoteTime(const ConnectionString& cluster, + unsigned long long maxNetSkew = MAX_LOCK_NET_SKEW); + + /** + * Namespace for lock pings + */ + static const std::string lockPingNS; + + /** + * Namespace for locks + */ + static const std::string locksNS; + + const ConnectionString _conn; + const std::string _name; + const std::string _processId; + + // Timeout for lock, usually LOCK_TIMEOUT + const unsigned long long _lockTimeout; + const unsigned long long _maxClockSkew; + const unsigned long long _maxNetSkew; + const unsigned long long _lockPing; +private: + void resetLastPing() { + lastPings.setLastPing(_conn, _name, DistLockPingInfo()); + } + + void setLastPing(const DistLockPingInfo& pd) { + lastPings.setLastPing(_conn, _name, pd); + } + + DistLockPingInfo getLastPing() { + return lastPings.getLastPing(_conn, _name); + } +}; +} diff --git a/src/mongo/s/catalog/legacy/legacy_dist_lock_manager.cpp b/src/mongo/s/catalog/legacy/legacy_dist_lock_manager.cpp index 18cc7d537cf..e375c792f28 100644 --- a/src/mongo/s/catalog/legacy/legacy_dist_lock_manager.cpp +++ b/src/mongo/s/catalog/legacy/legacy_dist_lock_manager.cpp @@ -39,187 +39,176 @@ namespace mongo { - using std::string; - using std::unique_ptr; - using stdx::chrono::milliseconds; +using std::string; +using std::unique_ptr; +using stdx::chrono::milliseconds; namespace { - const stdx::chrono::seconds kDefaultSocketTimeout(30); - const milliseconds kDefaultPingInterval(30 * 1000); -} // unnamed namespace - - LegacyDistLockManager::LegacyDistLockManager(ConnectionString configServer): - _configServer(std::move(configServer)), - _isStopped(false), - _pingerEnabled(true) { - } +const stdx::chrono::seconds kDefaultSocketTimeout(30); +const milliseconds kDefaultPingInterval(30 * 1000); +} // unnamed namespace - void LegacyDistLockManager::startUp() { - stdx::lock_guard<stdx::mutex> sl(_mutex); - invariant(!_pinger); - _pinger = stdx::make_unique<LegacyDistLockPinger>(); - } +LegacyDistLockManager::LegacyDistLockManager(ConnectionString configServer) + : _configServer(std::move(configServer)), _isStopped(false), _pingerEnabled(true) {} - void LegacyDistLockManager::shutDown() { - stdx::unique_lock<stdx::mutex> sl(_mutex); - _isStopped = true; +void LegacyDistLockManager::startUp() { + stdx::lock_guard<stdx::mutex> sl(_mutex); + invariant(!_pinger); + _pinger = stdx::make_unique<LegacyDistLockPinger>(); +} - while (!_lockMap.empty()) { - _noLocksCV.wait(sl); - } +void LegacyDistLockManager::shutDown() { + stdx::unique_lock<stdx::mutex> sl(_mutex); + _isStopped = true; - if (_pinger) { - _pinger->shutdown(); - } + while (!_lockMap.empty()) { + _noLocksCV.wait(sl); } - StatusWith<DistLockManager::ScopedDistLock> LegacyDistLockManager::lock( - StringData name, - StringData whyMessage, - milliseconds waitFor, - milliseconds lockTryInterval) { - - auto distLock = stdx::make_unique<DistributedLock>(_configServer, name.toString()); + if (_pinger) { + _pinger->shutdown(); + } +} - { - stdx::lock_guard<stdx::mutex> sl(_mutex); +StatusWith<DistLockManager::ScopedDistLock> LegacyDistLockManager::lock( + StringData name, StringData whyMessage, milliseconds waitFor, milliseconds lockTryInterval) { + auto distLock = stdx::make_unique<DistributedLock>(_configServer, name.toString()); - if (_isStopped) { - return Status(ErrorCodes::LockBusy, "legacy distlock manager is stopped"); - } + { + stdx::lock_guard<stdx::mutex> sl(_mutex); - if (_pingerEnabled) { - auto pingStatus = _pinger->startPing(*(distLock.get()), kDefaultPingInterval); - if (!pingStatus.isOK()) { - return pingStatus; - } - } + if (_isStopped) { + return Status(ErrorCodes::LockBusy, "legacy distlock manager is stopped"); } - auto lastStatus = Status(ErrorCodes::LockBusy, - str::stream() << "timed out waiting for " << name); - - Timer timer; - Timer msgTimer; - while (waitFor <= milliseconds::zero() || milliseconds(timer.millis()) < waitFor) { - bool acquired = false; - BSONObj lockDoc; - try { - acquired = distLock->lock_try(whyMessage.toString(), - &lockDoc, - kDefaultSocketTimeout.count()); - - if (!acquired) { - lastStatus = Status(ErrorCodes::LockBusy, - str::stream() << "Lock for " << whyMessage - << " is taken."); - } + if (_pingerEnabled) { + auto pingStatus = _pinger->startPing(*(distLock.get()), kDefaultPingInterval); + if (!pingStatus.isOK()) { + return pingStatus; } - catch (const LockException& lockExcep) { - OID needUnlockID(lockExcep.getMustUnlockID()); - if (needUnlockID.isSet()) { - _pinger->addUnlockOID(needUnlockID); - } + } + } - lastStatus = lockExcep.toStatus(); + auto lastStatus = + Status(ErrorCodes::LockBusy, str::stream() << "timed out waiting for " << name); + + Timer timer; + Timer msgTimer; + while (waitFor <= milliseconds::zero() || milliseconds(timer.millis()) < waitFor) { + bool acquired = false; + BSONObj lockDoc; + try { + acquired = + distLock->lock_try(whyMessage.toString(), &lockDoc, kDefaultSocketTimeout.count()); + + if (!acquired) { + lastStatus = Status(ErrorCodes::LockBusy, + str::stream() << "Lock for " << whyMessage << " is taken."); } - catch (...) { - lastStatus = exceptionToStatus(); + } catch (const LockException& lockExcep) { + OID needUnlockID(lockExcep.getMustUnlockID()); + if (needUnlockID.isSet()) { + _pinger->addUnlockOID(needUnlockID); } - if (acquired) { - verify(!lockDoc.isEmpty()); + lastStatus = lockExcep.toStatus(); + } catch (...) { + lastStatus = exceptionToStatus(); + } - LocksType lock; - string errMsg; - if (!lock.parseBSON(lockDoc, &errMsg)) { - return StatusWith<ScopedDistLock>( - ErrorCodes::UnsupportedFormat, - str::stream() << "error while parsing lock document: " << errMsg); - } + if (acquired) { + verify(!lockDoc.isEmpty()); - dassert(lock.isLockIDSet()); + LocksType lock; + string errMsg; + if (!lock.parseBSON(lockDoc, &errMsg)) { + return StatusWith<ScopedDistLock>( + ErrorCodes::UnsupportedFormat, + str::stream() << "error while parsing lock document: " << errMsg); + } - { - stdx::lock_guard<stdx::mutex> sl(_mutex); - _lockMap.insert(std::make_pair(lock.getLockID(), std::move(distLock))); - } + dassert(lock.isLockIDSet()); - return ScopedDistLock(lock.getLockID(), this); + { + stdx::lock_guard<stdx::mutex> sl(_mutex); + _lockMap.insert(std::make_pair(lock.getLockID(), std::move(distLock))); } - if (waitFor == milliseconds::zero()) break; + return ScopedDistLock(lock.getLockID(), this); + } - if (lastStatus != ErrorCodes::LockBusy) { - return lastStatus; - } + if (waitFor == milliseconds::zero()) + break; - // Periodically message for debugging reasons - if (msgTimer.seconds() > 10) { - log() << "waited " << timer.seconds() << "s for distributed lock " << name - << " for " << whyMessage << ": " << lastStatus.toString(); + if (lastStatus != ErrorCodes::LockBusy) { + return lastStatus; + } - msgTimer.reset(); - } + // Periodically message for debugging reasons + if (msgTimer.seconds() > 10) { + log() << "waited " << timer.seconds() << "s for distributed lock " << name << " for " + << whyMessage << ": " << lastStatus.toString(); - milliseconds timeRemaining = - std::max(milliseconds::zero(), waitFor - milliseconds(timer.millis())); - sleepFor(std::min(lockTryInterval, timeRemaining)); + msgTimer.reset(); } - return lastStatus; + milliseconds timeRemaining = + std::max(milliseconds::zero(), waitFor - milliseconds(timer.millis())); + sleepFor(std::min(lockTryInterval, timeRemaining)); } - void LegacyDistLockManager::unlock(const DistLockHandle& lockHandle) BOOST_NOEXCEPT { - unique_ptr<DistributedLock> distLock; + return lastStatus; +} - { - stdx::lock_guard<stdx::mutex> sl(_mutex); - auto iter = _lockMap.find(lockHandle); - invariant(iter != _lockMap.end()); +void LegacyDistLockManager::unlock(const DistLockHandle& lockHandle) BOOST_NOEXCEPT { + unique_ptr<DistributedLock> distLock; - distLock = std::move(iter->second); - _lockMap.erase(iter); - } + { + stdx::lock_guard<stdx::mutex> sl(_mutex); + auto iter = _lockMap.find(lockHandle); + invariant(iter != _lockMap.end()); - if (!distLock->unlock(lockHandle)) { - _pinger->addUnlockOID(lockHandle); - } + distLock = std::move(iter->second); + _lockMap.erase(iter); + } - { - stdx::lock_guard<stdx::mutex> sl(_mutex); - if (_lockMap.empty()) { - _noLocksCV.notify_all(); - } - } + if (!distLock->unlock(lockHandle)) { + _pinger->addUnlockOID(lockHandle); } - Status LegacyDistLockManager::checkStatus(const DistLockHandle& lockHandle) { - // Note: this should not happen when locks are managed through ScopedDistLock. - if (_pinger->willUnlockOID(lockHandle)) { - return Status(ErrorCodes::LockFailed, - str::stream() << "lock " << lockHandle << " is not held and " - << "is currently being scheduled for lazy unlock"); + { + stdx::lock_guard<stdx::mutex> sl(_mutex); + if (_lockMap.empty()) { + _noLocksCV.notify_all(); } + } +} - DistributedLock* distLock = nullptr; +Status LegacyDistLockManager::checkStatus(const DistLockHandle& lockHandle) { + // Note: this should not happen when locks are managed through ScopedDistLock. + if (_pinger->willUnlockOID(lockHandle)) { + return Status(ErrorCodes::LockFailed, + str::stream() << "lock " << lockHandle << " is not held and " + << "is currently being scheduled for lazy unlock"); + } - { - // Assumption: lockHandles are never shared across threads. - stdx::lock_guard<stdx::mutex> sl(_mutex); - auto iter = _lockMap.find(lockHandle); - invariant(iter != _lockMap.end()); + DistributedLock* distLock = nullptr; - distLock = iter->second.get(); - } + { + // Assumption: lockHandles are never shared across threads. + stdx::lock_guard<stdx::mutex> sl(_mutex); + auto iter = _lockMap.find(lockHandle); + invariant(iter != _lockMap.end()); - return distLock->checkStatus(kDefaultSocketTimeout.count()); + distLock = iter->second.get(); } - void LegacyDistLockManager::enablePinger(bool enable) { - stdx::lock_guard<stdx::mutex> sl(_mutex); - _pingerEnabled = enable; - } + return distLock->checkStatus(kDefaultSocketTimeout.count()); +} +void LegacyDistLockManager::enablePinger(bool enable) { + stdx::lock_guard<stdx::mutex> sl(_mutex); + _pingerEnabled = enable; +} } diff --git a/src/mongo/s/catalog/legacy/legacy_dist_lock_manager.h b/src/mongo/s/catalog/legacy/legacy_dist_lock_manager.h index 50a6088c333..238989a3cdb 100644 --- a/src/mongo/s/catalog/legacy/legacy_dist_lock_manager.h +++ b/src/mongo/s/catalog/legacy/legacy_dist_lock_manager.h @@ -41,46 +41,43 @@ namespace mongo { - class DistributedLock; +class DistributedLock; - class LegacyDistLockManager: public DistLockManager { - public: - explicit LegacyDistLockManager(ConnectionString configServer); +class LegacyDistLockManager : public DistLockManager { +public: + explicit LegacyDistLockManager(ConnectionString configServer); - virtual ~LegacyDistLockManager() = default; + virtual ~LegacyDistLockManager() = default; - virtual void startUp() override; - virtual void shutDown() override; + virtual void startUp() override; + virtual void shutDown() override; - virtual StatusWith<DistLockManager::ScopedDistLock> lock( - StringData name, - StringData whyMessage, - stdx::chrono::milliseconds waitFor, - stdx::chrono::milliseconds lockTryInterval) override; + virtual StatusWith<DistLockManager::ScopedDistLock> lock( + StringData name, + StringData whyMessage, + stdx::chrono::milliseconds waitFor, + stdx::chrono::milliseconds lockTryInterval) override; - // For testing only. - void enablePinger(bool enable); + // For testing only. + void enablePinger(bool enable); - protected: +protected: + virtual void unlock(const DistLockHandle& lockHandle) BOOST_NOEXCEPT override; - virtual void unlock(const DistLockHandle& lockHandle) BOOST_NOEXCEPT override; + virtual Status checkStatus(const DistLockHandle& lockHandle) override; - virtual Status checkStatus(const DistLockHandle& lockHandle) override; +private: + const ConnectionString _configServer; - private: + stdx::mutex _mutex; + stdx::condition_variable _noLocksCV; + std::map<DistLockHandle, std::unique_ptr<DistributedLock>> _lockMap; - const ConnectionString _configServer; + std::unique_ptr<LegacyDistLockPinger> _pinger; - stdx::mutex _mutex; - stdx::condition_variable _noLocksCV; - std::map<DistLockHandle, std::unique_ptr<DistributedLock>> _lockMap; - - std::unique_ptr<LegacyDistLockPinger> _pinger; - - bool _isStopped; - - // For testing only. - bool _pingerEnabled; - }; + bool _isStopped; + // For testing only. + bool _pingerEnabled; +}; } diff --git a/src/mongo/s/catalog/legacy/legacy_dist_lock_pinger.cpp b/src/mongo/s/catalog/legacy/legacy_dist_lock_pinger.cpp index 5e5f03b0864..ef5299fd558 100644 --- a/src/mongo/s/catalog/legacy/legacy_dist_lock_pinger.cpp +++ b/src/mongo/s/catalog/legacy/legacy_dist_lock_pinger.cpp @@ -42,318 +42,304 @@ namespace mongo { - using std::string; +using std::string; namespace { - string pingThreadId(const ConnectionString& conn, const string& processId) { - return conn.toString() + "/" + processId; - } +string pingThreadId(const ConnectionString& conn, const string& processId) { + return conn.toString() + "/" + processId; +} } - void LegacyDistLockPinger::_distLockPingThread(ConnectionString addr, - const string& process, - Milliseconds sleepTime) { - setThreadName("LockPinger"); - - string pingId = pingThreadId(addr, process); - - LOG(0) << "creating distributed lock ping thread for " << addr - << " and process " << process << " (sleeping for " << sleepTime.count() << "ms)"; - - static int loops = 0; - Date_t lastPingTime = jsTime(); - while (!shouldStopPinging(addr, process)) { +void LegacyDistLockPinger::_distLockPingThread(ConnectionString addr, + const string& process, + Milliseconds sleepTime) { + setThreadName("LockPinger"); - LOG(3) << "distributed lock pinger '" << pingId << "' about to ping."; + string pingId = pingThreadId(addr, process); - Date_t pingTime; + LOG(0) << "creating distributed lock ping thread for " << addr << " and process " << process + << " (sleeping for " << sleepTime.count() << "ms)"; - try { - ScopedDbConnection conn(addr.toString(), 30.0); + static int loops = 0; + Date_t lastPingTime = jsTime(); + while (!shouldStopPinging(addr, process)) { + LOG(3) << "distributed lock pinger '" << pingId << "' about to ping."; - pingTime = jsTime(); - const auto elapsed = pingTime - lastPingTime; - if (elapsed > 10 * sleepTime) { - warning() << "Lock pinger for addr: " << addr - << ", proc: " << process - << " was inactive for " << elapsed; - } + Date_t pingTime; - lastPingTime = pingTime; + try { + ScopedDbConnection conn(addr.toString(), 30.0); - // Refresh the entry corresponding to this process in the lockpings collection. - conn->update(LockpingsType::ConfigNS, - BSON(LockpingsType::process(process)), - BSON("$set" << BSON(LockpingsType::ping(pingTime))), - true); + pingTime = jsTime(); + const auto elapsed = pingTime - lastPingTime; + if (elapsed > 10 * sleepTime) { + warning() << "Lock pinger for addr: " << addr << ", proc: " << process + << " was inactive for " << elapsed; + } - string err = conn->getLastError(); - if (!err.empty()) { - warning() << "pinging failed for distributed lock pinger '" << pingId << "'." - << causedBy(err); - conn.done(); + lastPingTime = pingTime; - if (!shouldStopPinging(addr, process)) { - waitTillNextPingTime(sleepTime); - } - continue; - } + // Refresh the entry corresponding to this process in the lockpings collection. + conn->update(LockpingsType::ConfigNS, + BSON(LockpingsType::process(process)), + BSON("$set" << BSON(LockpingsType::ping(pingTime))), + true); - // Remove really old entries from the lockpings collection if they're not - // holding a lock. This may happen if an instance of a process was taken down - // and no new instance came up to replace it for a quite a while. - // NOTE this is NOT the same as the standard take-over mechanism, which forces - // the lock entry. - BSONObj fieldsToReturn = BSON(LocksType::state() << 1 - << LocksType::process() << 1); - auto activeLocks = - conn->query(LocksType::ConfigNS, - BSON(LocksType::state() << NE << LocksType::UNLOCKED)); - - uassert(16060, - str::stream() << "cannot query locks collection on config server " - << conn.getHost(), - activeLocks.get()); - - std::set<string> pids; - while (activeLocks->more()) { - BSONObj lock = activeLocks->nextSafe(); - - if (!lock[LocksType::process()].eoo()) { - pids.insert(lock[LocksType::process()].str()); - } - else { - warning() << "found incorrect lock document during lock ping cleanup: " - << lock.toString(); - } - } + string err = conn->getLastError(); + if (!err.empty()) { + warning() << "pinging failed for distributed lock pinger '" << pingId << "'." + << causedBy(err); + conn.done(); - // This can potentially delete ping entries that are actually active (if the clock - // of another pinger is too skewed). This is still fine as the lock logic only - // checks if there is a change in the ping document and the document going away - // is a valid change. - Date_t fourDays = pingTime - stdx::chrono::hours{4 * 24}; - conn->remove(LockpingsType::ConfigNS, - BSON(LockpingsType::process() << NIN << pids - << LockpingsType::ping() << LT << fourDays)); - err = conn->getLastError(); - - if (!err.empty()) { - warning() << "ping cleanup for distributed lock pinger '" << pingId - << " failed." << causedBy(err); - conn.done(); - - if (!shouldStopPinging(addr, process)) { - waitTillNextPingTime(sleepTime); - } - continue; + if (!shouldStopPinging(addr, process)) { + waitTillNextPingTime(sleepTime); } + continue; + } - LOG(1 - (loops % 10 == 0 ? 1 : 0)) << "cluster " << addr - << " pinged successfully at " << pingTime - << " by distributed lock pinger '" << pingId - << "', sleeping for " << sleepTime.count() << "ms"; - - // Remove old locks, if possible - // Make sure no one else is adding to this list at the same time - stdx::lock_guard<stdx::mutex> lk(_mutex); - - int numOldLocks = _unlockList.size(); - if (numOldLocks > 0) { - LOG(0) << "trying to delete " << _unlockList.size() - << " old lock entries for process " << process; + // Remove really old entries from the lockpings collection if they're not + // holding a lock. This may happen if an instance of a process was taken down + // and no new instance came up to replace it for a quite a while. + // NOTE this is NOT the same as the standard take-over mechanism, which forces + // the lock entry. + BSONObj fieldsToReturn = BSON(LocksType::state() << 1 << LocksType::process() << 1); + auto activeLocks = conn->query(LocksType::ConfigNS, + BSON(LocksType::state() << NE << LocksType::UNLOCKED)); + + uassert(16060, + str::stream() << "cannot query locks collection on config server " + << conn.getHost(), + activeLocks.get()); + + std::set<string> pids; + while (activeLocks->more()) { + BSONObj lock = activeLocks->nextSafe(); + + if (!lock[LocksType::process()].eoo()) { + pids.insert(lock[LocksType::process()].str()); + } else { + warning() << "found incorrect lock document during lock ping cleanup: " + << lock.toString(); } + } - bool removed = false; - for (auto iter = _unlockList.begin(); iter != _unlockList.end(); - iter = (removed ? _unlockList.erase(iter) : ++iter)) { - removed = false; - try { - // Got DistLockHandle from lock, so we don't need to specify _id again - conn->update(LocksType::ConfigNS, - BSON(LocksType::lockID(*iter)), - BSON("$set" << BSON( LocksType::state(LocksType::UNLOCKED)))); - - // Either the update went through or it didn't, - // either way we're done trying to unlock. - LOG(0) << "handled late remove of old distributed lock with ts " << *iter; - removed = true; - } - catch (UpdateNotTheSame&) { - LOG(0) << "partially removed old distributed lock with ts " << *iter; - removed = true; - } - catch (std::exception& e) { - warning() << "could not remove old distributed lock with ts " << *iter - << causedBy(e); - } + // This can potentially delete ping entries that are actually active (if the clock + // of another pinger is too skewed). This is still fine as the lock logic only + // checks if there is a change in the ping document and the document going away + // is a valid change. + Date_t fourDays = pingTime - stdx::chrono::hours{4 * 24}; + conn->remove(LockpingsType::ConfigNS, + BSON(LockpingsType::process() << NIN << pids << LockpingsType::ping() << LT + << fourDays)); + err = conn->getLastError(); + + if (!err.empty()) { + warning() << "ping cleanup for distributed lock pinger '" << pingId << " failed." + << causedBy(err); + conn.done(); + if (!shouldStopPinging(addr, process)) { + waitTillNextPingTime(sleepTime); } + continue; + } - if (numOldLocks > 0 && _unlockList.size() > 0) { - LOG(0) << "not all old lock entries could be removed for process " << process; - } + LOG(1 - (loops % 10 == 0 ? 1 : 0)) << "cluster " << addr << " pinged successfully at " + << pingTime << " by distributed lock pinger '" + << pingId << "', sleeping for " << sleepTime.count() + << "ms"; - conn.done(); + // Remove old locks, if possible + // Make sure no one else is adding to this list at the same time + stdx::lock_guard<stdx::mutex> lk(_mutex); + int numOldLocks = _unlockList.size(); + if (numOldLocks > 0) { + LOG(0) << "trying to delete " << _unlockList.size() + << " old lock entries for process " << process; } - catch (std::exception& e) { - warning() << "distributed lock pinger '" << pingId - << "' detected an exception while pinging." << causedBy(e); + + bool removed = false; + for (auto iter = _unlockList.begin(); iter != _unlockList.end(); + iter = (removed ? _unlockList.erase(iter) : ++iter)) { + removed = false; + try { + // Got DistLockHandle from lock, so we don't need to specify _id again + conn->update(LocksType::ConfigNS, + BSON(LocksType::lockID(*iter)), + BSON("$set" << BSON(LocksType::state(LocksType::UNLOCKED)))); + + // Either the update went through or it didn't, + // either way we're done trying to unlock. + LOG(0) << "handled late remove of old distributed lock with ts " << *iter; + removed = true; + } catch (UpdateNotTheSame&) { + LOG(0) << "partially removed old distributed lock with ts " << *iter; + removed = true; + } catch (std::exception& e) { + warning() << "could not remove old distributed lock with ts " << *iter + << causedBy(e); + } } - if (!shouldStopPinging(addr, process)) { - waitTillNextPingTime(sleepTime); + if (numOldLocks > 0 && _unlockList.size() > 0) { + LOG(0) << "not all old lock entries could be removed for process " << process; } - } - warning() << "removing distributed lock ping thread '" << pingId << "'"; + conn.done(); - if (shouldStopPinging(addr, process)) { - acknowledgeStopPing(addr, process); + } catch (std::exception& e) { + warning() << "distributed lock pinger '" << pingId + << "' detected an exception while pinging." << causedBy(e); } - } - void LegacyDistLockPinger::distLockPingThread(ConnectionString addr, - long long clockSkew, - const std::string& processId, - stdx::chrono::milliseconds sleepTime) { - try { - jsTimeVirtualThreadSkew(clockSkew); - _distLockPingThread(addr, processId, sleepTime); - } - catch (std::exception& e) { - error() << "unexpected error while running distributed lock pinger for " << addr - << ", process " << processId << causedBy(e); - } - catch (...) { - error() << "unknown error while running distributed lock pinger for " << addr - << ", process " << processId; + if (!shouldStopPinging(addr, process)) { + waitTillNextPingTime(sleepTime); } } - Status LegacyDistLockPinger::startPing(const DistributedLock& lock, - stdx::chrono::milliseconds sleepTime) { - const ConnectionString& conn = lock.getRemoteConnection(); - const string& processId = lock.getProcessId(); - string pingID = pingThreadId(conn, processId); + warning() << "removing distributed lock ping thread '" << pingId << "'"; - { - // Make sure we don't start multiple threads for a process id. - stdx::lock_guard<stdx::mutex> lk(_mutex); + if (shouldStopPinging(addr, process)) { + acknowledgeStopPing(addr, process); + } +} - if (_inShutdown) { - return Status(ErrorCodes::ShutdownInProgress, - "shutting down, will not start ping"); - } +void LegacyDistLockPinger::distLockPingThread(ConnectionString addr, + long long clockSkew, + const std::string& processId, + stdx::chrono::milliseconds sleepTime) { + try { + jsTimeVirtualThreadSkew(clockSkew); + _distLockPingThread(addr, processId, sleepTime); + } catch (std::exception& e) { + error() << "unexpected error while running distributed lock pinger for " << addr + << ", process " << processId << causedBy(e); + } catch (...) { + error() << "unknown error while running distributed lock pinger for " << addr + << ", process " << processId; + } +} - // Ignore if we already have a pinging thread for this process. - if (_seen.count(pingID) > 0) { - return Status::OK(); - } +Status LegacyDistLockPinger::startPing(const DistributedLock& lock, + stdx::chrono::milliseconds sleepTime) { + const ConnectionString& conn = lock.getRemoteConnection(); + const string& processId = lock.getProcessId(); + string pingID = pingThreadId(conn, processId); - // Check the config server clock skew. - if (lock.isRemoteTimeSkewed()) { - return Status(ErrorCodes::DistributedClockSkewed, - str::stream() << "clock skew of the cluster " << conn.toString() - << " is too far out of bounds " - << "to allow distributed locking."); - } + { + // Make sure we don't start multiple threads for a process id. + stdx::lock_guard<stdx::mutex> lk(_mutex); + + if (_inShutdown) { + return Status(ErrorCodes::ShutdownInProgress, "shutting down, will not start ping"); } - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - stdx::thread thread(stdx::bind(&LegacyDistLockPinger::distLockPingThread, - this, - conn, - getJSTimeVirtualThreadSkew(), - processId, - sleepTime)); - _pingThreads.insert(std::make_pair(pingID, std::move(thread))); - - _seen.insert(pingID); + // Ignore if we already have a pinging thread for this process. + if (_seen.count(pingID) > 0) { + return Status::OK(); } - return Status::OK(); + // Check the config server clock skew. + if (lock.isRemoteTimeSkewed()) { + return Status(ErrorCodes::DistributedClockSkewed, + str::stream() << "clock skew of the cluster " << conn.toString() + << " is too far out of bounds " + << "to allow distributed locking."); + } } - void LegacyDistLockPinger::addUnlockOID(const DistLockHandle& lockID) { - // Modifying the lock from some other thread + { stdx::lock_guard<stdx::mutex> lk(_mutex); - _unlockList.push_back(lockID); + stdx::thread thread(stdx::bind(&LegacyDistLockPinger::distLockPingThread, + this, + conn, + getJSTimeVirtualThreadSkew(), + processId, + sleepTime)); + _pingThreads.insert(std::make_pair(pingID, std::move(thread))); + + _seen.insert(pingID); } - bool LegacyDistLockPinger::willUnlockOID(const DistLockHandle& lockID) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - return find(_unlockList.begin(), _unlockList.end(), lockID) != _unlockList.end(); - } + return Status::OK(); +} - void LegacyDistLockPinger::stopPing(const ConnectionString& conn, const string& processId) { - { - stdx::lock_guard<stdx::mutex> lk(_mutex); +void LegacyDistLockPinger::addUnlockOID(const DistLockHandle& lockID) { + // Modifying the lock from some other thread + stdx::lock_guard<stdx::mutex> lk(_mutex); + _unlockList.push_back(lockID); +} - string pingId = pingThreadId(conn, processId); +bool LegacyDistLockPinger::willUnlockOID(const DistLockHandle& lockID) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + return find(_unlockList.begin(), _unlockList.end(), lockID) != _unlockList.end(); +} - verify(_seen.count(pingId) > 0); - _kill.insert(pingId); - _pingStoppedCV.notify_all(); - } - } +void LegacyDistLockPinger::stopPing(const ConnectionString& conn, const string& processId) { + { + stdx::lock_guard<stdx::mutex> lk(_mutex); - void LegacyDistLockPinger::shutdown() { - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _inShutdown = true; - _pingStoppedCV.notify_all(); - } + string pingId = pingThreadId(conn, processId); - // Don't grab _mutex, otherwise will deadlock trying to join. Safe to read - // _pingThreads since it cannot be modified once _shutdown is true. - for (auto& idToThread : _pingThreads) { - if (idToThread.second.joinable()) { - idToThread.second.join(); - } - } + verify(_seen.count(pingId) > 0); + _kill.insert(pingId); + _pingStoppedCV.notify_all(); } +} - bool LegacyDistLockPinger::shouldStopPinging(const ConnectionString& conn, - const string& processId) { - if (inShutdown()) { - return true; - } - +void LegacyDistLockPinger::shutdown() { + { stdx::lock_guard<stdx::mutex> lk(_mutex); + _inShutdown = true; + _pingStoppedCV.notify_all(); + } - if (_inShutdown) { - return true; + // Don't grab _mutex, otherwise will deadlock trying to join. Safe to read + // _pingThreads since it cannot be modified once _shutdown is true. + for (auto& idToThread : _pingThreads) { + if (idToThread.second.joinable()) { + idToThread.second.join(); } + } +} - return _kill.count(pingThreadId(conn, processId)) > 0; +bool LegacyDistLockPinger::shouldStopPinging(const ConnectionString& conn, + const string& processId) { + if (inShutdown()) { + return true; } - void LegacyDistLockPinger::acknowledgeStopPing(const ConnectionString& addr, - const string& processId) { - { - stdx::lock_guard<stdx::mutex> lk(_mutex); + stdx::lock_guard<stdx::mutex> lk(_mutex); - string pingId = pingThreadId(addr, processId); + if (_inShutdown) { + return true; + } - _kill.erase(pingId); - _seen.erase(pingId); - } + return _kill.count(pingThreadId(conn, processId)) > 0; +} - try { - ScopedDbConnection conn(addr.toString(), 30.0); - conn->remove(LockpingsType::ConfigNS, BSON(LockpingsType::process(processId))); - } - catch (const DBException& ex) { - warning() << "Error encountered while stopping ping on " << processId - << causedBy(ex); - } +void LegacyDistLockPinger::acknowledgeStopPing(const ConnectionString& addr, + const string& processId) { + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + + string pingId = pingThreadId(addr, processId); + + _kill.erase(pingId); + _seen.erase(pingId); } - void LegacyDistLockPinger::waitTillNextPingTime(stdx::chrono::milliseconds duration) { - stdx::unique_lock<stdx::mutex> lk(_mutex); - _pingStoppedCV.wait_for(lk, duration); + try { + ScopedDbConnection conn(addr.toString(), 30.0); + conn->remove(LockpingsType::ConfigNS, BSON(LockpingsType::process(processId))); + } catch (const DBException& ex) { + warning() << "Error encountered while stopping ping on " << processId << causedBy(ex); } } + +void LegacyDistLockPinger::waitTillNextPingTime(stdx::chrono::milliseconds duration) { + stdx::unique_lock<stdx::mutex> lk(_mutex); + _pingStoppedCV.wait_for(lk, duration); +} +} diff --git a/src/mongo/s/catalog/legacy/legacy_dist_lock_pinger.h b/src/mongo/s/catalog/legacy/legacy_dist_lock_pinger.h index 777cb39314b..41874a59d13 100644 --- a/src/mongo/s/catalog/legacy/legacy_dist_lock_pinger.h +++ b/src/mongo/s/catalog/legacy/legacy_dist_lock_pinger.h @@ -43,98 +43,97 @@ namespace mongo { - class DistributedLock; - - class LegacyDistLockPinger { - public: - LegacyDistLockPinger() = default; - - /** - * Starts pinging the process id for the given lock. - */ - Status startPing(const DistributedLock& lock, Milliseconds sleepTime); - - /** - * Adds a distributed lock that has the given id to the unlock list. The unlock list - * contains the list of locks that this pinger will repeatedly attempt to unlock until - * it succeeds. - */ - void addUnlockOID(const DistLockHandle& lockID); - - /** - * Returns true if the given lock id is currently in the unlock queue. - */ - bool willUnlockOID(const DistLockHandle& lockID); - - /** - * For testing only: non-blocking call to stop pinging the given process id. - */ - void stopPing(const ConnectionString& conn, const std::string& processId); - - /** - * Kills all ping threads and wait for them to cleanup. - */ - void shutdown(); - - private: - /** - * Helper method for calling _distLockPingThread. - */ - void distLockPingThread(ConnectionString addr, - long long clockSkew, - const std::string& processId, - Milliseconds sleepTime); - - /** - * Function for repeatedly pinging the process id. Also attempts to unlock all the - * locks in the unlock list. - */ - void _distLockPingThread(ConnectionString addr, - const std::string& process, - Milliseconds sleepTime); - - /** - * Returns true if a request has been made to stop pinging the give process id. - */ - bool shouldStopPinging(const ConnectionString& conn, const std::string& processId); - - /** - * Acknowledge the stop ping request and performs the necessary cleanup. - */ - void acknowledgeStopPing(const ConnectionString& conn, const std::string& processId); - - /** - * Blocks until duration has elapsed or if the ping thread is interrupted. - */ - void waitTillNextPingTime(Milliseconds duration); - - // - // All member variables are labeled with one of the following codes indicating the - // synchronization rules for accessing them. - // - // (M) Must hold _mutex for access. - - stdx::mutex _mutex; - - // Triggered everytime a pinger thread is stopped. - stdx::condition_variable _pingStoppedCV; // (M) - - // pingID -> thread - // This can contain multiple elements in tests, but in tne normal case, this will - // contain only a single element. - // Note: can be safely read when _inShutdown is true. - std::map<std::string, stdx::thread> _pingThreads; // (M*) - - // Contains the list of process id to stopPing. - std::set<std::string> _kill; // (M) - - // Contains all of the process id to ping. - std::set<std::string> _seen; // (M) - - // Contains all lock ids to keeping on retrying to unlock until success. - std::list<DistLockHandle> _unlockList; // (M) - - bool _inShutdown = false; // (M) - }; - +class DistributedLock; + +class LegacyDistLockPinger { +public: + LegacyDistLockPinger() = default; + + /** + * Starts pinging the process id for the given lock. + */ + Status startPing(const DistributedLock& lock, Milliseconds sleepTime); + + /** + * Adds a distributed lock that has the given id to the unlock list. The unlock list + * contains the list of locks that this pinger will repeatedly attempt to unlock until + * it succeeds. + */ + void addUnlockOID(const DistLockHandle& lockID); + + /** + * Returns true if the given lock id is currently in the unlock queue. + */ + bool willUnlockOID(const DistLockHandle& lockID); + + /** + * For testing only: non-blocking call to stop pinging the given process id. + */ + void stopPing(const ConnectionString& conn, const std::string& processId); + + /** + * Kills all ping threads and wait for them to cleanup. + */ + void shutdown(); + +private: + /** + * Helper method for calling _distLockPingThread. + */ + void distLockPingThread(ConnectionString addr, + long long clockSkew, + const std::string& processId, + Milliseconds sleepTime); + + /** + * Function for repeatedly pinging the process id. Also attempts to unlock all the + * locks in the unlock list. + */ + void _distLockPingThread(ConnectionString addr, + const std::string& process, + Milliseconds sleepTime); + + /** + * Returns true if a request has been made to stop pinging the give process id. + */ + bool shouldStopPinging(const ConnectionString& conn, const std::string& processId); + + /** + * Acknowledge the stop ping request and performs the necessary cleanup. + */ + void acknowledgeStopPing(const ConnectionString& conn, const std::string& processId); + + /** + * Blocks until duration has elapsed or if the ping thread is interrupted. + */ + void waitTillNextPingTime(Milliseconds duration); + + // + // All member variables are labeled with one of the following codes indicating the + // synchronization rules for accessing them. + // + // (M) Must hold _mutex for access. + + stdx::mutex _mutex; + + // Triggered everytime a pinger thread is stopped. + stdx::condition_variable _pingStoppedCV; // (M) + + // pingID -> thread + // This can contain multiple elements in tests, but in tne normal case, this will + // contain only a single element. + // Note: can be safely read when _inShutdown is true. + std::map<std::string, stdx::thread> _pingThreads; // (M*) + + // Contains the list of process id to stopPing. + std::set<std::string> _kill; // (M) + + // Contains all of the process id to ping. + std::set<std::string> _seen; // (M) + + // Contains all lock ids to keeping on retrying to unlock until success. + std::list<DistLockHandle> _unlockList; // (M) + + bool _inShutdown = false; // (M) +}; } diff --git a/src/mongo/s/catalog/replset/catalog_manager_replica_set.cpp b/src/mongo/s/catalog/replset/catalog_manager_replica_set.cpp index 9cc35e05bcd..15c7913feb6 100644 --- a/src/mongo/s/catalog/replset/catalog_manager_replica_set.cpp +++ b/src/mongo/s/catalog/replset/catalog_manager_replica_set.cpp @@ -63,465 +63,449 @@ namespace mongo { - using std::set; - using std::string; - using std::unique_ptr; - using std::vector; - using str::stream; +using std::set; +using std::string; +using std::unique_ptr; +using std::vector; +using str::stream; namespace { - const Status notYetImplemented(ErrorCodes::InternalError, "Not yet implemented"); // todo remove +const Status notYetImplemented(ErrorCodes::InternalError, "Not yet implemented"); // todo remove - // Until read committed is supported always write to the primary with majoirty write and read - // from the secondary. That way we ensure that reads will see a consistent data. - const ReadPreferenceSetting kConfigWriteSelector(ReadPreference::PrimaryOnly, TagSet{}); - const ReadPreferenceSetting kConfigReadSelector(ReadPreference::SecondaryOnly, TagSet{}); +// Until read committed is supported always write to the primary with majoirty write and read +// from the secondary. That way we ensure that reads will see a consistent data. +const ReadPreferenceSetting kConfigWriteSelector(ReadPreference::PrimaryOnly, TagSet{}); +const ReadPreferenceSetting kConfigReadSelector(ReadPreference::SecondaryOnly, TagSet{}); - const int kNotMasterNumRetries = 3; - const Milliseconds kNotMasterRetryInterval{500}; +const int kNotMasterNumRetries = 3; +const Milliseconds kNotMasterRetryInterval{500}; - void _toBatchError(const Status& status, BatchedCommandResponse* response) { - response->clear(); - response->setErrCode(status.code()); - response->setErrMessage(status.reason()); - response->setOk(false); - } - -} // namespace +void _toBatchError(const Status& status, BatchedCommandResponse* response) { + response->clear(); + response->setErrCode(status.code()); + response->setErrMessage(status.reason()); + response->setOk(false); +} - CatalogManagerReplicaSet::CatalogManagerReplicaSet() = default; +} // namespace - CatalogManagerReplicaSet::~CatalogManagerReplicaSet() = default; +CatalogManagerReplicaSet::CatalogManagerReplicaSet() = default; - Status CatalogManagerReplicaSet::init(const ConnectionString& configCS, - std::unique_ptr<DistLockManager> distLockManager) { +CatalogManagerReplicaSet::~CatalogManagerReplicaSet() = default; - invariant(configCS.type() == ConnectionString::SET); +Status CatalogManagerReplicaSet::init(const ConnectionString& configCS, + std::unique_ptr<DistLockManager> distLockManager) { + invariant(configCS.type() == ConnectionString::SET); - _configServerConnectionString = configCS; - _distLockManager = std::move(distLockManager); + _configServerConnectionString = configCS; + _distLockManager = std::move(distLockManager); - return Status::OK(); - } - - Status CatalogManagerReplicaSet::startup(bool upgrade) { - return Status::OK(); - } + return Status::OK(); +} - ConnectionString CatalogManagerReplicaSet::connectionString() const { - return _configServerConnectionString; - } +Status CatalogManagerReplicaSet::startup(bool upgrade) { + return Status::OK(); +} - void CatalogManagerReplicaSet::shutDown() { - LOG(1) << "CatalogManagerReplicaSet::shutDown() called."; - { - std::lock_guard<std::mutex> lk(_mutex); - _inShutdown = true; - } +ConnectionString CatalogManagerReplicaSet::connectionString() const { + return _configServerConnectionString; +} - invariant(_distLockManager); - _distLockManager->shutDown(); +void CatalogManagerReplicaSet::shutDown() { + LOG(1) << "CatalogManagerReplicaSet::shutDown() called."; + { + std::lock_guard<std::mutex> lk(_mutex); + _inShutdown = true; } - Status CatalogManagerReplicaSet::enableSharding(const std::string& dbName) { - return notYetImplemented; + invariant(_distLockManager); + _distLockManager->shutDown(); +} + +Status CatalogManagerReplicaSet::enableSharding(const std::string& dbName) { + return notYetImplemented; +} + +Status CatalogManagerReplicaSet::shardCollection(const string& ns, + const ShardKeyPattern& fieldsAndOrder, + bool unique, + vector<BSONObj>* initPoints, + set<ShardId>* initShardsIds) { + return notYetImplemented; +} + +Status CatalogManagerReplicaSet::createDatabase(const std::string& dbName) { + return notYetImplemented; +} + +StatusWith<string> CatalogManagerReplicaSet::addShard(const string& name, + const ConnectionString& shardConnectionString, + const long long maxSize) { + return notYetImplemented; +} + +StatusWith<ShardDrainingStatus> CatalogManagerReplicaSet::removeShard(OperationContext* txn, + const std::string& name) { + return notYetImplemented; +} + +Status CatalogManagerReplicaSet::updateDatabase(const std::string& dbName, const DatabaseType& db) { + fassert(28684, db.validate()); + + return notYetImplemented; +} + +StatusWith<DatabaseType> CatalogManagerReplicaSet::getDatabase(const std::string& dbName) { + invariant(nsIsDbOnly(dbName)); + + // The two databases that are hosted on the config server are config and admin + if (dbName == "config" || dbName == "admin") { + DatabaseType dbt; + dbt.setName(dbName); + dbt.setSharded(false); + dbt.setPrimary("config"); + + return dbt; } - Status CatalogManagerReplicaSet::shardCollection(const string& ns, - const ShardKeyPattern& fieldsAndOrder, - bool unique, - vector<BSONObj>* initPoints, - set<ShardId>* initShardsIds) { - return notYetImplemented; + const auto configShard = grid.shardRegistry()->getShard("config"); + const auto readHost = configShard->getTargeter()->findHost(kConfigReadSelector); + if (!readHost.isOK()) { + return readHost.getStatus(); } - Status CatalogManagerReplicaSet::createDatabase(const std::string& dbName) { - return notYetImplemented; - } + auto findStatus = grid.shardRegistry()->exhaustiveFind(readHost.getValue(), + NamespaceString(DatabaseType::ConfigNS), + BSON(DatabaseType::name(dbName)), + 1); - StatusWith<string> CatalogManagerReplicaSet::addShard( - const string& name, - const ConnectionString& shardConnectionString, - const long long maxSize) { - return notYetImplemented; + if (!findStatus.isOK()) { + return findStatus.getStatus(); } - StatusWith<ShardDrainingStatus> CatalogManagerReplicaSet::removeShard(OperationContext* txn, - const std::string& name) { - return notYetImplemented; + const auto& docs = findStatus.getValue(); + if (docs.empty()) { + return {ErrorCodes::NamespaceNotFound, stream() << "database " << dbName << " not found"}; } - Status CatalogManagerReplicaSet::updateDatabase(const std::string& dbName, - const DatabaseType& db) { - fassert(28684, db.validate()); - - return notYetImplemented; + invariant(docs.size() == 1); + + return DatabaseType::fromBSON(docs.front()); +} + +Status CatalogManagerReplicaSet::updateCollection(const std::string& collNs, + const CollectionType& coll) { + fassert(28683, coll.validate()); + + BatchedCommandResponse response; + Status status = update(CollectionType::ConfigNS, + BSON(CollectionType::fullNs(collNs)), + coll.toBSON(), + true, // upsert + false, // multi + &response); + if (!status.isOK()) { + return Status(status.code(), + str::stream() << "collection metadata write failed: " << response.toBSON() + << "; status: " << status.toString()); } - StatusWith<DatabaseType> CatalogManagerReplicaSet::getDatabase(const std::string& dbName) { - invariant(nsIsDbOnly(dbName)); - - // The two databases that are hosted on the config server are config and admin - if (dbName == "config" || dbName == "admin") { - DatabaseType dbt; - dbt.setName(dbName); - dbt.setSharded(false); - dbt.setPrimary("config"); + return Status::OK(); +} - return dbt; - } - - const auto configShard = grid.shardRegistry()->getShard("config"); - const auto readHost = configShard->getTargeter()->findHost(kConfigReadSelector); - if (!readHost.isOK()) { - return readHost.getStatus(); - } - - auto findStatus = grid.shardRegistry()->exhaustiveFind( - readHost.getValue(), - NamespaceString(DatabaseType::ConfigNS), - BSON(DatabaseType::name(dbName)), - 1); - - if (!findStatus.isOK()) { - return findStatus.getStatus(); - } +StatusWith<CollectionType> CatalogManagerReplicaSet::getCollection(const std::string& collNs) { + auto configShard = grid.shardRegistry()->getShard("config"); - const auto& docs = findStatus.getValue(); - if (docs.empty()) { - return {ErrorCodes::NamespaceNotFound, - stream() << "database " << dbName << " not found"}; - } + auto readHostStatus = configShard->getTargeter()->findHost(kConfigReadSelector); + if (!readHostStatus.isOK()) { + return readHostStatus.getStatus(); + } - invariant(docs.size() == 1); + auto statusFind = + grid.shardRegistry()->exhaustiveFind(readHostStatus.getValue(), + NamespaceString(CollectionType::ConfigNS), + BSON(CollectionType::fullNs(collNs)), + 1); - return DatabaseType::fromBSON(docs.front()); + if (!statusFind.isOK()) { + return statusFind.getStatus(); } - Status CatalogManagerReplicaSet::updateCollection(const std::string& collNs, - const CollectionType& coll) { - fassert(28683, coll.validate()); - - BatchedCommandResponse response; - Status status = update(CollectionType::ConfigNS, - BSON(CollectionType::fullNs(collNs)), - coll.toBSON(), - true, // upsert - false, // multi - &response); - if (!status.isOK()) { - return Status(status.code(), - str::stream() << "collection metadata write failed: " - << response.toBSON() << "; status: " << status.toString()); - } - - return Status::OK(); + const auto& retVal = statusFind.getValue(); + if (retVal.empty()) { + return Status(ErrorCodes::NamespaceNotFound, + stream() << "collection " << collNs << " not found"); } - StatusWith<CollectionType> CatalogManagerReplicaSet::getCollection(const std::string& collNs) { - auto configShard = grid.shardRegistry()->getShard("config"); + invariant(retVal.size() == 1); - auto readHostStatus = configShard->getTargeter()->findHost(kConfigReadSelector); - if (!readHostStatus.isOK()) { - return readHostStatus.getStatus(); - } + return CollectionType::fromBSON(retVal.front()); +} - auto statusFind = grid.shardRegistry()->exhaustiveFind( - readHostStatus.getValue(), - NamespaceString(CollectionType::ConfigNS), - BSON(CollectionType::fullNs(collNs)), - 1); +Status CatalogManagerReplicaSet::getCollections(const std::string* dbName, + std::vector<CollectionType>* collections) { + return notYetImplemented; +} - if (!statusFind.isOK()) { - return statusFind.getStatus(); - } +Status CatalogManagerReplicaSet::dropCollection(const std::string& collectionNs) { + return notYetImplemented; +} - const auto& retVal = statusFind.getValue(); - if (retVal.empty()) { - return Status(ErrorCodes::NamespaceNotFound, - stream() << "collection " << collNs << " not found"); - } +void CatalogManagerReplicaSet::logAction(const ActionLogType& actionLog) {} - invariant(retVal.size() == 1); +void CatalogManagerReplicaSet::logChange(OperationContext* opCtx, + const string& what, + const string& ns, + const BSONObj& detail) {} - return CollectionType::fromBSON(retVal.front()); +StatusWith<SettingsType> CatalogManagerReplicaSet::getGlobalSettings(const string& key) { + const auto configShard = grid.shardRegistry()->getShard("config"); + const auto readHost = configShard->getTargeter()->findHost(kConfigReadSelector); + if (!readHost.isOK()) { + return readHost.getStatus(); } - Status CatalogManagerReplicaSet::getCollections(const std::string* dbName, - std::vector<CollectionType>* collections) { - return notYetImplemented; - } + auto findStatus = grid.shardRegistry()->exhaustiveFind(readHost.getValue(), + NamespaceString(SettingsType::ConfigNS), + BSON(SettingsType::key(key)), + 1); - Status CatalogManagerReplicaSet::dropCollection(const std::string& collectionNs) { - return notYetImplemented; + if (!findStatus.isOK()) { + return findStatus.getStatus(); } - void CatalogManagerReplicaSet::logAction(const ActionLogType& actionLog) { - + const auto& docs = findStatus.getValue(); + if (docs.empty()) { + return {ErrorCodes::NoMatchingDocument, + str::stream() << "can't find settings document with key: " << key}; } - void CatalogManagerReplicaSet::logChange(OperationContext* opCtx, - const string& what, - const string& ns, - const BSONObj& detail) { + BSONObj settingsDoc = docs.front(); + StatusWith<SettingsType> settingsResult = SettingsType::fromBSON(settingsDoc); + if (!settingsResult.isOK()) { + return {ErrorCodes::FailedToParse, + str::stream() << "error while parsing settings document: " << settingsDoc << " : " + << settingsResult.getStatus().toString()}; } - StatusWith<SettingsType> CatalogManagerReplicaSet::getGlobalSettings(const string& key) { - const auto configShard = grid.shardRegistry()->getShard("config"); - const auto readHost = configShard->getTargeter()->findHost(kConfigReadSelector); - if (!readHost.isOK()) { - return readHost.getStatus(); - } - - auto findStatus = grid.shardRegistry()->exhaustiveFind( - readHost.getValue(), - NamespaceString(SettingsType::ConfigNS), - BSON(SettingsType::key(key)), - 1); - - if (!findStatus.isOK()) { - return findStatus.getStatus(); - } - - const auto& docs = findStatus.getValue(); - if (docs.empty()) { - return {ErrorCodes::NoMatchingDocument, - str::stream() << "can't find settings document with key: " << key}; - } - - BSONObj settingsDoc = docs.front(); - StatusWith<SettingsType> settingsResult = SettingsType::fromBSON(settingsDoc); - if (!settingsResult.isOK()) { - return {ErrorCodes::FailedToParse, - str::stream() << "error while parsing settings document: " << settingsDoc - << " : " << settingsResult.getStatus().toString()}; - } - - const SettingsType& settings = settingsResult.getValue(); + const SettingsType& settings = settingsResult.getValue(); - Status validationStatus = settings.validate(); - if (!validationStatus.isOK()) { - return validationStatus; - } - - return settingsResult; + Status validationStatus = settings.validate(); + if (!validationStatus.isOK()) { + return validationStatus; } - Status CatalogManagerReplicaSet::getDatabasesForShard(const string& shardName, - vector<string>* dbs) { - return notYetImplemented; + return settingsResult; +} + +Status CatalogManagerReplicaSet::getDatabasesForShard(const string& shardName, + vector<string>* dbs) { + return notYetImplemented; +} + +Status CatalogManagerReplicaSet::getChunks(const Query& query, + int nToReturn, + vector<ChunkType>* chunks) { + auto configShard = grid.shardRegistry()->getShard("config"); + auto readHostStatus = configShard->getTargeter()->findHost(kConfigReadSelector); + if (!readHostStatus.isOK()) { + return readHostStatus.getStatus(); } - Status CatalogManagerReplicaSet::getChunks(const Query& query, - int nToReturn, - vector<ChunkType>* chunks) { + auto findStatus = grid.shardRegistry()->exhaustiveFind(readHostStatus.getValue(), + NamespaceString(ChunkType::ConfigNS), + query.obj, + boost::none); // no limit + if (!findStatus.isOK()) { + return findStatus.getStatus(); + } - auto configShard = grid.shardRegistry()->getShard("config"); - auto readHostStatus = configShard->getTargeter()->findHost(kConfigReadSelector); - if (!readHostStatus.isOK()) { - return readHostStatus.getStatus(); + for (const BSONObj& obj : findStatus.getValue()) { + auto chunkRes = ChunkType::fromBSON(obj); + if (!chunkRes.isOK()) { + chunks->clear(); + return {ErrorCodes::FailedToParse, + stream() << "Failed to parse chunk with id (" + << obj[ChunkType::name()].toString() + << "): " << chunkRes.getStatus().reason()}; } - auto findStatus = grid.shardRegistry()->exhaustiveFind(readHostStatus.getValue(), - NamespaceString(ChunkType::ConfigNS), - query.obj, - boost::none); // no limit - if (!findStatus.isOK()) { - return findStatus.getStatus(); - } + chunks->push_back(chunkRes.getValue()); + } - for (const BSONObj& obj : findStatus.getValue()) { - auto chunkRes = ChunkType::fromBSON(obj); - if (!chunkRes.isOK()) { - chunks->clear(); - return {ErrorCodes::FailedToParse, - stream() << "Failed to parse chunk with id (" - << obj[ChunkType::name()].toString() << "): " - << chunkRes.getStatus().reason()}; - } + return Status::OK(); +} - chunks->push_back(chunkRes.getValue()); - } +Status CatalogManagerReplicaSet::getTagsForCollection(const std::string& collectionNs, + std::vector<TagsType>* tags) { + return notYetImplemented; +} - return Status::OK(); - } +StatusWith<string> CatalogManagerReplicaSet::getTagForChunk(const std::string& collectionNs, + const ChunkType& chunk) { + return notYetImplemented; +} - Status CatalogManagerReplicaSet::getTagsForCollection(const std::string& collectionNs, - std::vector<TagsType>* tags) { - return notYetImplemented; +Status CatalogManagerReplicaSet::getAllShards(vector<ShardType>* shards) { + const auto configShard = grid.shardRegistry()->getShard("config"); + const auto readHost = configShard->getTargeter()->findHost(kConfigReadSelector); + if (!readHost.isOK()) { + return readHost.getStatus(); } - StatusWith<string> CatalogManagerReplicaSet::getTagForChunk(const std::string& collectionNs, - const ChunkType& chunk) { - return notYetImplemented; + auto findStatus = grid.shardRegistry()->exhaustiveFind(readHost.getValue(), + NamespaceString(ShardType::ConfigNS), + BSONObj(), // no query filter + boost::none); // no limit + if (!findStatus.isOK()) { + return findStatus.getStatus(); } - Status CatalogManagerReplicaSet::getAllShards(vector<ShardType>* shards) { - const auto configShard = grid.shardRegistry()->getShard("config"); - const auto readHost = configShard->getTargeter()->findHost(kConfigReadSelector); - if (!readHost.isOK()) { - return readHost.getStatus(); - } - - auto findStatus = grid.shardRegistry()->exhaustiveFind(readHost.getValue(), - NamespaceString(ShardType::ConfigNS), - BSONObj(), // no query filter - boost::none); // no limit - if (!findStatus.isOK()) { - return findStatus.getStatus(); - } - - for (const BSONObj& doc : findStatus.getValue()) { - auto shardRes = ShardType::fromBSON(doc); - if (!shardRes.isOK()) { - shards->clear(); - return {ErrorCodes::FailedToParse, - stream() << "Failed to parse shard with id (" - << doc[ShardType::name()].toString() << "): " - << shardRes.getStatus().reason()}; - } - - shards->push_back(shardRes.getValue()); + for (const BSONObj& doc : findStatus.getValue()) { + auto shardRes = ShardType::fromBSON(doc); + if (!shardRes.isOK()) { + shards->clear(); + return {ErrorCodes::FailedToParse, + stream() << "Failed to parse shard with id (" + << doc[ShardType::name()].toString() + << "): " << shardRes.getStatus().reason()}; } - return Status::OK(); - } - - bool CatalogManagerReplicaSet::isShardHost(const ConnectionString& connectionString) { - return false; + shards->push_back(shardRes.getValue()); } - bool CatalogManagerReplicaSet::runUserManagementWriteCommand(const std::string& commandName, - const std::string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) { - auto scopedDistLock = getDistLockManager()->lock("authorizationData", - commandName, - Seconds{5}); - if (!scopedDistLock.isOK()) { - return Command::appendCommandStatus(*result, scopedDistLock.getStatus()); - } - - auto targeter = grid.shardRegistry()->getShard("config")->getTargeter(); + return Status::OK(); +} - Status notMasterStatus{ErrorCodes::InternalError, "status not set"}; - for (int i = 0; i < kNotMasterNumRetries; ++i) { +bool CatalogManagerReplicaSet::isShardHost(const ConnectionString& connectionString) { + return false; +} - auto target = targeter->findHost(kConfigWriteSelector); - if (!target.isOK()) { - if (ErrorCodes::NotMaster == target.getStatus()) { - notMasterStatus = target.getStatus(); - sleepmillis(kNotMasterRetryInterval.count()); - continue; - } - return Command::appendCommandStatus(*result, target.getStatus()); - } +bool CatalogManagerReplicaSet::runUserManagementWriteCommand(const std::string& commandName, + const std::string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) { + auto scopedDistLock = getDistLockManager()->lock("authorizationData", commandName, Seconds{5}); + if (!scopedDistLock.isOK()) { + return Command::appendCommandStatus(*result, scopedDistLock.getStatus()); + } - auto response = grid.shardRegistry()->runCommand(target.getValue(), dbname, cmdObj); - if (!response.isOK()) { - return Command::appendCommandStatus(*result, response.getStatus()); - } + auto targeter = grid.shardRegistry()->getShard("config")->getTargeter(); - Status commandStatus = Command::getStatusFromCommandResult(response.getValue()); - if (ErrorCodes::NotMaster == commandStatus) { - notMasterStatus = commandStatus; + Status notMasterStatus{ErrorCodes::InternalError, "status not set"}; + for (int i = 0; i < kNotMasterNumRetries; ++i) { + auto target = targeter->findHost(kConfigWriteSelector); + if (!target.isOK()) { + if (ErrorCodes::NotMaster == target.getStatus()) { + notMasterStatus = target.getStatus(); sleepmillis(kNotMasterRetryInterval.count()); continue; } - - result->appendElements(response.getValue()); - - return commandStatus.isOK(); + return Command::appendCommandStatus(*result, target.getStatus()); } - invariant(ErrorCodes::NotMaster == notMasterStatus); - return Command::appendCommandStatus(*result, notMasterStatus); - } - - bool CatalogManagerReplicaSet::runUserManagementReadCommand(const std::string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) { - auto targeter = grid.shardRegistry()->getShard("config")->getTargeter(); - auto target = targeter->findHost(kConfigReadSelector); - if (!target.isOK()) { - return Command::appendCommandStatus(*result, target.getStatus()); + auto response = grid.shardRegistry()->runCommand(target.getValue(), dbname, cmdObj); + if (!response.isOK()) { + return Command::appendCommandStatus(*result, response.getStatus()); } - auto resultStatus = grid.shardRegistry()->runCommand(target.getValue(), dbname, cmdObj); - if (!resultStatus.isOK()) { - return Command::appendCommandStatus(*result, resultStatus.getStatus()); + Status commandStatus = Command::getStatusFromCommandResult(response.getValue()); + if (ErrorCodes::NotMaster == commandStatus) { + notMasterStatus = commandStatus; + sleepmillis(kNotMasterRetryInterval.count()); + continue; } - result->appendElements(resultStatus.getValue()); + result->appendElements(response.getValue()); - return Command::getStatusFromCommandResult(resultStatus.getValue()).isOK(); - return false; + return commandStatus.isOK(); } - Status CatalogManagerReplicaSet::applyChunkOpsDeprecated(const BSONArray& updateOps, - const BSONArray& preCondition) { - return notYetImplemented; + invariant(ErrorCodes::NotMaster == notMasterStatus); + return Command::appendCommandStatus(*result, notMasterStatus); +} + +bool CatalogManagerReplicaSet::runUserManagementReadCommand(const std::string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) { + auto targeter = grid.shardRegistry()->getShard("config")->getTargeter(); + auto target = targeter->findHost(kConfigReadSelector); + if (!target.isOK()) { + return Command::appendCommandStatus(*result, target.getStatus()); } - DistLockManager* CatalogManagerReplicaSet::getDistLockManager() { - invariant(_distLockManager); - return _distLockManager.get(); + auto resultStatus = grid.shardRegistry()->runCommand(target.getValue(), dbname, cmdObj); + if (!resultStatus.isOK()) { + return Command::appendCommandStatus(*result, resultStatus.getStatus()); } - void CatalogManagerReplicaSet::writeConfigServerDirect( - const BatchedCommandRequest& batchRequest, - BatchedCommandResponse* batchResponse) { - std::string dbname = batchRequest.getNSS().db().toString(); - invariant (dbname == "config" || dbname == "admin"); - const BSONObj cmdObj = batchRequest.toBSON(); - auto targeter = grid.shardRegistry()->getShard("config")->getTargeter(); - - Status notMasterStatus{ErrorCodes::InternalError, "status not set"}; - for (int i = 0; i < kNotMasterNumRetries; ++i) { - - auto target = targeter->findHost(kConfigWriteSelector); - if (!target.isOK()) { - if (ErrorCodes::NotMaster == target.getStatus()) { - notMasterStatus = target.getStatus(); - sleepmillis(kNotMasterRetryInterval.count()); - continue; - } - _toBatchError(target.getStatus(), batchResponse); - return; - } + result->appendElements(resultStatus.getValue()); - auto resultStatus = grid.shardRegistry()->runCommand(target.getValue(), - batchRequest.getNSS().db().toString(), - batchRequest.toBSON()); + return Command::getStatusFromCommandResult(resultStatus.getValue()).isOK(); + return false; +} - if (!resultStatus.isOK()) { - _toBatchError(resultStatus.getStatus(), batchResponse); - return; - } +Status CatalogManagerReplicaSet::applyChunkOpsDeprecated(const BSONArray& updateOps, + const BSONArray& preCondition) { + return notYetImplemented; +} + +DistLockManager* CatalogManagerReplicaSet::getDistLockManager() { + invariant(_distLockManager); + return _distLockManager.get(); +} - const BSONObj& commandResponse = resultStatus.getValue(); +void CatalogManagerReplicaSet::writeConfigServerDirect(const BatchedCommandRequest& batchRequest, + BatchedCommandResponse* batchResponse) { + std::string dbname = batchRequest.getNSS().db().toString(); + invariant(dbname == "config" || dbname == "admin"); + const BSONObj cmdObj = batchRequest.toBSON(); + auto targeter = grid.shardRegistry()->getShard("config")->getTargeter(); - Status commandStatus = getStatusFromCommandResult(commandResponse); - if (commandStatus == ErrorCodes::NotMaster) { - notMasterStatus = commandStatus; + Status notMasterStatus{ErrorCodes::InternalError, "status not set"}; + for (int i = 0; i < kNotMasterNumRetries; ++i) { + auto target = targeter->findHost(kConfigWriteSelector); + if (!target.isOK()) { + if (ErrorCodes::NotMaster == target.getStatus()) { + notMasterStatus = target.getStatus(); sleepmillis(kNotMasterRetryInterval.count()); continue; } + _toBatchError(target.getStatus(), batchResponse); + return; + } - string errmsg; - if (!batchResponse->parseBSON(commandResponse, &errmsg)) { - _toBatchError(Status(ErrorCodes::FailedToParse, - str::stream() << "Failed to parse config server response: " << - errmsg), - batchResponse); - return; - } - return; // The normal case return point. + auto resultStatus = grid.shardRegistry()->runCommand( + target.getValue(), batchRequest.getNSS().db().toString(), batchRequest.toBSON()); + + if (!resultStatus.isOK()) { + _toBatchError(resultStatus.getStatus(), batchResponse); + return; } - invariant(ErrorCodes::NotMaster == notMasterStatus); - _toBatchError(notMasterStatus, batchResponse); + const BSONObj& commandResponse = resultStatus.getValue(); + + Status commandStatus = getStatusFromCommandResult(commandResponse); + if (commandStatus == ErrorCodes::NotMaster) { + notMasterStatus = commandStatus; + sleepmillis(kNotMasterRetryInterval.count()); + continue; + } + + string errmsg; + if (!batchResponse->parseBSON(commandResponse, &errmsg)) { + _toBatchError( + Status(ErrorCodes::FailedToParse, + str::stream() << "Failed to parse config server response: " << errmsg), + batchResponse); + return; + } + return; // The normal case return point. } -} // namespace mongo + invariant(ErrorCodes::NotMaster == notMasterStatus); + _toBatchError(notMasterStatus, batchResponse); +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/replset/catalog_manager_replica_set.h b/src/mongo/s/catalog/replset/catalog_manager_replica_set.h index 9b85640d1fb..c4b93093c26 100644 --- a/src/mongo/s/catalog/replset/catalog_manager_replica_set.h +++ b/src/mongo/s/catalog/replset/catalog_manager_replica_set.h @@ -39,115 +39,111 @@ namespace mongo { +/** + * Implements the catalog manager for talking to replica set config servers. + */ +class CatalogManagerReplicaSet final : public CatalogManager { +public: + CatalogManagerReplicaSet(); + virtual ~CatalogManagerReplicaSet(); + /** - * Implements the catalog manager for talking to replica set config servers. + * Initializes the catalog manager. + * Can only be called once for the lifetime of the catalog manager. + * TODO(spencer): Take pointer to ShardRegistry rather than getting it from the global + * "grid" object. */ - class CatalogManagerReplicaSet final : public CatalogManager { - public: - CatalogManagerReplicaSet(); - virtual ~CatalogManagerReplicaSet(); - - /** - * Initializes the catalog manager. - * Can only be called once for the lifetime of the catalog manager. - * TODO(spencer): Take pointer to ShardRegistry rather than getting it from the global - * "grid" object. - */ - Status init(const ConnectionString& configCS, - std::unique_ptr<DistLockManager> distLockManager); - - Status startup(bool upgrade) override; + Status init(const ConnectionString& configCS, std::unique_ptr<DistLockManager> distLockManager); - ConnectionString connectionString() const override; + Status startup(bool upgrade) override; - void shutDown() override; + ConnectionString connectionString() const override; - Status enableSharding(const std::string& dbName) override; + void shutDown() override; - Status shardCollection(const std::string& ns, - const ShardKeyPattern& fieldsAndOrder, - bool unique, - std::vector<BSONObj>* initPoints, - std::set<ShardId>* initShardsIds = nullptr) override; + Status enableSharding(const std::string& dbName) override; - StatusWith<std::string> addShard(const std::string& name, - const ConnectionString& shardConnectionString, - const long long maxSize) override; + Status shardCollection(const std::string& ns, + const ShardKeyPattern& fieldsAndOrder, + bool unique, + std::vector<BSONObj>* initPoints, + std::set<ShardId>* initShardsIds = nullptr) override; - StatusWith<ShardDrainingStatus> removeShard(OperationContext* txn, - const std::string& name) override; + StatusWith<std::string> addShard(const std::string& name, + const ConnectionString& shardConnectionString, + const long long maxSize) override; - Status createDatabase(const std::string& dbName) override; + StatusWith<ShardDrainingStatus> removeShard(OperationContext* txn, + const std::string& name) override; - Status updateDatabase(const std::string& dbName, const DatabaseType& db) override; + Status createDatabase(const std::string& dbName) override; - StatusWith<DatabaseType> getDatabase(const std::string& dbName) override; + Status updateDatabase(const std::string& dbName, const DatabaseType& db) override; - Status updateCollection(const std::string& collNs, const CollectionType& coll) override; + StatusWith<DatabaseType> getDatabase(const std::string& dbName) override; - StatusWith<CollectionType> getCollection(const std::string& collNs) override; + Status updateCollection(const std::string& collNs, const CollectionType& coll) override; - Status getCollections(const std::string* dbName, - std::vector<CollectionType>* collections) override; + StatusWith<CollectionType> getCollection(const std::string& collNs) override; - Status dropCollection(const std::string& collectionNs) override; + Status getCollections(const std::string* dbName, + std::vector<CollectionType>* collections) override; - Status getDatabasesForShard(const std::string& shardName, - std::vector<std::string>* dbs) override; + Status dropCollection(const std::string& collectionNs) override; - Status getChunks(const Query& query, - int nToReturn, - std::vector<ChunkType>* chunks) override; + Status getDatabasesForShard(const std::string& shardName, + std::vector<std::string>* dbs) override; - Status getTagsForCollection(const std::string& collectionNs, - std::vector<TagsType>* tags) override; + Status getChunks(const Query& query, int nToReturn, std::vector<ChunkType>* chunks) override; - StatusWith<std::string> getTagForChunk(const std::string& collectionNs, - const ChunkType& chunk) override; + Status getTagsForCollection(const std::string& collectionNs, + std::vector<TagsType>* tags) override; - Status getAllShards(std::vector<ShardType>* shards) override; + StatusWith<std::string> getTagForChunk(const std::string& collectionNs, + const ChunkType& chunk) override; - bool isShardHost(const ConnectionString& shardConnectionString) override; + Status getAllShards(std::vector<ShardType>* shards) override; - bool runUserManagementWriteCommand(const std::string& commandName, - const std::string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) override; + bool isShardHost(const ConnectionString& shardConnectionString) override; - bool runUserManagementReadCommand(const std::string& dbname, - const BSONObj& cmdObj, - BSONObjBuilder* result) override; + bool runUserManagementWriteCommand(const std::string& commandName, + const std::string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) override; - Status applyChunkOpsDeprecated(const BSONArray& updateOps, - const BSONArray& preCondition) override; + bool runUserManagementReadCommand(const std::string& dbname, + const BSONObj& cmdObj, + BSONObjBuilder* result) override; - void logAction(const ActionLogType& actionLog) override; + Status applyChunkOpsDeprecated(const BSONArray& updateOps, + const BSONArray& preCondition) override; - void logChange(OperationContext* txn, - const std::string& what, - const std::string& ns, - const BSONObj& detail) override; + void logAction(const ActionLogType& actionLog) override; - StatusWith<SettingsType> getGlobalSettings(const std::string& key) override; + void logChange(OperationContext* txn, + const std::string& what, + const std::string& ns, + const BSONObj& detail) override; - void writeConfigServerDirect(const BatchedCommandRequest& request, - BatchedCommandResponse* response) override; + StatusWith<SettingsType> getGlobalSettings(const std::string& key) override; - DistLockManager* getDistLockManager() override; + void writeConfigServerDirect(const BatchedCommandRequest& request, + BatchedCommandResponse* response) override; - private: + DistLockManager* getDistLockManager() override; - // Config server connection string - ConnectionString _configServerConnectionString; +private: + // Config server connection string + ConnectionString _configServerConnectionString; - // Distribted lock manager singleton. - std::unique_ptr<DistLockManager> _distLockManager; + // Distribted lock manager singleton. + std::unique_ptr<DistLockManager> _distLockManager; - // protects _inShutdown - std::mutex _mutex; + // protects _inShutdown + std::mutex _mutex; - // True if shutDown() has been called. False, otherwise. - bool _inShutdown = false; - }; + // True if shutDown() has been called. False, otherwise. + bool _inShutdown = false; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/replset/catalog_manager_replica_set_test.cpp b/src/mongo/s/catalog/replset/catalog_manager_replica_set_test.cpp index 7ba386e12f0..2d551a855f5 100644 --- a/src/mongo/s/catalog/replset/catalog_manager_replica_set_test.cpp +++ b/src/mongo/s/catalog/replset/catalog_manager_replica_set_test.cpp @@ -56,357 +56,473 @@ namespace mongo { namespace { - using executor::NetworkInterfaceMock; - using executor::TaskExecutor; - using std::async; - using std::string; - using std::vector; - using stdx::chrono::milliseconds; - using unittest::assertGet; - - static const std::chrono::seconds kFutureTimeout{5}; - - TEST_F(CatalogManagerReplSetTestFixture, GetCollectionExisting) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - CollectionType expectedColl; - expectedColl.setNs(NamespaceString("TestDB.TestNS")); - expectedColl.setKeyPattern(BSON("KeyName" << 1)); - expectedColl.setUpdatedAt(Date_t()); - expectedColl.setEpoch(OID::gen()); - - auto future = async(std::launch::async, [this, &expectedColl] { - return assertGet(catalogManager()->getCollection(expectedColl.getNs())); - }); - - onFindCommand([&expectedColl](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), CollectionType::ConfigNS); - - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - - // Ensure the query is correct - ASSERT_EQ(query->ns(), CollectionType::ConfigNS); - ASSERT_EQ(query->getFilter(), BSON(CollectionType::fullNs(expectedColl.getNs()))); - - return vector<BSONObj>{ expectedColl.toBSON() }; - }); - - // Now wait for the getCollection call to return - const auto& actualColl = future.get(); - ASSERT_EQ(expectedColl.toBSON(), actualColl.toBSON()); - } - - TEST_F(CatalogManagerReplSetTestFixture, GetCollectionNotExisting) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - auto future = async(std::launch::async, [this] { - auto status = catalogManager()->getCollection("NonExistent"); - ASSERT_EQUALS(status.getStatus(), ErrorCodes::NamespaceNotFound); - }); - - onFindCommand([](const RemoteCommandRequest& request) { - return vector<BSONObj>{ }; - }); - - // Now wait for the getCollection call to return - future.get(); - } - - TEST_F(CatalogManagerReplSetTestFixture, GetDatabaseExisting) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - DatabaseType expectedDb; - expectedDb.setName("bigdata"); - expectedDb.setPrimary("shard0000"); - expectedDb.setSharded(true); - - auto future = async(std::launch::async, [this, &expectedDb] { - return assertGet(catalogManager()->getDatabase(expectedDb.getName())); - }); - - onFindCommand([&expectedDb](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), DatabaseType::ConfigNS); - - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - - ASSERT_EQ(query->ns(), DatabaseType::ConfigNS); - ASSERT_EQ(query->getFilter(), BSON(DatabaseType::name(expectedDb.getName()))); +using executor::NetworkInterfaceMock; +using executor::TaskExecutor; +using std::async; +using std::string; +using std::vector; +using stdx::chrono::milliseconds; +using unittest::assertGet; + +static const std::chrono::seconds kFutureTimeout{5}; + +TEST_F(CatalogManagerReplSetTestFixture, GetCollectionExisting) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + CollectionType expectedColl; + expectedColl.setNs(NamespaceString("TestDB.TestNS")); + expectedColl.setKeyPattern(BSON("KeyName" << 1)); + expectedColl.setUpdatedAt(Date_t()); + expectedColl.setEpoch(OID::gen()); + + auto future = async(std::launch::async, + [this, &expectedColl] { + return assertGet(catalogManager()->getCollection(expectedColl.getNs())); + }); + + onFindCommand([&expectedColl](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), CollectionType::ConfigNS); + + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + + // Ensure the query is correct + ASSERT_EQ(query->ns(), CollectionType::ConfigNS); + ASSERT_EQ(query->getFilter(), BSON(CollectionType::fullNs(expectedColl.getNs()))); + + return vector<BSONObj>{expectedColl.toBSON()}; + }); + + // Now wait for the getCollection call to return + const auto& actualColl = future.get(); + ASSERT_EQ(expectedColl.toBSON(), actualColl.toBSON()); +} + +TEST_F(CatalogManagerReplSetTestFixture, GetCollectionNotExisting) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + auto future = async(std::launch::async, + [this] { + auto status = catalogManager()->getCollection("NonExistent"); + ASSERT_EQUALS(status.getStatus(), ErrorCodes::NamespaceNotFound); + }); + + onFindCommand([](const RemoteCommandRequest& request) { return vector<BSONObj>{}; }); + + // Now wait for the getCollection call to return + future.get(); +} + +TEST_F(CatalogManagerReplSetTestFixture, GetDatabaseExisting) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + DatabaseType expectedDb; + expectedDb.setName("bigdata"); + expectedDb.setPrimary("shard0000"); + expectedDb.setSharded(true); + + auto future = async(std::launch::async, + [this, &expectedDb] { + return assertGet(catalogManager()->getDatabase(expectedDb.getName())); + }); + + onFindCommand([&expectedDb](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), DatabaseType::ConfigNS); + + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + + ASSERT_EQ(query->ns(), DatabaseType::ConfigNS); + ASSERT_EQ(query->getFilter(), BSON(DatabaseType::name(expectedDb.getName()))); + + return vector<BSONObj>{expectedDb.toBSON()}; + }); + + const auto& actualDb = future.get(); + ASSERT_EQ(expectedDb.toBSON(), actualDb.toBSON()); +} + +TEST_F(CatalogManagerReplSetTestFixture, GetDatabaseNotExisting) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + auto future = async(std::launch::async, + [this] { + auto dbResult = catalogManager()->getDatabase("NonExistent"); + ASSERT_EQ(dbResult.getStatus(), ErrorCodes::NamespaceNotFound); + }); + + onFindCommand([](const RemoteCommandRequest& request) { return vector<BSONObj>{}; }); + + future.get(); +} + +TEST_F(CatalogManagerReplSetTestFixture, UpdateCollection) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + CollectionType collection; + collection.setNs(NamespaceString("db.coll")); + collection.setUpdatedAt(network()->now()); + collection.setUnique(true); + collection.setEpoch(OID::gen()); + collection.setKeyPattern(KeyPattern(BSON("_id" << 1))); + + auto future = async(std::launch::async, + [this, collection] { + auto status = catalogManager()->updateCollection( + collection.getNs().toString(), collection); + ASSERT_OK(status); + }); + + onCommand([collection](const RemoteCommandRequest& request) { + ASSERT_EQUALS("config", request.dbname); + + BatchedUpdateRequest actualBatchedUpdate; + std::string errmsg; + ASSERT_TRUE(actualBatchedUpdate.parseBSON(request.cmdObj, &errmsg)); + ASSERT_EQUALS(CollectionType::ConfigNS, actualBatchedUpdate.getCollName()); + auto updates = actualBatchedUpdate.getUpdates(); + ASSERT_EQUALS(1U, updates.size()); + auto update = updates.front(); + + ASSERT_TRUE(update->getUpsert()); + ASSERT_FALSE(update->getMulti()); + ASSERT_EQUALS(update->getQuery(), + BSON(CollectionType::fullNs(collection.getNs().toString()))); + ASSERT_EQUALS(update->getUpdateExpr(), collection.toBSON()); + + BatchedCommandResponse response; + response.setOk(true); + response.setNModified(1); + + return response.toBSON(); + }); + + // Now wait for the updateCollection call to return + future.wait_for(kFutureTimeout); +} + +TEST_F(CatalogManagerReplSetTestFixture, UpdateCollectionNotMaster) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + CollectionType collection; + collection.setNs(NamespaceString("db.coll")); + collection.setUpdatedAt(network()->now()); + collection.setUnique(true); + collection.setEpoch(OID::gen()); + collection.setKeyPattern(KeyPattern(BSON("_id" << 1))); + + auto future = async(std::launch::async, + [this, collection] { + auto status = catalogManager()->updateCollection( + collection.getNs().toString(), collection); + ASSERT_EQUALS(ErrorCodes::NotMaster, status); + }); + + for (int i = 0; i < 3; ++i) { + onCommand([](const RemoteCommandRequest& request) { + BatchedCommandResponse response; + response.setOk(false); + response.setErrCode(ErrorCodes::NotMaster); + response.setErrMessage("not master"); - return vector<BSONObj>{ expectedDb.toBSON() }; + return response.toBSON(); }); - - const auto& actualDb = future.get(); - ASSERT_EQ(expectedDb.toBSON(), actualDb.toBSON()); } - TEST_F(CatalogManagerReplSetTestFixture, GetDatabaseNotExisting) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - auto future = async(std::launch::async, [this] { - auto dbResult = catalogManager()->getDatabase("NonExistent"); - ASSERT_EQ(dbResult.getStatus(), ErrorCodes::NamespaceNotFound); - }); - - onFindCommand([](const RemoteCommandRequest& request) { - return vector<BSONObj>{ }; - }); - - future.get(); + // Now wait for the updateCollection call to return + future.wait_for(kFutureTimeout); +} + +TEST_F(CatalogManagerReplSetTestFixture, UpdateCollectionNotMasterRetrySuccess) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + HostAndPort host1("TestHost1"); + HostAndPort host2("TestHost2"); + targeter->setFindHostReturnValue(host1); + + CollectionType collection; + collection.setNs(NamespaceString("db.coll")); + collection.setUpdatedAt(network()->now()); + collection.setUnique(true); + collection.setEpoch(OID::gen()); + collection.setKeyPattern(KeyPattern(BSON("_id" << 1))); + + auto future = async(std::launch::async, + [this, collection] { + auto status = catalogManager()->updateCollection( + collection.getNs().toString(), collection); + ASSERT_OK(status); + }); + + onCommand([host1, host2, targeter](const RemoteCommandRequest& request) { + ASSERT_EQUALS(host1, request.target); + + BatchedCommandResponse response; + response.setOk(false); + response.setErrCode(ErrorCodes::NotMaster); + response.setErrMessage("not master"); + + // Ensure that when the catalog manager tries to retarget after getting the + // NotMaster response, it will get back a new target. + targeter->setFindHostReturnValue(host2); + return response.toBSON(); + }); + + onCommand([host2, collection](const RemoteCommandRequest& request) { + ASSERT_EQUALS(host2, request.target); + + BatchedUpdateRequest actualBatchedUpdate; + std::string errmsg; + ASSERT_TRUE(actualBatchedUpdate.parseBSON(request.cmdObj, &errmsg)); + ASSERT_EQUALS(CollectionType::ConfigNS, actualBatchedUpdate.getCollName()); + auto updates = actualBatchedUpdate.getUpdates(); + ASSERT_EQUALS(1U, updates.size()); + auto update = updates.front(); + + ASSERT_TRUE(update->getUpsert()); + ASSERT_FALSE(update->getMulti()); + ASSERT_EQUALS(update->getQuery(), + BSON(CollectionType::fullNs(collection.getNs().toString()))); + ASSERT_EQUALS(update->getUpdateExpr(), collection.toBSON()); + + BatchedCommandResponse response; + response.setOk(true); + response.setNModified(1); + + return response.toBSON(); + }); + + // Now wait for the updateCollection call to return + future.wait_for(kFutureTimeout); +} + +TEST_F(CatalogManagerReplSetTestFixture, GetAllShardsValid) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + ShardType s1; + s1.setName("shard0000"); + s1.setHost("ShardHost"); + s1.setDraining(false); + s1.setMaxSizeMB(50); + s1.setTags({"tag1", "tag2", "tag3"}); + + ShardType s2; + s2.setName("shard0001"); + s2.setHost("ShardHost"); + + ShardType s3; + s3.setName("shard0002"); + s3.setHost("ShardHost"); + s3.setMaxSizeMB(65); + + const vector<ShardType> expectedShardsList = {s1, s2, s3}; + + auto future = async(std::launch::async, + [this] { + vector<ShardType> shards; + ASSERT_OK(catalogManager()->getAllShards(&shards)); + return shards; + }); + + onFindCommand([&s1, &s2, &s3](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), ShardType::ConfigNS); + + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + + ASSERT_EQ(query->ns(), ShardType::ConfigNS); + ASSERT_EQ(query->getFilter(), BSONObj()); + + return vector<BSONObj>{s1.toBSON(), s2.toBSON(), s3.toBSON()}; + }); + + const vector<ShardType> actualShardsList = future.get(); + ASSERT_EQ(actualShardsList.size(), expectedShardsList.size()); + + for (size_t i = 0; i < actualShardsList.size(); ++i) { + ASSERT_EQ(actualShardsList[i].toBSON(), expectedShardsList[i].toBSON()); } +} - TEST_F(CatalogManagerReplSetTestFixture, UpdateCollection) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - CollectionType collection; - collection.setNs(NamespaceString("db.coll")); - collection.setUpdatedAt(network()->now()); - collection.setUnique(true); - collection.setEpoch(OID::gen()); - collection.setKeyPattern(KeyPattern(BSON("_id" << 1))); - - auto future = async(std::launch::async, [this, collection] { - auto status = catalogManager()->updateCollection(collection.getNs().toString(), - collection); - ASSERT_OK(status); - }); - - onCommand([collection](const RemoteCommandRequest& request) { - ASSERT_EQUALS("config", request.dbname); - - BatchedUpdateRequest actualBatchedUpdate; - std::string errmsg; - ASSERT_TRUE(actualBatchedUpdate.parseBSON(request.cmdObj, &errmsg)); - ASSERT_EQUALS(CollectionType::ConfigNS, actualBatchedUpdate.getCollName()); - auto updates = actualBatchedUpdate.getUpdates(); - ASSERT_EQUALS(1U, updates.size()); - auto update = updates.front(); +TEST_F(CatalogManagerReplSetTestFixture, GetAllShardsWithInvalidShard) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - ASSERT_TRUE(update->getUpsert()); - ASSERT_FALSE(update->getMulti()); - ASSERT_EQUALS(update->getQuery(), - BSON(CollectionType::fullNs(collection.getNs().toString()))); - ASSERT_EQUALS(update->getUpdateExpr(), collection.toBSON()); + auto future = async(std::launch::async, + [this] { + vector<ShardType> shards; + Status status = catalogManager()->getAllShards(&shards); - BatchedCommandResponse response; - response.setOk(true); - response.setNModified(1); + ASSERT_NOT_OK(status); + ASSERT_EQ(0U, shards.size()); + }); - return response.toBSON(); - }); + onFindCommand([](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), ShardType::ConfigNS); - // Now wait for the updateCollection call to return - future.wait_for(kFutureTimeout); - } + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - TEST_F(CatalogManagerReplSetTestFixture, UpdateCollectionNotMaster) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - CollectionType collection; - collection.setNs(NamespaceString("db.coll")); - collection.setUpdatedAt(network()->now()); - collection.setUnique(true); - collection.setEpoch(OID::gen()); - collection.setKeyPattern(KeyPattern(BSON("_id" << 1))); - - auto future = async(std::launch::async, [this, collection] { - auto status = catalogManager()->updateCollection(collection.getNs().toString(), - collection); - ASSERT_EQUALS(ErrorCodes::NotMaster, status); - }); + ASSERT_EQ(query->ns(), ShardType::ConfigNS); + ASSERT_EQ(query->getFilter(), BSONObj()); - for (int i = 0; i < 3; ++i) { - onCommand([](const RemoteCommandRequest& request) { - BatchedCommandResponse response; - response.setOk(false); - response.setErrCode(ErrorCodes::NotMaster); - response.setErrMessage("not master"); + // valid ShardType + ShardType s1; + s1.setName("shard0001"); + s1.setHost("ShardHost"); - return response.toBSON(); - }); - } + return vector<BSONObj>{ + s1.toBSON(), + BSONObj() // empty document is invalid + }; + }); - // Now wait for the updateCollection call to return - future.wait_for(kFutureTimeout); - } + future.get(); +} - TEST_F(CatalogManagerReplSetTestFixture, UpdateCollectionNotMasterRetrySuccess) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - HostAndPort host1("TestHost1"); - HostAndPort host2("TestHost2"); - targeter->setFindHostReturnValue(host1); - - CollectionType collection; - collection.setNs(NamespaceString("db.coll")); - collection.setUpdatedAt(network()->now()); - collection.setUnique(true); - collection.setEpoch(OID::gen()); - collection.setKeyPattern(KeyPattern(BSON("_id" << 1))); - - auto future = async(std::launch::async, [this, collection] { - auto status = catalogManager()->updateCollection(collection.getNs().toString(), - collection); - ASSERT_OK(status); - }); +TEST_F(CatalogManagerReplSetTestFixture, GetChunksForNS) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - onCommand([host1, host2, targeter](const RemoteCommandRequest& request) { - ASSERT_EQUALS(host1, request.target); + OID oid = OID::gen(); - BatchedCommandResponse response; - response.setOk(false); - response.setErrCode(ErrorCodes::NotMaster); - response.setErrMessage("not master"); + ChunkType chunkA; + chunkA.setName("chunk0000"); + chunkA.setNS("TestDB.TestColl"); + chunkA.setMin(BSON("a" << 1)); + chunkA.setMax(BSON("a" << 100)); + chunkA.setVersion({1, 2, oid}); + chunkA.setShard("shard0000"); - // Ensure that when the catalog manager tries to retarget after getting the - // NotMaster response, it will get back a new target. - targeter->setFindHostReturnValue(host2); - return response.toBSON(); - }); + ChunkType chunkB; + chunkB.setName("chunk0001"); + chunkB.setNS("TestDB.TestColl"); + chunkB.setMin(BSON("a" << 100)); + chunkB.setMax(BSON("a" << 200)); + chunkB.setVersion({3, 4, oid}); + chunkB.setShard("shard0001"); - onCommand([host2, collection](const RemoteCommandRequest& request) { - ASSERT_EQUALS(host2, request.target); + ChunkVersion queryChunkVersion({1, 2, oid}); - BatchedUpdateRequest actualBatchedUpdate; - std::string errmsg; - ASSERT_TRUE(actualBatchedUpdate.parseBSON(request.cmdObj, &errmsg)); - ASSERT_EQUALS(CollectionType::ConfigNS, actualBatchedUpdate.getCollName()); - auto updates = actualBatchedUpdate.getUpdates(); - ASSERT_EQUALS(1U, updates.size()); - auto update = updates.front(); + const Query chunksQuery( + BSON(ChunkType::ns("TestDB.TestColl") + << ChunkType::DEPRECATED_lastmod() + << BSON("$gte" << static_cast<long long>(queryChunkVersion.toLong())))); - ASSERT_TRUE(update->getUpsert()); - ASSERT_FALSE(update->getMulti()); - ASSERT_EQUALS(update->getQuery(), - BSON(CollectionType::fullNs(collection.getNs().toString()))); - ASSERT_EQUALS(update->getUpdateExpr(), collection.toBSON()); + auto future = async(std::launch::async, + [this, &chunksQuery] { + vector<ChunkType> chunks; - BatchedCommandResponse response; - response.setOk(true); - response.setNModified(1); + ASSERT_OK(catalogManager()->getChunks(chunksQuery, 0, &chunks)); + ASSERT_EQ(2U, chunks.size()); - return response.toBSON(); - }); + return chunks; + }); - // Now wait for the updateCollection call to return - future.wait_for(kFutureTimeout); - } + onFindCommand([&chunksQuery, chunkA, chunkB](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), ChunkType::ConfigNS); - TEST_F(CatalogManagerReplSetTestFixture, GetAllShardsValid) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - ShardType s1; - s1.setName("shard0000"); - s1.setHost("ShardHost"); - s1.setDraining(false); - s1.setMaxSizeMB(50); - s1.setTags({ "tag1", "tag2", "tag3" }); + ASSERT_EQ(query->ns(), ChunkType::ConfigNS); + ASSERT_EQ(query->getFilter(), chunksQuery.getFilter()); - ShardType s2; - s2.setName("shard0001"); - s2.setHost("ShardHost"); + return vector<BSONObj>{chunkA.toBSON(), chunkB.toBSON()}; + }); - ShardType s3; - s3.setName("shard0002"); - s3.setHost("ShardHost"); - s3.setMaxSizeMB(65); + const auto& chunks = future.get(); + ASSERT_EQ(chunkA.toBSON(), chunks[0].toBSON()); + ASSERT_EQ(chunkB.toBSON(), chunks[1].toBSON()); +} - const vector<ShardType> expectedShardsList = { s1, s2, s3 }; +TEST_F(CatalogManagerReplSetTestFixture, GetChunksForNSNoChunks) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - auto future = async(std::launch::async, [this] { - vector<ShardType> shards; - ASSERT_OK(catalogManager()->getAllShards(&shards)); - return shards; - }); + ChunkVersion queryChunkVersion({1, 2, OID::gen()}); - onFindCommand([&s1, &s2, &s3](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), ShardType::ConfigNS); + const Query chunksQuery( + BSON(ChunkType::ns("TestDB.TestColl") + << ChunkType::DEPRECATED_lastmod() + << BSON("$gte" << static_cast<long long>(queryChunkVersion.toLong())))); - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + auto future = async(std::launch::async, + [this, &chunksQuery] { + vector<ChunkType> chunks; - ASSERT_EQ(query->ns(), ShardType::ConfigNS); - ASSERT_EQ(query->getFilter(), BSONObj()); + ASSERT_OK(catalogManager()->getChunks(chunksQuery, 0, &chunks)); + ASSERT_EQ(0U, chunks.size()); + }); - return vector<BSONObj>{ s1.toBSON(), s2.toBSON(), s3.toBSON() }; - }); + onFindCommand([&chunksQuery](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), ChunkType::ConfigNS); - const vector<ShardType> actualShardsList = future.get(); - ASSERT_EQ(actualShardsList.size(), expectedShardsList.size()); + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - for (size_t i = 0; i < actualShardsList.size(); ++i) { - ASSERT_EQ(actualShardsList[i].toBSON(), expectedShardsList[i].toBSON()); - } - } + ASSERT_EQ(query->ns(), ChunkType::ConfigNS); + ASSERT_EQ(query->getFilter(), chunksQuery.getFilter()); - TEST_F(CatalogManagerReplSetTestFixture, GetAllShardsWithInvalidShard) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + return vector<BSONObj>{}; + }); - auto future = async(std::launch::async, [this] { - vector<ShardType> shards; - Status status = catalogManager()->getAllShards(&shards); + future.get(); +} - ASSERT_NOT_OK(status); - ASSERT_EQ(0U, shards.size()); - }); +TEST_F(CatalogManagerReplSetTestFixture, GetChunksForNSInvalidChunk) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - onFindCommand([](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), ShardType::ConfigNS); + ChunkVersion queryChunkVersion({1, 2, OID::gen()}); - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + const Query chunksQuery( + BSON(ChunkType::ns("TestDB.TestColl") + << ChunkType::DEPRECATED_lastmod() + << BSON("$gte" << static_cast<long long>(queryChunkVersion.toLong())))); - ASSERT_EQ(query->ns(), ShardType::ConfigNS); - ASSERT_EQ(query->getFilter(), BSONObj()); + auto future = async(std::launch::async, + [this, &chunksQuery] { + vector<ChunkType> chunks; + Status status = catalogManager()->getChunks(chunksQuery, 0, &chunks); - // valid ShardType - ShardType s1; - s1.setName("shard0001"); - s1.setHost("ShardHost"); + ASSERT_EQUALS(ErrorCodes::FailedToParse, status); + ASSERT_EQ(0U, chunks.size()); + }); - return vector<BSONObj> { - s1.toBSON(), - BSONObj() // empty document is invalid - }; - }); + onFindCommand([&chunksQuery](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), ChunkType::ConfigNS); - future.get(); - } - - TEST_F(CatalogManagerReplSetTestFixture, GetChunksForNS) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - OID oid = OID::gen(); + ASSERT_EQ(query->ns(), ChunkType::ConfigNS); + ASSERT_EQ(query->getFilter(), chunksQuery.getFilter()); ChunkType chunkA; chunkA.setName("chunk0000"); chunkA.setNS("TestDB.TestColl"); chunkA.setMin(BSON("a" << 1)); chunkA.setMax(BSON("a" << 100)); - chunkA.setVersion({ 1, 2, oid }); + chunkA.setVersion({1, 2, OID::gen()}); chunkA.setShard("shard0000"); ChunkType chunkB; @@ -414,450 +530,355 @@ namespace { chunkB.setNS("TestDB.TestColl"); chunkB.setMin(BSON("a" << 100)); chunkB.setMax(BSON("a" << 200)); - chunkB.setVersion({ 3, 4, oid }); - chunkB.setShard("shard0001"); - - ChunkVersion queryChunkVersion({ 1, 2, oid }); - - const Query chunksQuery(BSON(ChunkType::ns("TestDB.TestColl") << - ChunkType::DEPRECATED_lastmod() << - BSON("$gte" << static_cast<long long>( - queryChunkVersion.toLong())))); - - auto future = async(std::launch::async, [this, &chunksQuery] { - vector<ChunkType> chunks; - - ASSERT_OK(catalogManager()->getChunks(chunksQuery, 0, &chunks)); - ASSERT_EQ(2U, chunks.size()); - - return chunks; - }); - - onFindCommand([&chunksQuery, chunkA, chunkB](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), ChunkType::ConfigNS); - - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - - ASSERT_EQ(query->ns(), ChunkType::ConfigNS); - ASSERT_EQ(query->getFilter(), chunksQuery.getFilter()); - - return vector<BSONObj>{ chunkA.toBSON(), chunkB.toBSON() }; - }); - - const auto& chunks = future.get(); - ASSERT_EQ(chunkA.toBSON(), chunks[0].toBSON()); - ASSERT_EQ(chunkB.toBSON(), chunks[1].toBSON()); - } - - TEST_F(CatalogManagerReplSetTestFixture, GetChunksForNSNoChunks) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - ChunkVersion queryChunkVersion({ 1, 2, OID::gen() }); - - const Query chunksQuery(BSON(ChunkType::ns("TestDB.TestColl") << - ChunkType::DEPRECATED_lastmod() << - BSON("$gte" << static_cast<long long>( - queryChunkVersion.toLong())))); - - auto future = async(std::launch::async, [this, &chunksQuery] { - vector<ChunkType> chunks; - - ASSERT_OK(catalogManager()->getChunks(chunksQuery, 0, &chunks)); - ASSERT_EQ(0U, chunks.size()); - }); - - onFindCommand([&chunksQuery](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), ChunkType::ConfigNS); - - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - - ASSERT_EQ(query->ns(), ChunkType::ConfigNS); - ASSERT_EQ(query->getFilter(), chunksQuery.getFilter()); - - return vector<BSONObj>{ }; - }); - - future.get(); - } - - TEST_F(CatalogManagerReplSetTestFixture, GetChunksForNSInvalidChunk) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - ChunkVersion queryChunkVersion({ 1, 2, OID::gen() }); - - const Query chunksQuery(BSON(ChunkType::ns("TestDB.TestColl") << - ChunkType::DEPRECATED_lastmod() << - BSON("$gte" << static_cast<long long>( - queryChunkVersion.toLong())))); - - auto future = async(std::launch::async, [this, &chunksQuery] { - vector<ChunkType> chunks; - Status status = catalogManager()->getChunks(chunksQuery, 0, &chunks); - - ASSERT_EQUALS(ErrorCodes::FailedToParse, status); - ASSERT_EQ(0U, chunks.size()); - }); - - onFindCommand([&chunksQuery](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), ChunkType::ConfigNS); - - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - - ASSERT_EQ(query->ns(), ChunkType::ConfigNS); - ASSERT_EQ(query->getFilter(), chunksQuery.getFilter()); - - ChunkType chunkA; - chunkA.setName("chunk0000"); - chunkA.setNS("TestDB.TestColl"); - chunkA.setMin(BSON("a" << 1)); - chunkA.setMax(BSON("a" << 100)); - chunkA.setVersion({ 1, 2, OID::gen() }); - chunkA.setShard("shard0000"); - - ChunkType chunkB; - chunkB.setName("chunk0001"); - chunkB.setNS("TestDB.TestColl"); - chunkB.setMin(BSON("a" << 100)); - chunkB.setMax(BSON("a" << 200)); - chunkB.setVersion({ 3, 4, OID::gen() }); - // Missing shard id - - return vector<BSONObj>{ chunkA.toBSON(), chunkB.toBSON() }; - }); - - future.get(); - } - - TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementReadCommand) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - auto future = async(std::launch::async, [this] { - BSONObjBuilder responseBuilder; - bool ok = catalogManager()->runUserManagementReadCommand("test", - BSON("usersInfo" << 1), - &responseBuilder); - ASSERT_TRUE(ok); - - BSONObj response = responseBuilder.obj(); - ASSERT_TRUE(response["ok"].trueValue()); - auto users = response["users"].Array(); - ASSERT_EQUALS(0U, users.size()); - }); - - onCommand([](const RemoteCommandRequest& request) { - ASSERT_EQUALS("test", request.dbname); - ASSERT_EQUALS(BSON("usersInfo" << 1), request.cmdObj); - - return BSON("ok" << 1 << "users" << BSONArrayBuilder().arr()); - }); - - // Now wait for the runUserManagementReadCommand call to return - future.wait_for(kFutureTimeout); - } - - TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementReadCommandUnsatisfiedReadPref) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(Status(ErrorCodes::FailedToSatisfyReadPreference, - "no nodes up")); - - BSONObjBuilder responseBuilder; - bool ok = catalogManager()->runUserManagementReadCommand("test", - BSON("usersInfo" << 1), - &responseBuilder); - ASSERT_FALSE(ok); - - Status commandStatus = Command::getStatusFromCommandResult(responseBuilder.obj()); - ASSERT_EQUALS(ErrorCodes::FailedToSatisfyReadPreference, commandStatus); - } - - TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementWriteCommandDistLockHeld) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - distLock()->expectLock( - [](StringData name, - StringData whyMessage, - milliseconds waitFor, - milliseconds lockTryInterval) { + chunkB.setVersion({3, 4, OID::gen()}); + // Missing shard id + + return vector<BSONObj>{chunkA.toBSON(), chunkB.toBSON()}; + }); + + future.get(); +} + +TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementReadCommand) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + auto future = async(std::launch::async, + [this] { + BSONObjBuilder responseBuilder; + bool ok = catalogManager()->runUserManagementReadCommand( + "test", BSON("usersInfo" << 1), &responseBuilder); + ASSERT_TRUE(ok); + + BSONObj response = responseBuilder.obj(); + ASSERT_TRUE(response["ok"].trueValue()); + auto users = response["users"].Array(); + ASSERT_EQUALS(0U, users.size()); + }); + + onCommand([](const RemoteCommandRequest& request) { + ASSERT_EQUALS("test", request.dbname); + ASSERT_EQUALS(BSON("usersInfo" << 1), request.cmdObj); + + return BSON("ok" << 1 << "users" << BSONArrayBuilder().arr()); + }); + + // Now wait for the runUserManagementReadCommand call to return + future.wait_for(kFutureTimeout); +} + +TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementReadCommandUnsatisfiedReadPref) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue( + Status(ErrorCodes::FailedToSatisfyReadPreference, "no nodes up")); + + BSONObjBuilder responseBuilder; + bool ok = catalogManager()->runUserManagementReadCommand( + "test", BSON("usersInfo" << 1), &responseBuilder); + ASSERT_FALSE(ok); + + Status commandStatus = Command::getStatusFromCommandResult(responseBuilder.obj()); + ASSERT_EQUALS(ErrorCodes::FailedToSatisfyReadPreference, commandStatus); +} + +TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementWriteCommandDistLockHeld) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + distLock()->expectLock( + [](StringData name, + StringData whyMessage, + milliseconds waitFor, + milliseconds lockTryInterval) { + ASSERT_EQUALS("authorizationData", name); + ASSERT_EQUALS("dropUser", whyMessage); + }, + Status(ErrorCodes::LockBusy, "lock already held")); + + BSONObjBuilder responseBuilder; + bool ok = catalogManager()->runUserManagementWriteCommand("dropUser", + "test", + BSON("dropUser" + << "test"), + &responseBuilder); + ASSERT_FALSE(ok); + BSONObj response = responseBuilder.obj(); + ASSERT_EQUALS(ErrorCodes::LockBusy, Command::getStatusFromCommandResult(response)); +} + +TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementWriteCommandSuccess) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + distLock()->expectLock( + [](StringData name, + StringData whyMessage, + milliseconds waitFor, + milliseconds lockTryInterval) { ASSERT_EQUALS("authorizationData", name); ASSERT_EQUALS("dropUser", whyMessage); - }, Status(ErrorCodes::LockBusy, "lock already held")); + }, + Status::OK()); + + auto future = + async(std::launch::async, + [this] { + BSONObjBuilder responseBuilder; + bool ok = catalogManager()->runUserManagementWriteCommand("dropUser", + "test", + BSON("dropUser" + << "test"), + &responseBuilder); + ASSERT_FALSE(ok); + + Status commandStatus = Command::getStatusFromCommandResult(responseBuilder.obj()); + ASSERT_EQUALS(ErrorCodes::UserNotFound, commandStatus); + }); + + onCommand([](const RemoteCommandRequest& request) { + ASSERT_EQUALS("test", request.dbname); + ASSERT_EQUALS(BSON("dropUser" + << "test"), + request.cmdObj); BSONObjBuilder responseBuilder; - bool ok = catalogManager()->runUserManagementWriteCommand("dropUser", - "test", - BSON("dropUser" << "test"), - &responseBuilder); - ASSERT_FALSE(ok); - BSONObj response = responseBuilder.obj(); - ASSERT_EQUALS(ErrorCodes::LockBusy, Command::getStatusFromCommandResult(response)); - } - - TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementWriteCommandSuccess) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - distLock()->expectLock( - [](StringData name, - StringData whyMessage, - milliseconds waitFor, - milliseconds lockTryInterval) { + Command::appendCommandStatus(responseBuilder, + Status(ErrorCodes::UserNotFound, "User test@test not found")); + return responseBuilder.obj(); + }); + + // Now wait for the runUserManagementWriteCommand call to return + future.wait_for(kFutureTimeout); +} + +TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementWriteCommandNotMaster) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + distLock()->expectLock( + [](StringData name, + StringData whyMessage, + milliseconds waitFor, + milliseconds lockTryInterval) { ASSERT_EQUALS("authorizationData", name); ASSERT_EQUALS("dropUser", whyMessage); - }, Status::OK()); - - auto future = async(std::launch::async, [this] { - BSONObjBuilder responseBuilder; - bool ok = catalogManager()->runUserManagementWriteCommand("dropUser", - "test", - BSON("dropUser" << "test"), - &responseBuilder); - ASSERT_FALSE(ok); - - Status commandStatus = Command::getStatusFromCommandResult(responseBuilder.obj()); - ASSERT_EQUALS(ErrorCodes::UserNotFound, commandStatus); - }); - + }, + Status::OK()); + + auto future = + async(std::launch::async, + [this] { + BSONObjBuilder responseBuilder; + bool ok = catalogManager()->runUserManagementWriteCommand("dropUser", + "test", + BSON("dropUser" + << "test"), + &responseBuilder); + ASSERT_FALSE(ok); + + Status commandStatus = Command::getStatusFromCommandResult(responseBuilder.obj()); + ASSERT_EQUALS(ErrorCodes::NotMaster, commandStatus); + }); + + for (int i = 0; i < 3; ++i) { onCommand([](const RemoteCommandRequest& request) { - ASSERT_EQUALS("test", request.dbname); - ASSERT_EQUALS(BSON("dropUser" << "test"), request.cmdObj); - BSONObjBuilder responseBuilder; Command::appendCommandStatus(responseBuilder, - Status(ErrorCodes::UserNotFound, - "User test@test not found")); + Status(ErrorCodes::NotMaster, "not master")); return responseBuilder.obj(); }); - - // Now wait for the runUserManagementWriteCommand call to return - future.wait_for(kFutureTimeout); } - TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementWriteCommandNotMaster) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - - distLock()->expectLock( - [](StringData name, - StringData whyMessage, - milliseconds waitFor, - milliseconds lockTryInterval) { + // Now wait for the runUserManagementWriteCommand call to return + future.wait_for(kFutureTimeout); +} + +TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementWriteCommandNotMasterRetrySuccess) { + HostAndPort host1("TestHost1"); + HostAndPort host2("TestHost2"); + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(host1); + + distLock()->expectLock( + [](StringData name, + StringData whyMessage, + milliseconds waitFor, + milliseconds lockTryInterval) { ASSERT_EQUALS("authorizationData", name); ASSERT_EQUALS("dropUser", whyMessage); - }, Status::OK()); - - auto future = async(std::launch::async, [this] { - BSONObjBuilder responseBuilder; - bool ok = catalogManager()->runUserManagementWriteCommand("dropUser", - "test", - BSON("dropUser" << "test"), - &responseBuilder); - ASSERT_FALSE(ok); - - Status commandStatus = Command::getStatusFromCommandResult(responseBuilder.obj()); - ASSERT_EQUALS(ErrorCodes::NotMaster, commandStatus); - }); - - for (int i = 0; i < 3; ++i) { - onCommand([](const RemoteCommandRequest& request) { - BSONObjBuilder responseBuilder; - Command::appendCommandStatus(responseBuilder, - Status(ErrorCodes::NotMaster, "not master")); - return responseBuilder.obj(); - }); - } - - // Now wait for the runUserManagementWriteCommand call to return - future.wait_for(kFutureTimeout); - } + }, + Status::OK()); + + auto future = + async(std::launch::async, + [this] { + BSONObjBuilder responseBuilder; + bool ok = catalogManager()->runUserManagementWriteCommand("dropUser", + "test", + BSON("dropUser" + << "test"), + &responseBuilder); + ASSERT_TRUE(ok); + + Status commandStatus = Command::getStatusFromCommandResult(responseBuilder.obj()); + ASSERT_OK(commandStatus); + }); + + onCommand([targeter, host1, host2](const RemoteCommandRequest& request) { + ASSERT_EQUALS(host1, request.target); - TEST_F(CatalogManagerReplSetTestFixture, RunUserManagementWriteCommandNotMasterRetrySuccess) { - HostAndPort host1("TestHost1"); - HostAndPort host2("TestHost2"); - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(host1); - - distLock()->expectLock( - [](StringData name, - StringData whyMessage, - milliseconds waitFor, - milliseconds lockTryInterval) { - ASSERT_EQUALS("authorizationData", name); - ASSERT_EQUALS("dropUser", whyMessage); - }, Status::OK()); - - auto future = async(std::launch::async, [this] { - BSONObjBuilder responseBuilder; - bool ok = catalogManager()->runUserManagementWriteCommand("dropUser", - "test", - BSON("dropUser" << "test"), - &responseBuilder); - ASSERT_TRUE(ok); - - Status commandStatus = Command::getStatusFromCommandResult(responseBuilder.obj()); - ASSERT_OK(commandStatus); - }); - - onCommand([targeter, host1, host2](const RemoteCommandRequest& request) { - ASSERT_EQUALS(host1, request.target); - - BSONObjBuilder responseBuilder; - Command::appendCommandStatus(responseBuilder, - Status(ErrorCodes::NotMaster, "not master")); - - // Ensure that when the catalog manager tries to retarget after getting the - // NotMaster response, it will get back a new target. - targeter->setFindHostReturnValue(host2); - return responseBuilder.obj(); - }); - - onCommand([host2](const RemoteCommandRequest& request) { - ASSERT_EQUALS(host2, request.target); - ASSERT_EQUALS("test", request.dbname); - ASSERT_EQUALS(BSON("dropUser" << "test"), request.cmdObj); - - return BSON("ok" << 1); - }); + BSONObjBuilder responseBuilder; + Command::appendCommandStatus(responseBuilder, Status(ErrorCodes::NotMaster, "not master")); + + // Ensure that when the catalog manager tries to retarget after getting the + // NotMaster response, it will get back a new target. + targeter->setFindHostReturnValue(host2); + return responseBuilder.obj(); + }); - // Now wait for the runUserManagementWriteCommand call to return - future.wait_for(kFutureTimeout); - } + onCommand([host2](const RemoteCommandRequest& request) { + ASSERT_EQUALS(host2, request.target); + ASSERT_EQUALS("test", request.dbname); + ASSERT_EQUALS(BSON("dropUser" + << "test"), + request.cmdObj); - TEST_F(CatalogManagerReplSetTestFixture, GetGlobalSettingsBalancerDoc) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + return BSON("ok" << 1); + }); - // sample balancer doc - SettingsType st1; - st1.setKey(SettingsType::BalancerDocKey); - st1.setBalancerStopped(true); + // Now wait for the runUserManagementWriteCommand call to return + future.wait_for(kFutureTimeout); +} - auto future = async(std::launch::async, [this] { - return assertGet(catalogManager()->getGlobalSettings(SettingsType::BalancerDocKey)); - }); +TEST_F(CatalogManagerReplSetTestFixture, GetGlobalSettingsBalancerDoc) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); + + // sample balancer doc + SettingsType st1; + st1.setKey(SettingsType::BalancerDocKey); + st1.setBalancerStopped(true); + + auto future = async(std::launch::async, + [this] { + return assertGet( + catalogManager()->getGlobalSettings(SettingsType::BalancerDocKey)); + }); - onFindCommand([st1](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), SettingsType::ConfigNS); + onFindCommand([st1](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), SettingsType::ConfigNS); - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - ASSERT_EQ(query->ns(), SettingsType::ConfigNS); - ASSERT_EQ(query->getFilter(), BSON(SettingsType::key(SettingsType::BalancerDocKey))); + ASSERT_EQ(query->ns(), SettingsType::ConfigNS); + ASSERT_EQ(query->getFilter(), BSON(SettingsType::key(SettingsType::BalancerDocKey))); - return vector<BSONObj>{ st1.toBSON() }; - }); + return vector<BSONObj>{st1.toBSON()}; + }); - const auto& actualBalSettings = future.get(); - ASSERT_EQ(actualBalSettings.toBSON(), st1.toBSON()); - } + const auto& actualBalSettings = future.get(); + ASSERT_EQ(actualBalSettings.toBSON(), st1.toBSON()); +} - TEST_F(CatalogManagerReplSetTestFixture, GetGlobalSettingsChunkSizeDoc) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); +TEST_F(CatalogManagerReplSetTestFixture, GetGlobalSettingsChunkSizeDoc) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - // sample chunk size doc - SettingsType st1; - st1.setKey(SettingsType::ChunkSizeDocKey); - st1.setChunkSizeMB(80); + // sample chunk size doc + SettingsType st1; + st1.setKey(SettingsType::ChunkSizeDocKey); + st1.setChunkSizeMB(80); - auto future = async(std::launch::async, [this] { - return assertGet(catalogManager()->getGlobalSettings(SettingsType::ChunkSizeDocKey)); - }); + auto future = async(std::launch::async, + [this] { + return assertGet( + catalogManager()->getGlobalSettings(SettingsType::ChunkSizeDocKey)); + }); - onFindCommand([st1](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), SettingsType::ConfigNS); + onFindCommand([st1](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), SettingsType::ConfigNS); - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - ASSERT_EQ(query->ns(), SettingsType::ConfigNS); - ASSERT_EQ(query->getFilter(), BSON(SettingsType::key(SettingsType::ChunkSizeDocKey))); + ASSERT_EQ(query->ns(), SettingsType::ConfigNS); + ASSERT_EQ(query->getFilter(), BSON(SettingsType::key(SettingsType::ChunkSizeDocKey))); - return vector<BSONObj>{ st1.toBSON() }; - }); + return vector<BSONObj>{st1.toBSON()}; + }); - const auto& actualBalSettings = future.get(); - ASSERT_EQ(actualBalSettings.toBSON(), st1.toBSON()); - } + const auto& actualBalSettings = future.get(); + ASSERT_EQ(actualBalSettings.toBSON(), st1.toBSON()); +} - TEST_F(CatalogManagerReplSetTestFixture, GetGlobalSettingsInvalidDoc) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); +TEST_F(CatalogManagerReplSetTestFixture, GetGlobalSettingsInvalidDoc) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - auto future = async(std::launch::async, [this] { - const auto balSettings = catalogManager()->getGlobalSettings("invalidKey"); + auto future = async(std::launch::async, + [this] { + const auto balSettings = + catalogManager()->getGlobalSettings("invalidKey"); - ASSERT_EQ(balSettings.getStatus(), ErrorCodes::FailedToParse); - }); + ASSERT_EQ(balSettings.getStatus(), ErrorCodes::FailedToParse); + }); - onFindCommand([](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), SettingsType::ConfigNS); + onFindCommand([](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), SettingsType::ConfigNS); - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - ASSERT_EQ(query->ns(), SettingsType::ConfigNS); - ASSERT_EQ(query->getFilter(), BSON(SettingsType::key("invalidKey"))); + ASSERT_EQ(query->ns(), SettingsType::ConfigNS); + ASSERT_EQ(query->getFilter(), BSON(SettingsType::key("invalidKey"))); - return vector<BSONObj> { - BSON("invalidKey" << "some value") // invalid settings document -- key is required - }; - }); + return vector<BSONObj>{ + BSON("invalidKey" + << "some value") // invalid settings document -- key is required + }; + }); - future.get(); - } + future.get(); +} - TEST_F(CatalogManagerReplSetTestFixture, GetGlobalSettingsNonExistent) { - RemoteCommandTargeterMock* targeter = - RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); - targeter->setFindHostReturnValue(HostAndPort("TestHost1")); +TEST_F(CatalogManagerReplSetTestFixture, GetGlobalSettingsNonExistent) { + RemoteCommandTargeterMock* targeter = + RemoteCommandTargeterMock::get(shardRegistry()->getShard("config")->getTargeter()); + targeter->setFindHostReturnValue(HostAndPort("TestHost1")); - auto future = async(std::launch::async, [this] { - const auto chunkSizeSettings = catalogManager()->getGlobalSettings( - SettingsType::ChunkSizeDocKey); + auto future = + async(std::launch::async, + [this] { + const auto chunkSizeSettings = + catalogManager()->getGlobalSettings(SettingsType::ChunkSizeDocKey); - ASSERT_EQ(chunkSizeSettings.getStatus(), ErrorCodes::NoMatchingDocument); - }); + ASSERT_EQ(chunkSizeSettings.getStatus(), ErrorCodes::NoMatchingDocument); + }); - onFindCommand([](const RemoteCommandRequest& request) { - const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); - ASSERT_EQ(nss.toString(), SettingsType::ConfigNS); + onFindCommand([](const RemoteCommandRequest& request) { + const NamespaceString nss(request.dbname, request.cmdObj.firstElement().String()); + ASSERT_EQ(nss.toString(), SettingsType::ConfigNS); - auto query = assertGet( - LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); + auto query = assertGet(LiteParsedQuery::makeFromFindCommand(nss, request.cmdObj, false)); - ASSERT_EQ(query->ns(), SettingsType::ConfigNS); - ASSERT_EQ(query->getFilter(), BSON(SettingsType::key(SettingsType::ChunkSizeDocKey))); + ASSERT_EQ(query->ns(), SettingsType::ConfigNS); + ASSERT_EQ(query->getFilter(), BSON(SettingsType::key(SettingsType::ChunkSizeDocKey))); - return vector<BSONObj> { }; - }); + return vector<BSONObj>{}; + }); - future.get(); - } + future.get(); +} -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/catalog/replset/catalog_manager_replica_set_test_fixture.cpp b/src/mongo/s/catalog/replset/catalog_manager_replica_set_test_fixture.cpp index 3f0d78f451b..795c1d329db 100644 --- a/src/mongo/s/catalog/replset/catalog_manager_replica_set_test_fixture.cpp +++ b/src/mongo/s/catalog/replset/catalog_manager_replica_set_test_fixture.cpp @@ -48,133 +48,127 @@ namespace mongo { - using executor::NetworkInterfaceMock; - using std::vector; +using executor::NetworkInterfaceMock; +using std::vector; - CatalogManagerReplSetTestFixture::CatalogManagerReplSetTestFixture() = default; +CatalogManagerReplSetTestFixture::CatalogManagerReplSetTestFixture() = default; - CatalogManagerReplSetTestFixture::~CatalogManagerReplSetTestFixture() = default; +CatalogManagerReplSetTestFixture::~CatalogManagerReplSetTestFixture() = default; - void CatalogManagerReplSetTestFixture::setUp() { - std::unique_ptr<NetworkInterfaceMock> network( - stdx::make_unique<executor::NetworkInterfaceMock>()); +void CatalogManagerReplSetTestFixture::setUp() { + std::unique_ptr<NetworkInterfaceMock> network( + stdx::make_unique<executor::NetworkInterfaceMock>()); - _mockNetwork = network.get(); + _mockNetwork = network.get(); - std::unique_ptr<repl::ReplicationExecutor> executor( - stdx::make_unique<repl::ReplicationExecutor>(network.release(), - nullptr, - 0)); + std::unique_ptr<repl::ReplicationExecutor> executor( + stdx::make_unique<repl::ReplicationExecutor>(network.release(), nullptr, 0)); - // The executor thread might run after the executor unique_ptr above has been moved to the - // ShardRegistry, so make sure we get the underlying pointer before that. - _executorThread = std::thread(std::bind([](repl::ReplicationExecutor* executorPtr) { - executorPtr->run(); - }, - executor.get())); + // The executor thread might run after the executor unique_ptr above has been moved to the + // ShardRegistry, so make sure we get the underlying pointer before that. + _executorThread = std::thread(std::bind( + [](repl::ReplicationExecutor* executorPtr) { executorPtr->run(); }, executor.get())); - std::unique_ptr<CatalogManagerReplicaSet> cm( - stdx::make_unique<CatalogManagerReplicaSet>()); + std::unique_ptr<CatalogManagerReplicaSet> cm(stdx::make_unique<CatalogManagerReplicaSet>()); - ASSERT_OK(cm->init(ConnectionString::forReplicaSet("CatalogManagerReplSetTest", - { HostAndPort{ "TestHost1" }, - HostAndPort{ "TestHost2" } }), - stdx::make_unique<DistLockManagerMock>())); + ASSERT_OK(cm->init( + ConnectionString::forReplicaSet("CatalogManagerReplSetTest", + {HostAndPort{"TestHost1"}, HostAndPort{"TestHost2"}}), + stdx::make_unique<DistLockManagerMock>())); - std::unique_ptr<ShardRegistry> shardRegistry( - stdx::make_unique<ShardRegistry>(stdx::make_unique<RemoteCommandTargeterFactoryMock>(), - stdx::make_unique<RemoteCommandRunnerMock>(), - std::move(executor), - cm.get())); + std::unique_ptr<ShardRegistry> shardRegistry( + stdx::make_unique<ShardRegistry>(stdx::make_unique<RemoteCommandTargeterFactoryMock>(), + stdx::make_unique<RemoteCommandRunnerMock>(), + std::move(executor), + cm.get())); - // For now initialize the global grid object. All sharding objects will be accessible - // from there until we get rid of it. - grid.init(std::move(cm), std::move(shardRegistry)); - } + // For now initialize the global grid object. All sharding objects will be accessible + // from there until we get rid of it. + grid.init(std::move(cm), std::move(shardRegistry)); +} - void CatalogManagerReplSetTestFixture::tearDown() { - // Stop the executor and wait for the executor thread to complete. This means that there - // will be no more calls into the executor and it can be safely deleted. - shardRegistry()->getExecutor()->shutdown(); - _executorThread.join(); +void CatalogManagerReplSetTestFixture::tearDown() { + // Stop the executor and wait for the executor thread to complete. This means that there + // will be no more calls into the executor and it can be safely deleted. + shardRegistry()->getExecutor()->shutdown(); + _executorThread.join(); - // This call will delete the shard registry, which will terminate the executor - grid.clearForUnitTests(); - } + // This call will delete the shard registry, which will terminate the executor + grid.clearForUnitTests(); +} - CatalogManagerReplicaSet* CatalogManagerReplSetTestFixture::catalogManager() const { - auto cm = dynamic_cast<CatalogManagerReplicaSet*>(grid.catalogManager()); - invariant(cm); +CatalogManagerReplicaSet* CatalogManagerReplSetTestFixture::catalogManager() const { + auto cm = dynamic_cast<CatalogManagerReplicaSet*>(grid.catalogManager()); + invariant(cm); - return cm; - } + return cm; +} - ShardRegistry* CatalogManagerReplSetTestFixture::shardRegistry() const { - return grid.shardRegistry(); - } +ShardRegistry* CatalogManagerReplSetTestFixture::shardRegistry() const { + return grid.shardRegistry(); +} - RemoteCommandRunnerMock* CatalogManagerReplSetTestFixture::commandRunner() const { - return RemoteCommandRunnerMock::get(shardRegistry()->getCommandRunner()); - } +RemoteCommandRunnerMock* CatalogManagerReplSetTestFixture::commandRunner() const { + return RemoteCommandRunnerMock::get(shardRegistry()->getCommandRunner()); +} - executor::NetworkInterfaceMock* CatalogManagerReplSetTestFixture::network() const { - return _mockNetwork; - } +executor::NetworkInterfaceMock* CatalogManagerReplSetTestFixture::network() const { + return _mockNetwork; +} - DistLockManagerMock* CatalogManagerReplSetTestFixture::distLock() const { - auto distLock = dynamic_cast<DistLockManagerMock*>(catalogManager()->getDistLockManager()); - invariant(distLock); +DistLockManagerMock* CatalogManagerReplSetTestFixture::distLock() const { + auto distLock = dynamic_cast<DistLockManagerMock*>(catalogManager()->getDistLockManager()); + invariant(distLock); - return distLock; - } + return distLock; +} - void CatalogManagerReplSetTestFixture::onCommand(OnCommandFunction func) { - network()->enterNetwork(); +void CatalogManagerReplSetTestFixture::onCommand(OnCommandFunction func) { + network()->enterNetwork(); - const NetworkInterfaceMock::NetworkOperationIterator noi = - network()->getNextReadyRequest(); - const RemoteCommandRequest& request = noi->getRequest(); + const NetworkInterfaceMock::NetworkOperationIterator noi = network()->getNextReadyRequest(); + const RemoteCommandRequest& request = noi->getRequest(); - const auto& resultStatus = func(request); + const auto& resultStatus = func(request); - BSONObjBuilder result; + BSONObjBuilder result; - if (resultStatus.isOK()) { - result.appendElements(resultStatus.getValue()); - } + if (resultStatus.isOK()) { + result.appendElements(resultStatus.getValue()); + } - Command::appendCommandStatus(result, resultStatus.getStatus()); + Command::appendCommandStatus(result, resultStatus.getStatus()); - const RemoteCommandResponse response(result.obj(), Milliseconds(1)); + const RemoteCommandResponse response(result.obj(), Milliseconds(1)); - network()->scheduleResponse(noi, network()->now(), response); + network()->scheduleResponse(noi, network()->now(), response); - network()->runReadyNetworkOperations(); + network()->runReadyNetworkOperations(); - network()->exitNetwork(); - } + network()->exitNetwork(); +} - void CatalogManagerReplSetTestFixture::onFindCommand(OnFindCommandFunction func) { - onCommand([&func](const RemoteCommandRequest& request) -> StatusWith<BSONObj> { +void CatalogManagerReplSetTestFixture::onFindCommand(OnFindCommandFunction func) { + onCommand([&func](const RemoteCommandRequest& request) -> StatusWith<BSONObj> { - const auto& resultStatus = func(request); + const auto& resultStatus = func(request); - if (!resultStatus.isOK()) { - return resultStatus.getStatus(); - } + if (!resultStatus.isOK()) { + return resultStatus.getStatus(); + } - BSONArrayBuilder arr; - for (const auto& obj : resultStatus.getValue()) { - arr.append(obj); - } + BSONArrayBuilder arr; + for (const auto& obj : resultStatus.getValue()) { + arr.append(obj); + } - const NamespaceString nss = NamespaceString(request.dbname, - request.cmdObj.firstElement().String()); - BSONObjBuilder result; - appendCursorResponseObject(0LL, nss.toString(), arr.arr(), &result); + const NamespaceString nss = + NamespaceString(request.dbname, request.cmdObj.firstElement().String()); + BSONObjBuilder result; + appendCursorResponseObject(0LL, nss.toString(), arr.arr(), &result); - return result.obj(); - }); - } + return result.obj(); + }); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/replset/catalog_manager_replica_set_test_fixture.h b/src/mongo/s/catalog/replset/catalog_manager_replica_set_test_fixture.h index 05a5dcf4e1f..9331f44ad4f 100644 --- a/src/mongo/s/catalog/replset/catalog_manager_replica_set_test_fixture.h +++ b/src/mongo/s/catalog/replset/catalog_manager_replica_set_test_fixture.h @@ -36,74 +36,74 @@ namespace mongo { - class BSONObj; - class CatalogManagerReplicaSet; - class DistLockManagerMock; - struct RemoteCommandRequest; - class RemoteCommandRunnerMock; - class ShardRegistry; - template<typename T> class StatusWith; +class BSONObj; +class CatalogManagerReplicaSet; +class DistLockManagerMock; +struct RemoteCommandRequest; +class RemoteCommandRunnerMock; +class ShardRegistry; +template <typename T> +class StatusWith; namespace executor { - class NetworkInterfaceMock; +class NetworkInterfaceMock; -} // namespace executor +} // namespace executor +/** + * Sets up the mocked out objects for testing the replica-set backed catalog manager. + */ +class CatalogManagerReplSetTestFixture : public mongo::unittest::Test { +public: + CatalogManagerReplSetTestFixture(); + ~CatalogManagerReplSetTestFixture(); + +protected: /** - * Sets up the mocked out objects for testing the replica-set backed catalog manager. + * Shortcut function to be used for generating mock responses to network requests. + * + * @param dbName Name of the database for which this request came. + * @param cmdObj Contents of the request. + * + * Return the BSON object representing the response(s) or an error, which will be passed + * back on the network. */ - class CatalogManagerReplSetTestFixture : public mongo::unittest::Test { - public: - CatalogManagerReplSetTestFixture(); - ~CatalogManagerReplSetTestFixture(); + using OnCommandFunction = std::function<StatusWith<BSONObj>(const RemoteCommandRequest&)>; - protected: - /** - * Shortcut function to be used for generating mock responses to network requests. - * - * @param dbName Name of the database for which this request came. - * @param cmdObj Contents of the request. - * - * Return the BSON object representing the response(s) or an error, which will be passed - * back on the network. - */ - using OnCommandFunction = - std::function<StatusWith<BSONObj>(const RemoteCommandRequest&)>; + using OnFindCommandFunction = + std::function<StatusWith<std::vector<BSONObj>>(const RemoteCommandRequest&)>; - using OnFindCommandFunction = - std::function<StatusWith<std::vector<BSONObj>>(const RemoteCommandRequest&)>; + CatalogManagerReplicaSet* catalogManager() const; - CatalogManagerReplicaSet* catalogManager() const; + ShardRegistry* shardRegistry() const; - ShardRegistry* shardRegistry() const; + RemoteCommandRunnerMock* commandRunner() const; - RemoteCommandRunnerMock* commandRunner() const; + executor::NetworkInterfaceMock* network() const; - executor::NetworkInterfaceMock* network() const; + DistLockManagerMock* distLock() const; - DistLockManagerMock* distLock() const; - - /** - * Blocking methods, which receive one message from the network and respond using the - * responses returned from the input function. This is a syntactic sugar for simple, - * single request + response or find tests. - */ - void onCommand(OnCommandFunction func); - void onFindCommand(OnFindCommandFunction func); + /** + * Blocking methods, which receive one message from the network and respond using the + * responses returned from the input function. This is a syntactic sugar for simple, + * single request + response or find tests. + */ + void onCommand(OnCommandFunction func); + void onFindCommand(OnFindCommandFunction func); - private: - void setUp() override; +private: + void setUp() override; - void tearDown() override; + void tearDown() override; - // Mocked out network under the task executor. This pointer is owned by the executor on - // the ShardRegistry, so it must not be accessed once the executor has been shut down. - executor::NetworkInterfaceMock* _mockNetwork; + // Mocked out network under the task executor. This pointer is owned by the executor on + // the ShardRegistry, so it must not be accessed once the executor has been shut down. + executor::NetworkInterfaceMock* _mockNetwork; - // Thread used to execute the task executor's loop. This thread will be busy until the - // shutdown is called on the shard registry's task executor. - std::thread _executorThread; - }; + // Thread used to execute the task executor's loop. This thread will be busy until the + // shutdown is called on the shard registry's task executor. + std::thread _executorThread; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/replset/replset_dist_lock_manager.cpp b/src/mongo/s/catalog/replset/replset_dist_lock_manager.cpp index 5d38ea337cd..ef329ded1eb 100644 --- a/src/mongo/s/catalog/replset/replset_dist_lock_manager.cpp +++ b/src/mongo/s/catalog/replset/replset_dist_lock_manager.cpp @@ -48,342 +48,321 @@ namespace mongo { - using std::string; - using std::unique_ptr; - using stdx::chrono::milliseconds; - using std::chrono::duration_cast; - - ReplSetDistLockManager::ReplSetDistLockManager(ServiceContext* globalContext, - StringData processID, - unique_ptr<DistLockCatalog> catalog, - milliseconds pingInterval, - milliseconds lockExpiration): - _serviceContext(globalContext), - _processID(processID.toString()), - _catalog(std::move(catalog)), - _pingInterval(pingInterval), - _lockExpiration(lockExpiration) { +using std::string; +using std::unique_ptr; +using stdx::chrono::milliseconds; +using std::chrono::duration_cast; + +ReplSetDistLockManager::ReplSetDistLockManager(ServiceContext* globalContext, + StringData processID, + unique_ptr<DistLockCatalog> catalog, + milliseconds pingInterval, + milliseconds lockExpiration) + : _serviceContext(globalContext), + _processID(processID.toString()), + _catalog(std::move(catalog)), + _pingInterval(pingInterval), + _lockExpiration(lockExpiration) {} + +ReplSetDistLockManager::~ReplSetDistLockManager() = default; + +void ReplSetDistLockManager::startUp() { + _execThread = stdx::make_unique<stdx::thread>(&ReplSetDistLockManager::doTask, this); +} + +void ReplSetDistLockManager::shutDown() { + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _isShutDown = true; + _shutDownCV.notify_all(); } - ReplSetDistLockManager::~ReplSetDistLockManager() = default; + // Don't grab _mutex, otherwise will deadlock trying to join. Safe to read + // _execThread since it is modified only at statrUp(). + if (_execThread && _execThread->joinable()) { + _execThread->join(); + _execThread.reset(); + } - void ReplSetDistLockManager::startUp() { - _execThread = stdx::make_unique<stdx::thread>(&ReplSetDistLockManager::doTask, this); + auto status = _catalog->stopPing(_processID); + if (!status.isOK()) { + warning() << "error encountered while cleaning up distributed ping entry for " << _processID + << causedBy(status); } +} - void ReplSetDistLockManager::shutDown() { - { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _isShutDown = true; - _shutDownCV.notify_all(); - } +bool ReplSetDistLockManager::isShutDown() { + stdx::lock_guard<stdx::mutex> lk(_mutex); + return _isShutDown; +} - // Don't grab _mutex, otherwise will deadlock trying to join. Safe to read - // _execThread since it is modified only at statrUp(). - if (_execThread && _execThread->joinable()) { - _execThread->join(); - _execThread.reset(); - } +void ReplSetDistLockManager::doTask() { + LOG(0) << "creating distributed lock ping thread for process " << _processID + << " (sleeping for " << duration_cast<milliseconds>(_pingInterval).count() << " ms)"; + + Timer elapsedSincelastPing(_serviceContext->getTickSource()); + while (!isShutDown()) { + auto pingStatus = _catalog->ping(_processID, Date_t::now()); - auto status = _catalog->stopPing(_processID); - if (!status.isOK()) { - warning() << "error encountered while cleaning up distributed ping entry for " - << _processID << causedBy(status); + if (!pingStatus.isOK()) { + warning() << "pinging failed for distributed lock pinger" << causedBy(pingStatus); } - } - bool ReplSetDistLockManager::isShutDown() { - stdx::lock_guard<stdx::mutex> lk(_mutex); - return _isShutDown; - } + const milliseconds elapsed(elapsedSincelastPing.millis()); + if (elapsed > 10 * _pingInterval) { + warning() << "Lock pinger for proc: " << _processID << " was inactive for " << elapsed + << " ms"; + } + elapsedSincelastPing.reset(); - void ReplSetDistLockManager::doTask() { - LOG(0) << "creating distributed lock ping thread for process " << _processID - << " (sleeping for " - << duration_cast<milliseconds>(_pingInterval).count() << " ms)"; + std::deque<DistLockHandle> toUnlockBatch; + { + stdx::unique_lock<stdx::mutex> lk(_mutex); + toUnlockBatch.swap(_unlockList); + } - Timer elapsedSincelastPing(_serviceContext->getTickSource()); - while (!isShutDown()) { - auto pingStatus = _catalog->ping(_processID, Date_t::now()); + for (const auto& toUnlock : toUnlockBatch) { + auto unlockStatus = _catalog->unlock(toUnlock); - if (!pingStatus.isOK()) { - warning() << "pinging failed for distributed lock pinger" << causedBy(pingStatus); + if (!unlockStatus.isOK()) { + warning() << "Failed to unlock lock with " << LocksType::lockID() << ": " + << toUnlock << causedBy(unlockStatus); + queueUnlock(toUnlock); + } else { + LOG(0) << "distributed lock with " << LocksType::lockID() << ": " << toUnlock + << "' unlocked."; } - const milliseconds elapsed(elapsedSincelastPing.millis()); - if (elapsed > 10 * _pingInterval) { - warning() << "Lock pinger for proc: " << _processID - << " was inactive for " << elapsed << " ms"; + if (isShutDown()) { + return; } - elapsedSincelastPing.reset(); + } - std::deque<DistLockHandle> toUnlockBatch; - { - stdx::unique_lock<stdx::mutex> lk(_mutex); - toUnlockBatch.swap(_unlockList); - } + stdx::unique_lock<stdx::mutex> lk(_mutex); + _shutDownCV.wait_for(lk, _pingInterval); + } +} - for (const auto& toUnlock : toUnlockBatch) { - auto unlockStatus = _catalog->unlock(toUnlock); +StatusWith<bool> ReplSetDistLockManager::canOvertakeLock(LocksType lockDoc) { + const auto& processID = lockDoc.getProcess(); + auto pingStatus = _catalog->getPing(processID); + if (!pingStatus.isOK()) { + return pingStatus.getStatus(); + } - if (!unlockStatus.isOK()) { - warning() << "Failed to unlock lock with " << LocksType::lockID() - << ": " << toUnlock << causedBy(unlockStatus); - queueUnlock(toUnlock); - } - else { - LOG(0) << "distributed lock with " << LocksType::lockID() - << ": " << toUnlock << "' unlocked."; - } + const auto& pingDoc = pingStatus.getValue(); + string errMsg; + if (!pingDoc.isValid(&errMsg)) { + return {ErrorCodes::UnsupportedFormat, + str::stream() << "invalid ping document for " << processID << ": " << errMsg}; + } - if (isShutDown()) { - return; - } - } + Timer timer(_serviceContext->getTickSource()); + auto serverInfoStatus = _catalog->getServerInfo(); + if (!serverInfoStatus.isOK()) { + return serverInfoStatus.getStatus(); + } - stdx::unique_lock<stdx::mutex> lk(_mutex); - _shutDownCV.wait_for(lk, _pingInterval); - } + // Be conservative when determining that lock expiration has elapsed by + // taking into account the roundtrip delay of trying to get the local + // time from the config server. + milliseconds delay(timer.millis() / 2); // Assuming symmetrical delay. + + Date_t pingValue = pingDoc.getPing(); + const auto& serverInfo = serverInfoStatus.getValue(); + + stdx::lock_guard<stdx::mutex> lk(_mutex); + auto pingIter = _pingHistory.find(lockDoc.getName()); + + if (pingIter == _pingHistory.end()) { + // We haven't seen this lock before so we don't have any point of reference + // to compare and determine the elapsed time. Save the current ping info + // for this lock. + _pingHistory.emplace(std::piecewise_construct, + std::forward_as_tuple(lockDoc.getName()), + std::forward_as_tuple(processID, + pingValue, + serverInfo.serverTime, + lockDoc.getLockID(), + serverInfo.electionId)); + return false; } - StatusWith<bool> ReplSetDistLockManager::canOvertakeLock(LocksType lockDoc) { - const auto& processID = lockDoc.getProcess(); - auto pingStatus = _catalog->getPing(processID); - if (!pingStatus.isOK()) { - return pingStatus.getStatus(); - } + auto configServerLocalTime = serverInfo.serverTime - delay; - const auto& pingDoc = pingStatus.getValue(); - string errMsg; - if (!pingDoc.isValid(&errMsg)) { - return {ErrorCodes::UnsupportedFormat, - str::stream() << "invalid ping document for " << processID - << ": " << errMsg}; - } + auto* pingInfo = &pingIter->second; - Timer timer(_serviceContext->getTickSource()); - auto serverInfoStatus = _catalog->getServerInfo(); - if (!serverInfoStatus.isOK()) { - return serverInfoStatus.getStatus(); - } + LOG(1) << "checking last ping for lock '" << lockDoc.getName() << "' against last seen process " + << pingInfo->processId << " and ping " << pingInfo->lastPing; - // Be conservative when determining that lock expiration has elapsed by - // taking into account the roundtrip delay of trying to get the local - // time from the config server. - milliseconds delay(timer.millis() / 2); // Assuming symmetrical delay. + if (pingInfo->lastPing != pingValue || // ping is active - Date_t pingValue = pingDoc.getPing(); - const auto& serverInfo = serverInfoStatus.getValue(); + // Owner of this lock is now different from last time so we can't + // use the ping data. + pingInfo->lockSessionId != lockDoc.getLockID() || - stdx::lock_guard<stdx::mutex> lk(_mutex); - auto pingIter = _pingHistory.find(lockDoc.getName()); - - if (pingIter == _pingHistory.end()) { - // We haven't seen this lock before so we don't have any point of reference - // to compare and determine the elapsed time. Save the current ping info - // for this lock. - _pingHistory.emplace(std::piecewise_construct, - std::forward_as_tuple(lockDoc.getName()), - std::forward_as_tuple(processID, - pingValue, - serverInfo.serverTime, - lockDoc.getLockID(), - serverInfo.electionId)); - return false; - } + // Primary changed, we can't trust that clocks are synchronized so + // treat as if this is a new entry. + pingInfo->electionId != serverInfo.electionId) { + pingInfo->lastPing = pingValue; + pingInfo->electionId = serverInfo.electionId; + pingInfo->configLocalTime = configServerLocalTime; + pingInfo->lockSessionId = lockDoc.getLockID(); + return false; + } + + if (configServerLocalTime < pingInfo->configLocalTime) { + warning() << "config server local time went backwards, from last seen: " + << pingInfo->configLocalTime << " to " << configServerLocalTime; + return false; + } - auto configServerLocalTime = serverInfo.serverTime - delay; + milliseconds elapsedSinceLastPing(configServerLocalTime - pingInfo->configLocalTime); + if (elapsedSinceLastPing >= _lockExpiration) { + LOG(0) << "forcing lock '" << lockDoc.getName() << "' because elapsed time " + << duration_cast<milliseconds>(elapsedSinceLastPing).count() + << " ms >= takeover time " << duration_cast<milliseconds>(_lockExpiration).count() + << " ms"; + return true; + } - auto* pingInfo = &pingIter->second; + LOG(1) << "could not force lock '" << lockDoc.getName() << "' because elapsed time " + << duration_cast<milliseconds>(elapsedSinceLastPing).count() << " ms < takeover time " + << duration_cast<milliseconds>(_lockExpiration).count() << " ms"; + return false; +} - LOG(1) << "checking last ping for lock '" << lockDoc.getName() - << "' against last seen process " << pingInfo->processId - << " and ping " << pingInfo->lastPing; +StatusWith<DistLockManager::ScopedDistLock> ReplSetDistLockManager::lock( + StringData name, StringData whyMessage, milliseconds waitFor, milliseconds lockTryInterval) { + Timer timer(_serviceContext->getTickSource()); + Timer msgTimer(_serviceContext->getTickSource()); - if (pingInfo->lastPing != pingValue || // ping is active + while (waitFor <= milliseconds::zero() || milliseconds(timer.millis()) < waitFor) { + OID lockSessionID = OID::gen(); + string who = str::stream() << _processID << ":" << getThreadName(); - // Owner of this lock is now different from last time so we can't - // use the ping data. - pingInfo->lockSessionId != lockDoc.getLockID() || + LOG(1) << "trying to acquire new distributed lock for " << name + << " ( lock timeout : " << duration_cast<milliseconds>(_lockExpiration).count() + << " ms, ping interval : " << duration_cast<milliseconds>(_pingInterval).count() + << " ms, process : " << _processID << " )" + << " with lockSessionID: " << lockSessionID << ", why: " << whyMessage; - // Primary changed, we can't trust that clocks are synchronized so - // treat as if this is a new entry. - pingInfo->electionId != serverInfo.electionId) { - pingInfo->lastPing = pingValue; - pingInfo->electionId = serverInfo.electionId; - pingInfo->configLocalTime = configServerLocalTime; - pingInfo->lockSessionId = lockDoc.getLockID(); - return false; - } + auto lockResult = + _catalog->grabLock(name, lockSessionID, who, _processID, Date_t::now(), whyMessage); + + auto status = lockResult.getStatus(); - if (configServerLocalTime < pingInfo->configLocalTime) { - warning() << "config server local time went backwards, from last seen: " - << pingInfo->configLocalTime - << " to " << configServerLocalTime; - return false; + if (status.isOK()) { + // Lock is acquired since findAndModify was able to successfully modify + // the lock document. + LOG(0) << "distributed lock '" << name << "' acquired, ts : " << lockSessionID; + return ScopedDistLock(lockSessionID, this); } - milliseconds elapsedSinceLastPing(configServerLocalTime - pingInfo->configLocalTime); - if (elapsedSinceLastPing >= _lockExpiration) { - LOG(0) << "forcing lock '" << lockDoc.getName() - << "' because elapsed time " - << duration_cast<milliseconds>(elapsedSinceLastPing).count() - << " ms >= takeover time " - << duration_cast<milliseconds>(_lockExpiration).count() << " ms"; - return true; + if (status != ErrorCodes::LockStateChangeFailed) { + // An error occurred but the write might have actually been applied on the + // other side. Schedule an unlock to clean it up just in case. + queueUnlock(lockSessionID); + return status; } - LOG(1) << "could not force lock '" << lockDoc.getName() - << "' because elapsed time " - << duration_cast<milliseconds>(elapsedSinceLastPing).count() - << " ms < takeover time " - << duration_cast<milliseconds>(_lockExpiration).count() << " ms"; - return false; - } + // Get info from current lock and check if we can overtake it. + auto getLockStatusResult = _catalog->getLockByName(name); + const auto& getLockStatus = getLockStatusResult.getStatus(); - StatusWith<DistLockManager::ScopedDistLock> ReplSetDistLockManager::lock( - StringData name, - StringData whyMessage, - milliseconds waitFor, - milliseconds lockTryInterval) { - Timer timer(_serviceContext->getTickSource()); - Timer msgTimer(_serviceContext->getTickSource()); - - while (waitFor <= milliseconds::zero() || milliseconds(timer.millis()) < waitFor) { - OID lockSessionID = OID::gen(); - string who = str::stream() << _processID << ":" << getThreadName(); - - LOG(1) << "trying to acquire new distributed lock for " << name - << " ( lock timeout : " - << duration_cast<milliseconds>(_lockExpiration).count() - << " ms, ping interval : " - << duration_cast<milliseconds>(_pingInterval).count() - << " ms, process : " << _processID << " )" - << " with lockSessionID: " << lockSessionID - << ", why: " << whyMessage; - - auto lockResult = _catalog->grabLock(name, - lockSessionID, - who, - _processID, - Date_t::now(), - whyMessage); - - auto status = lockResult.getStatus(); - - if (status.isOK()) { - // Lock is acquired since findAndModify was able to successfully modify - // the lock document. - LOG(0) << "distributed lock '" << name << "' acquired, ts : " << lockSessionID; - return ScopedDistLock(lockSessionID, this); - } + if (!getLockStatusResult.isOK() && getLockStatus != ErrorCodes::LockNotFound) { + return getLockStatus; + } + + // Note: Only attempt to overtake locks that actually exists. If lock was not + // found, use the normal grab lock path to acquire it. + if (getLockStatusResult.isOK()) { + auto currentLock = getLockStatusResult.getValue(); + auto canOvertakeResult = canOvertakeLock(currentLock); - if (status != ErrorCodes::LockStateChangeFailed) { - // An error occurred but the write might have actually been applied on the - // other side. Schedule an unlock to clean it up just in case. - queueUnlock(lockSessionID); - return status; + if (!canOvertakeResult.isOK()) { + return canOvertakeResult.getStatus(); } - // Get info from current lock and check if we can overtake it. - auto getLockStatusResult = _catalog->getLockByName(name); - const auto& getLockStatus = getLockStatusResult.getStatus(); + if (canOvertakeResult.getValue()) { + auto overtakeResult = _catalog->overtakeLock(name, + lockSessionID, + currentLock.getLockID(), + who, + _processID, + Date_t::now(), + whyMessage); - if (!getLockStatusResult.isOK() && getLockStatus != ErrorCodes::LockNotFound) { - return getLockStatus; - } + const auto& overtakeStatus = overtakeResult.getStatus(); - // Note: Only attempt to overtake locks that actually exists. If lock was not - // found, use the normal grab lock path to acquire it. - if (getLockStatusResult.isOK()) { - auto currentLock = getLockStatusResult.getValue(); - auto canOvertakeResult = canOvertakeLock(currentLock); + if (overtakeResult.isOK()) { + // Lock is acquired since findAndModify was able to successfully modify + // the lock document. - if (!canOvertakeResult.isOK()) { - return canOvertakeResult.getStatus(); + LOG(0) << "lock '" << name << "' successfully forced"; + LOG(0) << "distributed lock '" << name << "' acquired, ts : " << lockSessionID; + return ScopedDistLock(lockSessionID, this); } - if (canOvertakeResult.getValue()) { - auto overtakeResult = _catalog->overtakeLock(name, - lockSessionID, - currentLock.getLockID(), - who, - _processID, - Date_t::now(), - whyMessage); - - const auto& overtakeStatus = overtakeResult.getStatus(); - - if (overtakeResult.isOK()) { - // Lock is acquired since findAndModify was able to successfully modify - // the lock document. - - LOG(0) << "lock '" << name << "' successfully forced"; - LOG(0) << "distributed lock '" << name - << "' acquired, ts : " << lockSessionID; - return ScopedDistLock(lockSessionID, this); - } - - if (overtakeStatus != ErrorCodes::LockStateChangeFailed) { - // An error occurred but the write might have actually been applied on the - // other side. Schedule an unlock to clean it up just in case. - queueUnlock(lockSessionID); - return overtakeStatus; - } + if (overtakeStatus != ErrorCodes::LockStateChangeFailed) { + // An error occurred but the write might have actually been applied on the + // other side. Schedule an unlock to clean it up just in case. + queueUnlock(lockSessionID); + return overtakeStatus; } } + } - LOG(1) << "distributed lock '" << name << "' was not acquired."; - - if (waitFor == milliseconds::zero()) { - break; - } + LOG(1) << "distributed lock '" << name << "' was not acquired."; - // Periodically message for debugging reasons - if (msgTimer.seconds() > 10) { - LOG(0) << "waited " << timer.seconds() << "s for distributed lock " << name - << " for " << whyMessage; + if (waitFor == milliseconds::zero()) { + break; + } - msgTimer.reset(); - } + // Periodically message for debugging reasons + if (msgTimer.seconds() > 10) { + LOG(0) << "waited " << timer.seconds() << "s for distributed lock " << name << " for " + << whyMessage; - milliseconds timeRemaining = - std::max(milliseconds::zero(), waitFor - milliseconds(timer.millis())); - sleepFor(std::min(lockTryInterval, timeRemaining)); + msgTimer.reset(); } - return {ErrorCodes::LockBusy, str::stream() << "timed out waiting for " << name}; + milliseconds timeRemaining = + std::max(milliseconds::zero(), waitFor - milliseconds(timer.millis())); + sleepFor(std::min(lockTryInterval, timeRemaining)); } - void ReplSetDistLockManager::unlock(const DistLockHandle& lockSessionID) { - auto unlockStatus = _catalog->unlock(lockSessionID); - - if (!unlockStatus.isOK()) { - queueUnlock(lockSessionID); - } - else { - LOG(0) << "distributed lock with " << LocksType::lockID() - << ": " << lockSessionID << "' unlocked."; - } - } + return {ErrorCodes::LockBusy, str::stream() << "timed out waiting for " << name}; +} - Status ReplSetDistLockManager::checkStatus(const DistLockHandle& lockHandle) { - auto lockStatus = _catalog->getLockByTS(lockHandle); +void ReplSetDistLockManager::unlock(const DistLockHandle& lockSessionID) { + auto unlockStatus = _catalog->unlock(lockSessionID); - if (!lockStatus.isOK()) { - return lockStatus.getStatus(); - } + if (!unlockStatus.isOK()) { + queueUnlock(lockSessionID); + } else { + LOG(0) << "distributed lock with " << LocksType::lockID() << ": " << lockSessionID + << "' unlocked."; + } +} - auto lockDoc = lockStatus.getValue(); - if (!lockDoc.isValid(nullptr)) { - return {ErrorCodes::LockNotFound, "lock owner changed"}; - } +Status ReplSetDistLockManager::checkStatus(const DistLockHandle& lockHandle) { + auto lockStatus = _catalog->getLockByTS(lockHandle); - return Status::OK(); + if (!lockStatus.isOK()) { + return lockStatus.getStatus(); } - void ReplSetDistLockManager::queueUnlock(const DistLockHandle& lockSessionID) { - stdx::unique_lock<stdx::mutex> lk(_mutex); - _unlockList.push_back(lockSessionID); + auto lockDoc = lockStatus.getValue(); + if (!lockDoc.isValid(nullptr)) { + return {ErrorCodes::LockNotFound, "lock owner changed"}; } + + return Status::OK(); +} + +void ReplSetDistLockManager::queueUnlock(const DistLockHandle& lockSessionID) { + stdx::unique_lock<stdx::mutex> lk(_mutex); + _unlockList.push_back(lockSessionID); +} } diff --git a/src/mongo/s/catalog/replset/replset_dist_lock_manager.h b/src/mongo/s/catalog/replset/replset_dist_lock_manager.h index 4e042e2fa41..1779925b8ec 100644 --- a/src/mongo/s/catalog/replset/replset_dist_lock_manager.h +++ b/src/mongo/s/catalog/replset/replset_dist_lock_manager.h @@ -43,88 +43,86 @@ namespace mongo { - class ServiceContext; - - class ReplSetDistLockManager final : public DistLockManager { - public: - ReplSetDistLockManager(ServiceContext* globalContext, - StringData processID, - std::unique_ptr<DistLockCatalog> catalog, - stdx::chrono::milliseconds pingInterval, - stdx::chrono::milliseconds lockExpiration); - - virtual ~ReplSetDistLockManager(); - - virtual void startUp() override; - virtual void shutDown() override; - - virtual StatusWith<DistLockManager::ScopedDistLock> lock( - StringData name, - StringData whyMessage, - stdx::chrono::milliseconds waitFor, - stdx::chrono::milliseconds lockTryInterval) override; - - protected: - - virtual void unlock(const DistLockHandle& lockSessionID) override; - - virtual Status checkStatus(const DistLockHandle& lockSessionID) override; - - private: - - /** - * Queue a lock to be unlocked asynchronously with retry until it doesn't error. - */ - void queueUnlock(const DistLockHandle& lockSessionID); - - /** - * Periodically pings and checks if there are locks queued that needs unlocking. - */ - void doTask(); - - /** - * Returns true if shutDown was called. - */ - bool isShutDown(); - - /** - * Returns true if the current process that owns the lock has no fresh pings since - * the lock expiration threshold. - */ - StatusWith<bool> canOvertakeLock(const LocksType lockDoc); - - // - // All member variables are labeled with one of the following codes indicating the - // synchronization rules for accessing them. - // - // (F) Self synchronizing. - // (M) Must hold _mutex for access. - // (I) Immutable, no synchronization needed. - // (S) Can only be called inside startUp/shutDown. - // - - ServiceContext* const _serviceContext; // (F) - - const std::string _processID; // (I) - const std::unique_ptr<DistLockCatalog> _catalog; // (I) - const stdx::chrono::milliseconds _pingInterval; // (I) - const stdx::chrono::milliseconds _lockExpiration; // (I) - - stdx::mutex _mutex; - std::unique_ptr<stdx::thread> _execThread; // (S) - - // Contains the list of locks queued for unlocking. Cases when unlock operation can - // be queued include: - // 1. First attempt on unlocking resulted in an error. - // 2. Attempting to grab or overtake a lock resulted in an error where we are uncertain - // whether the modification was actually applied or not, and call unlock to make - // sure that it was cleaned up. - std::deque<DistLockHandle> _unlockList; // (M) - - bool _isShutDown = false; // (M) - stdx::condition_variable _shutDownCV; // (M) - - // Map of lockName to last ping information. - std::unordered_map<std::string, DistLockPingInfo> _pingHistory; // (M) - }; +class ServiceContext; + +class ReplSetDistLockManager final : public DistLockManager { +public: + ReplSetDistLockManager(ServiceContext* globalContext, + StringData processID, + std::unique_ptr<DistLockCatalog> catalog, + stdx::chrono::milliseconds pingInterval, + stdx::chrono::milliseconds lockExpiration); + + virtual ~ReplSetDistLockManager(); + + virtual void startUp() override; + virtual void shutDown() override; + + virtual StatusWith<DistLockManager::ScopedDistLock> lock( + StringData name, + StringData whyMessage, + stdx::chrono::milliseconds waitFor, + stdx::chrono::milliseconds lockTryInterval) override; + +protected: + virtual void unlock(const DistLockHandle& lockSessionID) override; + + virtual Status checkStatus(const DistLockHandle& lockSessionID) override; + +private: + /** + * Queue a lock to be unlocked asynchronously with retry until it doesn't error. + */ + void queueUnlock(const DistLockHandle& lockSessionID); + + /** + * Periodically pings and checks if there are locks queued that needs unlocking. + */ + void doTask(); + + /** + * Returns true if shutDown was called. + */ + bool isShutDown(); + + /** + * Returns true if the current process that owns the lock has no fresh pings since + * the lock expiration threshold. + */ + StatusWith<bool> canOvertakeLock(const LocksType lockDoc); + + // + // All member variables are labeled with one of the following codes indicating the + // synchronization rules for accessing them. + // + // (F) Self synchronizing. + // (M) Must hold _mutex for access. + // (I) Immutable, no synchronization needed. + // (S) Can only be called inside startUp/shutDown. + // + + ServiceContext* const _serviceContext; // (F) + + const std::string _processID; // (I) + const std::unique_ptr<DistLockCatalog> _catalog; // (I) + const stdx::chrono::milliseconds _pingInterval; // (I) + const stdx::chrono::milliseconds _lockExpiration; // (I) + + stdx::mutex _mutex; + std::unique_ptr<stdx::thread> _execThread; // (S) + + // Contains the list of locks queued for unlocking. Cases when unlock operation can + // be queued include: + // 1. First attempt on unlocking resulted in an error. + // 2. Attempting to grab or overtake a lock resulted in an error where we are uncertain + // whether the modification was actually applied or not, and call unlock to make + // sure that it was cleaned up. + std::deque<DistLockHandle> _unlockList; // (M) + + bool _isShutDown = false; // (M) + stdx::condition_variable _shutDownCV; // (M) + + // Map of lockName to last ping information. + std::unordered_map<std::string, DistLockPingInfo> _pingHistory; // (M) +}; } diff --git a/src/mongo/s/catalog/replset/replset_dist_lock_manager_test.cpp b/src/mongo/s/catalog/replset/replset_dist_lock_manager_test.cpp index 50e05ece2cc..52395a65e62 100644 --- a/src/mongo/s/catalog/replset/replset_dist_lock_manager_test.cpp +++ b/src/mongo/s/catalog/replset/replset_dist_lock_manager_test.cpp @@ -61,138 +61,132 @@ namespace mongo { namespace { - using std::map; - using std::string; - using std::vector; +using std::map; +using std::string; +using std::vector; - const Seconds kUnlockTimeout(30); - const Milliseconds kPingInterval(2); - const Seconds kLockExpiration(10); +const Seconds kUnlockTimeout(30); +const Milliseconds kPingInterval(2); +const Seconds kLockExpiration(10); + +/** + * Basic fixture for ReplSetDistLockManager that starts it up before the test begins + * and shuts it down when a test finishes. + */ +class ReplSetDistLockManagerFixture : public mongo::unittest::Test { +public: + ReplSetDistLockManagerFixture() + : _dummyDoNotUse(stdx::make_unique<DistLockCatalogMock>()), + _mockCatalog(_dummyDoNotUse.get()), + _processID("test"), + _mgr(&_context, _processID, std::move(_dummyDoNotUse), kPingInterval, kLockExpiration) {} /** - * Basic fixture for ReplSetDistLockManager that starts it up before the test begins - * and shuts it down when a test finishes. + * Returns the lock manager instance that is being tested. */ - class ReplSetDistLockManagerFixture: public mongo::unittest::Test { - public: - ReplSetDistLockManagerFixture(): - _dummyDoNotUse(stdx::make_unique<DistLockCatalogMock>()), - _mockCatalog(_dummyDoNotUse.get()), - _processID("test"), - _mgr(&_context, - _processID, - std::move(_dummyDoNotUse), - kPingInterval, - kLockExpiration) { - } - - /** - * Returns the lock manager instance that is being tested. - */ - ReplSetDistLockManager* getMgr() { - return &_mgr; - } + ReplSetDistLockManager* getMgr() { + return &_mgr; + } - /** - * Returns the mocked catalog used by the lock manager being tested. - */ - DistLockCatalogMock* getMockCatalog() { - return _mockCatalog; - } + /** + * Returns the mocked catalog used by the lock manager being tested. + */ + DistLockCatalogMock* getMockCatalog() { + return _mockCatalog; + } - /** - * Get the process id that was initialiezd with the lock manager being tested. - */ - string getProcessID() const { - return _processID; - } + /** + * Get the process id that was initialiezd with the lock manager being tested. + */ + string getProcessID() const { + return _processID; + } - protected: - void setUp() override { - _context.setTickSource(stdx::make_unique<SystemTickSource>()); - _mgr.startUp(); - } +protected: + void setUp() override { + _context.setTickSource(stdx::make_unique<SystemTickSource>()); + _mgr.startUp(); + } - void tearDown() override { - // Don't care about what shutDown passes to stopPing here. - _mockCatalog->expectStopPing([](StringData){}, Status::OK()); - _mgr.shutDown(); - } + void tearDown() override { + // Don't care about what shutDown passes to stopPing here. + _mockCatalog->expectStopPing([](StringData) {}, Status::OK()); + _mgr.shutDown(); + } - TickSourceMock _tickSource; - std::unique_ptr<DistLockCatalogMock> _dummyDoNotUse; // dummy placeholder - DistLockCatalogMock* _mockCatalog; - string _processID; - ServiceContextNoop _context; - ReplSetDistLockManager _mgr; - }; + TickSourceMock _tickSource; + std::unique_ptr<DistLockCatalogMock> _dummyDoNotUse; // dummy placeholder + DistLockCatalogMock* _mockCatalog; + string _processID; + ServiceContextNoop _context; + ReplSetDistLockManager _mgr; +}; - class RSDistLockMgrWithMockTickSource: public ReplSetDistLockManagerFixture { - public: - /** - * Returns the mock tick source. - */ - TickSourceMock* getMockTickSource() { - return dynamic_cast<TickSourceMock*>(_context.getTickSource()); - } +class RSDistLockMgrWithMockTickSource : public ReplSetDistLockManagerFixture { +public: + /** + * Returns the mock tick source. + */ + TickSourceMock* getMockTickSource() { + return dynamic_cast<TickSourceMock*>(_context.getTickSource()); + } - protected: - void setUp() override { - _context.setTickSource(stdx::make_unique<TickSourceMock>()); - _mgr.startUp(); - } - }; +protected: + void setUp() override { + _context.setTickSource(stdx::make_unique<TickSourceMock>()); + _mgr.startUp(); + } +}; - std::string mapToString(const std::map<OID, int>& map) { - StringBuilder str; +std::string mapToString(const std::map<OID, int>& map) { + StringBuilder str; - for (const auto& entry : map) { - str << "(" << entry.first.toString() << ": " << entry.second << ")"; - } + for (const auto& entry : map) { + str << "(" << entry.first.toString() << ": " << entry.second << ")"; + } - return str.str(); - }; + return str.str(); +}; - std::string vectorToString(const std::vector<OID>& list) { - StringBuilder str; +std::string vectorToString(const std::vector<OID>& list) { + StringBuilder str; - for (const auto& entry : list) { - str << "(" << entry.toString() << ")"; - } + for (const auto& entry : list) { + str << "(" << entry.toString() << ")"; + } - return str.str(); - }; + return str.str(); +}; - /** - * Test scenario: - * 1. Grab lock. - * 2. Unlock (on destructor of ScopedDistLock). - * 3. Check lock id used in lock and unlock are the same. - */ - TEST_F(ReplSetDistLockManagerFixture, BasicLockLifeCycle) { - string lockName("test"); - Date_t now(Date_t::now()); - string whyMsg("because"); - - LocksType retLockDoc; - retLockDoc.setName(lockName); - retLockDoc.setState(LocksType::LOCKED); - retLockDoc.setProcess(getProcessID()); - retLockDoc.setWho("me"); - retLockDoc.setWhy(whyMsg); - // Will be different from the actual lock session id. For testing only. - retLockDoc.setLockID(OID::gen()); - - OID lockSessionIDPassed; - - getMockCatalog()->expectGrabLock( - [this, &lockName, &now, &whyMsg, &lockSessionIDPassed]( - StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why) { +/** + * Test scenario: + * 1. Grab lock. + * 2. Unlock (on destructor of ScopedDistLock). + * 3. Check lock id used in lock and unlock are the same. + */ +TEST_F(ReplSetDistLockManagerFixture, BasicLockLifeCycle) { + string lockName("test"); + Date_t now(Date_t::now()); + string whyMsg("because"); + + LocksType retLockDoc; + retLockDoc.setName(lockName); + retLockDoc.setState(LocksType::LOCKED); + retLockDoc.setProcess(getProcessID()); + retLockDoc.setWho("me"); + retLockDoc.setWhy(whyMsg); + // Will be different from the actual lock session id. For testing only. + retLockDoc.setLockID(OID::gen()); + + OID lockSessionIDPassed; + + getMockCatalog()->expectGrabLock( + [this, &lockName, &now, &whyMsg, &lockSessionIDPassed](StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS(lockName, lockID); ASSERT_TRUE(lockSessionID.isSet()); ASSERT_EQUALS(getProcessID(), processId); @@ -200,72 +194,73 @@ namespace { ASSERT_EQUALS(whyMsg, why); lockSessionIDPassed = lockSessionID; - getMockCatalog()->expectNoGrabLock(); // Call only once. - }, retLockDoc); + getMockCatalog()->expectNoGrabLock(); // Call only once. + }, + retLockDoc); - int unlockCallCount = 0; - OID unlockSessionIDPassed; + int unlockCallCount = 0; + OID unlockSessionIDPassed; - { - auto lockStatus = getMgr()->lock(lockName, - whyMsg, - DistLockManager::kDefaultSingleLockAttemptTimeout, - DistLockManager::kDefaultLockRetryInterval); - ASSERT_OK(lockStatus.getStatus()); + { + auto lockStatus = getMgr()->lock(lockName, + whyMsg, + DistLockManager::kDefaultSingleLockAttemptTimeout, + DistLockManager::kDefaultLockRetryInterval); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock( - [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { unlockCallCount++; unlockSessionIDPassed = lockSessionID; - }, Status::OK()); - } - - ASSERT_EQUALS(1, unlockCallCount); - ASSERT_EQUALS(lockSessionIDPassed, unlockSessionIDPassed); + }, + Status::OK()); } - /** - * Test scenario: - * 1. Grab lock fails up to 3 times. - * 2. Check that each attempt uses a unique lock session id. - * 3. Unlock (on destructor of ScopedDistLock). - * 4. Check lock id used in lock and unlock are the same. - */ - TEST_F(RSDistLockMgrWithMockTickSource, LockSuccessAfterRetry) { - string lockName("test"); - string me("me"); - OID lastTS; - Date_t lastTime(Date_t::now()); - string whyMsg("because"); - - int retryAttempt = 0; - const int kMaxRetryAttempt = 3; - - LocksType goodLockDoc; - goodLockDoc.setName(lockName); - goodLockDoc.setState(LocksType::LOCKED); - goodLockDoc.setProcess(getProcessID()); - goodLockDoc.setWho("me"); - goodLockDoc.setWhy(whyMsg); - goodLockDoc.setLockID(OID::gen()); - - getMockCatalog()->expectGrabLock( - [this, - &lockName, - &lastTS, - &me, - &lastTime, - &whyMsg, - &retryAttempt, - &kMaxRetryAttempt, - &goodLockDoc]( - StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why) { + ASSERT_EQUALS(1, unlockCallCount); + ASSERT_EQUALS(lockSessionIDPassed, unlockSessionIDPassed); +} + +/** + * Test scenario: + * 1. Grab lock fails up to 3 times. + * 2. Check that each attempt uses a unique lock session id. + * 3. Unlock (on destructor of ScopedDistLock). + * 4. Check lock id used in lock and unlock are the same. + */ +TEST_F(RSDistLockMgrWithMockTickSource, LockSuccessAfterRetry) { + string lockName("test"); + string me("me"); + OID lastTS; + Date_t lastTime(Date_t::now()); + string whyMsg("because"); + + int retryAttempt = 0; + const int kMaxRetryAttempt = 3; + + LocksType goodLockDoc; + goodLockDoc.setName(lockName); + goodLockDoc.setState(LocksType::LOCKED); + goodLockDoc.setProcess(getProcessID()); + goodLockDoc.setWho("me"); + goodLockDoc.setWhy(whyMsg); + goodLockDoc.setLockID(OID::gen()); + + getMockCatalog()->expectGrabLock( + [this, + &lockName, + &lastTS, + &me, + &lastTime, + &whyMsg, + &retryAttempt, + &kMaxRetryAttempt, + &goodLockDoc](StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS(lockName, lockID); // Every attempt should have a unique sesssion ID. ASSERT_NOT_EQUALS(lastTS, lockSessionID); @@ -279,119 +274,107 @@ namespace { getMockTickSource()->advance(Milliseconds(1)); if (++retryAttempt >= kMaxRetryAttempt) { - getMockCatalog()->expectGrabLock([this, - &lockName, - &lastTS, - &me, - &lastTime, - &whyMsg]( - StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why) { - ASSERT_EQUALS(lockName, lockID); - // Every attempt should have a unique sesssion ID. - ASSERT_NOT_EQUALS(lastTS, lockSessionID); - lastTS = lockSessionID; - ASSERT_TRUE(lockSessionID.isSet()); - ASSERT_EQUALS(getProcessID(), processId); - ASSERT_GREATER_THAN_OR_EQUALS(time, lastTime); - ASSERT_EQUALS(whyMsg, why); - - getMockCatalog()->expectNoGrabLock(); - - getMockCatalog()->expectGetLockByName([](StringData name) { - FAIL("should not attempt to overtake lock after successful lock"); - }, LocksType()); - }, goodLockDoc); + getMockCatalog()->expectGrabLock( + [this, &lockName, &lastTS, &me, &lastTime, &whyMsg](StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { + ASSERT_EQUALS(lockName, lockID); + // Every attempt should have a unique sesssion ID. + ASSERT_NOT_EQUALS(lastTS, lockSessionID); + lastTS = lockSessionID; + ASSERT_TRUE(lockSessionID.isSet()); + ASSERT_EQUALS(getProcessID(), processId); + ASSERT_GREATER_THAN_OR_EQUALS(time, lastTime); + ASSERT_EQUALS(whyMsg, why); + + getMockCatalog()->expectNoGrabLock(); + + getMockCatalog()->expectGetLockByName([](StringData name) { + FAIL("should not attempt to overtake lock after successful lock"); + }, LocksType()); + }, + goodLockDoc); } - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - // - // Setup mock for lock overtaking. - // + // + // Setup mock for lock overtaking. + // - LocksType currentLockDoc; - currentLockDoc.setName("test"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID::gen()); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); + LocksType currentLockDoc; + currentLockDoc.setName("test"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID::gen()); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("test", name); - }, currentLockDoc); + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("test", name); }, + currentLockDoc); - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); - // Config server time is fixed, so overtaking will never succeed. - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t(), OID())); + // Config server time is fixed, so overtaking will never succeed. + getMockCatalog()->expectGetServerInfo([]() {}, DistLockCatalog::ServerInfo(Date_t(), OID())); - // - // Try grabbing lock. - // + // + // Try grabbing lock. + // - int unlockCallCount = 0; - OID unlockSessionIDPassed; + int unlockCallCount = 0; + OID unlockSessionIDPassed; - { - auto lockStatus = getMgr()->lock(lockName, whyMsg, Milliseconds(10), Milliseconds(1)); - ASSERT_OK(lockStatus.getStatus()); + { + auto lockStatus = getMgr()->lock(lockName, whyMsg, Milliseconds(10), Milliseconds(1)); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock( - [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { unlockCallCount++; unlockSessionIDPassed = lockSessionID; - }, Status::OK()); - } - - ASSERT_EQUALS(1, unlockCallCount); - ASSERT_EQUALS(lastTS, unlockSessionIDPassed); + }, + Status::OK()); } - /** - * Test scenario: - * 1. Grab lock fails up to 3 times. - * 2. Check that each attempt uses a unique lock session id. - * 3. Grab lock errors out on the fourth try. - * 4. Make sure that unlock is called to cleanup the last lock attempted that error out. - */ - TEST_F(RSDistLockMgrWithMockTickSource, LockFailsAfterRetry) { - string lockName("test"); - string me("me"); - OID lastTS; - Date_t lastTime(Date_t::now()); - string whyMsg("because"); - - int retryAttempt = 0; - const int kMaxRetryAttempt = 3; - - getMockCatalog()->expectGrabLock( - [this, - &lockName, - &lastTS, - &me, - &lastTime, - &whyMsg, - &retryAttempt, - &kMaxRetryAttempt]( - StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why) { + ASSERT_EQUALS(1, unlockCallCount); + ASSERT_EQUALS(lastTS, unlockSessionIDPassed); +} + +/** + * Test scenario: + * 1. Grab lock fails up to 3 times. + * 2. Check that each attempt uses a unique lock session id. + * 3. Grab lock errors out on the fourth try. + * 4. Make sure that unlock is called to cleanup the last lock attempted that error out. + */ +TEST_F(RSDistLockMgrWithMockTickSource, LockFailsAfterRetry) { + string lockName("test"); + string me("me"); + OID lastTS; + Date_t lastTime(Date_t::now()); + string whyMsg("because"); + + int retryAttempt = 0; + const int kMaxRetryAttempt = 3; + + getMockCatalog()->expectGrabLock( + [this, &lockName, &lastTS, &me, &lastTime, &whyMsg, &retryAttempt, &kMaxRetryAttempt]( + StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS(lockName, lockID); // Every attempt should have a unique sesssion ID. ASSERT_NOT_EQUALS(lastTS, lockSessionID); @@ -405,124 +388,116 @@ namespace { getMockTickSource()->advance(Milliseconds(1)); if (++retryAttempt >= kMaxRetryAttempt) { - getMockCatalog()->expectGrabLock([this, - &lockName, - &lastTS, - &me, - &lastTime, - &whyMsg]( - StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why) { - ASSERT_EQUALS(lockName, lockID); - // Every attempt should have a unique sesssion ID. - ASSERT_NOT_EQUALS(lastTS, lockSessionID); - lastTS = lockSessionID; - ASSERT_TRUE(lockSessionID.isSet()); - ASSERT_EQUALS(getProcessID(), processId); - ASSERT_GREATER_THAN_OR_EQUALS(time, lastTime); - ASSERT_EQUALS(whyMsg, why); - - getMockCatalog()->expectNoGrabLock(); - }, {ErrorCodes::NetworkTimeout, "bad test network"}); + getMockCatalog()->expectGrabLock( + [this, &lockName, &lastTS, &me, &lastTime, &whyMsg](StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { + ASSERT_EQUALS(lockName, lockID); + // Every attempt should have a unique sesssion ID. + ASSERT_NOT_EQUALS(lastTS, lockSessionID); + lastTS = lockSessionID; + ASSERT_TRUE(lockSessionID.isSet()); + ASSERT_EQUALS(getProcessID(), processId); + ASSERT_GREATER_THAN_OR_EQUALS(time, lastTime); + ASSERT_EQUALS(whyMsg, why); + + getMockCatalog()->expectNoGrabLock(); + }, + {ErrorCodes::NetworkTimeout, "bad test network"}); } - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - // Make mock return lock not found to skip lock overtaking. - getMockCatalog()->expectGetLockByName([](StringData) {}, - {ErrorCodes::LockNotFound, "not found!"}); + // Make mock return lock not found to skip lock overtaking. + getMockCatalog()->expectGetLockByName([](StringData) {}, + {ErrorCodes::LockNotFound, "not found!"}); - stdx::mutex unlockMutex; - stdx::condition_variable unlockCV; - OID unlockSessionIDPassed; - int unlockCallCount = 0; + stdx::mutex unlockMutex; + stdx::condition_variable unlockCV; + OID unlockSessionIDPassed; + int unlockCallCount = 0; - getMockCatalog()->expectUnLock( - [&unlockMutex, &unlockCV, &unlockCallCount, &unlockSessionIDPassed]( - const OID& lockSessionID) { + getMockCatalog()->expectUnLock( + [&unlockMutex, &unlockCV, &unlockCallCount, &unlockSessionIDPassed]( + const OID& lockSessionID) { stdx::unique_lock<stdx::mutex> lk(unlockMutex); unlockCallCount++; unlockSessionIDPassed = lockSessionID; unlockCV.notify_all(); - }, Status::OK()); + }, + Status::OK()); - { - auto lockStatus = getMgr()->lock(lockName, whyMsg, Milliseconds(10), Milliseconds(1)); - ASSERT_NOT_OK(lockStatus.getStatus()); - } + { + auto lockStatus = getMgr()->lock(lockName, whyMsg, Milliseconds(10), Milliseconds(1)); + ASSERT_NOT_OK(lockStatus.getStatus()); + } - bool didTimeout = false; - { - stdx::unique_lock<stdx::mutex> lk(unlockMutex); - if (unlockCallCount == 0) { - didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; - } + bool didTimeout = false; + { + stdx::unique_lock<stdx::mutex> lk(unlockMutex); + if (unlockCallCount == 0) { + didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; } - - // Join the background thread before trying to call asserts. Shutdown calls - // stopPing and we don't care in this test. - getMockCatalog()->expectStopPing([](StringData){}, Status::OK()); - getMgr()->shutDown(); - - // No assert until shutDown has been called to make sure that the background thread - // won't be trying to access the local variables that were captured by lamdas that - // may have gone out of scope when the assert unwinds the stack. - // No need to grab unlockMutex since there is only one thread running at this point. - - ASSERT_FALSE(didTimeout); - ASSERT_EQUALS(1, unlockCallCount); - ASSERT_EQUALS(lastTS, unlockSessionIDPassed); } - TEST_F(ReplSetDistLockManagerFixture, LockBusyNoRetry) { - getMockCatalog()->expectGrabLock([this]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { - getMockCatalog()->expectNoGrabLock(); // Call only once. - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + // Join the background thread before trying to call asserts. Shutdown calls + // stopPing and we don't care in this test. + getMockCatalog()->expectStopPing([](StringData) {}, Status::OK()); + getMgr()->shutDown(); + + // No assert until shutDown has been called to make sure that the background thread + // won't be trying to access the local variables that were captured by lamdas that + // may have gone out of scope when the assert unwinds the stack. + // No need to grab unlockMutex since there is only one thread running at this point. + + ASSERT_FALSE(didTimeout); + ASSERT_EQUALS(1, unlockCallCount); + ASSERT_EQUALS(lastTS, unlockSessionIDPassed); +} + +TEST_F(ReplSetDistLockManagerFixture, LockBusyNoRetry) { + getMockCatalog()->expectGrabLock( + [this](StringData, const OID&, StringData, StringData, Date_t, StringData) { + getMockCatalog()->expectNoGrabLock(); // Call only once. + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + // Make mock return lock not found to skip lock overtaking. + getMockCatalog()->expectGetLockByName([](StringData) {}, + {ErrorCodes::LockNotFound, "not found!"}); + + auto status = getMgr()->lock("", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); +} - // Make mock return lock not found to skip lock overtaking. - getMockCatalog()->expectGetLockByName([](StringData) {}, - {ErrorCodes::LockNotFound, "not found!"}); - - auto status = getMgr()->lock("", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } - - /** - * Test scenario: - * 1. Attempt to grab lock. - * 2. Check that each attempt uses a unique lock session id. - * 3. Times out trying. - * 4. Checks result is error. - * 5. Implicitly check that unlock is not called (default setting of mock catalog). - */ - TEST_F(RSDistLockMgrWithMockTickSource, LockRetryTimeout) { - string lockName("test"); - string me("me"); - OID lastTS; - Date_t lastTime(Date_t::now()); - string whyMsg("because"); - - int retryAttempt = 0; - - getMockCatalog()->expectGrabLock( - [this, - &lockName, - &lastTS, - &me, - &lastTime, - &whyMsg, - &retryAttempt]( - StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why) { +/** + * Test scenario: + * 1. Attempt to grab lock. + * 2. Check that each attempt uses a unique lock session id. + * 3. Times out trying. + * 4. Checks result is error. + * 5. Implicitly check that unlock is not called (default setting of mock catalog). + */ +TEST_F(RSDistLockMgrWithMockTickSource, LockRetryTimeout) { + string lockName("test"); + string me("me"); + OID lastTS; + Date_t lastTime(Date_t::now()); + string whyMsg("because"); + + int retryAttempt = 0; + + getMockCatalog()->expectGrabLock( + [this, &lockName, &lastTS, &me, &lastTime, &whyMsg, &retryAttempt](StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS(lockName, lockID); // Every attempt should have a unique sesssion ID. ASSERT_NOT_EQUALS(lastTS, lockSessionID); @@ -535,47 +510,41 @@ namespace { retryAttempt++; getMockTickSource()->advance(Milliseconds(1)); - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - // Make mock return lock not found to skip lock overtaking. - getMockCatalog()->expectGetLockByName([](StringData) {}, - {ErrorCodes::LockNotFound, "not found!"}); + // Make mock return lock not found to skip lock overtaking. + getMockCatalog()->expectGetLockByName([](StringData) {}, + {ErrorCodes::LockNotFound, "not found!"}); - auto lockStatus = getMgr()->lock(lockName, - whyMsg, - Milliseconds(5), - Milliseconds(1)).getStatus(); - ASSERT_NOT_OK(lockStatus); + auto lockStatus = + getMgr()->lock(lockName, whyMsg, Milliseconds(5), Milliseconds(1)).getStatus(); + ASSERT_NOT_OK(lockStatus); - ASSERT_EQUALS(ErrorCodes::LockBusy, lockStatus.code()); - ASSERT_GREATER_THAN(retryAttempt, 1); - } + ASSERT_EQUALS(ErrorCodes::LockBusy, lockStatus.code()); + ASSERT_GREATER_THAN(retryAttempt, 1); +} - /** - * Test scenario: - * 1. Set mock to error on grab lock. - * 2. Grab lock attempted. - * 3. Wait for unlock to be called. - * 4. Check that lockSessionID used on all unlock is the same as the one used to grab lock. - */ - TEST_F(ReplSetDistLockManagerFixture, MustUnlockOnLockError) { - string lockName("test"); - string me("me"); - OID lastTS; - string whyMsg("because"); - - getMockCatalog()->expectGrabLock( - [this, - &lockName, - &lastTS, - &me, - &whyMsg]( - StringData lockID, - const OID& lockSessionID, - StringData who, - StringData processId, - Date_t time, - StringData why) { +/** + * Test scenario: + * 1. Set mock to error on grab lock. + * 2. Grab lock attempted. + * 3. Wait for unlock to be called. + * 4. Check that lockSessionID used on all unlock is the same as the one used to grab lock. + */ +TEST_F(ReplSetDistLockManagerFixture, MustUnlockOnLockError) { + string lockName("test"); + string me("me"); + OID lastTS; + string whyMsg("because"); + + getMockCatalog()->expectGrabLock( + [this, &lockName, &lastTS, &me, &whyMsg](StringData lockID, + const OID& lockSessionID, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS(lockName, lockID); // Every attempt should have a unique sesssion ID. ASSERT_TRUE(lockSessionID.isSet()); @@ -584,69 +553,69 @@ namespace { lastTS = lockSessionID; getMockCatalog()->expectNoGrabLock(); - }, {ErrorCodes::NetworkTimeout, "bad test network"}); + }, + {ErrorCodes::NetworkTimeout, "bad test network"}); - stdx::mutex unlockMutex; - stdx::condition_variable unlockCV; - int unlockCallCount = 0; - OID unlockSessionIDPassed; + stdx::mutex unlockMutex; + stdx::condition_variable unlockCV; + int unlockCallCount = 0; + OID unlockSessionIDPassed; - getMockCatalog()->expectUnLock( - [&unlockMutex, &unlockCV, &unlockCallCount, &unlockSessionIDPassed]( - const OID& lockSessionID) { + getMockCatalog()->expectUnLock( + [&unlockMutex, &unlockCV, &unlockCallCount, &unlockSessionIDPassed]( + const OID& lockSessionID) { stdx::unique_lock<stdx::mutex> lk(unlockMutex); unlockCallCount++; unlockSessionIDPassed = lockSessionID; unlockCV.notify_all(); - }, Status::OK()); - - auto lockStatus = getMgr()->lock(lockName, - whyMsg, - Milliseconds(10), - Milliseconds(1)).getStatus(); - ASSERT_NOT_OK(lockStatus); - ASSERT_EQUALS(ErrorCodes::NetworkTimeout, lockStatus.code()); - - bool didTimeout = false; - { - stdx::unique_lock<stdx::mutex> lk(unlockMutex); - if (unlockCallCount == 0) { - didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; - } + }, + Status::OK()); + + auto lockStatus = + getMgr()->lock(lockName, whyMsg, Milliseconds(10), Milliseconds(1)).getStatus(); + ASSERT_NOT_OK(lockStatus); + ASSERT_EQUALS(ErrorCodes::NetworkTimeout, lockStatus.code()); + + bool didTimeout = false; + { + stdx::unique_lock<stdx::mutex> lk(unlockMutex); + if (unlockCallCount == 0) { + didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; } + } - // Join the background thread before trying to call asserts. Shutdown calls - // stopPing and we don't care in this test. - getMockCatalog()->expectStopPing([](StringData){}, Status::OK()); - getMgr()->shutDown(); + // Join the background thread before trying to call asserts. Shutdown calls + // stopPing and we don't care in this test. + getMockCatalog()->expectStopPing([](StringData) {}, Status::OK()); + getMgr()->shutDown(); - // No assert until shutDown has been called to make sure that the background thread - // won't be trying to access the local variables that were captured by lamdas that - // may have gone out of scope when the assert unwinds the stack. - // No need to grab unlockMutex since there is only one thread running at this point. + // No assert until shutDown has been called to make sure that the background thread + // won't be trying to access the local variables that were captured by lamdas that + // may have gone out of scope when the assert unwinds the stack. + // No need to grab unlockMutex since there is only one thread running at this point. - ASSERT_FALSE(didTimeout); - ASSERT_EQUALS(1, unlockCallCount); - ASSERT_EQUALS(lastTS, unlockSessionIDPassed); - } + ASSERT_FALSE(didTimeout); + ASSERT_EQUALS(1, unlockCallCount); + ASSERT_EQUALS(lastTS, unlockSessionIDPassed); +} - /** - * Test scenario: - * 1. Ping thread started during setUp of fixture. - * 2. Wait until ping was called at least 3 times. - * 3. Check that correct process is being pinged. - * 4. Check that ping values are unique (based on the assumption that the system - * clock supports 2ms granularity). - */ - TEST_F(ReplSetDistLockManagerFixture, LockPinging) { - stdx::mutex testMutex; - stdx::condition_variable ping3TimesCV; - vector<Date_t> pingValues; - vector<string> processIDList; - - getMockCatalog()->expectPing( - [&testMutex, &ping3TimesCV, &processIDList, &pingValues]( - StringData processIDArg, Date_t ping) { +/** + * Test scenario: + * 1. Ping thread started during setUp of fixture. + * 2. Wait until ping was called at least 3 times. + * 3. Check that correct process is being pinged. + * 4. Check that ping values are unique (based on the assumption that the system + * clock supports 2ms granularity). + */ +TEST_F(ReplSetDistLockManagerFixture, LockPinging) { + stdx::mutex testMutex; + stdx::condition_variable ping3TimesCV; + vector<Date_t> pingValues; + vector<string> processIDList; + + getMockCatalog()->expectPing( + [&testMutex, &ping3TimesCV, &processIDList, &pingValues](StringData processIDArg, + Date_t ping) { stdx::lock_guard<stdx::mutex> lk(testMutex); processIDList.push_back(processIDArg.toString()); pingValues.push_back(ping); @@ -654,1152 +623,1127 @@ namespace { if (processIDList.size() >= 3) { ping3TimesCV.notify_all(); } - }, Status::OK()); - - bool didTimeout = false; - { - stdx::unique_lock<stdx::mutex> lk(testMutex); - if (processIDList.size() < 3) { - didTimeout = ping3TimesCV.wait_for(lk, Milliseconds(50)) == - stdx::cv_status::timeout; - } + }, + Status::OK()); + + bool didTimeout = false; + { + stdx::unique_lock<stdx::mutex> lk(testMutex); + if (processIDList.size() < 3) { + didTimeout = ping3TimesCV.wait_for(lk, Milliseconds(50)) == stdx::cv_status::timeout; } + } - // Join the background thread before trying to call asserts. Shutdown calls - // stopPing and we don't care in this test. - getMockCatalog()->expectStopPing([](StringData){}, Status::OK()); - getMgr()->shutDown(); + // Join the background thread before trying to call asserts. Shutdown calls + // stopPing and we don't care in this test. + getMockCatalog()->expectStopPing([](StringData) {}, Status::OK()); + getMgr()->shutDown(); - // No assert until shutDown has been called to make sure that the background thread - // won't be trying to access the local variables that were captured by lamdas that - // may have gone out of scope when the assert unwinds the stack. - // No need to grab testMutex since there is only one thread running at this point. + // No assert until shutDown has been called to make sure that the background thread + // won't be trying to access the local variables that were captured by lamdas that + // may have gone out of scope when the assert unwinds the stack. + // No need to grab testMutex since there is only one thread running at this point. - ASSERT_FALSE(didTimeout); + ASSERT_FALSE(didTimeout); - Date_t lastPing; - for (const auto& ping : pingValues) { - ASSERT_NOT_EQUALS(lastPing, ping); - lastPing = ping; - } - - for (const auto& processIDArg : processIDList) { - ASSERT_EQUALS(getProcessID(), processIDArg); - } + Date_t lastPing; + for (const auto& ping : pingValues) { + ASSERT_NOT_EQUALS(lastPing, ping); + lastPing = ping; } - /** - * Test scenario: - * 1. Grab lock. - * 2. Unlock fails 3 times. - * 3. Unlock finally succeeds at the 4th time. - * 4. Check that lockSessionID used on all unlock is the same as the one used to grab lock. - */ - TEST_F(ReplSetDistLockManagerFixture, UnlockUntilNoError) { - stdx::mutex unlockMutex; - stdx::condition_variable unlockCV; - const unsigned int kUnlockErrorCount = 3; - vector<OID> lockSessionIDPassed; + for (const auto& processIDArg : processIDList) { + ASSERT_EQUALS(getProcessID(), processIDArg); + } +} - getMockCatalog()->expectUnLock( - [this, &unlockMutex, &unlockCV, &kUnlockErrorCount, &lockSessionIDPassed]( - const OID& lockSessionID) { +/** + * Test scenario: + * 1. Grab lock. + * 2. Unlock fails 3 times. + * 3. Unlock finally succeeds at the 4th time. + * 4. Check that lockSessionID used on all unlock is the same as the one used to grab lock. + */ +TEST_F(ReplSetDistLockManagerFixture, UnlockUntilNoError) { + stdx::mutex unlockMutex; + stdx::condition_variable unlockCV; + const unsigned int kUnlockErrorCount = 3; + vector<OID> lockSessionIDPassed; + + getMockCatalog()->expectUnLock( + [this, &unlockMutex, &unlockCV, &kUnlockErrorCount, &lockSessionIDPassed]( + const OID& lockSessionID) { stdx::unique_lock<stdx::mutex> lk(unlockMutex); lockSessionIDPassed.push_back(lockSessionID); if (lockSessionIDPassed.size() >= kUnlockErrorCount) { getMockCatalog()->expectUnLock( - [&lockSessionIDPassed, &unlockMutex, &unlockCV]( - const OID& lockSessionID) { - stdx::unique_lock<stdx::mutex> lk(unlockMutex); - lockSessionIDPassed.push_back(lockSessionID); - unlockCV.notify_all(); - }, Status::OK()); - } - }, {ErrorCodes::NetworkTimeout, "bad test network"}); - - OID lockSessionID; - LocksType retLockDoc; - retLockDoc.setName("test"); - retLockDoc.setState(LocksType::LOCKED); - retLockDoc.setProcess(getProcessID()); - retLockDoc.setWho("me"); - retLockDoc.setWhy("why"); - // Will be different from the actual lock session id. For testing only. - retLockDoc.setLockID(OID::gen()); - - getMockCatalog()->expectGrabLock([&lockSessionID]( - StringData lockID, - const OID& lockSessionIDArg, - StringData who, - StringData processId, - Date_t time, - StringData why) { - lockSessionID = lockSessionIDArg; - }, retLockDoc); - - { - auto lockStatus = getMgr()->lock("test", "why", Milliseconds(0), Milliseconds(0)); - } - - bool didTimeout = false; - { - stdx::unique_lock<stdx::mutex> lk(unlockMutex); - if (lockSessionIDPassed.size() < kUnlockErrorCount) { - didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; + [&lockSessionIDPassed, &unlockMutex, &unlockCV](const OID& lockSessionID) { + stdx::unique_lock<stdx::mutex> lk(unlockMutex); + lockSessionIDPassed.push_back(lockSessionID); + unlockCV.notify_all(); + }, + Status::OK()); } + }, + {ErrorCodes::NetworkTimeout, "bad test network"}); + + OID lockSessionID; + LocksType retLockDoc; + retLockDoc.setName("test"); + retLockDoc.setState(LocksType::LOCKED); + retLockDoc.setProcess(getProcessID()); + retLockDoc.setWho("me"); + retLockDoc.setWhy("why"); + // Will be different from the actual lock session id. For testing only. + retLockDoc.setLockID(OID::gen()); + + getMockCatalog()->expectGrabLock( + [&lockSessionID](StringData lockID, + const OID& lockSessionIDArg, + StringData who, + StringData processId, + Date_t time, + StringData why) { lockSessionID = lockSessionIDArg; }, + retLockDoc); + + { auto lockStatus = getMgr()->lock("test", "why", Milliseconds(0), Milliseconds(0)); } + + bool didTimeout = false; + { + stdx::unique_lock<stdx::mutex> lk(unlockMutex); + if (lockSessionIDPassed.size() < kUnlockErrorCount) { + didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; } + } - // Join the background thread before trying to call asserts. Shutdown calls - // stopPing and we don't care in this test. - getMockCatalog()->expectStopPing([](StringData){}, Status::OK()); - getMgr()->shutDown(); + // Join the background thread before trying to call asserts. Shutdown calls + // stopPing and we don't care in this test. + getMockCatalog()->expectStopPing([](StringData) {}, Status::OK()); + getMgr()->shutDown(); - // No assert until shutDown has been called to make sure that the background thread - // won't be trying to access the local variables that were captured by lamdas that - // may have gone out of scope when the assert unwinds the stack. - // No need to grab testMutex since there is only one thread running at this point. + // No assert until shutDown has been called to make sure that the background thread + // won't be trying to access the local variables that were captured by lamdas that + // may have gone out of scope when the assert unwinds the stack. + // No need to grab testMutex since there is only one thread running at this point. - ASSERT_FALSE(didTimeout); + ASSERT_FALSE(didTimeout); - for (const auto& id : lockSessionIDPassed) { - ASSERT_EQUALS(lockSessionID, id); - } + for (const auto& id : lockSessionIDPassed) { + ASSERT_EQUALS(lockSessionID, id); } +} + +/** + * Test scenario: + * 1. Grab 2 locks. + * 2. Trigger unlocks by making ScopedDistLock go out of scope. + * 3. Unlocks fail and will be queued for retry. + * 4. Unlocks will keep on failing until we see at least 3 unique ids being unlocked more + * than once. This implies that both ids have been retried at least 3 times. + * 5. Check that the lock session id used when lock was called matches with unlock. + */ +TEST_F(ReplSetDistLockManagerFixture, MultipleQueuedUnlock) { + stdx::mutex testMutex; + stdx::condition_variable unlockCV; + + vector<OID> lockSessionIDPassed; + map<OID, int> unlockIDMap; // id -> count /** - * Test scenario: - * 1. Grab 2 locks. - * 2. Trigger unlocks by making ScopedDistLock go out of scope. - * 3. Unlocks fail and will be queued for retry. - * 4. Unlocks will keep on failing until we see at least 3 unique ids being unlocked more - * than once. This implies that both ids have been retried at least 3 times. - * 5. Check that the lock session id used when lock was called matches with unlock. + * Returns true if all values in the map are greater than 2. */ - TEST_F(ReplSetDistLockManagerFixture, MultipleQueuedUnlock) { - stdx::mutex testMutex; - stdx::condition_variable unlockCV; - - vector<OID> lockSessionIDPassed; - map<OID, int> unlockIDMap; // id -> count - - /** - * Returns true if all values in the map are greater than 2. - */ - auto mapEntriesGreaterThanTwo = [](const decltype(unlockIDMap)& map) -> bool { - auto iter = find_if(map.begin(), map.end(), - [](const std::remove_reference<decltype(map)>::type::value_type& entry) - -> bool { - return entry.second < 3; - }); - - return iter == map.end(); - }; + auto mapEntriesGreaterThanTwo = [](const decltype(unlockIDMap)& map) -> bool { + auto iter = find_if(map.begin(), + map.end(), + [](const std::remove_reference<decltype(map)>::type::value_type& entry) + -> bool { return entry.second < 3; }); - getMockCatalog()->expectUnLock( - [this, &unlockIDMap, &testMutex, &unlockCV, &mapEntriesGreaterThanTwo]( - const OID& lockSessionID) { + return iter == map.end(); + }; + + getMockCatalog()->expectUnLock( + [this, &unlockIDMap, &testMutex, &unlockCV, &mapEntriesGreaterThanTwo]( + const OID& lockSessionID) { stdx::unique_lock<stdx::mutex> lk(testMutex); unlockIDMap[lockSessionID]++; // Wait until we see at least 2 unique lockSessionID more than twice. if (unlockIDMap.size() >= 2 && mapEntriesGreaterThanTwo(unlockIDMap)) { - getMockCatalog()->expectUnLock( - [&testMutex, &unlockCV](const OID& lockSessionID) { + getMockCatalog()->expectUnLock([&testMutex, &unlockCV](const OID& lockSessionID) { stdx::unique_lock<stdx::mutex> lk(testMutex); unlockCV.notify_all(); }, Status::OK()); } - }, {ErrorCodes::NetworkTimeout, "bad test network"}); - - LocksType retLockDoc; - retLockDoc.setName("test"); - retLockDoc.setState(LocksType::LOCKED); - retLockDoc.setProcess(getProcessID()); - retLockDoc.setWho("me"); - retLockDoc.setWhy("why"); - // Will be different from the actual lock session id. For testing only. - retLockDoc.setLockID(OID::gen()); - - getMockCatalog()->expectGrabLock([&testMutex, &lockSessionIDPassed]( - StringData lockID, - const OID& lockSessionIDArg, - StringData who, - StringData processId, - Date_t time, - StringData why) { + }, + {ErrorCodes::NetworkTimeout, "bad test network"}); + + LocksType retLockDoc; + retLockDoc.setName("test"); + retLockDoc.setState(LocksType::LOCKED); + retLockDoc.setProcess(getProcessID()); + retLockDoc.setWho("me"); + retLockDoc.setWhy("why"); + // Will be different from the actual lock session id. For testing only. + retLockDoc.setLockID(OID::gen()); + + getMockCatalog()->expectGrabLock( + [&testMutex, &lockSessionIDPassed](StringData lockID, + const OID& lockSessionIDArg, + StringData who, + StringData processId, + Date_t time, + StringData why) { stdx::unique_lock<stdx::mutex> lk(testMutex); lockSessionIDPassed.push_back(lockSessionIDArg); - }, retLockDoc); - - { - auto lockStatus = getMgr()->lock("test", "why", Milliseconds(0), Milliseconds(0)); - auto otherStatus = getMgr()->lock("lock", "why", Milliseconds(0), Milliseconds(0)); - } - - bool didTimeout = false; - { - stdx::unique_lock<stdx::mutex> lk(testMutex); - - if (unlockIDMap.size() < 2 || !mapEntriesGreaterThanTwo(unlockIDMap)) { - didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; - } - } - - // Join the background thread before trying to call asserts. Shutdown calls - // stopPing and we don't care in this test. - getMockCatalog()->expectStopPing([](StringData){}, Status::OK()); - getMgr()->shutDown(); + }, + retLockDoc); - // No assert until shutDown has been called to make sure that the background thread - // won't be trying to access the local variables that were captured by lamdas that - // may have gone out of scope when the assert unwinds the stack. - // No need to grab testMutex since there is only one thread running at this point. + { + auto lockStatus = getMgr()->lock("test", "why", Milliseconds(0), Milliseconds(0)); + auto otherStatus = getMgr()->lock("lock", "why", Milliseconds(0), Milliseconds(0)); + } - ASSERT_FALSE(didTimeout); - ASSERT_EQUALS(2u, lockSessionIDPassed.size()); + bool didTimeout = false; + { + stdx::unique_lock<stdx::mutex> lk(testMutex); - for (const auto& id : lockSessionIDPassed) { - ASSERT_GREATER_THAN(unlockIDMap[id], 2) - << "lockIDList: " << vectorToString(lockSessionIDPassed) - << ", map: " << mapToString(unlockIDMap); + if (unlockIDMap.size() < 2 || !mapEntriesGreaterThanTwo(unlockIDMap)) { + didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; } } - TEST_F(ReplSetDistLockManagerFixture, CleanupPingOnShutdown) { - bool stopPingCalled = false; - getMockCatalog()->expectStopPing([this, & stopPingCalled]( - StringData processID) { - ASSERT_EQUALS(getProcessID(), processID); - stopPingCalled = true; - }, Status::OK()); - - getMgr()->shutDown(); - ASSERT_TRUE(stopPingCalled); - } + // Join the background thread before trying to call asserts. Shutdown calls + // stopPing and we don't care in this test. + getMockCatalog()->expectStopPing([](StringData) {}, Status::OK()); + getMgr()->shutDown(); - TEST_F(ReplSetDistLockManagerFixture, CheckLockStatusOK) { - LocksType retLockDoc; - retLockDoc.setName("test"); - retLockDoc.setState(LocksType::LOCKED); - retLockDoc.setProcess(getProcessID()); - retLockDoc.setWho("me"); - retLockDoc.setWhy("why"); - // Will be different from the actual lock session id. For testing only. - retLockDoc.setLockID(OID::gen()); + // No assert until shutDown has been called to make sure that the background thread + // won't be trying to access the local variables that were captured by lamdas that + // may have gone out of scope when the assert unwinds the stack. + // No need to grab testMutex since there is only one thread running at this point. - OID lockSessionID; + ASSERT_FALSE(didTimeout); + ASSERT_EQUALS(2u, lockSessionIDPassed.size()); - getMockCatalog()->expectGrabLock([&lockSessionID] - (StringData, const OID& ts, StringData, StringData, Date_t, StringData) { + for (const auto& id : lockSessionIDPassed) { + ASSERT_GREATER_THAN(unlockIDMap[id], 2) + << "lockIDList: " << vectorToString(lockSessionIDPassed) + << ", map: " << mapToString(unlockIDMap); + } +} + +TEST_F(ReplSetDistLockManagerFixture, CleanupPingOnShutdown) { + bool stopPingCalled = false; + getMockCatalog()->expectStopPing([this, &stopPingCalled](StringData processID) { + ASSERT_EQUALS(getProcessID(), processID); + stopPingCalled = true; + }, Status::OK()); + + getMgr()->shutDown(); + ASSERT_TRUE(stopPingCalled); +} + +TEST_F(ReplSetDistLockManagerFixture, CheckLockStatusOK) { + LocksType retLockDoc; + retLockDoc.setName("test"); + retLockDoc.setState(LocksType::LOCKED); + retLockDoc.setProcess(getProcessID()); + retLockDoc.setWho("me"); + retLockDoc.setWhy("why"); + // Will be different from the actual lock session id. For testing only. + retLockDoc.setLockID(OID::gen()); + + OID lockSessionID; + + getMockCatalog()->expectGrabLock( + [&lockSessionID](StringData, const OID& ts, StringData, StringData, Date_t, StringData) { lockSessionID = ts; - }, retLockDoc); + }, + retLockDoc); - auto lockStatus = getMgr()->lock("a", "", Milliseconds(0), Milliseconds(0)); - ASSERT_OK(lockStatus.getStatus()); + auto lockStatus = getMgr()->lock("a", "", Milliseconds(0), Milliseconds(0)); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock([](const OID&) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [](const OID&) { // Don't care - }, Status::OK()); + }, + Status::OK()); - auto& scopedLock = lockStatus.getValue(); + auto& scopedLock = lockStatus.getValue(); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectGetLockByTS([&lockSessionID](const OID& ts) { - ASSERT_EQUALS(lockSessionID, ts); - }, retLockDoc); + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectGetLockByTS( + [&lockSessionID](const OID& ts) { ASSERT_EQUALS(lockSessionID, ts); }, retLockDoc); - ASSERT_OK(scopedLock.checkStatus()); - } + ASSERT_OK(scopedLock.checkStatus()); +} - TEST_F(ReplSetDistLockManagerFixture, CheckLockStatusNoLongerOwn) { - LocksType retLockDoc; - retLockDoc.setName("test"); - retLockDoc.setState(LocksType::LOCKED); - retLockDoc.setProcess(getProcessID()); - retLockDoc.setWho("me"); - retLockDoc.setWhy("why"); - // Will be different from the actual lock session id. For testing only. - retLockDoc.setLockID(OID::gen()); +TEST_F(ReplSetDistLockManagerFixture, CheckLockStatusNoLongerOwn) { + LocksType retLockDoc; + retLockDoc.setName("test"); + retLockDoc.setState(LocksType::LOCKED); + retLockDoc.setProcess(getProcessID()); + retLockDoc.setWho("me"); + retLockDoc.setWhy("why"); + // Will be different from the actual lock session id. For testing only. + retLockDoc.setLockID(OID::gen()); - OID lockSessionID; + OID lockSessionID; - getMockCatalog()->expectGrabLock([&lockSessionID] - (StringData, const OID& ts, StringData, StringData, Date_t, StringData) { + getMockCatalog()->expectGrabLock( + [&lockSessionID](StringData, const OID& ts, StringData, StringData, Date_t, StringData) { lockSessionID = ts; - }, retLockDoc); + }, + retLockDoc); - auto lockStatus = getMgr()->lock("a", "", Milliseconds(0), Milliseconds(0)); - ASSERT_OK(lockStatus.getStatus()); + auto lockStatus = getMgr()->lock("a", "", Milliseconds(0), Milliseconds(0)); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock([](const OID&) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [](const OID&) { // Don't care - }, Status::OK()); + }, + Status::OK()); - auto& scopedLock = lockStatus.getValue(); + auto& scopedLock = lockStatus.getValue(); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectGetLockByTS([&lockSessionID](const OID& ts) { - ASSERT_EQUALS(lockSessionID, ts); - }, {ErrorCodes::LockNotFound, "no lock"}); + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectGetLockByTS([&lockSessionID](const OID& ts) { + ASSERT_EQUALS(lockSessionID, ts); + }, {ErrorCodes::LockNotFound, "no lock"}); - ASSERT_NOT_OK(scopedLock.checkStatus()); - } + ASSERT_NOT_OK(scopedLock.checkStatus()); +} - TEST_F(ReplSetDistLockManagerFixture, CheckLockStatusError) { - LocksType retLockDoc; - retLockDoc.setName("test"); - retLockDoc.setState(LocksType::LOCKED); - retLockDoc.setProcess(getProcessID()); - retLockDoc.setWho("me"); - retLockDoc.setWhy("why"); - // Will be different from the actual lock session id. For testing only. - retLockDoc.setLockID(OID::gen()); +TEST_F(ReplSetDistLockManagerFixture, CheckLockStatusError) { + LocksType retLockDoc; + retLockDoc.setName("test"); + retLockDoc.setState(LocksType::LOCKED); + retLockDoc.setProcess(getProcessID()); + retLockDoc.setWho("me"); + retLockDoc.setWhy("why"); + // Will be different from the actual lock session id. For testing only. + retLockDoc.setLockID(OID::gen()); - OID lockSessionID; + OID lockSessionID; - getMockCatalog()->expectGrabLock([&lockSessionID] - (StringData, const OID& ts, StringData, StringData, Date_t, StringData) { + getMockCatalog()->expectGrabLock( + [&lockSessionID](StringData, const OID& ts, StringData, StringData, Date_t, StringData) { lockSessionID = ts; - }, retLockDoc); + }, + retLockDoc); - auto lockStatus = getMgr()->lock("a", "", Milliseconds(0), Milliseconds(0)); - ASSERT_OK(lockStatus.getStatus()); + auto lockStatus = getMgr()->lock("a", "", Milliseconds(0), Milliseconds(0)); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock([](const OID&) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [](const OID&) { // Don't care - }, Status::OK()); + }, + Status::OK()); - auto& scopedLock = lockStatus.getValue(); + auto& scopedLock = lockStatus.getValue(); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectGetLockByTS([&lockSessionID](const OID& ts) { - ASSERT_EQUALS(lockSessionID, ts); - }, {ErrorCodes::NetworkTimeout, "bad test network"}); + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectGetLockByTS([&lockSessionID](const OID& ts) { + ASSERT_EQUALS(lockSessionID, ts); + }, {ErrorCodes::NetworkTimeout, "bad test network"}); - ASSERT_NOT_OK(scopedLock.checkStatus()); - } + ASSERT_NOT_OK(scopedLock.checkStatus()); +} - /** - * Test scenario: - * 1. Attempt to grab lock fails because lock is already owned. - * 2. Try to get ping data and config server clock. - * 3. Since we don't have previous ping data to compare with, we cannot - * decide whether it's ok to overtake, so we can't. - * 4. Lock expiration has elapsed and the ping has not been updated since. - * 5. 2nd attempt to grab lock still fails for the same reason. - * 6. But since the ping is not fresh anymore, dist lock manager should overtake lock. - */ - TEST_F(ReplSetDistLockManagerFixture, BasicLockOvertaking) { - OID lastTS; +/** + * Test scenario: + * 1. Attempt to grab lock fails because lock is already owned. + * 2. Try to get ping data and config server clock. + * 3. Since we don't have previous ping data to compare with, we cannot + * decide whether it's ok to overtake, so we can't. + * 4. Lock expiration has elapsed and the ping has not been updated since. + * 5. 2nd attempt to grab lock still fails for the same reason. + * 6. But since the ping is not fresh anymore, dist lock manager should overtake lock. + */ +TEST_F(ReplSetDistLockManagerFixture, BasicLockOvertaking) { + OID lastTS; - getMockCatalog()->expectGrabLock([&lastTS]( - StringData, const OID& lockSessionID, StringData, StringData, Date_t, StringData) { + getMockCatalog()->expectGrabLock( + [&lastTS]( + StringData, const OID& lockSessionID, StringData, StringData, Date_t, StringData) { lastTS = lockSessionID; - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); - - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); - - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); - - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t(), OID())); - - // First attempt will record the ping data. - { - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); + + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); + + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); - // Advance config server time to exceed lock expiration. - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t() + kLockExpiration + Milliseconds(1), OID())); - - getMockCatalog()->expectOvertakeLock( - [this, &lastTS, ¤tLockDoc] - (StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { + getMockCatalog()->expectGetServerInfo([]() {}, DistLockCatalog::ServerInfo(Date_t(), OID())); + + // First attempt will record the ping data. + { + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); + } + + // Advance config server time to exceed lock expiration. + getMockCatalog()->expectGetServerInfo( + []() {}, DistLockCatalog::ServerInfo(Date_t() + kLockExpiration + Milliseconds(1), OID())); + + getMockCatalog()->expectOvertakeLock( + [this, &lastTS, ¤tLockDoc](StringData lockID, + const OID& lockSessionID, + const OID& currentHolderTS, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS("bar", lockID); ASSERT_EQUALS(lastTS, lockSessionID); ASSERT_EQUALS(currentLockDoc.getLockID(), currentHolderTS); ASSERT_EQUALS(getProcessID(), processId); ASSERT_EQUALS("foo", why); - }, currentLockDoc); // return arbitrary valid lock document, for testing purposes only. + }, + currentLockDoc); // return arbitrary valid lock document, for testing purposes only. - int unlockCallCount = 0; - OID unlockSessionIDPassed; + int unlockCallCount = 0; + OID unlockSessionIDPassed; - // Second attempt should overtake lock. - { - auto lockStatus = getMgr()->lock("bar", - "foo", - Milliseconds(0), - Milliseconds(0)); + // Second attempt should overtake lock. + { + auto lockStatus = getMgr()->lock("bar", "foo", Milliseconds(0), Milliseconds(0)); - ASSERT_OK(lockStatus.getStatus()); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock( - [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { unlockCallCount++; unlockSessionIDPassed = lockSessionID; - }, Status::OK()); - } - - ASSERT_EQUALS(1, unlockCallCount); - ASSERT_EQUALS(lastTS, unlockSessionIDPassed); + }, + Status::OK()); } - TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfExpirationHasNotElapsed) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { - // Don't care. - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); - - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); - - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); - - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t(), OID())); - - // First attempt will record the ping data. - { - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + ASSERT_EQUALS(1, unlockCallCount); + ASSERT_EQUALS(lastTS, unlockSessionIDPassed); +} - // Advance config server time to 1 millisecond before lock expiration. - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t() + kLockExpiration - Milliseconds(1), OID())); +TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfExpirationHasNotElapsed) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { + // Don't care. + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - // Second attempt should still not overtake lock. - { - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } - } + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); - TEST_F(ReplSetDistLockManagerFixture, GetPingErrorWhileOvertaking) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { - // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, {ErrorCodes::NetworkTimeout, "bad test network"}); + getMockCatalog()->expectGetServerInfo([]() {}, DistLockCatalog::ServerInfo(Date_t(), OID())); + // First attempt will record the ping data. + { auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::NetworkTimeout, status.code()); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); } - TEST_F(ReplSetDistLockManagerFixture, GetInvalidPingDocumentWhileOvertaking) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { - // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); - - LockpingsType invalidPing; - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, invalidPing); + // Advance config server time to 1 millisecond before lock expiration. + getMockCatalog()->expectGetServerInfo( + []() {}, DistLockCatalog::ServerInfo(Date_t() + kLockExpiration - Milliseconds(1), OID())); + // Second attempt should still not overtake lock. + { auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); } +} - TEST_F(ReplSetDistLockManagerFixture, GetServerInfoErrorWhileOvertaking) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { +TEST_F(ReplSetDistLockManagerFixture, GetPingErrorWhileOvertaking) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); + + getMockCatalog()->expectGetPing([](StringData process) { + ASSERT_EQUALS("otherProcess", process); + }, {ErrorCodes::NetworkTimeout, "bad test network"}); + + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::NetworkTimeout, status.code()); +} + +TEST_F(ReplSetDistLockManagerFixture, GetInvalidPingDocumentWhileOvertaking) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { + // Don't care + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); + + LockpingsType invalidPing; + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, invalidPing); + + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::UnsupportedFormat, status.code()); +} + +TEST_F(ReplSetDistLockManagerFixture, GetServerInfoErrorWhileOvertaking) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { + // Don't care + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); + + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); + + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); + + getMockCatalog()->expectGetServerInfo([]() {}, + {ErrorCodes::NetworkTimeout, "bad test network"}); + + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::NetworkTimeout, status.code()); +} + +TEST_F(ReplSetDistLockManagerFixture, GetLockErrorWhileOvertaking) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { + // Don't care + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + {ErrorCodes::NetworkTimeout, "bad test network"}); - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::NetworkTimeout, status.code()); +} - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); +TEST_F(ReplSetDistLockManagerFixture, GetLockDisappearedWhileOvertaking) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { + // Don't care + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - getMockCatalog()->expectGetServerInfo([]() { - }, {ErrorCodes::NetworkTimeout, "bad test network"}); + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + {ErrorCodes::LockNotFound, "disappeared!"}); - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::NetworkTimeout, status.code()); - } + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); +} - TEST_F(ReplSetDistLockManagerFixture, GetLockErrorWhileOvertaking) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { +/** + * 1. Try to grab lock multiple times. + * 2. For each attempt, the ping is updated and the config server clock is advanced + * by increments of lock expiration duration. + * 3. All of the previous attempt should result in lock busy. + * 4. Try to grab lock again when the ping was not updated and lock expiration has elapsed. + */ +TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfPingIsActive) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, {ErrorCodes::NetworkTimeout, "bad test network"}); + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + Date_t currentPing; + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + + Date_t configServerLocalTime; + int getServerInfoCallCount = 0; + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); + + const int kLoopCount = 5; + for (int x = 0; x < kLoopCount; x++) { + // Advance config server time to reach lock expiration. + configServerLocalTime += kLockExpiration; - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::NetworkTimeout, status.code()); - } + currentPing += Milliseconds(1); + pingDoc.setPing(currentPing); - TEST_F(ReplSetDistLockManagerFixture, GetLockDisappearedWhileOvertaking) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { - // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, {ErrorCodes::LockNotFound, "disappeared!"}); + getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { + getServerInfoCallCount++; + }, DistLockCatalog::ServerInfo(configServerLocalTime, OID())); auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); ASSERT_NOT_OK(status); ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); } - /** - * 1. Try to grab lock multiple times. - * 2. For each attempt, the ping is updated and the config server clock is advanced - * by increments of lock expiration duration. - * 3. All of the previous attempt should result in lock busy. - * 4. Try to grab lock again when the ping was not updated and lock expiration has elapsed. - */ - TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfPingIsActive) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { - // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - Date_t currentPing; - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - - Date_t configServerLocalTime; - int getServerInfoCallCount = 0; - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); - - const int kLoopCount = 5; - for (int x = 0; x < kLoopCount; x++) { - // Advance config server time to reach lock expiration. - configServerLocalTime += kLockExpiration; - - currentPing += Milliseconds(1); - pingDoc.setPing(currentPing); - - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); - - getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { - getServerInfoCallCount++; - }, DistLockCatalog::ServerInfo(configServerLocalTime, OID())); - - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } - - ASSERT_EQUALS(kLoopCount, getServerInfoCallCount); - - configServerLocalTime += kLockExpiration; - getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { - getServerInfoCallCount++; - }, DistLockCatalog::ServerInfo(configServerLocalTime, OID())); - - OID lockTS; - // Make sure that overtake is now ok since ping is no longer updated. - getMockCatalog()->expectOvertakeLock( - [this, &lockTS, ¤tLockDoc] - (StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { + ASSERT_EQUALS(kLoopCount, getServerInfoCallCount); + + configServerLocalTime += kLockExpiration; + getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { + getServerInfoCallCount++; + }, DistLockCatalog::ServerInfo(configServerLocalTime, OID())); + + OID lockTS; + // Make sure that overtake is now ok since ping is no longer updated. + getMockCatalog()->expectOvertakeLock( + [this, &lockTS, ¤tLockDoc](StringData lockID, + const OID& lockSessionID, + const OID& currentHolderTS, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS("bar", lockID); lockTS = lockSessionID; ASSERT_EQUALS(currentLockDoc.getLockID(), currentHolderTS); ASSERT_EQUALS(getProcessID(), processId); ASSERT_EQUALS("foo", why); - }, currentLockDoc); // return arbitrary valid lock document, for testing purposes only. + }, + currentLockDoc); // return arbitrary valid lock document, for testing purposes only. - int unlockCallCount = 0; - OID unlockSessionIDPassed; + int unlockCallCount = 0; + OID unlockSessionIDPassed; - { - auto lockStatus = getMgr()->lock("bar", - "foo", - Milliseconds(0), - Milliseconds(0)); + { + auto lockStatus = getMgr()->lock("bar", "foo", Milliseconds(0), Milliseconds(0)); - ASSERT_OK(lockStatus.getStatus()); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock( - [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { unlockCallCount++; unlockSessionIDPassed = lockSessionID; - }, Status::OK()); - } - - ASSERT_EQUALS(1, unlockCallCount); - ASSERT_EQUALS(lockTS, unlockSessionIDPassed); + }, + Status::OK()); } - /** - * 1. Try to grab lock multiple times. - * 2. For each attempt, the owner of the lock is different and the config server clock is - * advanced by increments of lock expiration duration. - * 3. All of the previous attempt should result in lock busy. - * 4. Try to grab lock again when the ping was not updated and lock expiration has elapsed. - */ - TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfOwnerJustChanged) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { + ASSERT_EQUALS(1, unlockCallCount); + ASSERT_EQUALS(lockTS, unlockSessionIDPassed); +} + +/** + * 1. Try to grab lock multiple times. + * 2. For each attempt, the owner of the lock is different and the config server clock is + * advanced by increments of lock expiration duration. + * 3. All of the previous attempt should result in lock busy. + * 4. Try to grab lock again when the ping was not updated and lock expiration has elapsed. + */ +TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfOwnerJustChanged) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - Date_t currentPing; - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); - - Date_t configServerLocalTime; - int getServerInfoCallCount = 0; - - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); - - const int kLoopCount = 5; - for (int x = 0; x < kLoopCount; x++) { - // Advance config server time to reach lock expiration. - configServerLocalTime += kLockExpiration; - - currentLockDoc.setLockID(OID::gen()); - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); - - getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { - getServerInfoCallCount++; - }, DistLockCatalog::ServerInfo(configServerLocalTime, OID())); - - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + Date_t currentPing; + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); + + Date_t configServerLocalTime; + int getServerInfoCallCount = 0; + + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); + + const int kLoopCount = 5; + for (int x = 0; x < kLoopCount; x++) { + // Advance config server time to reach lock expiration. + configServerLocalTime += kLockExpiration; - ASSERT_EQUALS(kLoopCount, getServerInfoCallCount); + currentLockDoc.setLockID(OID::gen()); + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); - configServerLocalTime += kLockExpiration; getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { getServerInfoCallCount++; }, DistLockCatalog::ServerInfo(configServerLocalTime, OID())); - OID lockTS; - // Make sure that overtake is now ok since lock owner didn't change. - getMockCatalog()->expectOvertakeLock( - [this, &lockTS, ¤tLockDoc] - (StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); + } + + ASSERT_EQUALS(kLoopCount, getServerInfoCallCount); + + configServerLocalTime += kLockExpiration; + getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { + getServerInfoCallCount++; + }, DistLockCatalog::ServerInfo(configServerLocalTime, OID())); + + OID lockTS; + // Make sure that overtake is now ok since lock owner didn't change. + getMockCatalog()->expectOvertakeLock( + [this, &lockTS, ¤tLockDoc](StringData lockID, + const OID& lockSessionID, + const OID& currentHolderTS, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS("bar", lockID); lockTS = lockSessionID; ASSERT_EQUALS(currentLockDoc.getLockID(), currentHolderTS); ASSERT_EQUALS(getProcessID(), processId); ASSERT_EQUALS("foo", why); - }, currentLockDoc); // return arbitrary valid lock document, for testing purposes only. + }, + currentLockDoc); // return arbitrary valid lock document, for testing purposes only. - int unlockCallCount = 0; - OID unlockSessionIDPassed; + int unlockCallCount = 0; + OID unlockSessionIDPassed; - { - auto lockStatus = getMgr()->lock("bar", - "foo", - Milliseconds(0), - Milliseconds(0)); + { + auto lockStatus = getMgr()->lock("bar", "foo", Milliseconds(0), Milliseconds(0)); - ASSERT_OK(lockStatus.getStatus()); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock( - [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { unlockCallCount++; unlockSessionIDPassed = lockSessionID; - }, Status::OK()); - } - - ASSERT_EQUALS(1, unlockCallCount); - ASSERT_EQUALS(lockTS, unlockSessionIDPassed); + }, + Status::OK()); } - /** - * 1. Try to grab lock multiple times. - * 2. For each attempt, the electionId of the config server is different and the - * config server clock is advanced by increments of lock expiration duration. - * 3. All of the previous attempt should result in lock busy. - * 4. Try to grab lock again when the ping was not updated and lock expiration has elapsed. - */ - TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfElectionIdChanged) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { + ASSERT_EQUALS(1, unlockCallCount); + ASSERT_EQUALS(lockTS, unlockSessionIDPassed); +} + +/** + * 1. Try to grab lock multiple times. + * 2. For each attempt, the electionId of the config server is different and the + * config server clock is advanced by increments of lock expiration duration. + * 3. All of the previous attempt should result in lock busy. + * 4. Try to grab lock again when the ping was not updated and lock expiration has elapsed. + */ +TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfElectionIdChanged) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - Date_t currentPing; - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); - - Date_t configServerLocalTime; - int getServerInfoCallCount = 0; - - const LocksType& fixedLockDoc = currentLockDoc; - const LockpingsType& fixedPingDoc = pingDoc; - - const int kLoopCount = 5; - OID lastElectionId; - for (int x = 0; x < kLoopCount; x++) { - // Advance config server time to reach lock expiration. - configServerLocalTime += kLockExpiration; - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, fixedLockDoc); - - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, fixedPingDoc); - - lastElectionId = OID::gen(); - getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { - getServerInfoCallCount++; - }, DistLockCatalog::ServerInfo(configServerLocalTime, lastElectionId)); - - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + Date_t currentPing; + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); + + Date_t configServerLocalTime; + int getServerInfoCallCount = 0; + + const LocksType& fixedLockDoc = currentLockDoc; + const LockpingsType& fixedPingDoc = pingDoc; + + const int kLoopCount = 5; + OID lastElectionId; + for (int x = 0; x < kLoopCount; x++) { + // Advance config server time to reach lock expiration. + configServerLocalTime += kLockExpiration; - ASSERT_EQUALS(kLoopCount, getServerInfoCallCount); + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + fixedLockDoc); - configServerLocalTime += kLockExpiration; + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, fixedPingDoc); + + lastElectionId = OID::gen(); getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { getServerInfoCallCount++; }, DistLockCatalog::ServerInfo(configServerLocalTime, lastElectionId)); - OID lockTS; - // Make sure that overtake is now ok since electionId didn't change. - getMockCatalog()->expectOvertakeLock( - [this, &lockTS, ¤tLockDoc] - (StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); + } + + ASSERT_EQUALS(kLoopCount, getServerInfoCallCount); + + configServerLocalTime += kLockExpiration; + getMockCatalog()->expectGetServerInfo([&getServerInfoCallCount]() { + getServerInfoCallCount++; + }, DistLockCatalog::ServerInfo(configServerLocalTime, lastElectionId)); + + OID lockTS; + // Make sure that overtake is now ok since electionId didn't change. + getMockCatalog()->expectOvertakeLock( + [this, &lockTS, ¤tLockDoc](StringData lockID, + const OID& lockSessionID, + const OID& currentHolderTS, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS("bar", lockID); lockTS = lockSessionID; ASSERT_EQUALS(currentLockDoc.getLockID(), currentHolderTS); ASSERT_EQUALS(getProcessID(), processId); ASSERT_EQUALS("foo", why); - }, currentLockDoc); // return arbitrary valid lock document, for testing purposes only. + }, + currentLockDoc); // return arbitrary valid lock document, for testing purposes only. - int unlockCallCount = 0; - OID unlockSessionIDPassed; + int unlockCallCount = 0; + OID unlockSessionIDPassed; - { - auto lockStatus = getMgr()->lock("bar", - "foo", - Milliseconds(0), - Milliseconds(0)); + { + auto lockStatus = getMgr()->lock("bar", "foo", Milliseconds(0), Milliseconds(0)); - ASSERT_OK(lockStatus.getStatus()); + ASSERT_OK(lockStatus.getStatus()); - getMockCatalog()->expectNoGrabLock(); - getMockCatalog()->expectUnLock( - [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { + getMockCatalog()->expectNoGrabLock(); + getMockCatalog()->expectUnLock( + [&unlockCallCount, &unlockSessionIDPassed](const OID& lockSessionID) { unlockCallCount++; unlockSessionIDPassed = lockSessionID; - }, Status::OK()); - } - - ASSERT_EQUALS(1, unlockCallCount); - ASSERT_EQUALS(lockTS, unlockSessionIDPassed); + }, + Status::OK()); } - /** - * Test scenario: - * 1. Attempt to grab lock fails because lock is already owned. - * 2. Try to get ping data and config server clock. - * 3. Since we don't have previous ping data to compare with, we cannot - * decide whether it's ok to overtake, so we can't. - * 4. Lock expiration has elapsed and the ping has not been updated since. - * 5. 2nd attempt to grab lock still fails for the same reason. - * 6. But since the ping is not fresh anymore, dist lock manager should overtake lock. - * 7. Attempt to overtake resulted in an error. - * 8. Check that unlock was called. - */ - TEST_F(ReplSetDistLockManagerFixture, LockOvertakingResultsInError) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { + ASSERT_EQUALS(1, unlockCallCount); + ASSERT_EQUALS(lockTS, unlockSessionIDPassed); +} + +/** + * Test scenario: + * 1. Attempt to grab lock fails because lock is already owned. + * 2. Try to get ping data and config server clock. + * 3. Since we don't have previous ping data to compare with, we cannot + * decide whether it's ok to overtake, so we can't. + * 4. Lock expiration has elapsed and the ping has not been updated since. + * 5. 2nd attempt to grab lock still fails for the same reason. + * 6. But since the ping is not fresh anymore, dist lock manager should overtake lock. + * 7. Attempt to overtake resulted in an error. + * 8. Check that unlock was called. + */ +TEST_F(ReplSetDistLockManagerFixture, LockOvertakingResultsInError) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); - - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); - - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); - - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t(), OID())); - - // First attempt will record the ping data. - { - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); + + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); + + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); - // Advance config server time to exceed lock expiration. - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t() + kLockExpiration + Milliseconds(1), OID())); - - OID lastTS; - getMockCatalog()->expectOvertakeLock( - [this, &lastTS, ¤tLockDoc] - (StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { + getMockCatalog()->expectGetServerInfo([]() {}, DistLockCatalog::ServerInfo(Date_t(), OID())); + + // First attempt will record the ping data. + { + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); + } + + // Advance config server time to exceed lock expiration. + getMockCatalog()->expectGetServerInfo( + []() {}, DistLockCatalog::ServerInfo(Date_t() + kLockExpiration + Milliseconds(1), OID())); + + OID lastTS; + getMockCatalog()->expectOvertakeLock( + [this, &lastTS, ¤tLockDoc](StringData lockID, + const OID& lockSessionID, + const OID& currentHolderTS, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS("bar", lockID); lastTS = lockSessionID; ASSERT_EQUALS(currentLockDoc.getLockID(), currentHolderTS); ASSERT_EQUALS(getProcessID(), processId); ASSERT_EQUALS("foo", why); - }, {ErrorCodes::NetworkTimeout, "bad test network"}); + }, + {ErrorCodes::NetworkTimeout, "bad test network"}); - OID unlockSessionIDPassed; + OID unlockSessionIDPassed; - stdx::mutex unlockMutex; - stdx::condition_variable unlockCV; - getMockCatalog()->expectUnLock( - [&unlockSessionIDPassed, &unlockMutex, &unlockCV]( - const OID& lockSessionID) { + stdx::mutex unlockMutex; + stdx::condition_variable unlockCV; + getMockCatalog()->expectUnLock( + [&unlockSessionIDPassed, &unlockMutex, &unlockCV](const OID& lockSessionID) { stdx::unique_lock<stdx::mutex> lk(unlockMutex); unlockSessionIDPassed = lockSessionID; unlockCV.notify_all(); - }, Status::OK()); + }, + Status::OK()); - // Second attempt should overtake lock. - auto lockStatus = getMgr()->lock("bar", - "foo", - Milliseconds(0), - Milliseconds(0)); + // Second attempt should overtake lock. + auto lockStatus = getMgr()->lock("bar", "foo", Milliseconds(0), Milliseconds(0)); - ASSERT_NOT_OK(lockStatus.getStatus()); + ASSERT_NOT_OK(lockStatus.getStatus()); - bool didTimeout = false; - { - stdx::unique_lock<stdx::mutex> lk(unlockMutex); - if (!unlockSessionIDPassed.isSet()) { - didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; - } + bool didTimeout = false; + { + stdx::unique_lock<stdx::mutex> lk(unlockMutex); + if (!unlockSessionIDPassed.isSet()) { + didTimeout = unlockCV.wait_for(lk, kUnlockTimeout) == stdx::cv_status::timeout; } + } - // Join the background thread before trying to call asserts. Shutdown calls - // stopPing and we don't care in this test. - getMockCatalog()->expectStopPing([](StringData){}, Status::OK()); - getMgr()->shutDown(); + // Join the background thread before trying to call asserts. Shutdown calls + // stopPing and we don't care in this test. + getMockCatalog()->expectStopPing([](StringData) {}, Status::OK()); + getMgr()->shutDown(); - // No assert until shutDown has been called to make sure that the background thread - // won't be trying to access the local variables that were captured by lamdas that - // may have gone out of scope when the assert unwinds the stack. - // No need to grab testMutex since there is only one thread running at this point. + // No assert until shutDown has been called to make sure that the background thread + // won't be trying to access the local variables that were captured by lamdas that + // may have gone out of scope when the assert unwinds the stack. + // No need to grab testMutex since there is only one thread running at this point. - ASSERT_FALSE(didTimeout); - ASSERT_EQUALS(lastTS, unlockSessionIDPassed); - } + ASSERT_FALSE(didTimeout); + ASSERT_EQUALS(lastTS, unlockSessionIDPassed); +} - /** - * Test scenario: - * 1. Attempt to grab lock fails because lock is already owned. - * 2. Try to get ping data and config server clock. - * 3. Since we don't have previous ping data to compare with, we cannot - * decide whether it's ok to overtake, so we can't. - * 4. Lock expiration has elapsed and the ping has not been updated since. - * 5. 2nd attempt to grab lock still fails for the same reason. - * 6. But since the ping is not fresh anymore, dist lock manager should overtake lock. - * 7. Attempt to overtake resulted failed because someone beat us into it. - */ - TEST_F(ReplSetDistLockManagerFixture, LockOvertakingFailed) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { +/** + * Test scenario: + * 1. Attempt to grab lock fails because lock is already owned. + * 2. Try to get ping data and config server clock. + * 3. Since we don't have previous ping data to compare with, we cannot + * decide whether it's ok to overtake, so we can't. + * 4. Lock expiration has elapsed and the ping has not been updated since. + * 5. 2nd attempt to grab lock still fails for the same reason. + * 6. But since the ping is not fresh anymore, dist lock manager should overtake lock. + * 7. Attempt to overtake resulted failed because someone beat us into it. + */ +TEST_F(ReplSetDistLockManagerFixture, LockOvertakingFailed) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); - - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); - - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); - - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t(), OID())); - - // First attempt will record the ping data. - { - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); + + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); + + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); - // Advance config server time to exceed lock expiration. - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(Date_t() + kLockExpiration + Milliseconds(1), OID())); - - // Second attempt should overtake lock. - getMockCatalog()->expectOvertakeLock([this, ¤tLockDoc] - (StringData lockID, - const OID& lockSessionID, - const OID& currentHolderTS, - StringData who, - StringData processId, - Date_t time, - StringData why) { + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); + + getMockCatalog()->expectGetServerInfo([]() {}, DistLockCatalog::ServerInfo(Date_t(), OID())); + + // First attempt will record the ping data. + { + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); + } + + // Advance config server time to exceed lock expiration. + getMockCatalog()->expectGetServerInfo( + []() {}, DistLockCatalog::ServerInfo(Date_t() + kLockExpiration + Milliseconds(1), OID())); + + // Second attempt should overtake lock. + getMockCatalog()->expectOvertakeLock( + [this, ¤tLockDoc](StringData lockID, + const OID& lockSessionID, + const OID& currentHolderTS, + StringData who, + StringData processId, + Date_t time, + StringData why) { ASSERT_EQUALS("bar", lockID); ASSERT_EQUALS(currentLockDoc.getLockID(), currentHolderTS); ASSERT_EQUALS(getProcessID(), processId); ASSERT_EQUALS("foo", why); - }, {ErrorCodes::LockStateChangeFailed, "nmod 0"}); - - { - auto status = getMgr()->lock("bar", - "foo", - Milliseconds(0), - Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + }, + {ErrorCodes::LockStateChangeFailed, "nmod 0"}); + + { + auto status = getMgr()->lock("bar", "foo", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); } +} - /** - * Test scenario: - * 1. Attempt to grab lock fails because lock is already owned. - * 2. Try to get ping data and config server clock. - * 3. Since we don't have previous ping data to compare with, we cannot - * decide whether it's ok to overtake, so we can't. - * 4. Lock expiration has elapsed and the ping has not been updated since. - * 5. 2nd attempt to grab lock still fails for the same reason. - * 6. But since the ping is not fresh anymore, dist lock manager should overtake lock. - * 7. Attempt to overtake resulted failed because someone beat us into it. - */ - TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfConfigServerClockGoesBackwards) { - getMockCatalog()->expectGrabLock([]( - StringData, const OID&, StringData, StringData, Date_t, StringData) { +/** + * Test scenario: + * 1. Attempt to grab lock fails because lock is already owned. + * 2. Try to get ping data and config server clock. + * 3. Since we don't have previous ping data to compare with, we cannot + * decide whether it's ok to overtake, so we can't. + * 4. Lock expiration has elapsed and the ping has not been updated since. + * 5. 2nd attempt to grab lock still fails for the same reason. + * 6. But since the ping is not fresh anymore, dist lock manager should overtake lock. + * 7. Attempt to overtake resulted failed because someone beat us into it. + */ +TEST_F(ReplSetDistLockManagerFixture, CannotOvertakeIfConfigServerClockGoesBackwards) { + getMockCatalog()->expectGrabLock( + [](StringData, const OID&, StringData, StringData, Date_t, StringData) { // Don't care - }, {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - - LocksType currentLockDoc; - currentLockDoc.setName("bar"); - currentLockDoc.setState(LocksType::LOCKED); - currentLockDoc.setProcess("otherProcess"); - currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); - currentLockDoc.setWho("me"); - currentLockDoc.setWhy("why"); - - getMockCatalog()->expectGetLockByName([](StringData name) { - ASSERT_EQUALS("bar", name); - }, currentLockDoc); - - LockpingsType pingDoc; - pingDoc.setProcess("otherProcess"); - pingDoc.setPing(Date_t()); - - getMockCatalog()->expectGetPing([](StringData process) { - ASSERT_EQUALS("otherProcess", process); - }, pingDoc); - - Date_t configClock(Date_t::now()); - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(configClock, OID())); - - // First attempt will record the ping data. - { - auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + }, + {ErrorCodes::LockStateChangeFailed, "nMod 0"}); - // Make config server time go backwards by lock expiration duration. - getMockCatalog()->expectGetServerInfo([]() { - }, DistLockCatalog::ServerInfo(configClock - kLockExpiration - Milliseconds(1), OID())); - - // Second attempt should not overtake lock. - { - auto status = getMgr()->lock("bar", - "foo", - Milliseconds(0), - Milliseconds(0)).getStatus(); - ASSERT_NOT_OK(status); - ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); - } + LocksType currentLockDoc; + currentLockDoc.setName("bar"); + currentLockDoc.setState(LocksType::LOCKED); + currentLockDoc.setProcess("otherProcess"); + currentLockDoc.setLockID(OID("5572007fda9e476582bf3716")); + currentLockDoc.setWho("me"); + currentLockDoc.setWhy("why"); + + getMockCatalog()->expectGetLockByName([](StringData name) { ASSERT_EQUALS("bar", name); }, + currentLockDoc); + + LockpingsType pingDoc; + pingDoc.setProcess("otherProcess"); + pingDoc.setPing(Date_t()); + + getMockCatalog()->expectGetPing( + [](StringData process) { ASSERT_EQUALS("otherProcess", process); }, pingDoc); + + Date_t configClock(Date_t::now()); + getMockCatalog()->expectGetServerInfo([]() {}, DistLockCatalog::ServerInfo(configClock, OID())); + + // First attempt will record the ping data. + { + auto status = getMgr()->lock("bar", "", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); + } + + // Make config server time go backwards by lock expiration duration. + getMockCatalog()->expectGetServerInfo([]() { + }, DistLockCatalog::ServerInfo(configClock - kLockExpiration - Milliseconds(1), OID())); + + // Second attempt should not overtake lock. + { + auto status = getMgr()->lock("bar", "foo", Milliseconds(0), Milliseconds(0)).getStatus(); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(ErrorCodes::LockBusy, status.code()); } +} -} // unnamed namespace -} // namespace mongo +} // unnamed namespace +} // namespace mongo diff --git a/src/mongo/s/catalog/type_actionlog.cpp b/src/mongo/s/catalog/type_actionlog.cpp index 49dd09b6958..1cefd2774dc 100644 --- a/src/mongo/s/catalog/type_actionlog.cpp +++ b/src/mongo/s/catalog/type_actionlog.cpp @@ -32,124 +32,129 @@ namespace mongo { - using std::string; +using std::string; - using mongo::str::stream; +using mongo::str::stream; - const std::string ActionLogType::ConfigNS = "config.actionlog"; +const std::string ActionLogType::ConfigNS = "config.actionlog"; - const BSONField<std::string> ActionLogType::server("server"); - const BSONField<std::string> ActionLogType::what("what"); - const BSONField<Date_t> ActionLogType::time("time"); - const BSONField<BSONObj> ActionLogType::details("details"); +const BSONField<std::string> ActionLogType::server("server"); +const BSONField<std::string> ActionLogType::what("what"); +const BSONField<Date_t> ActionLogType::time("time"); +const BSONField<BSONObj> ActionLogType::details("details"); - ActionLogType::ActionLogType() { - clear(); - } +ActionLogType::ActionLogType() { + clear(); +} - ActionLogType::~ActionLogType() { - } +ActionLogType::~ActionLogType() {} - bool ActionLogType::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isServerSet) { - *errMsg = stream() << "missing " << server.name() << " field"; - return false; - } - if (!_isTimeSet) { - *errMsg = stream() << "missing " << time.name() << " field"; - return false; - } - if (!_isWhatSet) { - *errMsg = stream() << "missing " << what.name() << " field"; - return false; - } - if (!_isDetailsSet) { - *errMsg = stream() << "missing " << details.name() << " field"; - return false; - } - - return true; +bool ActionLogType::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BSONObj ActionLogType::toBSON() const { - BSONObjBuilder builder; - - if (_isServerSet) builder.append(server(), _server); - if (_isTimeSet) builder.append(time(), _time); - if (_isWhatSet) builder.append(what(), _what); - if (_isDetailsSet) builder.append(details(), _details); - - return builder.obj(); + // All the mandatory fields must be present. + if (!_isServerSet) { + *errMsg = stream() << "missing " << server.name() << " field"; + return false; + } + if (!_isTimeSet) { + *errMsg = stream() << "missing " << time.name() << " field"; + return false; + } + if (!_isWhatSet) { + *errMsg = stream() << "missing " << what.name() << " field"; + return false; + } + if (!_isDetailsSet) { + *errMsg = stream() << "missing " << details.name() << " field"; + return false; } - bool ActionLogType::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + return true; +} - std::string dummy; - if (!errMsg) errMsg = &dummy; +BSONObj ActionLogType::toBSON() const { + BSONObjBuilder builder; - FieldParser::FieldState fieldState; + if (_isServerSet) + builder.append(server(), _server); + if (_isTimeSet) + builder.append(time(), _time); + if (_isWhatSet) + builder.append(what(), _what); + if (_isDetailsSet) + builder.append(details(), _details); - fieldState = FieldParser::extract(source, server, &_server, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isServerSet = fieldState == FieldParser::FIELD_SET; + return builder.obj(); +} - fieldState = FieldParser::extract(source, time, &_time, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isTimeSet = fieldState == FieldParser::FIELD_SET; +bool ActionLogType::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - fieldState = FieldParser::extract(source, what, &_what, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isWhatSet = fieldState == FieldParser::FIELD_SET; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - fieldState = FieldParser::extract(source, details, &_details, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isDetailsSet = fieldState == FieldParser::FIELD_SET; + FieldParser::FieldState fieldState; - return true; - } + fieldState = FieldParser::extract(source, server, &_server, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isServerSet = fieldState == FieldParser::FIELD_SET; - void ActionLogType::clear() { + fieldState = FieldParser::extract(source, time, &_time, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isTimeSet = fieldState == FieldParser::FIELD_SET; - _server.clear(); - _isServerSet = false; + fieldState = FieldParser::extract(source, what, &_what, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isWhatSet = fieldState == FieldParser::FIELD_SET; - _what.clear(); - _isWhatSet = false; + fieldState = FieldParser::extract(source, details, &_details, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isDetailsSet = fieldState == FieldParser::FIELD_SET; - _time = Date_t(); - _isTimeSet = false; + return true; +} - _details = BSONObj(); - _isDetailsSet = false; +void ActionLogType::clear() { + _server.clear(); + _isServerSet = false; - } + _what.clear(); + _isWhatSet = false; - void ActionLogType::cloneTo(ActionLogType* other) const { - other->clear(); + _time = Date_t(); + _isTimeSet = false; - other->_server = _server; - other->_isServerSet = _isServerSet; + _details = BSONObj(); + _isDetailsSet = false; +} - other->_what = _what; - other->_isWhatSet = _isWhatSet; +void ActionLogType::cloneTo(ActionLogType* other) const { + other->clear(); - other->_time = _time; - other->_isTimeSet = _isTimeSet; + other->_server = _server; + other->_isServerSet = _isServerSet; - other->_details = _details; - other->_isDetailsSet = _isDetailsSet; + other->_what = _what; + other->_isWhatSet = _isWhatSet; - } + other->_time = _time; + other->_isTimeSet = _isTimeSet; - std::string ActionLogType::toString() const { - return toBSON().toString(); - } + other->_details = _details; + other->_isDetailsSet = _isDetailsSet; +} + +std::string ActionLogType::toString() const { + return toBSON().toString(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/type_actionlog.h b/src/mongo/s/catalog/type_actionlog.h index bba7703371c..b7ff387a823 100644 --- a/src/mongo/s/catalog/type_actionlog.h +++ b/src/mongo/s/catalog/type_actionlog.h @@ -36,169 +36,184 @@ namespace mongo { +/** + * This class represents the layout and contents of documents contained in the + * config.actionlog collection. All manipulation of documents coming from that + * collection should be done with this class. + * + * Usage Example: + * + * // Contact the config. 'conn' has been obtained before. + * DBClientBase* conn; + * BSONObj query = QUERY(ActionLogType::exampleField("exampleFieldName")); + * exampleDoc = conn->findOne(ActionLogType::ConfigNS, query); + * + * // Process the response. + * ActionLogType exampleType; + * std::string errMsg; + * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { + * // Can't use 'exampleType'. Take action. + * } + * // use 'exampleType' + * + */ +class ActionLogType { + MONGO_DISALLOW_COPYING(ActionLogType); + +public: + // + // schema declarations + // + + // Name of the actionlog collection in the config server. + static const std::string ConfigNS; + + // Field names and types in the actionlog collection type. + static const BSONField<std::string> server; + static const BSONField<std::string> what; + static const BSONField<Date_t> time; + static const BSONField<BSONObj> details; + + // Common field names included under details + static const BSONField<int> candidateChunks; + static const BSONField<int> chunksMoved; + static const BSONField<bool> didError; + static const BSONField<long long> executionTimeMicros; + static const BSONField<std::string> errmsg; + + + // + // actionlog type methods + // + + ActionLogType(); + ~ActionLogType(); + /** - * This class represents the layout and contents of documents contained in the - * config.actionlog collection. All manipulation of documents coming from that - * collection should be done with this class. - * - * Usage Example: - * - * // Contact the config. 'conn' has been obtained before. - * DBClientBase* conn; - * BSONObj query = QUERY(ActionLogType::exampleField("exampleFieldName")); - * exampleDoc = conn->findOne(ActionLogType::ConfigNS, query); - * - * // Process the response. - * ActionLogType exampleType; - * std::string errMsg; - * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { - * // Can't use 'exampleType'. Take action. - * } - * // use 'exampleType' - * + * Returns true if all the mandatory fields are present and have valid + * representations. Otherwise returns false and fills in the optional 'errMsg' string. */ - class ActionLogType { - MONGO_DISALLOW_COPYING(ActionLogType); - public: - - // - // schema declarations - // - - // Name of the actionlog collection in the config server. - static const std::string ConfigNS; + bool isValid(std::string* errMsg) const; - // Field names and types in the actionlog collection type. - static const BSONField<std::string> server; - static const BSONField<std::string> what; - static const BSONField<Date_t> time; - static const BSONField<BSONObj> details; - - // Common field names included under details - static const BSONField<int> candidateChunks; - static const BSONField<int> chunksMoved; - static const BSONField<bool> didError; - static const BSONField<long long> executionTimeMicros; - static const BSONField<std::string> errmsg; + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; + void buildDetails(); - // - // actionlog type methods - // - - ActionLogType(); - ~ActionLogType(); - - /** - * Returns true if all the mandatory fields are present and have valid - * representations. Otherwise returns false and fills in the optional 'errMsg' string. - */ - bool isValid(std::string* errMsg) const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - void buildDetails(); - - /** - * Clears and populates the internal state using the 'source' BSON object if the - * latter contains valid values. Otherwise sets errMsg and returns false. - */ - bool parseBSON(const BSONObj& source, std::string* errMsg); - - /** - * Clears the internal state. - */ - void clear(); - - /** - * Copies all the fields present in 'this' to 'other'. - */ - void cloneTo(ActionLogType* other) const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - // - // individual field accessors - // - - void setServer(StringData server) { - _server = server.toString(); - _isServerSet = true; - } - - void unsetServer() { _isServerSet = false; } - - bool isServerSet() const { return _isServerSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getServer() const { - dassert(_isServerSet); - return _what; - } - - void setWhat(StringData what) { - _what = what.toString(); - _isWhatSet = true; - } - - void unsetWhat() { _isWhatSet = false; } - - bool isWhatSet() const { return _isWhatSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getWhat() const { - dassert(_isWhatSet); - return _what; - } - - void setTime(const Date_t time) { - _time = time; - _isTimeSet = true; - } - - void unsetTime() { _isTimeSet = false; } - - bool isTimeSet() const { return _isTimeSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const Date_t getTime() const { - dassert(_isTimeSet); - return _time; - } - - void setDetails(const BSONObj& details) { - _details = details.getOwned(); - _isDetailsSet = true; - } - - void unsetDetails() { _isDetailsSet = false; } - - bool isDetailsSet() const { return _isDetailsSet; } + /** + * Clears and populates the internal state using the 'source' BSON object if the + * latter contains valid values. Otherwise sets errMsg and returns false. + */ + bool parseBSON(const BSONObj& source, std::string* errMsg); - // Calling get*() methods when the member is not set results in undefined behavior - const BSONObj getDetails() const { - dassert(_isDetailsSet); - return _details; - } - - private: - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - std::string _server; // (M) hostname of server that we are making the change on. - // Does not include port. - bool _isServerSet; - std::string _what; // (M) what the action being performed was. - bool _isWhatSet; - Date_t _time; // (M) time this change was made - bool _isTimeSet; - BSONObj _details; // (M) A BSONObj containing extra information about some operations - bool _isDetailsSet; - - }; - -} // namespace mongo + /** + * Clears the internal state. + */ + void clear(); + + /** + * Copies all the fields present in 'this' to 'other'. + */ + void cloneTo(ActionLogType* other) const; + + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; + + // + // individual field accessors + // + + void setServer(StringData server) { + _server = server.toString(); + _isServerSet = true; + } + + void unsetServer() { + _isServerSet = false; + } + + bool isServerSet() const { + return _isServerSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getServer() const { + dassert(_isServerSet); + return _what; + } + + void setWhat(StringData what) { + _what = what.toString(); + _isWhatSet = true; + } + + void unsetWhat() { + _isWhatSet = false; + } + + bool isWhatSet() const { + return _isWhatSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getWhat() const { + dassert(_isWhatSet); + return _what; + } + + void setTime(const Date_t time) { + _time = time; + _isTimeSet = true; + } + + void unsetTime() { + _isTimeSet = false; + } + + bool isTimeSet() const { + return _isTimeSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const Date_t getTime() const { + dassert(_isTimeSet); + return _time; + } + + void setDetails(const BSONObj& details) { + _details = details.getOwned(); + _isDetailsSet = true; + } + + void unsetDetails() { + _isDetailsSet = false; + } + + bool isDetailsSet() const { + return _isDetailsSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const BSONObj getDetails() const { + dassert(_isDetailsSet); + return _details; + } + +private: + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + std::string _server; // (M) hostname of server that we are making the change on. + // Does not include port. + bool _isServerSet; + std::string _what; // (M) what the action being performed was. + bool _isWhatSet; + Date_t _time; // (M) time this change was made + bool _isTimeSet; + BSONObj _details; // (M) A BSONObj containing extra information about some operations + bool _isDetailsSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_changelog.cpp b/src/mongo/s/catalog/type_changelog.cpp index 40db60dbf94..d31b5a83806 100644 --- a/src/mongo/s/catalog/type_changelog.cpp +++ b/src/mongo/s/catalog/type_changelog.cpp @@ -35,171 +35,182 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; +using mongoutils::str::stream; - const std::string ChangelogType::ConfigNS = "config.changelog"; +const std::string ChangelogType::ConfigNS = "config.changelog"; - const BSONField<std::string> ChangelogType::changeID("_id"); - const BSONField<std::string> ChangelogType::server("server"); - const BSONField<std::string> ChangelogType::clientAddr("clientAddr"); - const BSONField<Date_t> ChangelogType::time("time"); - const BSONField<std::string> ChangelogType::what("what"); - const BSONField<std::string> ChangelogType::ns("ns"); - const BSONField<BSONObj> ChangelogType::details("details"); +const BSONField<std::string> ChangelogType::changeID("_id"); +const BSONField<std::string> ChangelogType::server("server"); +const BSONField<std::string> ChangelogType::clientAddr("clientAddr"); +const BSONField<Date_t> ChangelogType::time("time"); +const BSONField<std::string> ChangelogType::what("what"); +const BSONField<std::string> ChangelogType::ns("ns"); +const BSONField<BSONObj> ChangelogType::details("details"); - ChangelogType::ChangelogType() { - clear(); - } +ChangelogType::ChangelogType() { + clear(); +} - ChangelogType::~ChangelogType() { - } +ChangelogType::~ChangelogType() {} - bool ChangelogType::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isChangeIDSet) { - *errMsg = stream() << "missing " << changeID.name() << " field"; - return false; - } - if (!_isServerSet) { - *errMsg = stream() << "missing " << server.name() << " field"; - return false; - } - if (!_isClientAddrSet) { - *errMsg = stream() << "missing " << clientAddr.name() << " field"; - return false; - } - if (!_isTimeSet) { - *errMsg = stream() << "missing " << time.name() << " field"; - return false; - } - if (!_isWhatSet) { - *errMsg = stream() << "missing " << what.name() << " field"; - return false; - } - if (!_isNsSet) { - *errMsg = stream() << "missing " << ns.name() << " field"; - return false; - } - if (!_isDetailsSet) { - *errMsg = stream() << "missing " << details.name() << " field"; - return false; - } - - return true; +bool ChangelogType::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BSONObj ChangelogType::toBSON() const { - BSONObjBuilder builder; - - if (_isChangeIDSet) builder.append(changeID(), _changeID); - if (_isServerSet) builder.append(server(), _server); - if (_isClientAddrSet) builder.append(clientAddr(), _clientAddr); - if (_isTimeSet) builder.append(time(), _time); - if (_isWhatSet) builder.append(what(), _what); - if (_isNsSet) builder.append(ns(), _ns); - if (_isDetailsSet) builder.append(details(), _details); - - return builder.obj(); + // All the mandatory fields must be present. + if (!_isChangeIDSet) { + *errMsg = stream() << "missing " << changeID.name() << " field"; + return false; + } + if (!_isServerSet) { + *errMsg = stream() << "missing " << server.name() << " field"; + return false; + } + if (!_isClientAddrSet) { + *errMsg = stream() << "missing " << clientAddr.name() << " field"; + return false; + } + if (!_isTimeSet) { + *errMsg = stream() << "missing " << time.name() << " field"; + return false; + } + if (!_isWhatSet) { + *errMsg = stream() << "missing " << what.name() << " field"; + return false; + } + if (!_isNsSet) { + *errMsg = stream() << "missing " << ns.name() << " field"; + return false; + } + if (!_isDetailsSet) { + *errMsg = stream() << "missing " << details.name() << " field"; + return false; } - bool ChangelogType::parseBSON(const BSONObj& source, string* errMsg) { - clear(); - - std::string dummy; - if (!errMsg) errMsg = &dummy; + return true; +} + +BSONObj ChangelogType::toBSON() const { + BSONObjBuilder builder; - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, changeID, &_changeID, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isChangeIDSet = fieldState == FieldParser::FIELD_SET; + if (_isChangeIDSet) + builder.append(changeID(), _changeID); + if (_isServerSet) + builder.append(server(), _server); + if (_isClientAddrSet) + builder.append(clientAddr(), _clientAddr); + if (_isTimeSet) + builder.append(time(), _time); + if (_isWhatSet) + builder.append(what(), _what); + if (_isNsSet) + builder.append(ns(), _ns); + if (_isDetailsSet) + builder.append(details(), _details); - fieldState = FieldParser::extract(source, server, &_server, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isServerSet = fieldState == FieldParser::FIELD_SET; + return builder.obj(); +} - fieldState = FieldParser::extract(source, clientAddr, &_clientAddr, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isClientAddrSet = fieldState == FieldParser::FIELD_SET; +bool ChangelogType::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - fieldState = FieldParser::extract(source, time, &_time, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isTimeSet = fieldState == FieldParser::FIELD_SET; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - fieldState = FieldParser::extract(source, what, &_what, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isWhatSet = fieldState == FieldParser::FIELD_SET; + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, changeID, &_changeID, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isChangeIDSet = fieldState == FieldParser::FIELD_SET; - fieldState = FieldParser::extract(source, ns, &_ns, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isNsSet = fieldState == FieldParser::FIELD_SET; + fieldState = FieldParser::extract(source, server, &_server, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isServerSet = fieldState == FieldParser::FIELD_SET; - fieldState = FieldParser::extract(source, details, &_details, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isDetailsSet = fieldState == FieldParser::FIELD_SET; + fieldState = FieldParser::extract(source, clientAddr, &_clientAddr, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isClientAddrSet = fieldState == FieldParser::FIELD_SET; - return true; - } + fieldState = FieldParser::extract(source, time, &_time, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isTimeSet = fieldState == FieldParser::FIELD_SET; - void ChangelogType::clear() { + fieldState = FieldParser::extract(source, what, &_what, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isWhatSet = fieldState == FieldParser::FIELD_SET; - _changeID.clear(); - _isChangeIDSet = false; + fieldState = FieldParser::extract(source, ns, &_ns, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isNsSet = fieldState == FieldParser::FIELD_SET; - _server.clear(); - _isServerSet = false; + fieldState = FieldParser::extract(source, details, &_details, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isDetailsSet = fieldState == FieldParser::FIELD_SET; - _clientAddr.clear(); - _isClientAddrSet = false; + return true; +} - _time = Date_t(); - _isTimeSet = false; +void ChangelogType::clear() { + _changeID.clear(); + _isChangeIDSet = false; - _what.clear(); - _isWhatSet = false; + _server.clear(); + _isServerSet = false; - _ns.clear(); - _isNsSet = false; + _clientAddr.clear(); + _isClientAddrSet = false; - _details = BSONObj(); - _isDetailsSet = false; + _time = Date_t(); + _isTimeSet = false; - } + _what.clear(); + _isWhatSet = false; - void ChangelogType::cloneTo(ChangelogType* other) const { - other->clear(); + _ns.clear(); + _isNsSet = false; - other->_changeID = _changeID; - other->_isChangeIDSet = _isChangeIDSet; + _details = BSONObj(); + _isDetailsSet = false; +} - other->_server = _server; - other->_isServerSet = _isServerSet; +void ChangelogType::cloneTo(ChangelogType* other) const { + other->clear(); - other->_clientAddr = _clientAddr; - other->_isClientAddrSet = _isClientAddrSet; + other->_changeID = _changeID; + other->_isChangeIDSet = _isChangeIDSet; - other->_time = _time; - other->_isTimeSet = _isTimeSet; + other->_server = _server; + other->_isServerSet = _isServerSet; - other->_what = _what; - other->_isWhatSet = _isWhatSet; + other->_clientAddr = _clientAddr; + other->_isClientAddrSet = _isClientAddrSet; - other->_ns = _ns; - other->_isNsSet = _isNsSet; + other->_time = _time; + other->_isTimeSet = _isTimeSet; + + other->_what = _what; + other->_isWhatSet = _isWhatSet; - other->_details = _details; - other->_isDetailsSet = _isDetailsSet; + other->_ns = _ns; + other->_isNsSet = _isNsSet; - } + other->_details = _details; + other->_isDetailsSet = _isDetailsSet; +} - std::string ChangelogType::toString() const { - return toBSON().toString(); - } +std::string ChangelogType::toString() const { + return toBSON().toString(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/type_changelog.h b/src/mongo/s/catalog/type_changelog.h index 1124f6a16c3..7a0360fd7a1 100644 --- a/src/mongo/s/catalog/type_changelog.h +++ b/src/mongo/s/catalog/type_changelog.h @@ -35,214 +35,243 @@ namespace mongo { +/** + * This class represents the layout and contents of documents contained in the + * config.changelog collection. All manipulation of documents coming from that + * collection should be done with this class. + * + * Usage Example: + * + * // Contact the config. 'conn' has been obtained before. + * DBClientBase* conn; + * BSONObj query = QUERY(ChangelogType::exampleField("exampleFieldName")); + * exampleDoc = conn->findOne(ChangelogType::ConfigNS, query); + * + * // Process the response. + * ChangelogType exampleType; + * std::string errMsg; + * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { + * // Can't use 'exampleType'. Take action. + * } + * // use 'exampleType' + * + */ +class ChangelogType { + MONGO_DISALLOW_COPYING(ChangelogType); + +public: + // + // schema declarations + // + + // Name of the changelog collection in the config server. + static const std::string ConfigNS; + + // Field names and types in the changelog collection type. + static const BSONField<std::string> changeID; + static const BSONField<std::string> server; + static const BSONField<std::string> clientAddr; + static const BSONField<Date_t> time; + static const BSONField<std::string> what; + static const BSONField<std::string> ns; + static const BSONField<BSONObj> details; + + // + // changelog type methods + // + + ChangelogType(); + ~ChangelogType(); + /** - * This class represents the layout and contents of documents contained in the - * config.changelog collection. All manipulation of documents coming from that - * collection should be done with this class. - * - * Usage Example: - * - * // Contact the config. 'conn' has been obtained before. - * DBClientBase* conn; - * BSONObj query = QUERY(ChangelogType::exampleField("exampleFieldName")); - * exampleDoc = conn->findOne(ChangelogType::ConfigNS, query); - * - * // Process the response. - * ChangelogType exampleType; - * std::string errMsg; - * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { - * // Can't use 'exampleType'. Take action. - * } - * // use 'exampleType' - * + * Returns true if all the mandatory fields are present and have valid + * representations. Otherwise returns false and fills in the optional 'errMsg' string. */ - class ChangelogType { - MONGO_DISALLOW_COPYING(ChangelogType); - public: - - // - // schema declarations - // - - // Name of the changelog collection in the config server. - static const std::string ConfigNS; - - // Field names and types in the changelog collection type. - static const BSONField<std::string> changeID; - static const BSONField<std::string> server; - static const BSONField<std::string> clientAddr; - static const BSONField<Date_t> time; - static const BSONField<std::string> what; - static const BSONField<std::string> ns; - static const BSONField<BSONObj> details; - - // - // changelog type methods - // - - ChangelogType(); - ~ChangelogType(); - - /** - * Returns true if all the mandatory fields are present and have valid - * representations. Otherwise returns false and fills in the optional 'errMsg' string. - */ - bool isValid(std::string* errMsg) const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Clears and populates the internal state using the 'source' BSON object if the - * latter contains valid values. Otherwise sets errMsg and returns false. - */ - bool parseBSON(const BSONObj& source, std::string* errMsg); - - /** - * Clears the internal state. - */ - void clear(); - - /** - * Copies all the fields present in 'this' to 'other'. - */ - void cloneTo(ChangelogType* other) const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - // - // individual field accessors - // + bool isValid(std::string* errMsg) const; - // Mandatory Fields - void setChangeID(StringData changeID) { - _changeID = changeID.toString(); - _isChangeIDSet = true; - } - - void unsetChangeID() { _isChangeIDSet = false; } + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; + + /** + * Clears and populates the internal state using the 'source' BSON object if the + * latter contains valid values. Otherwise sets errMsg and returns false. + */ + bool parseBSON(const BSONObj& source, std::string* errMsg); + + /** + * Clears the internal state. + */ + void clear(); - bool isChangeIDSet() const { return _isChangeIDSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getChangeID() const { - dassert(_isChangeIDSet); - return _changeID; - } - - void setServer(StringData server) { - _server = server.toString(); - _isServerSet = true; - } - - void unsetServer() { _isServerSet = false; } - - bool isServerSet() const { return _isServerSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getServer() const { - dassert(_isServerSet); - return _server; - } - - void setClientAddr(StringData clientAddr) { - _clientAddr = clientAddr.toString(); - _isClientAddrSet = true; - } - - void unsetClientAddr() { _isClientAddrSet = false; } - - bool isClientAddrSet() const { return _isClientAddrSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getClientAddr() const { - dassert(_isClientAddrSet); - return _clientAddr; - } - - void setTime(const Date_t time) { - _time = time; - _isTimeSet = true; - } - - void unsetTime() { _isTimeSet = false; } - - bool isTimeSet() const { return _isTimeSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const Date_t getTime() const { - dassert(_isTimeSet); - return _time; - } - - void setWhat(StringData what) { - _what = what.toString(); - _isWhatSet = true; - } - - void unsetWhat() { _isWhatSet = false; } - - bool isWhatSet() const { return _isWhatSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getWhat() const { - dassert(_isWhatSet); - return _what; - } - - void setNS(StringData ns) { - _ns = ns.toString(); - _isNsSet = true; - } - - void unsetNS() { _isNsSet = false; } - - bool isNSSet() const { return _isNsSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getNS() const { - dassert(_isNsSet); - return _ns; - } - - void setDetails(const BSONObj& details) { - _details = details.getOwned(); - _isDetailsSet = true; - } - - void unsetDetails() { _isDetailsSet = false; } - - bool isDetailsSet() const { return _isDetailsSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const BSONObj getDetails() const { - dassert(_isDetailsSet); - return _details; - } + /** + * Copies all the fields present in 'this' to 'other'. + */ + void cloneTo(ChangelogType* other) const; - // Optional Fields - - private: - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - std::string _changeID; // (M) id for this change "<hostname>-<current_time>-<increment>" - bool _isChangeIDSet; - std::string _server; // (M) hostname of server that we are making the change on. Does not include port. - bool _isServerSet; - std::string _clientAddr; // (M) hostname:port of the client that made this change - bool _isClientAddrSet; - Date_t _time; // (M) time this change was made - bool _isTimeSet; - std::string _what; // (M) description of the change - bool _isWhatSet; - std::string _ns; // (M) database or collection this change applies to - bool _isNsSet; - BSONObj _details; // (M) A BSONObj containing extra information about some operations - bool _isDetailsSet; - }; - -} // namespace mongo + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; + + // + // individual field accessors + // + + // Mandatory Fields + void setChangeID(StringData changeID) { + _changeID = changeID.toString(); + _isChangeIDSet = true; + } + + void unsetChangeID() { + _isChangeIDSet = false; + } + + bool isChangeIDSet() const { + return _isChangeIDSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getChangeID() const { + dassert(_isChangeIDSet); + return _changeID; + } + + void setServer(StringData server) { + _server = server.toString(); + _isServerSet = true; + } + + void unsetServer() { + _isServerSet = false; + } + + bool isServerSet() const { + return _isServerSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getServer() const { + dassert(_isServerSet); + return _server; + } + + void setClientAddr(StringData clientAddr) { + _clientAddr = clientAddr.toString(); + _isClientAddrSet = true; + } + + void unsetClientAddr() { + _isClientAddrSet = false; + } + + bool isClientAddrSet() const { + return _isClientAddrSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getClientAddr() const { + dassert(_isClientAddrSet); + return _clientAddr; + } + + void setTime(const Date_t time) { + _time = time; + _isTimeSet = true; + } + + void unsetTime() { + _isTimeSet = false; + } + + bool isTimeSet() const { + return _isTimeSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const Date_t getTime() const { + dassert(_isTimeSet); + return _time; + } + + void setWhat(StringData what) { + _what = what.toString(); + _isWhatSet = true; + } + + void unsetWhat() { + _isWhatSet = false; + } + + bool isWhatSet() const { + return _isWhatSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getWhat() const { + dassert(_isWhatSet); + return _what; + } + + void setNS(StringData ns) { + _ns = ns.toString(); + _isNsSet = true; + } + + void unsetNS() { + _isNsSet = false; + } + + bool isNSSet() const { + return _isNsSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getNS() const { + dassert(_isNsSet); + return _ns; + } + + void setDetails(const BSONObj& details) { + _details = details.getOwned(); + _isDetailsSet = true; + } + + void unsetDetails() { + _isDetailsSet = false; + } + + bool isDetailsSet() const { + return _isDetailsSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const BSONObj getDetails() const { + dassert(_isDetailsSet); + return _details; + } + + // Optional Fields + +private: + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + std::string _changeID; // (M) id for this change "<hostname>-<current_time>-<increment>" + bool _isChangeIDSet; + std::string + _server; // (M) hostname of server that we are making the change on. Does not include port. + bool _isServerSet; + std::string _clientAddr; // (M) hostname:port of the client that made this change + bool _isClientAddrSet; + Date_t _time; // (M) time this change was made + bool _isTimeSet; + std::string _what; // (M) description of the change + bool _isWhatSet; + std::string _ns; // (M) database or collection this change applies to + bool _isNsSet; + BSONObj _details; // (M) A BSONObj containing extra information about some operations + bool _isDetailsSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_changelog_test.cpp b/src/mongo/s/catalog/type_changelog_test.cpp index 0844b231497..22d6746dfcc 100644 --- a/src/mongo/s/catalog/type_changelog_test.cpp +++ b/src/mongo/s/catalog/type_changelog_test.cpp @@ -34,145 +34,145 @@ namespace { - using std::string; - using mongo::ChangelogType; - using mongo::BSONObj; - using mongo::Date_t; +using std::string; +using mongo::ChangelogType; +using mongo::BSONObj; +using mongo::Date_t; - TEST(ChangelogType, Empty) { - ChangelogType logEntry; - BSONObj emptyObj = BSONObj(); - string errMsg; - ASSERT(logEntry.parseBSON(emptyObj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(logEntry.isValid(NULL)); - } +TEST(ChangelogType, Empty) { + ChangelogType logEntry; + BSONObj emptyObj = BSONObj(); + string errMsg; + ASSERT(logEntry.parseBSON(emptyObj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(logEntry.isValid(NULL)); +} - TEST(ChangelogType, Valid) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") << - ChangelogType::server("host.local") << - ChangelogType::clientAddr("192.168.0.189:51128") << - ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << - ChangelogType::what("split") << - ChangelogType::ns("test.test") << - ChangelogType::details(BSON("dummy" << "info"))); - string errMsg; - ASSERT(logEntry.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(logEntry.isValid(NULL)); - ASSERT_EQUALS(logEntry.getChangeID(), "host.local-2012-11-21T19:14:10-8"); - ASSERT_EQUALS(logEntry.getServer(), "host.local"); - ASSERT_EQUALS(logEntry.getClientAddr(), "192.168.0.189:51128"); - ASSERT_EQUALS(logEntry.getTime(), Date_t::fromMillisSinceEpoch(1)); - ASSERT_EQUALS(logEntry.getWhat(), "split"); - ASSERT_EQUALS(logEntry.getNS(), "test.test"); - ASSERT_EQUALS(logEntry.getDetails(), BSON("dummy" << "info")); - } +TEST(ChangelogType, Valid) { + ChangelogType logEntry; + BSONObj obj = BSON( + ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") + << ChangelogType::server("host.local") << ChangelogType::clientAddr("192.168.0.189:51128") + << ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << ChangelogType::what("split") + << ChangelogType::ns("test.test") << ChangelogType::details(BSON("dummy" + << "info"))); + string errMsg; + ASSERT(logEntry.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(logEntry.isValid(NULL)); + ASSERT_EQUALS(logEntry.getChangeID(), "host.local-2012-11-21T19:14:10-8"); + ASSERT_EQUALS(logEntry.getServer(), "host.local"); + ASSERT_EQUALS(logEntry.getClientAddr(), "192.168.0.189:51128"); + ASSERT_EQUALS(logEntry.getTime(), Date_t::fromMillisSinceEpoch(1)); + ASSERT_EQUALS(logEntry.getWhat(), "split"); + ASSERT_EQUALS(logEntry.getNS(), "test.test"); + ASSERT_EQUALS(logEntry.getDetails(), + BSON("dummy" + << "info")); +} - TEST(ChangelogType, MissingChangeID) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::server("host.local") << - ChangelogType::clientAddr("192.168.0.189:51128") << - ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << - ChangelogType::what("split") << - ChangelogType::ns("test.test") << - ChangelogType::details(BSON("dummy" << "info"))); - string errMsg; - ASSERT(logEntry.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(logEntry.isValid(NULL)); - } +TEST(ChangelogType, MissingChangeID) { + ChangelogType logEntry; + BSONObj obj = + BSON(ChangelogType::server("host.local") + << ChangelogType::clientAddr("192.168.0.189:51128") + << ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << ChangelogType::what("split") + << ChangelogType::ns("test.test") << ChangelogType::details(BSON("dummy" + << "info"))); + string errMsg; + ASSERT(logEntry.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(logEntry.isValid(NULL)); +} - TEST(ChangelogType, MissingServer) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") << - ChangelogType::clientAddr("192.168.0.189:51128") << - ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << - ChangelogType::what("split") << - ChangelogType::ns("test.test") << - ChangelogType::details(BSON("dummy" << "info"))); - string errMsg; - ASSERT(logEntry.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(logEntry.isValid(NULL)); - } +TEST(ChangelogType, MissingServer) { + ChangelogType logEntry; + BSONObj obj = + BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") + << ChangelogType::clientAddr("192.168.0.189:51128") + << ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << ChangelogType::what("split") + << ChangelogType::ns("test.test") << ChangelogType::details(BSON("dummy" + << "info"))); + string errMsg; + ASSERT(logEntry.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(logEntry.isValid(NULL)); +} - TEST(ChangelogType, MissingClientAddr) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") << - ChangelogType::server("host.local") << - ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << - ChangelogType::what("split") << - ChangelogType::ns("test.test") << - ChangelogType::details(BSON("dummy" << "info"))); - string errMsg; - ASSERT(logEntry.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(logEntry.isValid(NULL)); - } +TEST(ChangelogType, MissingClientAddr) { + ChangelogType logEntry; + BSONObj obj = + BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") + << ChangelogType::server("host.local") + << ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << ChangelogType::what("split") + << ChangelogType::ns("test.test") << ChangelogType::details(BSON("dummy" + << "info"))); + string errMsg; + ASSERT(logEntry.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(logEntry.isValid(NULL)); +} - TEST(ChangelogType, MissingTime) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") << - ChangelogType::server("host.local") << - ChangelogType::clientAddr("192.168.0.189:51128") << - ChangelogType::what("split") << - ChangelogType::ns("test.test") << - ChangelogType::details(BSON("dummy" << "info"))); - string errMsg; - ASSERT(logEntry.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(logEntry.isValid(NULL)); - } +TEST(ChangelogType, MissingTime) { + ChangelogType logEntry; + BSONObj obj = + BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") + << ChangelogType::server("host.local") + << ChangelogType::clientAddr("192.168.0.189:51128") << ChangelogType::what("split") + << ChangelogType::ns("test.test") << ChangelogType::details(BSON("dummy" + << "info"))); + string errMsg; + ASSERT(logEntry.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(logEntry.isValid(NULL)); +} - TEST(ChangelogType, MissingWhat) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") << - ChangelogType::server("host.local") << - ChangelogType::clientAddr("192.168.0.189:51128") << - ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << - ChangelogType::ns("test.test") << - ChangelogType::details(BSON("dummy" << "info"))); - string errMsg; - ASSERT(logEntry.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(logEntry.isValid(NULL)); - } +TEST(ChangelogType, MissingWhat) { + ChangelogType logEntry; + BSONObj obj = BSON( + ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") + << ChangelogType::server("host.local") << ChangelogType::clientAddr("192.168.0.189:51128") + << ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << ChangelogType::ns("test.test") + << ChangelogType::details(BSON("dummy" + << "info"))); + string errMsg; + ASSERT(logEntry.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(logEntry.isValid(NULL)); +} - TEST(ChangelogType, MissingNS) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") << - ChangelogType::server("host.local") << - ChangelogType::clientAddr("192.168.0.189:51128") << - ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << - ChangelogType::what("split") << - ChangelogType::details(BSON("dummy" << "info"))); - string errMsg; - ASSERT(logEntry.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(logEntry.isValid(NULL)); - } +TEST(ChangelogType, MissingNS) { + ChangelogType logEntry; + BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") + << ChangelogType::server("host.local") + << ChangelogType::clientAddr("192.168.0.189:51128") + << ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) + << ChangelogType::what("split") << ChangelogType::details(BSON("dummy" + << "info"))); + string errMsg; + ASSERT(logEntry.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(logEntry.isValid(NULL)); +} - TEST(ChangelogType, MissingDetails) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") << - ChangelogType::server("host.local") << - ChangelogType::clientAddr("192.168.0.189:51128") << - ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) << - ChangelogType::what("split") << - ChangelogType::ns("test.test")); - string errMsg; - ASSERT(logEntry.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(logEntry.isValid(NULL)); - } +TEST(ChangelogType, MissingDetails) { + ChangelogType logEntry; + BSONObj obj = BSON(ChangelogType::changeID("host.local-2012-11-21T19:14:10-8") + << ChangelogType::server("host.local") + << ChangelogType::clientAddr("192.168.0.189:51128") + << ChangelogType::time(Date_t::fromMillisSinceEpoch(1)) + << ChangelogType::what("split") << ChangelogType::ns("test.test")); + string errMsg; + ASSERT(logEntry.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(logEntry.isValid(NULL)); +} - TEST(ChangelogType, BadType) { - ChangelogType logEntry; - BSONObj obj = BSON(ChangelogType::changeID() << 0); - string errMsg; - ASSERT((!logEntry.parseBSON(obj, &errMsg)) && (errMsg != "")); - } +TEST(ChangelogType, BadType) { + ChangelogType logEntry; + BSONObj obj = BSON(ChangelogType::changeID() << 0); + string errMsg; + ASSERT((!logEntry.parseBSON(obj, &errMsg)) && (errMsg != "")); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/catalog/type_chunk.cpp b/src/mongo/s/catalog/type_chunk.cpp index 41498d5373a..68b9c016cae 100644 --- a/src/mongo/s/catalog/type_chunk.cpp +++ b/src/mongo/s/catalog/type_chunk.cpp @@ -41,198 +41,202 @@ namespace mongo { - const std::string ChunkType::ConfigNS = "config.chunks"; - - const BSONField<std::string> ChunkType::name("_id"); - const BSONField<std::string> ChunkType::ns("ns"); - const BSONField<BSONObj> ChunkType::min("min"); - const BSONField<BSONObj> ChunkType::max("max"); - const BSONField<BSONArray> ChunkType::version("version"); - const BSONField<std::string> ChunkType::shard("shard"); - const BSONField<bool> ChunkType::jumbo("jumbo"); - const BSONField<Date_t> ChunkType::DEPRECATED_lastmod("lastmod"); - const BSONField<OID> ChunkType::DEPRECATED_epoch("lastmodEpoch"); - - - StatusWith<ChunkType> ChunkType::fromBSON(const BSONObj& source) { - ChunkType chunk; - - { - std::string chunkName; - Status status = bsonExtractStringField(source, name.name(), &chunkName); - if (!status.isOK()) return status; - chunk._name = chunkName; - } - - { - std::string chunkNS; - Status status = bsonExtractStringField(source, ns.name(), &chunkNS); - if (!status.isOK()) return status; - chunk._ns = chunkNS; - } - - { - BSONElement chunkMinElement; - Status status = bsonExtractTypedField(source, min.name(), Object, &chunkMinElement); - if (!status.isOK()) return status; - chunk._min = chunkMinElement.Obj().getOwned(); - } - - { - BSONElement chunkMaxElement; - Status status = bsonExtractTypedField(source, max.name(), Object, &chunkMaxElement); - if (!status.isOK()) return status; - chunk._max = chunkMaxElement.Obj().getOwned(); - } - - { - std::string chunkShard; - Status status = bsonExtractStringField(source, shard.name(), &chunkShard); - if (!status.isOK()) return status; - chunk._shard = chunkShard; - } - - { - bool chunkJumbo; - Status status = bsonExtractBooleanField(source, jumbo.name(), &chunkJumbo); - if (status.isOK()) { - chunk._jumbo = chunkJumbo; - } - else if (status == ErrorCodes::NoSuchKey) { - // Jumbo status is missing, so it will be presumed false - } - else { - return status; - } - } - - // - // ChunkVersion backward compatibility logic contained in ChunkVersion - // - - // ChunkVersion is currently encoded as { 'version': [<TS>,<OID>] } - - if (ChunkVersion::canParseBSON(source, version())) { - chunk._version = ChunkVersion::fromBSON(source, version()); - } - else if (ChunkVersion::canParseBSON(source, DEPRECATED_lastmod())) { - chunk._version = ChunkVersion::fromBSON(source, DEPRECATED_lastmod()); - } - - return chunk; +const std::string ChunkType::ConfigNS = "config.chunks"; + +const BSONField<std::string> ChunkType::name("_id"); +const BSONField<std::string> ChunkType::ns("ns"); +const BSONField<BSONObj> ChunkType::min("min"); +const BSONField<BSONObj> ChunkType::max("max"); +const BSONField<BSONArray> ChunkType::version("version"); +const BSONField<std::string> ChunkType::shard("shard"); +const BSONField<bool> ChunkType::jumbo("jumbo"); +const BSONField<Date_t> ChunkType::DEPRECATED_lastmod("lastmod"); +const BSONField<OID> ChunkType::DEPRECATED_epoch("lastmodEpoch"); + + +StatusWith<ChunkType> ChunkType::fromBSON(const BSONObj& source) { + ChunkType chunk; + + { + std::string chunkName; + Status status = bsonExtractStringField(source, name.name(), &chunkName); + if (!status.isOK()) + return status; + chunk._name = chunkName; } - Status ChunkType::validate() const { - - if (!_name.is_initialized() || _name->empty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << name.name() << " field"); - } - - if (!_ns.is_initialized() || _ns->empty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << ns.name() << " field"); - } + { + std::string chunkNS; + Status status = bsonExtractStringField(source, ns.name(), &chunkNS); + if (!status.isOK()) + return status; + chunk._ns = chunkNS; + } - if (!_min.is_initialized() || _min->isEmpty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << min.name() << " field"); - } + { + BSONElement chunkMinElement; + Status status = bsonExtractTypedField(source, min.name(), Object, &chunkMinElement); + if (!status.isOK()) + return status; + chunk._min = chunkMinElement.Obj().getOwned(); + } - if (!_max.is_initialized() || _max->isEmpty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << max.name() << " field"); - } + { + BSONElement chunkMaxElement; + Status status = bsonExtractTypedField(source, max.name(), Object, &chunkMaxElement); + if (!status.isOK()) + return status; + chunk._max = chunkMaxElement.Obj().getOwned(); + } - if (!_version.is_initialized() || !_version->isSet()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << version.name() << " field"); - } + { + std::string chunkShard; + Status status = bsonExtractStringField(source, shard.name(), &chunkShard); + if (!status.isOK()) + return status; + chunk._shard = chunkShard; + } - if (!_shard.is_initialized() || _shard->empty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << shard.name() << " field"); - } - - // 'min' and 'max' must share the same fields. - if (_min->nFields() != _max->nFields()) { - return Status(ErrorCodes::BadValue, - str::stream() << "min and max have a different number of keys"); + { + bool chunkJumbo; + Status status = bsonExtractBooleanField(source, jumbo.name(), &chunkJumbo); + if (status.isOK()) { + chunk._jumbo = chunkJumbo; + } else if (status == ErrorCodes::NoSuchKey) { + // Jumbo status is missing, so it will be presumed false + } else { + return status; } + } - BSONObjIterator minIt(getMin()); - BSONObjIterator maxIt(getMax()); - while (minIt.more() && maxIt.more()) { - BSONElement minElem = minIt.next(); - BSONElement maxElem = maxIt.next(); - if (strcmp(minElem.fieldName(), maxElem.fieldName())) { - return Status(ErrorCodes::BadValue, - str::stream() << "min and max must have the same set of keys"); - } - } + // + // ChunkVersion backward compatibility logic contained in ChunkVersion + // - // 'max' should be greater than 'min'. - if (_min->woCompare(getMax()) >= 0) { - return Status(ErrorCodes::BadValue, - str::stream() << "max key must be greater than min key"); - } + // ChunkVersion is currently encoded as { 'version': [<TS>,<OID>] } - return Status::OK(); + if (ChunkVersion::canParseBSON(source, version())) { + chunk._version = ChunkVersion::fromBSON(source, version()); + } else if (ChunkVersion::canParseBSON(source, DEPRECATED_lastmod())) { + chunk._version = ChunkVersion::fromBSON(source, DEPRECATED_lastmod()); } - BSONObj ChunkType::toBSON() const { - BSONObjBuilder builder; - if (_name) builder.append(name.name(), getName()); - if (_ns) builder.append(ns.name(), getNS()); - if (_min) builder.append(min.name(), getMin()); - if (_max) builder.append(max.name(), getMax()); - if (_shard) builder.append(shard.name(), getShard()); - if (_version) { - // For now, write both the deprecated *and* the new fields - _version->addToBSON(builder, version()); - _version->addToBSON(builder, DEPRECATED_lastmod()); - } - if (_jumbo) builder.append(jumbo.name(), getJumbo()); + return chunk; +} - return builder.obj(); +Status ChunkType::validate() const { + if (!_name.is_initialized() || _name->empty()) { + return Status(ErrorCodes::NoSuchKey, + str::stream() << "missing " << name.name() << " field"); } - std::string ChunkType::toString() const { - return toBSON().toString(); + if (!_ns.is_initialized() || _ns->empty()) { + return Status(ErrorCodes::NoSuchKey, str::stream() << "missing " << ns.name() << " field"); } - void ChunkType::setName(const std::string& name) { - invariant(!name.empty()); - _name = name; + if (!_min.is_initialized() || _min->isEmpty()) { + return Status(ErrorCodes::NoSuchKey, str::stream() << "missing " << min.name() << " field"); } - void ChunkType::setNS(const std::string& ns) { - invariant(!ns.empty()); - _ns = ns; + if (!_max.is_initialized() || _max->isEmpty()) { + return Status(ErrorCodes::NoSuchKey, str::stream() << "missing " << max.name() << " field"); } - void ChunkType::setMin(const BSONObj& min) { - invariant(!min.isEmpty()); - _min = min; + if (!_version.is_initialized() || !_version->isSet()) { + return Status(ErrorCodes::NoSuchKey, + str::stream() << "missing " << version.name() << " field"); } - void ChunkType::setMax(const BSONObj& max) { - invariant(!max.isEmpty()); - _max = max; + if (!_shard.is_initialized() || _shard->empty()) { + return Status(ErrorCodes::NoSuchKey, + str::stream() << "missing " << shard.name() << " field"); } - void ChunkType::setVersion(const ChunkVersion& version) { - invariant(version.isSet()); - _version = version; + // 'min' and 'max' must share the same fields. + if (_min->nFields() != _max->nFields()) { + return Status(ErrorCodes::BadValue, + str::stream() << "min and max have a different number of keys"); } - void ChunkType::setShard(const std::string& shard) { - invariant(!shard.empty()); - _shard = shard; + BSONObjIterator minIt(getMin()); + BSONObjIterator maxIt(getMax()); + while (minIt.more() && maxIt.more()) { + BSONElement minElem = minIt.next(); + BSONElement maxElem = maxIt.next(); + if (strcmp(minElem.fieldName(), maxElem.fieldName())) { + return Status(ErrorCodes::BadValue, + str::stream() << "min and max must have the same set of keys"); + } } - void ChunkType::setJumbo(bool jumbo) { - _jumbo = jumbo; + // 'max' should be greater than 'min'. + if (_min->woCompare(getMax()) >= 0) { + return Status(ErrorCodes::BadValue, + str::stream() << "max key must be greater than min key"); } -} // namespace mongo + return Status::OK(); +} + +BSONObj ChunkType::toBSON() const { + BSONObjBuilder builder; + if (_name) + builder.append(name.name(), getName()); + if (_ns) + builder.append(ns.name(), getNS()); + if (_min) + builder.append(min.name(), getMin()); + if (_max) + builder.append(max.name(), getMax()); + if (_shard) + builder.append(shard.name(), getShard()); + if (_version) { + // For now, write both the deprecated *and* the new fields + _version->addToBSON(builder, version()); + _version->addToBSON(builder, DEPRECATED_lastmod()); + } + if (_jumbo) + builder.append(jumbo.name(), getJumbo()); + + return builder.obj(); +} + +std::string ChunkType::toString() const { + return toBSON().toString(); +} + +void ChunkType::setName(const std::string& name) { + invariant(!name.empty()); + _name = name; +} + +void ChunkType::setNS(const std::string& ns) { + invariant(!ns.empty()); + _ns = ns; +} + +void ChunkType::setMin(const BSONObj& min) { + invariant(!min.isEmpty()); + _min = min; +} + +void ChunkType::setMax(const BSONObj& max) { + invariant(!max.isEmpty()); + _max = max; +} + +void ChunkType::setVersion(const ChunkVersion& version) { + invariant(version.isSet()); + _version = version; +} + +void ChunkType::setShard(const std::string& shard) { + invariant(!shard.empty()); + _shard = shard; +} + +void ChunkType::setJumbo(bool jumbo) { + _jumbo = jumbo; +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_chunk.h b/src/mongo/s/catalog/type_chunk.h index 5945f722f6f..52a452a20f2 100644 --- a/src/mongo/s/catalog/type_chunk.h +++ b/src/mongo/s/catalog/type_chunk.h @@ -36,95 +36,110 @@ namespace mongo { - class BSONObj; - class Status; - template<typename T> class StatusWith; +class BSONObj; +class Status; +template <typename T> +class StatusWith; + +/** + * This class represents the layout and contents of documents contained in the + * config.chunks collection. All manipulation of documents coming from that + * collection should be done with this class. + */ +class ChunkType { +public: + // Name of the chunks collection in the config server. + static const std::string ConfigNS; + + // Field names and types in the chunks collection type. + static const BSONField<std::string> name; + static const BSONField<std::string> ns; + static const BSONField<BSONObj> min; + static const BSONField<BSONObj> max; + static const BSONField<BSONArray> version; + static const BSONField<std::string> shard; + static const BSONField<bool> jumbo; + static const BSONField<Date_t> DEPRECATED_lastmod; + static const BSONField<OID> DEPRECATED_epoch; + + + /** + * Constructs a new ChunkType object from BSON. + * Also does validation of the contents. + */ + static StatusWith<ChunkType> fromBSON(const BSONObj& source); + + /** + * Returns OK if all fields have been set. Otherwise returns NoSuchKey + * and information about the first field that is missing. + */ + Status validate() const; + + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; /** - * This class represents the layout and contents of documents contained in the - * config.chunks collection. All manipulation of documents coming from that - * collection should be done with this class. + * Returns a std::string representation of the current internal state. */ - class ChunkType { - public: - - // Name of the chunks collection in the config server. - static const std::string ConfigNS; - - // Field names and types in the chunks collection type. - static const BSONField<std::string> name; - static const BSONField<std::string> ns; - static const BSONField<BSONObj> min; - static const BSONField<BSONObj> max; - static const BSONField<BSONArray> version; - static const BSONField<std::string> shard; - static const BSONField<bool> jumbo; - static const BSONField<Date_t> DEPRECATED_lastmod; - static const BSONField<OID> DEPRECATED_epoch; - - - /** - * Constructs a new ChunkType object from BSON. - * Also does validation of the contents. - */ - static StatusWith<ChunkType> fromBSON(const BSONObj& source); - - /** - * Returns OK if all fields have been set. Otherwise returns NoSuchKey - * and information about the first field that is missing. - */ - Status validate() const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - const std::string& getName() const { return _name.get(); } - void setName(const std::string& name); - - const std::string& getNS() const { return _ns.get(); } - void setNS(const std::string& name); - - const BSONObj& getMin() const { return _min.get(); } - void setMin(const BSONObj& min); - - const BSONObj& getMax() const { return _max.get(); } - void setMax(const BSONObj& max); - - bool isVersionSet() const { return _version.is_initialized(); } - const ChunkVersion& getVersion() const { return _version.get(); } - void setVersion(const ChunkVersion& version); - - const std::string& getShard() const { return _shard.get(); } - void setShard(const std::string& shard); - - bool getJumbo() const { return _jumbo.get_value_or(false); } - void setJumbo(bool jumbo); - - private: - - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - - // (M) chunk's id - boost::optional<std::string> _name; - // (M) collection this chunk is in - boost::optional<std::string> _ns; - // (M) first key of the range, inclusive - boost::optional<BSONObj> _min; - // (M) last key of the range, non-inclusive - boost::optional<BSONObj> _max; - // (M) version of this chunk - boost::optional<ChunkVersion> _version; - // (M) shard this chunk lives in - boost::optional<std::string> _shard; - // (O) too big to move? - boost::optional<bool> _jumbo; - }; - -} // namespace mongo + std::string toString() const; + + const std::string& getName() const { + return _name.get(); + } + void setName(const std::string& name); + + const std::string& getNS() const { + return _ns.get(); + } + void setNS(const std::string& name); + + const BSONObj& getMin() const { + return _min.get(); + } + void setMin(const BSONObj& min); + + const BSONObj& getMax() const { + return _max.get(); + } + void setMax(const BSONObj& max); + + bool isVersionSet() const { + return _version.is_initialized(); + } + const ChunkVersion& getVersion() const { + return _version.get(); + } + void setVersion(const ChunkVersion& version); + + const std::string& getShard() const { + return _shard.get(); + } + void setShard(const std::string& shard); + + bool getJumbo() const { + return _jumbo.get_value_or(false); + } + void setJumbo(bool jumbo); + +private: + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + + // (M) chunk's id + boost::optional<std::string> _name; + // (M) collection this chunk is in + boost::optional<std::string> _ns; + // (M) first key of the range, inclusive + boost::optional<BSONObj> _min; + // (M) last key of the range, non-inclusive + boost::optional<BSONObj> _max; + // (M) version of this chunk + boost::optional<ChunkVersion> _version; + // (M) shard this chunk lives in + boost::optional<std::string> _shard; + // (O) too big to move? + boost::optional<bool> _jumbo; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_chunk_test.cpp b/src/mongo/s/catalog/type_chunk_test.cpp index 35d9f578d5f..bc22f23ddb1 100644 --- a/src/mongo/s/catalog/type_chunk_test.cpp +++ b/src/mongo/s/catalog/type_chunk_test.cpp @@ -37,139 +37,126 @@ namespace { - using namespace mongo; - - using std::string; - - TEST(ChunkType, MissingRequiredFields) { - ChunkType chunk; - BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << OID::gen()); - - BSONObj objModNS = BSON(ChunkType::name("test.mycol-a_MinKey") << - ChunkType::min(BSON("a" << 10 << "b" << 10)) << - ChunkType::max(BSON("a" << 20)) << - ChunkType::version(version) << - ChunkType::shard("shard0001")); - StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(objModNS); - ASSERT_FALSE(chunkRes.isOK()); - - BSONObj objModName = BSON(ChunkType::ns("test.mycol") << - ChunkType::min(BSON("a" << 10 << "b" << 10)) << - ChunkType::max(BSON("a" << 20)) << - ChunkType::version(version) << - ChunkType::shard("shard0001")); - chunkRes = ChunkType::fromBSON(objModName); - ASSERT_FALSE(chunkRes.isOK()); - - BSONObj objModKeys = BSON(ChunkType::name("test.mycol-a_MinKey") << - ChunkType::ns("test.mycol") << - ChunkType::version(version) << - ChunkType::shard("shard0001")); - chunkRes = ChunkType::fromBSON(objModKeys); - ASSERT_FALSE(chunkRes.isOK()); - - BSONObj objModShard = BSON(ChunkType::name("test.mycol-a_MinKey") << - ChunkType::ns("test.mycol") << - ChunkType::min(BSON("a" << 10 << "b" << 10)) << - ChunkType::max(BSON("a" << 20)) << - ChunkType::version(version)); - chunkRes = ChunkType::fromBSON(objModShard); - ASSERT_FALSE(chunkRes.isOK()); - } - - TEST(ChunkType, DifferentNumberOfColumns) { - BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << OID::gen()); - BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") << - ChunkType::ns("test.mycol") << - ChunkType::min(BSON("a" << 10 << "b" << 10)) << - ChunkType::max(BSON("a" << 20)) << - ChunkType::version(version) << - ChunkType::shard("shard0001")); - StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); - ASSERT(chunkRes.isOK()); - ASSERT_FALSE(chunkRes.getValue().validate().isOK()); - } - - TEST(ChunkType, DifferentColumns) { - BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << OID::gen()); - BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") << - ChunkType::ns("test.mycol") << - ChunkType::min(BSON("a" << 10)) << - ChunkType::max(BSON("b" << 20)) << - ChunkType::version(version) << - ChunkType::shard("shard0001")); - StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); - ASSERT(chunkRes.isOK()); - ASSERT_FALSE(chunkRes.getValue().validate().isOK()); - } - - TEST(ChunkType, NotAscending) { - BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << OID::gen()); - BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") << - ChunkType::ns("test.mycol") << - ChunkType::min(BSON("a" << 20)) << - ChunkType::max(BSON("a" << 10)) << - ChunkType::version(version) << - ChunkType::shard("shard0001")); - StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); - ASSERT(chunkRes.isOK()); - ASSERT_FALSE(chunkRes.getValue().validate().isOK()); - } - - TEST(ChunkType, NewFormatVersion) { - ChunkType chunk; - OID epoch = OID::gen(); - BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << epoch); - BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") << - ChunkType::ns("test.mycol") << - ChunkType::min(BSON("a" << 10)) << - ChunkType::max(BSON("a" << 20)) << - ChunkType::version(version) << - ChunkType::shard("shard0001")); - StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); - ASSERT(chunkRes.isOK()); - chunk = chunkRes.getValue(); - - ASSERT_EQUALS(chunk.getName(), "test.mycol-a_MinKey"); - ASSERT_EQUALS(chunk.getNS(), "test.mycol"); - ASSERT_EQUALS(chunk.getMin(), BSON("a" << 10)); - ASSERT_EQUALS(chunk.getMax(), BSON("a" << 20)); - ChunkVersion fetchedVersion = chunk.getVersion(); - ASSERT_EQUALS(fetchedVersion._combined, 1ULL); - ASSERT_EQUALS(fetchedVersion._epoch, epoch); - ASSERT_EQUALS(chunk.getShard(), "shard0001"); - ASSERT_TRUE(chunk.validate().isOK()); - } - - TEST(ChunkType, OldFormatVersion) { - ChunkType chunk; - OID epoch = OID::gen(); - BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") << - ChunkType::ns("test.mycol") << - ChunkType::min(BSON("a" << 10)) << - ChunkType::max(BSON("a" << 20)) << - ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch(1)) << - ChunkType::DEPRECATED_epoch(epoch) << - ChunkType::shard("shard0001")); - StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); - ASSERT(chunkRes.isOK()); - chunk = chunkRes.getValue(); - - ASSERT_EQUALS(chunk.getName(), "test.mycol-a_MinKey"); - ASSERT_EQUALS(chunk.getNS(), "test.mycol"); - ASSERT_EQUALS(chunk.getMin(), BSON("a" << 10)); - ASSERT_EQUALS(chunk.getMax(), BSON("a" << 20)); - ChunkVersion fetchedVersion = chunk.getVersion(); - ASSERT_EQUALS(fetchedVersion._combined, 1ULL); - ASSERT_EQUALS(fetchedVersion._epoch, epoch); - ASSERT_EQUALS(chunk.getShard(), "shard0001"); - ASSERT_TRUE(chunk.validate().isOK()); - } - - TEST(ChunkType, BadType) { - BSONObj obj = BSON(ChunkType::name() << 0); - StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); - ASSERT_FALSE(chunkRes.isOK()); - } - -} // unnamed namespace +using namespace mongo; + +using std::string; + +TEST(ChunkType, MissingRequiredFields) { + ChunkType chunk; + BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << OID::gen()); + + BSONObj objModNS = + BSON(ChunkType::name("test.mycol-a_MinKey") + << ChunkType::min(BSON("a" << 10 << "b" << 10)) << ChunkType::max(BSON("a" << 20)) + << ChunkType::version(version) << ChunkType::shard("shard0001")); + StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(objModNS); + ASSERT_FALSE(chunkRes.isOK()); + + BSONObj objModName = + BSON(ChunkType::ns("test.mycol") + << ChunkType::min(BSON("a" << 10 << "b" << 10)) << ChunkType::max(BSON("a" << 20)) + << ChunkType::version(version) << ChunkType::shard("shard0001")); + chunkRes = ChunkType::fromBSON(objModName); + ASSERT_FALSE(chunkRes.isOK()); + + BSONObj objModKeys = BSON(ChunkType::name("test.mycol-a_MinKey") + << ChunkType::ns("test.mycol") << ChunkType::version(version) + << ChunkType::shard("shard0001")); + chunkRes = ChunkType::fromBSON(objModKeys); + ASSERT_FALSE(chunkRes.isOK()); + + BSONObj objModShard = + BSON(ChunkType::name("test.mycol-a_MinKey") + << ChunkType::ns("test.mycol") << ChunkType::min(BSON("a" << 10 << "b" << 10)) + << ChunkType::max(BSON("a" << 20)) << ChunkType::version(version)); + chunkRes = ChunkType::fromBSON(objModShard); + ASSERT_FALSE(chunkRes.isOK()); +} + +TEST(ChunkType, DifferentNumberOfColumns) { + BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << OID::gen()); + BSONObj obj = + BSON(ChunkType::name("test.mycol-a_MinKey") + << ChunkType::ns("test.mycol") << ChunkType::min(BSON("a" << 10 << "b" << 10)) + << ChunkType::max(BSON("a" << 20)) << ChunkType::version(version) + << ChunkType::shard("shard0001")); + StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); + ASSERT(chunkRes.isOK()); + ASSERT_FALSE(chunkRes.getValue().validate().isOK()); +} + +TEST(ChunkType, DifferentColumns) { + BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << OID::gen()); + BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") + << ChunkType::ns("test.mycol") << ChunkType::min(BSON("a" << 10)) + << ChunkType::max(BSON("b" << 20)) << ChunkType::version(version) + << ChunkType::shard("shard0001")); + StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); + ASSERT(chunkRes.isOK()); + ASSERT_FALSE(chunkRes.getValue().validate().isOK()); +} + +TEST(ChunkType, NotAscending) { + BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << OID::gen()); + BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") + << ChunkType::ns("test.mycol") << ChunkType::min(BSON("a" << 20)) + << ChunkType::max(BSON("a" << 10)) << ChunkType::version(version) + << ChunkType::shard("shard0001")); + StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); + ASSERT(chunkRes.isOK()); + ASSERT_FALSE(chunkRes.getValue().validate().isOK()); +} + +TEST(ChunkType, NewFormatVersion) { + ChunkType chunk; + OID epoch = OID::gen(); + BSONArray version = BSON_ARRAY(Date_t::fromMillisSinceEpoch(1) << epoch); + BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") + << ChunkType::ns("test.mycol") << ChunkType::min(BSON("a" << 10)) + << ChunkType::max(BSON("a" << 20)) << ChunkType::version(version) + << ChunkType::shard("shard0001")); + StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); + ASSERT(chunkRes.isOK()); + chunk = chunkRes.getValue(); + + ASSERT_EQUALS(chunk.getName(), "test.mycol-a_MinKey"); + ASSERT_EQUALS(chunk.getNS(), "test.mycol"); + ASSERT_EQUALS(chunk.getMin(), BSON("a" << 10)); + ASSERT_EQUALS(chunk.getMax(), BSON("a" << 20)); + ChunkVersion fetchedVersion = chunk.getVersion(); + ASSERT_EQUALS(fetchedVersion._combined, 1ULL); + ASSERT_EQUALS(fetchedVersion._epoch, epoch); + ASSERT_EQUALS(chunk.getShard(), "shard0001"); + ASSERT_TRUE(chunk.validate().isOK()); +} + +TEST(ChunkType, OldFormatVersion) { + ChunkType chunk; + OID epoch = OID::gen(); + BSONObj obj = BSON(ChunkType::name("test.mycol-a_MinKey") + << ChunkType::ns("test.mycol") << ChunkType::min(BSON("a" << 10)) + << ChunkType::max(BSON("a" << 20)) + << ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch(1)) + << ChunkType::DEPRECATED_epoch(epoch) << ChunkType::shard("shard0001")); + StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); + ASSERT(chunkRes.isOK()); + chunk = chunkRes.getValue(); + + ASSERT_EQUALS(chunk.getName(), "test.mycol-a_MinKey"); + ASSERT_EQUALS(chunk.getNS(), "test.mycol"); + ASSERT_EQUALS(chunk.getMin(), BSON("a" << 10)); + ASSERT_EQUALS(chunk.getMax(), BSON("a" << 20)); + ChunkVersion fetchedVersion = chunk.getVersion(); + ASSERT_EQUALS(fetchedVersion._combined, 1ULL); + ASSERT_EQUALS(fetchedVersion._epoch, epoch); + ASSERT_EQUALS(chunk.getShard(), "shard0001"); + ASSERT_TRUE(chunk.validate().isOK()); +} + +TEST(ChunkType, BadType) { + BSONObj obj = BSON(ChunkType::name() << 0); + StatusWith<ChunkType> chunkRes = ChunkType::fromBSON(obj); + ASSERT_FALSE(chunkRes.isOK()); +} + +} // unnamed namespace diff --git a/src/mongo/s/catalog/type_collection.cpp b/src/mongo/s/catalog/type_collection.cpp index bd24ca9c44c..7a125e52062 100644 --- a/src/mongo/s/catalog/type_collection.cpp +++ b/src/mongo/s/catalog/type_collection.cpp @@ -38,198 +38,191 @@ namespace mongo { - const std::string CollectionType::ConfigNS = "config.collections"; +const std::string CollectionType::ConfigNS = "config.collections"; - const BSONField<std::string> CollectionType::fullNs("_id"); - const BSONField<OID> CollectionType::epoch("lastmodEpoch"); - const BSONField<Date_t> CollectionType::updatedAt("lastmod"); - const BSONField<BSONObj> CollectionType::keyPattern("key"); - const BSONField<bool> CollectionType::unique("unique"); - const BSONField<bool> CollectionType::noBalance("noBalance"); - const BSONField<bool> CollectionType::dropped("dropped"); +const BSONField<std::string> CollectionType::fullNs("_id"); +const BSONField<OID> CollectionType::epoch("lastmodEpoch"); +const BSONField<Date_t> CollectionType::updatedAt("lastmod"); +const BSONField<BSONObj> CollectionType::keyPattern("key"); +const BSONField<bool> CollectionType::unique("unique"); +const BSONField<bool> CollectionType::noBalance("noBalance"); +const BSONField<bool> CollectionType::dropped("dropped"); - StatusWith<CollectionType> CollectionType::fromBSON(const BSONObj& source) { - CollectionType coll; +StatusWith<CollectionType> CollectionType::fromBSON(const BSONObj& source) { + CollectionType coll; - { - std::string collFullNs; - Status status = bsonExtractStringField(source, fullNs.name(), &collFullNs); - if (!status.isOK()) return status; + { + std::string collFullNs; + Status status = bsonExtractStringField(source, fullNs.name(), &collFullNs); + if (!status.isOK()) + return status; - coll._fullNs = NamespaceString{collFullNs}; - } - - { - OID collEpoch; - Status status = bsonExtractOIDField(source, epoch.name(), &collEpoch); - if (!status.isOK()) return status; + coll._fullNs = NamespaceString{collFullNs}; + } - coll._epoch = collEpoch; - } + { + OID collEpoch; + Status status = bsonExtractOIDField(source, epoch.name(), &collEpoch); + if (!status.isOK()) + return status; - { - BSONElement collUpdatedAt; - Status status = bsonExtractTypedField(source, updatedAt.name(), Date, &collUpdatedAt); - if (!status.isOK()) return status; + coll._epoch = collEpoch; + } - coll._updatedAt = collUpdatedAt.Date(); - } + { + BSONElement collUpdatedAt; + Status status = bsonExtractTypedField(source, updatedAt.name(), Date, &collUpdatedAt); + if (!status.isOK()) + return status; - { - bool collDropped; - Status status = bsonExtractBooleanField(source, dropped.name(), &collDropped); - if (status.isOK()) { - coll._dropped = collDropped; - } - else if (status == ErrorCodes::NoSuchKey) { - // Dropped can be missing in which case it is presumed false - } - else { - return status; - } - } + coll._updatedAt = collUpdatedAt.Date(); + } - { - BSONElement collKeyPattern; - Status status = - bsonExtractTypedField(source, keyPattern.name(), Object, &collKeyPattern); - if (status.isOK()) { - BSONObj obj = collKeyPattern.Obj(); - if (obj.isEmpty()) { - return Status(ErrorCodes::ShardKeyNotFound, "empty shard key"); - } - - coll._keyPattern = KeyPattern(obj.getOwned()); - } - else if ((status == ErrorCodes::NoSuchKey) && coll.getDropped()) { - // Sharding key can be missing if the collection is dropped - } - else { - return status; - } + { + bool collDropped; + Status status = bsonExtractBooleanField(source, dropped.name(), &collDropped); + if (status.isOK()) { + coll._dropped = collDropped; + } else if (status == ErrorCodes::NoSuchKey) { + // Dropped can be missing in which case it is presumed false + } else { + return status; } + } - { - bool collUnique; - Status status = bsonExtractBooleanField(source, unique.name(), &collUnique); - if (status.isOK()) { - coll._unique = collUnique; + { + BSONElement collKeyPattern; + Status status = bsonExtractTypedField(source, keyPattern.name(), Object, &collKeyPattern); + if (status.isOK()) { + BSONObj obj = collKeyPattern.Obj(); + if (obj.isEmpty()) { + return Status(ErrorCodes::ShardKeyNotFound, "empty shard key"); } - else if (status == ErrorCodes::NoSuchKey) { - // Key uniqueness can be missing in which case it is presumed false - } - else { - return status; - } - } - { - bool collNoBalance; - Status status = bsonExtractBooleanField(source, noBalance.name(), &collNoBalance); - if (status.isOK()) { - coll._allowBalance = !collNoBalance; - } - else if (status == ErrorCodes::NoSuchKey) { - // No balance can be missing in which case it is presumed as false - } - else { - return status; - } + coll._keyPattern = KeyPattern(obj.getOwned()); + } else if ((status == ErrorCodes::NoSuchKey) && coll.getDropped()) { + // Sharding key can be missing if the collection is dropped + } else { + return status; } - - return StatusWith<CollectionType>(coll); } - Status CollectionType::validate() const { - // These fields must always be set - if (!_fullNs.is_initialized()) { - return Status(ErrorCodes::NoSuchKey, "missing ns"); - } - - if (!_fullNs->isValid()) { - return Status(ErrorCodes::BadValue, "invalid namespace " + _fullNs->toString()); - } - - if (!_epoch.is_initialized()) { - return Status(ErrorCodes::NoSuchKey, "missing epoch"); + { + bool collUnique; + Status status = bsonExtractBooleanField(source, unique.name(), &collUnique); + if (status.isOK()) { + coll._unique = collUnique; + } else if (status == ErrorCodes::NoSuchKey) { + // Key uniqueness can be missing in which case it is presumed false + } else { + return status; } + } - if (!_updatedAt.is_initialized()) { - return Status(ErrorCodes::NoSuchKey, "missing updated at timestamp"); + { + bool collNoBalance; + Status status = bsonExtractBooleanField(source, noBalance.name(), &collNoBalance); + if (status.isOK()) { + coll._allowBalance = !collNoBalance; + } else if (status == ErrorCodes::NoSuchKey) { + // No balance can be missing in which case it is presumed as false + } else { + return status; } + } - if (!_dropped.get_value_or(false)) { - if (!_epoch->isSet()) { - return Status(ErrorCodes::BadValue, "invalid epoch"); - } + return StatusWith<CollectionType>(coll); +} - if (Date_t() == _updatedAt.get()) { - return Status(ErrorCodes::BadValue, "invalid updated at timestamp"); - } +Status CollectionType::validate() const { + // These fields must always be set + if (!_fullNs.is_initialized()) { + return Status(ErrorCodes::NoSuchKey, "missing ns"); + } - if (!_keyPattern.is_initialized()) { - return Status(ErrorCodes::NoSuchKey, "missing key pattern"); - } - else { - invariant(!_keyPattern->toBSON().isEmpty()); - } - } + if (!_fullNs->isValid()) { + return Status(ErrorCodes::BadValue, "invalid namespace " + _fullNs->toString()); + } - return Status::OK(); + if (!_epoch.is_initialized()) { + return Status(ErrorCodes::NoSuchKey, "missing epoch"); } - BSONObj CollectionType::toBSON() const { - BSONObjBuilder builder; + if (!_updatedAt.is_initialized()) { + return Status(ErrorCodes::NoSuchKey, "missing updated at timestamp"); + } - if (_fullNs) { - builder.append(fullNs.name(), _fullNs->toString()); + if (!_dropped.get_value_or(false)) { + if (!_epoch->isSet()) { + return Status(ErrorCodes::BadValue, "invalid epoch"); } - builder.append(epoch.name(), _epoch.get_value_or(OID())); - builder.append(updatedAt.name(), _updatedAt.get_value_or(Date_t())); - // These fields are optional, so do not include them in the metadata for the purposes of - // consuming less space on the config servers. - - if (_dropped.is_initialized()) { - builder.append(dropped.name(), _dropped.get()); + if (Date_t() == _updatedAt.get()) { + return Status(ErrorCodes::BadValue, "invalid updated at timestamp"); } - if (_keyPattern.is_initialized()) { - builder.append(keyPattern.name(), _keyPattern->toBSON()); + if (!_keyPattern.is_initialized()) { + return Status(ErrorCodes::NoSuchKey, "missing key pattern"); + } else { + invariant(!_keyPattern->toBSON().isEmpty()); } + } - if (_unique.is_initialized()) { - builder.append(unique.name(), _unique.get()); - } + return Status::OK(); +} - if (_allowBalance.is_initialized()) { - builder.append(noBalance.name(), !_allowBalance.get()); - } +BSONObj CollectionType::toBSON() const { + BSONObjBuilder builder; - return builder.obj(); + if (_fullNs) { + builder.append(fullNs.name(), _fullNs->toString()); } + builder.append(epoch.name(), _epoch.get_value_or(OID())); + builder.append(updatedAt.name(), _updatedAt.get_value_or(Date_t())); - std::string CollectionType::toString() const { - return toBSON().toString(); - } + // These fields are optional, so do not include them in the metadata for the purposes of + // consuming less space on the config servers. - void CollectionType::setNs(const NamespaceString& fullNs) { - invariant(fullNs.isValid()); - _fullNs = fullNs; + if (_dropped.is_initialized()) { + builder.append(dropped.name(), _dropped.get()); } - void CollectionType::setEpoch(OID epoch) { - _epoch = epoch; + if (_keyPattern.is_initialized()) { + builder.append(keyPattern.name(), _keyPattern->toBSON()); } - void CollectionType::setUpdatedAt(Date_t updatedAt) { - _updatedAt = updatedAt; + if (_unique.is_initialized()) { + builder.append(unique.name(), _unique.get()); } - void CollectionType::setKeyPattern(const KeyPattern& keyPattern) { - invariant(!keyPattern.toBSON().isEmpty()); - _keyPattern = keyPattern; + if (_allowBalance.is_initialized()) { + builder.append(noBalance.name(), !_allowBalance.get()); } -} // namespace mongo + return builder.obj(); +} + +std::string CollectionType::toString() const { + return toBSON().toString(); +} + +void CollectionType::setNs(const NamespaceString& fullNs) { + invariant(fullNs.isValid()); + _fullNs = fullNs; +} + +void CollectionType::setEpoch(OID epoch) { + _epoch = epoch; +} + +void CollectionType::setUpdatedAt(Date_t updatedAt) { + _updatedAt = updatedAt; +} + +void CollectionType::setKeyPattern(const KeyPattern& keyPattern) { + invariant(!keyPattern.toBSON().isEmpty()); + _keyPattern = keyPattern; +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_collection.h b/src/mongo/s/catalog/type_collection.h index 12360d7fef3..545c949e81e 100644 --- a/src/mongo/s/catalog/type_collection.h +++ b/src/mongo/s/catalog/type_collection.h @@ -37,91 +37,110 @@ namespace mongo { - class Status; - template<typename T> class StatusWith; +class Status; +template <typename T> +class StatusWith; - /** - * This class represents the layout and contents of documents contained in the - * config.collections collection. All manipulation of documents coming from that collection - * should be done with this class. - */ - class CollectionType { - public: - // Name of the collections collection in the config server. - static const std::string ConfigNS; - - static const BSONField<std::string> fullNs; - static const BSONField<OID> epoch; - static const BSONField<Date_t> updatedAt; - static const BSONField<BSONObj> keyPattern; - static const BSONField<bool> unique; - static const BSONField<bool> noBalance; - static const BSONField<bool> dropped; - - - /** - * Constructs a new DatabaseType object from BSON. Also does validation of the contents. - */ - static StatusWith<CollectionType> fromBSON(const BSONObj& source); - - /** - * Returns OK if all fields have been set. Otherwise returns NoSuchKey and information - * about what is the first field which is missing. - */ - Status validate() const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - const NamespaceString& getNs() const { return _fullNs.get(); } - void setNs(const NamespaceString& fullNs); - - OID getEpoch() const { return _epoch.get(); } - void setEpoch(OID epoch); - - Date_t getUpdatedAt() const { return _updatedAt.get(); } - void setUpdatedAt(Date_t updatedAt); - - bool getDropped() const { return _dropped.get_value_or(false); } - void setDropped(bool dropped) { _dropped = dropped; } - - const KeyPattern& getKeyPattern() const { return _keyPattern.get(); } - void setKeyPattern(const KeyPattern& keyPattern); - - bool getUnique() const { return _unique.get_value_or(false); } - void setUnique(bool unique) { _unique = unique; } - - bool getAllowBalance() const { return _allowBalance.get_value_or(true); } - - private: - // Required full namespace (with the database prefix). - boost::optional<NamespaceString> _fullNs; - - // Required to disambiguate collection namespace incarnations. - boost::optional<OID> _epoch; +/** + * This class represents the layout and contents of documents contained in the + * config.collections collection. All manipulation of documents coming from that collection + * should be done with this class. + */ +class CollectionType { +public: + // Name of the collections collection in the config server. + static const std::string ConfigNS; - // Required last updated time. - boost::optional<Date_t> _updatedAt; + static const BSONField<std::string> fullNs; + static const BSONField<OID> epoch; + static const BSONField<Date_t> updatedAt; + static const BSONField<BSONObj> keyPattern; + static const BSONField<bool> unique; + static const BSONField<bool> noBalance; + static const BSONField<bool> dropped; - // Optional, whether the collection has been dropped. If missing, implies false. - boost::optional<bool> _dropped; - // Sharding key. Required, if collection is not dropped. - boost::optional<KeyPattern> _keyPattern; + /** + * Constructs a new DatabaseType object from BSON. Also does validation of the contents. + */ + static StatusWith<CollectionType> fromBSON(const BSONObj& source); - // Optional uniqueness of the sharding key. If missing, implies false. - boost::optional<bool> _unique; + /** + * Returns OK if all fields have been set. Otherwise returns NoSuchKey and information + * about what is the first field which is missing. + */ + Status validate() const; - // Optional whether balancing is allowed for this collection. If missing, implies true. - boost::optional<bool> _allowBalance; - }; + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; -} // namespace mongo + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; + + const NamespaceString& getNs() const { + return _fullNs.get(); + } + void setNs(const NamespaceString& fullNs); + + OID getEpoch() const { + return _epoch.get(); + } + void setEpoch(OID epoch); + + Date_t getUpdatedAt() const { + return _updatedAt.get(); + } + void setUpdatedAt(Date_t updatedAt); + + bool getDropped() const { + return _dropped.get_value_or(false); + } + void setDropped(bool dropped) { + _dropped = dropped; + } + + const KeyPattern& getKeyPattern() const { + return _keyPattern.get(); + } + void setKeyPattern(const KeyPattern& keyPattern); + + bool getUnique() const { + return _unique.get_value_or(false); + } + void setUnique(bool unique) { + _unique = unique; + } + + bool getAllowBalance() const { + return _allowBalance.get_value_or(true); + } + +private: + // Required full namespace (with the database prefix). + boost::optional<NamespaceString> _fullNs; + + // Required to disambiguate collection namespace incarnations. + boost::optional<OID> _epoch; + + // Required last updated time. + boost::optional<Date_t> _updatedAt; + + // Optional, whether the collection has been dropped. If missing, implies false. + boost::optional<bool> _dropped; + + // Sharding key. Required, if collection is not dropped. + boost::optional<KeyPattern> _keyPattern; + + // Optional uniqueness of the sharding key. If missing, implies false. + boost::optional<bool> _unique; + + // Optional whether balancing is allowed for this collection. If missing, implies true. + boost::optional<bool> _allowBalance; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_collection_test.cpp b/src/mongo/s/catalog/type_collection_test.cpp index cde1bff5ce9..9d3d3553ef7 100644 --- a/src/mongo/s/catalog/type_collection_test.cpp +++ b/src/mongo/s/catalog/type_collection_test.cpp @@ -36,58 +36,53 @@ namespace { - using namespace mongo; +using namespace mongo; - TEST(CollectionType, Empty) { - StatusWith<CollectionType> status = CollectionType::fromBSON(BSONObj()); - ASSERT_FALSE(status.isOK()); - } +TEST(CollectionType, Empty) { + StatusWith<CollectionType> status = CollectionType::fromBSON(BSONObj()); + ASSERT_FALSE(status.isOK()); +} - TEST(CollectionType, Basic) { - const OID oid = OID::gen(); - StatusWith<CollectionType> status = CollectionType::fromBSON( - BSON(CollectionType::fullNs("db.coll") << - CollectionType::epoch(oid) << - CollectionType::updatedAt(Date_t::fromMillisSinceEpoch(1)) << - CollectionType::keyPattern(BSON("a" << 1)) << - CollectionType::unique(true))); - ASSERT_TRUE(status.isOK()); +TEST(CollectionType, Basic) { + const OID oid = OID::gen(); + StatusWith<CollectionType> status = CollectionType::fromBSON(BSON( + CollectionType::fullNs("db.coll") + << CollectionType::epoch(oid) << CollectionType::updatedAt(Date_t::fromMillisSinceEpoch(1)) + << CollectionType::keyPattern(BSON("a" << 1)) << CollectionType::unique(true))); + ASSERT_TRUE(status.isOK()); - CollectionType coll = status.getValue(); - ASSERT_TRUE(coll.validate().isOK()); - ASSERT(coll.getNs() == NamespaceString{"db.coll"}); - ASSERT_EQUALS(coll.getEpoch(), oid); - ASSERT_EQUALS(coll.getUpdatedAt(), Date_t::fromMillisSinceEpoch(1)); - ASSERT_EQUALS(coll.getKeyPattern().toBSON(), BSON("a" << 1)); - ASSERT_EQUALS(coll.getUnique(), true); - ASSERT_EQUALS(coll.getAllowBalance(), true); - ASSERT_EQUALS(coll.getDropped(), false); - } + CollectionType coll = status.getValue(); + ASSERT_TRUE(coll.validate().isOK()); + ASSERT(coll.getNs() == NamespaceString{"db.coll"}); + ASSERT_EQUALS(coll.getEpoch(), oid); + ASSERT_EQUALS(coll.getUpdatedAt(), Date_t::fromMillisSinceEpoch(1)); + ASSERT_EQUALS(coll.getKeyPattern().toBSON(), BSON("a" << 1)); + ASSERT_EQUALS(coll.getUnique(), true); + ASSERT_EQUALS(coll.getAllowBalance(), true); + ASSERT_EQUALS(coll.getDropped(), false); +} - TEST(CollectionType, InvalidCollectionNamespace) { - const OID oid = OID::gen(); - StatusWith<CollectionType> result = CollectionType::fromBSON( - BSON(CollectionType::fullNs("foo\\bar.coll") << - CollectionType::epoch(oid) << - CollectionType::updatedAt(Date_t::fromMillisSinceEpoch(1)) << - CollectionType::keyPattern(BSON("a" << 1)) << - CollectionType::unique(true))); - ASSERT_TRUE(result.isOK()); - CollectionType collType = result.getValue(); - ASSERT_FALSE(collType.validate().isOK()); - } +TEST(CollectionType, InvalidCollectionNamespace) { + const OID oid = OID::gen(); + StatusWith<CollectionType> result = CollectionType::fromBSON(BSON( + CollectionType::fullNs("foo\\bar.coll") + << CollectionType::epoch(oid) << CollectionType::updatedAt(Date_t::fromMillisSinceEpoch(1)) + << CollectionType::keyPattern(BSON("a" << 1)) << CollectionType::unique(true))); + ASSERT_TRUE(result.isOK()); + CollectionType collType = result.getValue(); + ASSERT_FALSE(collType.validate().isOK()); +} - TEST(CollectionType, BadType) { - const OID oid = OID::gen(); - StatusWith<CollectionType> status = CollectionType::fromBSON( - BSON(CollectionType::fullNs() << 1 << - CollectionType::epoch(oid) << - CollectionType::updatedAt(Date_t::fromMillisSinceEpoch(1)) << - CollectionType::keyPattern(BSON("a" << 1)) << - CollectionType::unique(true))); +TEST(CollectionType, BadType) { + const OID oid = OID::gen(); + StatusWith<CollectionType> status = CollectionType::fromBSON( + BSON(CollectionType::fullNs() << 1 << CollectionType::epoch(oid) + << CollectionType::updatedAt(Date_t::fromMillisSinceEpoch(1)) + << CollectionType::keyPattern(BSON("a" << 1)) + << CollectionType::unique(true))); - ASSERT_FALSE(status.isOK()); - } + ASSERT_FALSE(status.isOK()); +} -} // namespace +} // namespace diff --git a/src/mongo/s/catalog/type_database.cpp b/src/mongo/s/catalog/type_database.cpp index 222f46cf301..24de84c838c 100644 --- a/src/mongo/s/catalog/type_database.cpp +++ b/src/mongo/s/catalog/type_database.cpp @@ -38,82 +38,86 @@ namespace mongo { - using std::string; +using std::string; - const std::string DatabaseType::ConfigNS = "config.databases"; +const std::string DatabaseType::ConfigNS = "config.databases"; - const BSONField<std::string> DatabaseType::name("_id"); - const BSONField<std::string> DatabaseType::primary("primary"); - const BSONField<bool> DatabaseType::sharded("partitioned"); +const BSONField<std::string> DatabaseType::name("_id"); +const BSONField<std::string> DatabaseType::primary("primary"); +const BSONField<bool> DatabaseType::sharded("partitioned"); - StatusWith<DatabaseType> DatabaseType::fromBSON(const BSONObj& source) { - DatabaseType dbt; +StatusWith<DatabaseType> DatabaseType::fromBSON(const BSONObj& source) { + DatabaseType dbt; - { - std::string dbtName; - Status status = bsonExtractStringField(source, name.name(), &dbtName); - if (!status.isOK()) return status; + { + std::string dbtName; + Status status = bsonExtractStringField(source, name.name(), &dbtName); + if (!status.isOK()) + return status; - dbt._name = dbtName; - } - - { - std::string dbtPrimary; - Status status = bsonExtractStringField(source, primary.name(), &dbtPrimary); - if (!status.isOK()) return status; + dbt._name = dbtName; + } - dbt._primary = dbtPrimary; - } + { + std::string dbtPrimary; + Status status = bsonExtractStringField(source, primary.name(), &dbtPrimary); + if (!status.isOK()) + return status; - { - bool dbtSharded; - Status status = bsonExtractBooleanFieldWithDefault(source, sharded.name(), false, &dbtSharded); - if (!status.isOK()) return status; + dbt._primary = dbtPrimary; + } - dbt._sharded = dbtSharded; - } + { + bool dbtSharded; + Status status = + bsonExtractBooleanFieldWithDefault(source, sharded.name(), false, &dbtSharded); + if (!status.isOK()) + return status; - return StatusWith<DatabaseType>(dbt); + dbt._sharded = dbtSharded; } - Status DatabaseType::validate() const { - if (!_name.is_initialized() || _name->empty()) { - return Status(ErrorCodes::NoSuchKey, "missing name"); - } + return StatusWith<DatabaseType>(dbt); +} - if (!_primary.is_initialized() || _primary->empty()) { - return Status(ErrorCodes::NoSuchKey, "missing primary"); - } +Status DatabaseType::validate() const { + if (!_name.is_initialized() || _name->empty()) { + return Status(ErrorCodes::NoSuchKey, "missing name"); + } - if (!_sharded.is_initialized()) { - return Status(ErrorCodes::NoSuchKey, "missing sharded"); - } + if (!_primary.is_initialized() || _primary->empty()) { + return Status(ErrorCodes::NoSuchKey, "missing primary"); + } - return Status::OK(); + if (!_sharded.is_initialized()) { + return Status(ErrorCodes::NoSuchKey, "missing sharded"); } - BSONObj DatabaseType::toBSON() const { - BSONObjBuilder builder; - builder.append(name.name(), _name.get_value_or("")); - builder.append(primary.name(), _primary.get_value_or("")); - builder.append(sharded.name(), _sharded.get_value_or(false)); + return Status::OK(); +} - return builder.obj(); - } +BSONObj DatabaseType::toBSON() const { + BSONObjBuilder builder; + builder.append(name.name(), _name.get_value_or("")); + builder.append(primary.name(), _primary.get_value_or("")); + builder.append(sharded.name(), _sharded.get_value_or(false)); - std::string DatabaseType::toString() const { - return toBSON().toString(); - } + return builder.obj(); +} - void DatabaseType::setName(const std::string& name) { - invariant(!name.empty()); - _name = name; - } +std::string DatabaseType::toString() const { + return toBSON().toString(); +} - void DatabaseType::setPrimary(const std::string& primary) { - invariant(!primary.empty()); - _primary = primary; - } +void DatabaseType::setName(const std::string& name) { + invariant(!name.empty()); + _name = name; +} + +void DatabaseType::setPrimary(const std::string& primary) { + invariant(!primary.empty()); + _primary = primary; +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/type_database.h b/src/mongo/s/catalog/type_database.h index 95735903307..e452bd826a2 100644 --- a/src/mongo/s/catalog/type_database.h +++ b/src/mongo/s/catalog/type_database.h @@ -35,67 +35,76 @@ namespace mongo { - class BSONObj; - class Status; - template<typename T> class StatusWith; +class BSONObj; +class Status; +template <typename T> +class StatusWith; - /** - * This class represents the layout and contents of documents contained in the config.databases - * collection. All manipulation of documents coming from that collection should be done with - * this class. - */ - class DatabaseType { - public: - // Name of the databases collection in the config server. - static const std::string ConfigNS; - - static const BSONField<std::string> name; - static const BSONField<std::string> primary; - static const BSONField<bool> sharded; - - - /** - * Constructs a new DatabaseType object from BSON. Also does validation of the contents. - */ - static StatusWith<DatabaseType> fromBSON(const BSONObj& source); - - /** - * Returns OK if all fields have been set. Otherwise returns NoSuchKey and information - * about what is the first field which is missing. - */ - Status validate() const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - const std::string& getName() const { return _name.get(); } - void setName(const std::string& name); +/** + * This class represents the layout and contents of documents contained in the config.databases + * collection. All manipulation of documents coming from that collection should be done with + * this class. + */ +class DatabaseType { +public: + // Name of the databases collection in the config server. + static const std::string ConfigNS; - const std::string& getPrimary() const { return _primary.get(); } - void setPrimary(const std::string& primary); + static const BSONField<std::string> name; + static const BSONField<std::string> primary; + static const BSONField<bool> sharded; - bool getSharded() const { return _sharded.get(); } - void setSharded(bool sharded) { _sharded = sharded; } - private: - // Requred database name - boost::optional<std::string> _name; + /** + * Constructs a new DatabaseType object from BSON. Also does validation of the contents. + */ + static StatusWith<DatabaseType> fromBSON(const BSONObj& source); - // Required primary shard (must be set even if the database is sharded, because there - // might be collections, which are unsharded). - boost::optional<std::string> _primary; + /** + * Returns OK if all fields have been set. Otherwise returns NoSuchKey and information + * about what is the first field which is missing. + */ + Status validate() const; - // Required whether sharding is enabled for this database. Even though this field is of - // type optional, it is only used as an indicator that the value was explicitly set. - boost::optional<bool> _sharded; - }; + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; -} // namespace mongo + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; + + const std::string& getName() const { + return _name.get(); + } + void setName(const std::string& name); + + const std::string& getPrimary() const { + return _primary.get(); + } + void setPrimary(const std::string& primary); + + bool getSharded() const { + return _sharded.get(); + } + void setSharded(bool sharded) { + _sharded = sharded; + } + +private: + // Requred database name + boost::optional<std::string> _name; + + // Required primary shard (must be set even if the database is sharded, because there + // might be collections, which are unsharded). + boost::optional<std::string> _primary; + + // Required whether sharding is enabled for this database. Even though this field is of + // type optional, it is only used as an indicator that the value was explicitly set. + boost::optional<bool> _sharded; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_database_test.cpp b/src/mongo/s/catalog/type_database_test.cpp index db7f9be9575..f038ee65bb4 100644 --- a/src/mongo/s/catalog/type_database_test.cpp +++ b/src/mongo/s/catalog/type_database_test.cpp @@ -35,36 +35,34 @@ namespace { - using namespace mongo; - using std::string; +using namespace mongo; +using std::string; - TEST(DatabaseType, Empty) { - StatusWith<DatabaseType> status = DatabaseType::fromBSON(BSONObj()); - ASSERT_FALSE(status.isOK()); - } +TEST(DatabaseType, Empty) { + StatusWith<DatabaseType> status = DatabaseType::fromBSON(BSONObj()); + ASSERT_FALSE(status.isOK()); +} - TEST(DatabaseType, Basic) { - StatusWith<DatabaseType> status = DatabaseType::fromBSON( - BSON(DatabaseType::name("mydb") << - DatabaseType::primary("shard") << - DatabaseType::sharded(true))); - ASSERT_TRUE(status.isOK()); +TEST(DatabaseType, Basic) { + StatusWith<DatabaseType> status = + DatabaseType::fromBSON(BSON(DatabaseType::name("mydb") << DatabaseType::primary("shard") + << DatabaseType::sharded(true))); + ASSERT_TRUE(status.isOK()); - DatabaseType db = status.getValue(); - ASSERT_EQUALS(db.getName(), "mydb"); - ASSERT_EQUALS(db.getPrimary(), "shard"); - ASSERT_TRUE(db.getSharded()); - } + DatabaseType db = status.getValue(); + ASSERT_EQUALS(db.getName(), "mydb"); + ASSERT_EQUALS(db.getPrimary(), "shard"); + ASSERT_TRUE(db.getSharded()); +} - TEST(DatabaseType, BadType) { - StatusWith<DatabaseType> status = DatabaseType::fromBSON(BSON(DatabaseType::name() << 0)); - ASSERT_FALSE(status.isOK()); - } +TEST(DatabaseType, BadType) { + StatusWith<DatabaseType> status = DatabaseType::fromBSON(BSON(DatabaseType::name() << 0)); + ASSERT_FALSE(status.isOK()); +} - TEST(DatabaseType, MissingRequired) { - StatusWith<DatabaseType> status = DatabaseType::fromBSON( - BSON(DatabaseType::name("mydb"))); - ASSERT_FALSE(status.isOK()); - } +TEST(DatabaseType, MissingRequired) { + StatusWith<DatabaseType> status = DatabaseType::fromBSON(BSON(DatabaseType::name("mydb"))); + ASSERT_FALSE(status.isOK()); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/catalog/type_settings.cpp b/src/mongo/s/catalog/type_settings.cpp index 1f71888d9cc..c19809efa30 100644 --- a/src/mongo/s/catalog/type_settings.cpp +++ b/src/mongo/s/catalog/type_settings.cpp @@ -46,261 +46,249 @@ namespace mongo { - const std::string SettingsType::ConfigNS = "config.settings"; - const std::string SettingsType::BalancerDocKey("balancer"); - const std::string SettingsType::ChunkSizeDocKey("chunksize"); - - const BSONField<std::string> SettingsType::key("_id"); - const BSONField<long long> SettingsType::chunkSizeMB("value"); - const BSONField<bool> SettingsType::balancerStopped("stopped"); - const BSONField<BSONObj> SettingsType::balancerActiveWindow("activeWindow"); - const BSONField<bool> SettingsType::deprecated_secondaryThrottle("_secondaryThrottle"); - const BSONField<BSONObj> SettingsType::migrationWriteConcern("_secondaryThrottle"); - const BSONField<bool> SettingsType::waitForDelete("_waitForDelete"); - - StatusWith<SettingsType> SettingsType::fromBSON(const BSONObj& source) { - SettingsType settings; +const std::string SettingsType::ConfigNS = "config.settings"; +const std::string SettingsType::BalancerDocKey("balancer"); +const std::string SettingsType::ChunkSizeDocKey("chunksize"); + +const BSONField<std::string> SettingsType::key("_id"); +const BSONField<long long> SettingsType::chunkSizeMB("value"); +const BSONField<bool> SettingsType::balancerStopped("stopped"); +const BSONField<BSONObj> SettingsType::balancerActiveWindow("activeWindow"); +const BSONField<bool> SettingsType::deprecated_secondaryThrottle("_secondaryThrottle"); +const BSONField<BSONObj> SettingsType::migrationWriteConcern("_secondaryThrottle"); +const BSONField<bool> SettingsType::waitForDelete("_waitForDelete"); + +StatusWith<SettingsType> SettingsType::fromBSON(const BSONObj& source) { + SettingsType settings; + + { + std::string settingsKey; + Status status = bsonExtractStringField(source, key.name(), &settingsKey); + if (!status.isOK()) + return status; + settings._key = settingsKey; + } + if (settings._key == ChunkSizeDocKey) { + long long settingsChunkSizeMB; + Status status = bsonExtractIntegerField(source, chunkSizeMB.name(), &settingsChunkSizeMB); + if (!status.isOK()) + return status; + settings._chunkSizeMB = settingsChunkSizeMB; + } else if (settings._key == BalancerDocKey) { { - std::string settingsKey; - Status status = bsonExtractStringField(source, key.name(), &settingsKey); - if (!status.isOK()) return status; - settings._key = settingsKey; + bool settingsBalancerStopped; + Status status = bsonExtractBooleanFieldWithDefault( + source, balancerStopped.name(), false, &settingsBalancerStopped); + if (!status.isOK()) + return status; + settings._balancerStopped = settingsBalancerStopped; } - if (settings._key == ChunkSizeDocKey) { - long long settingsChunkSizeMB; - Status status = bsonExtractIntegerField(source, - chunkSizeMB.name(), - &settingsChunkSizeMB); - if (!status.isOK()) return status; - settings._chunkSizeMB = settingsChunkSizeMB; - } - else if (settings._key == BalancerDocKey) { - { - bool settingsBalancerStopped; - Status status = bsonExtractBooleanFieldWithDefault(source, - balancerStopped.name(), - false, - &settingsBalancerStopped); - if (!status.isOK()) return status; - settings._balancerStopped = settingsBalancerStopped; - } - - { - BSONElement settingsBalancerActiveWindowElem; - Status status = bsonExtractTypedField(source, - balancerActiveWindow.name(), - Object, - &settingsBalancerActiveWindowElem); - if (status != ErrorCodes::NoSuchKey) { - if (!status.isOK()) return status; - StatusWith<BoostTimePair> timePairResult = - settings._parseBalancingWindow(settingsBalancerActiveWindowElem.Obj()); - if (!timePairResult.isOK()) return timePairResult.getStatus(); - settings._balancerActiveWindow = timePairResult.getValue(); - } - } - - { - BSONElement settingsMigrationWriteConcernElem; - Status status = bsonExtractTypedField(source, - migrationWriteConcern.name(), - Object, - &settingsMigrationWriteConcernElem); - if (status == ErrorCodes::TypeMismatch) { - bool settingsSecondaryThrottle; - status = bsonExtractBooleanFieldWithDefault(source, - deprecated_secondaryThrottle - .name(), - true, - &settingsSecondaryThrottle); - if (!status.isOK()) return status; - settings._secondaryThrottle = settingsSecondaryThrottle; - } - else if (status != ErrorCodes::NoSuchKey) { - if (!status.isOK()) return status; - settings._migrationWriteConcern = WriteConcernOptions(); - status = settings._migrationWriteConcern->parse( - settingsMigrationWriteConcernElem.Obj() - ); - if (!status.isOK()) return status; - } - } - - { - bool settingsWaitForDelete; - Status status = bsonExtractBooleanField(source, - waitForDelete.name(), - &settingsWaitForDelete); - if (status != ErrorCodes::NoSuchKey) { - if (!status.isOK()) return status; - settings._waitForDelete = settingsWaitForDelete; - } + { + BSONElement settingsBalancerActiveWindowElem; + Status status = bsonExtractTypedField( + source, balancerActiveWindow.name(), Object, &settingsBalancerActiveWindowElem); + if (status != ErrorCodes::NoSuchKey) { + if (!status.isOK()) + return status; + StatusWith<BoostTimePair> timePairResult = + settings._parseBalancingWindow(settingsBalancerActiveWindowElem.Obj()); + if (!timePairResult.isOK()) + return timePairResult.getStatus(); + settings._balancerActiveWindow = timePairResult.getValue(); } } - return settings; - } - - Status SettingsType::validate() const { - if (!_key.is_initialized() || _key->empty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << key.name() << " field"); - } - - if (_key == ChunkSizeDocKey) { - if (!(getChunkSizeMB() > 0)) { - return Status(ErrorCodes::BadValue, - str::stream() << "chunksize specified in " << chunkSizeMB.name() - << " field must be greater than zero"); + { + BSONElement settingsMigrationWriteConcernElem; + Status status = bsonExtractTypedField( + source, migrationWriteConcern.name(), Object, &settingsMigrationWriteConcernElem); + if (status == ErrorCodes::TypeMismatch) { + bool settingsSecondaryThrottle; + status = bsonExtractBooleanFieldWithDefault( + source, deprecated_secondaryThrottle.name(), true, &settingsSecondaryThrottle); + if (!status.isOK()) + return status; + settings._secondaryThrottle = settingsSecondaryThrottle; + } else if (status != ErrorCodes::NoSuchKey) { + if (!status.isOK()) + return status; + settings._migrationWriteConcern = WriteConcernOptions(); + status = + settings._migrationWriteConcern->parse(settingsMigrationWriteConcernElem.Obj()); + if (!status.isOK()) + return status; } } - else if (_key == BalancerDocKey) { - if (_secondaryThrottle.is_initialized() && - _migrationWriteConcern.is_initialized()) { - return Status(ErrorCodes::BadValue, - str::stream() << "cannot have both secondary throttle and " - << "migration write concern set at the same time"); + + { + bool settingsWaitForDelete; + Status status = + bsonExtractBooleanField(source, waitForDelete.name(), &settingsWaitForDelete); + if (status != ErrorCodes::NoSuchKey) { + if (!status.isOK()) + return status; + settings._waitForDelete = settingsWaitForDelete; } } - else { - return Status(ErrorCodes::UnsupportedFormat, - str::stream() << "unsupported key in " << key.name() << " field"); - } - - return Status::OK(); } - BSONObj SettingsType::toBSON() const { - BSONObjBuilder builder; + return settings; +} - if (_key) builder.append(key(), getKey()); - if (_chunkSizeMB) builder.append(chunkSizeMB(), getChunkSizeMB()); - if (_balancerStopped) builder.append(balancerStopped(), getBalancerStopped()); - if (_secondaryThrottle) { - builder.append(deprecated_secondaryThrottle(), getSecondaryThrottle()); - } - if (_migrationWriteConcern) { - builder.append(migrationWriteConcern(), getMigrationWriteConcern().toBSON()); - } - if (_waitForDelete) builder.append(waitForDelete(), getWaitForDelete()); - - return builder.obj(); +Status SettingsType::validate() const { + if (!_key.is_initialized() || _key->empty()) { + return Status(ErrorCodes::NoSuchKey, str::stream() << "missing " << key.name() << " field"); } - std::string SettingsType::toString() const { - return toBSON().toString(); - } - - std::unique_ptr<WriteConcernOptions> SettingsType::getWriteConcern() const { - dassert(_key.is_initialized()); - dassert(_key == BalancerDocKey); - - if (isSecondaryThrottleSet() && !getSecondaryThrottle()) { - return stdx::make_unique<WriteConcernOptions>(1, WriteConcernOptions::NONE, 0); - } - else if (!isMigrationWriteConcernSet()) { - // Default setting. - return nullptr; - } - else { - return stdx::make_unique<WriteConcernOptions>(getMigrationWriteConcern()); - } - } - - StatusWith<BoostTimePair> SettingsType::_parseBalancingWindow(const BSONObj& balancingWindowObj) { - if (balancingWindowObj.isEmpty()) { + if (_key == ChunkSizeDocKey) { + if (!(getChunkSizeMB() > 0)) { return Status(ErrorCodes::BadValue, - "'activeWindow' can't be empty"); + str::stream() << "chunksize specified in " << chunkSizeMB.name() + << " field must be greater than zero"); } - - // check if both 'start' and 'stop' are present - std::string start = balancingWindowObj.getField("start").str(); - std::string stop = balancingWindowObj.getField("stop").str(); - if (start.empty() || stop.empty()) { + } else if (_key == BalancerDocKey) { + if (_secondaryThrottle.is_initialized() && _migrationWriteConcern.is_initialized()) { return Status(ErrorCodes::BadValue, - str::stream() << "must specify both start and end of balancing window: " - << balancingWindowObj); + str::stream() << "cannot have both secondary throttle and " + << "migration write concern set at the same time"); } - - // check that both 'start' and 'stop' are valid time-of-day - boost::posix_time::ptime startTime, stopTime; - if (!toPointInTime(start, &startTime) || !toPointInTime(stop, &stopTime)) { - return Status(ErrorCodes::BadValue, - str::stream() << balancerActiveWindow.name() << " format is " - << " { start: \"hh:mm\" , stop: \"hh:mm\" }"); - } - - return std::make_pair(startTime, stopTime); + } else { + return Status(ErrorCodes::UnsupportedFormat, + str::stream() << "unsupported key in " << key.name() << " field"); } - bool SettingsType::inBalancingWindow(const boost::posix_time::ptime& now) const { - if (!_balancerActiveWindow.is_initialized()) { - return true; - } - const boost::posix_time::ptime& startTime = _balancerActiveWindow->first; - const boost::posix_time::ptime& stopTime = _balancerActiveWindow->second; - - LOG(1).stream() << "inBalancingWindow: " - << " now: " << now - << " startTime: " << startTime - << " stopTime: " << stopTime; - - // allow balancing if during the activeWindow - // note that a window may be open during the night - if (stopTime > startTime) { - if ((now >= startTime) && (now <= stopTime)) { - return true; - } - } - else if (startTime > stopTime) { - if ((now >= startTime) || (now <= stopTime)) { - return true; - } - } + return Status::OK(); +} - return false; - } +BSONObj SettingsType::toBSON() const { + BSONObjBuilder builder; - void SettingsType::setKey(const std::string& key) { - invariant(!key.empty()); - _key = key; + if (_key) + builder.append(key(), getKey()); + if (_chunkSizeMB) + builder.append(chunkSizeMB(), getChunkSizeMB()); + if (_balancerStopped) + builder.append(balancerStopped(), getBalancerStopped()); + if (_secondaryThrottle) { + builder.append(deprecated_secondaryThrottle(), getSecondaryThrottle()); } - - void SettingsType::setChunkSizeMB(const long long chunkSizeMB) { - invariant(_key == ChunkSizeDocKey); - invariant(chunkSizeMB > 0); - _chunkSizeMB = chunkSizeMB; + if (_migrationWriteConcern) { + builder.append(migrationWriteConcern(), getMigrationWriteConcern().toBSON()); + } + if (_waitForDelete) + builder.append(waitForDelete(), getWaitForDelete()); + + return builder.obj(); +} + +std::string SettingsType::toString() const { + return toBSON().toString(); +} + +std::unique_ptr<WriteConcernOptions> SettingsType::getWriteConcern() const { + dassert(_key.is_initialized()); + dassert(_key == BalancerDocKey); + + if (isSecondaryThrottleSet() && !getSecondaryThrottle()) { + return stdx::make_unique<WriteConcernOptions>(1, WriteConcernOptions::NONE, 0); + } else if (!isMigrationWriteConcernSet()) { + // Default setting. + return nullptr; + } else { + return stdx::make_unique<WriteConcernOptions>(getMigrationWriteConcern()); } +} - void SettingsType::setBalancerStopped(const bool balancerStopped) { - invariant(_key == BalancerDocKey); - _balancerStopped = balancerStopped; +StatusWith<BoostTimePair> SettingsType::_parseBalancingWindow(const BSONObj& balancingWindowObj) { + if (balancingWindowObj.isEmpty()) { + return Status(ErrorCodes::BadValue, "'activeWindow' can't be empty"); } - void SettingsType::setBalancerActiveWindow(const BSONObj& balancerActiveWindow) { - invariant(_key == BalancerDocKey); - StatusWith<BoostTimePair> timePairResult = _parseBalancingWindow(balancerActiveWindow); - invariant(timePairResult.isOK()); - _balancerActiveWindow = timePairResult.getValue(); + // check if both 'start' and 'stop' are present + std::string start = balancingWindowObj.getField("start").str(); + std::string stop = balancingWindowObj.getField("stop").str(); + if (start.empty() || stop.empty()) { + return Status(ErrorCodes::BadValue, + str::stream() << "must specify both start and end of balancing window: " + << balancingWindowObj); } - void SettingsType::setSecondaryThrottle(const bool secondaryThrottle) { - invariant(_key == BalancerDocKey); - _secondaryThrottle = secondaryThrottle; + // check that both 'start' and 'stop' are valid time-of-day + boost::posix_time::ptime startTime, stopTime; + if (!toPointInTime(start, &startTime) || !toPointInTime(stop, &stopTime)) { + return Status(ErrorCodes::BadValue, + str::stream() << balancerActiveWindow.name() << " format is " + << " { start: \"hh:mm\" , stop: \"hh:mm\" }"); } - void SettingsType::setMigrationWriteConcern(const BSONObj& migrationWCBSONObj) { - invariant(_key == BalancerDocKey); - invariant(!migrationWCBSONObj.isEmpty()); - Status status = _migrationWriteConcern->parse(migrationWCBSONObj); - invariant(status.isOK()); + return std::make_pair(startTime, stopTime); +} + +bool SettingsType::inBalancingWindow(const boost::posix_time::ptime& now) const { + if (!_balancerActiveWindow.is_initialized()) { + return true; } + const boost::posix_time::ptime& startTime = _balancerActiveWindow->first; + const boost::posix_time::ptime& stopTime = _balancerActiveWindow->second; + + LOG(1).stream() << "inBalancingWindow: " + << " now: " << now << " startTime: " << startTime << " stopTime: " << stopTime; - void SettingsType::setWaitForDelete(const bool waitForDelete) { - invariant(_key == BalancerDocKey); - _waitForDelete = waitForDelete; + // allow balancing if during the activeWindow + // note that a window may be open during the night + if (stopTime > startTime) { + if ((now >= startTime) && (now <= stopTime)) { + return true; + } + } else if (startTime > stopTime) { + if ((now >= startTime) || (now <= stopTime)) { + return true; + } } -} // namespace mongo + return false; +} + +void SettingsType::setKey(const std::string& key) { + invariant(!key.empty()); + _key = key; +} + +void SettingsType::setChunkSizeMB(const long long chunkSizeMB) { + invariant(_key == ChunkSizeDocKey); + invariant(chunkSizeMB > 0); + _chunkSizeMB = chunkSizeMB; +} + +void SettingsType::setBalancerStopped(const bool balancerStopped) { + invariant(_key == BalancerDocKey); + _balancerStopped = balancerStopped; +} + +void SettingsType::setBalancerActiveWindow(const BSONObj& balancerActiveWindow) { + invariant(_key == BalancerDocKey); + StatusWith<BoostTimePair> timePairResult = _parseBalancingWindow(balancerActiveWindow); + invariant(timePairResult.isOK()); + _balancerActiveWindow = timePairResult.getValue(); +} + +void SettingsType::setSecondaryThrottle(const bool secondaryThrottle) { + invariant(_key == BalancerDocKey); + _secondaryThrottle = secondaryThrottle; +} + +void SettingsType::setMigrationWriteConcern(const BSONObj& migrationWCBSONObj) { + invariant(_key == BalancerDocKey); + invariant(!migrationWCBSONObj.isEmpty()); + Status status = _migrationWriteConcern->parse(migrationWCBSONObj); + invariant(status.isOK()); +} + +void SettingsType::setWaitForDelete(const bool waitForDelete) { + invariant(_key == BalancerDocKey); + _waitForDelete = waitForDelete; +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_settings.h b/src/mongo/s/catalog/type_settings.h index 098d7466997..b6954f9d5e4 100644 --- a/src/mongo/s/catalog/type_settings.h +++ b/src/mongo/s/catalog/type_settings.h @@ -38,153 +38,178 @@ namespace mongo { - struct WriteConcernOptions; - class BSONObj; - template<typename T> class StatusWith; +struct WriteConcernOptions; +class BSONObj; +template <typename T> +class StatusWith; - using BoostTimePair = std::pair<boost::posix_time::ptime, boost::posix_time::ptime>; +using BoostTimePair = std::pair<boost::posix_time::ptime, boost::posix_time::ptime>; +/** + * This class represents the layout and contents of documents contained in the + * config.settings collection. All manipulation of documents coming from that + * collection should be done with this class. + * + * Usage Example: + * + * // Contact the config. 'conn' has been obtained before. + * DBClientBase* conn; + * BSONObj query = QUERY(SettingsType::exampleField(SettingsType::ExampleFieldName)); + * exampleDoc = conn->findOne(SettingsType::ConfigNS, query); + * + * // Process the response. + * StatusWith<SettingsType> exampleResult = SettingsType::fromBSON(exampleDoc); + * if (!exampleResult.isOK()) { + * if (exampleResult.getStatus() == ErrorCodes::NoSuchKey) { + * // exampleDoc has no key set or is empty + * } + * // handle error -- exampleResult.getStatus() + * } + * SettingsType exampleType = exampleResult.getValue(); + */ +class SettingsType { +public: + // Name of the settings collection in the config server. + static const std::string ConfigNS; + + static const std::string BalancerDocKey; + static const std::string ChunkSizeDocKey; + + // Field names and types in the settings collection type. + static const BSONField<std::string> key; + static const BSONField<long long> chunkSizeMB; + static const BSONField<bool> balancerStopped; + static const BSONField<BSONObj> balancerActiveWindow; + static const BSONField<bool> deprecated_secondaryThrottle; + static const BSONField<BSONObj> migrationWriteConcern; + static const BSONField<bool> waitForDelete; + + /** + * Returns OK if all mandatory fields have been set and their corresponding + * values are valid. + */ + Status validate() const; + + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; + + /** + * Constructs a new ShardType object from BSON. + * Also does validation of the contents. + */ + static StatusWith<SettingsType> fromBSON(const BSONObj& source); + + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; + + /** + * Returns the write concern to use for balancing. + * Uses *deprecated* secondary throttle if migration write concern is not set. + */ + std::unique_ptr<WriteConcernOptions> getWriteConcern() const; + + /** + * Returns true if either 'now' is in the balancing window or + * if no balancing window exists. + */ + bool inBalancingWindow(const boost::posix_time::ptime& now) const; + + bool isKeySet() const { + return _key.is_initialized(); + } + const std::string& getKey() const { + return _key.get(); + } + void setKey(const std::string& key); + + bool isChunkSizeMBSet() const { + return _chunkSizeMB.is_initialized(); + } + long long getChunkSizeMB() const { + return _chunkSizeMB.get(); + } + void setChunkSizeMB(const long long chunkSizeMB); + + bool isBalancerStoppedSet() const { + return _balancerStopped.is_initialized(); + } + bool getBalancerStopped() const { + return _balancerStopped.get(); + } + void setBalancerStopped(const bool balancerStopped); + + bool isBalancerActiveWindowSet() const { + return _balancerActiveWindow.is_initialized(); + } + const BoostTimePair& getBalancerActiveWindow() const { + return _balancerActiveWindow.get(); + } + void setBalancerActiveWindow(const BSONObj& balancerActiveWindow); + + bool isSecondaryThrottleSet() const { + return _secondaryThrottle.is_initialized(); + } + bool getSecondaryThrottle() const { + return _secondaryThrottle.get(); + } + void setSecondaryThrottle(const bool secondaryThrottle); + + bool isMigrationWriteConcernSet() const { + return _migrationWriteConcern.is_initialized(); + } + const WriteConcernOptions& getMigrationWriteConcern() const { + return _migrationWriteConcern.get(); + } + void setMigrationWriteConcern(const BSONObj& migrationWriteConcern); + + bool isWaitForDeleteSet() const { + return _waitForDelete.is_initialized(); + } + bool getWaitForDelete() const { + return _waitForDelete.get(); + } + void setWaitForDelete(const bool waitForDelete); + +private: /** - * This class represents the layout and contents of documents contained in the - * config.settings collection. All manipulation of documents coming from that - * collection should be done with this class. - * - * Usage Example: - * - * // Contact the config. 'conn' has been obtained before. - * DBClientBase* conn; - * BSONObj query = QUERY(SettingsType::exampleField(SettingsType::ExampleFieldName)); - * exampleDoc = conn->findOne(SettingsType::ConfigNS, query); - * - * // Process the response. - * StatusWith<SettingsType> exampleResult = SettingsType::fromBSON(exampleDoc); - * if (!exampleResult.isOK()) { - * if (exampleResult.getStatus() == ErrorCodes::NoSuchKey) { - * // exampleDoc has no key set or is empty - * } - * // handle error -- exampleResult.getStatus() - * } - * SettingsType exampleType = exampleResult.getValue(); + * Used to parse balancing 'activeWindow'. + * See '_balancerActiveWindow' member variable doc comments below. */ - class SettingsType { - public: - - // Name of the settings collection in the config server. - static const std::string ConfigNS; - - static const std::string BalancerDocKey; - static const std::string ChunkSizeDocKey; - - // Field names and types in the settings collection type. - static const BSONField<std::string> key; - static const BSONField<long long> chunkSizeMB; - static const BSONField<bool> balancerStopped; - static const BSONField<BSONObj> balancerActiveWindow; - static const BSONField<bool> deprecated_secondaryThrottle; - static const BSONField<BSONObj> migrationWriteConcern; - static const BSONField<bool> waitForDelete; - - /** - * Returns OK if all mandatory fields have been set and their corresponding - * values are valid. - */ - Status validate() const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Constructs a new ShardType object from BSON. - * Also does validation of the contents. - */ - static StatusWith<SettingsType> fromBSON(const BSONObj& source); - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - /** - * Returns the write concern to use for balancing. - * Uses *deprecated* secondary throttle if migration write concern is not set. - */ - std::unique_ptr<WriteConcernOptions> getWriteConcern() const; - - /** - * Returns true if either 'now' is in the balancing window or - * if no balancing window exists. - */ - bool inBalancingWindow(const boost::posix_time::ptime& now) const; - - bool isKeySet() const { return _key.is_initialized(); } - const std::string& getKey() const { return _key.get(); } - void setKey(const std::string& key); - - bool isChunkSizeMBSet() const { return _chunkSizeMB.is_initialized(); } - long long getChunkSizeMB() const { return _chunkSizeMB.get(); } - void setChunkSizeMB(const long long chunkSizeMB); - - bool isBalancerStoppedSet() const { return _balancerStopped.is_initialized(); } - bool getBalancerStopped() const { return _balancerStopped.get(); } - void setBalancerStopped(const bool balancerStopped); - - bool isBalancerActiveWindowSet() const { return _balancerActiveWindow.is_initialized(); } - const BoostTimePair& getBalancerActiveWindow() const { return _balancerActiveWindow.get(); } - void setBalancerActiveWindow(const BSONObj& balancerActiveWindow); - - bool isSecondaryThrottleSet() const { return _secondaryThrottle.is_initialized(); } - bool getSecondaryThrottle() const { return _secondaryThrottle.get(); } - void setSecondaryThrottle(const bool secondaryThrottle); - - bool isMigrationWriteConcernSet() const { return _migrationWriteConcern.is_initialized(); } - const WriteConcernOptions& getMigrationWriteConcern() const { - return _migrationWriteConcern.get(); - } - void setMigrationWriteConcern(const BSONObj& migrationWriteConcern); - - bool isWaitForDeleteSet() const { return _waitForDelete.is_initialized(); } - bool getWaitForDelete() const { return _waitForDelete.get(); } - void setWaitForDelete(const bool waitForDelete); - - private: - - /** - * Used to parse balancing 'activeWindow'. - * See '_balancerActiveWindow' member variable doc comments below. - */ - StatusWith<BoostTimePair> _parseBalancingWindow(const BSONObj& balancingWindowObj); - - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - - // (M) key determining the type of options to use - boost::optional<std::string> _key; - - // (S) size of the chunks in our cluster in MB - // Required if key is chunkSize - boost::optional<long long> _chunkSizeMB; - - // (O) is balancer enabled or disabled (default) - // Defaults to false. - boost::optional<bool> _balancerStopped; - - // (O) if present, balancerActiveWindow is an interval during the day - // when the balancer should be active. - // Stored as (<start time>, <stop time>) pair. strftime format is %H:%M - boost::optional<BoostTimePair> _balancerActiveWindow; - - // (O) only migrate chunks as fast as at least one secondary can keep up with - boost::optional<bool> _secondaryThrottle; - - // (O) detailed write concern for *individual* writes during migration. - // From side: deletes during cleanup. - // To side: deletes to clear the incoming range, deletes to undo migration at abort, - // and writes during cloning. - boost::optional<WriteConcernOptions> _migrationWriteConcern; - - // (O) synchronous migration cleanup. - boost::optional<bool> _waitForDelete; - }; - -} // namespace mongo + StatusWith<BoostTimePair> _parseBalancingWindow(const BSONObj& balancingWindowObj); + + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + + // (M) key determining the type of options to use + boost::optional<std::string> _key; + + // (S) size of the chunks in our cluster in MB + // Required if key is chunkSize + boost::optional<long long> _chunkSizeMB; + + // (O) is balancer enabled or disabled (default) + // Defaults to false. + boost::optional<bool> _balancerStopped; + + // (O) if present, balancerActiveWindow is an interval during the day + // when the balancer should be active. + // Stored as (<start time>, <stop time>) pair. strftime format is %H:%M + boost::optional<BoostTimePair> _balancerActiveWindow; + + // (O) only migrate chunks as fast as at least one secondary can keep up with + boost::optional<bool> _secondaryThrottle; + + // (O) detailed write concern for *individual* writes during migration. + // From side: deletes during cleanup. + // To side: deletes to clear the incoming range, deletes to undo migration at abort, + // and writes during cloning. + boost::optional<WriteConcernOptions> _migrationWriteConcern; + + // (O) synchronous migration cleanup. + boost::optional<bool> _waitForDelete; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_settings_test.cpp b/src/mongo/s/catalog/type_settings_test.cpp index 0fbe6fee3b6..b7ca2f0bd2f 100644 --- a/src/mongo/s/catalog/type_settings_test.cpp +++ b/src/mongo/s/catalog/type_settings_test.cpp @@ -36,178 +36,179 @@ namespace { - using namespace mongo; - - TEST(SettingsType, MissingKey) { - BSONObj objNoKey = BSONObj(); - StatusWith<SettingsType> result = SettingsType::fromBSON(objNoKey); - ASSERT_FALSE(result.isOK()); - ASSERT_EQUALS(result.getStatus(), ErrorCodes::NoSuchKey); - } - - TEST(SettingsType, ChunkSize) { - BSONObj objChunkSizeZero = BSON(SettingsType::key(SettingsType::ChunkSizeDocKey) << - SettingsType::chunkSizeMB(0)); - StatusWith<SettingsType> result = SettingsType::fromBSON(objChunkSizeZero); - ASSERT(result.isOK()); - SettingsType settings = result.getValue(); - ASSERT_EQUALS(settings.getChunkSizeMB(), 0); - Status validationStatus = settings.validate(); - ASSERT_FALSE(validationStatus.isOK()); - ASSERT_EQUALS(validationStatus, ErrorCodes::BadValue); - } - - TEST(SettingsType, UnsupportedSetting) { - BSONObj objBadSetting = BSON(SettingsType::key("badsetting")); - StatusWith<SettingsType> result = SettingsType::fromBSON(objBadSetting); - ASSERT(result.isOK()); - SettingsType settings = result.getValue(); - Status validationStatus = settings.validate(); - ASSERT_FALSE(validationStatus.isOK()); - ASSERT_EQUALS(validationStatus, ErrorCodes::UnsupportedFormat); - } - - TEST(SettingsType, InvalidBalancerWindow) { - BSONObj objBalancerBadKeys = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerActiveWindow(BSON("begin" << - "23:00" << - "end" << - "6:00" ))); - StatusWith<SettingsType> result = SettingsType::fromBSON(objBalancerBadKeys); - ASSERT_FALSE(result.isOK()); - - BSONObj objBalancerBadTimes = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerActiveWindow(BSON("start" << - "23" << - "stop" << - "6" ))); - result = SettingsType::fromBSON(objBalancerBadTimes); - ASSERT_FALSE(result.isOK()); - } - - TEST(SettingsType, ValidValues) { - BSONObj objChunkSize = BSON(SettingsType::key(SettingsType::ChunkSizeDocKey) << - SettingsType::chunkSizeMB(1)); - StatusWith<SettingsType> result = SettingsType::fromBSON(objChunkSize); - SettingsType settings = result.getValue(); - ASSERT(result.isOK()); - Status validationStatus = settings.validate(); - ASSERT(validationStatus.isOK()); - ASSERT_EQUALS(settings.getKey(), SettingsType::ChunkSizeDocKey); - ASSERT_EQUALS(settings.getChunkSizeMB(), 1); - - BSONObj objBalancer = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerStopped(true) << - SettingsType::balancerActiveWindow(BSON("start" << "23:00" << - "stop" << "6:00" )) << - SettingsType::migrationWriteConcern(BSON("w" << 2))); - result = SettingsType::fromBSON(objBalancer); - settings = result.getValue(); - ASSERT(result.isOK()); - validationStatus = settings.validate(); - ASSERT(validationStatus.isOK()); - ASSERT_EQUALS(settings.getKey(), SettingsType::BalancerDocKey); - ASSERT_EQUALS(settings.getBalancerStopped(), true); - - WriteConcernOptions wc; - wc.parse(BSON("w" << 2)); - ASSERT_EQUALS(settings.getMigrationWriteConcern().toBSON(), wc.toBSON()); - } - - TEST(SettingsType, ValidWithDeprecatedThrottle) { - BSONObj objChunkSize = BSON(SettingsType::key(SettingsType::ChunkSizeDocKey) << - SettingsType::chunkSizeMB(1)); - StatusWith<SettingsType> result = SettingsType::fromBSON(objChunkSize); - ASSERT(result.isOK()); - SettingsType settings = result.getValue(); - ASSERT_EQUALS(settings.getKey(), SettingsType::ChunkSizeDocKey); - ASSERT_EQUALS(settings.getChunkSizeMB(), 1); - - BSONObj objBalancer = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::deprecated_secondaryThrottle(true)); - result = SettingsType::fromBSON(objBalancer); - ASSERT_EQUALS(result.getStatus(), Status::OK()); - settings = result.getValue(); - ASSERT_EQUALS(settings.getKey(), SettingsType::BalancerDocKey); - ASSERT(settings.getSecondaryThrottle()); - } - - TEST(SettingsType, BadType) { - BSONObj badTypeObj = BSON(SettingsType::key() << 0); - StatusWith<SettingsType> result = SettingsType::fromBSON(badTypeObj); - ASSERT_FALSE(result.isOK()); - } - - TEST(SettingsType, BalancingWindow) { - // T0 < T1 < now < T2 < T3 and Error - const std::string T0 = "9:00"; - const std::string T1 = "11:00"; - boost::posix_time::ptime now(currentDate(), - boost::posix_time::hours(13) + - boost::posix_time::minutes(48)); - const std::string T2 = "17:00"; - const std::string T3 = "21:30"; - const std::string E = "28:35"; - - // closed in the past - BSONObj w1 = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerActiveWindow(BSON("start" << T0 << "stop" << T1))); - StatusWith<SettingsType> result = SettingsType::fromBSON(w1); - ASSERT(result.isOK()); - ASSERT_FALSE(result.getValue().inBalancingWindow(now)); - - // not opened until the future - BSONObj w2 = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerActiveWindow(BSON("start" << T2 << "stop" << T3))); - result = SettingsType::fromBSON(w2); - ASSERT(result.isOK()); - ASSERT_FALSE(result.getValue().inBalancingWindow(now)); - - // open now - BSONObj w3 = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerActiveWindow(BSON("start" << T1 << "stop" << T2))); - result = SettingsType::fromBSON(w3); - ASSERT(result.isOK()); - ASSERT(result.getValue().inBalancingWindow(now)); - - // open since last day - BSONObj w4 = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerActiveWindow(BSON("start" << T3 << "stop" << T2))); - result = SettingsType::fromBSON(w4); - ASSERT(result.isOK()); - ASSERT(result.getValue().inBalancingWindow(now)); - - // bad input should not stop the balancer - - // empty window - BSONObj w5 = BSON(SettingsType::key(SettingsType::BalancerDocKey)); - result = SettingsType::fromBSON(w5); - ASSERT(result.isOK()); - ASSERT(result.getValue().inBalancingWindow(now)); - - // missing stop - BSONObj w6 = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerActiveWindow(BSON("start" << 1))); - result = SettingsType::fromBSON(w6); - ASSERT_FALSE(result.isOK()); - - // missing start - BSONObj w7 = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - SettingsType::balancerActiveWindow(BSON("stop" << 1))); - result = SettingsType::fromBSON(w7); - ASSERT_FALSE(result.isOK()); - - // active window marker missing - BSONObj w8 = BSON(SettingsType::key(SettingsType::BalancerDocKey) << - "wrongMarker" << BSON("start" << 1 << "stop" << 1)); - result = SettingsType::fromBSON(w8); - ASSERT(result.isOK()); - ASSERT(result.getValue().inBalancingWindow(now)); - - // garbage in window - BSONObj w9 = BSON(SettingsType::balancerActiveWindow(BSON("start" << T3 << "stop" << E))); - result = SettingsType::fromBSON(w9); - ASSERT_FALSE(result.isOK()); - } - -} // unnamed namespace +using namespace mongo; + +TEST(SettingsType, MissingKey) { + BSONObj objNoKey = BSONObj(); + StatusWith<SettingsType> result = SettingsType::fromBSON(objNoKey); + ASSERT_FALSE(result.isOK()); + ASSERT_EQUALS(result.getStatus(), ErrorCodes::NoSuchKey); +} + +TEST(SettingsType, ChunkSize) { + BSONObj objChunkSizeZero = + BSON(SettingsType::key(SettingsType::ChunkSizeDocKey) << SettingsType::chunkSizeMB(0)); + StatusWith<SettingsType> result = SettingsType::fromBSON(objChunkSizeZero); + ASSERT(result.isOK()); + SettingsType settings = result.getValue(); + ASSERT_EQUALS(settings.getChunkSizeMB(), 0); + Status validationStatus = settings.validate(); + ASSERT_FALSE(validationStatus.isOK()); + ASSERT_EQUALS(validationStatus, ErrorCodes::BadValue); +} + +TEST(SettingsType, UnsupportedSetting) { + BSONObj objBadSetting = BSON(SettingsType::key("badsetting")); + StatusWith<SettingsType> result = SettingsType::fromBSON(objBadSetting); + ASSERT(result.isOK()); + SettingsType settings = result.getValue(); + Status validationStatus = settings.validate(); + ASSERT_FALSE(validationStatus.isOK()); + ASSERT_EQUALS(validationStatus, ErrorCodes::UnsupportedFormat); +} + +TEST(SettingsType, InvalidBalancerWindow) { + BSONObj objBalancerBadKeys = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerActiveWindow(BSON("begin" + << "23:00" + << "end" + << "6:00"))); + StatusWith<SettingsType> result = SettingsType::fromBSON(objBalancerBadKeys); + ASSERT_FALSE(result.isOK()); + + BSONObj objBalancerBadTimes = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerActiveWindow(BSON("start" + << "23" + << "stop" + << "6"))); + result = SettingsType::fromBSON(objBalancerBadTimes); + ASSERT_FALSE(result.isOK()); +} + +TEST(SettingsType, ValidValues) { + BSONObj objChunkSize = + BSON(SettingsType::key(SettingsType::ChunkSizeDocKey) << SettingsType::chunkSizeMB(1)); + StatusWith<SettingsType> result = SettingsType::fromBSON(objChunkSize); + SettingsType settings = result.getValue(); + ASSERT(result.isOK()); + Status validationStatus = settings.validate(); + ASSERT(validationStatus.isOK()); + ASSERT_EQUALS(settings.getKey(), SettingsType::ChunkSizeDocKey); + ASSERT_EQUALS(settings.getChunkSizeMB(), 1); + + BSONObj objBalancer = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerStopped(true) + << SettingsType::balancerActiveWindow(BSON("start" + << "23:00" + << "stop" + << "6:00")) + << SettingsType::migrationWriteConcern(BSON("w" << 2))); + result = SettingsType::fromBSON(objBalancer); + settings = result.getValue(); + ASSERT(result.isOK()); + validationStatus = settings.validate(); + ASSERT(validationStatus.isOK()); + ASSERT_EQUALS(settings.getKey(), SettingsType::BalancerDocKey); + ASSERT_EQUALS(settings.getBalancerStopped(), true); + + WriteConcernOptions wc; + wc.parse(BSON("w" << 2)); + ASSERT_EQUALS(settings.getMigrationWriteConcern().toBSON(), wc.toBSON()); +} + +TEST(SettingsType, ValidWithDeprecatedThrottle) { + BSONObj objChunkSize = + BSON(SettingsType::key(SettingsType::ChunkSizeDocKey) << SettingsType::chunkSizeMB(1)); + StatusWith<SettingsType> result = SettingsType::fromBSON(objChunkSize); + ASSERT(result.isOK()); + SettingsType settings = result.getValue(); + ASSERT_EQUALS(settings.getKey(), SettingsType::ChunkSizeDocKey); + ASSERT_EQUALS(settings.getChunkSizeMB(), 1); + + BSONObj objBalancer = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::deprecated_secondaryThrottle(true)); + result = SettingsType::fromBSON(objBalancer); + ASSERT_EQUALS(result.getStatus(), Status::OK()); + settings = result.getValue(); + ASSERT_EQUALS(settings.getKey(), SettingsType::BalancerDocKey); + ASSERT(settings.getSecondaryThrottle()); +} + +TEST(SettingsType, BadType) { + BSONObj badTypeObj = BSON(SettingsType::key() << 0); + StatusWith<SettingsType> result = SettingsType::fromBSON(badTypeObj); + ASSERT_FALSE(result.isOK()); +} + +TEST(SettingsType, BalancingWindow) { + // T0 < T1 < now < T2 < T3 and Error + const std::string T0 = "9:00"; + const std::string T1 = "11:00"; + boost::posix_time::ptime now(currentDate(), + boost::posix_time::hours(13) + boost::posix_time::minutes(48)); + const std::string T2 = "17:00"; + const std::string T3 = "21:30"; + const std::string E = "28:35"; + + // closed in the past + BSONObj w1 = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerActiveWindow(BSON("start" << T0 << "stop" << T1))); + StatusWith<SettingsType> result = SettingsType::fromBSON(w1); + ASSERT(result.isOK()); + ASSERT_FALSE(result.getValue().inBalancingWindow(now)); + + // not opened until the future + BSONObj w2 = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerActiveWindow(BSON("start" << T2 << "stop" << T3))); + result = SettingsType::fromBSON(w2); + ASSERT(result.isOK()); + ASSERT_FALSE(result.getValue().inBalancingWindow(now)); + + // open now + BSONObj w3 = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerActiveWindow(BSON("start" << T1 << "stop" << T2))); + result = SettingsType::fromBSON(w3); + ASSERT(result.isOK()); + ASSERT(result.getValue().inBalancingWindow(now)); + + // open since last day + BSONObj w4 = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerActiveWindow(BSON("start" << T3 << "stop" << T2))); + result = SettingsType::fromBSON(w4); + ASSERT(result.isOK()); + ASSERT(result.getValue().inBalancingWindow(now)); + + // bad input should not stop the balancer + + // empty window + BSONObj w5 = BSON(SettingsType::key(SettingsType::BalancerDocKey)); + result = SettingsType::fromBSON(w5); + ASSERT(result.isOK()); + ASSERT(result.getValue().inBalancingWindow(now)); + + // missing stop + BSONObj w6 = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerActiveWindow(BSON("start" << 1))); + result = SettingsType::fromBSON(w6); + ASSERT_FALSE(result.isOK()); + + // missing start + BSONObj w7 = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << SettingsType::balancerActiveWindow(BSON("stop" << 1))); + result = SettingsType::fromBSON(w7); + ASSERT_FALSE(result.isOK()); + + // active window marker missing + BSONObj w8 = BSON(SettingsType::key(SettingsType::BalancerDocKey) + << "wrongMarker" << BSON("start" << 1 << "stop" << 1)); + result = SettingsType::fromBSON(w8); + ASSERT(result.isOK()); + ASSERT(result.getValue().inBalancingWindow(now)); + + // garbage in window + BSONObj w9 = BSON(SettingsType::balancerActiveWindow(BSON("start" << T3 << "stop" << E))); + result = SettingsType::fromBSON(w9); + ASSERT_FALSE(result.isOK()); +} + +} // unnamed namespace diff --git a/src/mongo/s/catalog/type_shard.cpp b/src/mongo/s/catalog/type_shard.cpp index 8fafe50c549..898d784cf75 100644 --- a/src/mongo/s/catalog/type_shard.cpp +++ b/src/mongo/s/catalog/type_shard.cpp @@ -39,143 +39,143 @@ namespace mongo { - const std::string ShardType::ConfigNS = "config.shards"; +const std::string ShardType::ConfigNS = "config.shards"; - const BSONField<std::string> ShardType::name("_id"); - const BSONField<std::string> ShardType::host("host"); - const BSONField<bool> ShardType::draining("draining"); - const BSONField<long long> ShardType::maxSizeMB("maxSize"); - const BSONField<BSONArray> ShardType::tags("tags"); +const BSONField<std::string> ShardType::name("_id"); +const BSONField<std::string> ShardType::host("host"); +const BSONField<bool> ShardType::draining("draining"); +const BSONField<long long> ShardType::maxSizeMB("maxSize"); +const BSONField<BSONArray> ShardType::tags("tags"); - StatusWith<ShardType> ShardType::fromBSON(const BSONObj& source) { - ShardType shard; +StatusWith<ShardType> ShardType::fromBSON(const BSONObj& source) { + ShardType shard; - { - std::string shardName; - Status status = bsonExtractStringField(source, name.name(), &shardName); - if (!status.isOK()) return status; - shard._name = shardName; - } - - { - std::string shardHost; - Status status = bsonExtractStringField(source, host.name(), &shardHost); - if (!status.isOK()) return status; - shard._host = shardHost; - } - - { - bool isShardDraining; - Status status = bsonExtractBooleanField(source, - draining.name(), - &isShardDraining); - if (status.isOK()) { - shard._draining = isShardDraining; - } - else if (status == ErrorCodes::NoSuchKey) { - // draining field can be mssing in which case it is presumed false - } - else { - return status; - } - } - - { - long long shardMaxSizeMB; - // maxSizeMB == 0 means there's no limitation to space usage. - Status status = bsonExtractIntegerField(source, - maxSizeMB.name(), - &shardMaxSizeMB); - if (status.isOK()) { - shard._maxSizeMB = shardMaxSizeMB; - } - else if (status == ErrorCodes::NoSuchKey) { - // maxSizeMB field can be missing in which case it is presumed false - } - else { - return status; - } - } - - if (source.hasField(tags.name())) { - shard._tags = std::vector<std::string>(); - BSONElement tagsElement; - Status status = bsonExtractTypedField(source, tags.name(), Array, &tagsElement); - if (!status.isOK()) return status; - - BSONObjIterator it(tagsElement.Obj()); - while (it.more()) { - BSONElement tagElement = it.next(); - if (tagElement.type() != String) { - return Status(ErrorCodes::TypeMismatch, - str::stream() << "Elements in \"" << tags.name() - << "\" array must be strings but found " - << typeName(tagElement.type())); - } - shard._tags->push_back(tagElement.String()); - } - } - - return shard; + { + std::string shardName; + Status status = bsonExtractStringField(source, name.name(), &shardName); + if (!status.isOK()) + return status; + shard._name = shardName; } - Status ShardType::validate() const { - if (!_name.is_initialized() || _name->empty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << name.name() << " field"); - } - - if (!_host.is_initialized() || _host->empty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << host.name() << " field"); - } - - if (_maxSizeMB.is_initialized() && getMaxSizeMB() < 0) { - return Status(ErrorCodes::BadValue, str::stream() << "maxSize can't be negative"); - } - - return Status::OK(); + { + std::string shardHost; + Status status = bsonExtractStringField(source, host.name(), &shardHost); + if (!status.isOK()) + return status; + shard._host = shardHost; } - BSONObj ShardType::toBSON() const { - BSONObjBuilder builder; - - if (_name) builder.append(name(), getName()); - if (_host) builder.append(host(), getHost()); - if (_draining) builder.append(draining(), getDraining()); - if (_maxSizeMB) builder.append(maxSizeMB(), getMaxSizeMB()); - if (_tags) builder.append(tags(), getTags()); - - return builder.obj(); + { + bool isShardDraining; + Status status = bsonExtractBooleanField(source, draining.name(), &isShardDraining); + if (status.isOK()) { + shard._draining = isShardDraining; + } else if (status == ErrorCodes::NoSuchKey) { + // draining field can be mssing in which case it is presumed false + } else { + return status; + } } - std::string ShardType::toString() const { - return toBSON().toString(); + { + long long shardMaxSizeMB; + // maxSizeMB == 0 means there's no limitation to space usage. + Status status = bsonExtractIntegerField(source, maxSizeMB.name(), &shardMaxSizeMB); + if (status.isOK()) { + shard._maxSizeMB = shardMaxSizeMB; + } else if (status == ErrorCodes::NoSuchKey) { + // maxSizeMB field can be missing in which case it is presumed false + } else { + return status; + } } - void ShardType::setName(const std::string& name) { - invariant(!name.empty()); - _name = name; + if (source.hasField(tags.name())) { + shard._tags = std::vector<std::string>(); + BSONElement tagsElement; + Status status = bsonExtractTypedField(source, tags.name(), Array, &tagsElement); + if (!status.isOK()) + return status; + + BSONObjIterator it(tagsElement.Obj()); + while (it.more()) { + BSONElement tagElement = it.next(); + if (tagElement.type() != String) { + return Status(ErrorCodes::TypeMismatch, + str::stream() << "Elements in \"" << tags.name() + << "\" array must be strings but found " + << typeName(tagElement.type())); + } + shard._tags->push_back(tagElement.String()); + } } - void ShardType::setHost(const std::string& host) { - invariant(!host.empty()); - _host = host; - } + return shard; +} - void ShardType::setDraining(const bool isDraining) { - _draining = isDraining; +Status ShardType::validate() const { + if (!_name.is_initialized() || _name->empty()) { + return Status(ErrorCodes::NoSuchKey, + str::stream() << "missing " << name.name() << " field"); } - void ShardType::setMaxSizeMB(const long long maxSizeMB) { - invariant(maxSizeMB >= 0); - _maxSizeMB = maxSizeMB; + if (!_host.is_initialized() || _host->empty()) { + return Status(ErrorCodes::NoSuchKey, + str::stream() << "missing " << host.name() << " field"); } - void ShardType::setTags(const std::vector<std::string>& tags) { - invariant(tags.size() > 0); - _tags = tags; + if (_maxSizeMB.is_initialized() && getMaxSizeMB() < 0) { + return Status(ErrorCodes::BadValue, str::stream() << "maxSize can't be negative"); } -} // namespace mongo + return Status::OK(); +} + +BSONObj ShardType::toBSON() const { + BSONObjBuilder builder; + + if (_name) + builder.append(name(), getName()); + if (_host) + builder.append(host(), getHost()); + if (_draining) + builder.append(draining(), getDraining()); + if (_maxSizeMB) + builder.append(maxSizeMB(), getMaxSizeMB()); + if (_tags) + builder.append(tags(), getTags()); + + return builder.obj(); +} + +std::string ShardType::toString() const { + return toBSON().toString(); +} + +void ShardType::setName(const std::string& name) { + invariant(!name.empty()); + _name = name; +} + +void ShardType::setHost(const std::string& host) { + invariant(!host.empty()); + _host = host; +} + +void ShardType::setDraining(const bool isDraining) { + _draining = isDraining; +} + +void ShardType::setMaxSizeMB(const long long maxSizeMB) { + invariant(maxSizeMB >= 0); + _maxSizeMB = maxSizeMB; +} + +void ShardType::setTags(const std::vector<std::string>& tags) { + invariant(tags.size() > 0); + _tags = tags; +} + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_shard.h b/src/mongo/s/catalog/type_shard.h index 9a4a5c5c55b..9b30e1a9505 100644 --- a/src/mongo/s/catalog/type_shard.h +++ b/src/mongo/s/catalog/type_shard.h @@ -36,82 +36,90 @@ namespace mongo { - struct BSONArray; - class BSONObj; - class Status; - template<typename T> class StatusWith; +struct BSONArray; +class BSONObj; +class Status; +template <typename T> +class StatusWith; + +/** + * This class represents the layout and contents of documents contained in the + * config.shards collection. All manipulation of documents coming from that + * collection should be done with this class. + */ +class ShardType { +public: + // Name of the shards collection in the config server. + static const std::string ConfigNS; + + // Field names and types in the shards collection type. + static const BSONField<std::string> name; + static const BSONField<std::string> host; + static const BSONField<bool> draining; + static const BSONField<long long> maxSizeMB; + static const BSONField<BSONArray> tags; + + + /** + * Constructs a new ShardType object from BSON. + * Also does validation of the contents. + */ + static StatusWith<ShardType> fromBSON(const BSONObj& source); + + /** + * Returns OK if all fields have been set. Otherwise returns NoSuchKey + * and information about the first field that is missing. + */ + Status validate() const; + + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; /** - * This class represents the layout and contents of documents contained in the - * config.shards collection. All manipulation of documents coming from that - * collection should be done with this class. + * Returns a std::string representation of the current internal state. */ - class ShardType { - public: - - // Name of the shards collection in the config server. - static const std::string ConfigNS; - - // Field names and types in the shards collection type. - static const BSONField<std::string> name; - static const BSONField<std::string> host; - static const BSONField<bool> draining; - static const BSONField<long long> maxSizeMB; - static const BSONField<BSONArray> tags; - - - /** - * Constructs a new ShardType object from BSON. - * Also does validation of the contents. - */ - static StatusWith<ShardType> fromBSON(const BSONObj& source); - - /** - * Returns OK if all fields have been set. Otherwise returns NoSuchKey - * and information about the first field that is missing. - */ - Status validate() const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - const std::string& getName() const { return _name.get(); } - void setName(const std::string& name); - - const std::string& getHost() const { return _host.get(); } - void setHost(const std::string& host); - - bool getDraining() const { return _draining.value_or(false); } - void setDraining(const bool draining); - - long long getMaxSizeMB() const { return _maxSizeMB.value_or(0); } - void setMaxSizeMB(const long long maxSizeMB); - - const std::vector<std::string> getTags() const { - return _tags.value_or(std::vector<std::string>()); - } - void setTags(const std::vector<std::string>& tags); - - private: - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - - // (M) shard's id - boost::optional<std::string> _name; - // (M) connection string for the host(s) - boost::optional<std::string> _host; - // (O) is it draining chunks? - boost::optional<bool> _draining; - // (O) maximum allowed disk space in MB - boost::optional<long long> _maxSizeMB; - // (O) shard tags - boost::optional<std::vector<std::string>> _tags; - }; - -} // namespace mongo + std::string toString() const; + + const std::string& getName() const { + return _name.get(); + } + void setName(const std::string& name); + + const std::string& getHost() const { + return _host.get(); + } + void setHost(const std::string& host); + + bool getDraining() const { + return _draining.value_or(false); + } + void setDraining(const bool draining); + + long long getMaxSizeMB() const { + return _maxSizeMB.value_or(0); + } + void setMaxSizeMB(const long long maxSizeMB); + + const std::vector<std::string> getTags() const { + return _tags.value_or(std::vector<std::string>()); + } + void setTags(const std::vector<std::string>& tags); + +private: + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + + // (M) shard's id + boost::optional<std::string> _name; + // (M) connection string for the host(s) + boost::optional<std::string> _host; + // (O) is it draining chunks? + boost::optional<bool> _draining; + // (O) maximum allowed disk space in MB + boost::optional<long long> _maxSizeMB; + // (O) shard tags + boost::optional<std::vector<std::string>> _tags; +}; + +} // namespace mongo diff --git a/src/mongo/s/catalog/type_shard_test.cpp b/src/mongo/s/catalog/type_shard_test.cpp index a7db4f079f2..f89a54d5a3f 100644 --- a/src/mongo/s/catalog/type_shard_test.cpp +++ b/src/mongo/s/catalog/type_shard_test.cpp @@ -36,56 +36,53 @@ namespace { - using namespace mongo; +using namespace mongo; - using std::string; +using std::string; - TEST(ShardType, MissingName) { - BSONObj obj = BSON(ShardType::host("localhost:27017")); - StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); - ASSERT_FALSE(shardRes.isOK()); - } +TEST(ShardType, MissingName) { + BSONObj obj = BSON(ShardType::host("localhost:27017")); + StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); + ASSERT_FALSE(shardRes.isOK()); +} - TEST(ShardType, MissingHost) { - BSONObj obj = BSON(ShardType::name("shard0000")); - StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); - ASSERT_FALSE(shardRes.isOK()); - } +TEST(ShardType, MissingHost) { + BSONObj obj = BSON(ShardType::name("shard0000")); + StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); + ASSERT_FALSE(shardRes.isOK()); +} - TEST(ShardType, OnlyMandatory) { - BSONObj obj = BSON(ShardType::name("shard0000") << - ShardType::host("localhost:27017")); - StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); - ASSERT(shardRes.isOK()); - ShardType shard = shardRes.getValue(); - ASSERT(shard.validate().isOK()); - } +TEST(ShardType, OnlyMandatory) { + BSONObj obj = BSON(ShardType::name("shard0000") << ShardType::host("localhost:27017")); + StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); + ASSERT(shardRes.isOK()); + ShardType shard = shardRes.getValue(); + ASSERT(shard.validate().isOK()); +} - TEST(ShardType, AllOptionalsPresent) { - BSONObj obj = BSON(ShardType::name("shard0000") << - ShardType::host("localhost:27017") << - ShardType::draining(true) << - ShardType::maxSizeMB(100)); - StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); - ASSERT(shardRes.isOK()); - ShardType shard = shardRes.getValue(); - ASSERT(shard.validate().isOK()); - } +TEST(ShardType, AllOptionalsPresent) { + BSONObj obj = BSON(ShardType::name("shard0000") << ShardType::host("localhost:27017") + << ShardType::draining(true) + << ShardType::maxSizeMB(100)); + StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); + ASSERT(shardRes.isOK()); + ShardType shard = shardRes.getValue(); + ASSERT(shard.validate().isOK()); +} - TEST(ShardType, MaxSizeAsFloat) { - BSONObj obj = BSON(ShardType::name("shard0000") << - ShardType::host("localhost:27017") << - ShardType::maxSizeMB() << 100.0); - StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); - ASSERT(shardRes.isOK()); - ShardType shard = shardRes.getValue(); - ASSERT(shard.validate().isOK()); - } +TEST(ShardType, MaxSizeAsFloat) { + BSONObj obj = BSON(ShardType::name("shard0000") << ShardType::host("localhost:27017") + << ShardType::maxSizeMB() << 100.0); + StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); + ASSERT(shardRes.isOK()); + ShardType shard = shardRes.getValue(); + ASSERT(shard.validate().isOK()); +} - TEST(ShardType, BadType) { - BSONObj obj = BSON(ShardType::name() << 0); - StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); - ASSERT_FALSE(shardRes.isOK()); - } +TEST(ShardType, BadType) { + BSONObj obj = BSON(ShardType::name() << 0); + StatusWith<ShardType> shardRes = ShardType::fromBSON(obj); + ASSERT_FALSE(shardRes.isOK()); +} } // unnamed namespace diff --git a/src/mongo/s/catalog/type_tags.cpp b/src/mongo/s/catalog/type_tags.cpp index a0fd4dff70c..9b45e4553a8 100644 --- a/src/mongo/s/catalog/type_tags.cpp +++ b/src/mongo/s/catalog/type_tags.cpp @@ -39,111 +39,115 @@ namespace mongo { - using std::string; +using std::string; - const std::string TagsType::ConfigNS = "config.tags"; +const std::string TagsType::ConfigNS = "config.tags"; - const BSONField<std::string> TagsType::ns("ns"); - const BSONField<std::string> TagsType::tag("tag"); - const BSONField<BSONObj> TagsType::min("min"); - const BSONField<BSONObj> TagsType::max("max"); +const BSONField<std::string> TagsType::ns("ns"); +const BSONField<std::string> TagsType::tag("tag"); +const BSONField<BSONObj> TagsType::min("min"); +const BSONField<BSONObj> TagsType::max("max"); - StatusWith<TagsType> TagsType::fromBSON(const BSONObj& source) { - TagsType tags; +StatusWith<TagsType> TagsType::fromBSON(const BSONObj& source) { + TagsType tags; - { - std::string tagsNs; - Status status = bsonExtractStringField(source, ns.name(), &tagsNs); - if (!status.isOK()) return status; + { + std::string tagsNs; + Status status = bsonExtractStringField(source, ns.name(), &tagsNs); + if (!status.isOK()) + return status; - tags._ns = tagsNs; - } - - { - std::string tagsTag; - Status status = bsonExtractStringField(source, tag.name(), &tagsTag); - if (!status.isOK()) return status; + tags._ns = tagsNs; + } - tags._tag = tagsTag; - } + { + std::string tagsTag; + Status status = bsonExtractStringField(source, tag.name(), &tagsTag); + if (!status.isOK()) + return status; - { - BSONElement tagsMinKey; - Status status = bsonExtractTypedField(source, min.name(), Object, &tagsMinKey); - if (!status.isOK()) return status; + tags._tag = tagsTag; + } - tags._minKey = tagsMinKey.Obj().getOwned(); - } + { + BSONElement tagsMinKey; + Status status = bsonExtractTypedField(source, min.name(), Object, &tagsMinKey); + if (!status.isOK()) + return status; - { - BSONElement tagsMaxKey; - Status status = bsonExtractTypedField(source, max.name(), Object, &tagsMaxKey); - if (!status.isOK()) return status; + tags._minKey = tagsMinKey.Obj().getOwned(); + } - tags._maxKey = tagsMaxKey.Obj().getOwned(); - } + { + BSONElement tagsMaxKey; + Status status = bsonExtractTypedField(source, max.name(), Object, &tagsMaxKey); + if (!status.isOK()) + return status; - return tags; + tags._maxKey = tagsMaxKey.Obj().getOwned(); } - Status TagsType::validate() const { - if (!_ns.is_initialized() || _ns->empty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << ns.name() << " field"); - } + return tags; +} - if (!_tag.is_initialized() || _tag->empty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << tag.name() << " field"); - } +Status TagsType::validate() const { + if (!_ns.is_initialized() || _ns->empty()) { + return Status(ErrorCodes::NoSuchKey, str::stream() << "missing " << ns.name() << " field"); + } - if (!_minKey.is_initialized() || _minKey->isEmpty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << min.name() << " field"); - } + if (!_tag.is_initialized() || _tag->empty()) { + return Status(ErrorCodes::NoSuchKey, str::stream() << "missing " << tag.name() << " field"); + } - if (!_maxKey.is_initialized() || _maxKey->isEmpty()) { - return Status(ErrorCodes::NoSuchKey, - str::stream() << "missing " << max.name() << " field"); - } + if (!_minKey.is_initialized() || _minKey->isEmpty()) { + return Status(ErrorCodes::NoSuchKey, str::stream() << "missing " << min.name() << " field"); + } - // 'min' and 'max' must share the same fields. - if (_minKey->nFields() != _maxKey->nFields()) { - return Status(ErrorCodes::BadValue, "min and max have a different number of keys"); - } + if (!_maxKey.is_initialized() || _maxKey->isEmpty()) { + return Status(ErrorCodes::NoSuchKey, str::stream() << "missing " << max.name() << " field"); + } - BSONObjIterator minIt(_minKey.get()); - BSONObjIterator maxIt(_maxKey.get()); - while (minIt.more() && maxIt.more()) { - BSONElement minElem = minIt.next(); - BSONElement maxElem = maxIt.next(); - if (strcmp(minElem.fieldName(), maxElem.fieldName())) { - return Status(ErrorCodes::BadValue, "min and max have different set of keys"); - } - } + // 'min' and 'max' must share the same fields. + if (_minKey->nFields() != _maxKey->nFields()) { + return Status(ErrorCodes::BadValue, "min and max have a different number of keys"); + } - // 'max' should be greater than 'min'. - if (_minKey->woCompare(_maxKey.get()) >= 0) { - return Status(ErrorCodes::BadValue, "max key must be greater than min key"); + BSONObjIterator minIt(_minKey.get()); + BSONObjIterator maxIt(_maxKey.get()); + while (minIt.more() && maxIt.more()) { + BSONElement minElem = minIt.next(); + BSONElement maxElem = maxIt.next(); + if (strcmp(minElem.fieldName(), maxElem.fieldName())) { + return Status(ErrorCodes::BadValue, "min and max have different set of keys"); } + } - return Status::OK(); + // 'max' should be greater than 'min'. + if (_minKey->woCompare(_maxKey.get()) >= 0) { + return Status(ErrorCodes::BadValue, "max key must be greater than min key"); } - BSONObj TagsType::toBSON() const { - BSONObjBuilder builder; + return Status::OK(); +} - if (_ns) builder.append(ns.name(), getNS()); - if (_tag) builder.append(tag.name(), getTag()); - if (_minKey) builder.append(min.name(), getMinKey()); - if (_maxKey) builder.append(max.name(), getMaxKey()); +BSONObj TagsType::toBSON() const { + BSONObjBuilder builder; - return builder.obj(); - } + if (_ns) + builder.append(ns.name(), getNS()); + if (_tag) + builder.append(tag.name(), getTag()); + if (_minKey) + builder.append(min.name(), getMinKey()); + if (_maxKey) + builder.append(max.name(), getMaxKey()); - std::string TagsType::toString() const { - return toBSON().toString(); - } + return builder.obj(); +} + +std::string TagsType::toString() const { + return toBSON().toString(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/type_tags.h b/src/mongo/s/catalog/type_tags.h index a21e4cdf327..d6890af5a7f 100644 --- a/src/mongo/s/catalog/type_tags.h +++ b/src/mongo/s/catalog/type_tags.h @@ -35,74 +35,83 @@ namespace mongo { - class BSONObj; - class Status; - template<typename T> class StatusWith; +class BSONObj; +class Status; +template <typename T> +class StatusWith; - /** - * This class represents the layout and contents of documents contained in the config.tags - * collection. All manipulation of documents coming from that collection should be done with - * this class. - */ - class TagsType { - public: - // Name of the tags collection in the config server. - static const std::string ConfigNS; +/** + * This class represents the layout and contents of documents contained in the config.tags + * collection. All manipulation of documents coming from that collection should be done with + * this class. + */ +class TagsType { +public: + // Name of the tags collection in the config server. + static const std::string ConfigNS; - // Field names and types in the tags collection type. - static const BSONField<std::string> ns; - static const BSONField<std::string> tag; - static const BSONField<BSONObj> min; - static const BSONField<BSONObj> max; + // Field names and types in the tags collection type. + static const BSONField<std::string> ns; + static const BSONField<std::string> tag; + static const BSONField<BSONObj> min; + static const BSONField<BSONObj> max; - /** - * Constructs a new DatabaseType object from BSON. Validates that all required fields are - * present. - */ - static StatusWith<TagsType> fromBSON(const BSONObj& source); + /** + * Constructs a new DatabaseType object from BSON. Validates that all required fields are + * present. + */ + static StatusWith<TagsType> fromBSON(const BSONObj& source); - /** - * Returns OK if all fields have been set. Otherwise returns NoSuchKey and information - * about what is the first field which is missing. - */ - Status validate() const; + /** + * Returns OK if all fields have been set. Otherwise returns NoSuchKey and information + * about what is the first field which is missing. + */ + Status validate() const; - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; - const std::string& getNS() const { return _ns.get(); } - void setNS(const std::string& ns); + const std::string& getNS() const { + return _ns.get(); + } + void setNS(const std::string& ns); - const std::string& getTag() const { return _tag.get(); } - void setTag(const std::string& tag); + const std::string& getTag() const { + return _tag.get(); + } + void setTag(const std::string& tag); - const BSONObj& getMinKey() const { return _minKey.get(); } - void setMinKey(const BSONObj& minKey); + const BSONObj& getMinKey() const { + return _minKey.get(); + } + void setMinKey(const BSONObj& minKey); - const BSONObj& getMaxKey() const { return _maxKey.get(); } - void setMaxKey(const BSONObj& max); + const BSONObj& getMaxKey() const { + return _maxKey.get(); + } + void setMaxKey(const BSONObj& max); - private: - // Required namespace to which this tag belongs - boost::optional<std::string> _ns; +private: + // Required namespace to which this tag belongs + boost::optional<std::string> _ns; - // Required tag name - boost::optional<std::string> _tag; + // Required tag name + boost::optional<std::string> _tag; - // Required first key of the tag (inclusive) - boost::optional<BSONObj> _minKey; + // Required first key of the tag (inclusive) + boost::optional<BSONObj> _minKey; - // Required last key of the tag (not-inclusive) - boost::optional<BSONObj> _maxKey; - }; + // Required last key of the tag (not-inclusive) + boost::optional<BSONObj> _maxKey; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/catalog/type_tags_test.cpp b/src/mongo/s/catalog/type_tags_test.cpp index 58a0d4a7d95..3f1a7a1175d 100644 --- a/src/mongo/s/catalog/type_tags_test.cpp +++ b/src/mongo/s/catalog/type_tags_test.cpp @@ -36,105 +36,97 @@ namespace { - using namespace mongo; - - using std::string; - - TEST(TagsType, Valid) { - BSONObj obj = BSON(TagsType::ns("test.mycol") << - TagsType::tag("tag") << - TagsType::min(BSON("a" << 10)) << - TagsType::max(BSON("a" << 20))); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - ASSERT_TRUE(status.isOK()); - - TagsType tag = status.getValue(); - - ASSERT_EQUALS(tag.getNS(), "test.mycol"); - ASSERT_EQUALS(tag.getTag(), "tag"); - ASSERT_EQUALS(tag.getMinKey(), BSON("a" << 10)); - ASSERT_EQUALS(tag.getMaxKey(), BSON("a" << 20)); - } - - TEST(TagsType, MissingNsField) { - BSONObj obj = BSON(TagsType::tag("tag") << - TagsType::min(BSON("a" << 10)) << - TagsType::max(BSON("a" << 20))); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - ASSERT_FALSE(status.isOK()); - ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); - } - - TEST(TagsType, MissingTagField) { - BSONObj obj = BSON(TagsType::ns("test.mycol") << - TagsType::min(BSON("a" << 10)) << - TagsType::max(BSON("a" << 20))); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - ASSERT_FALSE(status.isOK()); - ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); - } - - TEST(TagsType, MissingMinKey) { - BSONObj obj = BSON(TagsType::ns("test.mycol") << - TagsType::tag("tag") << - TagsType::max(BSON("a" << 20))); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - ASSERT_FALSE(status.isOK()); - ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); - } - - TEST(TagsType, MissingMaxKey) { - BSONObj obj = BSON(TagsType::ns("test.mycol") << - TagsType::tag("tag") << - TagsType::min(BSON("a" << 10))); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - ASSERT_FALSE(status.isOK()); - ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); - } - - TEST(TagsType, KeysWithDifferentNumberOfColumns) { - BSONObj obj = BSON(TagsType::ns("test.mycol") << - TagsType::tag("tag") << - TagsType::min(BSON("a" << 10 << "b" << 10)) << - TagsType::max(BSON("a" << 20))); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - const TagsType& tag = status.getValue(); - ASSERT_EQUALS(ErrorCodes::BadValue, tag.validate()); - } - - TEST(TagsType, KeysWithDifferentColumns) { - BSONObj obj = BSON(TagsType::ns("test.mycol") << - TagsType::tag("tag") << - TagsType::min(BSON("a" << 10)) << - TagsType::max(BSON("b" << 20))); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - const TagsType& tag = status.getValue(); - ASSERT_EQUALS(ErrorCodes::BadValue, tag.validate()); - } - - TEST(TagsType, KeysNotAscending) { - BSONObj obj = BSON(TagsType::tag("tag") << - TagsType::ns("test.mycol") << - TagsType::min(BSON("a" << 20)) << - TagsType::max(BSON("a" << 10))); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - const TagsType& tag = status.getValue(); - ASSERT_EQUALS(ErrorCodes::BadValue, tag.validate()); - } - - TEST(TagsType, BadType) { - BSONObj obj = BSON(TagsType::tag() << 0); - - StatusWith<TagsType> status = TagsType::fromBSON(obj); - ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); - } - -} // namespace +using namespace mongo; + +using std::string; + +TEST(TagsType, Valid) { + BSONObj obj = + BSON(TagsType::ns("test.mycol") << TagsType::tag("tag") << TagsType::min(BSON("a" << 10)) + << TagsType::max(BSON("a" << 20))); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + ASSERT_TRUE(status.isOK()); + + TagsType tag = status.getValue(); + + ASSERT_EQUALS(tag.getNS(), "test.mycol"); + ASSERT_EQUALS(tag.getTag(), "tag"); + ASSERT_EQUALS(tag.getMinKey(), BSON("a" << 10)); + ASSERT_EQUALS(tag.getMaxKey(), BSON("a" << 20)); +} + +TEST(TagsType, MissingNsField) { + BSONObj obj = BSON(TagsType::tag("tag") << TagsType::min(BSON("a" << 10)) + << TagsType::max(BSON("a" << 20))); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + ASSERT_FALSE(status.isOK()); + ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); +} + +TEST(TagsType, MissingTagField) { + BSONObj obj = BSON(TagsType::ns("test.mycol") << TagsType::min(BSON("a" << 10)) + << TagsType::max(BSON("a" << 20))); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + ASSERT_FALSE(status.isOK()); + ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); +} + +TEST(TagsType, MissingMinKey) { + BSONObj obj = + BSON(TagsType::ns("test.mycol") << TagsType::tag("tag") << TagsType::max(BSON("a" << 20))); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + ASSERT_FALSE(status.isOK()); + ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); +} + +TEST(TagsType, MissingMaxKey) { + BSONObj obj = + BSON(TagsType::ns("test.mycol") << TagsType::tag("tag") << TagsType::min(BSON("a" << 10))); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + ASSERT_FALSE(status.isOK()); + ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); +} + +TEST(TagsType, KeysWithDifferentNumberOfColumns) { + BSONObj obj = BSON(TagsType::ns("test.mycol") << TagsType::tag("tag") + << TagsType::min(BSON("a" << 10 << "b" << 10)) + << TagsType::max(BSON("a" << 20))); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + const TagsType& tag = status.getValue(); + ASSERT_EQUALS(ErrorCodes::BadValue, tag.validate()); +} + +TEST(TagsType, KeysWithDifferentColumns) { + BSONObj obj = + BSON(TagsType::ns("test.mycol") << TagsType::tag("tag") << TagsType::min(BSON("a" << 10)) + << TagsType::max(BSON("b" << 20))); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + const TagsType& tag = status.getValue(); + ASSERT_EQUALS(ErrorCodes::BadValue, tag.validate()); +} + +TEST(TagsType, KeysNotAscending) { + BSONObj obj = + BSON(TagsType::tag("tag") << TagsType::ns("test.mycol") << TagsType::min(BSON("a" << 20)) + << TagsType::max(BSON("a" << 10))); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + const TagsType& tag = status.getValue(); + ASSERT_EQUALS(ErrorCodes::BadValue, tag.validate()); +} + +TEST(TagsType, BadType) { + BSONObj obj = BSON(TagsType::tag() << 0); + + StatusWith<TagsType> status = TagsType::fromBSON(obj); + ASSERT_EQUALS(ErrorCodes::NoSuchKey, status.getStatus()); +} + +} // namespace diff --git a/src/mongo/s/chunk.cpp b/src/mongo/s/chunk.cpp index 5e475771c07..ede622930f5 100644 --- a/src/mongo/s/chunk.cpp +++ b/src/mongo/s/chunk.cpp @@ -54,721 +54,689 @@ namespace mongo { - using std::shared_ptr; - using std::unique_ptr; - using std::map; - using std::ostringstream; - using std::set; - using std::string; - using std::stringstream; - using std::vector; +using std::shared_ptr; +using std::unique_ptr; +using std::map; +using std::ostringstream; +using std::set; +using std::string; +using std::stringstream; +using std::vector; namespace { - const int kTooManySplitPoints = 4; +const int kTooManySplitPoints = 4; - /** - * Attempts to move the given chunk to another shard. - * - * Returns true if the chunk was actually moved. - */ - bool tryMoveToOtherShard(const ChunkManager& manager, const ChunkType& chunk) { - // reload sharding metadata before starting migration - ChunkManagerPtr chunkMgr = manager.reload(false /* just reloaded in mulitsplit */); +/** + * Attempts to move the given chunk to another shard. + * + * Returns true if the chunk was actually moved. + */ +bool tryMoveToOtherShard(const ChunkManager& manager, const ChunkType& chunk) { + // reload sharding metadata before starting migration + ChunkManagerPtr chunkMgr = manager.reload(false /* just reloaded in mulitsplit */); - ShardInfoMap shardInfo; - Status loadStatus = DistributionStatus::populateShardInfoMap(&shardInfo); + ShardInfoMap shardInfo; + Status loadStatus = DistributionStatus::populateShardInfoMap(&shardInfo); - if (!loadStatus.isOK()) { - warning() << "failed to load shard metadata while trying to moveChunk after " - << "auto-splitting" << causedBy(loadStatus); - return false; - } + if (!loadStatus.isOK()) { + warning() << "failed to load shard metadata while trying to moveChunk after " + << "auto-splitting" << causedBy(loadStatus); + return false; + } - if (shardInfo.size() < 2) { - LOG(0) << "no need to move top chunk since there's only 1 shard"; - return false; - } + if (shardInfo.size() < 2) { + LOG(0) << "no need to move top chunk since there's only 1 shard"; + return false; + } - map<string, vector<ChunkType>> shardToChunkMap; - DistributionStatus::populateShardToChunksMap(shardInfo, *chunkMgr, &shardToChunkMap); + map<string, vector<ChunkType>> shardToChunkMap; + DistributionStatus::populateShardToChunksMap(shardInfo, *chunkMgr, &shardToChunkMap); - StatusWith<string> tagStatus = grid.catalogManager()->getTagForChunk(manager.getns(), - chunk); - if (!tagStatus.isOK()) { - warning() << "Not auto-moving chunk because of an error encountered while " - << "checking tag for chunk: " << tagStatus.getStatus(); - return false; - } + StatusWith<string> tagStatus = grid.catalogManager()->getTagForChunk(manager.getns(), chunk); + if (!tagStatus.isOK()) { + warning() << "Not auto-moving chunk because of an error encountered while " + << "checking tag for chunk: " << tagStatus.getStatus(); + return false; + } - DistributionStatus chunkDistribution(shardInfo, shardToChunkMap); - const string newLocation( - chunkDistribution.getBestReceieverShard(tagStatus.getValue())); + DistributionStatus chunkDistribution(shardInfo, shardToChunkMap); + const string newLocation(chunkDistribution.getBestReceieverShard(tagStatus.getValue())); - if (newLocation.empty()) { - LOG(1) << "recently split chunk: " << chunk - << " but no suitable shard to move to"; - return false; - } + if (newLocation.empty()) { + LOG(1) << "recently split chunk: " << chunk << " but no suitable shard to move to"; + return false; + } - if (chunk.getShard() == newLocation) { - // if this is the best shard, then we shouldn't do anything. - LOG(1) << "recently split chunk: " << chunk - << " already in the best shard"; - return false; - } + if (chunk.getShard() == newLocation) { + // if this is the best shard, then we shouldn't do anything. + LOG(1) << "recently split chunk: " << chunk << " already in the best shard"; + return false; + } - ChunkPtr toMove = chunkMgr->findIntersectingChunk(chunk.getMin()); + ChunkPtr toMove = chunkMgr->findIntersectingChunk(chunk.getMin()); - if (!(toMove->getMin() == chunk.getMin() && toMove->getMax() == chunk.getMax())) { - LOG(1) << "recently split chunk: " << chunk - << " modified before we could migrate " << toMove->toString(); - return false; - } + if (!(toMove->getMin() == chunk.getMin() && toMove->getMax() == chunk.getMax())) { + LOG(1) << "recently split chunk: " << chunk << " modified before we could migrate " + << toMove->toString(); + return false; + } - log() << "moving chunk (auto): " << toMove->toString() << " to: " << newLocation; + log() << "moving chunk (auto): " << toMove->toString() << " to: " << newLocation; - shared_ptr<Shard> newShard = grid.shardRegistry()->getShard(newLocation); - if (!newShard) { - warning() << "Newly selected shard " << newLocation << " could not be found."; - return false; - } + shared_ptr<Shard> newShard = grid.shardRegistry()->getShard(newLocation); + if (!newShard) { + warning() << "Newly selected shard " << newLocation << " could not be found."; + return false; + } - BSONObj res; - WriteConcernOptions noThrottle; - if (!toMove->moveAndCommit(newShard->getId(), - Chunk::MaxChunkSize, - &noThrottle, /* secondaryThrottle */ - false, /* waitForDelete - small chunk, no need */ - 0, /* maxTimeMS - don't time out */ - res)) { - msgassertedNoTrace(10412, str::stream() << "moveAndCommit failed: " << res); - } + BSONObj res; + WriteConcernOptions noThrottle; + if (!toMove->moveAndCommit(newShard->getId(), + Chunk::MaxChunkSize, + &noThrottle, /* secondaryThrottle */ + false, /* waitForDelete - small chunk, no need */ + 0, /* maxTimeMS - don't time out */ + res)) { + msgassertedNoTrace(10412, str::stream() << "moveAndCommit failed: " << res); + } - // update our config - manager.reload(); + // update our config + manager.reload(); - return true; - } + return true; +} -} // namespace +} // namespace - long long Chunk::MaxChunkSize = 1024 * 1024 * 64; - int Chunk::MaxObjectPerChunk = 250000; +long long Chunk::MaxChunkSize = 1024 * 1024 * 64; +int Chunk::MaxObjectPerChunk = 250000; - // Can be overridden from command line - bool Chunk::ShouldAutoSplit = true; +// Can be overridden from command line +bool Chunk::ShouldAutoSplit = true; - Chunk::Chunk(const ChunkManager * manager, BSONObj from) - : _manager(manager), _lastmod(0, 0, OID()), _dataWritten(mkDataWritten()) - { - string ns = from.getStringField(ChunkType::ns().c_str()); - _shardId = from.getStringField(ChunkType::shard().c_str()); +Chunk::Chunk(const ChunkManager* manager, BSONObj from) + : _manager(manager), _lastmod(0, 0, OID()), _dataWritten(mkDataWritten()) { + string ns = from.getStringField(ChunkType::ns().c_str()); + _shardId = from.getStringField(ChunkType::shard().c_str()); - _lastmod = ChunkVersion::fromBSON(from[ChunkType::DEPRECATED_lastmod()]); - verify( _lastmod.isSet() ); + _lastmod = ChunkVersion::fromBSON(from[ChunkType::DEPRECATED_lastmod()]); + verify(_lastmod.isSet()); - _min = from.getObjectField(ChunkType::min().c_str()).getOwned(); - _max = from.getObjectField(ChunkType::max().c_str()).getOwned(); - - _jumbo = from[ChunkType::jumbo()].trueValue(); + _min = from.getObjectField(ChunkType::min().c_str()).getOwned(); + _max = from.getObjectField(ChunkType::max().c_str()).getOwned(); - uassert( 10170 , "Chunk needs a ns" , ! ns.empty() ); - uassert( 13327 , "Chunk ns must match server ns" , ns == _manager->getns() ); + _jumbo = from[ChunkType::jumbo()].trueValue(); - { - const auto shard = grid.shardRegistry()->getShard(_shardId); - uassert(10171, "Chunk needs a server", shard); - } + uassert(10170, "Chunk needs a ns", !ns.empty()); + uassert(13327, "Chunk ns must match server ns", ns == _manager->getns()); - uassert( 10172 , "Chunk needs a min" , ! _min.isEmpty() ); - uassert( 10173 , "Chunk needs a max" , ! _max.isEmpty() ); + { + const auto shard = grid.shardRegistry()->getShard(_shardId); + uassert(10171, "Chunk needs a server", shard); } - Chunk::Chunk(const ChunkManager * info, - const BSONObj& min, - const BSONObj& max, - const ShardId& shardId, - ChunkVersion lastmod) - : _manager(info), - _min(min), - _max(max), - _shardId(shardId), - _lastmod(lastmod), - _jumbo(false), - _dataWritten(mkDataWritten()) - {} + uassert(10172, "Chunk needs a min", !_min.isEmpty()); + uassert(10173, "Chunk needs a max", !_max.isEmpty()); +} + +Chunk::Chunk(const ChunkManager* info, + const BSONObj& min, + const BSONObj& max, + const ShardId& shardId, + ChunkVersion lastmod) + : _manager(info), + _min(min), + _max(max), + _shardId(shardId), + _lastmod(lastmod), + _jumbo(false), + _dataWritten(mkDataWritten()) {} + +int Chunk::mkDataWritten() { + PseudoRandom r(static_cast<int64_t>(time(0))); + return r.nextInt32(MaxChunkSize / ChunkManager::SplitHeuristics::splitTestFactor); +} + +bool Chunk::containsKey(const BSONObj& shardKey) const { + return getMin().woCompare(shardKey) <= 0 && shardKey.woCompare(getMax()) < 0; +} + +bool ChunkRange::containsKey(const BSONObj& shardKey) const { + // same as Chunk method + return getMin().woCompare(shardKey) <= 0 && shardKey.woCompare(getMax()) < 0; +} + +bool Chunk::_minIsInf() const { + return 0 == _manager->getShardKeyPattern().getKeyPattern().globalMin().woCompare(getMin()); +} + +bool Chunk::_maxIsInf() const { + return 0 == _manager->getShardKeyPattern().getKeyPattern().globalMax().woCompare(getMax()); +} + +BSONObj Chunk::_getExtremeKey(bool doSplitAtLower) const { + Query q; + if (doSplitAtLower) { + q.sort(_manager->getShardKeyPattern().toBSON()); + } else { + // need to invert shard key pattern to sort backwards + // TODO: make a helper in ShardKeyPattern? + + BSONObj k = _manager->getShardKeyPattern().toBSON(); + BSONObjBuilder r; + + BSONObjIterator i(k); + while (i.more()) { + BSONElement e = i.next(); + uassert(10163, "can only handle numbers here - which i think is correct", e.isNumber()); + r.append(e.fieldName(), -1 * e.number()); + } - int Chunk::mkDataWritten() { - PseudoRandom r(static_cast<int64_t>(time(0))); - return r.nextInt32( MaxChunkSize / ChunkManager::SplitHeuristics::splitTestFactor ); + q.sort(r.obj()); } - bool Chunk::containsKey( const BSONObj& shardKey ) const { - return getMin().woCompare( shardKey ) <= 0 && shardKey.woCompare( getMax() ) < 0; - } + // find the extreme key + ScopedDbConnection conn(_getShardConnectionString()); + BSONObj end; + + if (doSplitAtLower) { + // Splitting close to the lower bound means that the split point will be the + // upper bound. Chunk range upper bounds are exclusive so skip a document to + // make the lower half of the split end up with a single document. + unique_ptr<DBClientCursor> cursor = conn->query(_manager->getns(), + q, + 1, /* nToReturn */ + 1 /* nToSkip */); - bool ChunkRange::containsKey( const BSONObj& shardKey ) const { - // same as Chunk method - return getMin().woCompare( shardKey ) <= 0 && shardKey.woCompare( getMax() ) < 0; + if (cursor->more()) { + end = cursor->next().getOwned(); + } + } else { + end = conn->findOne(_manager->getns(), q); } - bool Chunk::_minIsInf() const { - return 0 == - _manager->getShardKeyPattern().getKeyPattern().globalMin().woCompare(getMin()); + conn.done(); + if (end.isEmpty()) + return BSONObj(); + return _manager->getShardKeyPattern().extractShardKeyFromDoc(end); +} + +void Chunk::pickMedianKey(BSONObj& medianKey) const { + // Ask the mongod holding this chunk to figure out the split points. + ScopedDbConnection conn(_getShardConnectionString()); + BSONObj result; + BSONObjBuilder cmd; + cmd.append("splitVector", _manager->getns()); + cmd.append("keyPattern", _manager->getShardKeyPattern().toBSON()); + cmd.append("min", getMin()); + cmd.append("max", getMax()); + cmd.appendBool("force", true); + BSONObj cmdObj = cmd.obj(); + + if (!conn->runCommand("admin", cmdObj, result)) { + conn.done(); + ostringstream os; + os << "splitVector command (median key) failed: " << result; + uassert(13503, os.str(), 0); } - bool Chunk::_maxIsInf() const { - return 0 == - _manager->getShardKeyPattern().getKeyPattern().globalMax().woCompare(getMax()); + BSONObjIterator it(result.getObjectField("splitKeys")); + if (it.more()) { + medianKey = it.next().Obj().getOwned(); } - BSONObj Chunk::_getExtremeKey(bool doSplitAtLower) const { - Query q; - if (doSplitAtLower) { - q.sort( _manager->getShardKeyPattern().toBSON() ); - } - else { - // need to invert shard key pattern to sort backwards - // TODO: make a helper in ShardKeyPattern? - - BSONObj k = _manager->getShardKeyPattern().toBSON(); - BSONObjBuilder r; - - BSONObjIterator i(k); - while( i.more() ) { - BSONElement e = i.next(); - uassert( 10163 , "can only handle numbers here - which i think is correct" , e.isNumber() ); - r.append( e.fieldName() , -1 * e.number() ); - } + conn.done(); +} + +void Chunk::pickSplitVector(vector<BSONObj>& splitPoints, + long long chunkSize /* bytes */, + int maxPoints, + int maxObjs) const { + // Ask the mongod holding this chunk to figure out the split points. + ScopedDbConnection conn(_getShardConnectionString()); + BSONObj result; + BSONObjBuilder cmd; + cmd.append("splitVector", _manager->getns()); + cmd.append("keyPattern", _manager->getShardKeyPattern().toBSON()); + cmd.append("min", getMin()); + cmd.append("max", getMax()); + cmd.append("maxChunkSizeBytes", chunkSize); + cmd.append("maxSplitPoints", maxPoints); + cmd.append("maxChunkObjects", maxObjs); + BSONObj cmdObj = cmd.obj(); + + if (!conn->runCommand("admin", cmdObj, result)) { + conn.done(); + ostringstream os; + os << "splitVector command failed: " << result; + uassert(13345, os.str(), 0); + } - q.sort( r.obj() ); + BSONObjIterator it(result.getObjectField("splitKeys")); + while (it.more()) { + splitPoints.push_back(it.next().Obj().getOwned()); + } + conn.done(); +} + +void Chunk::determineSplitPoints(bool atMedian, vector<BSONObj>* splitPoints) const { + // if splitting is not obligatory we may return early if there are not enough data + // we cap the number of objects that would fall in the first half (before the split point) + // the rationale is we'll find a split point without traversing all the data + if (atMedian) { + BSONObj medianKey; + pickMedianKey(medianKey); + if (!medianKey.isEmpty()) + splitPoints->push_back(medianKey); + } else { + long long chunkSize = _manager->getCurrentDesiredChunkSize(); + + // Note: One split point for every 1/2 chunk size. + const int estNumSplitPoints = _dataWritten / chunkSize * 2; + if (estNumSplitPoints >= kTooManySplitPoints) { + // The current desired chunk size will split the chunk into lots of small chunks + // (At the worst case, this can result into thousands of chunks); so check and + // see if a bigger value can be used. + + chunkSize = std::min(_dataWritten, Chunk::MaxChunkSize); + } + + pickSplitVector(*splitPoints, chunkSize, 0, MaxObjectPerChunk); + + if (splitPoints->size() <= 1) { + // no split points means there isn't enough data to split on + // 1 split point means we have between half the chunk size to full chunk size + // so we shouldn't split + splitPoints->clear(); } + } +} - // find the extreme key - ScopedDbConnection conn(_getShardConnectionString()); - BSONObj end; - - if (doSplitAtLower) { - // Splitting close to the lower bound means that the split point will be the - // upper bound. Chunk range upper bounds are exclusive so skip a document to - // make the lower half of the split end up with a single document. - unique_ptr<DBClientCursor> cursor = conn->query(_manager->getns(), - q, - 1, /* nToReturn */ - 1 /* nToSkip */); - - if (cursor->more()) { - end = cursor->next().getOwned(); - } - } - else { - end = conn->findOne(_manager->getns(), q); - } +Status Chunk::split(SplitPointMode mode, size_t* resultingSplits, BSONObj* res) const { + size_t dummy; + if (resultingSplits == NULL) { + resultingSplits = &dummy; + } - conn.done(); - if ( end.isEmpty() ) - return BSONObj(); - return _manager->getShardKeyPattern().extractShardKeyFromDoc(end); - } - - void Chunk::pickMedianKey( BSONObj& medianKey ) const { - // Ask the mongod holding this chunk to figure out the split points. - ScopedDbConnection conn(_getShardConnectionString()); - BSONObj result; - BSONObjBuilder cmd; - cmd.append( "splitVector" , _manager->getns() ); - cmd.append( "keyPattern" , _manager->getShardKeyPattern().toBSON() ); - cmd.append( "min" , getMin() ); - cmd.append( "max" , getMax() ); - cmd.appendBool( "force" , true ); - BSONObj cmdObj = cmd.obj(); - - if ( ! conn->runCommand( "admin" , cmdObj , result )) { - conn.done(); - ostringstream os; - os << "splitVector command (median key) failed: " << result; - uassert( 13503 , os.str() , 0 ); - } + bool atMedian = mode == Chunk::atMedian; + vector<BSONObj> splitPoints; - BSONObjIterator it( result.getObjectField( "splitKeys" ) ); - if ( it.more() ) { - medianKey = it.next().Obj().getOwned(); + determineSplitPoints(atMedian, &splitPoints); + if (splitPoints.empty()) { + string msg; + if (atMedian) { + msg = "cannot find median in chunk, possibly empty"; + } else { + msg = "chunk not full enough to trigger auto-split"; } - conn.done(); + LOG(1) << msg; + return Status(ErrorCodes::CannotSplit, msg); } - void Chunk::pickSplitVector(vector<BSONObj>& splitPoints, - long long chunkSize /* bytes */, - int maxPoints, - int maxObjs) const { - // Ask the mongod holding this chunk to figure out the split points. - ScopedDbConnection conn(_getShardConnectionString()); - BSONObj result; - BSONObjBuilder cmd; - cmd.append( "splitVector" , _manager->getns() ); - cmd.append( "keyPattern" , _manager->getShardKeyPattern().toBSON() ); - cmd.append( "min" , getMin() ); - cmd.append( "max" , getMax() ); - cmd.append( "maxChunkSizeBytes" , chunkSize ); - cmd.append( "maxSplitPoints" , maxPoints ); - cmd.append( "maxChunkObjects" , maxObjs ); - BSONObj cmdObj = cmd.obj(); - - if ( ! conn->runCommand( "admin" , cmdObj , result )) { - conn.done(); - ostringstream os; - os << "splitVector command failed: " << result; - uassert( 13345 , os.str() , 0 ); - } - - BSONObjIterator it( result.getObjectField( "splitKeys" ) ); - while ( it.more() ) { - splitPoints.push_back( it.next().Obj().getOwned() ); + // We assume that if the chunk being split is the first (or last) one on the collection, + // this chunk is likely to see more insertions. Instead of splitting mid-chunk, we use + // the very first (or last) key as a split point. + // This heuristic is skipped for "special" shard key patterns that are not likely to + // produce monotonically increasing or decreasing values (e.g. hashed shard keys). + if (mode == Chunk::autoSplitInternal && + KeyPattern::isOrderedKeyPattern(_manager->getShardKeyPattern().toBSON())) { + if (_minIsInf()) { + BSONObj key = _getExtremeKey(true); + if (!key.isEmpty()) { + splitPoints[0] = key.getOwned(); + } + } else if (_maxIsInf()) { + BSONObj key = _getExtremeKey(false); + if (!key.isEmpty()) { + splitPoints.pop_back(); + splitPoints.push_back(key); + } } - conn.done(); } - void Chunk::determineSplitPoints(bool atMedian, vector<BSONObj>* splitPoints) const { - // if splitting is not obligatory we may return early if there are not enough data - // we cap the number of objects that would fall in the first half (before the split point) - // the rationale is we'll find a split point without traversing all the data - if ( atMedian ) { - BSONObj medianKey; - pickMedianKey( medianKey ); - if ( ! medianKey.isEmpty() ) - splitPoints->push_back( medianKey ); - } - else { - long long chunkSize = _manager->getCurrentDesiredChunkSize(); + // Normally, we'd have a sound split point here if the chunk is not empty. + // It's also a good place to sanity check. + if (_min == splitPoints.front()) { + string msg(str::stream() << "not splitting chunk " << toString() << ", split point " + << splitPoints.front() << " is exactly on chunk bounds"); + log() << msg; + return Status(ErrorCodes::CannotSplit, msg); + } - // Note: One split point for every 1/2 chunk size. - const int estNumSplitPoints = _dataWritten / chunkSize * 2; - if (estNumSplitPoints >= kTooManySplitPoints) { - // The current desired chunk size will split the chunk into lots of small chunks - // (At the worst case, this can result into thousands of chunks); so check and - // see if a bigger value can be used. + if (_max == splitPoints.back()) { + string msg(str::stream() << "not splitting chunk " << toString() << ", split point " + << splitPoints.back() << " is exactly on chunk bounds"); + log() << msg; + return Status(ErrorCodes::CannotSplit, msg); + } - chunkSize = std::min(_dataWritten, Chunk::MaxChunkSize); - } + Status status = multiSplit(splitPoints, res); + *resultingSplits = splitPoints.size(); + return status; +} + +Status Chunk::multiSplit(const vector<BSONObj>& m, BSONObj* res) const { + const size_t maxSplitPoints = 8192; + + uassert(10165, "can't split as shard doesn't have a manager", _manager); + uassert(13332, "need a split key to split chunk", !m.empty()); + uassert(13333, "can't split a chunk in that many parts", m.size() < maxSplitPoints); + uassert(13003, "can't split a chunk with only one distinct value", _min.woCompare(_max)); + + ScopedDbConnection conn(_getShardConnectionString()); + + BSONObjBuilder cmd; + cmd.append("splitChunk", _manager->getns()); + cmd.append("keyPattern", _manager->getShardKeyPattern().toBSON()); + cmd.append("min", getMin()); + cmd.append("max", getMax()); + cmd.append("from", getShardId()); + cmd.append("splitKeys", m); + cmd.append("configdb", grid.catalogManager()->connectionString().toString()); + cmd.append("epoch", _manager->getVersion().epoch()); + BSONObj cmdObj = cmd.obj(); + + BSONObj dummy; + if (res == NULL) { + res = &dummy; + } - pickSplitVector(*splitPoints, chunkSize, 0, MaxObjectPerChunk); + if (!conn->runCommand("admin", cmdObj, *res)) { + string msg(str::stream() << "splitChunk failed - cmd: " << cmdObj << " result: " << *res); + warning() << msg; + conn.done(); - if ( splitPoints->size() <= 1 ) { - // no split points means there isn't enough data to split on - // 1 split point means we have between half the chunk size to full chunk size - // so we shouldn't split - splitPoints->clear(); - } - } + return Status(ErrorCodes::SplitFailed, msg); } - Status Chunk::split(SplitPointMode mode, size_t* resultingSplits, BSONObj* res) const { - size_t dummy; - if (resultingSplits == NULL) { - resultingSplits = &dummy; - } - - bool atMedian = mode == Chunk::atMedian; - vector<BSONObj> splitPoints; + conn.done(); - determineSplitPoints( atMedian, &splitPoints ); - if (splitPoints.empty()) { - string msg; - if (atMedian) { - msg = "cannot find median in chunk, possibly empty"; - } - else { - msg = "chunk not full enough to trigger auto-split"; - } + // force reload of config + _manager->reload(); - LOG(1) << msg; - return Status(ErrorCodes::CannotSplit, msg); - } + return Status::OK(); +} - // We assume that if the chunk being split is the first (or last) one on the collection, - // this chunk is likely to see more insertions. Instead of splitting mid-chunk, we use - // the very first (or last) key as a split point. - // This heuristic is skipped for "special" shard key patterns that are not likely to - // produce monotonically increasing or decreasing values (e.g. hashed shard keys). - if (mode == Chunk::autoSplitInternal && - KeyPattern::isOrderedKeyPattern(_manager->getShardKeyPattern().toBSON())) { - - if (_minIsInf()) { - BSONObj key = _getExtremeKey(true); - if (!key.isEmpty()) { - splitPoints[0] = key.getOwned(); - } - } - else if (_maxIsInf()) { - BSONObj key = _getExtremeKey(false); - if (!key.isEmpty()) { - splitPoints.pop_back(); - splitPoints.push_back(key); - } - } - } +bool Chunk::moveAndCommit(const ShardId& toShardId, + long long chunkSize /* bytes */, + const WriteConcernOptions* writeConcern, + bool waitForDelete, + int maxTimeMS, + BSONObj& res) const { + uassert(10167, "can't move shard to its current location!", getShardId() != toShardId); - // Normally, we'd have a sound split point here if the chunk is not empty. - // It's also a good place to sanity check. - if ( _min == splitPoints.front() ) { - string msg(str::stream() << "not splitting chunk " << toString() - << ", split point " << splitPoints.front() - << " is exactly on chunk bounds"); - log() << msg; - return Status(ErrorCodes::CannotSplit, msg); - } + log() << "moving chunk ns: " << _manager->getns() << " moving ( " << toString() << ") " + << getShardId() << " -> " << toShardId; - if ( _max == splitPoints.back() ) { - string msg(str::stream() << "not splitting chunk " << toString() - << ", split point " << splitPoints.back() - << " is exactly on chunk bounds"); - log() << msg; - return Status(ErrorCodes::CannotSplit, msg); - } + const auto from = grid.shardRegistry()->getShard(getShardId()); - Status status = multiSplit(splitPoints, res); - *resultingSplits = splitPoints.size(); - return status; + BSONObjBuilder builder; + builder.append("moveChunk", _manager->getns()); + builder.append("from", from->getConnString().toString()); + { + const auto toShard = grid.shardRegistry()->getShard(toShardId); + builder.append("to", toShard->getConnString().toString()); + } + // NEEDED FOR 2.0 COMPATIBILITY + builder.append("fromShard", from->getId()); + builder.append("toShard", toShardId); + /////////////////////////////// + builder.append("min", _min); + builder.append("max", _max); + builder.append("maxChunkSizeBytes", chunkSize); + builder.append("configdb", grid.catalogManager()->connectionString().toString()); + + // For legacy secondary throttle setting. + bool secondaryThrottle = true; + if (writeConcern && writeConcern->wNumNodes <= 1 && writeConcern->wMode.empty()) { + secondaryThrottle = false; } - Status Chunk::multiSplit(const vector<BSONObj>& m, BSONObj* res) const { - const size_t maxSplitPoints = 8192; - - uassert( 10165 , "can't split as shard doesn't have a manager" , _manager ); - uassert( 13332 , "need a split key to split chunk" , !m.empty() ); - uassert( 13333 , "can't split a chunk in that many parts", m.size() < maxSplitPoints ); - uassert( 13003 , "can't split a chunk with only one distinct value" , _min.woCompare(_max) ); + builder.append("secondaryThrottle", secondaryThrottle); - ScopedDbConnection conn(_getShardConnectionString()); + if (secondaryThrottle && writeConcern) { + builder.append("writeConcern", writeConcern->toBSON()); + } - BSONObjBuilder cmd; - cmd.append( "splitChunk" , _manager->getns() ); - cmd.append( "keyPattern" , _manager->getShardKeyPattern().toBSON() ); - cmd.append( "min" , getMin() ); - cmd.append( "max" , getMax() ); - cmd.append( "from" , getShardId()); - cmd.append( "splitKeys" , m ); - cmd.append("configdb", grid.catalogManager()->connectionString().toString()); - cmd.append("epoch", _manager->getVersion().epoch()); - BSONObj cmdObj = cmd.obj(); + builder.append("waitForDelete", waitForDelete); + builder.append(LiteParsedQuery::cmdOptionMaxTimeMS, maxTimeMS); + builder.append("epoch", _manager->getVersion().epoch()); - BSONObj dummy; - if (res == NULL) { - res = &dummy; - } + ScopedDbConnection fromconn(from->getConnString()); + bool worked = fromconn->runCommand("admin", builder.done(), res); + fromconn.done(); - if (!conn->runCommand("admin", cmdObj, *res)) { - string msg(str::stream() << "splitChunk failed - cmd: " - << cmdObj << " result: " << *res); - warning() << msg; - conn.done(); + LOG(worked ? 1 : 0) << "moveChunk result: " << res; - return Status(ErrorCodes::SplitFailed, msg); - } + // if succeeded, needs to reload to pick up the new location + // if failed, mongos may be stale + // reload is excessive here as the failure could be simply because collection metadata is taken + _manager->reload(); - conn.done(); - - // force reload of config - _manager->reload(); - - return Status::OK(); - } - - bool Chunk::moveAndCommit(const ShardId& toShardId, - long long chunkSize /* bytes */, - const WriteConcernOptions* writeConcern, - bool waitForDelete, - int maxTimeMS, - BSONObj& res) const { - uassert(10167, - "can't move shard to its current location!", - getShardId() != toShardId); - - log() << "moving chunk ns: " << _manager->getns() << " moving ( " << toString() << ") " - << getShardId() << " -> " << toShardId; - - const auto from = grid.shardRegistry()->getShard(getShardId()); - - BSONObjBuilder builder; - builder.append("moveChunk", _manager->getns()); - builder.append("from", from->getConnString().toString()); - { - const auto toShard = grid.shardRegistry()->getShard(toShardId); - builder.append("to", toShard->getConnString().toString()); - } - // NEEDED FOR 2.0 COMPATIBILITY - builder.append("fromShard", from->getId()); - builder.append("toShard", toShardId); - /////////////////////////////// - builder.append("min", _min); - builder.append("max", _max); - builder.append("maxChunkSizeBytes", chunkSize); - builder.append("configdb", grid.catalogManager()->connectionString().toString()); - - // For legacy secondary throttle setting. - bool secondaryThrottle = true; - if (writeConcern && - writeConcern->wNumNodes <= 1 && - writeConcern->wMode.empty()) { - secondaryThrottle = false; - } + return worked; +} - builder.append("secondaryThrottle", secondaryThrottle); +bool Chunk::splitIfShould(long dataWritten) const { + dassert(ShouldAutoSplit); + LastError::Disabled d(&LastError::get(cc())); - if (secondaryThrottle && writeConcern) { - builder.append("writeConcern", writeConcern->toBSON()); + try { + _dataWritten += dataWritten; + int splitThreshold = getManager()->getCurrentDesiredChunkSize(); + if (_minIsInf() || _maxIsInf()) { + splitThreshold = (int)((double)splitThreshold * .9); } - builder.append("waitForDelete", waitForDelete); - builder.append(LiteParsedQuery::cmdOptionMaxTimeMS, maxTimeMS); - builder.append("epoch", _manager->getVersion().epoch()); - - ScopedDbConnection fromconn(from->getConnString()); - bool worked = fromconn->runCommand("admin", builder.done(), res); - fromconn.done(); + if (_dataWritten < splitThreshold / ChunkManager::SplitHeuristics::splitTestFactor) + return false; - LOG( worked ? 1 : 0 ) << "moveChunk result: " << res; + if (!getManager()->_splitHeuristics._splitTickets.tryAcquire()) { + LOG(1) << "won't auto split because not enough tickets: " << getManager()->getns(); + return false; + } + TicketHolderReleaser releaser(&(getManager()->_splitHeuristics._splitTickets)); - // if succeeded, needs to reload to pick up the new location - // if failed, mongos may be stale - // reload is excessive here as the failure could be simply because collection metadata is taken - _manager->reload(); + // this is a bit ugly + // we need it so that mongos blocks for the writes to actually be committed + // this does mean mongos has more back pressure than mongod alone + // since it nots 100% tcp queue bound + // this was implicit before since we did a splitVector on the same socket + ShardConnection::sync(); - return worked; - } + LOG(1) << "about to initiate autosplit: " << *this << " dataWritten: " << _dataWritten + << " splitThreshold: " << splitThreshold; - bool Chunk::splitIfShould( long dataWritten ) const { - dassert( ShouldAutoSplit ); - LastError::Disabled d(&LastError::get(cc())); + BSONObj res; + size_t splitCount = 0; + Status status = split(Chunk::autoSplitInternal, &splitCount, &res); + if (!status.isOK()) { + // Split would have issued a message if we got here. This means there wasn't enough + // data to split, so don't want to try again until considerable more data + _dataWritten = 0; + return false; + } - try { - _dataWritten += dataWritten; - int splitThreshold = getManager()->getCurrentDesiredChunkSize(); - if (_minIsInf() || _maxIsInf()) { - splitThreshold = (int)((double)splitThreshold * .9); - } + if (_maxIsInf() || _minIsInf()) { + // we don't want to reset _dataWritten since we kind of want to check the other side right away + } else { + // we're splitting, so should wait a bit + _dataWritten = 0; + } - if ( _dataWritten < splitThreshold / ChunkManager::SplitHeuristics::splitTestFactor ) - return false; - - if ( ! getManager()->_splitHeuristics._splitTickets.tryAcquire() ) { - LOG(1) << "won't auto split because not enough tickets: " << getManager()->getns(); - return false; - } - TicketHolderReleaser releaser( &(getManager()->_splitHeuristics._splitTickets) ); - - // this is a bit ugly - // we need it so that mongos blocks for the writes to actually be committed - // this does mean mongos has more back pressure than mongod alone - // since it nots 100% tcp queue bound - // this was implicit before since we did a splitVector on the same socket - ShardConnection::sync(); - - LOG(1) << "about to initiate autosplit: " << *this << " dataWritten: " << _dataWritten << " splitThreshold: " << splitThreshold; - - BSONObj res; - size_t splitCount = 0; - Status status = split(Chunk::autoSplitInternal, - &splitCount, - &res); + bool shouldBalance = grid.getConfigShouldBalance(); + if (shouldBalance) { + auto status = grid.catalogManager()->getCollection(_manager->getns()); if (!status.isOK()) { - // Split would have issued a message if we got here. This means there wasn't enough - // data to split, so don't want to try again until considerable more data - _dataWritten = 0; + log() << "Auto-split for " << _manager->getns() + << " failed to load collection metadata due to " << status.getStatus(); return false; } - - if (_maxIsInf() || _minIsInf()) { - // we don't want to reset _dataWritten since we kind of want to check the other side right away - } - else { - // we're splitting, so should wait a bit - _dataWritten = 0; - } - - bool shouldBalance = grid.getConfigShouldBalance(); - if (shouldBalance) { - auto status = grid.catalogManager()->getCollection(_manager->getns()); - if (!status.isOK()) { - log() << "Auto-split for " << _manager->getns() - << " failed to load collection metadata due to " << status.getStatus(); - return false; - } - shouldBalance = status.getValue().getAllowBalance(); - } + shouldBalance = status.getValue().getAllowBalance(); + } - log() << "autosplitted " << _manager->getns() - << " shard: " << toString() - << " into " << (splitCount + 1) - << " (splitThreshold " << splitThreshold << ")" + log() << "autosplitted " << _manager->getns() << " shard: " << toString() << " into " + << (splitCount + 1) << " (splitThreshold " << splitThreshold << ")" #ifdef MONGO_CONFIG_DEBUG_BUILD - << " size: " << getPhysicalSize() // slow - but can be useful when debugging + << " size: " << getPhysicalSize() // slow - but can be useful when debugging #endif - << ( res["shouldMigrate"].eoo() ? "" : (string)" (migrate suggested" + - ( shouldBalance ? ")" : ", but no migrations allowed)" ) ); - - // Top chunk optimization - try to move the top chunk out of this shard - // to prevent the hot spot from staying on a single shard. This is based on - // the assumption that succeeding inserts will fall on the top chunk. - BSONElement shouldMigrate = res["shouldMigrate"]; // not in mongod < 1.9.1 but that is ok - if ( ! shouldMigrate.eoo() && shouldBalance ){ - BSONObj range = shouldMigrate.embeddedObject(); - - ChunkType chunkToMove; - { - const auto shard = grid.shardRegistry()->getShard(getShardId()); - chunkToMove.setShard(shard->toString()); - } - chunkToMove.setMin(range["min"].embeddedObject()); - chunkToMove.setMax(range["max"].embeddedObject()); - - tryMoveToOtherShard(*_manager, chunkToMove); + << (res["shouldMigrate"].eoo() ? "" : (string) " (migrate suggested" + + (shouldBalance ? ")" : ", but no migrations allowed)")); + + // Top chunk optimization - try to move the top chunk out of this shard + // to prevent the hot spot from staying on a single shard. This is based on + // the assumption that succeeding inserts will fall on the top chunk. + BSONElement shouldMigrate = res["shouldMigrate"]; // not in mongod < 1.9.1 but that is ok + if (!shouldMigrate.eoo() && shouldBalance) { + BSONObj range = shouldMigrate.embeddedObject(); + + ChunkType chunkToMove; + { + const auto shard = grid.shardRegistry()->getShard(getShardId()); + chunkToMove.setShard(shard->toString()); } + chunkToMove.setMin(range["min"].embeddedObject()); + chunkToMove.setMax(range["max"].embeddedObject()); - return true; - + tryMoveToOtherShard(*_manager, chunkToMove); } - catch ( DBException& e ) { - // TODO: Make this better - there are lots of reasons a split could fail - // Random so that we don't sync up with other failed splits - _dataWritten = mkDataWritten(); - - // if the collection lock is taken (e.g. we're migrating), it is fine for the split to fail. - warning() << "could not autosplit collection " << _manager->getns() << causedBy( e ); - return false; - } - } - - const ConnectionString& Chunk::_getShardConnectionString() const { - const auto shard = grid.shardRegistry()->getShard(getShardId()); - return shard->getConnString(); - } - - long Chunk::getPhysicalSize() const { - ScopedDbConnection conn(_getShardConnectionString()); + return true; - BSONObj result; - uassert( 10169 , "datasize failed!" , conn->runCommand( "admin" , - BSON( "datasize" << _manager->getns() - << "keyPattern" << _manager->getShardKeyPattern().toBSON() - << "min" << getMin() - << "max" << getMax() - << "maxSize" << ( MaxChunkSize + 1 ) - << "estimate" << true - ) , result ) ); + } catch (DBException& e) { + // TODO: Make this better - there are lots of reasons a split could fail + // Random so that we don't sync up with other failed splits + _dataWritten = mkDataWritten(); - conn.done(); - return (long)result["size"].number(); + // if the collection lock is taken (e.g. we're migrating), it is fine for the split to fail. + warning() << "could not autosplit collection " << _manager->getns() << causedBy(e); + return false; } - - void Chunk::appendShortVersion( const char * name , BSONObjBuilder& b ) const { - BSONObjBuilder bb( b.subobjStart( name ) ); - bb.append(ChunkType::min(), _min); - bb.append(ChunkType::max(), _max); - bb.done(); +} + +const ConnectionString& Chunk::_getShardConnectionString() const { + const auto shard = grid.shardRegistry()->getShard(getShardId()); + return shard->getConnString(); +} + +long Chunk::getPhysicalSize() const { + ScopedDbConnection conn(_getShardConnectionString()); + + BSONObj result; + uassert(10169, + "datasize failed!", + conn->runCommand("admin", + BSON("datasize" << _manager->getns() << "keyPattern" + << _manager->getShardKeyPattern().toBSON() << "min" + << getMin() << "max" << getMax() << "maxSize" + << (MaxChunkSize + 1) << "estimate" << true), + result)); + + conn.done(); + return (long)result["size"].number(); +} + +void Chunk::appendShortVersion(const char* name, BSONObjBuilder& b) const { + BSONObjBuilder bb(b.subobjStart(name)); + bb.append(ChunkType::min(), _min); + bb.append(ChunkType::max(), _max); + bb.done(); +} + +bool Chunk::operator==(const Chunk& s) const { + return _min.woCompare(s._min) == 0 && _max.woCompare(s._max) == 0; +} + +void Chunk::serialize(BSONObjBuilder& to, ChunkVersion myLastMod) { + to.append("_id", genID(_manager->getns(), _min)); + + if (myLastMod.isSet()) { + myLastMod.addToBSON(to, ChunkType::DEPRECATED_lastmod()); + } else if (_lastmod.isSet()) { + _lastmod.addToBSON(to, ChunkType::DEPRECATED_lastmod()); + } else { + verify(0); } - bool Chunk::operator==( const Chunk& s ) const { - return _min.woCompare( s._min ) == 0 && _max.woCompare( s._max ) == 0; - } + to << ChunkType::ns(_manager->getns()); + to << ChunkType::min(_min); + to << ChunkType::max(_max); + to << ChunkType::shard(_shardId); +} - void Chunk::serialize(BSONObjBuilder& to,ChunkVersion myLastMod) { +string Chunk::genID() const { + return genID(_manager->getns(), _min); +} - to.append( "_id" , genID( _manager->getns() , _min ) ); +string Chunk::genID(const string& ns, const BSONObj& o) { + StringBuilder buf; + buf << ns << "-"; - if ( myLastMod.isSet() ) { - myLastMod.addToBSON(to, ChunkType::DEPRECATED_lastmod()); - } - else if ( _lastmod.isSet() ) { - _lastmod.addToBSON(to, ChunkType::DEPRECATED_lastmod()); - } - else { - verify(0); - } - - to << ChunkType::ns(_manager->getns()); - to << ChunkType::min(_min); - to << ChunkType::max(_max); - to << ChunkType::shard(_shardId); + BSONObjIterator i(o); + while (i.more()) { + BSONElement e = i.next(); + buf << e.fieldName() << "_" << e.toString(false, true); } - string Chunk::genID() const { - return genID(_manager->getns(), _min); + return buf.str(); +} + +string Chunk::toString() const { + stringstream ss; + ss << ChunkType::ns() << ": " << _manager->getns() << ", " << ChunkType::shard() << ": " + << _shardId << ", " << ChunkType::DEPRECATED_lastmod() << ": " << _lastmod.toString() << ", " + << ChunkType::min() << ": " << _min << ", " << ChunkType::max() << ": " << _max; + return ss.str(); +} + +void Chunk::markAsJumbo() const { + // set this first + // even if we can't set it in the db + // at least this mongos won't try and keep moving + _jumbo = true; + + Status result = grid.catalogManager()->update(ChunkType::ConfigNS, + BSON(ChunkType::name(genID())), + BSON("$set" << BSON(ChunkType::jumbo(true))), + false, // upsert + false, // multi + NULL); + if (!result.isOK()) { + warning() << "couldn't set jumbo for chunk: " << genID() << result.reason(); } - - string Chunk::genID( const string& ns , const BSONObj& o ) { - StringBuilder buf; - buf << ns << "-"; - - BSONObjIterator i(o); - while ( i.more() ) { - BSONElement e = i.next(); - buf << e.fieldName() << "_" << e.toString(false, true); - } - - return buf.str(); - } - - string Chunk::toString() const { - stringstream ss; - ss << ChunkType::ns() << ": " << _manager->getns() << ", " - << ChunkType::shard() << ": " << _shardId << ", " - << ChunkType::DEPRECATED_lastmod() << ": " << _lastmod.toString() << ", " - << ChunkType::min() << ": " << _min << ", " - << ChunkType::max() << ": " << _max; - return ss.str(); - } - - void Chunk::markAsJumbo() const { - // set this first - // even if we can't set it in the db - // at least this mongos won't try and keep moving - _jumbo = true; - - Status result = grid.catalogManager()->update(ChunkType::ConfigNS, - BSON(ChunkType::name(genID())), - BSON("$set" << BSON(ChunkType::jumbo(true))), - false, // upsert - false, // multi - NULL); - if (!result.isOK()) { - warning() << "couldn't set jumbo for chunk: " << genID() << result.reason(); - } +} + +void Chunk::refreshChunkSize() { + auto chunkSizeSettingsResult = + grid.catalogManager()->getGlobalSettings(SettingsType::ChunkSizeDocKey); + if (!chunkSizeSettingsResult.isOK()) { + log() << chunkSizeSettingsResult.getStatus(); + return; } + SettingsType chunkSizeSettings = chunkSizeSettingsResult.getValue(); + int csize = chunkSizeSettings.getChunkSizeMB(); - void Chunk::refreshChunkSize() { - auto chunkSizeSettingsResult = - grid.catalogManager()->getGlobalSettings(SettingsType::ChunkSizeDocKey); - if (!chunkSizeSettingsResult.isOK()) { - log() << chunkSizeSettingsResult.getStatus(); - return; - } - SettingsType chunkSizeSettings = chunkSizeSettingsResult.getValue(); - int csize = chunkSizeSettings.getChunkSizeMB(); - - LOG(1) << "Refreshing MaxChunkSize: " << csize << "MB"; - - if (csize != Chunk::MaxChunkSize/(1024*1024)) { - log() << "MaxChunkSize changing from " << Chunk::MaxChunkSize/(1024*1024) << "MB" - << " to " << csize << "MB"; - } + LOG(1) << "Refreshing MaxChunkSize: " << csize << "MB"; - if ( !setMaxChunkSizeSizeMB( csize ) ) { - warning() << "invalid MaxChunkSize: " << csize; - } + if (csize != Chunk::MaxChunkSize / (1024 * 1024)) { + log() << "MaxChunkSize changing from " << Chunk::MaxChunkSize / (1024 * 1024) << "MB" + << " to " << csize << "MB"; } - bool Chunk::setMaxChunkSizeSizeMB( int newMaxChunkSize ) { - if ( newMaxChunkSize < 1 ) - return false; - if ( newMaxChunkSize > 1024 ) - return false; - MaxChunkSize = newMaxChunkSize * 1024 * 1024; - return true; + if (!setMaxChunkSizeSizeMB(csize)) { + warning() << "invalid MaxChunkSize: " << csize; } - -} // namespace mongo +} + +bool Chunk::setMaxChunkSizeSizeMB(int newMaxChunkSize) { + if (newMaxChunkSize < 1) + return false; + if (newMaxChunkSize > 1024) + return false; + MaxChunkSize = newMaxChunkSize * 1024 * 1024; + return true; +} + +} // namespace mongo diff --git a/src/mongo/s/chunk.h b/src/mongo/s/chunk.h index b90da174961..d4fd7857d30 100644 --- a/src/mongo/s/chunk.h +++ b/src/mongo/s/chunk.h @@ -33,251 +33,271 @@ namespace mongo { - class ChunkManager; - struct WriteConcernOptions; +class ChunkManager; +struct WriteConcernOptions; + +/** + config.chunks + { ns : "alleyinsider.fs.chunks" , min : {} , max : {} , server : "localhost:30001" } + + x is in a shard iff + min <= x < max + */ +class Chunk { + MONGO_DISALLOW_COPYING(Chunk); + +public: + enum SplitPointMode { + // Determines the split points that will make the current chunk smaller than + // the current chunk size setting. Gives empty result if chunk is not big enough. + normal, + + // Will get a split which approximately splits the chunk into 2 halves, + // regardless of the size of the chunk. + atMedian, + + // Behaves like normal, with additional special heuristics for "top chunks" + // (the 1 or 2 chunks in the extreme ends of the chunk key space). + autoSplitInternal + }; + + Chunk(const ChunkManager* info, BSONObj from); + Chunk(const ChunkManager* info, + const BSONObj& min, + const BSONObj& max, + const ShardId& shardId, + ChunkVersion lastmod = ChunkVersion()); + + // + // serialization support + // + + void serialize(BSONObjBuilder& to, ChunkVersion myLastMod = ChunkVersion(0, 0, OID())); + + // + // chunk boundary support + // + + const BSONObj& getMin() const { + return _min; + } + const BSONObj& getMax() const { + return _max; + } + + // Returns true if this chunk contains the given shard key, and false otherwise + // + // Note: this function takes an extracted *key*, not an original document + // (the point may be computed by, say, hashing a given field or projecting + // to a subset of fields). + bool containsKey(const BSONObj& shardKey) const; + + std::string genID() const; + static std::string genID(const std::string& ns, const BSONObj& min); + + // + // chunk version support + // + + void appendShortVersion(const char* name, BSONObjBuilder& b) const; + + ChunkVersion getLastmod() const { + return _lastmod; + } + void setLastmod(ChunkVersion v) { + _lastmod = v; + } + + // + // split support + // + + long long getBytesWritten() const { + return _dataWritten; + } + // Const since _dataWritten is mutable and a heuristic + // TODO: Split data tracking and chunk information + void setBytesWritten(long long bytesWritten) const { + _dataWritten = bytesWritten; + } /** - config.chunks - { ns : "alleyinsider.fs.chunks" , min : {} , max : {} , server : "localhost:30001" } + * if the amount of data written nears the max size of a shard + * then we check the real size, and if its too big, we split + * @return if something was split + */ + bool splitIfShould(long dataWritten) const; - x is in a shard iff - min <= x < max + /** + * Splits this chunk at a non-specificed split key to be chosen by the mongod holding this chunk. + * + * @param mode + * @param res the object containing details about the split execution + * @param resultingSplits the number of resulting split points. Set to NULL to ignore. + * + * @throws UserException */ - class Chunk { - MONGO_DISALLOW_COPYING(Chunk); - public: - enum SplitPointMode { - // Determines the split points that will make the current chunk smaller than - // the current chunk size setting. Gives empty result if chunk is not big enough. - normal, - - // Will get a split which approximately splits the chunk into 2 halves, - // regardless of the size of the chunk. - atMedian, - - // Behaves like normal, with additional special heuristics for "top chunks" - // (the 1 or 2 chunks in the extreme ends of the chunk key space). - autoSplitInternal - }; - - Chunk( const ChunkManager * info , BSONObj from); - Chunk( const ChunkManager * info , - const BSONObj& min, - const BSONObj& max, - const ShardId& shardId, - ChunkVersion lastmod = ChunkVersion() ); - - // - // serialization support - // - - void serialize(BSONObjBuilder& to, ChunkVersion myLastMod = ChunkVersion(0, 0, OID())); - - // - // chunk boundary support - // - - const BSONObj& getMin() const { return _min; } - const BSONObj& getMax() const { return _max; } - - // Returns true if this chunk contains the given shard key, and false otherwise - // - // Note: this function takes an extracted *key*, not an original document - // (the point may be computed by, say, hashing a given field or projecting - // to a subset of fields). - bool containsKey( const BSONObj& shardKey ) const; - - std::string genID() const; - static std::string genID( const std::string& ns , const BSONObj& min ); - - // - // chunk version support - // - - void appendShortVersion( const char * name , BSONObjBuilder& b ) const; - - ChunkVersion getLastmod() const { return _lastmod; } - void setLastmod( ChunkVersion v ) { _lastmod = v; } - - // - // split support - // - - long long getBytesWritten() const { return _dataWritten; } - // Const since _dataWritten is mutable and a heuristic - // TODO: Split data tracking and chunk information - void setBytesWritten(long long bytesWritten) const { _dataWritten = bytesWritten; } - - /** - * if the amount of data written nears the max size of a shard - * then we check the real size, and if its too big, we split - * @return if something was split - */ - bool splitIfShould( long dataWritten ) const; - - /** - * Splits this chunk at a non-specificed split key to be chosen by the mongod holding this chunk. - * - * @param mode - * @param res the object containing details about the split execution - * @param resultingSplits the number of resulting split points. Set to NULL to ignore. - * - * @throws UserException - */ - Status split(SplitPointMode mode, - size_t* resultingSplits, - BSONObj* res) const; - - /** - * Splits this chunk at the given key (or keys) - * - * @param splitPoints the vector of keys that should be used to divide this chunk - * @param res the object containing details about the split execution - * - * @throws UserException - */ - Status multiSplit(const std::vector<BSONObj>& splitPoints, BSONObj* res) const; - - /** - * Asks the mongod holding this chunk to find a key that approximately divides this chunk in two - * - * @param medianKey the key that divides this chunk, if there is one, or empty - */ - void pickMedianKey( BSONObj& medianKey ) const; - - /** - * @param splitPoints vector to be filled in - * @param chunkSize chunk size to target in bytes - * @param maxPoints limits the number of split points that are needed, zero is max (optional) - * @param maxObjs limits the number of objects in each chunk, zero is as max (optional) - */ - void pickSplitVector(std::vector<BSONObj>& splitPoints, - long long chunkSize, - int maxPoints = 0, - int maxObjs = 0) const; - - // - // migration support - // - - /** - * Issues a migrate request for this chunk - * - * @param to shard to move this chunk to - * @param chunSize maximum number of bytes beyond which the migrate should no go trhough - * @param writeConcern detailed write concern. NULL means the default write concern. - * @param waitForDelete whether chunk move should wait for cleanup or return immediately - * @param maxTimeMS max time for the migrate request - * @param res the object containing details about the migrate execution - * @return true if move was successful - */ - bool moveAndCommit(const ShardId& to, - long long chunkSize, - const WriteConcernOptions* writeConcern, - bool waitForDelete, - int maxTimeMS, - BSONObj& res) const; - - /** - * @return size of shard in bytes - * talks to mongod to do this - */ - long getPhysicalSize() const; - - /** - * marks this chunk as a jumbo chunk - * that means the chunk will be inelligble for migrates - */ - void markAsJumbo() const; - - bool isJumbo() const { return _jumbo; } - - /** - * Attempt to refresh maximum chunk size from config. - */ - static void refreshChunkSize(); - - /** - * sets MaxChunkSize - * 1 <= newMaxChunkSize <= 1024 - * @return true if newMaxChunkSize is valid and was set - */ - static bool setMaxChunkSizeSizeMB( int newMaxChunkSize ); - - // - // public constants - // - - static long long MaxChunkSize; - static int MaxObjectPerChunk; - static bool ShouldAutoSplit; - - // - // accessors and helpers - // - - std::string toString() const; - - friend std::ostream& operator << (std::ostream& out, const Chunk& c) { return (out << c.toString()); } - - // chunk equality is determined by comparing the min and max bounds of the chunk - bool operator==(const Chunk& s) const; - bool operator!=(const Chunk& s) const { return ! ( *this == s ); } - - ShardId getShardId() const { return _shardId; } - const ChunkManager* getManager() const { return _manager; } - - private: - - /** - * Returns the connection string for the shard on which this chunk lives on. - */ - const ConnectionString& _getShardConnectionString() const; - - // if min/max key is pos/neg infinity - bool _minIsInf() const; - bool _maxIsInf() const; - - // The chunk manager, which owns this chunk. Not owned by the chunk. - const ChunkManager* _manager; - - BSONObj _min; - BSONObj _max; - ShardId _shardId; - ChunkVersion _lastmod; - mutable bool _jumbo; - - // transient stuff - - mutable long long _dataWritten; - - // methods, etc.. - - /** - * Returns the split point that will result in one of the chunk having exactly one - * document. Also returns an empty document if the split point cannot be determined. - * - * @param doSplitAtLower determines which side of the split will have exactly one document. - * True means that the split point chosen will be closer to the lower bound. - * - * Warning: this assumes that the shard key is not "special"- that is, the shardKeyPattern - * is simply an ordered list of ascending/descending field names. Examples: - * {a : 1, b : -1} is not special. {a : "hashed"} is. - */ - BSONObj _getExtremeKey(bool doSplitAtLower) const; - - /** - * Determines the appropriate split points for this chunk. - * - * @param atMedian perform a single split at the middle of this chunk. - * @param splitPoints out parameter containing the chosen split points. Can be empty. - */ - void determineSplitPoints(bool atMedian, std::vector<BSONObj>* splitPoints) const; - - /** initializes _dataWritten with a random value so that a mongos restart wouldn't cause delay in splitting */ - static int mkDataWritten(); - }; + Status split(SplitPointMode mode, size_t* resultingSplits, BSONObj* res) const; + + /** + * Splits this chunk at the given key (or keys) + * + * @param splitPoints the vector of keys that should be used to divide this chunk + * @param res the object containing details about the split execution + * + * @throws UserException + */ + Status multiSplit(const std::vector<BSONObj>& splitPoints, BSONObj* res) const; + + /** + * Asks the mongod holding this chunk to find a key that approximately divides this chunk in two + * + * @param medianKey the key that divides this chunk, if there is one, or empty + */ + void pickMedianKey(BSONObj& medianKey) const; + + /** + * @param splitPoints vector to be filled in + * @param chunkSize chunk size to target in bytes + * @param maxPoints limits the number of split points that are needed, zero is max (optional) + * @param maxObjs limits the number of objects in each chunk, zero is as max (optional) + */ + void pickSplitVector(std::vector<BSONObj>& splitPoints, + long long chunkSize, + int maxPoints = 0, + int maxObjs = 0) const; + + // + // migration support + // + + /** + * Issues a migrate request for this chunk + * + * @param to shard to move this chunk to + * @param chunSize maximum number of bytes beyond which the migrate should no go trhough + * @param writeConcern detailed write concern. NULL means the default write concern. + * @param waitForDelete whether chunk move should wait for cleanup or return immediately + * @param maxTimeMS max time for the migrate request + * @param res the object containing details about the migrate execution + * @return true if move was successful + */ + bool moveAndCommit(const ShardId& to, + long long chunkSize, + const WriteConcernOptions* writeConcern, + bool waitForDelete, + int maxTimeMS, + BSONObj& res) const; + + /** + * @return size of shard in bytes + * talks to mongod to do this + */ + long getPhysicalSize() const; + + /** + * marks this chunk as a jumbo chunk + * that means the chunk will be inelligble for migrates + */ + void markAsJumbo() const; + + bool isJumbo() const { + return _jumbo; + } + + /** + * Attempt to refresh maximum chunk size from config. + */ + static void refreshChunkSize(); + + /** + * sets MaxChunkSize + * 1 <= newMaxChunkSize <= 1024 + * @return true if newMaxChunkSize is valid and was set + */ + static bool setMaxChunkSizeSizeMB(int newMaxChunkSize); + + // + // public constants + // + + static long long MaxChunkSize; + static int MaxObjectPerChunk; + static bool ShouldAutoSplit; + + // + // accessors and helpers + // + + std::string toString() const; + + friend std::ostream& operator<<(std::ostream& out, const Chunk& c) { + return (out << c.toString()); + } + + // chunk equality is determined by comparing the min and max bounds of the chunk + bool operator==(const Chunk& s) const; + bool operator!=(const Chunk& s) const { + return !(*this == s); + } + + ShardId getShardId() const { + return _shardId; + } + const ChunkManager* getManager() const { + return _manager; + } + +private: + /** + * Returns the connection string for the shard on which this chunk lives on. + */ + const ConnectionString& _getShardConnectionString() const; + + // if min/max key is pos/neg infinity + bool _minIsInf() const; + bool _maxIsInf() const; + + // The chunk manager, which owns this chunk. Not owned by the chunk. + const ChunkManager* _manager; + + BSONObj _min; + BSONObj _max; + ShardId _shardId; + ChunkVersion _lastmod; + mutable bool _jumbo; + + // transient stuff + + mutable long long _dataWritten; + + // methods, etc.. + + /** + * Returns the split point that will result in one of the chunk having exactly one + * document. Also returns an empty document if the split point cannot be determined. + * + * @param doSplitAtLower determines which side of the split will have exactly one document. + * True means that the split point chosen will be closer to the lower bound. + * + * Warning: this assumes that the shard key is not "special"- that is, the shardKeyPattern + * is simply an ordered list of ascending/descending field names. Examples: + * {a : 1, b : -1} is not special. {a : "hashed"} is. + */ + BSONObj _getExtremeKey(bool doSplitAtLower) const; + + /** + * Determines the appropriate split points for this chunk. + * + * @param atMedian perform a single split at the middle of this chunk. + * @param splitPoints out parameter containing the chosen split points. Can be empty. + */ + void determineSplitPoints(bool atMedian, std::vector<BSONObj>* splitPoints) const; + + /** initializes _dataWritten with a random value so that a mongos restart wouldn't cause delay in splitting */ + static int mkDataWritten(); +}; - typedef std::shared_ptr<const Chunk> ChunkPtr; +typedef std::shared_ptr<const Chunk> ChunkPtr; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/chunk_diff.cpp b/src/mongo/s/chunk_diff.cpp index 488ea85728b..b270a54d9d3 100644 --- a/src/mongo/s/chunk_diff.cpp +++ b/src/mongo/s/chunk_diff.cpp @@ -42,223 +42,215 @@ namespace mongo { - template <class ValType, class ShardType> - ConfigDiffTracker<ValType, ShardType>::ConfigDiffTracker() { - _ns.clear(); - _currMap = NULL; - _maxVersion = NULL; - _maxShardVersions = NULL; - _validDiffs = 0; - } +template <class ValType, class ShardType> +ConfigDiffTracker<ValType, ShardType>::ConfigDiffTracker() { + _ns.clear(); + _currMap = NULL; + _maxVersion = NULL; + _maxShardVersions = NULL; + _validDiffs = 0; +} + +template <class ValType, class ShardType> +ConfigDiffTracker<ValType, ShardType>::~ConfigDiffTracker() = default; + +template <class ValType, class ShardType> +void ConfigDiffTracker<ValType, ShardType>::attach(const std::string& ns, + RangeMap& currMap, + ChunkVersion& maxVersion, + MaxChunkVersionMap& maxShardVersions) { + _ns = ns; + _currMap = &currMap; + _maxVersion = &maxVersion; + _maxShardVersions = &maxShardVersions; + _validDiffs = 0; +} + +template <class ValType, class ShardType> +bool ConfigDiffTracker<ValType, ShardType>::isOverlapping(const BSONObj& min, const BSONObj& max) { + RangeOverlap overlap = overlappingRange(min, max); + + return overlap.first != overlap.second; +} + +template <class ValType, class ShardType> +void ConfigDiffTracker<ValType, ShardType>::removeOverlapping(const BSONObj& min, + const BSONObj& max) { + _assertAttached(); - template <class ValType, class ShardType> - ConfigDiffTracker<ValType, ShardType>::~ConfigDiffTracker() = default; - - template <class ValType, class ShardType> - void ConfigDiffTracker<ValType, ShardType>::attach(const std::string& ns, - RangeMap& currMap, - ChunkVersion& maxVersion, - MaxChunkVersionMap& maxShardVersions) { - _ns = ns; - _currMap = &currMap; - _maxVersion = &maxVersion; - _maxShardVersions = &maxShardVersions; - _validDiffs = 0; - } + RangeOverlap overlap = overlappingRange(min, max); - template <class ValType, class ShardType> - bool ConfigDiffTracker<ValType, ShardType>::isOverlapping(const BSONObj& min, - const BSONObj& max) { - RangeOverlap overlap = overlappingRange(min, max); + _currMap->erase(overlap.first, overlap.second); +} - return overlap.first != overlap.second; - } +template <class ValType, class ShardType> +typename ConfigDiffTracker<ValType, ShardType>::RangeOverlap +ConfigDiffTracker<ValType, ShardType>::overlappingRange(const BSONObj& min, const BSONObj& max) { + _assertAttached(); + + typename RangeMap::iterator low; + typename RangeMap::iterator high; - template <class ValType, class ShardType> - void ConfigDiffTracker<ValType, ShardType>::removeOverlapping(const BSONObj& min, - const BSONObj& max) { - _assertAttached(); + if (isMinKeyIndexed()) { + // Returns the first chunk with a min key that is >= min - implies the + // previous chunk cannot overlap min + low = _currMap->lower_bound(min); - RangeOverlap overlap = overlappingRange(min, max); + // Returns the first chunk with a min key that is >= max - implies the + // chunk does not overlap max + high = _currMap->lower_bound(max); + } else { + // Returns the first chunk with a max key that is > min - implies that + // the chunk overlaps min + low = _currMap->upper_bound(min); - _currMap->erase(overlap.first, overlap.second); + // Returns the first chunk with a max key that is > max - implies that + // the next chunk cannot not overlap max + high = _currMap->upper_bound(max); } - template <class ValType, class ShardType> - typename ConfigDiffTracker<ValType, ShardType>::RangeOverlap - ConfigDiffTracker<ValType, ShardType>::overlappingRange(const BSONObj& min, - const BSONObj& max) { - _assertAttached(); + return RangeOverlap(low, high); +} - typename RangeMap::iterator low; - typename RangeMap::iterator high; +template <class ValType, class ShardType> +int ConfigDiffTracker<ValType, ShardType>::calculateConfigDiff(CatalogManager* catalogManager) { + _assertAttached(); - if (isMinKeyIndexed()) { - // Returns the first chunk with a min key that is >= min - implies the - // previous chunk cannot overlap min - low = _currMap->lower_bound(min); + // Get the diff query required + Query diffQuery = configDiffQuery(); - // Returns the first chunk with a min key that is >= max - implies the - // chunk does not overlap max - high = _currMap->lower_bound(max); - } - else { - // Returns the first chunk with a max key that is > min - implies that - // the chunk overlaps min - low = _currMap->upper_bound(min); - - // Returns the first chunk with a max key that is > max - implies that - // the next chunk cannot not overlap max - high = _currMap->upper_bound(max); - } + try { + std::vector<ChunkType> chunks; + uassertStatusOK(catalogManager->getChunks(diffQuery, 0, &chunks)); - return RangeOverlap(low, high); + return calculateConfigDiff(chunks); + } catch (DBException& e) { + // Should only happen on connection errors + e.addContext(str::stream() << "could not calculate config difference for ns " << _ns); + throw; } +} + +template <class ValType, class ShardType> +int ConfigDiffTracker<ValType, ShardType>::calculateConfigDiff( + const std::vector<ChunkType>& chunks) { + _assertAttached(); + + // Apply the chunk changes to the ranges and versions + // + // Overall idea here is to work in two steps : + // 1. For all the new chunks we find, increment the maximum version per-shard and + // per-collection, and remove any conflicting chunks from the ranges. + // 2. For all the new chunks we're interested in (all of them for mongos, just chunks on + // the shard for mongod) add them to the ranges. - template<class ValType, class ShardType> - int ConfigDiffTracker<ValType, ShardType>::calculateConfigDiff( - CatalogManager* catalogManager) { - _assertAttached(); + std::vector<ChunkType> newTracked; - // Get the diff query required - Query diffQuery = configDiffQuery(); + // Store epoch now so it doesn't change when we change max + OID currEpoch = _maxVersion->epoch(); - try { - std::vector<ChunkType> chunks; - uassertStatusOK(catalogManager->getChunks(diffQuery, 0, &chunks)); + _validDiffs = 0; - return calculateConfigDiff(chunks); + for (const ChunkType& chunk : chunks) { + ChunkVersion chunkVersion = + ChunkVersion::fromBSON(chunk.toBSON(), ChunkType::DEPRECATED_lastmod()); + + if (!chunkVersion.isSet() || !chunkVersion.hasEqualEpoch(currEpoch)) { + warning() << "got invalid chunk version " << chunkVersion << " in document " + << chunk.toString() << " when trying to load differing chunks at version " + << ChunkVersion( + _maxVersion->majorVersion(), _maxVersion->minorVersion(), currEpoch); + + // Don't keep loading, since we know we'll be broken here + return -1; } - catch (DBException& e) { - // Should only happen on connection errors - e.addContext(str::stream() << "could not calculate config difference for ns " << _ns); - throw; + + _validDiffs++; + + // Get max changed version and chunk version + if (chunkVersion > *_maxVersion) { + *_maxVersion = chunkVersion; } - } - template<class ValType, class ShardType> - int ConfigDiffTracker<ValType, ShardType>::calculateConfigDiff( - const std::vector<ChunkType>& chunks) { - _assertAttached(); + // Chunk version changes + ShardType shard = shardFor(chunk.getShard()); - // Apply the chunk changes to the ranges and versions - // - // Overall idea here is to work in two steps : - // 1. For all the new chunks we find, increment the maximum version per-shard and - // per-collection, and remove any conflicting chunks from the ranges. - // 2. For all the new chunks we're interested in (all of them for mongos, just chunks on - // the shard for mongod) add them to the ranges. - - std::vector<ChunkType> newTracked; - - // Store epoch now so it doesn't change when we change max - OID currEpoch = _maxVersion->epoch(); - - _validDiffs = 0; - - for (const ChunkType& chunk : chunks) { - ChunkVersion chunkVersion = ChunkVersion::fromBSON(chunk.toBSON(), - ChunkType::DEPRECATED_lastmod()); - - if (!chunkVersion.isSet() || !chunkVersion.hasEqualEpoch(currEpoch)) { - warning() << "got invalid chunk version " << chunkVersion - << " in document " << chunk.toString() - << " when trying to load differing chunks at version " - << ChunkVersion(_maxVersion->majorVersion(), - _maxVersion->minorVersion(), - currEpoch); - - // Don't keep loading, since we know we'll be broken here - return -1; - } - - _validDiffs++; - - // Get max changed version and chunk version - if (chunkVersion > *_maxVersion) { - *_maxVersion = chunkVersion; - } - - // Chunk version changes - ShardType shard = shardFor(chunk.getShard()); - - typename MaxChunkVersionMap::const_iterator shardVersionIt = _maxShardVersions->find(shard); - if (shardVersionIt == _maxShardVersions->end() || - shardVersionIt->second < chunkVersion) { - (*_maxShardVersions)[shard] = chunkVersion; - } - - // See if we need to remove any chunks we are currently tracking because of this - // chunk's changes - removeOverlapping(chunk.getMin(), chunk.getMax()); - - // Figure out which of the new chunks we need to track - // Important - we need to actually own this doc, in case the cursor decides to getMore - // or unbuffer. - if (isTracked(chunk)) { - newTracked.push_back(chunk); - } + typename MaxChunkVersionMap::const_iterator shardVersionIt = _maxShardVersions->find(shard); + if (shardVersionIt == _maxShardVersions->end() || shardVersionIt->second < chunkVersion) { + (*_maxShardVersions)[shard] = chunkVersion; } - LOG(3) << "found " << _validDiffs << " new chunks for collection " << _ns - << " (tracking " << newTracked.size() << "), new version is " << *_maxVersion; + // See if we need to remove any chunks we are currently tracking because of this + // chunk's changes + removeOverlapping(chunk.getMin(), chunk.getMax()); - for (const ChunkType& chunk : newTracked) { - // Invariant enforced by sharding - it's possible to read inconsistent state due to - // getMore and yielding, so we want to detect it as early as possible. - // - // TODO: This checks for overlap, we also should check for holes here iff we're - // tracking all chunks. - if (isOverlapping(chunk.getMin(), chunk.getMax())) { - return -1; - } + // Figure out which of the new chunks we need to track + // Important - we need to actually own this doc, in case the cursor decides to getMore + // or unbuffer. + if (isTracked(chunk)) { + newTracked.push_back(chunk); + } + } + + LOG(3) << "found " << _validDiffs << " new chunks for collection " << _ns << " (tracking " + << newTracked.size() << "), new version is " << *_maxVersion; - _currMap->insert(rangeFor(chunk)); + for (const ChunkType& chunk : newTracked) { + // Invariant enforced by sharding - it's possible to read inconsistent state due to + // getMore and yielding, so we want to detect it as early as possible. + // + // TODO: This checks for overlap, we also should check for holes here iff we're + // tracking all chunks. + if (isOverlapping(chunk.getMin(), chunk.getMax())) { + return -1; } - return _validDiffs; + _currMap->insert(rangeFor(chunk)); } - template<class ValType, class ShardType> - Query ConfigDiffTracker<ValType, ShardType>::configDiffQuery() const { - _assertAttached(); + return _validDiffs; +} - // Basic idea behind the query is to find all the chunks $gte the current max version. - // Currently, any splits and merges will increment the current max version. - BSONObjBuilder queryB; - queryB.append(ChunkType::ns(), _ns); +template <class ValType, class ShardType> +Query ConfigDiffTracker<ValType, ShardType>::configDiffQuery() const { + _assertAttached(); - { - BSONObjBuilder tsBuilder(queryB.subobjStart(ChunkType::DEPRECATED_lastmod())); - tsBuilder.appendTimestamp("$gte", _maxVersion->toLong()); - tsBuilder.done(); - } + // Basic idea behind the query is to find all the chunks $gte the current max version. + // Currently, any splits and merges will increment the current max version. + BSONObjBuilder queryB; + queryB.append(ChunkType::ns(), _ns); - // NOTE: IT IS IMPORTANT FOR CONSISTENCY THAT WE SORT BY ASC VERSION, IN ORDER TO HANDLE - // CURSOR YIELDING BETWEEN CHUNKS BEING MIGRATED. - // - // This ensures that changes to chunk version (which will always be higher) will always - // come *after* our current position in the chunk cursor. + { + BSONObjBuilder tsBuilder(queryB.subobjStart(ChunkType::DEPRECATED_lastmod())); + tsBuilder.appendTimestamp("$gte", _maxVersion->toLong()); + tsBuilder.done(); + } - Query queryObj(queryB.obj()); - queryObj.sort(BSON("lastmod" << 1)); + // NOTE: IT IS IMPORTANT FOR CONSISTENCY THAT WE SORT BY ASC VERSION, IN ORDER TO HANDLE + // CURSOR YIELDING BETWEEN CHUNKS BEING MIGRATED. + // + // This ensures that changes to chunk version (which will always be higher) will always + // come *after* our current position in the chunk cursor. - LOG(2) << "major version query from " << *_maxVersion - << " and over " << _maxShardVersions->size() << " shards is " << queryObj; + Query queryObj(queryB.obj()); + queryObj.sort(BSON("lastmod" << 1)); - return queryObj; - } + LOG(2) << "major version query from " << *_maxVersion << " and over " + << _maxShardVersions->size() << " shards is " << queryObj; - template <class ValType, class ShardType> - void ConfigDiffTracker<ValType, ShardType>::_assertAttached() const { - invariant(_currMap); - invariant(_maxVersion); - invariant(_maxShardVersions); - } + return queryObj; +} + +template <class ValType, class ShardType> +void ConfigDiffTracker<ValType, ShardType>::_assertAttached() const { + invariant(_currMap); + invariant(_maxVersion); + invariant(_maxShardVersions); +} - // Ensures that these instances of the template are compiled - template class ConfigDiffTracker<BSONObj, std::string>; - template class ConfigDiffTracker<std::shared_ptr<Chunk>, std::string>; +// Ensures that these instances of the template are compiled +template class ConfigDiffTracker<BSONObj, std::string>; +template class ConfigDiffTracker<std::shared_ptr<Chunk>, std::string>; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/chunk_diff.h b/src/mongo/s/chunk_diff.h index 9a67583a53b..00d713addd5 100644 --- a/src/mongo/s/chunk_diff.h +++ b/src/mongo/s/chunk_diff.h @@ -35,104 +35,108 @@ namespace mongo { - class ChunkType; - struct ChunkVersion; - class CatalogManager; - class Query; +class ChunkType; +struct ChunkVersion; +class CatalogManager; +class Query; - /** - * This class manages and applies diffs from partial config server data reloads. Because the - * config data can be large, we want to update it in small parts, not all-at-once. Once a - * ConfigDiffTracker is created, the current config data is *attached* to it, and it is then - * able to modify the data. - * - * The current form is templated b/c the overall algorithm is identical between mongos and - * mongod, but the actual chunk maps used differ in implementation. We don't want to copy the - * implementation, because the logic is identical, or the chunk data, because that would be - * slow for big clusters, so this is the alternative for now. - * - * TODO: Standardize between mongos and mongod and convert template parameters to types. - */ - template <class ValType, class ShardType> - class ConfigDiffTracker { - public: - // Stores ranges indexed by max or min key - typedef typename std::map<BSONObj, ValType, BSONObjCmp> RangeMap; +/** + * This class manages and applies diffs from partial config server data reloads. Because the + * config data can be large, we want to update it in small parts, not all-at-once. Once a + * ConfigDiffTracker is created, the current config data is *attached* to it, and it is then + * able to modify the data. + * + * The current form is templated b/c the overall algorithm is identical between mongos and + * mongod, but the actual chunk maps used differ in implementation. We don't want to copy the + * implementation, because the logic is identical, or the chunk data, because that would be + * slow for big clusters, so this is the alternative for now. + * + * TODO: Standardize between mongos and mongod and convert template parameters to types. + */ +template <class ValType, class ShardType> +class ConfigDiffTracker { +public: + // Stores ranges indexed by max or min key + typedef typename std::map<BSONObj, ValType, BSONObjCmp> RangeMap; - // Pair of iterators defining a subset of ranges - typedef typename std::pair<typename RangeMap::iterator, - typename RangeMap::iterator> RangeOverlap; + // Pair of iterators defining a subset of ranges + typedef + typename std::pair<typename RangeMap::iterator, typename RangeMap::iterator> RangeOverlap; - // Map of shard identifiers to the maximum chunk version on that shard - typedef typename std::map<ShardType, ChunkVersion> MaxChunkVersionMap; + // Map of shard identifiers to the maximum chunk version on that shard + typedef typename std::map<ShardType, ChunkVersion> MaxChunkVersionMap; - ConfigDiffTracker(); - virtual ~ConfigDiffTracker(); + ConfigDiffTracker(); + virtual ~ConfigDiffTracker(); - /** - * The tracker attaches to a set of ranges with versions, and uses a config server - * connection to update these. Because the set of ranges and versions may be large, they - * aren't owned by the tracker, they're just passed in and updated. Therefore they must all - * stay in scope while the tracker is working. - * - * TODO: Make a standard VersionedRange to encapsulate this info in both mongod and mongos? - */ - void attach(const std::string& ns, - RangeMap& currMap, - ChunkVersion& maxVersion, - MaxChunkVersionMap& maxShardVersions); + /** + * The tracker attaches to a set of ranges with versions, and uses a config server + * connection to update these. Because the set of ranges and versions may be large, they + * aren't owned by the tracker, they're just passed in and updated. Therefore they must all + * stay in scope while the tracker is working. + * + * TODO: Make a standard VersionedRange to encapsulate this info in both mongod and mongos? + */ + void attach(const std::string& ns, + RangeMap& currMap, + ChunkVersion& maxVersion, + MaxChunkVersionMap& maxShardVersions); - // Call after load for more information - int numValidDiffs() const { return _validDiffs; } + // Call after load for more information + int numValidDiffs() const { + return _validDiffs; + } - // Whether or not a range exists in the min/max region - bool isOverlapping(const BSONObj& min, const BSONObj& max); + // Whether or not a range exists in the min/max region + bool isOverlapping(const BSONObj& min, const BSONObj& max); - // Removes all ranges in the region from min/max - void removeOverlapping(const BSONObj& min, const BSONObj& max); + // Removes all ranges in the region from min/max + void removeOverlapping(const BSONObj& min, const BSONObj& max); - // Returns a subset of ranges overlapping the region min/max - RangeOverlap overlappingRange(const BSONObj& min, const BSONObj& max); + // Returns a subset of ranges overlapping the region min/max + RangeOverlap overlappingRange(const BSONObj& min, const BSONObj& max); - // Finds and applies the changes to a collection from the config servers via - // the catalog manager. - // Also includes minor version changes for particular major-version chunks if explicitly - // specified. - // Returns the number of diffs processed, or -1 if the diffs were inconsistent - // Throws a DBException on connection errors - int calculateConfigDiff(CatalogManager* catalogManager); + // Finds and applies the changes to a collection from the config servers via + // the catalog manager. + // Also includes minor version changes for particular major-version chunks if explicitly + // specified. + // Returns the number of diffs processed, or -1 if the diffs were inconsistent + // Throws a DBException on connection errors + int calculateConfigDiff(CatalogManager* catalogManager); - // Applies changes to the config data from a vector of chunks passed in - // Returns the number of diffs processed, or -1 if the diffs were inconsistent - // Throws a DBException on connection errors - int calculateConfigDiff(const std::vector<ChunkType>& chunks); + // Applies changes to the config data from a vector of chunks passed in + // Returns the number of diffs processed, or -1 if the diffs were inconsistent + // Throws a DBException on connection errors + int calculateConfigDiff(const std::vector<ChunkType>& chunks); - // Returns the query needed to find new changes to a collection from the config server - // Needed only if a custom connection is required to the config server - Query configDiffQuery() const; + // Returns the query needed to find new changes to a collection from the config server + // Needed only if a custom connection is required to the config server + Query configDiffQuery() const; - protected: - // Determines which chunks are actually being remembered by our RangeMap - virtual bool isTracked(const ChunkType& chunk) const = 0; +protected: + // Determines which chunks are actually being remembered by our RangeMap + virtual bool isTracked(const ChunkType& chunk) const = 0; - // Whether or not our RangeMap uses min or max keys - virtual bool isMinKeyIndexed() const { return true; } + // Whether or not our RangeMap uses min or max keys + virtual bool isMinKeyIndexed() const { + return true; + } - virtual std::pair<BSONObj, ValType> rangeFor(const ChunkType& chunk) const = 0; + virtual std::pair<BSONObj, ValType> rangeFor(const ChunkType& chunk) const = 0; - virtual ShardType shardFor(const std::string& name) const = 0; + virtual ShardType shardFor(const std::string& name) const = 0; - private: - void _assertAttached() const; +private: + void _assertAttached() const; - std::string _ns; - RangeMap* _currMap; - ChunkVersion* _maxVersion; - MaxChunkVersionMap* _maxShardVersions; + std::string _ns; + RangeMap* _currMap; + ChunkVersion* _maxVersion; + MaxChunkVersionMap* _maxShardVersions; - // Store for later use - int _validDiffs; - }; + // Store for later use + int _validDiffs; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/chunk_diff_test.cpp b/src/mongo/s/chunk_diff_test.cpp index d63cb236fee..067b80ac0f9 100644 --- a/src/mongo/s/chunk_diff_test.cpp +++ b/src/mongo/s/chunk_diff_test.cpp @@ -37,38 +37,41 @@ namespace { - using mongo::BSONObj; - using mongo::ChunkType; - using mongo::ConfigDiffTracker; - using std::string; - using std::pair; - using std::make_pair; +using mongo::BSONObj; +using mongo::ChunkType; +using mongo::ConfigDiffTracker; +using std::string; +using std::pair; +using std::make_pair; - // XXX - // We'd move ChunkDiffUnitTest here - // We can check the queries it generates. - // We can check if is populating the attaching structures properly - // +// XXX +// We'd move ChunkDiffUnitTest here +// We can check the queries it generates. +// We can check if is populating the attaching structures properly +// - // The default pass-through adapter for using config diffs. - class DefaultDiffAdapter : public ConfigDiffTracker<BSONObj,string> { - public: +// The default pass-through adapter for using config diffs. +class DefaultDiffAdapter : public ConfigDiffTracker<BSONObj, string> { +public: + DefaultDiffAdapter() {} + virtual ~DefaultDiffAdapter() {} - DefaultDiffAdapter() {} - virtual ~DefaultDiffAdapter() {} - - virtual bool isTracked(const ChunkType& chunk) const { return true; } - - virtual pair<BSONObj,BSONObj> rangeFor(const ChunkType& chunk) const { - return make_pair(chunk.getMin(), chunk.getMax()); - } + virtual bool isTracked(const ChunkType& chunk) const { + return true; + } - virtual string shardFor(const string& name) const { return name; } - }; + virtual pair<BSONObj, BSONObj> rangeFor(const ChunkType& chunk) const { + return make_pair(chunk.getMin(), chunk.getMax()); + } - TEST(Basics, Simple) { - DefaultDiffAdapter differ; - ASSERT_TRUE(true); + virtual string shardFor(const string& name) const { + return name; } +}; + +TEST(Basics, Simple) { + DefaultDiffAdapter differ; + ASSERT_TRUE(true); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/chunk_manager.cpp b/src/mongo/s/chunk_manager.cpp index bcf7db273fa..73c1461c92f 100644 --- a/src/mongo/s/chunk_manager.cpp +++ b/src/mongo/s/chunk_manager.cpp @@ -55,752 +55,745 @@ namespace mongo { - using std::shared_ptr; +using std::shared_ptr; - using std::make_pair; - using std::map; - using std::max; - using std::pair; - using std::set; - using std::string; - using std::vector; +using std::make_pair; +using std::map; +using std::max; +using std::pair; +using std::set; +using std::string; +using std::vector; namespace { - /** - * This is an adapter so we can use config diffs - mongos and mongod do them slightly - * differently - * - * The mongos adapter here tracks all shards, and stores ranges by (max, Chunk) in the map. - */ - class CMConfigDiffTracker : public ConfigDiffTracker<shared_ptr<Chunk>, string> { - public: - CMConfigDiffTracker(ChunkManager* manager) : _manager(manager) { } - - bool isTracked(const ChunkType& chunk) const final { - // Mongos tracks all shards - return true; - } +/** + * This is an adapter so we can use config diffs - mongos and mongod do them slightly + * differently + * + * The mongos adapter here tracks all shards, and stores ranges by (max, Chunk) in the map. + */ +class CMConfigDiffTracker : public ConfigDiffTracker<shared_ptr<Chunk>, string> { +public: + CMConfigDiffTracker(ChunkManager* manager) : _manager(manager) {} - bool isMinKeyIndexed() const final { return false; } + bool isTracked(const ChunkType& chunk) const final { + // Mongos tracks all shards + return true; + } - pair<BSONObj, shared_ptr<Chunk>> rangeFor(const ChunkType& chunk) const final { - shared_ptr<Chunk> c(new Chunk(_manager, chunk.toBSON())); - return make_pair(chunk.getMax(), c); - } + bool isMinKeyIndexed() const final { + return false; + } - string shardFor(const string& hostName) const final { - const auto shard = grid.shardRegistry()->getShard(hostName); - return shard->getId(); - } + pair<BSONObj, shared_ptr<Chunk>> rangeFor(const ChunkType& chunk) const final { + shared_ptr<Chunk> c(new Chunk(_manager, chunk.toBSON())); + return make_pair(chunk.getMax(), c); + } - private: - ChunkManager* const _manager; - }; + string shardFor(const string& hostName) const final { + const auto shard = grid.shardRegistry()->getShard(hostName); + return shard->getId(); + } +private: + ChunkManager* const _manager; +}; - bool allOfType(BSONType type, const BSONObj& o) { - BSONObjIterator it(o); - while(it.more()) { - if (it.next().type() != type) { - return false; - } + +bool allOfType(BSONType type, const BSONObj& o) { + BSONObjIterator it(o); + while (it.more()) { + if (it.next().type() != type) { + return false; } + } + return true; +} + +bool isChunkMapValid(const ChunkMap& chunkMap) { +#define ENSURE(x) \ + do { \ + if (!(x)) { \ + log() << "ChunkManager::_isValid failed: " #x; \ + return false; \ + } \ + } while (0) + + if (chunkMap.empty()) { return true; } - bool isChunkMapValid(const ChunkMap& chunkMap) { -#define ENSURE(x) do { if(!(x)) { log() << "ChunkManager::_isValid failed: " #x; return false; } } while(0) + // Check endpoints + ENSURE(allOfType(MinKey, chunkMap.begin()->second->getMin())); + ENSURE(allOfType(MaxKey, boost::prior(chunkMap.end())->second->getMax())); - if (chunkMap.empty()) { - return true; - } - - // Check endpoints - ENSURE(allOfType(MinKey, chunkMap.begin()->second->getMin())); - ENSURE(allOfType(MaxKey, boost::prior(chunkMap.end())->second->getMax())); + // Make sure there are no gaps or overlaps + for (ChunkMap::const_iterator it = boost::next(chunkMap.begin()), end = chunkMap.end(); + it != end; + ++it) { + ChunkMap::const_iterator last = boost::prior(it); - // Make sure there are no gaps or overlaps - for (ChunkMap::const_iterator it = boost::next(chunkMap.begin()), end = chunkMap.end(); it != end; ++it) { - ChunkMap::const_iterator last = boost::prior(it); - - if (!(it->second->getMin() == last->second->getMax())) { - log() << last->second->toString(); - log() << it->second->toString(); - log() << it->second->getMin(); - log() << last->second->getMax(); - } - - ENSURE(it->second->getMin() == last->second->getMax()); + if (!(it->second->getMin() == last->second->getMax())) { + log() << last->second->toString(); + log() << it->second->toString(); + log() << it->second->getMin(); + log() << last->second->getMax(); } - return true; - -#undef ENSURE + ENSURE(it->second->getMin() == last->second->getMax()); } -} // namespace + return true; - AtomicUInt32 ChunkManager::NextSequenceNumber(1U); - - ChunkManager::ChunkManager(const string& ns, const ShardKeyPattern& pattern, bool unique) - : _ns( ns ), - _keyPattern( pattern.getKeyPattern() ), - _unique( unique ), - _sequenceNumber(NextSequenceNumber.addAndFetch(1)), - _chunkRanges() { +#undef ENSURE +} + +} // namespace + +AtomicUInt32 ChunkManager::NextSequenceNumber(1U); + +ChunkManager::ChunkManager(const string& ns, const ShardKeyPattern& pattern, bool unique) + : _ns(ns), + _keyPattern(pattern.getKeyPattern()), + _unique(unique), + _sequenceNumber(NextSequenceNumber.addAndFetch(1)), + _chunkRanges() {} + +ChunkManager::ChunkManager(const CollectionType& coll) + : _ns(coll.getNs()), + _keyPattern(coll.getKeyPattern()), + _unique(coll.getUnique()), + _sequenceNumber(NextSequenceNumber.addAndFetch(1)), + _chunkRanges() { + _version = ChunkVersion::fromBSON(coll.toBSON()); +} + +void ChunkManager::loadExistingRanges(const ChunkManager* oldManager) { + int tries = 3; + + while (tries--) { + ChunkMap chunkMap; + set<ShardId> shardIds; + ShardVersionMap shardVersions; + + Timer t; + + bool success = _load(chunkMap, shardIds, &shardVersions, oldManager); + if (success) { + log() << "ChunkManager: time to load chunks for " << _ns << ": " << t.millis() << "ms" + << " sequenceNumber: " << _sequenceNumber << " version: " << _version.toString() + << " based on: " + << (oldManager ? oldManager->getVersion().toString() : "(empty)"); + + // TODO: Merge into diff code above, so we validate in one place + if (isChunkMapValid(chunkMap)) { + _chunkMap.swap(chunkMap); + _shardIds.swap(shardIds); + _shardVersions.swap(shardVersions); + _chunkRanges.reloadAll(_chunkMap); + + return; + } + } - } + if (_chunkMap.size() < 10) { + _printChunks(); + } - ChunkManager::ChunkManager(const CollectionType& coll) - : _ns(coll.getNs()), - _keyPattern(coll.getKeyPattern()), - _unique(coll.getUnique()), - _sequenceNumber(NextSequenceNumber.addAndFetch(1)), - _chunkRanges() { + warning() << "ChunkManager loaded an invalid config for " << _ns << ", trying again"; - _version = ChunkVersion::fromBSON(coll.toBSON()); + sleepmillis(10 * (3 - tries)); } - void ChunkManager::loadExistingRanges(const ChunkManager* oldManager) { - int tries = 3; + // This will abort construction so we should never have a reference to an invalid config + msgasserted(13282, + str::stream() << "Couldn't load a valid config for " << _ns + << " after 3 attempts. Please try again."); +} - while (tries--) { - ChunkMap chunkMap; - set<ShardId> shardIds; - ShardVersionMap shardVersions; +bool ChunkManager::_load(ChunkMap& chunkMap, + set<ShardId>& shardIds, + ShardVersionMap* shardVersions, + const ChunkManager* oldManager) { + // Reset the max version, but not the epoch, when we aren't loading from the oldManager + _version = ChunkVersion(0, 0, _version.epoch()); - Timer t; + // If we have a previous version of the ChunkManager to work from, use that info to reduce + // our config query + if (oldManager && oldManager->getVersion().isSet()) { + // Get the old max version + _version = oldManager->getVersion(); - bool success = _load(chunkMap, shardIds, &shardVersions, oldManager); - if (success) { - log() << "ChunkManager: time to load chunks for " << _ns << ": " - << t.millis() << "ms" - << " sequenceNumber: " << _sequenceNumber - << " version: " << _version.toString() - << " based on: " - << (oldManager ? oldManager->getVersion().toString() : "(empty)"); + // Load a copy of the old versions + *shardVersions = oldManager->_shardVersions; - // TODO: Merge into diff code above, so we validate in one place - if (isChunkMapValid(chunkMap)) { - _chunkMap.swap(chunkMap); - _shardIds.swap(shardIds); - _shardVersions.swap(shardVersions); - _chunkRanges.reloadAll(_chunkMap); - - return; - } - } + // Load a copy of the chunk map, replacing the chunk manager with our own + const ChunkMap& oldChunkMap = oldManager->getChunkMap(); - if (_chunkMap.size() < 10) { - _printChunks(); - } + // Could be v.expensive + // TODO: If chunks were immutable and didn't reference the manager, we could do more + // interesting things here + for (const auto& oldChunkMapEntry : oldChunkMap) { + shared_ptr<Chunk> oldC = oldChunkMapEntry.second; + shared_ptr<Chunk> newC(new Chunk( + this, oldC->getMin(), oldC->getMax(), oldC->getShardId(), oldC->getLastmod())); - warning() << "ChunkManager loaded an invalid config for " << _ns << ", trying again"; + newC->setBytesWritten(oldC->getBytesWritten()); - sleepmillis(10 * (3 - tries)); + chunkMap.insert(make_pair(oldC->getMax(), newC)); } - // This will abort construction so we should never have a reference to an invalid config - msgasserted(13282, str::stream() << "Couldn't load a valid config for " << _ns - << " after 3 attempts. Please try again."); + LOG(2) << "loading chunk manager for collection " << _ns + << " using old chunk manager w/ version " << _version.toString() << " and " + << oldChunkMap.size() << " chunks"; } - bool ChunkManager::_load(ChunkMap& chunkMap, - set<ShardId>& shardIds, - ShardVersionMap* shardVersions, - const ChunkManager* oldManager) { - - // Reset the max version, but not the epoch, when we aren't loading from the oldManager - _version = ChunkVersion(0, 0, _version.epoch()); - - // If we have a previous version of the ChunkManager to work from, use that info to reduce - // our config query - if (oldManager && oldManager->getVersion().isSet()) { - // Get the old max version - _version = oldManager->getVersion(); - - // Load a copy of the old versions - *shardVersions = oldManager->_shardVersions; - - // Load a copy of the chunk map, replacing the chunk manager with our own - const ChunkMap& oldChunkMap = oldManager->getChunkMap(); - - // Could be v.expensive - // TODO: If chunks were immutable and didn't reference the manager, we could do more - // interesting things here - for(const auto& oldChunkMapEntry : oldChunkMap){ - shared_ptr<Chunk> oldC = oldChunkMapEntry.second; - shared_ptr<Chunk> newC(new Chunk(this, - oldC->getMin(), - oldC->getMax(), - oldC->getShardId(), - oldC->getLastmod())); - - newC->setBytesWritten(oldC->getBytesWritten()); - - chunkMap.insert(make_pair(oldC->getMax(), newC)); + // Attach a diff tracker for the versioned chunk data + CMConfigDiffTracker differ(this); + differ.attach(_ns, chunkMap, _version, *shardVersions); + + // Diff tracker should *always* find at least one chunk if collection exists + int diffsApplied = differ.calculateConfigDiff(grid.catalogManager()); + if (diffsApplied > 0) { + LOG(2) << "loaded " << diffsApplied << " chunks into new chunk manager for " << _ns + << " with version " << _version; + + // Add all existing shards we find to the shards set + for (ShardVersionMap::iterator it = shardVersions->begin(); it != shardVersions->end();) { + shared_ptr<Shard> shard = grid.shardRegistry()->getShard(it->first); + if (shard) { + shardIds.insert(it->first); + ++it; + } else { + shardVersions->erase(it++); } - - LOG(2) << "loading chunk manager for collection " << _ns - << " using old chunk manager w/ version " << _version.toString() - << " and " << oldChunkMap.size() << " chunks"; } - // Attach a diff tracker for the versioned chunk data - CMConfigDiffTracker differ(this); - differ.attach(_ns, chunkMap, _version, *shardVersions); - - // Diff tracker should *always* find at least one chunk if collection exists - int diffsApplied = differ.calculateConfigDiff(grid.catalogManager()); - if (diffsApplied > 0) { - LOG(2) << "loaded " << diffsApplied << " chunks into new chunk manager for " << _ns - << " with version " << _version; - - // Add all existing shards we find to the shards set - for (ShardVersionMap::iterator it = shardVersions->begin(); - it != shardVersions->end(); ) { - - shared_ptr<Shard> shard = grid.shardRegistry()->getShard(it->first); - if (shard) { - shardIds.insert(it->first); - ++it; - } - else { - shardVersions->erase(it++); - } - } + return true; + } else if (diffsApplied == 0) { + // No chunks were found for the ns + warning() << "no chunks found when reloading " << _ns << ", previous version was " + << _version; - return true; - } - else if (diffsApplied == 0) { - // No chunks were found for the ns - warning() << "no chunks found when reloading " << _ns - << ", previous version was " << _version; + // Set all our data to empty + chunkMap.clear(); + shardVersions->clear(); - // Set all our data to empty - chunkMap.clear(); - shardVersions->clear(); + _version = ChunkVersion(0, 0, OID()); - _version = ChunkVersion(0, 0, OID()); + return true; + } else { // diffsApplied < 0 - return true; + bool allInconsistent = (differ.numValidDiffs() == 0); + if (allInconsistent) { + // All versions are different, this can be normal + warning() << "major change in chunk information found when reloading " << _ns + << ", previous version was " << _version; + } else { + // Inconsistent load halfway through (due to yielding cursor during load) + // should be rare + warning() << "inconsistent chunks found when reloading " << _ns + << ", previous version was " << _version << ", this should be rare"; } - else { // diffsApplied < 0 - bool allInconsistent = (differ.numValidDiffs() == 0); - if (allInconsistent) { - // All versions are different, this can be normal - warning() << "major change in chunk information found when reloading " - << _ns << ", previous version was " << _version; - } - else { - // Inconsistent load halfway through (due to yielding cursor during load) - // should be rare - warning() << "inconsistent chunks found when reloading " << _ns - << ", previous version was " << _version << ", this should be rare"; - } - - // Set all our data to empty to be extra safe - chunkMap.clear(); - shardVersions->clear(); + // Set all our data to empty to be extra safe + chunkMap.clear(); + shardVersions->clear(); - _version = ChunkVersion(0, 0, OID()); + _version = ChunkVersion(0, 0, OID()); - return allInconsistent; - } + return allInconsistent; } +} + +shared_ptr<ChunkManager> ChunkManager::reload(bool force) const { + const NamespaceString nss(_ns); + auto status = grid.catalogCache()->getDatabase(nss.db().toString()); + shared_ptr<DBConfig> config = uassertStatusOK(status); - shared_ptr<ChunkManager> ChunkManager::reload(bool force) const { - const NamespaceString nss(_ns); - auto status = grid.catalogCache()->getDatabase(nss.db().toString()); - shared_ptr<DBConfig> config = uassertStatusOK(status); + return config->getChunkManager(getns(), force); +} - return config->getChunkManager(getns(), force); +void ChunkManager::_printChunks() const { + for (ChunkMap::const_iterator it = _chunkMap.begin(), end = _chunkMap.end(); it != end; ++it) { + log() << *it->second; } +} + +void ChunkManager::calcInitSplitsAndShards(const ShardId& primaryShardId, + const vector<BSONObj>* initPoints, + const set<ShardId>* initShardIds, + vector<BSONObj>* splitPoints, + vector<ShardId>* shardIds) const { + verify(_chunkMap.size() == 0); + + unsigned long long numObjects = 0; + Chunk c(this, + _keyPattern.getKeyPattern().globalMin(), + _keyPattern.getKeyPattern().globalMax(), + primaryShardId); + + if (!initPoints || !initPoints->size()) { + // discover split points + { + const auto primaryShard = grid.shardRegistry()->getShard(primaryShardId); + // get stats to see if there is any data + ScopedDbConnection shardConn(primaryShard->getConnString()); - void ChunkManager::_printChunks() const { - for (ChunkMap::const_iterator it=_chunkMap.begin(), end=_chunkMap.end(); it != end; ++it) { - log() << *it->second ; + numObjects = shardConn->count(getns()); + shardConn.done(); } - } - void ChunkManager::calcInitSplitsAndShards( const ShardId& primaryShardId, - const vector<BSONObj>* initPoints, - const set<ShardId>* initShardIds, - vector<BSONObj>* splitPoints, - vector<ShardId>* shardIds ) const - { - verify( _chunkMap.size() == 0 ); - - unsigned long long numObjects = 0; - Chunk c(this, - _keyPattern.getKeyPattern().globalMin(), - _keyPattern.getKeyPattern().globalMax(), - primaryShardId); - - if ( !initPoints || !initPoints->size() ) { - // discover split points - { - const auto primaryShard = grid.shardRegistry()->getShard(primaryShardId); - // get stats to see if there is any data - ScopedDbConnection shardConn(primaryShard->getConnString()); - - numObjects = shardConn->count( getns() ); - shardConn.done(); - } + if (numObjects > 0) + c.pickSplitVector(*splitPoints, Chunk::MaxChunkSize); - if ( numObjects > 0 ) - c.pickSplitVector( *splitPoints , Chunk::MaxChunkSize ); + // since docs alread exists, must use primary shard + shardIds->push_back(primaryShardId); + } else { + // make sure points are unique and ordered + set<BSONObj> orderedPts; + for (unsigned i = 0; i < initPoints->size(); ++i) { + BSONObj pt = (*initPoints)[i]; + orderedPts.insert(pt); + } + for (set<BSONObj>::iterator it = orderedPts.begin(); it != orderedPts.end(); ++it) { + splitPoints->push_back(*it); + } - // since docs alread exists, must use primary shard + if (!initShardIds || !initShardIds->size()) { + // If not specified, only use the primary shard (note that it's not safe for mongos + // to put initial chunks on other shards without the primary mongod knowing). shardIds->push_back(primaryShardId); } else { - // make sure points are unique and ordered - set<BSONObj> orderedPts; - for ( unsigned i = 0; i < initPoints->size(); ++i ) { - BSONObj pt = (*initPoints)[i]; - orderedPts.insert( pt ); - } - for ( set<BSONObj>::iterator it = orderedPts.begin(); it != orderedPts.end(); ++it ) { - splitPoints->push_back( *it ); - } - - if ( !initShardIds || !initShardIds->size() ) { - // If not specified, only use the primary shard (note that it's not safe for mongos - // to put initial chunks on other shards without the primary mongod knowing). - shardIds->push_back(primaryShardId); - } else { - std::copy(initShardIds->begin() , initShardIds->end() , std::back_inserter(*shardIds)); - } + std::copy(initShardIds->begin(), initShardIds->end(), std::back_inserter(*shardIds)); } } +} - void ChunkManager::createFirstChunks(const ShardId& primaryShardId, - const vector<BSONObj>* initPoints, - const set<ShardId>* initShardIds) - { - // TODO distlock? - // TODO: Race condition if we shard the collection and insert data while we split across - // the non-primary shard. +void ChunkManager::createFirstChunks(const ShardId& primaryShardId, + const vector<BSONObj>* initPoints, + const set<ShardId>* initShardIds) { + // TODO distlock? + // TODO: Race condition if we shard the collection and insert data while we split across + // the non-primary shard. - vector<BSONObj> splitPoints; - vector<ShardId> shardIds; - calcInitSplitsAndShards(primaryShardId, initPoints, initShardIds, &splitPoints, &shardIds); + vector<BSONObj> splitPoints; + vector<ShardId> shardIds; + calcInitSplitsAndShards(primaryShardId, initPoints, initShardIds, &splitPoints, &shardIds); - // this is the first chunk; start the versioning from scratch - ChunkVersion version; - version.incEpoch(); - version.incMajor(); - - log() << "going to create " << splitPoints.size() + 1 << " chunk(s) for: " << _ns - << " using new epoch " << version.epoch() ; + // this is the first chunk; start the versioning from scratch + ChunkVersion version; + version.incEpoch(); + version.incMajor(); - for ( unsigned i=0; i<=splitPoints.size(); i++ ) { - BSONObj min = i == 0 ? _keyPattern.getKeyPattern().globalMin() : splitPoints[i-1]; - BSONObj max = i < splitPoints.size() ? - splitPoints[i] : _keyPattern.getKeyPattern().globalMax(); + log() << "going to create " << splitPoints.size() + 1 << " chunk(s) for: " << _ns + << " using new epoch " << version.epoch(); - Chunk temp( this , min , max , shardIds[ i % shardIds.size() ], version ); + for (unsigned i = 0; i <= splitPoints.size(); i++) { + BSONObj min = i == 0 ? _keyPattern.getKeyPattern().globalMin() : splitPoints[i - 1]; + BSONObj max = + i < splitPoints.size() ? splitPoints[i] : _keyPattern.getKeyPattern().globalMax(); - BSONObjBuilder chunkBuilder; - temp.serialize(chunkBuilder); + Chunk temp(this, min, max, shardIds[i % shardIds.size()], version); - BSONObj chunkObj = chunkBuilder.obj(); + BSONObjBuilder chunkBuilder; + temp.serialize(chunkBuilder); - Status result = grid.catalogManager()->update(ChunkType::ConfigNS, - BSON(ChunkType::name(temp.genID())), - chunkObj, - true, - false, - NULL); + BSONObj chunkObj = chunkBuilder.obj(); - version.incMinor(); + Status result = grid.catalogManager()->update( + ChunkType::ConfigNS, BSON(ChunkType::name(temp.genID())), chunkObj, true, false, NULL); - if (!result.isOK()) { - string ss = str::stream() << "creating first chunks failed. result: " - << result.reason(); - error() << ss; - msgasserted(15903, ss); - } - } + version.incMinor(); - _version = ChunkVersion(0, 0, version.epoch()); + if (!result.isOK()) { + string ss = str::stream() + << "creating first chunks failed. result: " << result.reason(); + error() << ss; + msgasserted(15903, ss); + } } - ChunkPtr ChunkManager::findIntersectingChunk( const BSONObj& shardKey ) const { - { - BSONObj chunkMin; - ChunkPtr chunk; - { - ChunkMap::const_iterator it = _chunkMap.upper_bound( shardKey ); - if (it != _chunkMap.end()) { - chunkMin = it->first; - chunk = it->second; - } - } + _version = ChunkVersion(0, 0, version.epoch()); +} - if ( chunk ) { - if ( chunk->containsKey( shardKey ) ){ - return chunk; - } - - log() << chunkMin; - log() << *chunk; - log() << shardKey; - - reload(); - msgasserted(13141, "Chunk map pointed to incorrect chunk"); +ChunkPtr ChunkManager::findIntersectingChunk(const BSONObj& shardKey) const { + { + BSONObj chunkMin; + ChunkPtr chunk; + { + ChunkMap::const_iterator it = _chunkMap.upper_bound(shardKey); + if (it != _chunkMap.end()) { + chunkMin = it->first; + chunk = it->second; } } - msgasserted( 8070 , - str::stream() << "couldn't find a chunk intersecting: " << shardKey - << " for ns: " << _ns - << " at version: " << _version.toString() - << ", number of chunks: " << _chunkMap.size() ); - } - - void ChunkManager::getShardIdsForQuery(set<ShardId>& shardIds, const BSONObj& query) const { - CanonicalQuery* canonicalQuery = NULL; - Status status = CanonicalQuery::canonicalize( - _ns, - query, - &canonicalQuery, - WhereCallbackNoop()); - - std::unique_ptr<CanonicalQuery> canonicalQueryPtr(canonicalQuery); - - uassert(status.code(), status.reason(), status.isOK()); - - // Query validation - if (QueryPlannerCommon::hasNode(canonicalQuery->root(), MatchExpression::GEO_NEAR)) { - uassert(13501, "use geoNear command rather than $near query", false); - } - - // Transforms query into bounds for each field in the shard key - // for example : - // Key { a: 1, b: 1 }, - // Query { a : { $gte : 1, $lt : 2 }, - // b : { $gte : 3, $lt : 4 } } - // => Bounds { a : [1, 2), b : [3, 4) } - IndexBounds bounds = getIndexBoundsForQuery(_keyPattern.toBSON(), canonicalQuery); - - // Transforms bounds for each shard key field into full shard key ranges - // for example : - // Key { a : 1, b : 1 } - // Bounds { a : [1, 2), b : [3, 4) } - // => Ranges { a : 1, b : 3 } => { a : 2, b : 4 } - BoundList ranges = _keyPattern.flattenBounds(bounds); - - for (BoundList::const_iterator it = ranges.begin(); it != ranges.end(); - ++it) { - - getShardIdsForRange(shardIds, it->first /*min*/, it->second /*max*/); + if (chunk) { + if (chunk->containsKey(shardKey)) { + return chunk; + } - // once we know we need to visit all shards no need to keep looping - if(shardIds.size() == _shardIds.size()) break; - } + log() << chunkMin; + log() << *chunk; + log() << shardKey; - // SERVER-4914 Some clients of getShardIdsForQuery() assume at least one shard will be - // returned. For now, we satisfy that assumption by adding a shard with no matches rather - // than return an empty set of shards. - if (shardIds.empty()) { - massert( 16068, "no chunk ranges available", !_chunkRanges.ranges().empty() ); - shardIds.insert(_chunkRanges.ranges().begin()->second->getShardId()); + reload(); + msgasserted(13141, "Chunk map pointed to incorrect chunk"); } } - void ChunkManager::getShardIdsForRange(set<ShardId>& shardIds, - const BSONObj& min, - const BSONObj& max) const { - - ChunkRangeMap::const_iterator it = _chunkRanges.upper_bound(min); - ChunkRangeMap::const_iterator end = _chunkRanges.upper_bound(max); + msgasserted(8070, + str::stream() << "couldn't find a chunk intersecting: " << shardKey + << " for ns: " << _ns << " at version: " << _version.toString() + << ", number of chunks: " << _chunkMap.size()); +} - massert(13507, - str::stream() << "no chunks found between bounds " << min << " and " << max, - it != _chunkRanges.ranges().end()); +void ChunkManager::getShardIdsForQuery(set<ShardId>& shardIds, const BSONObj& query) const { + CanonicalQuery* canonicalQuery = NULL; + Status status = CanonicalQuery::canonicalize(_ns, query, &canonicalQuery, WhereCallbackNoop()); - if( end != _chunkRanges.ranges().end() ) ++end; + std::unique_ptr<CanonicalQuery> canonicalQueryPtr(canonicalQuery); - for( ; it != end; ++it ){ - shardIds.insert(it->second->getShardId()); + uassert(status.code(), status.reason(), status.isOK()); - // once we know we need to visit all shards no need to keep looping - if (shardIds.size() == _shardIds.size()) break; - } + // Query validation + if (QueryPlannerCommon::hasNode(canonicalQuery->root(), MatchExpression::GEO_NEAR)) { + uassert(13501, "use geoNear command rather than $near query", false); } - void ChunkManager::getAllShardIds(set<ShardId>* all) const { - dassert(all); + // Transforms query into bounds for each field in the shard key + // for example : + // Key { a: 1, b: 1 }, + // Query { a : { $gte : 1, $lt : 2 }, + // b : { $gte : 3, $lt : 4 } } + // => Bounds { a : [1, 2), b : [3, 4) } + IndexBounds bounds = getIndexBoundsForQuery(_keyPattern.toBSON(), canonicalQuery); + + // Transforms bounds for each shard key field into full shard key ranges + // for example : + // Key { a : 1, b : 1 } + // Bounds { a : [1, 2), b : [3, 4) } + // => Ranges { a : 1, b : 3 } => { a : 2, b : 4 } + BoundList ranges = _keyPattern.flattenBounds(bounds); + + for (BoundList::const_iterator it = ranges.begin(); it != ranges.end(); ++it) { + getShardIdsForRange(shardIds, it->first /*min*/, it->second /*max*/); + + // once we know we need to visit all shards no need to keep looping + if (shardIds.size() == _shardIds.size()) + break; + } - all->insert(_shardIds.begin(), _shardIds.end()); + // SERVER-4914 Some clients of getShardIdsForQuery() assume at least one shard will be + // returned. For now, we satisfy that assumption by adding a shard with no matches rather + // than return an empty set of shards. + if (shardIds.empty()) { + massert(16068, "no chunk ranges available", !_chunkRanges.ranges().empty()); + shardIds.insert(_chunkRanges.ranges().begin()->second->getShardId()); } +} - IndexBounds ChunkManager::getIndexBoundsForQuery(const BSONObj& key, const CanonicalQuery* canonicalQuery) { - // $text is not allowed in planning since we don't have text index on mongos. - // - // TODO: Treat $text query as a no-op in planning. So with shard key {a: 1}, - // the query { a: 2, $text: { ... } } will only target to {a: 2}. - if (QueryPlannerCommon::hasNode(canonicalQuery->root(), MatchExpression::TEXT)) { - IndexBounds bounds; - IndexBoundsBuilder::allValuesBounds(key, &bounds); // [minKey, maxKey] - return bounds; - } +void ChunkManager::getShardIdsForRange(set<ShardId>& shardIds, + const BSONObj& min, + const BSONObj& max) const { + ChunkRangeMap::const_iterator it = _chunkRanges.upper_bound(min); + ChunkRangeMap::const_iterator end = _chunkRanges.upper_bound(max); - // Consider shard key as an index - string accessMethod = IndexNames::findPluginName(key); - dassert(accessMethod == IndexNames::BTREE || accessMethod == IndexNames::HASHED); + massert(13507, + str::stream() << "no chunks found between bounds " << min << " and " << max, + it != _chunkRanges.ranges().end()); - // Use query framework to generate index bounds - QueryPlannerParams plannerParams; - // Must use "shard key" index - plannerParams.options = QueryPlannerParams::NO_TABLE_SCAN; - IndexEntry indexEntry(key, accessMethod, false /* multiKey */, false /* sparse */, - false /* unique */, "shardkey", NULL /* filterExpr */, BSONObj()); - plannerParams.indices.push_back(indexEntry); + if (end != _chunkRanges.ranges().end()) + ++end; - OwnedPointerVector<QuerySolution> solutions; - Status status = QueryPlanner::plan(*canonicalQuery, plannerParams, &solutions.mutableVector()); - uassert(status.code(), status.reason(), status.isOK()); + for (; it != end; ++it) { + shardIds.insert(it->second->getShardId()); - IndexBounds bounds; + // once we know we need to visit all shards no need to keep looping + if (shardIds.size() == _shardIds.size()) + break; + } +} - for (vector<QuerySolution*>::const_iterator it = solutions.begin(); - bounds.size() == 0 && it != solutions.end(); it++) { - // Try next solution if we failed to generate index bounds, i.e. bounds.size() == 0 - bounds = collapseQuerySolution((*it)->root.get()); - } +void ChunkManager::getAllShardIds(set<ShardId>* all) const { + dassert(all); - if (bounds.size() == 0) { - // We cannot plan the query without collection scan, so target to all shards. - IndexBoundsBuilder::allValuesBounds(key, &bounds); // [minKey, maxKey] - } + all->insert(_shardIds.begin(), _shardIds.end()); +} + +IndexBounds ChunkManager::getIndexBoundsForQuery(const BSONObj& key, + const CanonicalQuery* canonicalQuery) { + // $text is not allowed in planning since we don't have text index on mongos. + // + // TODO: Treat $text query as a no-op in planning. So with shard key {a: 1}, + // the query { a: 2, $text: { ... } } will only target to {a: 2}. + if (QueryPlannerCommon::hasNode(canonicalQuery->root(), MatchExpression::TEXT)) { + IndexBounds bounds; + IndexBoundsBuilder::allValuesBounds(key, &bounds); // [minKey, maxKey] return bounds; } - IndexBounds ChunkManager::collapseQuerySolution( const QuerySolutionNode* node ) { - if (node->children.size() == 0) { - invariant(node->getType() == STAGE_IXSCAN); + // Consider shard key as an index + string accessMethod = IndexNames::findPluginName(key); + dassert(accessMethod == IndexNames::BTREE || accessMethod == IndexNames::HASHED); + + // Use query framework to generate index bounds + QueryPlannerParams plannerParams; + // Must use "shard key" index + plannerParams.options = QueryPlannerParams::NO_TABLE_SCAN; + IndexEntry indexEntry(key, + accessMethod, + false /* multiKey */, + false /* sparse */, + false /* unique */, + "shardkey", + NULL /* filterExpr */, + BSONObj()); + plannerParams.indices.push_back(indexEntry); + + OwnedPointerVector<QuerySolution> solutions; + Status status = QueryPlanner::plan(*canonicalQuery, plannerParams, &solutions.mutableVector()); + uassert(status.code(), status.reason(), status.isOK()); + + IndexBounds bounds; + + for (vector<QuerySolution*>::const_iterator it = solutions.begin(); + bounds.size() == 0 && it != solutions.end(); + it++) { + // Try next solution if we failed to generate index bounds, i.e. bounds.size() == 0 + bounds = collapseQuerySolution((*it)->root.get()); + } - const IndexScanNode* ixNode = static_cast<const IndexScanNode*>( node ); - return ixNode->bounds; - } + if (bounds.size() == 0) { + // We cannot plan the query without collection scan, so target to all shards. + IndexBoundsBuilder::allValuesBounds(key, &bounds); // [minKey, maxKey] + } + return bounds; +} - if (node->children.size() == 1) { - // e.g. FETCH -> IXSCAN - return collapseQuerySolution( node->children.front() ); - } +IndexBounds ChunkManager::collapseQuerySolution(const QuerySolutionNode* node) { + if (node->children.size() == 0) { + invariant(node->getType() == STAGE_IXSCAN); - // children.size() > 1, assert it's OR / SORT_MERGE. - if ( node->getType() != STAGE_OR && node->getType() != STAGE_SORT_MERGE ) { - // Unexpected node. We should never reach here. - error() << "could not generate index bounds on query solution tree: " << node->toString(); - dassert(false); // We'd like to know this error in testing. + const IndexScanNode* ixNode = static_cast<const IndexScanNode*>(node); + return ixNode->bounds; + } - // Bail out with all shards in production, since this isn't a fatal error. - return IndexBounds(); - } + if (node->children.size() == 1) { + // e.g. FETCH -> IXSCAN + return collapseQuerySolution(node->children.front()); + } - IndexBounds bounds; - for ( vector<QuerySolutionNode*>::const_iterator it = node->children.begin(); - it != node->children.end(); it++ ) - { - // The first branch under OR - if ( it == node->children.begin() ) { - invariant(bounds.size() == 0); - bounds = collapseQuerySolution( *it ); - if (bounds.size() == 0) { // Got unexpected node in query solution tree - return IndexBounds(); - } - continue; - } + // children.size() > 1, assert it's OR / SORT_MERGE. + if (node->getType() != STAGE_OR && node->getType() != STAGE_SORT_MERGE) { + // Unexpected node. We should never reach here. + error() << "could not generate index bounds on query solution tree: " << node->toString(); + dassert(false); // We'd like to know this error in testing. - IndexBounds childBounds = collapseQuerySolution( *it ); - if (childBounds.size() == 0) { // Got unexpected node in query solution tree - return IndexBounds(); - } + // Bail out with all shards in production, since this isn't a fatal error. + return IndexBounds(); + } - invariant(childBounds.size() == bounds.size()); - for ( size_t i = 0; i < bounds.size(); i++ ) { - bounds.fields[i].intervals.insert( bounds.fields[i].intervals.end(), - childBounds.fields[i].intervals.begin(), - childBounds.fields[i].intervals.end() ); + IndexBounds bounds; + for (vector<QuerySolutionNode*>::const_iterator it = node->children.begin(); + it != node->children.end(); + it++) { + // The first branch under OR + if (it == node->children.begin()) { + invariant(bounds.size() == 0); + bounds = collapseQuerySolution(*it); + if (bounds.size() == 0) { // Got unexpected node in query solution tree + return IndexBounds(); } + continue; } - for ( size_t i = 0; i < bounds.size(); i++ ) { - IndexBoundsBuilder::unionize( &bounds.fields[i] ); + IndexBounds childBounds = collapseQuerySolution(*it); + if (childBounds.size() == 0) { // Got unexpected node in query solution tree + return IndexBounds(); } - return bounds; - } - - bool ChunkManager::compatibleWith(const ChunkManager& other, const string& shardName) const { - // Return true if the shard version is the same in the two chunk managers - // TODO: This doesn't need to be so strong, just major vs - return other.getVersion(shardName).equals(getVersion(shardName)); + invariant(childBounds.size() == bounds.size()); + for (size_t i = 0; i < bounds.size(); i++) { + bounds.fields[i].intervals.insert(bounds.fields[i].intervals.end(), + childBounds.fields[i].intervals.begin(), + childBounds.fields[i].intervals.end()); + } } - ChunkVersion ChunkManager::getVersion(const std::string& shardName) const { - ShardVersionMap::const_iterator i = _shardVersions.find(shardName); - if ( i == _shardVersions.end() ) { - // Shards without explicitly tracked shard versions (meaning they have - // no chunks) always have a version of (0, 0, epoch). Note this is - // *different* from the dropped chunk version of (0, 0, OID(000...)). - // See s/chunk_version.h. - return ChunkVersion( 0, 0, _version.epoch() ); - } - return i->second; + for (size_t i = 0; i < bounds.size(); i++) { + IndexBoundsBuilder::unionize(&bounds.fields[i]); } - ChunkVersion ChunkManager::getVersion() const { - return _version; + return bounds; +} + +bool ChunkManager::compatibleWith(const ChunkManager& other, const string& shardName) const { + // Return true if the shard version is the same in the two chunk managers + // TODO: This doesn't need to be so strong, just major vs + return other.getVersion(shardName).equals(getVersion(shardName)); +} + +ChunkVersion ChunkManager::getVersion(const std::string& shardName) const { + ShardVersionMap::const_iterator i = _shardVersions.find(shardName); + if (i == _shardVersions.end()) { + // Shards without explicitly tracked shard versions (meaning they have + // no chunks) always have a version of (0, 0, epoch). Note this is + // *different* from the dropped chunk version of (0, 0, OID(000...)). + // See s/chunk_version.h. + return ChunkVersion(0, 0, _version.epoch()); } + return i->second; +} - string ChunkManager::toString() const { - StringBuilder sb; - sb << "ChunkManager: " << _ns << " key:" << _keyPattern.toString() << '\n'; +ChunkVersion ChunkManager::getVersion() const { + return _version; +} - for (ChunkMap::const_iterator i = _chunkMap.begin(); i != _chunkMap.end(); ++i) { - sb << "\t" << i->second->toString() << '\n'; - } +string ChunkManager::toString() const { + StringBuilder sb; + sb << "ChunkManager: " << _ns << " key:" << _keyPattern.toString() << '\n'; - return sb.str(); + for (ChunkMap::const_iterator i = _chunkMap.begin(); i != _chunkMap.end(); ++i) { + sb << "\t" << i->second->toString() << '\n'; } + return sb.str(); +} - ChunkRange::ChunkRange(ChunkMap::const_iterator begin, const ChunkMap::const_iterator end) - : _manager(begin->second->getManager()), - _shardId(begin->second->getShardId()), - _min(begin->second->getMin()), - _max(boost::prior(end)->second->getMax()) { - invariant(begin != end); +ChunkRange::ChunkRange(ChunkMap::const_iterator begin, const ChunkMap::const_iterator end) + : _manager(begin->second->getManager()), + _shardId(begin->second->getShardId()), + _min(begin->second->getMin()), + _max(boost::prior(end)->second->getMax()) { + invariant(begin != end); - DEV while (begin != end) { - dassert(begin->second->getManager() == _manager); - dassert(begin->second->getShardId() == _shardId); - ++begin; - } - } - - ChunkRange::ChunkRange(const ChunkRange& min, const ChunkRange& max) - : _manager(min.getManager()), - _shardId(min.getShardId()), - _min(min.getMin()), - _max(max.getMax()) { - - invariant(min.getShardId() == max.getShardId()); - invariant(min.getManager() == max.getManager()); - invariant(min.getMax() == max.getMin()); + DEV while (begin != end) { + dassert(begin->second->getManager() == _manager); + dassert(begin->second->getShardId() == _shardId); + ++begin; } +} - string ChunkRange::toString() const { - StringBuilder sb; - sb << "ChunkRange(min=" << _min << ", max=" << _max - << ", shard=" << _shardId << ")"; +ChunkRange::ChunkRange(const ChunkRange& min, const ChunkRange& max) + : _manager(min.getManager()), + _shardId(min.getShardId()), + _min(min.getMin()), + _max(max.getMax()) { + invariant(min.getShardId() == max.getShardId()); + invariant(min.getManager() == max.getManager()); + invariant(min.getMax() == max.getMin()); +} - return sb.str(); - } +string ChunkRange::toString() const { + StringBuilder sb; + sb << "ChunkRange(min=" << _min << ", max=" << _max << ", shard=" << _shardId << ")"; + return sb.str(); +} - void ChunkRangeManager::assertValid() const { - if (_ranges.empty()) - return; - try { - // No Nulls - for (ChunkRangeMap::const_iterator it=_ranges.begin(), end=_ranges.end(); it != end; ++it) { - verify(it->second); - } +void ChunkRangeManager::assertValid() const { + if (_ranges.empty()) + return; - // Check endpoints - verify(allOfType(MinKey, _ranges.begin()->second->getMin())); - verify(allOfType(MaxKey, boost::prior(_ranges.end())->second->getMax())); + try { + // No Nulls + for (ChunkRangeMap::const_iterator it = _ranges.begin(), end = _ranges.end(); it != end; + ++it) { + verify(it->second); + } - // Make sure there are no gaps or overlaps - for (ChunkRangeMap::const_iterator it=boost::next(_ranges.begin()), end=_ranges.end(); it != end; ++it) { - ChunkRangeMap::const_iterator last = boost::prior(it); - verify(it->second->getMin() == last->second->getMax()); - } + // Check endpoints + verify(allOfType(MinKey, _ranges.begin()->second->getMin())); + verify(allOfType(MaxKey, boost::prior(_ranges.end())->second->getMax())); - // Check Map keys - for (ChunkRangeMap::const_iterator it=_ranges.begin(), end=_ranges.end(); it != end; ++it) { - verify(it->first == it->second->getMax()); - } + // Make sure there are no gaps or overlaps + for (ChunkRangeMap::const_iterator it = boost::next(_ranges.begin()), end = _ranges.end(); + it != end; + ++it) { + ChunkRangeMap::const_iterator last = boost::prior(it); + verify(it->second->getMin() == last->second->getMax()); + } - // Make sure we match the original chunks - const ChunkMap chunks = _ranges.begin()->second->getManager()->_chunkMap; - for ( ChunkMap::const_iterator i=chunks.begin(); i!=chunks.end(); ++i ) { - const ChunkPtr chunk = i->second; + // Check Map keys + for (ChunkRangeMap::const_iterator it = _ranges.begin(), end = _ranges.end(); it != end; + ++it) { + verify(it->first == it->second->getMax()); + } - ChunkRangeMap::const_iterator min = _ranges.upper_bound(chunk->getMin()); - ChunkRangeMap::const_iterator max = _ranges.lower_bound(chunk->getMax()); + // Make sure we match the original chunks + const ChunkMap chunks = _ranges.begin()->second->getManager()->_chunkMap; + for (ChunkMap::const_iterator i = chunks.begin(); i != chunks.end(); ++i) { + const ChunkPtr chunk = i->second; - verify(min != _ranges.end()); - verify(max != _ranges.end()); - verify(min == max); - verify(min->second->getShardId() == chunk->getShardId()); - verify(min->second->containsKey( chunk->getMin() )); - verify(min->second->containsKey( chunk->getMax() ) || (min->second->getMax() == chunk->getMax())); - } + ChunkRangeMap::const_iterator min = _ranges.upper_bound(chunk->getMin()); + ChunkRangeMap::const_iterator max = _ranges.lower_bound(chunk->getMax()); + verify(min != _ranges.end()); + verify(max != _ranges.end()); + verify(min == max); + verify(min->second->getShardId() == chunk->getShardId()); + verify(min->second->containsKey(chunk->getMin())); + verify(min->second->containsKey(chunk->getMax()) || + (min->second->getMax() == chunk->getMax())); } - catch (...) { - error() << "\t invalid ChunkRangeMap! printing ranges:"; - for (ChunkRangeMap::const_iterator it = _ranges.begin(), end = _ranges.end(); it != end; ++it) { - log() << it->first << ": " << it->second->toString(); - } + } catch (...) { + error() << "\t invalid ChunkRangeMap! printing ranges:"; - throw; + for (ChunkRangeMap::const_iterator it = _ranges.begin(), end = _ranges.end(); it != end; + ++it) { + log() << it->first << ": " << it->second->toString(); } + + throw; } +} - void ChunkRangeManager::reloadAll(const ChunkMap& chunks) { - _ranges.clear(); - _insertRange(chunks.begin(), chunks.end()); +void ChunkRangeManager::reloadAll(const ChunkMap& chunks) { + _ranges.clear(); + _insertRange(chunks.begin(), chunks.end()); - DEV assertValid(); - } + DEV assertValid(); +} - void ChunkRangeManager::_insertRange(ChunkMap::const_iterator begin, const ChunkMap::const_iterator end) { - while (begin != end) { - ChunkMap::const_iterator first = begin; - ShardId shardId = first->second->getShardId(); - while (begin != end && (begin->second->getShardId() == shardId)) - ++begin; +void ChunkRangeManager::_insertRange(ChunkMap::const_iterator begin, + const ChunkMap::const_iterator end) { + while (begin != end) { + ChunkMap::const_iterator first = begin; + ShardId shardId = first->second->getShardId(); + while (begin != end && (begin->second->getShardId() == shardId)) + ++begin; - shared_ptr<ChunkRange> cr (new ChunkRange(first, begin)); - _ranges[cr->getMax()] = cr; - } + shared_ptr<ChunkRange> cr(new ChunkRange(first, begin)); + _ranges[cr->getMax()] = cr; } +} - int ChunkManager::getCurrentDesiredChunkSize() const { - // split faster in early chunks helps spread out an initial load better - const int minChunkSize = 1 << 20; // 1 MBytes +int ChunkManager::getCurrentDesiredChunkSize() const { + // split faster in early chunks helps spread out an initial load better + const int minChunkSize = 1 << 20; // 1 MBytes - int splitThreshold = Chunk::MaxChunkSize; + int splitThreshold = Chunk::MaxChunkSize; - int nc = numChunks(); - - if ( nc <= 1 ) { - return 1024; - } - else if ( nc < 3 ) { - return minChunkSize / 2; - } - else if ( nc < 10 ) { - splitThreshold = max( splitThreshold / 4 , minChunkSize ); - } - else if ( nc < 20 ) { - splitThreshold = max( splitThreshold / 2 , minChunkSize ); - } + int nc = numChunks(); - return splitThreshold; + if (nc <= 1) { + return 1024; + } else if (nc < 3) { + return minChunkSize / 2; + } else if (nc < 10) { + splitThreshold = max(splitThreshold / 4, minChunkSize); + } else if (nc < 20) { + splitThreshold = max(splitThreshold / 2, minChunkSize); } -} // namespace mongo + return splitThreshold; +} + +} // namespace mongo diff --git a/src/mongo/s/chunk_manager.h b/src/mongo/s/chunk_manager.h index e3b61a30b62..c2c7a48286d 100644 --- a/src/mongo/s/chunk_manager.h +++ b/src/mongo/s/chunk_manager.h @@ -38,240 +38,270 @@ namespace mongo { - class CanonicalQuery; - class ChunkManager; - class CollectionType; - struct QuerySolutionNode; - - typedef std::shared_ptr<ChunkManager> ChunkManagerPtr; - - // The key for the map is max for each Chunk or ChunkRange - typedef std::map<BSONObj, std::shared_ptr<Chunk>, BSONObjCmp> ChunkMap; - - class ChunkRange { - public: - ChunkRange(ChunkMap::const_iterator begin, const ChunkMap::const_iterator end); - - // Merge min and max (must be adjacent ranges) - ChunkRange(const ChunkRange& min, const ChunkRange& max); - - const ChunkManager* getManager() const { return _manager; } - ShardId getShardId() const { return _shardId; } - - const BSONObj& getMin() const { return _min; } - const BSONObj& getMax() const { return _max; } - - // clones of Chunk methods - // Returns true if this ChunkRange contains the given shard key, and false otherwise - // - // Note: this function takes an extracted *key*, not an original document - // (the point may be computed by, say, hashing a given field or projecting - // to a subset of fields). - bool containsKey( const BSONObj& shardKey ) const; - - std::string toString() const; - - private: - const ChunkManager* _manager; - const ShardId _shardId; - const BSONObj _min; - const BSONObj _max; - }; - - typedef std::map<BSONObj, std::shared_ptr<ChunkRange>, BSONObjCmp> ChunkRangeMap; - - - class ChunkRangeManager { +class CanonicalQuery; +class ChunkManager; +class CollectionType; +struct QuerySolutionNode; + +typedef std::shared_ptr<ChunkManager> ChunkManagerPtr; + +// The key for the map is max for each Chunk or ChunkRange +typedef std::map<BSONObj, std::shared_ptr<Chunk>, BSONObjCmp> ChunkMap; + +class ChunkRange { +public: + ChunkRange(ChunkMap::const_iterator begin, const ChunkMap::const_iterator end); + + // Merge min and max (must be adjacent ranges) + ChunkRange(const ChunkRange& min, const ChunkRange& max); + + const ChunkManager* getManager() const { + return _manager; + } + ShardId getShardId() const { + return _shardId; + } + + const BSONObj& getMin() const { + return _min; + } + const BSONObj& getMax() const { + return _max; + } + + // clones of Chunk methods + // Returns true if this ChunkRange contains the given shard key, and false otherwise + // + // Note: this function takes an extracted *key*, not an original document + // (the point may be computed by, say, hashing a given field or projecting + // to a subset of fields). + bool containsKey(const BSONObj& shardKey) const; + + std::string toString() const; + +private: + const ChunkManager* _manager; + const ShardId _shardId; + const BSONObj _min; + const BSONObj _max; +}; + +typedef std::map<BSONObj, std::shared_ptr<ChunkRange>, BSONObjCmp> ChunkRangeMap; + + +class ChunkRangeManager { +public: + const ChunkRangeMap& ranges() const { + return _ranges; + } + + void clear() { + _ranges.clear(); + } + + void reloadAll(const ChunkMap& chunks); + + // Slow operation -- wrap with DEV + void assertValid() const; + + ChunkRangeMap::const_iterator upper_bound(const BSONObj& o) const { + return _ranges.upper_bound(o); + } + ChunkRangeMap::const_iterator lower_bound(const BSONObj& o) const { + return _ranges.lower_bound(o); + } + +private: + // assumes nothing in this range exists in _ranges + void _insertRange(ChunkMap::const_iterator begin, const ChunkMap::const_iterator end); + + ChunkRangeMap _ranges; +}; + + +/* config.sharding + { ns: 'alleyinsider.fs.chunks' , + key: { ts : 1 } , + shards: [ { min: 1, max: 100, server: a } , { min: 101, max: 200 , server : b } ] + } +*/ +class ChunkManager { +public: + typedef std::map<std::string, ChunkVersion> ShardVersionMap; + + // Loads a new chunk manager from a collection document + explicit ChunkManager(const CollectionType& coll); + + // Creates an empty chunk manager for the namespace + ChunkManager(const std::string& ns, const ShardKeyPattern& pattern, bool unique); + + const std::string& getns() const { + return _ns; + } + const ShardKeyPattern& getShardKeyPattern() const { + return _keyPattern; + } + bool isUnique() const { + return _unique; + } + + /** + * this is just an increasing number of how many ChunkManagers we have so we know if something has been updated + */ + unsigned long long getSequenceNumber() const { + return _sequenceNumber; + } + + // + // After constructor is invoked, we need to call loadExistingRanges. If this is a new + // sharded collection, we can call createFirstChunks first. + // + + // Creates new chunks based on info in chunk manager + void createFirstChunks(const ShardId& primaryShardId, + const std::vector<BSONObj>* initPoints, + const std::set<ShardId>* initShardIds); + + // Loads existing ranges based on info in chunk manager + void loadExistingRanges(const ChunkManager* oldManager); + + + // Helpers for load + void calcInitSplitsAndShards(const ShardId& primaryShardId, + const std::vector<BSONObj>* initPoints, + const std::set<ShardId>* initShardIds, + std::vector<BSONObj>* splitPoints, + std::vector<ShardId>* shardIds) const; + + // + // Methods to use once loaded / created + // + + int numChunks() const { + return _chunkMap.size(); + } + + /** + * Given a key that has been extracted from a document, returns the + * chunk that contains that key. + * + * For instance, to locate the chunk for document {a : "foo" , b : "bar"} + * when the shard key is {a : "hashed"}, you can call + * findIntersectingChunk() on {a : hash("foo") } + */ + ChunkPtr findIntersectingChunk(const BSONObj& shardKey) const; + + void getShardIdsForQuery(std::set<ShardId>& shardIds, const BSONObj& query) const; + void getAllShardIds(std::set<ShardId>* all) const; + /** @param shardIds set to the shard ids for shards + * covered by the interval [min, max], see SERVER-4791 + */ + void getShardIdsForRange(std::set<ShardId>& shardIds, + const BSONObj& min, + const BSONObj& max) const; + + // Transforms query into bounds for each field in the shard key + // for example : + // Key { a: 1, b: 1 }, + // Query { a : { $gte : 1, $lt : 2 }, + // b : { $gte : 3, $lt : 4 } } + // => Bounds { a : [1, 2), b : [3, 4) } + static IndexBounds getIndexBoundsForQuery(const BSONObj& key, + const CanonicalQuery* canonicalQuery); + + // Collapse query solution tree. + // + // If it has OR node, the result could be a superset of the index bounds generated. + // Since to give a single IndexBounds, this gives the union of bounds on each field. + // for example: + // OR: { a: (0, 1), b: (0, 1) }, + // { a: (2, 3), b: (2, 3) } + // => { a: (0, 1), (2, 3), b: (0, 1), (2, 3) } + static IndexBounds collapseQuerySolution(const QuerySolutionNode* node); + + const ChunkMap& getChunkMap() const { + return _chunkMap; + } + + /** + * Returns true if, for this shard, the chunks are identical in both chunk managers + */ + bool compatibleWith(const ChunkManager& other, const std::string& shard) const; + + std::string toString() const; + + ChunkVersion getVersion(const std::string& shardName) const; + ChunkVersion getVersion() const; + + void _printChunks() const; + + int getCurrentDesiredChunkSize() const; + + std::shared_ptr<ChunkManager> reload(bool force = true) const; // doesn't modify self! + +private: + // returns true if load was consistent + bool _load(ChunkMap& chunks, + std::set<ShardId>& shardIds, + ShardVersionMap* shardVersions, + const ChunkManager* oldManager); + + + // All members should be const for thread-safety + const std::string _ns; + const ShardKeyPattern _keyPattern; + const bool _unique; + + // The shard versioning mechanism hinges on keeping track of the number of times we reload + // ChunkManagers. Increasing this number here will prompt checkShardVersion to refresh the + // connection-level versions to the most up to date value. + const unsigned long long _sequenceNumber; + + ChunkMap _chunkMap; + ChunkRangeManager _chunkRanges; + + std::set<ShardId> _shardIds; + + // Max known version per shard + ShardVersionMap _shardVersions; + + // Max version across all chunks + ChunkVersion _version; + + // + // Split Heuristic info + // + class SplitHeuristics { public: - const ChunkRangeMap& ranges() const { return _ranges; } - - void clear() { _ranges.clear(); } - - void reloadAll(const ChunkMap& chunks); - - // Slow operation -- wrap with DEV - void assertValid() const; - - ChunkRangeMap::const_iterator upper_bound(const BSONObj& o) const { return _ranges.upper_bound(o); } - ChunkRangeMap::const_iterator lower_bound(const BSONObj& o) const { return _ranges.lower_bound(o); } - - private: - // assumes nothing in this range exists in _ranges - void _insertRange(ChunkMap::const_iterator begin, const ChunkMap::const_iterator end); - - ChunkRangeMap _ranges; + SplitHeuristics() : _splitTickets(maxParallelSplits) {} + + TicketHolder _splitTickets; + + // Test whether we should split once data * splitTestFactor > chunkSize (approximately) + static const int splitTestFactor = 5; + // Maximum number of parallel threads requesting a split + static const int maxParallelSplits = 5; + + // The idea here is that we're over-aggressive on split testing by a factor of + // splitTestFactor, so we can safely wait until we get to splitTestFactor invalid splits + // before changing. Unfortunately, we also potentially over-request the splits by a + // factor of maxParallelSplits, but since the factors are identical it works out + // (for now) for parallel or sequential oversplitting. + // TODO: Make splitting a separate thread with notifications? + static const int staleMinorReloadThreshold = maxParallelSplits; }; + mutable SplitHeuristics _splitHeuristics; - /* config.sharding - { ns: 'alleyinsider.fs.chunks' , - key: { ts : 1 } , - shards: [ { min: 1, max: 100, server: a } , { min: 101, max: 200 , server : b } ] - } - */ - class ChunkManager { - public: - typedef std::map<std::string, ChunkVersion> ShardVersionMap; - - // Loads a new chunk manager from a collection document - explicit ChunkManager(const CollectionType& coll); - - // Creates an empty chunk manager for the namespace - ChunkManager( const std::string& ns, const ShardKeyPattern& pattern, bool unique ); - - const std::string& getns() const { return _ns; } - const ShardKeyPattern& getShardKeyPattern() const { return _keyPattern; } - bool isUnique() const { return _unique; } - - /** - * this is just an increasing number of how many ChunkManagers we have so we know if something has been updated - */ - unsigned long long getSequenceNumber() const { return _sequenceNumber; } - - // - // After constructor is invoked, we need to call loadExistingRanges. If this is a new - // sharded collection, we can call createFirstChunks first. - // - - // Creates new chunks based on info in chunk manager - void createFirstChunks(const ShardId& primaryShardId, - const std::vector<BSONObj>* initPoints, - const std::set<ShardId>* initShardIds); + // + // End split heuristics + // - // Loads existing ranges based on info in chunk manager - void loadExistingRanges(const ChunkManager* oldManager); - - - // Helpers for load - void calcInitSplitsAndShards(const ShardId& primaryShardId, - const std::vector<BSONObj>* initPoints, - const std::set<ShardId>* initShardIds, - std::vector<BSONObj>* splitPoints, - std::vector<ShardId>* shardIds) const; - - // - // Methods to use once loaded / created - // - - int numChunks() const { return _chunkMap.size(); } - - /** - * Given a key that has been extracted from a document, returns the - * chunk that contains that key. - * - * For instance, to locate the chunk for document {a : "foo" , b : "bar"} - * when the shard key is {a : "hashed"}, you can call - * findIntersectingChunk() on {a : hash("foo") } - */ - ChunkPtr findIntersectingChunk( const BSONObj& shardKey ) const; - - void getShardIdsForQuery(std::set<ShardId>& shardIds, const BSONObj& query) const; - void getAllShardIds(std::set<ShardId>* all) const; - /** @param shardIds set to the shard ids for shards - * covered by the interval [min, max], see SERVER-4791 - */ - void getShardIdsForRange(std::set<ShardId>& shardIds, const BSONObj& min, const BSONObj& max) const; - - // Transforms query into bounds for each field in the shard key - // for example : - // Key { a: 1, b: 1 }, - // Query { a : { $gte : 1, $lt : 2 }, - // b : { $gte : 3, $lt : 4 } } - // => Bounds { a : [1, 2), b : [3, 4) } - static IndexBounds getIndexBoundsForQuery(const BSONObj& key, const CanonicalQuery* canonicalQuery); - - // Collapse query solution tree. - // - // If it has OR node, the result could be a superset of the index bounds generated. - // Since to give a single IndexBounds, this gives the union of bounds on each field. - // for example: - // OR: { a: (0, 1), b: (0, 1) }, - // { a: (2, 3), b: (2, 3) } - // => { a: (0, 1), (2, 3), b: (0, 1), (2, 3) } - static IndexBounds collapseQuerySolution( const QuerySolutionNode* node ); - - const ChunkMap& getChunkMap() const { return _chunkMap; } - - /** - * Returns true if, for this shard, the chunks are identical in both chunk managers - */ - bool compatibleWith(const ChunkManager& other, const std::string& shard) const; - - std::string toString() const; - - ChunkVersion getVersion(const std::string& shardName) const; - ChunkVersion getVersion() const; - - void _printChunks() const; - - int getCurrentDesiredChunkSize() const; - - std::shared_ptr<ChunkManager> reload(bool force = true) const; // doesn't modify self! - - private: - // returns true if load was consistent - bool _load(ChunkMap& chunks, - std::set<ShardId>& shardIds, - ShardVersionMap* shardVersions, - const ChunkManager* oldManager); - - - // All members should be const for thread-safety - const std::string _ns; - const ShardKeyPattern _keyPattern; - const bool _unique; - - // The shard versioning mechanism hinges on keeping track of the number of times we reload - // ChunkManagers. Increasing this number here will prompt checkShardVersion to refresh the - // connection-level versions to the most up to date value. - const unsigned long long _sequenceNumber; - - ChunkMap _chunkMap; - ChunkRangeManager _chunkRanges; - - std::set<ShardId> _shardIds; - - // Max known version per shard - ShardVersionMap _shardVersions; - - // Max version across all chunks - ChunkVersion _version; - - // - // Split Heuristic info - // - class SplitHeuristics { - public: - SplitHeuristics() : _splitTickets(maxParallelSplits) { - } - - TicketHolder _splitTickets; - - // Test whether we should split once data * splitTestFactor > chunkSize (approximately) - static const int splitTestFactor = 5; - // Maximum number of parallel threads requesting a split - static const int maxParallelSplits = 5; - - // The idea here is that we're over-aggressive on split testing by a factor of - // splitTestFactor, so we can safely wait until we get to splitTestFactor invalid splits - // before changing. Unfortunately, we also potentially over-request the splits by a - // factor of maxParallelSplits, but since the factors are identical it works out - // (for now) for parallel or sequential oversplitting. - // TODO: Make splitting a separate thread with notifications? - static const int staleMinorReloadThreshold = maxParallelSplits; - }; - - mutable SplitHeuristics _splitHeuristics; - - // - // End split heuristics - // - - friend class Chunk; - friend class ChunkRangeManager; // only needed for CRM::assertValid() - static AtomicUInt32 NextSequenceNumber; + friend class Chunk; + friend class ChunkRangeManager; // only needed for CRM::assertValid() + static AtomicUInt32 NextSequenceNumber; - friend class TestableChunkManager; - }; + friend class TestableChunkManager; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/chunk_manager_targeter.cpp b/src/mongo/s/chunk_manager_targeter.cpp index bac563fc027..446d4a5fafd 100644 --- a/src/mongo/s/chunk_manager_targeter.cpp +++ b/src/mongo/s/chunk_manager_targeter.cpp @@ -43,738 +43,678 @@ namespace mongo { - using std::shared_ptr; - using mongoutils::str::stream; - using std::map; - using std::set; - using std::string; - using std::vector; +using std::shared_ptr; +using mongoutils::str::stream; +using std::map; +using std::set; +using std::string; +using std::vector; namespace { - enum UpdateType { - UpdateType_Replacement, UpdateType_OpStyle, UpdateType_Unknown - }; - - enum CompareResult { - CompareResult_Unknown, CompareResult_GTE, CompareResult_LT - }; - - const ShardKeyPattern virtualIdShardKey(BSON("_id" << 1)); - - // To match legacy reload behavior, we have to backoff on config reload per-thread - // TODO: Centralize this behavior better by refactoring config reload in mongos - boost::thread_specific_ptr<Backoff> perThreadBackoff; - const int maxWaitMillis = 500; - - /** - * There are two styles of update expressions: - * - * Replacement style: coll.update({ x : 1 }, { y : 2 }) - * OpStyle: coll.update({ x : 1 }, { $set : { y : 2 } }) - */ - UpdateType getUpdateExprType(const BSONObj& updateExpr) { - // Empty update is replacement-style, by default - if (updateExpr.isEmpty()) { - return UpdateType_Replacement; - } +enum UpdateType { UpdateType_Replacement, UpdateType_OpStyle, UpdateType_Unknown }; - UpdateType updateType = UpdateType_Unknown; +enum CompareResult { CompareResult_Unknown, CompareResult_GTE, CompareResult_LT }; - BSONObjIterator it(updateExpr); - while (it.more()) { - BSONElement next = it.next(); +const ShardKeyPattern virtualIdShardKey(BSON("_id" << 1)); - if (next.fieldName()[0] == '$') { - if (updateType == UpdateType_Unknown) { - updateType = UpdateType_OpStyle; - } - else if (updateType == UpdateType_Replacement) { - return UpdateType_Unknown; - } - } - else { - if (updateType == UpdateType_Unknown) { - updateType = UpdateType_Replacement; - } - else if (updateType == UpdateType_OpStyle) { - return UpdateType_Unknown; - } - } - } +// To match legacy reload behavior, we have to backoff on config reload per-thread +// TODO: Centralize this behavior better by refactoring config reload in mongos +boost::thread_specific_ptr<Backoff> perThreadBackoff; +const int maxWaitMillis = 500; - return updateType; - } - - /** - * This returns "does the query have an _id field" and "is the _id field querying for a direct - * value like _id : 3 and not _id : { $gt : 3 }" - * - * Ex: { _id : 1 } => true - * { foo : <anything>, _id : 1 } => true - * { _id : { $lt : 30 } } => false - * { foo : <anything> } => false - */ - bool isExactIdQuery(const BSONObj& query) { - StatusWith<BSONObj> status = virtualIdShardKey.extractShardKeyFromQuery(query); - if (!status.isOK()) { - return false; - } - - return !status.getValue()["_id"].eoo(); +/** + * There are two styles of update expressions: + * + * Replacement style: coll.update({ x : 1 }, { y : 2 }) + * OpStyle: coll.update({ x : 1 }, { $set : { y : 2 } }) + */ +UpdateType getUpdateExprType(const BSONObj& updateExpr) { + // Empty update is replacement-style, by default + if (updateExpr.isEmpty()) { + return UpdateType_Replacement; } - void refreshBackoff() { - if (!perThreadBackoff.get()) { - perThreadBackoff.reset(new Backoff(maxWaitMillis, maxWaitMillis * 2)); - } - - perThreadBackoff.get()->nextSleepMillis(); - } + UpdateType updateType = UpdateType_Unknown; + BSONObjIterator it(updateExpr); + while (it.more()) { + BSONElement next = it.next(); - // - // Utilities to compare shard versions - // - - /** - * Returns the relationship of two shard versions. Shard versions of a collection that has not - * been dropped and recreated and where there is at least one chunk on a shard are comparable, - * otherwise the result is ambiguous. - */ - CompareResult compareShardVersions(const ChunkVersion& shardVersionA, - const ChunkVersion& shardVersionB) { - - // Collection may have been dropped - if (!shardVersionA.hasEqualEpoch(shardVersionB)) { - return CompareResult_Unknown; - } - - // Zero shard versions are only comparable to themselves - if (!shardVersionA.isSet() || !shardVersionB.isSet()) { - // If both are zero... - if (!shardVersionA.isSet() && !shardVersionB.isSet()) { - return CompareResult_GTE; + if (next.fieldName()[0] == '$') { + if (updateType == UpdateType_Unknown) { + updateType = UpdateType_OpStyle; + } else if (updateType == UpdateType_Replacement) { + return UpdateType_Unknown; + } + } else { + if (updateType == UpdateType_Unknown) { + updateType = UpdateType_Replacement; + } else if (updateType == UpdateType_OpStyle) { + return UpdateType_Unknown; } - - return CompareResult_Unknown; } + } - if (shardVersionA < shardVersionB) { - return CompareResult_LT; - } + return updateType; +} - else return CompareResult_GTE; +/** + * This returns "does the query have an _id field" and "is the _id field querying for a direct + * value like _id : 3 and not _id : { $gt : 3 }" + * + * Ex: { _id : 1 } => true + * { foo : <anything>, _id : 1 } => true + * { _id : { $lt : 30 } } => false + * { foo : <anything> } => false + */ +bool isExactIdQuery(const BSONObj& query) { + StatusWith<BSONObj> status = virtualIdShardKey.extractShardKeyFromQuery(query); + if (!status.isOK()) { + return false; } - ChunkVersion getShardVersion(StringData shardName, - const ChunkManager* manager, - const Shard* primary) { + return !status.getValue()["_id"].eoo(); +} - dassert(!(manager && primary)); - dassert(manager || primary); +void refreshBackoff() { + if (!perThreadBackoff.get()) { + perThreadBackoff.reset(new Backoff(maxWaitMillis, maxWaitMillis * 2)); + } - if (primary) { - return ChunkVersion::UNSHARDED(); - } + perThreadBackoff.get()->nextSleepMillis(); +} - return manager->getVersion(shardName.toString()); - } - - /** - * Returns the relationship between two maps of shard versions. As above, these maps are often - * comparable when the collection has not been dropped and there is at least one chunk on the - * shards. If any versions in the maps are not comparable, the result is _Unknown. - * - * If any versions in the first map (cached) are _LT the versions in the second map (remote), - * the first (cached) versions are _LT the second (remote) versions. - * - * Note that the signature here is weird since our cached map of chunk versions is stored in a - * ChunkManager or is implicit in the primary shard of the collection. - */ - CompareResult compareAllShardVersions(const ChunkManager* cachedChunkManager, - const Shard* cachedPrimary, - const map<string, ChunkVersion>& remoteShardVersions) { - - CompareResult finalResult = CompareResult_GTE; - - for (map<string, ChunkVersion>::const_iterator it = remoteShardVersions.begin(); - it != remoteShardVersions.end(); - ++it) { - - // Get the remote and cached version for the next shard - const string& shardName = it->first; - const ChunkVersion& remoteShardVersion = it->second; - - ChunkVersion cachedShardVersion; - - try { - // Throws b/c shard constructor throws - cachedShardVersion = getShardVersion(shardName, - cachedChunkManager, - cachedPrimary); - } - catch (const DBException& ex) { - warning() << "could not lookup shard " << shardName - << " in local cache, shard metadata may have changed" - << " or be unavailable" << causedBy(ex); - return CompareResult_Unknown; - } +// +// Utilities to compare shard versions +// - // Compare the remote and cached versions - CompareResult result = compareShardVersions(cachedShardVersion, remoteShardVersion); - - if (result == CompareResult_Unknown) return result; - if (result == CompareResult_LT) finalResult = CompareResult_LT; +/** + * Returns the relationship of two shard versions. Shard versions of a collection that has not + * been dropped and recreated and where there is at least one chunk on a shard are comparable, + * otherwise the result is ambiguous. + */ +CompareResult compareShardVersions(const ChunkVersion& shardVersionA, + const ChunkVersion& shardVersionB) { + // Collection may have been dropped + if (!shardVersionA.hasEqualEpoch(shardVersionB)) { + return CompareResult_Unknown; + } - // Note that we keep going after _LT b/c there could be more _Unknowns. + // Zero shard versions are only comparable to themselves + if (!shardVersionA.isSet() || !shardVersionB.isSet()) { + // If both are zero... + if (!shardVersionA.isSet() && !shardVersionB.isSet()) { + return CompareResult_GTE; } - return finalResult; + return CompareResult_Unknown; } - /** - * Whether or not the manager/primary pair is different from the other manager/primary pair. - */ - bool isMetadataDifferent(const ChunkManagerPtr& managerA, - const ShardPtr& primaryA, - const ChunkManagerPtr& managerB, - const ShardPtr& primaryB) { + if (shardVersionA < shardVersionB) { + return CompareResult_LT; + } - if ((managerA && !managerB) || (!managerA && managerB) || (primaryA && !primaryB) || (!primaryA && primaryB)) return true; + else + return CompareResult_GTE; +} - if (managerA) { - return !managerA->getVersion().isStrictlyEqualTo(managerB->getVersion()); - } +ChunkVersion getShardVersion(StringData shardName, + const ChunkManager* manager, + const Shard* primary) { + dassert(!(manager && primary)); + dassert(manager || primary); - dassert(NULL != primaryA.get()); - return primaryA->getId() != primaryB->getId(); + if (primary) { + return ChunkVersion::UNSHARDED(); } - /** - * Whether or not the manager/primary pair was changed or refreshed from a previous version - * of the metadata. - */ - bool wasMetadataRefreshed(const ChunkManagerPtr& managerA, - const ShardPtr& primaryA, - const ChunkManagerPtr& managerB, - const ShardPtr& primaryB) { + return manager->getVersion(shardName.toString()); +} - if (isMetadataDifferent(managerA, primaryA, managerB, primaryB)) - return true; +/** + * Returns the relationship between two maps of shard versions. As above, these maps are often + * comparable when the collection has not been dropped and there is at least one chunk on the + * shards. If any versions in the maps are not comparable, the result is _Unknown. + * + * If any versions in the first map (cached) are _LT the versions in the second map (remote), + * the first (cached) versions are _LT the second (remote) versions. + * + * Note that the signature here is weird since our cached map of chunk versions is stored in a + * ChunkManager or is implicit in the primary shard of the collection. + */ +CompareResult compareAllShardVersions(const ChunkManager* cachedChunkManager, + const Shard* cachedPrimary, + const map<string, ChunkVersion>& remoteShardVersions) { + CompareResult finalResult = CompareResult_GTE; + + for (map<string, ChunkVersion>::const_iterator it = remoteShardVersions.begin(); + it != remoteShardVersions.end(); + ++it) { + // Get the remote and cached version for the next shard + const string& shardName = it->first; + const ChunkVersion& remoteShardVersion = it->second; + + ChunkVersion cachedShardVersion; + + try { + // Throws b/c shard constructor throws + cachedShardVersion = getShardVersion(shardName, cachedChunkManager, cachedPrimary); + } catch (const DBException& ex) { + warning() << "could not lookup shard " << shardName + << " in local cache, shard metadata may have changed" + << " or be unavailable" << causedBy(ex); - if (managerA) { - dassert(managerB.get()); // otherwise metadata would be different - return managerA->getSequenceNumber() != managerB->getSequenceNumber(); + return CompareResult_Unknown; } - return false; - } - -} // namespace + // Compare the remote and cached versions + CompareResult result = compareShardVersions(cachedShardVersion, remoteShardVersion); - ChunkManagerTargeter::ChunkManagerTargeter(const NamespaceString& nss) - : _nss(nss), - _needsTargetingRefresh(false) { + if (result == CompareResult_Unknown) + return result; + if (result == CompareResult_LT) + finalResult = CompareResult_LT; + // Note that we keep going after _LT b/c there could be more _Unknowns. } - Status ChunkManagerTargeter::init() { - auto status = grid.implicitCreateDb(_nss.db().toString()); - if (!status.isOK()) { - return status.getStatus(); - } - - shared_ptr<DBConfig> config = status.getValue(); - config->getChunkManagerOrPrimary(_nss.ns(), _manager, _primary); - - return Status::OK(); - } + return finalResult; +} - const NamespaceString& ChunkManagerTargeter::getNS() const { - return _nss; +/** + * Whether or not the manager/primary pair is different from the other manager/primary pair. + */ +bool isMetadataDifferent(const ChunkManagerPtr& managerA, + const ShardPtr& primaryA, + const ChunkManagerPtr& managerB, + const ShardPtr& primaryB) { + if ((managerA && !managerB) || (!managerA && managerB) || (primaryA && !primaryB) || + (!primaryA && primaryB)) + return true; + + if (managerA) { + return !managerA->getVersion().isStrictlyEqualTo(managerB->getVersion()); } - Status ChunkManagerTargeter::targetInsert( const BSONObj& doc, - ShardEndpoint** endpoint ) const { + dassert(NULL != primaryA.get()); + return primaryA->getId() != primaryB->getId(); +} - BSONObj shardKey; +/** +* Whether or not the manager/primary pair was changed or refreshed from a previous version +* of the metadata. +*/ +bool wasMetadataRefreshed(const ChunkManagerPtr& managerA, + const ShardPtr& primaryA, + const ChunkManagerPtr& managerB, + const ShardPtr& primaryB) { + if (isMetadataDifferent(managerA, primaryA, managerB, primaryB)) + return true; + + if (managerA) { + dassert(managerB.get()); // otherwise metadata would be different + return managerA->getSequenceNumber() != managerB->getSequenceNumber(); + } - if ( _manager ) { + return false; +} - // - // Sharded collections have the following requirements for targeting: - // - // Inserts must contain the exact shard key. - // +} // namespace - shardKey = _manager->getShardKeyPattern().extractShardKeyFromDoc(doc); +ChunkManagerTargeter::ChunkManagerTargeter(const NamespaceString& nss) + : _nss(nss), _needsTargetingRefresh(false) {} - // Check shard key exists - if (shardKey.isEmpty()) { - return Status(ErrorCodes::ShardKeyNotFound, - stream() << "document " << doc - << " does not contain shard key for pattern " - << _manager->getShardKeyPattern().toString()); - } - - // Check shard key size on insert - Status status = ShardKeyPattern::checkShardKeySize(shardKey); - if (!status.isOK()) - return status; - } +Status ChunkManagerTargeter::init() { + auto status = grid.implicitCreateDb(_nss.db().toString()); + if (!status.isOK()) { + return status.getStatus(); + } - // Target the shard key or database primary - if (!shardKey.isEmpty()) { - return targetShardKey(shardKey, doc.objsize(), endpoint); - } - else { + shared_ptr<DBConfig> config = status.getValue(); + config->getChunkManagerOrPrimary(_nss.ns(), _manager, _primary); - if (!_primary) { - return Status(ErrorCodes::NamespaceNotFound, - str::stream() << "could not target insert in collection " - << getNS().ns() << "; no metadata found"); - } + return Status::OK(); +} - *endpoint = new ShardEndpoint(_primary->getId(), ChunkVersion::UNSHARDED()); - return Status::OK(); - } - } +const NamespaceString& ChunkManagerTargeter::getNS() const { + return _nss; +} - Status ChunkManagerTargeter::targetUpdate( const BatchedUpdateDocument& updateDoc, - vector<ShardEndpoint*>* endpoints ) const { +Status ChunkManagerTargeter::targetInsert(const BSONObj& doc, ShardEndpoint** endpoint) const { + BSONObj shardKey; + if (_manager) { // - // Update targeting may use either the query or the update. This is to support save-style - // updates, of the form: - // - // coll.update({ _id : xxx }, { _id : xxx, shardKey : 1, foo : bar }, { upsert : true }) - // - // Because drivers do not know the shard key, they can't pull the shard key automatically - // into the query doc, and to correctly support upsert we must target a single shard. + // Sharded collections have the following requirements for targeting: // - // The rule is simple - If the update is replacement style (no '$set'), we target using the - // update. If the update is replacement style, we target using the query. + // Inserts must contain the exact shard key. // - // If we have the exact shard key in either the query or replacement doc, we target using - // that extracted key. - // - - BSONObj query = updateDoc.getQuery(); - BSONObj updateExpr = updateDoc.getUpdateExpr(); - UpdateType updateType = getUpdateExprType( updateDoc.getUpdateExpr() ); + shardKey = _manager->getShardKeyPattern().extractShardKeyFromDoc(doc); - if ( updateType == UpdateType_Unknown ) { - return Status( ErrorCodes::UnsupportedFormat, - stream() << "update document " << updateExpr - << " has mixed $operator and non-$operator style fields" ); + // Check shard key exists + if (shardKey.isEmpty()) { + return Status(ErrorCodes::ShardKeyNotFound, + stream() << "document " << doc + << " does not contain shard key for pattern " + << _manager->getShardKeyPattern().toString()); } - BSONObj shardKey; - - if ( _manager ) { - - // - // Sharded collections have the following futher requirements for targeting: - // - // Upserts must be targeted exactly by shard key. - // Non-multi updates must be targeted exactly by shard key *or* exact _id. - // - - // Get the shard key - if (updateType == UpdateType_OpStyle) { - - // Target using the query - StatusWith<BSONObj> status = - _manager->getShardKeyPattern().extractShardKeyFromQuery(query); - - // Bad query - if (!status.isOK()) - return status.getStatus(); - - shardKey = status.getValue(); - } - else { - // Target using the replacement document - shardKey = _manager->getShardKeyPattern().extractShardKeyFromDoc(updateExpr); - } + // Check shard key size on insert + Status status = ShardKeyPattern::checkShardKeySize(shardKey); + if (!status.isOK()) + return status; + } - // - // Extra sharded update validation - // + // Target the shard key or database primary + if (!shardKey.isEmpty()) { + return targetShardKey(shardKey, doc.objsize(), endpoint); + } else { + if (!_primary) { + return Status(ErrorCodes::NamespaceNotFound, + str::stream() << "could not target insert in collection " << getNS().ns() + << "; no metadata found"); + } - if (updateDoc.getUpsert()) { + *endpoint = new ShardEndpoint(_primary->getId(), ChunkVersion::UNSHARDED()); + return Status::OK(); + } +} - // Sharded upserts *always* need to be exactly targeted by shard key - if (shardKey.isEmpty()) { - return Status(ErrorCodes::ShardKeyNotFound, - stream() << "upsert " << updateDoc.toBSON() - << " does not contain shard key for pattern " - << _manager->getShardKeyPattern().toString()); - } +Status ChunkManagerTargeter::targetUpdate(const BatchedUpdateDocument& updateDoc, + vector<ShardEndpoint*>* endpoints) const { + // + // Update targeting may use either the query or the update. This is to support save-style + // updates, of the form: + // + // coll.update({ _id : xxx }, { _id : xxx, shardKey : 1, foo : bar }, { upsert : true }) + // + // Because drivers do not know the shard key, they can't pull the shard key automatically + // into the query doc, and to correctly support upsert we must target a single shard. + // + // The rule is simple - If the update is replacement style (no '$set'), we target using the + // update. If the update is replacement style, we target using the query. + // + // If we have the exact shard key in either the query or replacement doc, we target using + // that extracted key. + // - // Also check shard key size on upsert - Status status = ShardKeyPattern::checkShardKeySize(shardKey); - if (!status.isOK()) - return status; - } + BSONObj query = updateDoc.getQuery(); + BSONObj updateExpr = updateDoc.getUpdateExpr(); - // Validate that single (non-multi) sharded updates are targeted by shard key or _id - if (!updateDoc.getMulti() && shardKey.isEmpty() - && !isExactIdQuery(updateDoc.getQuery())) { - return Status(ErrorCodes::ShardKeyNotFound, - stream() << "update " << updateDoc.toBSON() - << " does not contain _id or shard key for pattern " - << _manager->getShardKeyPattern().toString()); - } - } + UpdateType updateType = getUpdateExprType(updateDoc.getUpdateExpr()); - // Target the shard key, query, or replacement doc - if (!shardKey.isEmpty()) { - // We can't rely on our query targeting to be exact - ShardEndpoint* endpoint = NULL; - Status result = targetShardKey(shardKey, - (query.objsize() + updateExpr.objsize()), - &endpoint); - endpoints->push_back(endpoint); - return result; - } - else if (updateType == UpdateType_OpStyle) { - return targetQuery(query, endpoints); - } - else { - return targetDoc(updateExpr, endpoints); - } + if (updateType == UpdateType_Unknown) { + return Status(ErrorCodes::UnsupportedFormat, + stream() << "update document " << updateExpr + << " has mixed $operator and non-$operator style fields"); } - Status ChunkManagerTargeter::targetDelete( const BatchedDeleteDocument& deleteDoc, - vector<ShardEndpoint*>* endpoints ) const { - - BSONObj shardKey; - - if ( _manager ) { + BSONObj shardKey; - // - // Sharded collections have the following further requirements for targeting: - // - // Limit-1 deletes must be targeted exactly by shard key *or* exact _id - // + if (_manager) { + // + // Sharded collections have the following futher requirements for targeting: + // + // Upserts must be targeted exactly by shard key. + // Non-multi updates must be targeted exactly by shard key *or* exact _id. + // - // Get the shard key + // Get the shard key + if (updateType == UpdateType_OpStyle) { + // Target using the query StatusWith<BSONObj> status = - _manager->getShardKeyPattern().extractShardKeyFromQuery(deleteDoc.getQuery()); + _manager->getShardKeyPattern().extractShardKeyFromQuery(query); // Bad query if (!status.isOK()) return status.getStatus(); shardKey = status.getValue(); + } else { + // Target using the replacement document + shardKey = _manager->getShardKeyPattern().extractShardKeyFromDoc(updateExpr); + } - // Validate that single (limit-1) sharded deletes are targeted by shard key or _id - if (deleteDoc.getLimit() == 1 && shardKey.isEmpty() - && !isExactIdQuery(deleteDoc.getQuery())) { + // + // Extra sharded update validation + // + + if (updateDoc.getUpsert()) { + // Sharded upserts *always* need to be exactly targeted by shard key + if (shardKey.isEmpty()) { return Status(ErrorCodes::ShardKeyNotFound, - stream() << "delete " << deleteDoc.toBSON() - << " does not contain _id or shard key for pattern " + stream() << "upsert " << updateDoc.toBSON() + << " does not contain shard key for pattern " << _manager->getShardKeyPattern().toString()); } - } - // Target the shard key or delete query - if (!shardKey.isEmpty()) { - // We can't rely on our query targeting to be exact - ShardEndpoint* endpoint = NULL; - Status result = targetShardKey(shardKey, 0, &endpoint); - endpoints->push_back(endpoint); - return result; + // Also check shard key size on upsert + Status status = ShardKeyPattern::checkShardKeySize(shardKey); + if (!status.isOK()) + return status; } - else { - return targetQuery(deleteDoc.getQuery(), endpoints); + + // Validate that single (non-multi) sharded updates are targeted by shard key or _id + if (!updateDoc.getMulti() && shardKey.isEmpty() && !isExactIdQuery(updateDoc.getQuery())) { + return Status(ErrorCodes::ShardKeyNotFound, + stream() << "update " << updateDoc.toBSON() + << " does not contain _id or shard key for pattern " + << _manager->getShardKeyPattern().toString()); } } - Status ChunkManagerTargeter::targetDoc(const BSONObj& doc, - vector<ShardEndpoint*>* endpoints) const { - // NOTE: This is weird and fragile, but it's the way our language works right now - - // documents are either A) invalid or B) valid equality queries over themselves. - return targetQuery(doc, endpoints); + // Target the shard key, query, or replacement doc + if (!shardKey.isEmpty()) { + // We can't rely on our query targeting to be exact + ShardEndpoint* endpoint = NULL; + Status result = + targetShardKey(shardKey, (query.objsize() + updateExpr.objsize()), &endpoint); + endpoints->push_back(endpoint); + return result; + } else if (updateType == UpdateType_OpStyle) { + return targetQuery(query, endpoints); + } else { + return targetDoc(updateExpr, endpoints); } +} - Status ChunkManagerTargeter::targetQuery( const BSONObj& query, - vector<ShardEndpoint*>* endpoints ) const { - - if ( !_primary && !_manager ) { - return Status(ErrorCodes::NamespaceNotFound, - stream() << "could not target query in " - << getNS().ns() << "; no metadata found"); - } - - set<ShardId> shardIds; - if ( _manager ) { - try { - _manager->getShardIdsForQuery(shardIds, query); - } catch ( const DBException& ex ) { - return ex.toStatus(); - } - } - else { - shardIds.insert(_primary->getId()); - } +Status ChunkManagerTargeter::targetDelete(const BatchedDeleteDocument& deleteDoc, + vector<ShardEndpoint*>* endpoints) const { + BSONObj shardKey; - for (const ShardId& shardId : shardIds) { - endpoints->push_back(new ShardEndpoint(shardId, - _manager ? _manager->getVersion(shardId) : - ChunkVersion::UNSHARDED())); - } + if (_manager) { + // + // Sharded collections have the following further requirements for targeting: + // + // Limit-1 deletes must be targeted exactly by shard key *or* exact _id + // - return Status::OK(); - } + // Get the shard key + StatusWith<BSONObj> status = + _manager->getShardKeyPattern().extractShardKeyFromQuery(deleteDoc.getQuery()); - Status ChunkManagerTargeter::targetShardKey(const BSONObj& shardKey, - long long estDataSize, - ShardEndpoint** endpoint) const { - invariant(NULL != _manager); + // Bad query + if (!status.isOK()) + return status.getStatus(); - ChunkPtr chunk = _manager->findIntersectingChunk(shardKey); + shardKey = status.getValue(); - // Track autosplit stats for sharded collections - // Note: this is only best effort accounting and is not accurate. - if (estDataSize > 0) { - _stats.chunkSizeDelta[chunk->getMin()] += estDataSize; + // Validate that single (limit-1) sharded deletes are targeted by shard key or _id + if (deleteDoc.getLimit() == 1 && shardKey.isEmpty() && + !isExactIdQuery(deleteDoc.getQuery())) { + return Status(ErrorCodes::ShardKeyNotFound, + stream() << "delete " << deleteDoc.toBSON() + << " does not contain _id or shard key for pattern " + << _manager->getShardKeyPattern().toString()); } + } - *endpoint = new ShardEndpoint(chunk->getShardId(), - _manager->getVersion(chunk->getShardId())); + // Target the shard key or delete query + if (!shardKey.isEmpty()) { + // We can't rely on our query targeting to be exact + ShardEndpoint* endpoint = NULL; + Status result = targetShardKey(shardKey, 0, &endpoint); + endpoints->push_back(endpoint); + return result; + } else { + return targetQuery(deleteDoc.getQuery(), endpoints); + } +} + +Status ChunkManagerTargeter::targetDoc(const BSONObj& doc, + vector<ShardEndpoint*>* endpoints) const { + // NOTE: This is weird and fragile, but it's the way our language works right now - + // documents are either A) invalid or B) valid equality queries over themselves. + return targetQuery(doc, endpoints); +} + +Status ChunkManagerTargeter::targetQuery(const BSONObj& query, + vector<ShardEndpoint*>* endpoints) const { + if (!_primary && !_manager) { + return Status(ErrorCodes::NamespaceNotFound, + stream() << "could not target query in " << getNS().ns() + << "; no metadata found"); + } - return Status::OK(); + set<ShardId> shardIds; + if (_manager) { + try { + _manager->getShardIdsForQuery(shardIds, query); + } catch (const DBException& ex) { + return ex.toStatus(); + } + } else { + shardIds.insert(_primary->getId()); } - Status ChunkManagerTargeter::targetCollection( vector<ShardEndpoint*>* endpoints ) const { + for (const ShardId& shardId : shardIds) { + endpoints->push_back(new ShardEndpoint( + shardId, _manager ? _manager->getVersion(shardId) : ChunkVersion::UNSHARDED())); + } - if ( !_primary && !_manager ) { - return Status( ErrorCodes::NamespaceNotFound, - str::stream() << "could not target full range of " - << getNS().ns() - << "; metadata not found" ); - } + return Status::OK(); +} - set<ShardId> shardIds; - if ( _manager ) { - _manager->getAllShardIds(&shardIds); - } - else { - shardIds.insert(_primary->getId()); - } +Status ChunkManagerTargeter::targetShardKey(const BSONObj& shardKey, + long long estDataSize, + ShardEndpoint** endpoint) const { + invariant(NULL != _manager); - for (const ShardId& shardId : shardIds) { - endpoints->push_back(new ShardEndpoint(shardId, - _manager ? _manager->getVersion(shardId) : - ChunkVersion::UNSHARDED())); - } + ChunkPtr chunk = _manager->findIntersectingChunk(shardKey); - return Status::OK(); + // Track autosplit stats for sharded collections + // Note: this is only best effort accounting and is not accurate. + if (estDataSize > 0) { + _stats.chunkSizeDelta[chunk->getMin()] += estDataSize; } - Status ChunkManagerTargeter::targetAllShards( vector<ShardEndpoint*>* endpoints ) const { + *endpoint = new ShardEndpoint(chunk->getShardId(), _manager->getVersion(chunk->getShardId())); - if ( !_primary && !_manager ) { - return Status( ErrorCodes::NamespaceNotFound, - str::stream() << "could not target every shard with versions for " - << getNS().ns() - << "; metadata not found" ); - } - - vector<ShardId> shardIds; - grid.shardRegistry()->getAllShardIds(&shardIds); + return Status::OK(); +} - for (const ShardId& shardId : shardIds) { - endpoints->push_back(new ShardEndpoint(shardId, - _manager ? - _manager->getVersion(shardId) : - ChunkVersion::UNSHARDED())); - } +Status ChunkManagerTargeter::targetCollection(vector<ShardEndpoint*>* endpoints) const { + if (!_primary && !_manager) { + return Status(ErrorCodes::NamespaceNotFound, + str::stream() << "could not target full range of " << getNS().ns() + << "; metadata not found"); + } - return Status::OK(); + set<ShardId> shardIds; + if (_manager) { + _manager->getAllShardIds(&shardIds); + } else { + shardIds.insert(_primary->getId()); } - void ChunkManagerTargeter::noteStaleResponse( const ShardEndpoint& endpoint, - const BSONObj& staleInfo ) { - dassert( !_needsTargetingRefresh ); + for (const ShardId& shardId : shardIds) { + endpoints->push_back(new ShardEndpoint( + shardId, _manager ? _manager->getVersion(shardId) : ChunkVersion::UNSHARDED())); + } - ChunkVersion remoteShardVersion; - if ( staleInfo["vWanted"].eoo() ) { - // If we don't have a vWanted sent, assume the version is higher than our current - // version. - remoteShardVersion = - getShardVersion(endpoint.shardName, _manager.get(), _primary.get()); - remoteShardVersion.incMajor(); - } - else { - remoteShardVersion = ChunkVersion::fromBSON( staleInfo, "vWanted" ); - } + return Status::OK(); +} - ShardVersionMap::iterator it = _remoteShardVersions.find( endpoint.shardName ); - if ( it == _remoteShardVersions.end() ) { - _remoteShardVersions.insert( make_pair( endpoint.shardName, remoteShardVersion ) ); - } - else { - ChunkVersion& previouslyNotedVersion = it->second; - if ( previouslyNotedVersion.hasEqualEpoch( remoteShardVersion )) { - if ( previouslyNotedVersion.isOlderThan( remoteShardVersion )) { - remoteShardVersion.cloneTo( &previouslyNotedVersion ); - } - } - else { - // Epoch changed midway while applying the batch so set the version to - // something unique and non-existent to force a reload when - // refreshIsNeeded is called. - ChunkVersion::IGNORED().cloneTo( &previouslyNotedVersion ); - } - } +Status ChunkManagerTargeter::targetAllShards(vector<ShardEndpoint*>* endpoints) const { + if (!_primary && !_manager) { + return Status(ErrorCodes::NamespaceNotFound, + str::stream() << "could not target every shard with versions for " + << getNS().ns() << "; metadata not found"); } - void ChunkManagerTargeter::noteCouldNotTarget() { - dassert( _remoteShardVersions.empty() ); - _needsTargetingRefresh = true; - } + vector<ShardId> shardIds; + grid.shardRegistry()->getAllShardIds(&shardIds); - const TargeterStats* ChunkManagerTargeter::getStats() const { - return &_stats; + for (const ShardId& shardId : shardIds) { + endpoints->push_back(new ShardEndpoint( + shardId, _manager ? _manager->getVersion(shardId) : ChunkVersion::UNSHARDED())); } - Status ChunkManagerTargeter::refreshIfNeeded( bool *wasChanged ) { + return Status::OK(); +} + +void ChunkManagerTargeter::noteStaleResponse(const ShardEndpoint& endpoint, + const BSONObj& staleInfo) { + dassert(!_needsTargetingRefresh); + + ChunkVersion remoteShardVersion; + if (staleInfo["vWanted"].eoo()) { + // If we don't have a vWanted sent, assume the version is higher than our current + // version. + remoteShardVersion = getShardVersion(endpoint.shardName, _manager.get(), _primary.get()); + remoteShardVersion.incMajor(); + } else { + remoteShardVersion = ChunkVersion::fromBSON(staleInfo, "vWanted"); + } - bool dummy; - if (!wasChanged) { - wasChanged = &dummy; + ShardVersionMap::iterator it = _remoteShardVersions.find(endpoint.shardName); + if (it == _remoteShardVersions.end()) { + _remoteShardVersions.insert(make_pair(endpoint.shardName, remoteShardVersion)); + } else { + ChunkVersion& previouslyNotedVersion = it->second; + if (previouslyNotedVersion.hasEqualEpoch(remoteShardVersion)) { + if (previouslyNotedVersion.isOlderThan(remoteShardVersion)) { + remoteShardVersion.cloneTo(&previouslyNotedVersion); + } + } else { + // Epoch changed midway while applying the batch so set the version to + // something unique and non-existent to force a reload when + // refreshIsNeeded is called. + ChunkVersion::IGNORED().cloneTo(&previouslyNotedVersion); } + } +} - *wasChanged = false; +void ChunkManagerTargeter::noteCouldNotTarget() { + dassert(_remoteShardVersions.empty()); + _needsTargetingRefresh = true; +} - // - // Did we have any stale config or targeting errors at all? - // +const TargeterStats* ChunkManagerTargeter::getStats() const { + return &_stats; +} - if (!_needsTargetingRefresh && _remoteShardVersions.empty()) { - return Status::OK(); - } +Status ChunkManagerTargeter::refreshIfNeeded(bool* wasChanged) { + bool dummy; + if (!wasChanged) { + wasChanged = &dummy; + } - // - // Get the latest metadata information from the cache if there were issues - // + *wasChanged = false; - ChunkManagerPtr lastManager = _manager; - ShardPtr lastPrimary = _primary; + // + // Did we have any stale config or targeting errors at all? + // - auto status = grid.implicitCreateDb(_nss.db().toString()); - if (!status.isOK()) { - return status.getStatus(); - } + if (!_needsTargetingRefresh && _remoteShardVersions.empty()) { + return Status::OK(); + } - shared_ptr<DBConfig> config = status.getValue(); - config->getChunkManagerOrPrimary( _nss.ns(), _manager, _primary ); + // + // Get the latest metadata information from the cache if there were issues + // - // We now have the latest metadata from the cache. + ChunkManagerPtr lastManager = _manager; + ShardPtr lastPrimary = _primary; - // - // See if and how we need to do a remote refresh. - // Either we couldn't target at all, or we have stale versions, but not both. - // + auto status = grid.implicitCreateDb(_nss.db().toString()); + if (!status.isOK()) { + return status.getStatus(); + } - dassert( !( _needsTargetingRefresh && !_remoteShardVersions.empty() ) ); + shared_ptr<DBConfig> config = status.getValue(); + config->getChunkManagerOrPrimary(_nss.ns(), _manager, _primary); - if ( _needsTargetingRefresh ) { + // We now have the latest metadata from the cache. - // Reset the field - _needsTargetingRefresh = false; + // + // See if and how we need to do a remote refresh. + // Either we couldn't target at all, or we have stale versions, but not both. + // - // If we couldn't target, we might need to refresh if we haven't remotely refreshed the - // metadata since we last got it from the cache. + dassert(!(_needsTargetingRefresh && !_remoteShardVersions.empty())); - bool alreadyRefreshed = wasMetadataRefreshed( lastManager, - lastPrimary, - _manager, - _primary ); + if (_needsTargetingRefresh) { + // Reset the field + _needsTargetingRefresh = false; - // If didn't already refresh the targeting information, refresh it - if ( !alreadyRefreshed ) { - // To match previous behavior, we just need an incremental refresh here - return refreshNow( RefreshType_RefreshChunkManager ); - } + // If we couldn't target, we might need to refresh if we haven't remotely refreshed the + // metadata since we last got it from the cache. - *wasChanged = isMetadataDifferent( lastManager, lastPrimary, _manager, _primary ); - return Status::OK(); - } - else if ( !_remoteShardVersions.empty() ) { + bool alreadyRefreshed = wasMetadataRefreshed(lastManager, lastPrimary, _manager, _primary); - // If we got stale shard versions from remote shards, we may need to refresh - // NOTE: Not sure yet if this can happen simultaneously with targeting issues + // If didn't already refresh the targeting information, refresh it + if (!alreadyRefreshed) { + // To match previous behavior, we just need an incremental refresh here + return refreshNow(RefreshType_RefreshChunkManager); + } - CompareResult result = compareAllShardVersions(_manager.get(), - _primary.get(), - _remoteShardVersions); - // Reset the versions - _remoteShardVersions.clear(); + *wasChanged = isMetadataDifferent(lastManager, lastPrimary, _manager, _primary); + return Status::OK(); + } else if (!_remoteShardVersions.empty()) { + // If we got stale shard versions from remote shards, we may need to refresh + // NOTE: Not sure yet if this can happen simultaneously with targeting issues - if ( result == CompareResult_Unknown ) { - // Our current shard versions aren't all comparable to the old versions, maybe drop - return refreshNow( RefreshType_ReloadDatabase ); - } - else if ( result == CompareResult_LT ) { - // Our current shard versions are less than the remote versions, but no drop - return refreshNow( RefreshType_RefreshChunkManager ); - } + CompareResult result = + compareAllShardVersions(_manager.get(), _primary.get(), _remoteShardVersions); + // Reset the versions + _remoteShardVersions.clear(); - *wasChanged = isMetadataDifferent( lastManager, lastPrimary, _manager, _primary ); - return Status::OK(); + if (result == CompareResult_Unknown) { + // Our current shard versions aren't all comparable to the old versions, maybe drop + return refreshNow(RefreshType_ReloadDatabase); + } else if (result == CompareResult_LT) { + // Our current shard versions are less than the remote versions, but no drop + return refreshNow(RefreshType_RefreshChunkManager); } - // unreachable - dassert( false ); + *wasChanged = isMetadataDifferent(lastManager, lastPrimary, _manager, _primary); return Status::OK(); } - Status ChunkManagerTargeter::refreshNow( RefreshType refreshType ) { - auto status = grid.implicitCreateDb(_nss.db().toString()); - if (!status.isOK()) { - return status.getStatus(); - } + // unreachable + dassert(false); + return Status::OK(); +} - shared_ptr<DBConfig> config = status.getValue(); +Status ChunkManagerTargeter::refreshNow(RefreshType refreshType) { + auto status = grid.implicitCreateDb(_nss.db().toString()); + if (!status.isOK()) { + return status.getStatus(); + } - // Try not to spam the configs - refreshBackoff(); + shared_ptr<DBConfig> config = status.getValue(); - // TODO: Improve synchronization and make more explicit - if ( refreshType == RefreshType_RefreshChunkManager ) { - try { - // Forces a remote check of the collection info, synchronization between threads - // happens internally. - config->getChunkManagerIfExists( _nss.ns(), true ); - } - catch ( const DBException& ex ) { - return Status( ErrorCodes::UnknownError, ex.toString() ); - } - config->getChunkManagerOrPrimary( _nss.ns(), _manager, _primary ); - } - else if ( refreshType == RefreshType_ReloadDatabase ) { - try { - // Dumps the db info, reloads it all, synchronization between threads happens - // internally. - config->reload(); - config->getChunkManagerIfExists( _nss.ns(), true, true ); - } - catch ( const DBException& ex ) { - return Status( ErrorCodes::UnknownError, ex.toString() ); - } + // Try not to spam the configs + refreshBackoff(); - config->getChunkManagerOrPrimary( _nss.ns(), _manager, _primary ); + // TODO: Improve synchronization and make more explicit + if (refreshType == RefreshType_RefreshChunkManager) { + try { + // Forces a remote check of the collection info, synchronization between threads + // happens internally. + config->getChunkManagerIfExists(_nss.ns(), true); + } catch (const DBException& ex) { + return Status(ErrorCodes::UnknownError, ex.toString()); + } + config->getChunkManagerOrPrimary(_nss.ns(), _manager, _primary); + } else if (refreshType == RefreshType_ReloadDatabase) { + try { + // Dumps the db info, reloads it all, synchronization between threads happens + // internally. + config->reload(); + config->getChunkManagerIfExists(_nss.ns(), true, true); + } catch (const DBException& ex) { + return Status(ErrorCodes::UnknownError, ex.toString()); } - return Status::OK(); + config->getChunkManagerOrPrimary(_nss.ns(), _manager, _primary); } -} // namespace mongo + return Status::OK(); +} +} // namespace mongo diff --git a/src/mongo/s/chunk_manager_targeter.h b/src/mongo/s/chunk_manager_targeter.h index 3c5b9d68a09..e7c754448c8 100644 --- a/src/mongo/s/chunk_manager_targeter.h +++ b/src/mongo/s/chunk_manager_targeter.h @@ -36,130 +36,130 @@ namespace mongo { - class ChunkManager; - struct ChunkVersion; - class Shard; - - struct TargeterStats { - // Map of chunk shard minKey -> approximate delta. This is used for deciding - // whether a chunk might need splitting or not. - std::map<BSONObj, int> chunkSizeDelta; - }; +class ChunkManager; +struct ChunkVersion; +class Shard; + +struct TargeterStats { + // Map of chunk shard minKey -> approximate delta. This is used for deciding + // whether a chunk might need splitting or not. + std::map<BSONObj, int> chunkSizeDelta; +}; + +/** + * NSTargeter based on a ChunkManager implementation. Wraps all exception codepaths and + * returns DatabaseNotFound statuses on applicable failures. + * + * Must be initialized before use, and initialization may fail. + */ +class ChunkManagerTargeter : public NSTargeter { +public: + ChunkManagerTargeter(const NamespaceString& nss); + + /** + * Initializes the ChunkManagerTargeter with the latest targeting information for the + * namespace. May need to block and load information from a remote config server. + * + * Returns !OK if the information could not be initialized. + */ + Status init(); + + const NamespaceString& getNS() const; + + // Returns ShardKeyNotFound if document does not have a full shard key. + Status targetInsert(const BSONObj& doc, ShardEndpoint** endpoint) const; + + // Returns ShardKeyNotFound if the update can't be targeted without a shard key. + Status targetUpdate(const BatchedUpdateDocument& updateDoc, + std::vector<ShardEndpoint*>* endpoints) const; + + // Returns ShardKeyNotFound if the delete can't be targeted without a shard key. + Status targetDelete(const BatchedDeleteDocument& deleteDoc, + std::vector<ShardEndpoint*>* endpoints) const; + + Status targetCollection(std::vector<ShardEndpoint*>* endpoints) const; + + Status targetAllShards(std::vector<ShardEndpoint*>* endpoints) const; + + void noteStaleResponse(const ShardEndpoint& endpoint, const BSONObj& staleInfo); + + void noteCouldNotTarget(); /** - * NSTargeter based on a ChunkManager implementation. Wraps all exception codepaths and - * returns DatabaseNotFound statuses on applicable failures. + * Replaces the targeting information with the latest information from the cache. If this + * information is stale WRT the noted stale responses or a remote refresh is needed due + * to a targeting failure, will contact the config servers to reload the metadata. + * + * Reports wasChanged = true if the metadata is different after this reload. * - * Must be initialized before use, and initialization may fail. + * Also see NSTargeter::refreshIfNeeded(). + */ + Status refreshIfNeeded(bool* wasChanged); + + /** + * Returns the stats. Note that the returned stats object is still owned by this targeter. */ - class ChunkManagerTargeter : public NSTargeter { - public: - ChunkManagerTargeter(const NamespaceString& nss); - - /** - * Initializes the ChunkManagerTargeter with the latest targeting information for the - * namespace. May need to block and load information from a remote config server. - * - * Returns !OK if the information could not be initialized. - */ - Status init(); - - const NamespaceString& getNS() const; - - // Returns ShardKeyNotFound if document does not have a full shard key. - Status targetInsert( const BSONObj& doc, ShardEndpoint** endpoint ) const; - - // Returns ShardKeyNotFound if the update can't be targeted without a shard key. - Status targetUpdate( const BatchedUpdateDocument& updateDoc, - std::vector<ShardEndpoint*>* endpoints ) const; - - // Returns ShardKeyNotFound if the delete can't be targeted without a shard key. - Status targetDelete( const BatchedDeleteDocument& deleteDoc, - std::vector<ShardEndpoint*>* endpoints ) const; - - Status targetCollection( std::vector<ShardEndpoint*>* endpoints ) const; - - Status targetAllShards( std::vector<ShardEndpoint*>* endpoints ) const; - - void noteStaleResponse( const ShardEndpoint& endpoint, const BSONObj& staleInfo ); - - void noteCouldNotTarget(); - - /** - * Replaces the targeting information with the latest information from the cache. If this - * information is stale WRT the noted stale responses or a remote refresh is needed due - * to a targeting failure, will contact the config servers to reload the metadata. - * - * Reports wasChanged = true if the metadata is different after this reload. - * - * Also see NSTargeter::refreshIfNeeded(). - */ - Status refreshIfNeeded( bool* wasChanged ); - - /** - * Returns the stats. Note that the returned stats object is still owned by this targeter. - */ - const TargeterStats* getStats() const; - - private: - // Different ways we can refresh metadata - enum RefreshType { - // No refresh is needed - RefreshType_None, - // The version has gone up, but the collection hasn't been dropped - RefreshType_RefreshChunkManager, - // The collection may have been dropped, so we need to reload the db - RefreshType_ReloadDatabase - }; - - typedef std::map<std::string, ChunkVersion> ShardVersionMap; - - - /** - * Performs an actual refresh from the config server. - */ - Status refreshNow( RefreshType refreshType ); - - /** - * Returns a vector of ShardEndpoints where a document might need to be placed. - * - * Returns !OK with message if replacement could not be targeted - */ - Status targetDoc(const BSONObj& doc, std::vector<ShardEndpoint*>* endpoints) const; - - /** - * Returns a vector of ShardEndpoints for a potentially multi-shard query. - * - * Returns !OK with message if query could not be targeted. - */ - Status targetQuery( const BSONObj& query, std::vector<ShardEndpoint*>* endpoints ) const; - - /** - * Returns a ShardEndpoint for an exact shard key query. - * - * Also has the side effect of updating the chunks stats with an estimate of the amount of - * data targeted at this shard key. - */ - Status targetShardKey(const BSONObj& doc, - long long estDataSize, - ShardEndpoint** endpoint) const; - - // Full namespace of the collection for this targeter - const NamespaceString _nss; - - // Stores whether we need to check the remote server on refresh - bool _needsTargetingRefresh; - - // Represents only the view and not really part of the targeter state. - mutable TargeterStats _stats; - - // Zero or one of these are filled at all times - // If sharded, _manager, if unsharded, _primary, on error, neither - std::shared_ptr<ChunkManager> _manager; - std::shared_ptr<Shard> _primary; - - // Map of shard->remote shard version reported from stale errors - ShardVersionMap _remoteShardVersions; + const TargeterStats* getStats() const; + +private: + // Different ways we can refresh metadata + enum RefreshType { + // No refresh is needed + RefreshType_None, + // The version has gone up, but the collection hasn't been dropped + RefreshType_RefreshChunkManager, + // The collection may have been dropped, so we need to reload the db + RefreshType_ReloadDatabase }; -} // namespace mongo + typedef std::map<std::string, ChunkVersion> ShardVersionMap; + + + /** + * Performs an actual refresh from the config server. + */ + Status refreshNow(RefreshType refreshType); + + /** + * Returns a vector of ShardEndpoints where a document might need to be placed. + * + * Returns !OK with message if replacement could not be targeted + */ + Status targetDoc(const BSONObj& doc, std::vector<ShardEndpoint*>* endpoints) const; + + /** + * Returns a vector of ShardEndpoints for a potentially multi-shard query. + * + * Returns !OK with message if query could not be targeted. + */ + Status targetQuery(const BSONObj& query, std::vector<ShardEndpoint*>* endpoints) const; + + /** + * Returns a ShardEndpoint for an exact shard key query. + * + * Also has the side effect of updating the chunks stats with an estimate of the amount of + * data targeted at this shard key. + */ + Status targetShardKey(const BSONObj& doc, + long long estDataSize, + ShardEndpoint** endpoint) const; + + // Full namespace of the collection for this targeter + const NamespaceString _nss; + + // Stores whether we need to check the remote server on refresh + bool _needsTargetingRefresh; + + // Represents only the view and not really part of the targeter state. + mutable TargeterStats _stats; + + // Zero or one of these are filled at all times + // If sharded, _manager, if unsharded, _primary, on error, neither + std::shared_ptr<ChunkManager> _manager; + std::shared_ptr<Shard> _primary; + + // Map of shard->remote shard version reported from stale errors + ShardVersionMap _remoteShardVersions; +}; + +} // namespace mongo diff --git a/src/mongo/s/chunk_manager_targeter_test.cpp b/src/mongo/s/chunk_manager_targeter_test.cpp index 85e674fd1c1..4f43580d40c 100644 --- a/src/mongo/s/chunk_manager_targeter_test.cpp +++ b/src/mongo/s/chunk_manager_targeter_test.cpp @@ -40,496 +40,475 @@ namespace { - using namespace mongo; - - using std::unique_ptr; - using std::make_pair; - - /** - * ChunkManager targeting test - * - * TODO: - * Pull the implementation out of chunk.cpp - */ - - // Utility function to create a CanonicalQuery - CanonicalQuery* canonicalize(const char* queryStr) { - BSONObj queryObj = fromjson(queryStr); - CanonicalQuery* cq; - Status result = CanonicalQuery::canonicalize( - "test.foo", queryObj, &cq, WhereCallbackNoop()); - ASSERT_OK(result); - return cq; - } - - void checkIndexBoundsWithKey(const char* keyStr, - const char* queryStr, - const IndexBounds& expectedBounds) { - - unique_ptr<CanonicalQuery> query(canonicalize(queryStr)); - ASSERT(query.get() != NULL); - - BSONObj key = fromjson(keyStr); - - IndexBounds indexBounds = ChunkManager::getIndexBoundsForQuery(key, query.get()); - ASSERT_EQUALS(indexBounds.size(), expectedBounds.size()); - for (size_t i = 0; i < indexBounds.size(); i++) { - const OrderedIntervalList& oil = indexBounds.fields[i]; - const OrderedIntervalList& expectedOil = expectedBounds.fields[i]; - ASSERT_EQUALS(oil.intervals.size(), expectedOil.intervals.size()); - for (size_t i = 0; i < oil.intervals.size(); i++) { - if (Interval::INTERVAL_EQUALS != oil.intervals[i].compare(expectedOil.intervals[i])) { - log() << oil.intervals[i] << " != " << expectedOil.intervals[i]; - } - ASSERT_EQUALS(Interval::INTERVAL_EQUALS, oil.intervals[i].compare(expectedOil.intervals[i])); - } - } - } - - // Assume shard key is { a: 1 } - void checkIndexBounds(const char* queryStr, const OrderedIntervalList& expectedOil) { - unique_ptr<CanonicalQuery> query(canonicalize(queryStr)); - ASSERT(query.get() != NULL); - - BSONObj key = fromjson("{a: 1}"); +using namespace mongo; - IndexBounds indexBounds = ChunkManager::getIndexBoundsForQuery(key, query.get()); - ASSERT_EQUALS(indexBounds.size(), 1U); - const OrderedIntervalList& oil = indexBounds.fields.front(); +using std::unique_ptr; +using std::make_pair; - if (oil.intervals.size() != expectedOil.intervals.size()) { - for (size_t i = 0; i < oil.intervals.size(); i++) { - log() << oil.intervals[i]; - } - } +/** + * ChunkManager targeting test + * + * TODO: + * Pull the implementation out of chunk.cpp + */ +// Utility function to create a CanonicalQuery +CanonicalQuery* canonicalize(const char* queryStr) { + BSONObj queryObj = fromjson(queryStr); + CanonicalQuery* cq; + Status result = CanonicalQuery::canonicalize("test.foo", queryObj, &cq, WhereCallbackNoop()); + ASSERT_OK(result); + return cq; +} + +void checkIndexBoundsWithKey(const char* keyStr, + const char* queryStr, + const IndexBounds& expectedBounds) { + unique_ptr<CanonicalQuery> query(canonicalize(queryStr)); + ASSERT(query.get() != NULL); + + BSONObj key = fromjson(keyStr); + + IndexBounds indexBounds = ChunkManager::getIndexBoundsForQuery(key, query.get()); + ASSERT_EQUALS(indexBounds.size(), expectedBounds.size()); + for (size_t i = 0; i < indexBounds.size(); i++) { + const OrderedIntervalList& oil = indexBounds.fields[i]; + const OrderedIntervalList& expectedOil = expectedBounds.fields[i]; ASSERT_EQUALS(oil.intervals.size(), expectedOil.intervals.size()); for (size_t i = 0; i < oil.intervals.size(); i++) { - ASSERT_EQUALS(Interval::INTERVAL_EQUALS, oil.intervals[i].compare(expectedOil.intervals[i])); + if (Interval::INTERVAL_EQUALS != oil.intervals[i].compare(expectedOil.intervals[i])) { + log() << oil.intervals[i] << " != " << expectedOil.intervals[i]; + } + ASSERT_EQUALS(Interval::INTERVAL_EQUALS, + oil.intervals[i].compare(expectedOil.intervals[i])); } } +} - const double INF = std::numeric_limits<double>::infinity(); +// Assume shard key is { a: 1 } +void checkIndexBounds(const char* queryStr, const OrderedIntervalList& expectedOil) { + unique_ptr<CanonicalQuery> query(canonicalize(queryStr)); + ASSERT(query.get() != NULL); - // { a: 2 } -> a: [2, 2] - TEST(CMCollapseTreeTest, Basic) { - OrderedIntervalList expected; - expected.intervals.push_back(Interval(BSON("" << 2 << "" << 2), true, true)); - checkIndexBounds("{a: 2}", expected); - } + BSONObj key = fromjson("{a: 1}"); - // { b: 2 } -> a: [MinKey, MaxKey] - TEST(CMCollapseTreeTest, AllValue) { - OrderedIntervalList expected; - BSONObjBuilder builder; - builder.appendMinKey(""); - builder.appendMaxKey(""); - expected.intervals.push_back(Interval(builder.obj(), true, true)); - checkIndexBounds("{b: 2}", expected); - } + IndexBounds indexBounds = ChunkManager::getIndexBoundsForQuery(key, query.get()); + ASSERT_EQUALS(indexBounds.size(), 1U); + const OrderedIntervalList& oil = indexBounds.fields.front(); - // { 'a' : { '$not' : { '$gt' : 1 } } } -> a: [MinKey, 1.0], (inf.0, MaxKey] - TEST(CMCollapseTreeTest, NegativeGT) { - OrderedIntervalList expected; - { - BSONObjBuilder builder; - builder.appendMinKey(""); - builder.appendNumber("", 1.0); - expected.intervals.push_back(Interval(builder.obj(), true, true)); - } - { - BSONObjBuilder builder; - builder.append("", std::numeric_limits<double>::infinity()); - builder.appendMaxKey(""); - expected.intervals.push_back(Interval(builder.obj(), false, true)); + if (oil.intervals.size() != expectedOil.intervals.size()) { + for (size_t i = 0; i < oil.intervals.size(); i++) { + log() << oil.intervals[i]; } - checkIndexBounds("{ 'a' : { '$not' : { '$gt' : 1 } } }", expected); - } - - // {$or: [{a: 20}, {$and: [{a:1}, {b:7}]}]} -> a: [1.0, 1.0], [20.0, 20.0] - TEST(CMCollapseTreeTest, OrWithAndChild) { - OrderedIntervalList expected; - expected.intervals.push_back(Interval(BSON("" << 1.0 << "" << 1.0), true, true)); - expected.intervals.push_back(Interval(BSON("" << 20.0 << "" << 20.0), true, true)); - checkIndexBounds("{$or: [{a: 20}, {$and: [{a:1}, {b:7}]}]}", expected); - } - - // {a:20, $or: [{b:1}, {c:7}]} -> a: [20.0, 20.0] - TEST(CMCollapseTreeTest, AndWithUnindexedOrChild) { - // Logic rewrite could give a tree with root OR. - OrderedIntervalList expected; - expected.intervals.push_back(Interval(BSON("" << 20.0 << "" << 20.0), true, true)); - checkIndexBounds("{a:20, $or: [{b:1}, {c:7}]}", expected); - } - - // {$or: [{a:{$gt:2,$lt:10}}, {a:{$gt:0,$lt:5}}]} -> a: (0.0, 10.0) - TEST(CMCollapseTreeTest, OrOfAnd) { - OrderedIntervalList expected; - expected.intervals.push_back(Interval(BSON("" << 0.0 << "" << 10.0), false, false)); - checkIndexBounds("{$or: [{a:{$gt:2,$lt:10}}, {a:{$gt:0,$lt:5}}]}", expected); - } - - // {$or: [{a:{$gt:2,$lt:10}}, {a:{$gt:0,$lt:15}}, {a:{$gt:20}}]} - // -> a: (0.0, 15.0), (20.0, inf.0] - TEST(CMCollapseTreeTest, OrOfAnd2) { - OrderedIntervalList expected; - expected.intervals.push_back(Interval(BSON("" << 0.0 << "" << 15.0), false, false)); - expected.intervals.push_back(Interval(BSON("" << 20.0 << "" << INF), false, true)); - checkIndexBounds("{$or: [{a:{$gt:2,$lt:10}}, {a:{$gt:0,$lt:15}}, {a:{$gt:20}}]}", expected); - } - - // "{$or: [{a:{$gt:1,$lt:5},b:6}, {a:3,b:{$gt:0,$lt:10}}]}" -> a: (1.0, 5.0) - TEST(CMCollapseTreeTest, OrOfAnd3) { - OrderedIntervalList expected; - expected.intervals.push_back(Interval(BSON("" << 1.0 << "" << 5.0), false, false)); - checkIndexBounds("{$or: [{a:{$gt:1,$lt:5},b:6}, {a:3,b:{$gt:0,$lt:10}}]}", expected); - } - - // - // Compound shard key - // - - // "{$or: [{a:{$gt:1,$lt:5}, b:{$gt:0,$lt:3}, c:6}, " - // "{a:3, b:{$gt:1,$lt:2}, c:{$gt:0,$lt:10}}]}", - // -> a: (1.0, 5.0), b: (0.0, 3.0) - TEST(CMCollapseTreeTest, OrOfAnd4) { - IndexBounds expectedBounds; - expectedBounds.fields.push_back(OrderedIntervalList()); - expectedBounds.fields.push_back(OrderedIntervalList()); - - expectedBounds.fields[0].intervals.push_back( - Interval(BSON("" << 1.0 << "" << 5.0), false, false)); - expectedBounds.fields[1].intervals.push_back( - Interval(BSON("" << 0.0 << "" << 3.0), false, false)); - - checkIndexBoundsWithKey( - "{a: 1, b: 1}", // shard key - "{$or: [{a:{$gt:1,$lt:5}, b:{$gt:0,$lt:3}, c:6}, " - "{a:3, b:{$gt:1,$lt:2}, c:{$gt:0,$lt:10}}]}", - expectedBounds); - } - - // "{$or: [{a:{$gt:1,$lt:5}, c:6}, " - // "{a:3, b:{$gt:1,$lt:2}, c:{$gt:0,$lt:10}}]}")); - // -> - TEST(CMCollapseTreeTest, OrOfAnd5) { - IndexBounds expectedBounds; - expectedBounds.fields.push_back(OrderedIntervalList()); - expectedBounds.fields.push_back(OrderedIntervalList()); - - expectedBounds.fields[0].intervals.push_back( - Interval(BSON("" << 1.0 << "" << 5.0), false, false)); - BSONObjBuilder builder; - builder.appendMinKey(""); - builder.appendMaxKey(""); - expectedBounds.fields[1].intervals.push_back( - Interval(builder.obj(), true, true)); - - checkIndexBoundsWithKey( - "{a: 1, b: 1}", // shard key - "{$or: [{a:{$gt:1,$lt:5}, c:6}, " - "{a:3, b:{$gt:1,$lt:2}, c:{$gt:0,$lt:10}}]}", - expectedBounds); - } - - // {$or: [{a:{$in:[1]},b:{$in:[1]}}, {a:{$in:[1,5]},b:{$in:[1,5]}}]} - // -> a: [1], [5]; b: [1], [5] - TEST(CMCollapseTreeTest, OrOfAnd6) { - IndexBounds expectedBounds; - expectedBounds.fields.push_back(OrderedIntervalList()); - expectedBounds.fields.push_back(OrderedIntervalList()); - - // a: [1], [5] - expectedBounds.fields[0].intervals.push_back( - Interval(BSON("" << 1.0 << "" << 1.0), true, true)); - expectedBounds.fields[0].intervals.push_back( - Interval(BSON("" << 5.0 << "" << 5.0), true, true)); - - // b: [1], [5] - expectedBounds.fields[1].intervals.push_back( - Interval(BSON("" << 1.0 << "" << 1.0), true, true)); - expectedBounds.fields[1].intervals.push_back( - Interval(BSON("" << 5.0 << "" << 5.0), true, true)); - - checkIndexBoundsWithKey( - "{a: 1, b: 1}", // shard key - "{$or: [{a:{$in:[1]},b:{$in:[1]}}, {a:{$in:[1,5]},b:{$in:[1,5]}}]}", - expectedBounds); - } - - // - // Array operators - // - - // {a : {$elemMatch: {b:1}}} -> a.b: [MinKey, MaxKey] - // Shard key doesn't allow multikey, but query on array should succeed without error. - TEST(CMCollapseTreeTest, ElemMatchOneField) { - IndexBounds expectedBounds; - expectedBounds.fields.push_back(OrderedIntervalList()); - OrderedIntervalList& oil = expectedBounds.fields.front(); - oil.intervals.push_back(Interval(BSON("" << 1 << "" << 1), true, true)); - checkIndexBoundsWithKey("{'a.b': 1}", "{a : {$elemMatch: {b:1}}}", expectedBounds); - } - - // {foo: {$all: [ {$elemMatch: {a:1, b:1}}, {$elemMatch: {a:2, b:2}}]}} - // -> foo.a: [1, 1] - // Or -> foo.a: [2, 2] - TEST(CMCollapseTreeTest, BasicAllElemMatch) { - Interval expectedInterval(BSON("" << 1 << "" << 1), true, true); - - const char* queryStr = "{foo: {$all: [ {$elemMatch: {a:1, b:1}} ]}}"; - unique_ptr<CanonicalQuery> query(canonicalize(queryStr)); - ASSERT(query.get() != NULL); - - BSONObj key = fromjson("{'foo.a': 1}"); - - IndexBounds indexBounds = ChunkManager::getIndexBoundsForQuery(key, query.get()); - ASSERT_EQUALS(indexBounds.size(), 1U); - const OrderedIntervalList& oil = indexBounds.fields.front(); - ASSERT_EQUALS(oil.intervals.size(), 1U); - const Interval& interval = oil.intervals.front(); - - // Choose one of the two possible solutions. - // Two solutions differ only by assignment of index tags. - ASSERT(Interval::INTERVAL_EQUALS == interval.compare(expectedInterval)); - } - - // {a : [1, 2, 3]} -> a: [1, 1], [[1, 2, 3], [1, 2, 3]] - TEST(CMCollapseTreeTest, ArrayEquality) { - OrderedIntervalList expected; - expected.intervals.push_back(Interval(BSON("" << 1 << "" << 1), true, true)); - BSONArray array(BSON_ARRAY(1 << 2 << 3)); - - Interval interval(BSON("" << array << "" << array), true, true); - expected.intervals.push_back(interval); - checkIndexBounds("{a : [1, 2, 3]}", expected); } - - // - // Features: Regex, $where, $text, hashed key - // - - // { a: /abc/ } -> a: ["", {}), [/abc/, /abc/] - TEST(CMCollapseTreeTest, Regex) { - OrderedIntervalList expected; - expected.intervals.push_back( - Interval(BSON("" << "" << "" << BSONObj()), true, false)); - BSONObjBuilder builder; - builder.appendRegex("", "abc"); - builder.appendRegex("", "abc"); - expected.intervals.push_back(Interval(builder.obj(), true, true)); - checkIndexBounds("{ a: /abc/ }", expected); + ASSERT_EQUALS(oil.intervals.size(), expectedOil.intervals.size()); + for (size_t i = 0; i < oil.intervals.size(); i++) { + ASSERT_EQUALS(Interval::INTERVAL_EQUALS, + oil.intervals[i].compare(expectedOil.intervals[i])); } - - // {$where: 'this.credits == this.debits' } - TEST(CMCollapseTreeTest, Where) { - OrderedIntervalList expected; +} + +const double INF = std::numeric_limits<double>::infinity(); + +// { a: 2 } -> a: [2, 2] +TEST(CMCollapseTreeTest, Basic) { + OrderedIntervalList expected; + expected.intervals.push_back(Interval(BSON("" << 2 << "" << 2), true, true)); + checkIndexBounds("{a: 2}", expected); +} + +// { b: 2 } -> a: [MinKey, MaxKey] +TEST(CMCollapseTreeTest, AllValue) { + OrderedIntervalList expected; + BSONObjBuilder builder; + builder.appendMinKey(""); + builder.appendMaxKey(""); + expected.intervals.push_back(Interval(builder.obj(), true, true)); + checkIndexBounds("{b: 2}", expected); +} + +// { 'a' : { '$not' : { '$gt' : 1 } } } -> a: [MinKey, 1.0], (inf.0, MaxKey] +TEST(CMCollapseTreeTest, NegativeGT) { + OrderedIntervalList expected; + { BSONObjBuilder builder; builder.appendMinKey(""); - builder.appendMaxKey(""); + builder.appendNumber("", 1.0); expected.intervals.push_back(Interval(builder.obj(), true, true)); - checkIndexBounds("{$where: 'this.credits == this.debits' }", expected); - } - - // { $text: { $search: "coffee -cake" } } - TEST(CMCollapseTreeTest, Text) { - OrderedIntervalList expected; - BSONObjBuilder builder; - builder.appendMinKey(""); - builder.appendMaxKey(""); - expected.intervals.push_back(Interval(builder.obj(), true, true)); - checkIndexBounds("{ $text: { $search: 'coffee -cake' } }", expected); - } - - // { a: 2, $text: { $search: "leche", $language: "es" } } - TEST(CMCollapseTreeTest, TextWithQuery) { - OrderedIntervalList expected; - BSONObjBuilder builder; - builder.appendMinKey(""); - builder.appendMaxKey(""); - expected.intervals.push_back(Interval(builder.obj(), true, true)); - checkIndexBounds("{ a: 2, $text: { $search: 'leche', $language: 'es' } }", expected); - } - - // { a: 0 } -> hashed a: [hash(0), hash(0)] - TEST(CMCollapseTreeTest, HashedSinglePoint) { - const char* queryStr = "{ a: 0 }"; - unique_ptr<CanonicalQuery> query(canonicalize(queryStr)); - ASSERT(query.get() != NULL); - - BSONObj key = fromjson("{a: 'hashed'}"); - - IndexBounds indexBounds = ChunkManager::getIndexBoundsForQuery(key, query.get()); - ASSERT_EQUALS(indexBounds.size(), 1U); - const OrderedIntervalList& oil = indexBounds.fields.front(); - ASSERT_EQUALS(oil.intervals.size(), 1U); - const Interval& interval = oil.intervals.front(); - ASSERT(interval.isPoint()); - } - - // { a: { $lt: 2, $gt: 1} } -> hashed a: [Minkey, Maxkey] - TEST(CMCollapseTreeTest, HashedRange) { - IndexBounds expectedBounds; - expectedBounds.fields.push_back(OrderedIntervalList()); - OrderedIntervalList& expectedOil = expectedBounds.fields.front(); - BSONObjBuilder builder; - builder.appendMinKey(""); - builder.appendMaxKey(""); - expectedOil.intervals.push_back(Interval(builder.obj(), true, true)); - checkIndexBoundsWithKey("{a: 'hashed'}", "{ a: { $lt: 2, $gt: 1} }", expectedBounds); } - - // { a: /abc/ } -> hashed a: [Minkey, Maxkey] - TEST(CMCollapseTreeTest, HashedRegex) { - IndexBounds expectedBounds; - expectedBounds.fields.push_back(OrderedIntervalList()); - OrderedIntervalList& expectedOil = expectedBounds.fields.front(); + { BSONObjBuilder builder; - builder.appendMinKey(""); + builder.append("", std::numeric_limits<double>::infinity()); builder.appendMaxKey(""); - expectedOil.intervals.push_back(Interval(builder.obj(), true, true)); - - checkIndexBoundsWithKey("{a: 'hashed'}", "{ a: /abc/ }", expectedBounds); + expected.intervals.push_back(Interval(builder.obj(), false, true)); } + checkIndexBounds("{ 'a' : { '$not' : { '$gt' : 1 } } }", expected); +} + +// {$or: [{a: 20}, {$and: [{a:1}, {b:7}]}]} -> a: [1.0, 1.0], [20.0, 20.0] +TEST(CMCollapseTreeTest, OrWithAndChild) { + OrderedIntervalList expected; + expected.intervals.push_back(Interval(BSON("" << 1.0 << "" << 1.0), true, true)); + expected.intervals.push_back(Interval(BSON("" << 20.0 << "" << 20.0), true, true)); + checkIndexBounds("{$or: [{a: 20}, {$and: [{a:1}, {b:7}]}]}", expected); +} + +// {a:20, $or: [{b:1}, {c:7}]} -> a: [20.0, 20.0] +TEST(CMCollapseTreeTest, AndWithUnindexedOrChild) { + // Logic rewrite could give a tree with root OR. + OrderedIntervalList expected; + expected.intervals.push_back(Interval(BSON("" << 20.0 << "" << 20.0), true, true)); + checkIndexBounds("{a:20, $or: [{b:1}, {c:7}]}", expected); +} + +// {$or: [{a:{$gt:2,$lt:10}}, {a:{$gt:0,$lt:5}}]} -> a: (0.0, 10.0) +TEST(CMCollapseTreeTest, OrOfAnd) { + OrderedIntervalList expected; + expected.intervals.push_back(Interval(BSON("" << 0.0 << "" << 10.0), false, false)); + checkIndexBounds("{$or: [{a:{$gt:2,$lt:10}}, {a:{$gt:0,$lt:5}}]}", expected); +} + +// {$or: [{a:{$gt:2,$lt:10}}, {a:{$gt:0,$lt:15}}, {a:{$gt:20}}]} +// -> a: (0.0, 15.0), (20.0, inf.0] +TEST(CMCollapseTreeTest, OrOfAnd2) { + OrderedIntervalList expected; + expected.intervals.push_back(Interval(BSON("" << 0.0 << "" << 15.0), false, false)); + expected.intervals.push_back(Interval(BSON("" << 20.0 << "" << INF), false, true)); + checkIndexBounds("{$or: [{a:{$gt:2,$lt:10}}, {a:{$gt:0,$lt:15}}, {a:{$gt:20}}]}", expected); +} + +// "{$or: [{a:{$gt:1,$lt:5},b:6}, {a:3,b:{$gt:0,$lt:10}}]}" -> a: (1.0, 5.0) +TEST(CMCollapseTreeTest, OrOfAnd3) { + OrderedIntervalList expected; + expected.intervals.push_back(Interval(BSON("" << 1.0 << "" << 5.0), false, false)); + checkIndexBounds("{$or: [{a:{$gt:1,$lt:5},b:6}, {a:3,b:{$gt:0,$lt:10}}]}", expected); +} + +// +// Compound shard key +// + +// "{$or: [{a:{$gt:1,$lt:5}, b:{$gt:0,$lt:3}, c:6}, " +// "{a:3, b:{$gt:1,$lt:2}, c:{$gt:0,$lt:10}}]}", +// -> a: (1.0, 5.0), b: (0.0, 3.0) +TEST(CMCollapseTreeTest, OrOfAnd4) { + IndexBounds expectedBounds; + expectedBounds.fields.push_back(OrderedIntervalList()); + expectedBounds.fields.push_back(OrderedIntervalList()); + + expectedBounds.fields[0].intervals.push_back( + Interval(BSON("" << 1.0 << "" << 5.0), false, false)); + expectedBounds.fields[1].intervals.push_back( + Interval(BSON("" << 0.0 << "" << 3.0), false, false)); + + checkIndexBoundsWithKey("{a: 1, b: 1}", // shard key + "{$or: [{a:{$gt:1,$lt:5}, b:{$gt:0,$lt:3}, c:6}, " + "{a:3, b:{$gt:1,$lt:2}, c:{$gt:0,$lt:10}}]}", + expectedBounds); +} + +// "{$or: [{a:{$gt:1,$lt:5}, c:6}, " +// "{a:3, b:{$gt:1,$lt:2}, c:{$gt:0,$lt:10}}]}")); +// -> +TEST(CMCollapseTreeTest, OrOfAnd5) { + IndexBounds expectedBounds; + expectedBounds.fields.push_back(OrderedIntervalList()); + expectedBounds.fields.push_back(OrderedIntervalList()); + + expectedBounds.fields[0].intervals.push_back( + Interval(BSON("" << 1.0 << "" << 5.0), false, false)); + BSONObjBuilder builder; + builder.appendMinKey(""); + builder.appendMaxKey(""); + expectedBounds.fields[1].intervals.push_back(Interval(builder.obj(), true, true)); + + checkIndexBoundsWithKey("{a: 1, b: 1}", // shard key + "{$or: [{a:{$gt:1,$lt:5}, c:6}, " + "{a:3, b:{$gt:1,$lt:2}, c:{$gt:0,$lt:10}}]}", + expectedBounds); +} + +// {$or: [{a:{$in:[1]},b:{$in:[1]}}, {a:{$in:[1,5]},b:{$in:[1,5]}}]} +// -> a: [1], [5]; b: [1], [5] +TEST(CMCollapseTreeTest, OrOfAnd6) { + IndexBounds expectedBounds; + expectedBounds.fields.push_back(OrderedIntervalList()); + expectedBounds.fields.push_back(OrderedIntervalList()); + + // a: [1], [5] + expectedBounds.fields[0].intervals.push_back( + Interval(BSON("" << 1.0 << "" << 1.0), true, true)); + expectedBounds.fields[0].intervals.push_back( + Interval(BSON("" << 5.0 << "" << 5.0), true, true)); + + // b: [1], [5] + expectedBounds.fields[1].intervals.push_back( + Interval(BSON("" << 1.0 << "" << 1.0), true, true)); + expectedBounds.fields[1].intervals.push_back( + Interval(BSON("" << 5.0 << "" << 5.0), true, true)); + + checkIndexBoundsWithKey("{a: 1, b: 1}", // shard key + "{$or: [{a:{$in:[1]},b:{$in:[1]}}, {a:{$in:[1,5]},b:{$in:[1,5]}}]}", + expectedBounds); +} + +// +// Array operators +// + +// {a : {$elemMatch: {b:1}}} -> a.b: [MinKey, MaxKey] +// Shard key doesn't allow multikey, but query on array should succeed without error. +TEST(CMCollapseTreeTest, ElemMatchOneField) { + IndexBounds expectedBounds; + expectedBounds.fields.push_back(OrderedIntervalList()); + OrderedIntervalList& oil = expectedBounds.fields.front(); + oil.intervals.push_back(Interval(BSON("" << 1 << "" << 1), true, true)); + checkIndexBoundsWithKey("{'a.b': 1}", "{a : {$elemMatch: {b:1}}}", expectedBounds); +} + +// {foo: {$all: [ {$elemMatch: {a:1, b:1}}, {$elemMatch: {a:2, b:2}}]}} +// -> foo.a: [1, 1] +// Or -> foo.a: [2, 2] +TEST(CMCollapseTreeTest, BasicAllElemMatch) { + Interval expectedInterval(BSON("" << 1 << "" << 1), true, true); + + const char* queryStr = "{foo: {$all: [ {$elemMatch: {a:1, b:1}} ]}}"; + unique_ptr<CanonicalQuery> query(canonicalize(queryStr)); + ASSERT(query.get() != NULL); + + BSONObj key = fromjson("{'foo.a': 1}"); + + IndexBounds indexBounds = ChunkManager::getIndexBoundsForQuery(key, query.get()); + ASSERT_EQUALS(indexBounds.size(), 1U); + const OrderedIntervalList& oil = indexBounds.fields.front(); + ASSERT_EQUALS(oil.intervals.size(), 1U); + const Interval& interval = oil.intervals.front(); + + // Choose one of the two possible solutions. + // Two solutions differ only by assignment of index tags. + ASSERT(Interval::INTERVAL_EQUALS == interval.compare(expectedInterval)); +} + +// {a : [1, 2, 3]} -> a: [1, 1], [[1, 2, 3], [1, 2, 3]] +TEST(CMCollapseTreeTest, ArrayEquality) { + OrderedIntervalList expected; + expected.intervals.push_back(Interval(BSON("" << 1 << "" << 1), true, true)); + BSONArray array(BSON_ARRAY(1 << 2 << 3)); + + Interval interval(BSON("" << array << "" << array), true, true); + expected.intervals.push_back(interval); + checkIndexBounds("{a : [1, 2, 3]}", expected); +} + + +// +// Features: Regex, $where, $text, hashed key +// + +// { a: /abc/ } -> a: ["", {}), [/abc/, /abc/] +TEST(CMCollapseTreeTest, Regex) { + OrderedIntervalList expected; + expected.intervals.push_back(Interval(BSON("" + << "" + << "" << BSONObj()), + true, + false)); + BSONObjBuilder builder; + builder.appendRegex("", "abc"); + builder.appendRegex("", "abc"); + expected.intervals.push_back(Interval(builder.obj(), true, true)); + checkIndexBounds("{ a: /abc/ }", expected); +} + +// {$where: 'this.credits == this.debits' } +TEST(CMCollapseTreeTest, Where) { + OrderedIntervalList expected; + BSONObjBuilder builder; + builder.appendMinKey(""); + builder.appendMaxKey(""); + expected.intervals.push_back(Interval(builder.obj(), true, true)); + checkIndexBounds("{$where: 'this.credits == this.debits' }", expected); +} + +// { $text: { $search: "coffee -cake" } } +TEST(CMCollapseTreeTest, Text) { + OrderedIntervalList expected; + BSONObjBuilder builder; + builder.appendMinKey(""); + builder.appendMaxKey(""); + expected.intervals.push_back(Interval(builder.obj(), true, true)); + checkIndexBounds("{ $text: { $search: 'coffee -cake' } }", expected); +} + +// { a: 2, $text: { $search: "leche", $language: "es" } } +TEST(CMCollapseTreeTest, TextWithQuery) { + OrderedIntervalList expected; + BSONObjBuilder builder; + builder.appendMinKey(""); + builder.appendMaxKey(""); + expected.intervals.push_back(Interval(builder.obj(), true, true)); + checkIndexBounds("{ a: 2, $text: { $search: 'leche', $language: 'es' } }", expected); +} + +// { a: 0 } -> hashed a: [hash(0), hash(0)] +TEST(CMCollapseTreeTest, HashedSinglePoint) { + const char* queryStr = "{ a: 0 }"; + unique_ptr<CanonicalQuery> query(canonicalize(queryStr)); + ASSERT(query.get() != NULL); + + BSONObj key = fromjson("{a: 'hashed'}"); + + IndexBounds indexBounds = ChunkManager::getIndexBoundsForQuery(key, query.get()); + ASSERT_EQUALS(indexBounds.size(), 1U); + const OrderedIntervalList& oil = indexBounds.fields.front(); + ASSERT_EQUALS(oil.intervals.size(), 1U); + const Interval& interval = oil.intervals.front(); + ASSERT(interval.isPoint()); +} + +// { a: { $lt: 2, $gt: 1} } -> hashed a: [Minkey, Maxkey] +TEST(CMCollapseTreeTest, HashedRange) { + IndexBounds expectedBounds; + expectedBounds.fields.push_back(OrderedIntervalList()); + OrderedIntervalList& expectedOil = expectedBounds.fields.front(); + BSONObjBuilder builder; + builder.appendMinKey(""); + builder.appendMaxKey(""); + expectedOil.intervals.push_back(Interval(builder.obj(), true, true)); + checkIndexBoundsWithKey("{a: 'hashed'}", "{ a: { $lt: 2, $gt: 1} }", expectedBounds); +} + +// { a: /abc/ } -> hashed a: [Minkey, Maxkey] +TEST(CMCollapseTreeTest, HashedRegex) { + IndexBounds expectedBounds; + expectedBounds.fields.push_back(OrderedIntervalList()); + OrderedIntervalList& expectedOil = expectedBounds.fields.front(); + BSONObjBuilder builder; + builder.appendMinKey(""); + builder.appendMaxKey(""); + expectedOil.intervals.push_back(Interval(builder.obj(), true, true)); + + checkIndexBoundsWithKey("{a: 'hashed'}", "{ a: /abc/ }", expectedBounds); +} - /** - * KeyPattern key bounds generation test - */ - - void CheckBoundList(const BoundList& list, const BoundList& expected) { - ASSERT_EQUALS(list.size(), expected.size()); - for (size_t i = 0; i < list.size(); i++) { - ASSERT_EQUALS(list[i].first.woCompare(expected[i].first), 0); - ASSERT_EQUALS(list[i].second.woCompare(expected[i].second), 0); - } - } - - // Key { a: 1 }, Bounds a: [0] - // => { a: 0 } -> { a: 0 } - TEST(CMKeyBoundsTest, Basic) { - IndexBounds indexBounds; - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields.front().intervals.push_back( - Interval(BSON("" << 0 << "" << 0), true, true)); - - BoundList expectedList; - expectedList.push_back(make_pair(fromjson("{a: 0}"), fromjson("{a: 0}"))); - - ShardKeyPattern skeyPattern(fromjson("{a: 1}")); - BoundList list = skeyPattern.flattenBounds(indexBounds); - CheckBoundList(list, expectedList); - } - - // Key { a: 1 }, Bounds a: [2, 3) - // => { a: 2 } -> { a: 3 } // bound inclusion is ignored. - TEST(CMKeyBoundsTest, SingleInterval) { - IndexBounds indexBounds; - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields.front().intervals.push_back( - Interval(BSON("" << 2 << "" << 3), true, false)); - - BoundList expectedList; - expectedList.push_back(make_pair(fromjson("{a: 2}"), fromjson("{a: 3}"))); - - ShardKeyPattern skeyPattern(fromjson("{a: 1}")); - BoundList list = skeyPattern.flattenBounds(indexBounds); - CheckBoundList(list, expectedList); - } - - // Key { a: 1, b: 1, c: 1 }, Bounds a: [2, 3), b: [2, 3), c: [2: 3) - // => { a: 2, b: 2, c: 2 } -> { a: 3, b: 3, c: 3 } - TEST(CMKeyBoundsTest, MultiIntervals) { - IndexBounds indexBounds; - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields[0].intervals.push_back( - Interval(BSON("" << 2 << "" << 3), true, false)); - indexBounds.fields[1].intervals.push_back( - Interval(BSON("" << 2 << "" << 3), true, false)); - indexBounds.fields[2].intervals.push_back( - Interval(BSON("" << 2 << "" << 3), true, false)); - - BoundList expectedList; - expectedList.push_back(make_pair( - fromjson("{ a: 2, b: 2, c: 2 }"), - fromjson("{ a: 3, b: 3, c: 3 }"))); - - ShardKeyPattern skeyPattern(fromjson("{a: 1, b: 1, c: 1}")); - BoundList list = skeyPattern.flattenBounds(indexBounds); - CheckBoundList(list, expectedList); - } - - // Key { a: 1, b: 1, c: 1 }, Bounds a: [0, 0], b: { $in: [4, 5, 6] }, c: [2: 3) - // => { a: 0, b: 4, c: 2 } -> { a: 0, b: 4, c: 3 } - // { a: 0, b: 5, c: 2 } -> { a: 0, b: 5, c: 3 } - // { a: 0, b: 6, c: 2 } -> { a: 0, b: 6, c: 3 } - TEST(CMKeyBoundsTest, IntervalExpansion) { - IndexBounds indexBounds; - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields.push_back(OrderedIntervalList()); - - indexBounds.fields[0].intervals.push_back( - Interval(BSON("" << 0 << "" << 0), true, true)); - - indexBounds.fields[1].intervals.push_back( - Interval(BSON("" << 4 << "" << 4), true, true)); - indexBounds.fields[1].intervals.push_back( - Interval(BSON("" << 5 << "" << 5), true, true)); - indexBounds.fields[1].intervals.push_back( - Interval(BSON("" << 6 << "" << 6), true, true)); - - indexBounds.fields[2].intervals.push_back( - Interval(BSON("" << 2 << "" << 3), true, false)); - - BoundList expectedList; - expectedList.push_back(make_pair( - fromjson("{ a: 0, b: 4, c: 2 }"), - fromjson("{ a: 0, b: 4, c: 3 }"))); - expectedList.push_back(make_pair( - fromjson("{ a: 0, b: 5, c: 2 }"), - fromjson("{ a: 0, b: 5, c: 3 }"))); - expectedList.push_back(make_pair( - fromjson("{ a: 0, b: 6, c: 2 }"), - fromjson("{ a: 0, b: 6, c: 3 }"))); - - ShardKeyPattern skeyPattern(fromjson("{a: 1, b: 1, c: 1}")); - BoundList list = skeyPattern.flattenBounds(indexBounds); - CheckBoundList(list, expectedList); - } +/** + * KeyPattern key bounds generation test + */ - // Key { a: 1, b: 1, c: 1 }, Bounds a: [0, 1], b: { $in: [4, 5, 6] }, c: [2: 3) - // => { a: 0, b: 4, c: 2 } -> { a: 1, b: 6, c: 3 } - // Since field "a" is not a point, expasion after "a" is not allowed. - TEST(CMKeyBoundsTest, NonPointIntervalExpasion) { - IndexBounds indexBounds; - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields.push_back(OrderedIntervalList()); - indexBounds.fields.push_back(OrderedIntervalList()); - - indexBounds.fields[0].intervals.push_back( - Interval(BSON("" << 0 << "" << 1), true, true)); - - indexBounds.fields[1].intervals.push_back( - Interval(BSON("" << 4 << "" << 4), true, true)); - indexBounds.fields[1].intervals.push_back( - Interval(BSON("" << 5 << "" << 5), true, true)); - indexBounds.fields[1].intervals.push_back( - Interval(BSON("" << 6 << "" << 6), true, true)); - - indexBounds.fields[2].intervals.push_back( - Interval(BSON("" << 2 << "" << 3), true, false)); - - BoundList expectedList; - expectedList.push_back(make_pair( - fromjson("{ a: 0, b: 4, c: 2 }"), - fromjson("{ a: 1, b: 6, c: 3 }"))); - - ShardKeyPattern skeyPattern(fromjson("{a: 1, b: 1, c: 1}")); - BoundList list = skeyPattern.flattenBounds(indexBounds); - CheckBoundList(list, expectedList); +void CheckBoundList(const BoundList& list, const BoundList& expected) { + ASSERT_EQUALS(list.size(), expected.size()); + for (size_t i = 0; i < list.size(); i++) { + ASSERT_EQUALS(list[i].first.woCompare(expected[i].first), 0); + ASSERT_EQUALS(list[i].second.woCompare(expected[i].second), 0); } - -} // namespace +} + +// Key { a: 1 }, Bounds a: [0] +// => { a: 0 } -> { a: 0 } +TEST(CMKeyBoundsTest, Basic) { + IndexBounds indexBounds; + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields.front().intervals.push_back(Interval(BSON("" << 0 << "" << 0), true, true)); + + BoundList expectedList; + expectedList.push_back(make_pair(fromjson("{a: 0}"), fromjson("{a: 0}"))); + + ShardKeyPattern skeyPattern(fromjson("{a: 1}")); + BoundList list = skeyPattern.flattenBounds(indexBounds); + CheckBoundList(list, expectedList); +} + +// Key { a: 1 }, Bounds a: [2, 3) +// => { a: 2 } -> { a: 3 } // bound inclusion is ignored. +TEST(CMKeyBoundsTest, SingleInterval) { + IndexBounds indexBounds; + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields.front().intervals.push_back(Interval(BSON("" << 2 << "" << 3), true, false)); + + BoundList expectedList; + expectedList.push_back(make_pair(fromjson("{a: 2}"), fromjson("{a: 3}"))); + + ShardKeyPattern skeyPattern(fromjson("{a: 1}")); + BoundList list = skeyPattern.flattenBounds(indexBounds); + CheckBoundList(list, expectedList); +} + +// Key { a: 1, b: 1, c: 1 }, Bounds a: [2, 3), b: [2, 3), c: [2: 3) +// => { a: 2, b: 2, c: 2 } -> { a: 3, b: 3, c: 3 } +TEST(CMKeyBoundsTest, MultiIntervals) { + IndexBounds indexBounds; + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields[0].intervals.push_back(Interval(BSON("" << 2 << "" << 3), true, false)); + indexBounds.fields[1].intervals.push_back(Interval(BSON("" << 2 << "" << 3), true, false)); + indexBounds.fields[2].intervals.push_back(Interval(BSON("" << 2 << "" << 3), true, false)); + + BoundList expectedList; + expectedList.push_back( + make_pair(fromjson("{ a: 2, b: 2, c: 2 }"), fromjson("{ a: 3, b: 3, c: 3 }"))); + + ShardKeyPattern skeyPattern(fromjson("{a: 1, b: 1, c: 1}")); + BoundList list = skeyPattern.flattenBounds(indexBounds); + CheckBoundList(list, expectedList); +} + +// Key { a: 1, b: 1, c: 1 }, Bounds a: [0, 0], b: { $in: [4, 5, 6] }, c: [2: 3) +// => { a: 0, b: 4, c: 2 } -> { a: 0, b: 4, c: 3 } +// { a: 0, b: 5, c: 2 } -> { a: 0, b: 5, c: 3 } +// { a: 0, b: 6, c: 2 } -> { a: 0, b: 6, c: 3 } +TEST(CMKeyBoundsTest, IntervalExpansion) { + IndexBounds indexBounds; + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields.push_back(OrderedIntervalList()); + + indexBounds.fields[0].intervals.push_back(Interval(BSON("" << 0 << "" << 0), true, true)); + + indexBounds.fields[1].intervals.push_back(Interval(BSON("" << 4 << "" << 4), true, true)); + indexBounds.fields[1].intervals.push_back(Interval(BSON("" << 5 << "" << 5), true, true)); + indexBounds.fields[1].intervals.push_back(Interval(BSON("" << 6 << "" << 6), true, true)); + + indexBounds.fields[2].intervals.push_back(Interval(BSON("" << 2 << "" << 3), true, false)); + + BoundList expectedList; + expectedList.push_back( + make_pair(fromjson("{ a: 0, b: 4, c: 2 }"), fromjson("{ a: 0, b: 4, c: 3 }"))); + expectedList.push_back( + make_pair(fromjson("{ a: 0, b: 5, c: 2 }"), fromjson("{ a: 0, b: 5, c: 3 }"))); + expectedList.push_back( + make_pair(fromjson("{ a: 0, b: 6, c: 2 }"), fromjson("{ a: 0, b: 6, c: 3 }"))); + + ShardKeyPattern skeyPattern(fromjson("{a: 1, b: 1, c: 1}")); + BoundList list = skeyPattern.flattenBounds(indexBounds); + CheckBoundList(list, expectedList); +} + +// Key { a: 1, b: 1, c: 1 }, Bounds a: [0, 1], b: { $in: [4, 5, 6] }, c: [2: 3) +// => { a: 0, b: 4, c: 2 } -> { a: 1, b: 6, c: 3 } +// Since field "a" is not a point, expasion after "a" is not allowed. +TEST(CMKeyBoundsTest, NonPointIntervalExpasion) { + IndexBounds indexBounds; + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields.push_back(OrderedIntervalList()); + indexBounds.fields.push_back(OrderedIntervalList()); + + indexBounds.fields[0].intervals.push_back(Interval(BSON("" << 0 << "" << 1), true, true)); + + indexBounds.fields[1].intervals.push_back(Interval(BSON("" << 4 << "" << 4), true, true)); + indexBounds.fields[1].intervals.push_back(Interval(BSON("" << 5 << "" << 5), true, true)); + indexBounds.fields[1].intervals.push_back(Interval(BSON("" << 6 << "" << 6), true, true)); + + indexBounds.fields[2].intervals.push_back(Interval(BSON("" << 2 << "" << 3), true, false)); + + BoundList expectedList; + expectedList.push_back( + make_pair(fromjson("{ a: 0, b: 4, c: 2 }"), fromjson("{ a: 1, b: 6, c: 3 }"))); + + ShardKeyPattern skeyPattern(fromjson("{a: 1, b: 1, c: 1}")); + BoundList list = skeyPattern.flattenBounds(indexBounds); + CheckBoundList(list, expectedList); +} + +} // namespace diff --git a/src/mongo/s/chunk_version.h b/src/mongo/s/chunk_version.h index 41b9e03bae2..fef1a581909 100644 --- a/src/mongo/s/chunk_version.h +++ b/src/mongo/s/chunk_version.h @@ -32,428 +32,423 @@ namespace mongo { - /** - * ChunkVersions consist of a major/minor version scoped to a version epoch - * - * Version configurations (format: major version, epoch): - * - * 1. (0, 0) - collection is dropped. - * 2. (0, n), n > 0 - applicable only to shardVersion; shard has no chunk. - * 3. (n, 0), n > 0 - invalid configuration. - * 4. (n, m), n > 0, m > 0 - normal sharded collection version. - * - * TODO: This is a "manual type" but, even so, still needs to comform to what's - * expected from types. - */ - struct ChunkVersion { - - union { - struct { - int _minor; - int _major; - }; - unsigned long long _combined; +/** + * ChunkVersions consist of a major/minor version scoped to a version epoch + * + * Version configurations (format: major version, epoch): + * + * 1. (0, 0) - collection is dropped. + * 2. (0, n), n > 0 - applicable only to shardVersion; shard has no chunk. + * 3. (n, 0), n > 0 - invalid configuration. + * 4. (n, m), n > 0, m > 0 - normal sharded collection version. + * + * TODO: This is a "manual type" but, even so, still needs to comform to what's + * expected from types. + */ +struct ChunkVersion { + union { + struct { + int _minor; + int _major; }; - OID _epoch; - - ChunkVersion() : _minor(0), _major(0), _epoch(OID()) {} - - // - // Constructors shouldn't have default parameters here, since it's vital we track from - // here on the epochs of versions, even if not used. - // + unsigned long long _combined; + }; + OID _epoch; - ChunkVersion( int major, int minor, const OID& epoch ) - : _minor(minor),_major(major), _epoch(epoch) { - } + ChunkVersion() : _minor(0), _major(0), _epoch(OID()) {} - static ChunkVersion DROPPED() { - return ChunkVersion( 0, 0, OID() ); // dropped OID is zero time, zero machineId/inc - } + // + // Constructors shouldn't have default parameters here, since it's vital we track from + // here on the epochs of versions, even if not used. + // - static ChunkVersion UNSHARDED() { - // TODO: Distinguish between these cases - return DROPPED(); - } + ChunkVersion(int major, int minor, const OID& epoch) + : _minor(minor), _major(major), _epoch(epoch) {} - static ChunkVersion IGNORED() { - ChunkVersion version = ChunkVersion(); - version._epoch.init(Date_t(), true); // ignored OID is zero time, max machineId/inc - return version; - } + static ChunkVersion DROPPED() { + return ChunkVersion(0, 0, OID()); // dropped OID is zero time, zero machineId/inc + } - static ChunkVersion fromDeprecatedLong(unsigned long long num, const OID& epoch) { - ChunkVersion version(0, 0, epoch); - version._combined = num; - return version; - } + static ChunkVersion UNSHARDED() { + // TODO: Distinguish between these cases + return DROPPED(); + } - static bool isIgnoredVersion( const ChunkVersion& version ) { - return version.majorVersion() == 0 && version.minorVersion() == 0 - && version.epoch() == IGNORED().epoch(); - } + static ChunkVersion IGNORED() { + ChunkVersion version = ChunkVersion(); + version._epoch.init(Date_t(), true); // ignored OID is zero time, max machineId/inc + return version; + } - void incMajor() { - _major++; - _minor = 0; - } + static ChunkVersion fromDeprecatedLong(unsigned long long num, const OID& epoch) { + ChunkVersion version(0, 0, epoch); + version._combined = num; + return version; + } - void incMinor() { - _minor++; - } + static bool isIgnoredVersion(const ChunkVersion& version) { + return version.majorVersion() == 0 && version.minorVersion() == 0 && + version.epoch() == IGNORED().epoch(); + } - // Incrementing an epoch creates a new, randomly generated identifier - void incEpoch() { - _epoch = OID::gen(); - _major = 0; - _minor = 0; - } + void incMajor() { + _major++; + _minor = 0; + } - // Note: this shouldn't be used as a substitute for version except in specific cases - - // epochs make versions more complex - unsigned long long toLong() const { - return _combined; - } + void incMinor() { + _minor++; + } - bool isSet() const { - return _combined > 0; - } + // Incrementing an epoch creates a new, randomly generated identifier + void incEpoch() { + _epoch = OID::gen(); + _major = 0; + _minor = 0; + } - bool isEpochSet() const { - return _epoch.isSet(); - } + // Note: this shouldn't be used as a substitute for version except in specific cases - + // epochs make versions more complex + unsigned long long toLong() const { + return _combined; + } - std::string toString() const { - std::stringstream ss; - // Similar to month/day/year. For the most part when debugging, we care about major - // so it's first - ss << _major << "|" << _minor << "||" << _epoch; - return ss.str(); - } + bool isSet() const { + return _combined > 0; + } - int majorVersion() const { return _major; } - int minorVersion() const { return _minor; } - OID epoch() const { return _epoch; } + bool isEpochSet() const { + return _epoch.isSet(); + } - // - // Explicit comparison operators - versions with epochs have non-trivial comparisons. - // > < operators do not check epoch cases. Generally if using == we need to handle - // more complex cases. - // + std::string toString() const { + std::stringstream ss; + // Similar to month/day/year. For the most part when debugging, we care about major + // so it's first + ss << _major << "|" << _minor << "||" << _epoch; + return ss.str(); + } - bool operator>( const ChunkVersion& otherVersion ) const { - return this->_combined > otherVersion._combined; - } + int majorVersion() const { + return _major; + } + int minorVersion() const { + return _minor; + } + OID epoch() const { + return _epoch; + } - bool operator>=( const ChunkVersion& otherVersion ) const { - return this->_combined >= otherVersion._combined; - } + // + // Explicit comparison operators - versions with epochs have non-trivial comparisons. + // > < operators do not check epoch cases. Generally if using == we need to handle + // more complex cases. + // - bool operator<( const ChunkVersion& otherVersion ) const { - return this->_combined < otherVersion._combined; - } + bool operator>(const ChunkVersion& otherVersion) const { + return this->_combined > otherVersion._combined; + } - bool operator<=( const ChunkVersion& otherVersion ) const { - return this->_combined <= otherVersion._combined; - } + bool operator>=(const ChunkVersion& otherVersion) const { + return this->_combined >= otherVersion._combined; + } - // - // Equivalence comparison types. - // + bool operator<(const ChunkVersion& otherVersion) const { + return this->_combined < otherVersion._combined; + } - // Can we write to this data and not have a problem? - bool isWriteCompatibleWith( const ChunkVersion& otherVersion ) const { - if( ! hasEqualEpoch( otherVersion ) ) return false; - return otherVersion._major == _major; - } + bool operator<=(const ChunkVersion& otherVersion) const { + return this->_combined <= otherVersion._combined; + } - // Is this the same version? - bool equals( const ChunkVersion& otherVersion ) const { - if( ! hasEqualEpoch( otherVersion ) ) return false; - return otherVersion._combined == _combined; - } + // + // Equivalence comparison types. + // - /** - * Returns true if the otherVersion is the same as this version and enforces strict epoch - * checking (empty epochs are not wildcards). - */ - bool isStrictlyEqualTo( const ChunkVersion& otherVersion ) const { - if ( otherVersion._epoch != _epoch ) - return false; - return otherVersion._combined == _combined; - } + // Can we write to this data and not have a problem? + bool isWriteCompatibleWith(const ChunkVersion& otherVersion) const { + if (!hasEqualEpoch(otherVersion)) + return false; + return otherVersion._major == _major; + } - /** - * Returns true if this version is (strictly) in the same epoch as the other version and - * this version is older. Returns false if we're not sure because the epochs are different - * or if this version is newer. - */ - bool isOlderThan( const ChunkVersion& otherVersion ) const { - if ( otherVersion._epoch != _epoch ) - return false; - - if ( _major != otherVersion._major ) - return _major < otherVersion._major; - - return _minor < otherVersion._minor; - } + // Is this the same version? + bool equals(const ChunkVersion& otherVersion) const { + if (!hasEqualEpoch(otherVersion)) + return false; + return otherVersion._combined == _combined; + } - // Is this in the same epoch? - bool hasEqualEpoch( const ChunkVersion& otherVersion ) const { - return hasEqualEpoch( otherVersion._epoch ); - } + /** + * Returns true if the otherVersion is the same as this version and enforces strict epoch + * checking (empty epochs are not wildcards). + */ + bool isStrictlyEqualTo(const ChunkVersion& otherVersion) const { + if (otherVersion._epoch != _epoch) + return false; + return otherVersion._combined == _combined; + } - bool hasEqualEpoch( const OID& otherEpoch ) const { - return _epoch == otherEpoch; - } + /** + * Returns true if this version is (strictly) in the same epoch as the other version and + * this version is older. Returns false if we're not sure because the epochs are different + * or if this version is newer. + */ + bool isOlderThan(const ChunkVersion& otherVersion) const { + if (otherVersion._epoch != _epoch) + return false; - // - // BSON input/output - // - // The idea here is to make the BSON input style very flexible right now, so we - // can then tighten it up in the next version. We can accept either a BSONObject field - // with version and epoch, or version and epoch in different fields (either is optional). - // In this case, epoch always is stored in a field name of the version field name + "Epoch" - // - - // - // { version : <TS> } and { version : [<TS>,<OID>] } format - // - - static bool canParseBSON( const BSONElement& el, const std::string& prefix="" ){ - bool canParse; - fromBSON( el, prefix, &canParse ); - return canParse; - } + if (_major != otherVersion._major) + return _major < otherVersion._major; - static ChunkVersion fromBSON( const BSONElement& el, const std::string& prefix="" ){ - bool canParse; - return fromBSON( el, prefix, &canParse ); - } + return _minor < otherVersion._minor; + } - static ChunkVersion fromBSON( const BSONElement& el, - const std::string& prefix, - bool* canParse ) - { - *canParse = true; + // Is this in the same epoch? + bool hasEqualEpoch(const ChunkVersion& otherVersion) const { + return hasEqualEpoch(otherVersion._epoch); + } - int type = el.type(); + bool hasEqualEpoch(const OID& otherEpoch) const { + return _epoch == otherEpoch; + } - if( type == Array ){ - return fromBSON( BSONArray( el.Obj() ), canParse ); - } + // + // BSON input/output + // + // The idea here is to make the BSON input style very flexible right now, so we + // can then tighten it up in the next version. We can accept either a BSONObject field + // with version and epoch, or version and epoch in different fields (either is optional). + // In this case, epoch always is stored in a field name of the version field name + "Epoch" + // + + // + // { version : <TS> } and { version : [<TS>,<OID>] } format + // + + static bool canParseBSON(const BSONElement& el, const std::string& prefix = "") { + bool canParse; + fromBSON(el, prefix, &canParse); + return canParse; + } - if( type == jstOID ){ - return ChunkVersion( 0, 0, el.OID() ); - } + static ChunkVersion fromBSON(const BSONElement& el, const std::string& prefix = "") { + bool canParse; + return fromBSON(el, prefix, &canParse); + } - if( type == bsonTimestamp || type == Date ){ - return fromDeprecatedLong( el._numberLong(), OID() ); - } + static ChunkVersion fromBSON(const BSONElement& el, const std::string& prefix, bool* canParse) { + *canParse = true; - *canParse = false; + int type = el.type(); - return ChunkVersion( 0, 0, OID() ); + if (type == Array) { + return fromBSON(BSONArray(el.Obj()), canParse); } - // - // { version : <TS>, versionEpoch : <OID> } object format - // - - static bool canParseBSON( const BSONObj& obj, const std::string& prefix="" ){ - bool canParse; - fromBSON( obj, prefix, &canParse ); - return canParse; + if (type == jstOID) { + return ChunkVersion(0, 0, el.OID()); } - static ChunkVersion fromBSON( const BSONObj& obj, const std::string& prefix="" ){ - bool canParse; - return fromBSON( obj, prefix, &canParse ); + if (type == bsonTimestamp || type == Date) { + return fromDeprecatedLong(el._numberLong(), OID()); } - static ChunkVersion fromBSON( const BSONObj& obj, - const std::string& prefixIn, - bool* canParse ) - { - *canParse = true; + *canParse = false; - std::string prefix = prefixIn; - // "version" doesn't have a "cluster constanst" because that field is never - // written to the config. - if( prefixIn == "" && ! obj[ "version" ].eoo() ){ - prefix = (std::string)"version"; - } - // TODO: use ChunkType::DEPRECATED_lastmod() - // NOTE: type_chunk.h includes this file - else if( prefixIn == "" && ! obj["lastmod"].eoo() ){ - prefix = (std::string)"lastmod"; - } + return ChunkVersion(0, 0, OID()); + } - ChunkVersion version = fromBSON( obj[ prefix ], prefixIn, canParse ); + // + // { version : <TS>, versionEpoch : <OID> } object format + // - if( obj[ prefix + "Epoch" ].type() == jstOID ){ - version._epoch = obj[ prefix + "Epoch" ].OID(); - *canParse = true; - } + static bool canParseBSON(const BSONObj& obj, const std::string& prefix = "") { + bool canParse; + fromBSON(obj, prefix, &canParse); + return canParse; + } - return version; - } + static ChunkVersion fromBSON(const BSONObj& obj, const std::string& prefix = "") { + bool canParse; + return fromBSON(obj, prefix, &canParse); + } - // - // { version : [<TS>, <OID>] } format - // + static ChunkVersion fromBSON(const BSONObj& obj, const std::string& prefixIn, bool* canParse) { + *canParse = true; - static bool canParseBSON( const BSONArray& arr ){ - bool canParse; - fromBSON( arr, &canParse ); - return canParse; + std::string prefix = prefixIn; + // "version" doesn't have a "cluster constanst" because that field is never + // written to the config. + if (prefixIn == "" && !obj["version"].eoo()) { + prefix = (std::string) "version"; + } + // TODO: use ChunkType::DEPRECATED_lastmod() + // NOTE: type_chunk.h includes this file + else if (prefixIn == "" && !obj["lastmod"].eoo()) { + prefix = (std::string) "lastmod"; } - static ChunkVersion fromBSON( const BSONArray& arr ){ - bool canParse; - return fromBSON( arr, &canParse ); + ChunkVersion version = fromBSON(obj[prefix], prefixIn, canParse); + + if (obj[prefix + "Epoch"].type() == jstOID) { + version._epoch = obj[prefix + "Epoch"].OID(); + *canParse = true; } - static ChunkVersion fromBSON( const BSONArray& arr, - bool* canParse ) - { - *canParse = false; + return version; + } - ChunkVersion version; + // + // { version : [<TS>, <OID>] } format + // - BSONObjIterator it( arr ); - if( ! it.more() ) return version; + static bool canParseBSON(const BSONArray& arr) { + bool canParse; + fromBSON(arr, &canParse); + return canParse; + } - version = fromBSON( it.next(), "", canParse ); - if( ! (*canParse) ) return version; + static ChunkVersion fromBSON(const BSONArray& arr) { + bool canParse; + return fromBSON(arr, &canParse); + } - *canParse = true; + static ChunkVersion fromBSON(const BSONArray& arr, bool* canParse) { + *canParse = false; - if( ! it.more() ) return version; - BSONElement next = it.next(); - if( next.type() != jstOID ) return version; + ChunkVersion version; - version._epoch = next.OID(); + BSONObjIterator it(arr); + if (!it.more()) + return version; + version = fromBSON(it.next(), "", canParse); + if (!(*canParse)) return version; - } - enum VersionChoice { - VersionChoice_Local, - VersionChoice_Remote, - VersionChoice_Unknown - }; + *canParse = true; - /** - * Compares a remotely-loaded version 'remoteVersion' to the latest local version of a - * collection, 'localVersion', and returns the newest. - * - * Because it isn't clear during epoch changes which epoch is newer, the local version - * before the reload occurred, 'prevLocalVersion', is used to determine whether the remote - * epoch is definitely newer, or we're not sure. - */ - static VersionChoice chooseNewestVersion( ChunkVersion prevLocalVersion, - ChunkVersion localVersion, - ChunkVersion remoteVersion ) - { - OID prevEpoch = prevLocalVersion.epoch(); - OID localEpoch = localVersion.epoch(); - OID remoteEpoch = remoteVersion.epoch(); - - // Everything changed in-flight, so we need to try again - if ( prevEpoch != localEpoch && localEpoch != remoteEpoch ) { - return VersionChoice_Unknown; - } + if (!it.more()) + return version; + BSONElement next = it.next(); + if (next.type() != jstOID) + return version; - // We're in the same (zero) epoch as the latest metadata, nothing to do - if ( localEpoch == remoteEpoch && !remoteEpoch.isSet() ) { - return VersionChoice_Local; - } + version._epoch = next.OID(); - // We're in the same (non-zero) epoch as the latest metadata, so increment the version - if ( localEpoch == remoteEpoch && remoteEpoch.isSet() ) { + return version; + } - // Use the newer version if possible - if ( localVersion < remoteVersion ) { - return VersionChoice_Remote; - } - else { - return VersionChoice_Local; - } - } + enum VersionChoice { VersionChoice_Local, VersionChoice_Remote, VersionChoice_Unknown }; - // We're now sure we're installing a new epoch and the epoch didn't change during reload - dassert( prevEpoch == localEpoch && localEpoch != remoteEpoch ); - return VersionChoice_Remote; + /** + * Compares a remotely-loaded version 'remoteVersion' to the latest local version of a + * collection, 'localVersion', and returns the newest. + * + * Because it isn't clear during epoch changes which epoch is newer, the local version + * before the reload occurred, 'prevLocalVersion', is used to determine whether the remote + * epoch is definitely newer, or we're not sure. + */ + static VersionChoice chooseNewestVersion(ChunkVersion prevLocalVersion, + ChunkVersion localVersion, + ChunkVersion remoteVersion) { + OID prevEpoch = prevLocalVersion.epoch(); + OID localEpoch = localVersion.epoch(); + OID remoteEpoch = remoteVersion.epoch(); + + // Everything changed in-flight, so we need to try again + if (prevEpoch != localEpoch && localEpoch != remoteEpoch) { + return VersionChoice_Unknown; } - // - // Currently our BSON output is to two different fields, to cleanly work with older - // versions that know nothing about epochs. - // + // We're in the same (zero) epoch as the latest metadata, nothing to do + if (localEpoch == remoteEpoch && !remoteEpoch.isSet()) { + return VersionChoice_Local; + } - BSONObj toBSONWithPrefix( const std::string& prefixIn ) const { - BSONObjBuilder b; + // We're in the same (non-zero) epoch as the latest metadata, so increment the version + if (localEpoch == remoteEpoch && remoteEpoch.isSet()) { + // Use the newer version if possible + if (localVersion < remoteVersion) { + return VersionChoice_Remote; + } else { + return VersionChoice_Local; + } + } - std::string prefix = prefixIn; - if( prefix == "" ) prefix = "version"; + // We're now sure we're installing a new epoch and the epoch didn't change during reload + dassert(prevEpoch == localEpoch && localEpoch != remoteEpoch); + return VersionChoice_Remote; + } - b.appendTimestamp( prefix, _combined ); - b.append( prefix + "Epoch", _epoch ); - return b.obj(); - } + // + // Currently our BSON output is to two different fields, to cleanly work with older + // versions that know nothing about epochs. + // - void addToBSON( BSONObjBuilder& b, const std::string& prefix="" ) const { - b.appendElements( toBSONWithPrefix( prefix ) ); - } + BSONObj toBSONWithPrefix(const std::string& prefixIn) const { + BSONObjBuilder b; - void addEpochToBSON( BSONObjBuilder& b, const std::string& prefix="" ) const { - b.append( prefix + "Epoch", _epoch ); - } + std::string prefix = prefixIn; + if (prefix == "") + prefix = "version"; - BSONObj toBSON() const { - // ChunkVersion wants to be an array. - BSONArrayBuilder b; - b.appendTimestamp(_combined); - b.append(_epoch); - return b.arr(); - } + b.appendTimestamp(prefix, _combined); + b.append(prefix + "Epoch", _epoch); + return b.obj(); + } - bool parseBSON(const BSONObj& source, std::string* errMsg) { - // ChunkVersion wants to be an array. - BSONArray arrSource = static_cast<BSONArray>(source); + void addToBSON(BSONObjBuilder& b, const std::string& prefix = "") const { + b.appendElements(toBSONWithPrefix(prefix)); + } - bool canParse; - ChunkVersion version = fromBSON(arrSource, &canParse); - if (!canParse) { - *errMsg = "Could not parse version structure"; - return false; - } + void addEpochToBSON(BSONObjBuilder& b, const std::string& prefix = "") const { + b.append(prefix + "Epoch", _epoch); + } - _minor = version._minor; - _major = version._major; - _epoch = version._epoch; - return true; - } + BSONObj toBSON() const { + // ChunkVersion wants to be an array. + BSONArrayBuilder b; + b.appendTimestamp(_combined); + b.append(_epoch); + return b.arr(); + } - void clear() { - _minor = 0; - _major = 0; - _epoch = OID(); - } + bool parseBSON(const BSONObj& source, std::string* errMsg) { + // ChunkVersion wants to be an array. + BSONArray arrSource = static_cast<BSONArray>(source); - void cloneTo(ChunkVersion* other) const { - other->clear(); - other->_minor = _minor; - other->_major = _major; - other->_epoch = _epoch; + bool canParse; + ChunkVersion version = fromBSON(arrSource, &canParse); + if (!canParse) { + *errMsg = "Could not parse version structure"; + return false; } - }; + _minor = version._minor; + _major = version._major; + _epoch = version._epoch; + return true; + } - inline std::ostream& operator<<( std::ostream &s , const ChunkVersion& v) { - s << v.toString(); - return s; + void clear() { + _minor = 0; + _major = 0; + _epoch = OID(); } -} // namespace mongo + void cloneTo(ChunkVersion* other) const { + other->clear(); + other->_minor = _minor; + other->_major = _major; + other->_epoch = _epoch; + } +}; + +inline std::ostream& operator<<(std::ostream& s, const ChunkVersion& v) { + s << v.toString(); + return s; +} + +} // namespace mongo diff --git a/src/mongo/s/chunk_version_test.cpp b/src/mongo/s/chunk_version_test.cpp index 07f2b54f883..f660d3f104d 100644 --- a/src/mongo/s/chunk_version_test.cpp +++ b/src/mongo/s/chunk_version_test.cpp @@ -35,94 +35,89 @@ namespace mongo { namespace { - /** - * Tests parsing of BSON for versions. In version 2.2, this parsing is meant to be very - * flexible so different formats can be tried and enforced later. - * - * Formats are: - * - * A) { vFieldName : <TSTYPE>, [ vFieldNameEpoch : <OID> ], ... } - * B) { fieldName : [ <TSTYPE>, <OID> ], ... } - * - * vFieldName is a specifyable name - usually "version" (default) or "lastmod". <TSTYPE> is a - * type convertible to Timestamp, ideally Timestamp but also numeric. - * <OID> is a value of type OID. - * - */ - TEST(Compatibility, LegacyFormatA) { - BSONObjBuilder versionObjB; - versionObjB.appendTimestamp( "testVersion", - ChunkVersion( 1, 1, OID() ).toLong() ); - versionObjB.append( "testVersionEpoch", OID::gen() ); - BSONObj versionObj = versionObjB.obj(); - - ChunkVersion parsed = - ChunkVersion::fromBSON( versionObj[ "testVersion" ] ); - - ASSERT( ChunkVersion::canParseBSON( versionObj[ "testVersion" ] ) ); - ASSERT( parsed.majorVersion() == 1 ); - ASSERT( parsed.minorVersion() == 1 ); - ASSERT( ! parsed.epoch().isSet() ); - - parsed = ChunkVersion::fromBSON( versionObj, "testVersion" ); - - ASSERT( ChunkVersion::canParseBSON( versionObj, "testVersion" ) ); - ASSERT( parsed.majorVersion() == 1 ); - ASSERT( parsed.minorVersion() == 1 ); - ASSERT( parsed.epoch().isSet() ); - } - - TEST(Compatibility, SubArrayFormatB) { - BSONObjBuilder tsObjB; - tsObjB.appendTimestamp( "ts", ChunkVersion( 1, 1, OID() ).toLong() ); - BSONObj tsObj = tsObjB.obj(); - - BSONObjBuilder versionObjB; - BSONArrayBuilder subArrB( versionObjB.subarrayStart( "testVersion" ) ); - // Append this weird way so we're sure we get a timestamp type - subArrB.append( tsObj.firstElement() ); - subArrB.append( OID::gen() ); - subArrB.done(); - BSONObj versionObj = versionObjB.obj(); - - ChunkVersion parsed = - ChunkVersion::fromBSON( versionObj[ "testVersion" ] ); - - ASSERT( ChunkVersion::canParseBSON( versionObj[ "testVersion" ] ) ); - ASSERT( ChunkVersion::canParseBSON( BSONArray( versionObj[ "testVersion" ].Obj() ) ) ); - ASSERT( parsed.majorVersion() == 1 ); - ASSERT( parsed.minorVersion() == 1 ); - ASSERT( parsed.epoch().isSet() ); - } - - TEST(Comparison, StrictEqual) { - - OID epoch = OID::gen(); - - ASSERT( ChunkVersion( 3, 1, epoch ).isStrictlyEqualTo( ChunkVersion( 3, 1, epoch ) ) ); - ASSERT( !ChunkVersion( 3, 1, epoch ).isStrictlyEqualTo( ChunkVersion( 3, 1, OID() ) ) ); - ASSERT( !ChunkVersion( 3, 1, OID() ).isStrictlyEqualTo( ChunkVersion( 3, 1, epoch ) ) ); - ASSERT( ChunkVersion( 3, 1, OID() ).isStrictlyEqualTo( ChunkVersion( 3, 1, OID() ) ) ); - ASSERT( !ChunkVersion( 4, 2, epoch ).isStrictlyEqualTo( ChunkVersion( 4, 1, epoch ) ) ); - } - - TEST(Comparison, OlderThan) { - - OID epoch = OID::gen(); - - ASSERT( ChunkVersion( 3, 1, epoch ).isOlderThan( ChunkVersion( 4, 1, epoch ) ) ); - ASSERT( !ChunkVersion( 4, 1, epoch ).isOlderThan( ChunkVersion( 3, 1, epoch ) ) ); - - ASSERT( ChunkVersion( 3, 1, epoch ).isOlderThan( ChunkVersion( 3, 2, epoch ) ) ); - ASSERT( !ChunkVersion( 3, 2, epoch ).isOlderThan( ChunkVersion( 3, 1, epoch ) ) ); - - ASSERT( !ChunkVersion( 3, 1, epoch ).isOlderThan( ChunkVersion( 4, 1, OID() ) ) ); - ASSERT( !ChunkVersion( 4, 1, OID() ).isOlderThan( ChunkVersion( 3, 1, epoch ) ) ); - - ASSERT( ChunkVersion( 3, 2, epoch ).isOlderThan( ChunkVersion( 4, 1, epoch ) ) ); - - ASSERT( !ChunkVersion( 3, 1, epoch ).isOlderThan( ChunkVersion( 3, 1, epoch ) ) ); - } - -} // unnamed namespace -} // namespace mongo +/** + * Tests parsing of BSON for versions. In version 2.2, this parsing is meant to be very + * flexible so different formats can be tried and enforced later. + * + * Formats are: + * + * A) { vFieldName : <TSTYPE>, [ vFieldNameEpoch : <OID> ], ... } + * B) { fieldName : [ <TSTYPE>, <OID> ], ... } + * + * vFieldName is a specifyable name - usually "version" (default) or "lastmod". <TSTYPE> is a + * type convertible to Timestamp, ideally Timestamp but also numeric. + * <OID> is a value of type OID. + * + */ +TEST(Compatibility, LegacyFormatA) { + BSONObjBuilder versionObjB; + versionObjB.appendTimestamp("testVersion", ChunkVersion(1, 1, OID()).toLong()); + versionObjB.append("testVersionEpoch", OID::gen()); + BSONObj versionObj = versionObjB.obj(); + + ChunkVersion parsed = ChunkVersion::fromBSON(versionObj["testVersion"]); + + ASSERT(ChunkVersion::canParseBSON(versionObj["testVersion"])); + ASSERT(parsed.majorVersion() == 1); + ASSERT(parsed.minorVersion() == 1); + ASSERT(!parsed.epoch().isSet()); + + parsed = ChunkVersion::fromBSON(versionObj, "testVersion"); + + ASSERT(ChunkVersion::canParseBSON(versionObj, "testVersion")); + ASSERT(parsed.majorVersion() == 1); + ASSERT(parsed.minorVersion() == 1); + ASSERT(parsed.epoch().isSet()); +} + +TEST(Compatibility, SubArrayFormatB) { + BSONObjBuilder tsObjB; + tsObjB.appendTimestamp("ts", ChunkVersion(1, 1, OID()).toLong()); + BSONObj tsObj = tsObjB.obj(); + + BSONObjBuilder versionObjB; + BSONArrayBuilder subArrB(versionObjB.subarrayStart("testVersion")); + // Append this weird way so we're sure we get a timestamp type + subArrB.append(tsObj.firstElement()); + subArrB.append(OID::gen()); + subArrB.done(); + BSONObj versionObj = versionObjB.obj(); + + ChunkVersion parsed = ChunkVersion::fromBSON(versionObj["testVersion"]); + + ASSERT(ChunkVersion::canParseBSON(versionObj["testVersion"])); + ASSERT(ChunkVersion::canParseBSON(BSONArray(versionObj["testVersion"].Obj()))); + ASSERT(parsed.majorVersion() == 1); + ASSERT(parsed.minorVersion() == 1); + ASSERT(parsed.epoch().isSet()); +} + +TEST(Comparison, StrictEqual) { + OID epoch = OID::gen(); + + ASSERT(ChunkVersion(3, 1, epoch).isStrictlyEqualTo(ChunkVersion(3, 1, epoch))); + ASSERT(!ChunkVersion(3, 1, epoch).isStrictlyEqualTo(ChunkVersion(3, 1, OID()))); + ASSERT(!ChunkVersion(3, 1, OID()).isStrictlyEqualTo(ChunkVersion(3, 1, epoch))); + ASSERT(ChunkVersion(3, 1, OID()).isStrictlyEqualTo(ChunkVersion(3, 1, OID()))); + ASSERT(!ChunkVersion(4, 2, epoch).isStrictlyEqualTo(ChunkVersion(4, 1, epoch))); +} + +TEST(Comparison, OlderThan) { + OID epoch = OID::gen(); + + ASSERT(ChunkVersion(3, 1, epoch).isOlderThan(ChunkVersion(4, 1, epoch))); + ASSERT(!ChunkVersion(4, 1, epoch).isOlderThan(ChunkVersion(3, 1, epoch))); + + ASSERT(ChunkVersion(3, 1, epoch).isOlderThan(ChunkVersion(3, 2, epoch))); + ASSERT(!ChunkVersion(3, 2, epoch).isOlderThan(ChunkVersion(3, 1, epoch))); + + ASSERT(!ChunkVersion(3, 1, epoch).isOlderThan(ChunkVersion(4, 1, OID()))); + ASSERT(!ChunkVersion(4, 1, OID()).isOlderThan(ChunkVersion(3, 1, epoch))); + + ASSERT(ChunkVersion(3, 2, epoch).isOlderThan(ChunkVersion(4, 1, epoch))); + + ASSERT(!ChunkVersion(3, 1, epoch).isOlderThan(ChunkVersion(3, 1, epoch))); +} + +} // unnamed namespace +} // namespace mongo diff --git a/src/mongo/s/client/dbclient_multi_command.cpp b/src/mongo/s/client/dbclient_multi_command.cpp index 3d7202f72ac..6fc6048d663 100644 --- a/src/mongo/s/client/dbclient_multi_command.cpp +++ b/src/mongo/s/client/dbclient_multi_command.cpp @@ -42,208 +42,201 @@ namespace mongo { - using std::unique_ptr; - using std::deque; - using std::string; - - DBClientMultiCommand::PendingCommand::PendingCommand( const ConnectionString& endpoint, - StringData dbName, - const BSONObj& cmdObj ) : - endpoint( endpoint ), - dbName( dbName.toString() ), - cmdObj( cmdObj ), - conn( NULL ), - status( Status::OK() ) { - } - - void DBClientMultiCommand::addCommand( const ConnectionString& endpoint, - StringData dbName, - const BSONSerializable& request ) { - PendingCommand* command = new PendingCommand( endpoint, dbName, request.toBSON() ); - _pendingCommands.push_back( command ); - } - - namespace { +using std::unique_ptr; +using std::deque; +using std::string; + +DBClientMultiCommand::PendingCommand::PendingCommand(const ConnectionString& endpoint, + StringData dbName, + const BSONObj& cmdObj) + : endpoint(endpoint), + dbName(dbName.toString()), + cmdObj(cmdObj), + conn(NULL), + status(Status::OK()) {} + +void DBClientMultiCommand::addCommand(const ConnectionString& endpoint, + StringData dbName, + const BSONSerializable& request) { + PendingCommand* command = new PendingCommand(endpoint, dbName, request.toBSON()); + _pendingCommands.push_back(command); +} - // - // For sanity checks of batch write operations - // +namespace { + +// +// For sanity checks of batch write operations +// + +BatchedCommandRequest::BatchType getBatchWriteType(const BSONObj& cmdObj) { + string cmdName = cmdObj.firstElement().fieldName(); + if (cmdName == "insert") + return BatchedCommandRequest::BatchType_Insert; + if (cmdName == "update") + return BatchedCommandRequest::BatchType_Update; + if (cmdName == "delete") + return BatchedCommandRequest::BatchType_Delete; + return BatchedCommandRequest::BatchType_Unknown; +} - BatchedCommandRequest::BatchType getBatchWriteType( const BSONObj& cmdObj ) { - string cmdName = cmdObj.firstElement().fieldName(); - if ( cmdName == "insert" ) return BatchedCommandRequest::BatchType_Insert; - if ( cmdName == "update" ) return BatchedCommandRequest::BatchType_Update; - if ( cmdName == "delete" ) return BatchedCommandRequest::BatchType_Delete; - return BatchedCommandRequest::BatchType_Unknown; - } +bool isBatchWriteCommand(const BSONObj& cmdObj) { + return getBatchWriteType(cmdObj) != BatchedCommandRequest::BatchType_Unknown; +} - bool isBatchWriteCommand( const BSONObj& cmdObj ) { - return getBatchWriteType( cmdObj ) != BatchedCommandRequest::BatchType_Unknown; - } +bool hasBatchWriteFeature(DBClientBase* conn) { + return conn->getMinWireVersion() <= BATCH_COMMANDS && + conn->getMaxWireVersion() >= BATCH_COMMANDS; +} +} - bool hasBatchWriteFeature( DBClientBase* conn ) { - return conn->getMinWireVersion() <= BATCH_COMMANDS - && conn->getMaxWireVersion() >= BATCH_COMMANDS; - } +// THROWS +static void sayAsCmd(DBClientBase* conn, StringData dbName, const BSONObj& cmdObj) { + auto requestBuilder = + rpc::makeRequestBuilder(conn->getClientRPCProtocols(), conn->getServerRPCProtocols()); + BSONObj upconvertedCmd; + BSONObj upconvertedMetadata; + + // Previous implementation had hardcoded flags of 0 - more specifically, writes + // are never secondaryOk. + std::tie(upconvertedCmd, upconvertedMetadata) = + uassertStatusOK(rpc::upconvertRequestMetadata(cmdObj, 0)); + + BSONObjBuilder metadataBob; + metadataBob.appendElements(upconvertedMetadata); + if (conn->getRequestMetadataWriter()) { + conn->getRequestMetadataWriter()(&metadataBob); } - // THROWS - static void sayAsCmd( DBClientBase* conn, StringData dbName, const BSONObj& cmdObj ) { - auto requestBuilder = rpc::makeRequestBuilder(conn->getClientRPCProtocols(), - conn->getServerRPCProtocols()); - BSONObj upconvertedCmd; - BSONObj upconvertedMetadata; - - // Previous implementation had hardcoded flags of 0 - more specifically, writes - // are never secondaryOk. - std::tie(upconvertedCmd, upconvertedMetadata) = uassertStatusOK( - rpc::upconvertRequestMetadata(cmdObj, 0) - ); - - BSONObjBuilder metadataBob; - metadataBob.appendElements(upconvertedMetadata); - if (conn->getRequestMetadataWriter()) { - conn->getRequestMetadataWriter()(&metadataBob); - } + requestBuilder->setDatabase(dbName); + requestBuilder->setCommandName(upconvertedCmd.firstElementFieldName()); + requestBuilder->setMetadata(metadataBob.done()); + requestBuilder->setCommandArgs(upconvertedCmd); + // Send our command + conn->say(*requestBuilder->done()); +} - requestBuilder->setDatabase(dbName); - requestBuilder->setCommandName(upconvertedCmd.firstElementFieldName()); - requestBuilder->setMetadata(metadataBob.done()); - requestBuilder->setCommandArgs(upconvertedCmd); - // Send our command - conn->say(*requestBuilder->done()); +// THROWS +static void recvAsCmd(DBClientBase* conn, Message* toRecv, BSONObj* result) { + if (!conn->recv(*toRecv)) { + // Confusingly, socket exceptions here are written to the log, not thrown. + uasserted(17255, + "error receiving write command response, " + "possible socket exception - see logs"); } - // THROWS - static void recvAsCmd( DBClientBase* conn, Message* toRecv, BSONObj* result ) { - - if ( !conn->recv( *toRecv ) ) { - // Confusingly, socket exceptions here are written to the log, not thrown. - uasserted( 17255, "error receiving write command response, " - "possible socket exception - see logs" ); - } - - auto reply = rpc::makeReply(toRecv); - - if (conn->getReplyMetadataReader()) { - conn->getReplyMetadataReader()(reply->getMetadata(), conn->getServerAddress()); - } + auto reply = rpc::makeReply(toRecv); - *result = reply->getCommandReply(); + if (conn->getReplyMetadataReader()) { + conn->getReplyMetadataReader()(reply->getMetadata(), conn->getServerAddress()); } - void DBClientMultiCommand::sendAll() { - - for ( deque<PendingCommand*>::iterator it = _pendingCommands.begin(); - it != _pendingCommands.end(); ++it ) { - - PendingCommand* command = *it; - dassert( NULL == command->conn ); - - try { - dassert( command->endpoint.type() == ConnectionString::MASTER || - command->endpoint.type() == ConnectionString::CUSTOM ); - - // TODO: Fix the pool up to take millis directly - int timeoutSecs = _timeoutMillis / 1000; - command->conn = shardConnectionPool.get( command->endpoint, timeoutSecs ); + *result = reply->getCommandReply(); +} - // Sanity check if we're sending a batch write that we're talking to a new-enough - // server. - massert(28563, str::stream() << "cannot send batch write operation to server " - << command->conn->toString(), - !isBatchWriteCommand(command->cmdObj) || - hasBatchWriteFeature(command->conn)); +void DBClientMultiCommand::sendAll() { + for (deque<PendingCommand*>::iterator it = _pendingCommands.begin(); + it != _pendingCommands.end(); + ++it) { + PendingCommand* command = *it; + dassert(NULL == command->conn); - sayAsCmd( command->conn, command->dbName, command->cmdObj ); - } - catch ( const DBException& ex ) { - command->status = ex.toStatus(); - - if ( NULL != command->conn ) { - - // Confusingly, the pool needs to know about failed connections so that it can - // invalidate other connections which might be bad. But if the connection - // doesn't seem bad, don't send it back, because we don't want to reuse it. - if ( !command->conn->isFailed() ) { - delete command->conn; - } - else { - shardConnectionPool.release( command->endpoint.toString(), command->conn ); - } - - command->conn = NULL; + try { + dassert(command->endpoint.type() == ConnectionString::MASTER || + command->endpoint.type() == ConnectionString::CUSTOM); + + // TODO: Fix the pool up to take millis directly + int timeoutSecs = _timeoutMillis / 1000; + command->conn = shardConnectionPool.get(command->endpoint, timeoutSecs); + + // Sanity check if we're sending a batch write that we're talking to a new-enough + // server. + massert(28563, + str::stream() << "cannot send batch write operation to server " + << command->conn->toString(), + !isBatchWriteCommand(command->cmdObj) || hasBatchWriteFeature(command->conn)); + + sayAsCmd(command->conn, command->dbName, command->cmdObj); + } catch (const DBException& ex) { + command->status = ex.toStatus(); + + if (NULL != command->conn) { + // Confusingly, the pool needs to know about failed connections so that it can + // invalidate other connections which might be bad. But if the connection + // doesn't seem bad, don't send it back, because we don't want to reuse it. + if (!command->conn->isFailed()) { + delete command->conn; + } else { + shardConnectionPool.release(command->endpoint.toString(), command->conn); } + + command->conn = NULL; } } } +} - int DBClientMultiCommand::numPending() const { - return static_cast<int>( _pendingCommands.size() ); - } - - Status DBClientMultiCommand::recvAny( ConnectionString* endpoint, BSONSerializable* response ) { - - unique_ptr<PendingCommand> command( _pendingCommands.front() ); - _pendingCommands.pop_front(); +int DBClientMultiCommand::numPending() const { + return static_cast<int>(_pendingCommands.size()); +} - *endpoint = command->endpoint; - if ( !command->status.isOK() ) return command->status; +Status DBClientMultiCommand::recvAny(ConnectionString* endpoint, BSONSerializable* response) { + unique_ptr<PendingCommand> command(_pendingCommands.front()); + _pendingCommands.pop_front(); - dassert( NULL != command->conn ); + *endpoint = command->endpoint; + if (!command->status.isOK()) + return command->status; - try { + dassert(NULL != command->conn); - // Holds the data and BSONObj for the command result - Message toRecv; - BSONObj result; + try { + // Holds the data and BSONObj for the command result + Message toRecv; + BSONObj result; - recvAsCmd( command->conn, &toRecv, &result ); + recvAsCmd(command->conn, &toRecv, &result); - shardConnectionPool.release( command->endpoint.toString(), command->conn ); - command->conn = NULL; + shardConnectionPool.release(command->endpoint.toString(), command->conn); + command->conn = NULL; - string errMsg; - if ( !response->parseBSON( result, &errMsg ) || !response->isValid( &errMsg ) ) { - return Status( ErrorCodes::FailedToParse, errMsg ); - } + string errMsg; + if (!response->parseBSON(result, &errMsg) || !response->isValid(&errMsg)) { + return Status(ErrorCodes::FailedToParse, errMsg); } - catch ( const DBException& ex ) { - - // Confusingly, the pool needs to know about failed connections so that it can - // invalidate other connections which might be bad. But if the connection doesn't seem - // bad, don't send it back, because we don't want to reuse it. - if ( !command->conn->isFailed() ) { - delete command->conn; - } - else { - shardConnectionPool.release( command->endpoint.toString(), command->conn ); - } - command->conn = NULL; - - return ex.toStatus(); + } catch (const DBException& ex) { + // Confusingly, the pool needs to know about failed connections so that it can + // invalidate other connections which might be bad. But if the connection doesn't seem + // bad, don't send it back, because we don't want to reuse it. + if (!command->conn->isFailed()) { + delete command->conn; + } else { + shardConnectionPool.release(command->endpoint.toString(), command->conn); } + command->conn = NULL; - return Status::OK(); + return ex.toStatus(); } - DBClientMultiCommand::~DBClientMultiCommand() { - - // Cleanup anything outstanding, do *not* return stuff to the pool, that might error - for ( deque<PendingCommand*>::iterator it = _pendingCommands.begin(); - it != _pendingCommands.end(); ++it ) { - - PendingCommand* command = *it; - - if ( NULL != command->conn ) delete command->conn; - delete command; - command = NULL; - } + return Status::OK(); +} - _pendingCommands.clear(); +DBClientMultiCommand::~DBClientMultiCommand() { + // Cleanup anything outstanding, do *not* return stuff to the pool, that might error + for (deque<PendingCommand*>::iterator it = _pendingCommands.begin(); + it != _pendingCommands.end(); + ++it) { + PendingCommand* command = *it; + + if (NULL != command->conn) + delete command->conn; + delete command; + command = NULL; } - void DBClientMultiCommand::setTimeoutMillis( int milliSecs ) { - _timeoutMillis = milliSecs; - } + _pendingCommands.clear(); +} + +void DBClientMultiCommand::setTimeoutMillis(int milliSecs) { + _timeoutMillis = milliSecs; +} } diff --git a/src/mongo/s/client/dbclient_multi_command.h b/src/mongo/s/client/dbclient_multi_command.h index 6cffa079fe7..0615fbfce66 100644 --- a/src/mongo/s/client/dbclient_multi_command.h +++ b/src/mongo/s/client/dbclient_multi_command.h @@ -35,55 +35,49 @@ namespace mongo { - /** - * A DBClientMultiCommand uses the client driver (DBClientConnections) to send and recv - * commands to different hosts in parallel. - * - * See MultiCommandDispatch for more details. - */ - class DBClientMultiCommand : public MultiCommandDispatch { - public: - - DBClientMultiCommand() : _timeoutMillis( 0 ) {} - - ~DBClientMultiCommand(); - - void addCommand( const ConnectionString& endpoint, - StringData dbName, - const BSONSerializable& request ); - - void sendAll(); +/** + * A DBClientMultiCommand uses the client driver (DBClientConnections) to send and recv + * commands to different hosts in parallel. + * + * See MultiCommandDispatch for more details. + */ +class DBClientMultiCommand : public MultiCommandDispatch { +public: + DBClientMultiCommand() : _timeoutMillis(0) {} - int numPending() const; + ~DBClientMultiCommand(); - Status recvAny( ConnectionString* endpoint, BSONSerializable* response ); + void addCommand(const ConnectionString& endpoint, + StringData dbName, + const BSONSerializable& request); - void setTimeoutMillis( int milliSecs ); + void sendAll(); - private: + int numPending() const; - // All info associated with an pre- or in-flight command - struct PendingCommand { + Status recvAny(ConnectionString* endpoint, BSONSerializable* response); - PendingCommand( const ConnectionString& endpoint, - StringData dbName, - const BSONObj& cmdObj ); + void setTimeoutMillis(int milliSecs); - // What to send - const ConnectionString endpoint; - const std::string dbName; - const BSONObj cmdObj; +private: + // All info associated with an pre- or in-flight command + struct PendingCommand { + PendingCommand(const ConnectionString& endpoint, StringData dbName, const BSONObj& cmdObj); - // Where to send it - DBClientBase* conn; + // What to send + const ConnectionString endpoint; + const std::string dbName; + const BSONObj cmdObj; - // If anything goes wrong - Status status; - }; + // Where to send it + DBClientBase* conn; - typedef std::deque<PendingCommand*> PendingQueue; - PendingQueue _pendingCommands; - int _timeoutMillis; + // If anything goes wrong + Status status; }; + typedef std::deque<PendingCommand*> PendingQueue; + PendingQueue _pendingCommands; + int _timeoutMillis; +}; } diff --git a/src/mongo/s/client/mock_multi_write_command.h b/src/mongo/s/client/mock_multi_write_command.h index cefdd00752c..5491c0b44f1 100644 --- a/src/mongo/s/client/mock_multi_write_command.h +++ b/src/mongo/s/client/mock_multi_write_command.h @@ -37,136 +37,127 @@ namespace mongo { - /** - * A ConnectionString endpoint registered with some kind of error, to simulate returning when - * the endpoint is used. - */ - struct MockWriteResult { - - MockWriteResult( const ConnectionString& endpoint, const WriteErrorDetail& error ) : - endpoint( endpoint ) { +/** + * A ConnectionString endpoint registered with some kind of error, to simulate returning when + * the endpoint is used. + */ +struct MockWriteResult { + MockWriteResult(const ConnectionString& endpoint, const WriteErrorDetail& error) + : endpoint(endpoint) { + WriteErrorDetail* errorCopy = new WriteErrorDetail; + error.cloneTo(errorCopy); + errorCopy->setIndex(0); + response.setOk(true); + response.setN(0); + response.addToErrDetails(errorCopy); + } + + MockWriteResult(const ConnectionString& endpoint, const WriteErrorDetail& error, int copies) + : endpoint(endpoint) { + response.setOk(true); + response.setN(0); + + for (int i = 0; i < copies; ++i) { WriteErrorDetail* errorCopy = new WriteErrorDetail; - error.cloneTo( errorCopy ); - errorCopy->setIndex( 0 ); - response.setOk(true); - response.setN(0); - response.addToErrDetails( errorCopy ); + error.cloneTo(errorCopy); + errorCopy->setIndex(i); + response.addToErrDetails(errorCopy); } + } - MockWriteResult( const ConnectionString& endpoint, - const WriteErrorDetail& error, - int copies ) : - endpoint( endpoint ) { - - response.setOk( true ); - response.setN( 0 ); - - for ( int i = 0; i < copies; ++i ) { - WriteErrorDetail* errorCopy = new WriteErrorDetail; - error.cloneTo( errorCopy ); - errorCopy->setIndex( i ); - response.addToErrDetails( errorCopy ); - } - } + MockWriteResult(const ConnectionString& endpoint, const BatchedCommandResponse& response) + : endpoint(endpoint) { + response.cloneTo(&this->response); + } - MockWriteResult( const ConnectionString& endpoint, const BatchedCommandResponse& response ) : - endpoint( endpoint ) { - response.cloneTo( &this->response ); - } + const ConnectionString endpoint; + BatchedCommandResponse response; +}; - const ConnectionString endpoint; - BatchedCommandResponse response; - }; +/** + * Implementation of the MultiCommandDispatch interface which allows registering a number of + * endpoints on which errors are returned. Note that *only* BatchedCommandResponses are + * supported here. + * + * The first matching MockEndpoint for a request in the MockEndpoint* vector is used for one + * request, then removed. This allows simulating retryable errors where a second request + * succeeds or has a different error reported. + * + * If an endpoint isn't registered with a MockEndpoint, just returns BatchedCommandResponses + * with ok : true. + */ +class MockMultiWriteCommand : public MultiCommandDispatch { +public: + void init(const std::vector<MockWriteResult*> mockEndpoints) { + ASSERT(!mockEndpoints.empty()); + _mockEndpoints.mutableVector().insert( + _mockEndpoints.mutableVector().end(), mockEndpoints.begin(), mockEndpoints.end()); + } + + void addCommand(const ConnectionString& endpoint, + StringData dbName, + const BSONSerializable& request) { + _pending.push_back(endpoint); + } + + void sendAll() { + // No-op + } + + int numPending() const { + return static_cast<int>(_pending.size()); + } /** - * Implementation of the MultiCommandDispatch interface which allows registering a number of - * endpoints on which errors are returned. Note that *only* BatchedCommandResponses are - * supported here. - * - * The first matching MockEndpoint for a request in the MockEndpoint* vector is used for one - * request, then removed. This allows simulating retryable errors where a second request - * succeeds or has a different error reported. - * - * If an endpoint isn't registered with a MockEndpoint, just returns BatchedCommandResponses - * with ok : true. + * Returns an error response if the next pending endpoint returned has a corresponding + * MockEndpoint. */ - class MockMultiWriteCommand : public MultiCommandDispatch { - public: - - void init( const std::vector<MockWriteResult*> mockEndpoints ) { - ASSERT( !mockEndpoints.empty() ); - _mockEndpoints.mutableVector().insert( _mockEndpoints.mutableVector().end(), - mockEndpoints.begin(), - mockEndpoints.end() ); - } - - void addCommand( const ConnectionString& endpoint, - StringData dbName, - const BSONSerializable& request ) { - _pending.push_back( endpoint ); + Status recvAny(ConnectionString* endpoint, BSONSerializable* response) { + BatchedCommandResponse* batchResponse = // + static_cast<BatchedCommandResponse*>(response); + + *endpoint = _pending.front(); + MockWriteResult* mockResponse = releaseByHost(_pending.front()); + _pending.pop_front(); + + if (NULL == mockResponse) { + batchResponse->setOk(true); + batchResponse->setN(0); // TODO: Make this accurate + } else { + mockResponse->response.cloneTo(batchResponse); + delete mockResponse; } - void sendAll() { - // No-op - } - - int numPending() const { - return static_cast<int>( _pending.size() ); - } - - /** - * Returns an error response if the next pending endpoint returned has a corresponding - * MockEndpoint. - */ - Status recvAny( ConnectionString* endpoint, BSONSerializable* response ) { - - BatchedCommandResponse* batchResponse = // - static_cast<BatchedCommandResponse*>( response ); - - *endpoint = _pending.front(); - MockWriteResult* mockResponse = releaseByHost( _pending.front() ); - _pending.pop_front(); - - if ( NULL == mockResponse ) { - batchResponse->setOk( true ); - batchResponse->setN( 0 ); // TODO: Make this accurate - } - else { - mockResponse->response.cloneTo( batchResponse ); - delete mockResponse; + ASSERT(batchResponse->isValid(NULL)); + return Status::OK(); + } + + const std::vector<MockWriteResult*>& getEndpoints() const { + return _mockEndpoints.vector(); + } + +private: + // Find a MockEndpoint* by host, and release it so we don't see it again + MockWriteResult* releaseByHost(const ConnectionString& endpoint) { + std::vector<MockWriteResult*>& endpoints = _mockEndpoints.mutableVector(); + + for (std::vector<MockWriteResult*>::iterator it = endpoints.begin(); it != endpoints.end(); + ++it) { + MockWriteResult* storedEndpoint = *it; + if (storedEndpoint->endpoint.toString().compare(endpoint.toString()) == 0) { + endpoints.erase(it); + return storedEndpoint; } - - ASSERT( batchResponse->isValid( NULL ) ); - return Status::OK(); - } - - const std::vector<MockWriteResult*>& getEndpoints() const { - return _mockEndpoints.vector(); } - private: - - // Find a MockEndpoint* by host, and release it so we don't see it again - MockWriteResult* releaseByHost( const ConnectionString& endpoint ) { - std::vector<MockWriteResult*>& endpoints = _mockEndpoints.mutableVector(); - - for ( std::vector<MockWriteResult*>::iterator it = endpoints.begin(); - it != endpoints.end(); ++it ) { - MockWriteResult* storedEndpoint = *it; - if ( storedEndpoint->endpoint.toString().compare( endpoint.toString() ) == 0 ) { - endpoints.erase( it ); - return storedEndpoint; - } - } - - return NULL; - } + return NULL; + } - // Manually-stored ranges - OwnedPointerVector<MockWriteResult> _mockEndpoints; + // Manually-stored ranges + OwnedPointerVector<MockWriteResult> _mockEndpoints; - std::deque<ConnectionString> _pending; - }; + std::deque<ConnectionString> _pending; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/client/multi_command_dispatch.h b/src/mongo/s/client/multi_command_dispatch.h index f9470091637..c16baec3f55 100644 --- a/src/mongo/s/client/multi_command_dispatch.h +++ b/src/mongo/s/client/multi_command_dispatch.h @@ -35,53 +35,50 @@ namespace mongo { +/** + * A MultiCommandDispatch is a send/recv operation for multiple commands at once. + * + * The commands are first registered with an endpoint and serializable request. Commands are + * sent out without waiting for responses, and then responses are read later one-at-a-time. + * + * If context must be tracked alongside these requests, it can be associated with the endpoint + * object. + */ +class MultiCommandDispatch { +public: + virtual ~MultiCommandDispatch() {} + /** - * A MultiCommandDispatch is a send/recv operation for multiple commands at once. + * Adds a command to this multi-command dispatch. Commands are registered with a + * ConnectionString endpoint and a serializable request. * - * The commands are first registered with an endpoint and serializable request. Commands are - * sent out without waiting for responses, and then responses are read later one-at-a-time. - * - * If context must be tracked alongside these requests, it can be associated with the endpoint - * object. + * Commands are not sent immediately, they are sent on sendAll. */ - class MultiCommandDispatch { - public: - - virtual ~MultiCommandDispatch() { - } - - /** - * Adds a command to this multi-command dispatch. Commands are registered with a - * ConnectionString endpoint and a serializable request. - * - * Commands are not sent immediately, they are sent on sendAll. - */ - virtual void addCommand( const ConnectionString& endpoint, - StringData dbName, - const BSONSerializable& request ) = 0; - - /** - * Sends all the commands in this dispatch to their endpoints, in undefined order and - * without waiting for responses. May block on full send queue (though this should be - * rare). - * - * Any error which occurs during sendAll will be reported on recvAny, *does not throw.* - */ - virtual void sendAll() = 0; + virtual void addCommand(const ConnectionString& endpoint, + StringData dbName, + const BSONSerializable& request) = 0; - /** - * Returns the number of sent requests that are still waiting to be recv'd. - */ - virtual int numPending() const = 0; + /** + * Sends all the commands in this dispatch to their endpoints, in undefined order and + * without waiting for responses. May block on full send queue (though this should be + * rare). + * + * Any error which occurs during sendAll will be reported on recvAny, *does not throw.* + */ + virtual void sendAll() = 0; - /** - * Blocks until a command response has come back. Any outstanding command response may be - * returned with associated endpoint. - * - * Returns !OK on send/recv/parse failure, otherwise command-level errors are returned in - * the response object itself. - */ - virtual Status recvAny( ConnectionString* endpoint, BSONSerializable* response ) = 0; - }; + /** + * Returns the number of sent requests that are still waiting to be recv'd. + */ + virtual int numPending() const = 0; + /** + * Blocks until a command response has come back. Any outstanding command response may be + * returned with associated endpoint. + * + * Returns !OK on send/recv/parse failure, otherwise command-level errors are returned in + * the response object itself. + */ + virtual Status recvAny(ConnectionString* endpoint, BSONSerializable* response) = 0; +}; } diff --git a/src/mongo/s/client/multi_host_query.cpp b/src/mongo/s/client/multi_host_query.cpp index 2fcb5906bf4..bfe9d36f55d 100644 --- a/src/mongo/s/client/multi_host_query.cpp +++ b/src/mongo/s/client/multi_host_query.cpp @@ -34,375 +34,339 @@ namespace mongo { - using std::shared_ptr; - using std::make_pair; - using std::string; - using std::vector; +using std::shared_ptr; +using std::make_pair; +using std::string; +using std::vector; - typedef stdx::unique_lock<stdx::mutex> boost_unique_lock; +typedef stdx::unique_lock<stdx::mutex> boost_unique_lock; - HostThreadPool::HostThreadPool(int poolSize, bool scopeAllWork) : - _scopeAllWork(scopeAllWork), _context(new PoolContext) { +HostThreadPool::HostThreadPool(int poolSize, bool scopeAllWork) + : _scopeAllWork(scopeAllWork), _context(new PoolContext) { + // All threads start as active to avoid races detecting idleness on thread startup - + // the pool isn't idle until all threads have started waiting. + _context->numActiveWorkers = poolSize; - // All threads start as active to avoid races detecting idleness on thread startup - - // the pool isn't idle until all threads have started waiting. - _context->numActiveWorkers = poolSize; - - for (int i = 0; i < poolSize; ++i) { - - // - // Each thread keeps a shared context allowing them to synchronize even if this - // dispatching pool has already been disposed. - // - - _threads.push_back(new stdx::thread(stdx::bind(&HostThreadPool::doWork, _context))); - } - } + for (int i = 0; i < poolSize; ++i) { + // + // Each thread keeps a shared context allowing them to synchronize even if this + // dispatching pool has already been disposed. + // - void HostThreadPool::schedule(Callback callback) { - boost_unique_lock lk(_context->mutex); - _context->scheduled.push_back(callback); - _context->workScheduledCV.notify_one(); + _threads.push_back(new stdx::thread(stdx::bind(&HostThreadPool::doWork, _context))); } +} - void HostThreadPool::doWork(std::shared_ptr<PoolContext> context) { - - while (true) { - - Callback callback; - - { - boost_unique_lock lk(context->mutex); - - --context->numActiveWorkers; - if (context->numActiveWorkers == 0) - context->isIdleCV.notify_all(); - - // Wait for work or until we're finished - while (context->isPoolActive && context->scheduled.empty()) { - context->workScheduledCV.wait(lk); - } +void HostThreadPool::schedule(Callback callback) { + boost_unique_lock lk(_context->mutex); + _context->scheduled.push_back(callback); + _context->workScheduledCV.notify_one(); +} - // - // Either the pool is no longer active, or the queue has some work we should do - // +void HostThreadPool::doWork(std::shared_ptr<PoolContext> context) { + while (true) { + Callback callback; - if (!context->isPoolActive) - return; + { + boost_unique_lock lk(context->mutex); - invariant( !context->scheduled.empty() ); - callback = context->scheduled.front(); - context->scheduled.pop_front(); + --context->numActiveWorkers; + if (context->numActiveWorkers == 0) + context->isIdleCV.notify_all(); - ++context->numActiveWorkers; + // Wait for work or until we're finished + while (context->isPoolActive && context->scheduled.empty()) { + context->workScheduledCV.wait(lk); } - callback(); - } - } - - void HostThreadPool::waitUntilIdle() { - boost_unique_lock lk(_context->mutex); - while (_context->numActiveWorkers > 0) { - _context->isIdleCV.wait(lk); - } - } + // + // Either the pool is no longer active, or the queue has some work we should do + // - HostThreadPool::~HostThreadPool() { + if (!context->isPoolActive) + return; - // Boost can throw on notify(), join(), detach() + invariant(!context->scheduled.empty()); + callback = context->scheduled.front(); + context->scheduled.pop_front(); - { - boost_unique_lock lk(_context->mutex); - _context->isPoolActive = false; - _context->scheduled.clear(); + ++context->numActiveWorkers; } - DESTRUCTOR_GUARD( _context->workScheduledCV.notify_all(); ) - - for (vector<stdx::thread*>::iterator it = _threads.begin(); it != _threads.end(); ++it) { - - if (_scopeAllWork) { - DESTRUCTOR_GUARD( ( *it )->join(); ) - } - else { - DESTRUCTOR_GUARD( ( *it )->detach(); ) - } + callback(); + } +} - delete *it; - } +void HostThreadPool::waitUntilIdle() { + boost_unique_lock lk(_context->mutex); + while (_context->numActiveWorkers > 0) { + _context->isIdleCV.wait(lk); } +} - HostThreadPools::HostThreadPools(int poolSize, bool scopeAllWork) : - _poolSize(poolSize), _scopeAllWork(scopeAllWork) { +HostThreadPool::~HostThreadPool() { + // Boost can throw on notify(), join(), detach() + + { + boost_unique_lock lk(_context->mutex); + _context->isPoolActive = false; + _context->scheduled.clear(); } - void HostThreadPools::schedule(const ConnectionString& host, - HostThreadPool::Callback callback) { - boost_unique_lock lk(_mutex); + DESTRUCTOR_GUARD(_context->workScheduledCV.notify_all();) - HostPoolMap::iterator seenIt = _pools.find(host); - if (seenIt == _pools.end()) { - seenIt = _pools.insert(make_pair(host, new HostThreadPool(_poolSize, _scopeAllWork))) - .first; + for (vector<stdx::thread*>::iterator it = _threads.begin(); it != _threads.end(); ++it) { + if (_scopeAllWork) { + DESTRUCTOR_GUARD((*it)->join();) + } else { + DESTRUCTOR_GUARD((*it)->detach();) } - seenIt->second->schedule(callback); + delete *it; } +} - void HostThreadPools::waitUntilIdle(const ConnectionString& host) { - - // Note that this prevents the creation of any new pools - it is only intended to be used - // for testing. - - boost_unique_lock lk(_mutex); +HostThreadPools::HostThreadPools(int poolSize, bool scopeAllWork) + : _poolSize(poolSize), _scopeAllWork(scopeAllWork) {} - HostPoolMap::iterator seenIt = _pools.find(host); - if (seenIt == _pools.end()) - return; +void HostThreadPools::schedule(const ConnectionString& host, HostThreadPool::Callback callback) { + boost_unique_lock lk(_mutex); - seenIt->second->waitUntilIdle(); + HostPoolMap::iterator seenIt = _pools.find(host); + if (seenIt == _pools.end()) { + seenIt = _pools.insert(make_pair(host, new HostThreadPool(_poolSize, _scopeAllWork))).first; } - HostThreadPools::~HostThreadPools() { + seenIt->second->schedule(callback); +} - boost_unique_lock lk(_mutex); - for (HostPoolMap::iterator it = _pools.begin(); it != _pools.end(); ++it) { - delete it->second; - } - } +void HostThreadPools::waitUntilIdle(const ConnectionString& host) { + // Note that this prevents the creation of any new pools - it is only intended to be used + // for testing. - MultiHostQueryOp::MultiHostQueryOp(SystemEnv* systemEnv, HostThreadPools* hostThreads) : - _systemEnv(systemEnv), _hostThreads(hostThreads) { - } + boost_unique_lock lk(_mutex); - StatusWith<DBClientCursor*> MultiHostQueryOp::queryAny(const vector<ConnectionString>& hosts, - const QuerySpec& query, - int timeoutMillis) { - - Date_t nowMillis = _systemEnv->currentTimeMillis(); - Date_t timeoutAtMillis = nowMillis + Milliseconds(timeoutMillis); + HostPoolMap::iterator seenIt = _pools.find(host); + if (seenIt == _pools.end()) + return; - // Send out all queries - scheduleQuery(hosts, query, timeoutAtMillis); + seenIt->second->waitUntilIdle(); +} - // Wait for them to come back - return waitForNextResult(timeoutAtMillis); +HostThreadPools::~HostThreadPools() { + boost_unique_lock lk(_mutex); + for (HostPoolMap::iterator it = _pools.begin(); it != _pools.end(); ++it) { + delete it->second; } +} - void MultiHostQueryOp::scheduleQuery(const vector<ConnectionString>& hosts, - const QuerySpec& query, - Date_t timeoutAtMillis) { - - invariant( _pending.empty() ); +MultiHostQueryOp::MultiHostQueryOp(SystemEnv* systemEnv, HostThreadPools* hostThreads) + : _systemEnv(systemEnv), _hostThreads(hostThreads) {} - for (vector<ConnectionString>::const_iterator it = hosts.begin(); it != hosts.end(); ++it) { +StatusWith<DBClientCursor*> MultiHostQueryOp::queryAny(const vector<ConnectionString>& hosts, + const QuerySpec& query, + int timeoutMillis) { + Date_t nowMillis = _systemEnv->currentTimeMillis(); + Date_t timeoutAtMillis = nowMillis + Milliseconds(timeoutMillis); - const ConnectionString& host = *it; + // Send out all queries + scheduleQuery(hosts, query, timeoutAtMillis); - shared_ptr<PendingQueryContext> pendingOp(new PendingQueryContext(host, - query, - timeoutAtMillis, - this)); + // Wait for them to come back + return waitForNextResult(timeoutAtMillis); +} - _pending.insert(make_pair(host, pendingOp)); +void MultiHostQueryOp::scheduleQuery(const vector<ConnectionString>& hosts, + const QuerySpec& query, + Date_t timeoutAtMillis) { + invariant(_pending.empty()); - HostThreadPool::Callback callback = - stdx::bind(&MultiHostQueryOp::PendingQueryContext::doBlockingQuery, pendingOp); + for (vector<ConnectionString>::const_iterator it = hosts.begin(); it != hosts.end(); ++it) { + const ConnectionString& host = *it; - _hostThreads->schedule(host, callback); - } - } + shared_ptr<PendingQueryContext> pendingOp( + new PendingQueryContext(host, query, timeoutAtMillis, this)); - StatusWith<DBClientCursor*> MultiHostQueryOp::waitForNextResult(Date_t timeoutAtMillis) { + _pending.insert(make_pair(host, pendingOp)); - StatusWith<DBClientCursor*> nextResult( NULL); + HostThreadPool::Callback callback = + stdx::bind(&MultiHostQueryOp::PendingQueryContext::doBlockingQuery, pendingOp); - boost_unique_lock lk(_resultsMutex); - while (!releaseResult_inlock(&nextResult)) { + _hostThreads->schedule(host, callback); + } +} - Date_t nowMillis = _systemEnv->currentTimeMillis(); +StatusWith<DBClientCursor*> MultiHostQueryOp::waitForNextResult(Date_t timeoutAtMillis) { + StatusWith<DBClientCursor*> nextResult(NULL); - if (nowMillis >= timeoutAtMillis) { - nextResult = StatusWith<DBClientCursor*>(combineErrorResults_inlock()); - break; - } + boost_unique_lock lk(_resultsMutex); + while (!releaseResult_inlock(&nextResult)) { + Date_t nowMillis = _systemEnv->currentTimeMillis(); - _nextResultCV.wait_for(lk, timeoutAtMillis - nowMillis); + if (nowMillis >= timeoutAtMillis) { + nextResult = StatusWith<DBClientCursor*>(combineErrorResults_inlock()); + break; } - dassert( !nextResult.isOK() || nextResult.getValue() ); - return nextResult; + _nextResultCV.wait_for(lk, timeoutAtMillis - nowMillis); } - void MultiHostQueryOp::noteResult(const ConnectionString& host, - StatusWith<DBClientCursor*> result) { - - boost_unique_lock lk(_resultsMutex); - dassert( _results.find( host ) == _results.end() ); - _results.insert(make_pair(host, result)); + dassert(!nextResult.isOK() || nextResult.getValue()); + return nextResult; +} - _nextResultCV.notify_one(); - } +void MultiHostQueryOp::noteResult(const ConnectionString& host, + StatusWith<DBClientCursor*> result) { + boost_unique_lock lk(_resultsMutex); + dassert(_results.find(host) == _results.end()); + _results.insert(make_pair(host, result)); - /** - * The results in the result map have four states: - * Nonexistent - query result still pending - * Status::OK w/ pointer - successful query result, not yet released to user - * Status::OK w/ NULL pointer - successful query result, user consumed the result - * Status::Not OK - error during query - * - * This function returns true and the next result to release to the user (or an error - * if there can be no successful results to release) or false to indicate the user - * should keep waiting. - */ - bool MultiHostQueryOp::releaseResult_inlock(StatusWith<DBClientCursor*>* nextResult) { - - int numErrors = 0; - int numReleased = 0; - for (ResultMap::iterator it = _results.begin(); it != _results.end(); ++it) { - - StatusWith<DBClientCursor*>& result = it->second; - - if (result.isOK() && result.getValue() != NULL) { - *nextResult = result; - it->second = StatusWith<DBClientCursor*>( NULL); - return true; - } - else if (result.isOK()) { - ++numReleased; - } - else { - ++numErrors; - } - } + _nextResultCV.notify_one(); +} - if (numErrors + numReleased == static_cast<int>(_pending.size())) { - *nextResult = StatusWith<DBClientCursor*>(combineErrorResults_inlock()); +/** + * The results in the result map have four states: + * Nonexistent - query result still pending + * Status::OK w/ pointer - successful query result, not yet released to user + * Status::OK w/ NULL pointer - successful query result, user consumed the result + * Status::Not OK - error during query + * + * This function returns true and the next result to release to the user (or an error + * if there can be no successful results to release) or false to indicate the user + * should keep waiting. + */ +bool MultiHostQueryOp::releaseResult_inlock(StatusWith<DBClientCursor*>* nextResult) { + int numErrors = 0; + int numReleased = 0; + for (ResultMap::iterator it = _results.begin(); it != _results.end(); ++it) { + StatusWith<DBClientCursor*>& result = it->second; + + if (result.isOK() && result.getValue() != NULL) { + *nextResult = result; + it->second = StatusWith<DBClientCursor*>(NULL); return true; + } else if (result.isOK()) { + ++numReleased; + } else { + ++numErrors; } - - return false; } - /** - * Goes through the set of results and combines all non-OK results into a single Status. - * If a single error is found, just returns that error. - * If no non-OK results are found, assumes the cause is a timeout. - */ - Status MultiHostQueryOp::combineErrorResults_inlock() { - - ErrorCodes::Error code = ErrorCodes::OK; - StringBuilder errMsg; - // Whether we should include human-readable codes in the msg - we don't need them if we're - // not aggregating multiple statuses together - bool includeHRCodes = false; - - for (ResultMap::const_iterator it = _results.begin(); it != _results.end(); ++it) { - - const StatusWith<DBClientCursor*>& result = it->second; - - if (!result.isOK()) { - - if (code == ErrorCodes::OK) { - code = result.getStatus().code(); - } - else { + if (numErrors + numReleased == static_cast<int>(_pending.size())) { + *nextResult = StatusWith<DBClientCursor*>(combineErrorResults_inlock()); + return true; + } - if (!includeHRCodes) { - includeHRCodes = true; - // Fixup the single error message to include a code - errMsg.reset(); - errMsg.append(Status(code, errMsg.str()).toString()); - } + return false; +} - code = ErrorCodes::MultipleErrorsOccurred; - errMsg.append(" :: and :: "); +/** + * Goes through the set of results and combines all non-OK results into a single Status. + * If a single error is found, just returns that error. + * If no non-OK results are found, assumes the cause is a timeout. + */ +Status MultiHostQueryOp::combineErrorResults_inlock() { + ErrorCodes::Error code = ErrorCodes::OK; + StringBuilder errMsg; + // Whether we should include human-readable codes in the msg - we don't need them if we're + // not aggregating multiple statuses together + bool includeHRCodes = false; + + for (ResultMap::const_iterator it = _results.begin(); it != _results.end(); ++it) { + const StatusWith<DBClientCursor*>& result = it->second; + + if (!result.isOK()) { + if (code == ErrorCodes::OK) { + code = result.getStatus().code(); + } else { + if (!includeHRCodes) { + includeHRCodes = true; + // Fixup the single error message to include a code + errMsg.reset(); + errMsg.append(Status(code, errMsg.str()).toString()); } - errMsg.append( - includeHRCodes ? result.getStatus().toString() : result.getStatus().reason()); - errMsg.append(string(", host ") + it->first.toString()); + code = ErrorCodes::MultipleErrorsOccurred; + errMsg.append(" :: and :: "); } - } - if (code == ErrorCodes::OK) { - return Status(ErrorCodes::NetworkTimeout, "no results were returned in time"); + errMsg.append(includeHRCodes ? result.getStatus().toString() + : result.getStatus().reason()); + errMsg.append(string(", host ") + it->first.toString()); } - - return Status(code, errMsg.str()); } - MultiHostQueryOp::PendingQueryContext::PendingQueryContext(const ConnectionString& host, - const QuerySpec& query, - const Date_t timeoutAtMillis, - MultiHostQueryOp* parentOp) : - host(host), query(query), timeoutAtMillis(timeoutAtMillis), parentOp(parentOp) { + if (code == ErrorCodes::OK) { + return Status(ErrorCodes::NetworkTimeout, "no results were returned in time"); } - void MultiHostQueryOp::PendingQueryContext::doBlockingQuery() { - - // This *NEEDS* to be around for as long as we're doing queries - i.e. as long as the - // HostThreadPools is. - MultiHostQueryOp::SystemEnv* systemEnv; + return Status(code, errMsg.str()); +} - // Extract means of doing query from the parent op - { - boost_unique_lock lk(parentMutex); +MultiHostQueryOp::PendingQueryContext::PendingQueryContext(const ConnectionString& host, + const QuerySpec& query, + const Date_t timeoutAtMillis, + MultiHostQueryOp* parentOp) + : host(host), query(query), timeoutAtMillis(timeoutAtMillis), parentOp(parentOp) {} - if (!parentOp) - return; +void MultiHostQueryOp::PendingQueryContext::doBlockingQuery() { + // This *NEEDS* to be around for as long as we're doing queries - i.e. as long as the + // HostThreadPools is. + MultiHostQueryOp::SystemEnv* systemEnv; - systemEnv = parentOp->_systemEnv; - } + // Extract means of doing query from the parent op + { + boost_unique_lock lk(parentMutex); - // Make sure we're not timed out - Date_t nowMillis = systemEnv->currentTimeMillis(); - if (nowMillis >= timeoutAtMillis) + if (!parentOp) return; - // Do query - StatusWith<DBClientCursor*> result = systemEnv->doBlockingQuery(host, query); - - // Push results back to parent op if it's still around - { - boost_unique_lock lk(parentMutex); - - if (parentOp) - parentOp->noteResult(host, result); - else if(result.isOK()) - delete result.getValue(); - } + systemEnv = parentOp->_systemEnv; } - MultiHostQueryOp::~MultiHostQueryOp() { + // Make sure we're not timed out + Date_t nowMillis = systemEnv->currentTimeMillis(); + if (nowMillis >= timeoutAtMillis) + return; - // - // Orphan all outstanding query contexts that haven't reported back - these will be gc'd - // once all scheduled query callbacks are finished. - // + // Do query + StatusWith<DBClientCursor*> result = systemEnv->doBlockingQuery(host, query); - for (PendingMap::iterator it = _pending.begin(); it != _pending.end(); ++it) { + // Push results back to parent op if it's still around + { + boost_unique_lock lk(parentMutex); - shared_ptr<PendingQueryContext>& pendingContext = it->second; + if (parentOp) + parentOp->noteResult(host, result); + else if (result.isOK()) + delete result.getValue(); + } +} - boost_unique_lock lk(pendingContext->parentMutex); - pendingContext->parentOp = NULL; - } +MultiHostQueryOp::~MultiHostQueryOp() { + // + // Orphan all outstanding query contexts that haven't reported back - these will be gc'd + // once all scheduled query callbacks are finished. + // - // - // Nobody else should be modifying _results now - callbacks don't have access to this op, - // and other clients should know the op is going out of scope - // + for (PendingMap::iterator it = _pending.begin(); it != _pending.end(); ++it) { + shared_ptr<PendingQueryContext>& pendingContext = it->second; - for (ResultMap::iterator it = _results.begin(); it != _results.end(); ++it) { + boost_unique_lock lk(pendingContext->parentMutex); + pendingContext->parentOp = NULL; + } - StatusWith<DBClientCursor*>& result = it->second; + // + // Nobody else should be modifying _results now - callbacks don't have access to this op, + // and other clients should know the op is going out of scope + // - if (result.isOK()) { - delete result.getValue(); - } + for (ResultMap::iterator it = _results.begin(); it != _results.end(); ++it) { + StatusWith<DBClientCursor*>& result = it->second; + + if (result.isOK()) { + delete result.getValue(); } } - +} } diff --git a/src/mongo/s/client/multi_host_query.h b/src/mongo/s/client/multi_host_query.h index 7b69ecc25f5..61bdf27bf4f 100644 --- a/src/mongo/s/client/multi_host_query.h +++ b/src/mongo/s/client/multi_host_query.h @@ -39,290 +39,280 @@ namespace mongo { - // - // NOTE TO DEVS - // This is probably not what we want long-term - think very carefully before letting any of the - // functionality below escape this file. - // +// +// NOTE TO DEVS +// This is probably not what we want long-term - think very carefully before letting any of the +// functionality below escape this file. +// + +class HostThreadPools; +class HostThreadPool; + +/** + * A MultiHostQueryOp manages a query operation across multiple hosts. Supports returning + * immediately when any host has results or when all hosts have (connectivity) errors. + * + * The QueryOp itself dispatches work to the thread pool, and does not wait for all work to be + * complete before destruction. This class is not intended to be used by multiple clients at + * once without external synchronization (for now). + * + * Cannot be reused once all query results and errors have been returned. + */ +class MultiHostQueryOp { + MONGO_DISALLOW_COPYING(MultiHostQueryOp); - class HostThreadPools; - class HostThreadPool; +public: + /** + * Network and time services interface + */ + class SystemEnv; /** - * A MultiHostQueryOp manages a query operation across multiple hosts. Supports returning - * immediately when any host has results or when all hosts have (connectivity) errors. + * Constructs a MultiHostQueryOp. Allows running a query across multiple hosts with a + * blocking interface. The lifetime of this class can be shorter than the lifetime of the + * queries sent via queryAny, freeing up the caller to do further work when any host is fast + * to respond. * - * The QueryOp itself dispatches work to the thread pool, and does not wait for all work to be - * complete before destruction. This class is not intended to be used by multiple clients at - * once without external synchronization (for now). + * The systemEnv and hostThreads must remain in scope while the query op remains in scope. * - * Cannot be reused once all query results and errors have been returned. + * NOTE: SystemEnv* MUST remain valid for as long as hostThreads remains valid, since this + * operation may schedule background queries but fall out of scope while one of those + * background queries is still in-progress. */ - class MultiHostQueryOp { - MONGO_DISALLOW_COPYING(MultiHostQueryOp); - public: - - /** - * Network and time services interface - */ - class SystemEnv; - - /** - * Constructs a MultiHostQueryOp. Allows running a query across multiple hosts with a - * blocking interface. The lifetime of this class can be shorter than the lifetime of the - * queries sent via queryAny, freeing up the caller to do further work when any host is fast - * to respond. - * - * The systemEnv and hostThreads must remain in scope while the query op remains in scope. - * - * NOTE: SystemEnv* MUST remain valid for as long as hostThreads remains valid, since this - * operation may schedule background queries but fall out of scope while one of those - * background queries is still in-progress. - */ - MultiHostQueryOp(SystemEnv* systemEnv, HostThreadPools* hostThreads); - - ~MultiHostQueryOp(); - - /** - * Blocks for a query to be run on any of the hosts, and returns the fastest result as soon - * as it becomes available. This function may only be executed once. - * - * If one or more hosts have an error sending/recv'ing the query, the error or composite - * error is returned if no other hosts are responsive after the timeout period. Note that - * this does not apply to errors successfully returned from remote hosts - this is a - * successful query with an error. - * - * Caller owns the returned result if OK. - */ - StatusWith<DBClientCursor*> queryAny(const std::vector<ConnectionString>& hosts, - const QuerySpec& query, - int timeoutMillis); - - // - // Below is exposed for testing *only* - // - - /** - * Schedules the query work on each of the hosts using the thread pool, with a timeout - * indicating how long the work is useful for. Can be called only once. - */ - void scheduleQuery(const std::vector<ConnectionString>& hosts, - const QuerySpec& query, - Date_t timeoutAtMillis); - - /** - * Blocks and waits for the next successful query result or any errors once the timeout is - * reached. - * Can be called multiple times until results from all hosts are returned or !OK. - */ - StatusWith<DBClientCursor*> waitForNextResult(Date_t timeoutAtMillis); - - private: - - /** - * Data required to execute a query operation by a callback on an arbitrary thread. - * Information from the dispatching parent op may not be available if the parent is no - * longer in scope. - */ - struct PendingQueryContext { - - PendingQueryContext(const ConnectionString& host, - const QuerySpec& query, - const Date_t timeoutAtMillis, - MultiHostQueryOp* parentOp); - - void doBlockingQuery(); - - const ConnectionString host; - const QuerySpec query; - const Date_t timeoutAtMillis; - - // Must be held to access the parent pointer below - stdx::mutex parentMutex; - // Set and unset by the parent operation on scheduling and destruction - MultiHostQueryOp* parentOp; - }; - - /** - * Called by a scheduled query (generally on a different thread from the waiting client) - * when a result is ready from a particular host. - */ - void noteResult(const ConnectionString& host, StatusWith<DBClientCursor*> result); - - /** - * Helper to check if any result is ready and extract that result - * Synchronized by _resultsMutex - */ - bool releaseResult_inlock(StatusWith<DBClientCursor*>* nextResult); - - /** - * Helper to return an error status from zero or more results - * Synchronized by _resultsMutex - */ - Status combineErrorResults_inlock(); - - // Not owned here - SystemEnv* _systemEnv; - - // Not owned here - HostThreadPools* _hostThreads; - - // Outstanding requests - typedef std::map<ConnectionString, std::shared_ptr<PendingQueryContext> > PendingMap; - PendingMap _pending; + MultiHostQueryOp(SystemEnv* systemEnv, HostThreadPools* hostThreads); - // Synchronizes below - stdx::mutex _resultsMutex; + ~MultiHostQueryOp(); - // Current results recv'd - typedef std::map<ConnectionString, StatusWith<DBClientCursor*> > ResultMap; - ResultMap _results; + /** + * Blocks for a query to be run on any of the hosts, and returns the fastest result as soon + * as it becomes available. This function may only be executed once. + * + * If one or more hosts have an error sending/recv'ing the query, the error or composite + * error is returned if no other hosts are responsive after the timeout period. Note that + * this does not apply to errors successfully returned from remote hosts - this is a + * successful query with an error. + * + * Caller owns the returned result if OK. + */ + StatusWith<DBClientCursor*> queryAny(const std::vector<ConnectionString>& hosts, + const QuerySpec& query, + int timeoutMillis); - stdx::condition_variable _nextResultCV; - }; + // + // Below is exposed for testing *only* + // /** - * Provides network and time services to allow unit testing of MultiHostQueryOp. + * Schedules the query work on each of the hosts using the thread pool, with a timeout + * indicating how long the work is useful for. Can be called only once. */ - class MultiHostQueryOp::SystemEnv { - public: - - virtual ~SystemEnv() { - } - - /** - * Returns the current time in milliseconds - */ - virtual Date_t currentTimeMillis() = 0; - - /** - * Executes a query against a given host. No timeout hint is given, but the query should - * not block forever. - * Note that no guarantees are given as to the state of the connection used after this - * returns, so the cursor must be self-contained. - * - * Caller owns any resulting cursor. - */ - virtual StatusWith<DBClientCursor*> doBlockingQuery(const ConnectionString& host, - const QuerySpec& query) = 0; - }; + void scheduleQuery(const std::vector<ConnectionString>& hosts, + const QuerySpec& query, + Date_t timeoutAtMillis); /** - * Object which encapsulates a thread pool per host, and allows scheduling operations against - * each of these hosts. - * - * Optionally supports not waiting for blocked threads before destruction. - * - * Thin wrapper for multiple hosts around HostThreadPool. + * Blocks and waits for the next successful query result or any errors once the timeout is + * reached. + * Can be called multiple times until results from all hosts are returned or !OK. + */ + StatusWith<DBClientCursor*> waitForNextResult(Date_t timeoutAtMillis); + +private: + /** + * Data required to execute a query operation by a callback on an arbitrary thread. + * Information from the dispatching parent op may not be available if the parent is no + * longer in scope. */ - class HostThreadPools { - MONGO_DISALLOW_COPYING(HostThreadPools); - public: - - typedef stdx::function<void(void)> Callback; - - /** - * Construct a HostThreadPools object, which lazily constructs thread pools per-host of the - * specified size. - * - * @param scopeAllWork true if the pool should wait for all work to be finished before - * going out of scope - */ - HostThreadPools(int poolSize, bool scopeAllWork); - ~HostThreadPools(); - - /** - * Schedules some work in the form of a callback for the pool of a particular host. - */ - void schedule(const ConnectionString& host, Callback callback); - - /** - * Blocks until pool is idle for a particular host. - * For testing. - */ - void waitUntilIdle(const ConnectionString& host); - - private: - - const int _poolSize; - const bool _scopeAllWork; - - stdx::mutex _mutex; - typedef std::map<ConnectionString, HostThreadPool*> HostPoolMap; - HostPoolMap _pools; + struct PendingQueryContext { + PendingQueryContext(const ConnectionString& host, + const QuerySpec& query, + const Date_t timeoutAtMillis, + MultiHostQueryOp* parentOp); + + void doBlockingQuery(); + + const ConnectionString host; + const QuerySpec query; + const Date_t timeoutAtMillis; + + // Must be held to access the parent pointer below + stdx::mutex parentMutex; + // Set and unset by the parent operation on scheduling and destruction + MultiHostQueryOp* parentOp; }; /** - * EXPOSED FOR TESTING ONLY. + * Called by a scheduled query (generally on a different thread from the waiting client) + * when a result is ready from a particular host. + */ + void noteResult(const ConnectionString& host, StatusWith<DBClientCursor*> result); + + /** + * Helper to check if any result is ready and extract that result + * Synchronized by _resultsMutex + */ + bool releaseResult_inlock(StatusWith<DBClientCursor*>* nextResult); + + /** + * Helper to return an error status from zero or more results + * Synchronized by _resultsMutex + */ + Status combineErrorResults_inlock(); + + // Not owned here + SystemEnv* _systemEnv; + + // Not owned here + HostThreadPools* _hostThreads; + + // Outstanding requests + typedef std::map<ConnectionString, std::shared_ptr<PendingQueryContext>> PendingMap; + PendingMap _pending; + + // Synchronizes below + stdx::mutex _resultsMutex; + + // Current results recv'd + typedef std::map<ConnectionString, StatusWith<DBClientCursor*>> ResultMap; + ResultMap _results; + + stdx::condition_variable _nextResultCV; +}; + +/** + * Provides network and time services to allow unit testing of MultiHostQueryOp. + */ +class MultiHostQueryOp::SystemEnv { +public: + virtual ~SystemEnv() {} + + /** + * Returns the current time in milliseconds + */ + virtual Date_t currentTimeMillis() = 0; + + /** + * Executes a query against a given host. No timeout hint is given, but the query should + * not block forever. + * Note that no guarantees are given as to the state of the connection used after this + * returns, so the cursor must be self-contained. * - * Thread pool allowing work to be scheduled against various hosts. - * Generic interface, but should not be used outside of this class. + * Caller owns any resulting cursor. */ - class HostThreadPool { - public: + virtual StatusWith<DBClientCursor*> doBlockingQuery(const ConnectionString& host, + const QuerySpec& query) = 0; +}; - typedef stdx::function<void(void)> Callback; +/** + * Object which encapsulates a thread pool per host, and allows scheduling operations against + * each of these hosts. + * + * Optionally supports not waiting for blocked threads before destruction. + * + * Thin wrapper for multiple hosts around HostThreadPool. + */ +class HostThreadPools { + MONGO_DISALLOW_COPYING(HostThreadPools); - /** - * Constructs a thread pool of a given size. - * - * Parameter scopeAllWork indicates whether the pool should wait for all work to be finished - * before going out of scope. - */ - HostThreadPool(int poolSize, bool scopeAllWork); +public: + typedef stdx::function<void(void)> Callback; - ~HostThreadPool(); + /** + * Construct a HostThreadPools object, which lazily constructs thread pools per-host of the + * specified size. + * + * @param scopeAllWork true if the pool should wait for all work to be finished before + * going out of scope + */ + HostThreadPools(int poolSize, bool scopeAllWork); + ~HostThreadPools(); - /** - * Schedules some work in the form of a callback to be done ASAP. - */ - void schedule(Callback callback); + /** + * Schedules some work in the form of a callback for the pool of a particular host. + */ + void schedule(const ConnectionString& host, Callback callback); + + /** + * Blocks until pool is idle for a particular host. + * For testing. + */ + void waitUntilIdle(const ConnectionString& host); - /** - * Blocks until all threads are idle. - */ - void waitUntilIdle(); +private: + const int _poolSize; + const bool _scopeAllWork; - private: + stdx::mutex _mutex; + typedef std::map<ConnectionString, HostThreadPool*> HostPoolMap; + HostPoolMap _pools; +}; - /** - * Synchronized work and activity information shared between the pool and the individual - * worker threads. - * This information must be shared, since if !scopeAllWork the parent pool is allowed to - * fall out of scope before the child thread completes. - */ - struct PoolContext { +/** + * EXPOSED FOR TESTING ONLY. + * + * Thread pool allowing work to be scheduled against various hosts. + * Generic interface, but should not be used outside of this class. + */ +class HostThreadPool { +public: + typedef stdx::function<void(void)> Callback; - PoolContext() : - numActiveWorkers(0), isPoolActive(true) { - } + /** + * Constructs a thread pool of a given size. + * + * Parameter scopeAllWork indicates whether the pool should wait for all work to be finished + * before going out of scope. + */ + HostThreadPool(int poolSize, bool scopeAllWork); - // Synchronizes below - stdx::mutex mutex; + ~HostThreadPool(); - // The scheduled work - std::deque<Callback> scheduled; - stdx::condition_variable workScheduledCV; + /** + * Schedules some work in the form of a callback to be done ASAP. + */ + void schedule(Callback callback); - // How many workers are currently active - int numActiveWorkers; - stdx::condition_variable isIdleCV; + /** + * Blocks until all threads are idle. + */ + void waitUntilIdle(); - // Whether the pool has been disposed of - bool isPoolActive; - }; +private: + /** + * Synchronized work and activity information shared between the pool and the individual + * worker threads. + * This information must be shared, since if !scopeAllWork the parent pool is allowed to + * fall out of scope before the child thread completes. + */ + struct PoolContext { + PoolContext() : numActiveWorkers(0), isPoolActive(true) {} - /** - * Worker loop run by each thread. - */ - static void doWork(std::shared_ptr<PoolContext> context); + // Synchronizes below + stdx::mutex mutex; - const bool _scopeAllWork; + // The scheduled work + std::deque<Callback> scheduled; + stdx::condition_variable workScheduledCV; - // For now, only modified in the constructor and destructor, but non-const - std::vector<stdx::thread*> _threads; + // How many workers are currently active + int numActiveWorkers; + stdx::condition_variable isIdleCV; - // Shared work and worker activity information - std::shared_ptr<PoolContext> _context; + // Whether the pool has been disposed of + bool isPoolActive; }; + + /** + * Worker loop run by each thread. + */ + static void doWork(std::shared_ptr<PoolContext> context); + + const bool _scopeAllWork; + + // For now, only modified in the constructor and destructor, but non-const + std::vector<stdx::thread*> _threads; + + // Shared work and worker activity information + std::shared_ptr<PoolContext> _context; +}; } diff --git a/src/mongo/s/client/multi_host_query_test.cpp b/src/mongo/s/client/multi_host_query_test.cpp index d63c4d311c8..c2d3f3688df 100644 --- a/src/mongo/s/client/multi_host_query_test.cpp +++ b/src/mongo/s/client/multi_host_query_test.cpp @@ -36,740 +36,698 @@ namespace { - using namespace mongo; - using std::unique_ptr; - using std::shared_ptr; - using std::make_pair; - using std::map; - using std::string; - using std::vector; - - class CallbackCheck { - public: - - enum LinkMode { - None, Notify_Other, Wait_For_Other - }; - - CallbackCheck() : - _status(ErrorCodes::OperationIncomplete, ""), _linkMode(None) { - } - - void blockUntil(CallbackCheck* other) { - - _otherNotification.reset(new Notification); - _linkMode = Wait_For_Other; - - other->_otherNotification = _otherNotification; - other->_linkMode = Notify_Other; - } - - HostThreadPool::Callback getCallback() { - return stdx::bind(&CallbackCheck::noteCallback, this); - } - - HostThreadPool::Callback getHostCallback(const ConnectionString& host) { - return stdx::bind(&CallbackCheck::noteHostCallback, this, host); - } - - void noteHostCallback(const ConnectionString& host) { - _host = host; - noteCallback(); - } - - void noteCallback() { +using namespace mongo; +using std::unique_ptr; +using std::shared_ptr; +using std::make_pair; +using std::map; +using std::string; +using std::vector; + +class CallbackCheck { +public: + enum LinkMode { None, Notify_Other, Wait_For_Other }; + + CallbackCheck() : _status(ErrorCodes::OperationIncomplete, ""), _linkMode(None) {} + + void blockUntil(CallbackCheck* other) { + _otherNotification.reset(new Notification); + _linkMode = Wait_For_Other; + + other->_otherNotification = _otherNotification; + other->_linkMode = Notify_Other; + } - _status = Status::OK(); - _notification.notifyOne(); + HostThreadPool::Callback getCallback() { + return stdx::bind(&CallbackCheck::noteCallback, this); + } - if (_linkMode == Wait_For_Other) - _otherNotification->waitToBeNotified(); - else if (_linkMode == Notify_Other) { - _otherNotification->notifyOne(); - } - } + HostThreadPool::Callback getHostCallback(const ConnectionString& host) { + return stdx::bind(&CallbackCheck::noteHostCallback, this, host); + } - void waitForCallback() { - _notification.waitToBeNotified(); - } + void noteHostCallback(const ConnectionString& host) { + _host = host; + noteCallback(); + } - Status getStatus() { - return _status; - } + void noteCallback() { + _status = Status::OK(); + _notification.notifyOne(); - const ConnectionString& getHost() { - return _host; + if (_linkMode == Wait_For_Other) + _otherNotification->waitToBeNotified(); + else if (_linkMode == Notify_Other) { + _otherNotification->notifyOne(); } - - private: - - Status _status; - Notification _notification; - - ConnectionString _host; - - shared_ptr<Notification> _otherNotification; - LinkMode _linkMode; - }; - - TEST(HostThreadPool, Schedule) { - - CallbackCheck cbCheck; - - // NOTE: pool must be initialized *after* the cbCheck that it executes - this avoids a - // subtle race where the cbCheck structure is disposed before the callback is complete. - HostThreadPool threadPool(1, true); - - threadPool.schedule(cbCheck.getCallback()); - - cbCheck.waitForCallback(); - ASSERT_OK(cbCheck.getStatus()); } - TEST(HostThreadPool, ScheduleTwoSerial) { - - CallbackCheck cbCheckA; - CallbackCheck cbCheckB; - - // NOTE: pool must be initialized *after* the cbCheck that it executes - HostThreadPool threadPool(1, true); - - threadPool.schedule(cbCheckA.getCallback()); - threadPool.schedule(cbCheckB.getCallback()); - - cbCheckB.waitForCallback(); - cbCheckA.waitForCallback(); - - ASSERT_OK(cbCheckA.getStatus()); - ASSERT_OK(cbCheckB.getStatus()); + void waitForCallback() { + _notification.waitToBeNotified(); } - TEST(HostThreadPool, ScheduleTwoParallel) { - - CallbackCheck cbCheckA; - CallbackCheck cbCheckB; - - // NOTE: pool must be initialized *after* the cbCheck that it executes - HostThreadPool threadPool(2, true); - - // Don't allow cbCheckA's callback to finish until cbCheckB's callback is processed - cbCheckA.blockUntil(&cbCheckB); - - threadPool.schedule(cbCheckA.getCallback()); - cbCheckA.waitForCallback(); - ASSERT_OK(cbCheckA.getStatus()); - // We're still blocking the thread processing cbCheckA's callback - - threadPool.schedule(cbCheckB.getCallback()); - cbCheckB.waitForCallback(); - ASSERT_OK(cbCheckB.getStatus()); + Status getStatus() { + return _status; } - TEST(HostThreadPool, ScheduleTwoHosts) { - - CallbackCheck cbCheckA; - CallbackCheck cbCheckB; - - // NOTE: pool must be initialized *after* the cbCheck that it executes - HostThreadPools threadPool(1, true); - - // Don't allow cbCheckA's callback to finish until cbCheckB's callback is processed. - // This means a single thread pool with a single thread would hang. - cbCheckA.blockUntil(&cbCheckB); - - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - - threadPool.schedule(hostA, cbCheckA.getHostCallback(hostA)); - cbCheckA.waitForCallback(); - ASSERT_OK(cbCheckA.getStatus()); - ASSERT_EQUALS(cbCheckA.getHost().toString(), hostA.toString()); - // We're still blocking the thread processing cbCheckA's callback - - threadPool.schedule(hostB, cbCheckB.getHostCallback(hostB)); - cbCheckB.waitForCallback(); - ASSERT_OK(cbCheckB.getStatus()); - ASSERT_EQUALS(cbCheckB.getHost().toString(), hostB.toString()); + const ConnectionString& getHost() { + return _host; } - class MockSystemEnv : public MultiHostQueryOp::SystemEnv { - private: +private: + Status _status; + Notification _notification; - struct MockHostInfo; - typedef map<ConnectionString, MockHostInfo*> HostInfoMap; + ConnectionString _host; - public: + shared_ptr<Notification> _otherNotification; + LinkMode _linkMode; +}; - MockSystemEnv(HostThreadPools* threadPool) : - _threadPool(threadPool), _mockTimeMillis(0) { - } +TEST(HostThreadPool, Schedule) { + CallbackCheck cbCheck; - virtual ~MockSystemEnv() { - for (HostInfoMap::iterator it = _mockHostInfo.begin(); it != _mockHostInfo.end(); - ++it) { - if (_threadPool) - _threadPool->waitUntilIdle(it->first); - delete it->second; - } - } + // NOTE: pool must be initialized *after* the cbCheck that it executes - this avoids a + // subtle race where the cbCheck structure is disposed before the callback is complete. + HostThreadPool threadPool(1, true); - void setHostThreadPools(HostThreadPools* threadPool) { - _threadPool = threadPool; - } + threadPool.schedule(cbCheck.getCallback()); - void addMockHostResultAt(const ConnectionString& host, int timeMillis) { - newMockHostResultAt(host, timeMillis, Status::OK(), NULL); - } + cbCheck.waitForCallback(); + ASSERT_OK(cbCheck.getStatus()); +} - void addMockHostErrorAt(const ConnectionString& host, int timeMillis, Status error) { - newMockHostResultAt(host, timeMillis, error, NULL); - } +TEST(HostThreadPool, ScheduleTwoSerial) { + CallbackCheck cbCheckA; + CallbackCheck cbCheckB; - void addMockHungHostAt(const ConnectionString& host, - int hangTimeMillis, - Notification* hangUntilNotify) { - newMockHostResultAt(host, hangTimeMillis, Status::OK(), hangUntilNotify); - } + // NOTE: pool must be initialized *after* the cbCheck that it executes + HostThreadPool threadPool(1, true); - void addMockTimestepAt(int timeMillis) { + threadPool.schedule(cbCheckA.getCallback()); + threadPool.schedule(cbCheckB.getCallback()); - // Add a mock query to a host we aren't using at the provided time - ConnectionString host = uassertStatusOK(ConnectionString::parse("$timestepHost:1000")); - newMockHostResultAt(host, timeMillis, Status::OK(), NULL); + cbCheckB.waitForCallback(); + cbCheckA.waitForCallback(); - // The query won't be scheduled by the multi op, so we need to do so ourselves - _threadPool->schedule(host, - stdx::bind(&MockSystemEnv::doBlockingQuerySwallowResult, - this, - host, - QuerySpec())); - } + ASSERT_OK(cbCheckA.getStatus()); + ASSERT_OK(cbCheckB.getStatus()); +} - Date_t currentTimeMillis() { - return Date_t::fromMillisSinceEpoch(_mockTimeMillis); - } +TEST(HostThreadPool, ScheduleTwoParallel) { + CallbackCheck cbCheckA; + CallbackCheck cbCheckB; - void doBlockingQuerySwallowResult(const ConnectionString& host, - const QuerySpec& query) { - StatusWith<DBClientCursor*> result = doBlockingQuery(host, query); - if (result.isOK()) - delete result.getValue(); - } + // NOTE: pool must be initialized *after* the cbCheck that it executes + HostThreadPool threadPool(2, true); - StatusWith<DBClientCursor*> doBlockingQuery(const ConnectionString& host, - const QuerySpec& query) { + // Don't allow cbCheckA's callback to finish until cbCheckB's callback is processed + cbCheckA.blockUntil(&cbCheckB); - ASSERT(_mockHostInfo.find(host) != _mockHostInfo.end()); + threadPool.schedule(cbCheckA.getCallback()); + cbCheckA.waitForCallback(); + ASSERT_OK(cbCheckA.getStatus()); + // We're still blocking the thread processing cbCheckA's callback - MockHostInfo& info = *(_mockHostInfo.find(host)->second); + threadPool.schedule(cbCheckB.getCallback()); + cbCheckB.waitForCallback(); + ASSERT_OK(cbCheckB.getStatus()); +} - if (info.prevHostActiveNotify) { - info.prevHostActiveNotify->waitToBeNotified(); - if (info.waitForPrevHostIdle) { - _threadPool->waitUntilIdle(info.prevHost); - } - } +TEST(HostThreadPool, ScheduleTwoHosts) { + CallbackCheck cbCheckA; + CallbackCheck cbCheckB; - _mockTimeMillis = info.queryTimeMillis; + // NOTE: pool must be initialized *after* the cbCheck that it executes + HostThreadPools threadPool(1, true); - if (info.nextHostActiveNotify) { - info.nextHostActiveNotify->notifyOne(); - } + // Don't allow cbCheckA's callback to finish until cbCheckB's callback is processed. + // This means a single thread pool with a single thread would hang. + cbCheckA.blockUntil(&cbCheckB); - if (info.hangUntilNotify) { - info.hangUntilNotify->waitToBeNotified(); - return StatusWith<DBClientCursor*>(ErrorCodes::InternalError, ""); - } - - if (!info.error.isOK()) { - return StatusWith<DBClientCursor*>(info.error); - } - - // - // Successful mock query - // + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - if (!info.conn) { - info.conn.reset(new DBClientConnection(false)); - // Need to do a connect failure so that we get an empty MessagingPort on the conn and - // the host name is set. - string errMsg; - ASSERT(!info.conn->connect(HostAndPort(host.toString()), errMsg)); - } - - return StatusWith<DBClientCursor*>(new DBClientCursor(info.conn.get(), - query.ns(), - query.query(), - query.ntoreturn(), - query.ntoskip(), - query.fieldsPtr(), - query.options(), - 0 /* batchSize */)); - } + threadPool.schedule(hostA, cbCheckA.getHostCallback(hostA)); + cbCheckA.waitForCallback(); + ASSERT_OK(cbCheckA.getStatus()); + ASSERT_EQUALS(cbCheckA.getHost().toString(), hostA.toString()); + // We're still blocking the thread processing cbCheckA's callback - private: + threadPool.schedule(hostB, cbCheckB.getHostCallback(hostB)); + cbCheckB.waitForCallback(); + ASSERT_OK(cbCheckB.getStatus()); + ASSERT_EQUALS(cbCheckB.getHost().toString(), hostB.toString()); +} - MockHostInfo* newMockHostResultAt(const ConnectionString& host, - int timeMillis, - const Status& error, - Notification* hangUntilNotify) { +class MockSystemEnv : public MultiHostQueryOp::SystemEnv { +private: + struct MockHostInfo; + typedef map<ConnectionString, MockHostInfo*> HostInfoMap; - ASSERT(_mockHostInfo.find(host) == _mockHostInfo.end()); +public: + MockSystemEnv(HostThreadPools* threadPool) : _threadPool(threadPool), _mockTimeMillis(0) {} - MockHostInfo* info = new MockHostInfo(timeMillis); - _mockHostInfo.insert(make_pair(host, info)); - info->error = error; - info->hangUntilNotify = hangUntilNotify; - - linkMockTimes(host, info); - return info; + virtual ~MockSystemEnv() { + for (HostInfoMap::iterator it = _mockHostInfo.begin(); it != _mockHostInfo.end(); ++it) { + if (_threadPool) + _threadPool->waitUntilIdle(it->first); + delete it->second; } + } - void linkMockTimes(const ConnectionString& host, MockHostInfo* info) { - - // - // This just basically sets up notifications between the processing of results such that - // the results are returned in the order defined by the _mockQueryTimes map. - // - // Idea is (second host result) waits for (first host result) thread to start and end, - // (third host result) waits for (second host result) thread to start and end, - // (fourth host result) waits for (third host result) thread to start and end, - // ... and so on ... - // + void setHostThreadPools(HostThreadPools* threadPool) { + _threadPool = threadPool; + } - ASSERT(_mockQueryTimes.find(info->queryTimeMillis) == _mockQueryTimes.end()); + void addMockHostResultAt(const ConnectionString& host, int timeMillis) { + newMockHostResultAt(host, timeMillis, Status::OK(), NULL); + } - HostQueryTimes::iterator prev = _mockQueryTimes.insert(make_pair(info->queryTimeMillis, - host)).first; + void addMockHostErrorAt(const ConnectionString& host, int timeMillis, Status error) { + newMockHostResultAt(host, timeMillis, error, NULL); + } - if (prev != _mockQueryTimes.begin()) - --prev; - else - prev = _mockQueryTimes.end(); + void addMockHungHostAt(const ConnectionString& host, + int hangTimeMillis, + Notification* hangUntilNotify) { + newMockHostResultAt(host, hangTimeMillis, Status::OK(), hangUntilNotify); + } - HostQueryTimes::iterator next = _mockQueryTimes.upper_bound(info->queryTimeMillis); + void addMockTimestepAt(int timeMillis) { + // Add a mock query to a host we aren't using at the provided time + ConnectionString host = uassertStatusOK(ConnectionString::parse("$timestepHost:1000")); + newMockHostResultAt(host, timeMillis, Status::OK(), NULL); - if (prev != _mockQueryTimes.end()) { + // The query won't be scheduled by the multi op, so we need to do so ourselves + _threadPool->schedule( + host, + stdx::bind(&MockSystemEnv::doBlockingQuerySwallowResult, this, host, QuerySpec())); + } - const ConnectionString& prevHost = prev->second; - MockHostInfo* prevInfo = _mockHostInfo.find(prevHost)->second; + Date_t currentTimeMillis() { + return Date_t::fromMillisSinceEpoch(_mockTimeMillis); + } - linkToNext(prevHost, prevInfo, info); - } + void doBlockingQuerySwallowResult(const ConnectionString& host, const QuerySpec& query) { + StatusWith<DBClientCursor*> result = doBlockingQuery(host, query); + if (result.isOK()) + delete result.getValue(); + } - if (next != _mockQueryTimes.end()) { + StatusWith<DBClientCursor*> doBlockingQuery(const ConnectionString& host, + const QuerySpec& query) { + ASSERT(_mockHostInfo.find(host) != _mockHostInfo.end()); - const ConnectionString& nextHost = next->second; - MockHostInfo* nextInfo = _mockHostInfo.find(nextHost)->second; + MockHostInfo& info = *(_mockHostInfo.find(host)->second); - linkToNext(host, info, nextInfo); + if (info.prevHostActiveNotify) { + info.prevHostActiveNotify->waitToBeNotified(); + if (info.waitForPrevHostIdle) { + _threadPool->waitUntilIdle(info.prevHost); } } - void linkToNext(const ConnectionString& host, MockHostInfo* info, MockHostInfo* nextInfo) { - - nextInfo->prevHost = host; - - nextInfo->prevHostActiveNotify.reset(new Notification()); - info->nextHostActiveNotify = nextInfo->prevHostActiveNotify.get(); + _mockTimeMillis = info.queryTimeMillis; - nextInfo->waitForPrevHostIdle = info->hangUntilNotify == NULL; + if (info.nextHostActiveNotify) { + info.nextHostActiveNotify->notifyOne(); } - // Not owned here, needed to allow ordering of mock queries - HostThreadPools* _threadPool; - - int _mockTimeMillis; - - typedef map<int, ConnectionString> HostQueryTimes; - HostQueryTimes _mockQueryTimes; - - struct MockHostInfo { - - MockHostInfo(int queryTimeMillis) : - nextHostActiveNotify( NULL), - waitForPrevHostIdle(false), - queryTimeMillis(queryTimeMillis), - hangUntilNotify( NULL), - error(Status::OK()) { - } - - Notification* nextHostActiveNotify; - - ConnectionString prevHost; - unique_ptr<Notification> prevHostActiveNotify; - bool waitForPrevHostIdle; - - int queryTimeMillis; + if (info.hangUntilNotify) { + info.hangUntilNotify->waitToBeNotified(); + return StatusWith<DBClientCursor*>(ErrorCodes::InternalError, ""); + } - unique_ptr<DBClientConnection> conn; - Notification* hangUntilNotify; - Status error; - }; + if (!info.error.isOK()) { + return StatusWith<DBClientCursor*>(info.error); + } - HostInfoMap _mockHostInfo; + // + // Successful mock query + // - }; + if (!info.conn) { + info.conn.reset(new DBClientConnection(false)); + // Need to do a connect failure so that we get an empty MessagingPort on the conn and + // the host name is set. + string errMsg; + ASSERT(!info.conn->connect(HostAndPort(host.toString()), errMsg)); + } - QuerySpec buildSpec(StringData ns, const BSONObj& query) { - return QuerySpec(ns.toString(), query, BSONObj(), 0, 0, 0); + return StatusWith<DBClientCursor*>(new DBClientCursor(info.conn.get(), + query.ns(), + query.query(), + query.ntoreturn(), + query.ntoskip(), + query.fieldsPtr(), + query.options(), + 0 /* batchSize */)); } - // - // Tests for the MultiHostQueryOp - // - - TEST(MultiHostQueryOp, SingleHost) { - - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); - - ConnectionString host = uassertStatusOK(ConnectionString::parse("$host:1000")); - vector<ConnectionString> hosts; - hosts.push_back(host); - - mockSystem.addMockHostResultAt(host, 1000); - - MultiHostQueryOp queryOp(&mockSystem, &threadPool); +private: + MockHostInfo* newMockHostResultAt(const ConnectionString& host, + int timeMillis, + const Status& error, + Notification* hangUntilNotify) { + ASSERT(_mockHostInfo.find(host) == _mockHostInfo.end()); - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 2000); + MockHostInfo* info = new MockHostInfo(timeMillis); + _mockHostInfo.insert(make_pair(host, info)); + info->error = error; + info->hangUntilNotify = hangUntilNotify; - ASSERT_OK(result.getStatus()); - ASSERT(NULL != result.getValue()); - ASSERT_EQUALS(result.getValue()->originalHost(), host.toString()); - delete result.getValue(); + linkMockTimes(host, info); + return info; } - TEST(MultiHostQueryOp, SingleHostError) { - - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); - - ConnectionString host = uassertStatusOK(ConnectionString::parse("$host:1000")); - vector<ConnectionString> hosts; - hosts.push_back(host); - - Status hostError = Status(ErrorCodes::InternalError, ""); - mockSystem.addMockHostErrorAt(host, 1000, hostError); - - MultiHostQueryOp queryOp(&mockSystem, &threadPool); - - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 2000); - - ASSERT_EQUALS(result.getStatus().code(), hostError.code()); - } + void linkMockTimes(const ConnectionString& host, MockHostInfo* info) { + // + // This just basically sets up notifications between the processing of results such that + // the results are returned in the order defined by the _mockQueryTimes map. + // + // Idea is (second host result) waits for (first host result) thread to start and end, + // (third host result) waits for (second host result) thread to start and end, + // (fourth host result) waits for (third host result) thread to start and end, + // ... and so on ... + // - TEST(MultiHostQueryOp, SingleHostHang) { + ASSERT(_mockQueryTimes.find(info->queryTimeMillis) == _mockQueryTimes.end()); - // Initialize notifier before the thread pool, otherwise we may dispose while threads are - // active - Notification unhangNotify; + HostQueryTimes::iterator prev = + _mockQueryTimes.insert(make_pair(info->queryTimeMillis, host)).first; - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); + if (prev != _mockQueryTimes.begin()) + --prev; + else + prev = _mockQueryTimes.end(); - ConnectionString host = uassertStatusOK(ConnectionString::parse("$host:1000")); - vector<ConnectionString> hosts; - hosts.push_back(host); + HostQueryTimes::iterator next = _mockQueryTimes.upper_bound(info->queryTimeMillis); - mockSystem.addMockHungHostAt(host, 4000, &unhangNotify); + if (prev != _mockQueryTimes.end()) { + const ConnectionString& prevHost = prev->second; + MockHostInfo* prevInfo = _mockHostInfo.find(prevHost)->second; - MultiHostQueryOp queryOp(&mockSystem, &threadPool); + linkToNext(prevHost, prevInfo, info); + } - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 2000); - // Unhang before checking status, in case it throws - unhangNotify.notifyOne(); + if (next != _mockQueryTimes.end()) { + const ConnectionString& nextHost = next->second; + MockHostInfo* nextInfo = _mockHostInfo.find(nextHost)->second; - ASSERT_EQUALS(result.getStatus().code(), ErrorCodes::NetworkTimeout); + linkToNext(host, info, nextInfo); + } } - TEST(MultiHostQueryOp, TwoHostResponses) { + void linkToNext(const ConnectionString& host, MockHostInfo* info, MockHostInfo* nextInfo) { + nextInfo->prevHost = host; - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); + nextInfo->prevHostActiveNotify.reset(new Notification()); + info->nextHostActiveNotify = nextInfo->prevHostActiveNotify.get(); - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); + nextInfo->waitForPrevHostIdle = info->hangUntilNotify == NULL; + } - // Make sure we return the first response, from hostB at time 1000 - mockSystem.addMockHostResultAt(hostA, 2000); - mockSystem.addMockHostResultAt(hostB, 1000); + // Not owned here, needed to allow ordering of mock queries + HostThreadPools* _threadPool; - MultiHostQueryOp queryOp(&mockSystem, &threadPool); + int _mockTimeMillis; - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); + typedef map<int, ConnectionString> HostQueryTimes; + HostQueryTimes _mockQueryTimes; - ASSERT_OK(result.getStatus()); - ASSERT(NULL != result.getValue()); - ASSERT_EQUALS(result.getValue()->originalHost(), hostB.toString()); - delete result.getValue(); - } + struct MockHostInfo { + MockHostInfo(int queryTimeMillis) + : nextHostActiveNotify(NULL), + waitForPrevHostIdle(false), + queryTimeMillis(queryTimeMillis), + hangUntilNotify(NULL), + error(Status::OK()) {} - TEST(MultiHostQueryOp, TwoHostsOneErrorResponse) { + Notification* nextHostActiveNotify; - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); + ConnectionString prevHost; + unique_ptr<Notification> prevHostActiveNotify; + bool waitForPrevHostIdle; - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); + int queryTimeMillis; - // The first response is a host error, the second is a successful result - Status hostError = Status(ErrorCodes::InternalError, ""); - mockSystem.addMockHostErrorAt(hostA, 1000, hostError); - mockSystem.addMockHostResultAt(hostB, 2000); - - MultiHostQueryOp queryOp(&mockSystem, &threadPool); + unique_ptr<DBClientConnection> conn; + Notification* hangUntilNotify; + Status error; + }; - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); + HostInfoMap _mockHostInfo; +}; - ASSERT_OK(result.getStatus()); - ASSERT(NULL != result.getValue()); - ASSERT_EQUALS(result.getValue()->originalHost(), hostB.toString()); - delete result.getValue(); - } +QuerySpec buildSpec(StringData ns, const BSONObj& query) { + return QuerySpec(ns.toString(), query, BSONObj(), 0, 0, 0); +} - TEST(MultiHostQueryOp, TwoHostsBothErrors) { +// +// Tests for the MultiHostQueryOp +// - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); +TEST(MultiHostQueryOp, SingleHost) { + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); + ConnectionString host = uassertStatusOK(ConnectionString::parse("$host:1000")); + vector<ConnectionString> hosts; + hosts.push_back(host); - // Both responses are errors - Status hostError = Status(ErrorCodes::InternalError, ""); - mockSystem.addMockHostErrorAt(hostA, 1000, hostError); - mockSystem.addMockHostErrorAt(hostB, 2000, hostError); + mockSystem.addMockHostResultAt(host, 1000); - MultiHostQueryOp queryOp(&mockSystem, &threadPool); + MultiHostQueryOp queryOp(&mockSystem, &threadPool); - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 2000); - ASSERT_EQUALS(result.getStatus().code(), ErrorCodes::MultipleErrorsOccurred); - } + ASSERT_OK(result.getStatus()); + ASSERT(NULL != result.getValue()); + ASSERT_EQUALS(result.getValue()->originalHost(), host.toString()); + delete result.getValue(); +} - TEST(MultiHostQueryOp, TwoHostsOneHang) { +TEST(MultiHostQueryOp, SingleHostError) { + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - // Initialize notifier before the thread pool - Notification unhangNotify; + ConnectionString host = uassertStatusOK(ConnectionString::parse("$host:1000")); + vector<ConnectionString> hosts; + hosts.push_back(host); - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); + Status hostError = Status(ErrorCodes::InternalError, ""); + mockSystem.addMockHostErrorAt(host, 1000, hostError); - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); + MultiHostQueryOp queryOp(&mockSystem, &threadPool); - // One host hangs - mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); - mockSystem.addMockHostResultAt(hostB, 2000); + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 2000); - MultiHostQueryOp queryOp(&mockSystem, &threadPool); + ASSERT_EQUALS(result.getStatus().code(), hostError.code()); +} - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); - // Unhang before checking status, in case it throws - unhangNotify.notifyOne(); +TEST(MultiHostQueryOp, SingleHostHang) { + // Initialize notifier before the thread pool, otherwise we may dispose while threads are + // active + Notification unhangNotify; - ASSERT_OK(result.getStatus()); - ASSERT(NULL != result.getValue()); - ASSERT_EQUALS(result.getValue()->originalHost(), hostB.toString()); - delete result.getValue(); - } + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - TEST(MultiHostQueryOp, TwoHostsOneHangOneError) { + ConnectionString host = uassertStatusOK(ConnectionString::parse("$host:1000")); + vector<ConnectionString> hosts; + hosts.push_back(host); - // Initialize notifier before the thread pool - Notification unhangNotify; + mockSystem.addMockHungHostAt(host, 4000, &unhangNotify); - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); + MultiHostQueryOp queryOp(&mockSystem, &threadPool); - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 2000); + // Unhang before checking status, in case it throws + unhangNotify.notifyOne(); - // One host hangs, one host has an error (at the mock timeout point so the query finishes) - Status hostError = Status(ErrorCodes::InternalError, ""); - mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); - mockSystem.addMockHostErrorAt(hostB, 3000, hostError); - mockSystem.addMockTimestepAt(4000); + ASSERT_EQUALS(result.getStatus().code(), ErrorCodes::NetworkTimeout); +} - MultiHostQueryOp queryOp(&mockSystem, &threadPool); +TEST(MultiHostQueryOp, TwoHostResponses) { + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 4000); - // Unhang before checking status, in case it throws - unhangNotify.notifyOne(); + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); - ASSERT_EQUALS(result.getStatus().code(), hostError.code()); - } + // Make sure we return the first response, from hostB at time 1000 + mockSystem.addMockHostResultAt(hostA, 2000); + mockSystem.addMockHostResultAt(hostB, 1000); - TEST(MultiHostQueryOp, ThreeHostsOneHang) { - // Initialize notifier before the thread pool - Notification unhangNotify; + MultiHostQueryOp queryOp(&mockSystem, &threadPool); - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - ConnectionString hostC = uassertStatusOK(ConnectionString::parse("$hostC:1000")); + ASSERT_OK(result.getStatus()); + ASSERT(NULL != result.getValue()); + ASSERT_EQUALS(result.getValue()->originalHost(), hostB.toString()); + delete result.getValue(); +} - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); - hosts.push_back(hostC); +TEST(MultiHostQueryOp, TwoHostsOneErrorResponse) { + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - // Host A hangs - mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); - mockSystem.addMockHostResultAt(hostB, 2000); - mockSystem.addMockHostResultAt(hostC, 3000); + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); - MultiHostQueryOp queryOp(&mockSystem, &threadPool); + // The first response is a host error, the second is a successful result + Status hostError = Status(ErrorCodes::InternalError, ""); + mockSystem.addMockHostErrorAt(hostA, 1000, hostError); + mockSystem.addMockHostResultAt(hostB, 2000); - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 4000); + MultiHostQueryOp queryOp(&mockSystem, &threadPool); - // Unhang before checking status, in case it throws - unhangNotify.notifyOne(); + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); - ASSERT_OK(result.getStatus()); - ASSERT(NULL != result.getValue()); + ASSERT_OK(result.getStatus()); + ASSERT(NULL != result.getValue()); + ASSERT_EQUALS(result.getValue()->originalHost(), hostB.toString()); + delete result.getValue(); +} - // We should never have results from hostA - ASSERT_NOT_EQUALS(result.getValue()->originalHost(), hostA.toString()); +TEST(MultiHostQueryOp, TwoHostsBothErrors) { + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - delete result.getValue(); - } + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); - TEST(MultiHostQueryOp, ThreeHostsTwoErrors) { + // Both responses are errors + Status hostError = Status(ErrorCodes::InternalError, ""); + mockSystem.addMockHostErrorAt(hostA, 1000, hostError); + mockSystem.addMockHostErrorAt(hostB, 2000, hostError); - // Initialize notifier before the thread pool - Notification unhangNotify; + MultiHostQueryOp queryOp(&mockSystem, &threadPool); - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - ConnectionString hostC = uassertStatusOK(ConnectionString::parse("$hostC:1000")); - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); - hosts.push_back(hostC); + ASSERT_EQUALS(result.getStatus().code(), ErrorCodes::MultipleErrorsOccurred); +} - // One host hangs, two hosts have errors (finish at mock timeout point so query ends) - Status hostError = Status(ErrorCodes::InternalError, ""); - mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); - mockSystem.addMockHostErrorAt(hostB, 4000, hostError); - mockSystem.addMockHostErrorAt(hostC, 2000, hostError); - mockSystem.addMockTimestepAt(5000); +TEST(MultiHostQueryOp, TwoHostsOneHang) { + // Initialize notifier before the thread pool + Notification unhangNotify; - MultiHostQueryOp queryOp(&mockSystem, &threadPool); + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 5000); - // Unhang before checking status, in case it throws - unhangNotify.notifyOne(); + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); - ASSERT_EQUALS(result.getStatus().code(), ErrorCodes::MultipleErrorsOccurred); - } + // One host hangs + mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); + mockSystem.addMockHostResultAt(hostB, 2000); + + MultiHostQueryOp queryOp(&mockSystem, &threadPool); + + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); + // Unhang before checking status, in case it throws + unhangNotify.notifyOne(); + + ASSERT_OK(result.getStatus()); + ASSERT(NULL != result.getValue()); + ASSERT_EQUALS(result.getValue()->originalHost(), hostB.toString()); + delete result.getValue(); +} - TEST(MultiHostQueryOp, ThreeHostsOneHangOneError) { +TEST(MultiHostQueryOp, TwoHostsOneHangOneError) { + // Initialize notifier before the thread pool + Notification unhangNotify; - // Initialize notifier before the thread pool - Notification unhangNotify; + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - HostThreadPools threadPool(1, true); - MockSystemEnv mockSystem(&threadPool); + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); + + // One host hangs, one host has an error (at the mock timeout point so the query finishes) + Status hostError = Status(ErrorCodes::InternalError, ""); + mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); + mockSystem.addMockHostErrorAt(hostB, 3000, hostError); + mockSystem.addMockTimestepAt(4000); + + MultiHostQueryOp queryOp(&mockSystem, &threadPool); + + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 4000); + // Unhang before checking status, in case it throws + unhangNotify.notifyOne(); + + ASSERT_EQUALS(result.getStatus().code(), hostError.code()); +} + +TEST(MultiHostQueryOp, ThreeHostsOneHang) { + // Initialize notifier before the thread pool + Notification unhangNotify; + + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); + + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + ConnectionString hostC = uassertStatusOK(ConnectionString::parse("$hostC:1000")); + + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); + hosts.push_back(hostC); + + // Host A hangs + mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); + mockSystem.addMockHostResultAt(hostB, 2000); + mockSystem.addMockHostResultAt(hostC, 3000); + + MultiHostQueryOp queryOp(&mockSystem, &threadPool); + + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 4000); + + // Unhang before checking status, in case it throws + unhangNotify.notifyOne(); + + ASSERT_OK(result.getStatus()); + ASSERT(NULL != result.getValue()); - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - ConnectionString hostC = uassertStatusOK(ConnectionString::parse("$hostC:1000")); - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); - hosts.push_back(hostC); + // We should never have results from hostA + ASSERT_NOT_EQUALS(result.getValue()->originalHost(), hostA.toString()); - // One host hangs, two hosts have errors (finish at mock timeout point so query ends) - Status hostError = Status(ErrorCodes::InternalError, ""); - mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); - mockSystem.addMockHostErrorAt(hostB, 2000, hostError); - mockSystem.addMockHostResultAt(hostC, 3000); + delete result.getValue(); +} + +TEST(MultiHostQueryOp, ThreeHostsTwoErrors) { + // Initialize notifier before the thread pool + Notification unhangNotify; + + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - MultiHostQueryOp queryOp(&mockSystem, &threadPool); + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + ConnectionString hostC = uassertStatusOK(ConnectionString::parse("$hostC:1000")); + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); + hosts.push_back(hostC); - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 4000); - // Unhang before checking status, in case it throws - unhangNotify.notifyOne(); + // One host hangs, two hosts have errors (finish at mock timeout point so query ends) + Status hostError = Status(ErrorCodes::InternalError, ""); + mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); + mockSystem.addMockHostErrorAt(hostB, 4000, hostError); + mockSystem.addMockHostErrorAt(hostC, 2000, hostError); + mockSystem.addMockTimestepAt(5000); - ASSERT_OK(result.getStatus()); - ASSERT(NULL != result.getValue()); - ASSERT_EQUALS(result.getValue()->originalHost(), hostC.toString()); - delete result.getValue(); - } + MultiHostQueryOp queryOp(&mockSystem, &threadPool); + + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 5000); + // Unhang before checking status, in case it throws + unhangNotify.notifyOne(); + + ASSERT_EQUALS(result.getStatus().code(), ErrorCodes::MultipleErrorsOccurred); +} - TEST(MultiHostQueryOp, TwoHostsOneHangUnscoped) { +TEST(MultiHostQueryOp, ThreeHostsOneHangOneError) { + // Initialize notifier before the thread pool + Notification unhangNotify; - // Initialize notifier before the thread pool - Notification unhangNotify; + HostThreadPools threadPool(1, true); + MockSystemEnv mockSystem(&threadPool); - // Create a thread pool which detaches itself from outstanding work on cleanup - unique_ptr<HostThreadPools> threadPool(new HostThreadPools(1, false)); - MockSystemEnv mockSystem(threadPool.get()); + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + ConnectionString hostC = uassertStatusOK(ConnectionString::parse("$hostC:1000")); + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); + hosts.push_back(hostC); - ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); - ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); - vector<ConnectionString> hosts; - hosts.push_back(hostA); - hosts.push_back(hostB); + // One host hangs, two hosts have errors (finish at mock timeout point so query ends) + Status hostError = Status(ErrorCodes::InternalError, ""); + mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); + mockSystem.addMockHostErrorAt(hostB, 2000, hostError); + mockSystem.addMockHostResultAt(hostC, 3000); - // One host hangs - mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); - mockSystem.addMockHostResultAt(hostB, 2000); + MultiHostQueryOp queryOp(&mockSystem, &threadPool); - MultiHostQueryOp queryOp(&mockSystem, threadPool.get()); + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 4000); + // Unhang before checking status, in case it throws + unhangNotify.notifyOne(); - QuerySpec query; - StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); + ASSERT_OK(result.getStatus()); + ASSERT(NULL != result.getValue()); + ASSERT_EQUALS(result.getValue()->originalHost(), hostC.toString()); + delete result.getValue(); +} - // Clean up the thread pool - mockSystem.setHostThreadPools( NULL); - threadPool.reset(); +TEST(MultiHostQueryOp, TwoHostsOneHangUnscoped) { + // Initialize notifier before the thread pool + Notification unhangNotify; - // Unhang before checking status, in case it throws - unhangNotify.notifyOne(); + // Create a thread pool which detaches itself from outstanding work on cleanup + unique_ptr<HostThreadPools> threadPool(new HostThreadPools(1, false)); + MockSystemEnv mockSystem(threadPool.get()); - ASSERT_OK(result.getStatus()); - ASSERT(NULL != result.getValue()); - ASSERT_EQUALS(result.getValue()->originalHost(), hostB.toString()); - delete result.getValue(); + ConnectionString hostA = uassertStatusOK(ConnectionString::parse("$hostA:1000")); + ConnectionString hostB = uassertStatusOK(ConnectionString::parse("$hostB:1000")); + vector<ConnectionString> hosts; + hosts.push_back(hostA); + hosts.push_back(hostB); - // Make sure we get the next result - result = queryOp.waitForNextResult(Date_t::fromMillisSinceEpoch(4000)); + // One host hangs + mockSystem.addMockHungHostAt(hostA, 1000, &unhangNotify); + mockSystem.addMockHostResultAt(hostB, 2000); - ASSERT_EQUALS(result.getStatus().code(), ErrorCodes::InternalError); - } + MultiHostQueryOp queryOp(&mockSystem, threadPool.get()); -} // unnamed namespace + QuerySpec query; + StatusWith<DBClientCursor*> result = queryOp.queryAny(hosts, query, 3000); + + // Clean up the thread pool + mockSystem.setHostThreadPools(NULL); + threadPool.reset(); + + // Unhang before checking status, in case it throws + unhangNotify.notifyOne(); + + ASSERT_OK(result.getStatus()); + ASSERT(NULL != result.getValue()); + ASSERT_EQUALS(result.getValue()->originalHost(), hostB.toString()); + delete result.getValue(); + + // Make sure we get the next result + result = queryOp.waitForNextResult(Date_t::fromMillisSinceEpoch(4000)); + + ASSERT_EQUALS(result.getStatus().code(), ErrorCodes::InternalError); +} + +} // unnamed namespace diff --git a/src/mongo/s/client/scc_fast_query_handler.cpp b/src/mongo/s/client/scc_fast_query_handler.cpp index f09dd767a21..20a8f1a0193 100644 --- a/src/mongo/s/client/scc_fast_query_handler.cpp +++ b/src/mongo/s/client/scc_fast_query_handler.cpp @@ -45,214 +45,196 @@ namespace mongo { - using std::unique_ptr; - using std::endl; - using std::string; - using std::vector; - - /** - * This parameter turns on fastest config reads for auth data only - *.system.users collections - * and the "usersInfo" command. This should be enough to prevent a non-responsive config from - * hanging other operations in a cluster. - */ - MONGO_EXPORT_SERVER_PARAMETER(internalSCCAllowFastestAuthConfigReads, bool, false); - - /** - * TESTING ONLY. - * - * This parameter turns on fastest config reads for *all* config.* collections except those - * deemed extremely critical (config.version, config.locks). - * - * NOT FOR PRODUCTION USE. - */ - MONGO_EXPORT_SERVER_PARAMETER(internalSCCAllowFastestMetadataConfigReads, bool, false); +using std::unique_ptr; +using std::endl; +using std::string; +using std::vector; - // - // The shared environment for MultiHostQueries - // - - namespace { - - class MultiQueryEnv : public MultiHostQueryOp::SystemEnv { - public: - - virtual ~MultiQueryEnv() { - } - - Date_t currentTimeMillis(); +/** + * This parameter turns on fastest config reads for auth data only - *.system.users collections + * and the "usersInfo" command. This should be enough to prevent a non-responsive config from + * hanging other operations in a cluster. + */ +MONGO_EXPORT_SERVER_PARAMETER(internalSCCAllowFastestAuthConfigReads, bool, false); - StatusWith<DBClientCursor*> doBlockingQuery(const ConnectionString& host, - const QuerySpec& query); - }; +/** + * TESTING ONLY. + * + * This parameter turns on fastest config reads for *all* config.* collections except those + * deemed extremely critical (config.version, config.locks). + * + * NOT FOR PRODUCTION USE. + */ +MONGO_EXPORT_SERVER_PARAMETER(internalSCCAllowFastestMetadataConfigReads, bool, false); - StatusWith<DBClientCursor*> MultiQueryEnv::doBlockingQuery(const ConnectionString& host, - const QuerySpec& query) { +// +// The shared environment for MultiHostQueries +// - // - // Note that this function may be active during shutdown. This means that we must - // handle connection pool shutdown exceptions (uasserts). The results of this - // operation must then be correctly discarded by the calling thread. - // +namespace { - unique_ptr<DBClientCursor> cursor; +class MultiQueryEnv : public MultiHostQueryOp::SystemEnv { +public: + virtual ~MultiQueryEnv() {} - try { + Date_t currentTimeMillis(); - ScopedDbConnection conn(host, 30.0 /* timeout secs */); + StatusWith<DBClientCursor*> doBlockingQuery(const ConnectionString& host, + const QuerySpec& query); +}; - cursor = conn->query(query.ns(), - query.filter(), - query.ntoreturn(), - query.ntoskip(), - query.fieldsPtr(), - query.options(), - 0); +StatusWith<DBClientCursor*> MultiQueryEnv::doBlockingQuery(const ConnectionString& host, + const QuerySpec& query) { + // + // Note that this function may be active during shutdown. This means that we must + // handle connection pool shutdown exceptions (uasserts). The results of this + // operation must then be correctly discarded by the calling thread. + // - if ( NULL == cursor.get()) { + unique_ptr<DBClientCursor> cursor; - // Confusingly, exceptions here are written to the log, not thrown + try { + ScopedDbConnection conn(host, 30.0 /* timeout secs */); - StringBuilder builder; - builder << "error querying server " << host.toString() - << ", could not create cursor for query"; + cursor = conn->query(query.ns(), + query.filter(), + query.ntoreturn(), + query.ntoskip(), + query.fieldsPtr(), + query.options(), + 0); - warning() << builder.str() << endl; - return StatusWith<DBClientCursor*>(ErrorCodes::HostUnreachable, builder.str()); - } + if (NULL == cursor.get()) { + // Confusingly, exceptions here are written to the log, not thrown - // Confusingly, this *detaches* the cursor from the connection it was established - // on, further getMore calls will use a (potentially different) ScopedDbConnection. - // - // Calls done() too. - cursor->attach(&conn); - } - catch (const DBException& ex) { - return StatusWith<DBClientCursor*>(ex.toStatus()); - } + StringBuilder builder; + builder << "error querying server " << host.toString() + << ", could not create cursor for query"; - return StatusWith<DBClientCursor*>(cursor.release()); + warning() << builder.str() << endl; + return StatusWith<DBClientCursor*>(ErrorCodes::HostUnreachable, builder.str()); } - Date_t MultiQueryEnv::currentTimeMillis() { - return jsTime(); - } - } - - // Shared networking environment which executes queries. - // NOTE: This environment must stay in scope as long as per-host threads are executing queries - - // i.e. for the lifetime of the server. - // Once the thread pools are disposed and connections shut down, the per-host threads should - // be self-contained and correctly shut down after discarding the results. - static MultiQueryEnv* _multiQueryEnv; - - namespace { - + // Confusingly, this *detaches* the cursor from the connection it was established + // on, further getMore calls will use a (potentially different) ScopedDbConnection. // - // Create the shared multi-query environment at startup - // - - MONGO_INITIALIZER(InitMultiQueryEnv)(InitializerContext* context) { - // Leaked intentionally - _multiQueryEnv = new MultiQueryEnv(); - return Status::OK(); - } + // Calls done() too. + cursor->attach(&conn); + } catch (const DBException& ex) { + return StatusWith<DBClientCursor*>(ex.toStatus()); } - // - // Per-SCC handling of queries - // + return StatusWith<DBClientCursor*>(cursor.release()); +} - SCCFastQueryHandler::SCCFastQueryHandler() : - _queryThreads(1, false) { - } +Date_t MultiQueryEnv::currentTimeMillis() { + return jsTime(); +} +} - bool SCCFastQueryHandler::canHandleQuery(const string& ns, Query query) { +// Shared networking environment which executes queries. +// NOTE: This environment must stay in scope as long as per-host threads are executing queries - +// i.e. for the lifetime of the server. +// Once the thread pools are disposed and connections shut down, the per-host threads should +// be self-contained and correctly shut down after discarding the results. +static MultiQueryEnv* _multiQueryEnv; - if (!internalSCCAllowFastestAuthConfigReads && - !internalSCCAllowFastestMetadataConfigReads) { - return false; - } +namespace { - // - // More operations can be added here - // - // NOTE: Not all operations actually pass through the SCC _queryOnActive path - notable - // exceptions include anything related to direct query ops and direct operations for - // connection maintenance. - // +// +// Create the shared multi-query environment at startup +// - NamespaceString nss(ns); - if (nss.isCommand()) { - BSONObj cmdObj = query.getFilter(); - string cmdName = cmdObj.firstElement().fieldName(); - if (cmdName == "usersInfo") - return true; - } - else if (nss.coll() == "system.users") { - return true; - } +MONGO_INITIALIZER(InitMultiQueryEnv)(InitializerContext* context) { + // Leaked intentionally + _multiQueryEnv = new MultiQueryEnv(); + return Status::OK(); +} +} - // - // Allow fastest config reads for all collections except for those involved in locks and - // cluster versioning. - // +// +// Per-SCC handling of queries +// - if (!internalSCCAllowFastestMetadataConfigReads) - return false; +SCCFastQueryHandler::SCCFastQueryHandler() : _queryThreads(1, false) {} - if (nss.db() != "config") - return false; +bool SCCFastQueryHandler::canHandleQuery(const string& ns, Query query) { + if (!internalSCCAllowFastestAuthConfigReads && !internalSCCAllowFastestMetadataConfigReads) { + return false; + } - if (nss.coll() != "version" && nss.coll() != "locks" && nss.coll() != "lockpings") { - return true; - } + // + // More operations can be added here + // + // NOTE: Not all operations actually pass through the SCC _queryOnActive path - notable + // exceptions include anything related to direct query ops and direct operations for + // connection maintenance. + // - return false; + NamespaceString nss(ns); + if (nss.isCommand()) { + BSONObj cmdObj = query.getFilter(); + string cmdName = cmdObj.firstElement().fieldName(); + if (cmdName == "usersInfo") + return true; + } else if (nss.coll() == "system.users") { + return true; } - static vector<ConnectionString> getHosts(const vector<string> hostStrings) { + // + // Allow fastest config reads for all collections except for those involved in locks and + // cluster versioning. + // - vector<ConnectionString> hosts; - for (vector<string>::const_iterator it = hostStrings.begin(); it != hostStrings.end(); - ++it) { + if (!internalSCCAllowFastestMetadataConfigReads) + return false; - string errMsg; - ConnectionString host; - hosts.push_back(ConnectionString::parse(*it, errMsg)); - invariant( hosts.back().type() != ConnectionString::INVALID ); - } + if (nss.db() != "config") + return false; - return hosts; + if (nss.coll() != "version" && nss.coll() != "locks" && nss.coll() != "lockpings") { + return true; } - unique_ptr<DBClientCursor> SCCFastQueryHandler::handleQuery(const vector<string>& hostStrings, - const string& ns, - Query query, - int nToReturn, - int nToSkip, - const BSONObj* fieldsToReturn, - int queryOptions, - int batchSize) { - - MultiHostQueryOp queryOp(_multiQueryEnv, &_queryThreads); - - QuerySpec querySpec(ns, - query.obj, - fieldsToReturn ? *fieldsToReturn : BSONObj(), - nToSkip, - nToReturn, - queryOptions); - - // TODO: Timeout must be passed down here as well - 30s timeout may not be applicable for - // all operations handled. - StatusWith<DBClientCursor*> status = queryOp.queryAny(getHosts(hostStrings), - querySpec, - 30 * 1000); - uassertStatusOK(status.getStatus()); - - unique_ptr<DBClientCursor> cursor(status.getValue()); - return cursor; + return false; +} + +static vector<ConnectionString> getHosts(const vector<string> hostStrings) { + vector<ConnectionString> hosts; + for (vector<string>::const_iterator it = hostStrings.begin(); it != hostStrings.end(); ++it) { + string errMsg; + ConnectionString host; + hosts.push_back(ConnectionString::parse(*it, errMsg)); + invariant(hosts.back().type() != ConnectionString::INVALID); } + return hosts; } +unique_ptr<DBClientCursor> SCCFastQueryHandler::handleQuery(const vector<string>& hostStrings, + const string& ns, + Query query, + int nToReturn, + int nToSkip, + const BSONObj* fieldsToReturn, + int queryOptions, + int batchSize) { + MultiHostQueryOp queryOp(_multiQueryEnv, &_queryThreads); + + QuerySpec querySpec(ns, + query.obj, + fieldsToReturn ? *fieldsToReturn : BSONObj(), + nToSkip, + nToReturn, + queryOptions); + + // TODO: Timeout must be passed down here as well - 30s timeout may not be applicable for + // all operations handled. + StatusWith<DBClientCursor*> status = + queryOp.queryAny(getHosts(hostStrings), querySpec, 30 * 1000); + uassertStatusOK(status.getStatus()); + + unique_ptr<DBClientCursor> cursor(status.getValue()); + return cursor; +} +} diff --git a/src/mongo/s/client/scc_fast_query_handler.h b/src/mongo/s/client/scc_fast_query_handler.h index bc942556389..c2079e717ff 100644 --- a/src/mongo/s/client/scc_fast_query_handler.h +++ b/src/mongo/s/client/scc_fast_query_handler.h @@ -33,44 +33,40 @@ namespace mongo { - /** - * Query handler that plugs in to a SyncClusterConnection and allows query on fastest host - * (if enabled). - * - * Glue code which shields the MultiHostQuery and server parameters from the separate client - * module which knows about neither. - * - * There is a *single* SCCFastQueryHandler for every SCC. Each SCCFastQueryHandler contains - * its own thread pool (lazily initialized) so that at maximum there is a thread-per-SCC-host - * and this thread may have an open connection to the host until it times out. - * If using the "fastestConfigReads" options, clients must be ready for the additional thread - * and connection load when configs are slow. - */ - class SCCFastQueryHandler : public SyncClusterConnection::QueryHandler { - public: - - SCCFastQueryHandler(); - - virtual ~SCCFastQueryHandler() { - } - - virtual bool canHandleQuery(const std::string& ns, Query query); +/** + * Query handler that plugs in to a SyncClusterConnection and allows query on fastest host + * (if enabled). + * + * Glue code which shields the MultiHostQuery and server parameters from the separate client + * module which knows about neither. + * + * There is a *single* SCCFastQueryHandler for every SCC. Each SCCFastQueryHandler contains + * its own thread pool (lazily initialized) so that at maximum there is a thread-per-SCC-host + * and this thread may have an open connection to the host until it times out. + * If using the "fastestConfigReads" options, clients must be ready for the additional thread + * and connection load when configs are slow. + */ +class SCCFastQueryHandler : public SyncClusterConnection::QueryHandler { +public: + SCCFastQueryHandler(); - virtual std::unique_ptr<DBClientCursor> handleQuery(const std::vector<std::string>& hostStrings, - const std::string &ns, - Query query, - int nToReturn, - int nToSkip, - const BSONObj *fieldsToReturn, - int queryOptions, - int batchSize); + virtual ~SCCFastQueryHandler() {} - private: + virtual bool canHandleQuery(const std::string& ns, Query query); - // The thread pool itself is scoped to the handler and SCC, and lazily creates threads - // per-host as needed. This ensures query starvation cannot occur due to other active - // client threads - though a thread must be created for every client. - HostThreadPools _queryThreads; - }; + virtual std::unique_ptr<DBClientCursor> handleQuery(const std::vector<std::string>& hostStrings, + const std::string& ns, + Query query, + int nToReturn, + int nToSkip, + const BSONObj* fieldsToReturn, + int queryOptions, + int batchSize); +private: + // The thread pool itself is scoped to the handler and SCC, and lazily creates threads + // per-host as needed. This ensures query starvation cannot occur due to other active + // client threads - though a thread must be created for every client. + HostThreadPools _queryThreads; +}; } diff --git a/src/mongo/s/client/shard.cpp b/src/mongo/s/client/shard.cpp index 54626a59321..767c7f7a7cc 100644 --- a/src/mongo/s/client/shard.cpp +++ b/src/mongo/s/client/shard.cpp @@ -48,149 +48,141 @@ namespace mongo { - using std::string; - using std::stringstream; - using std::vector; - - Shard::Shard(const ShardId& id, - const ConnectionString& connStr, - std::unique_ptr<RemoteCommandTargeter> targeter) - : _id(id), - _cs(connStr), - _targeter(std::move(targeter)) { - +using std::string; +using std::stringstream; +using std::vector; + +Shard::Shard(const ShardId& id, + const ConnectionString& connStr, + std::unique_ptr<RemoteCommandTargeter> targeter) + : _id(id), _cs(connStr), _targeter(std::move(targeter)) {} + +Shard::~Shard() = default; + +ShardPtr Shard::lookupRSName(const string& name) { + return grid.shardRegistry()->lookupRSName(name); +} + +BSONObj Shard::runCommand(const std::string& db, const std::string& simple) const { + return runCommand(db, BSON(simple << 1)); +} + +BSONObj Shard::runCommand(const string& db, const BSONObj& cmd) const { + BSONObj res; + bool ok = runCommand(db, cmd, res); + if (!ok) { + stringstream ss; + ss << "runCommand (" << cmd << ") on shard (" << _id << ") failed : " << res; + throw UserException(13136, ss.str()); } - - Shard::~Shard() = default; - - ShardPtr Shard::lookupRSName(const string& name) { - return grid.shardRegistry()->lookupRSName(name); + res = res.getOwned(); + return res; +} + +bool Shard::runCommand(const std::string& db, const std::string& simple, BSONObj& res) const { + return runCommand(db, BSON(simple << 1), res); +} + +bool Shard::runCommand(const string& db, const BSONObj& cmd, BSONObj& res) const { + const ReadPreferenceSetting readPref(ReadPreference::PrimaryOnly, TagSet::primaryOnly()); + auto selectedHost = getTargeter()->findHost(readPref); + if (!selectedHost.isOK()) { + return false; } - BSONObj Shard::runCommand(const std::string& db, const std::string& simple) const { - return runCommand(db, BSON(simple << 1)); - } - - BSONObj Shard::runCommand( const string& db , const BSONObj& cmd ) const { - BSONObj res; - bool ok = runCommand(db, cmd, res); - if ( ! ok ) { - stringstream ss; - ss << "runCommand (" << cmd << ") on shard (" << _id << ") failed : " << res; - throw UserException( 13136 , ss.str() ); - } - res = res.getOwned(); - return res; - } + const RemoteCommandRequest request(selectedHost.getValue(), db, cmd); - bool Shard::runCommand(const std::string& db, const std::string& simple, BSONObj& res) const { - return runCommand(db, BSON(simple << 1), res); + auto statusCommand = grid.shardRegistry()->getCommandRunner()->runCommand(request); + if (!statusCommand.isOK()) { + return false; } - bool Shard::runCommand(const string& db, const BSONObj& cmd, BSONObj& res) const { - const ReadPreferenceSetting readPref(ReadPreference::PrimaryOnly, TagSet::primaryOnly()); - auto selectedHost = getTargeter()->findHost(readPref); - if (!selectedHost.isOK()) { - return false; - } + res = statusCommand.getValue().data.getOwned(); - const RemoteCommandRequest request(selectedHost.getValue(), db, cmd); + return getStatusFromCommandResult(res).isOK(); +} - auto statusCommand = grid.shardRegistry()->getCommandRunner()->runCommand(request); - if (!statusCommand.isOK()) { - return false; - } +ShardStatus Shard::getStatus() const { + BSONObj listDatabases; + uassert(28589, + str::stream() << "call to listDatabases on " << getConnString().toString() + << " failed: " << listDatabases, + runCommand("admin", BSON("listDatabases" << 1), listDatabases)); - res = statusCommand.getValue().data.getOwned(); + BSONElement totalSizeElem = listDatabases["totalSize"]; + uassert(28590, "totalSize field not found in listDatabases", totalSizeElem.isNumber()); - return getStatusFromCommandResult(res).isOK(); - } + BSONObj serverStatus; + uassert(28591, + str::stream() << "call to serverStatus on " << getConnString().toString() + << " failed: " << serverStatus, + runCommand("admin", BSON("serverStatus" << 1), serverStatus)); - ShardStatus Shard::getStatus() const { - BSONObj listDatabases; - uassert(28589, - str::stream() << "call to listDatabases on " << getConnString().toString() - << " failed: " << listDatabases, - runCommand("admin", BSON("listDatabases" << 1), listDatabases)); + BSONElement versionElement = serverStatus["version"]; + uassert(28599, "version field not found in serverStatus", versionElement.type() == String); - BSONElement totalSizeElem = listDatabases["totalSize"]; - uassert(28590, "totalSize field not found in listDatabases", totalSizeElem.isNumber()); + return ShardStatus(totalSizeElem.numberLong(), versionElement.str()); +} - BSONObj serverStatus; - uassert(28591, - str::stream() << "call to serverStatus on " << getConnString().toString() - << " failed: " << serverStatus, - runCommand("admin", BSON("serverStatus" << 1), serverStatus)); +std::string Shard::toString() const { + return _id + ":" + _cs.toString(); +} - BSONElement versionElement = serverStatus["version"]; - uassert(28599, "version field not found in serverStatus", versionElement.type() == String); +void Shard::reloadShardInfo() { + grid.shardRegistry()->reload(); +} - return ShardStatus(totalSizeElem.numberLong(), versionElement.str()); - } +void Shard::removeShard(const ShardId& id) { + grid.shardRegistry()->remove(id); +} - std::string Shard::toString() const { - return _id + ":" + _cs.toString(); - } +ShardPtr Shard::pick() { + vector<ShardId> all; - void Shard::reloadShardInfo() { + grid.shardRegistry()->getAllShardIds(&all); + if (all.size() == 0) { grid.shardRegistry()->reload(); - } - - void Shard::removeShard(const ShardId& id) { - grid.shardRegistry()->remove(id); - } - - ShardPtr Shard::pick() { - vector<ShardId> all; - grid.shardRegistry()->getAllShardIds(&all); - if (all.size() == 0) { - grid.shardRegistry()->reload(); - grid.shardRegistry()->getAllShardIds(&all); - if (all.empty()) { - return nullptr; - } - } - - auto bestShard = grid.shardRegistry()->getShard(all[0]); - if (!bestShard) { + if (all.empty()) { return nullptr; } + } - ShardStatus bestStatus = bestShard->getStatus(); - - for (size_t i = 1; i < all.size(); i++) { - const auto shard = grid.shardRegistry()->getShard(all[i]); - if (!shard) { - continue; - } + auto bestShard = grid.shardRegistry()->getShard(all[0]); + if (!bestShard) { + return nullptr; + } - const ShardStatus status = shard->getStatus(); + ShardStatus bestStatus = bestShard->getStatus(); - if (status < bestStatus) { - bestShard = shard; - bestStatus = status; - } + for (size_t i = 1; i < all.size(); i++) { + const auto shard = grid.shardRegistry()->getShard(all[i]); + if (!shard) { + continue; } - LOG(1) << "best shard for new allocation is " << bestStatus; - return bestShard; + const ShardStatus status = shard->getStatus(); + + if (status < bestStatus) { + bestShard = shard; + bestStatus = status; + } } - ShardStatus::ShardStatus(long long dataSizeBytes, const string& mongoVersion) - : _dataSizeBytes(dataSizeBytes), - _mongoVersion(mongoVersion) { + LOG(1) << "best shard for new allocation is " << bestStatus; + return bestShard; +} - } +ShardStatus::ShardStatus(long long dataSizeBytes, const string& mongoVersion) + : _dataSizeBytes(dataSizeBytes), _mongoVersion(mongoVersion) {} - std::string ShardStatus::toString() const { - return str::stream() << " dataSizeBytes: " << _dataSizeBytes - << " version: " << _mongoVersion; - } +std::string ShardStatus::toString() const { + return str::stream() << " dataSizeBytes: " << _dataSizeBytes << " version: " << _mongoVersion; +} - bool ShardStatus::operator< (const ShardStatus& other) const { - return dataSizeBytes() < other.dataSizeBytes(); - } +bool ShardStatus::operator<(const ShardStatus& other) const { + return dataSizeBytes() < other.dataSizeBytes(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/client/shard.h b/src/mongo/s/client/shard.h index e5140dcee7e..f0d5d7833c2 100644 --- a/src/mongo/s/client/shard.h +++ b/src/mongo/s/client/shard.h @@ -35,98 +35,109 @@ namespace mongo { - class BSONObj; - class RemoteCommandTargeter; +class BSONObj; +class RemoteCommandTargeter; - using ShardId = std::string; +using ShardId = std::string; +/** + * Contains runtime information obtained from the shard. + */ +class ShardStatus { +public: + ShardStatus(long long dataSizeBytes, const std::string& version); + + long long dataSizeBytes() const { + return _dataSizeBytes; + } + const std::string& mongoVersion() const { + return _mongoVersion; + } + + std::string toString() const; + + bool operator<(const ShardStatus& other) const; + +private: + long long _dataSizeBytes; + std::string _mongoVersion; +}; + +class Shard; +using ShardPtr = std::shared_ptr<Shard>; + +/* + * Maintains the targeting and command execution logic for a single shard. Performs polling of + * the shard (if replica set). + */ +class Shard { + MONGO_DISALLOW_COPYING(Shard); + +public: /** - * Contains runtime information obtained from the shard. + * Instantiates a new shard connection management object for the specified shard. */ - class ShardStatus { - public: - ShardStatus(long long dataSizeBytes, const std::string& version); + Shard(const ShardId& id, + const ConnectionString& connStr, + std::unique_ptr<RemoteCommandTargeter> targeter); + + ~Shard(); + + const ShardId& getId() const { + return _id; + } - long long dataSizeBytes() const { return _dataSizeBytes; } - const std::string& mongoVersion() const { return _mongoVersion; } + const ConnectionString& getConnString() const { + return _cs; + } - std::string toString() const; + RemoteCommandTargeter* getTargeter() const { + return _targeter.get(); + } - bool operator< (const ShardStatus& other) const; + BSONObj runCommand(const std::string& db, const std::string& simple) const; + BSONObj runCommand(const std::string& db, const BSONObj& cmd) const; - private: - long long _dataSizeBytes; - std::string _mongoVersion; - }; + bool runCommand(const std::string& db, const std::string& simple, BSONObj& res) const; + bool runCommand(const std::string& db, const BSONObj& cmd, BSONObj& res) const; - class Shard; - using ShardPtr = std::shared_ptr<Shard>; + /** + * Returns metadata and stats for this shard. + */ + ShardStatus getStatus() const; - /* - * Maintains the targeting and command execution logic for a single shard. Performs polling of - * the shard (if replica set). + /** + * Returns a string description of this shard entry. */ - class Shard { - MONGO_DISALLOW_COPYING(Shard); - public: - /** - * Instantiates a new shard connection management object for the specified shard. - */ - Shard(const ShardId& id, - const ConnectionString& connStr, - std::unique_ptr<RemoteCommandTargeter> targeter); - - ~Shard(); - - const ShardId& getId() const { return _id; } - - const ConnectionString& getConnString() const { return _cs; } - - RemoteCommandTargeter* getTargeter() const { return _targeter.get(); } - - BSONObj runCommand(const std::string& db, const std::string& simple) const; - BSONObj runCommand(const std::string& db, const BSONObj& cmd) const; - - bool runCommand(const std::string& db, const std::string& simple, BSONObj& res) const; - bool runCommand(const std::string& db, const BSONObj& cmd, BSONObj& res) const; - - /** - * Returns metadata and stats for this shard. - */ - ShardStatus getStatus() const; - - /** - * Returns a string description of this shard entry. - */ - std::string toString() const; - - static ShardPtr lookupRSName(const std::string& name); - - /** - * @parm current - shard where the chunk/database currently lives in - * @return the currently emptiest shard, if best then current, or nullptr - */ - static ShardPtr pick(); - - static void reloadShardInfo(); - - static void removeShard(const ShardId& id); - - private: - /** - * Identifier of the shard as obtained from the configuration data (i.e. shard0000). - */ - const ShardId _id; - - /** - * Connection string for the shard. - */ - const ConnectionString _cs; - - /** - * Targeter for obtaining hosts from which to read or to which to write. - */ - const std::unique_ptr<RemoteCommandTargeter> _targeter; - }; - -} // namespace mongo + std::string toString() const; + + static ShardPtr lookupRSName(const std::string& name); + + /** + * @parm current - shard where the chunk/database currently lives in + * @return the currently emptiest shard, if best then current, or nullptr + */ + static ShardPtr pick(); + + static void reloadShardInfo(); + + static void removeShard(const ShardId& id); + +private: + /** + * Identifier of the shard as obtained from the configuration data (i.e. shard0000). + */ + const ShardId _id; + + /** + * Connection string for the shard. + */ + const ConnectionString _cs; + + /** + * Targeter for obtaining hosts from which to read or to which to write. + */ + const std::unique_ptr<RemoteCommandTargeter> _targeter; +}; + +} // namespace mongo diff --git a/src/mongo/s/client/shard_connection.cpp b/src/mongo/s/client/shard_connection.cpp index 226bea4fc2c..4607a14baae 100644 --- a/src/mongo/s/client/shard_connection.cpp +++ b/src/mongo/s/client/shard_connection.cpp @@ -49,514 +49,501 @@ namespace mongo { - using std::unique_ptr; - using std::map; - using std::set; - using std::string; - using std::stringstream; - using std::vector; +using std::unique_ptr; +using std::map; +using std::set; +using std::string; +using std::stringstream; +using std::vector; namespace { - class ClientConnections; +class ClientConnections; - /** - * Class which tracks ClientConnections (the client connection pool) for each incoming - * connection, allowing stats access. - */ - class ActiveClientConnections { - public: - void add(const ClientConnections* cc) { - stdx::lock_guard<stdx::mutex> lock(_mutex); - _clientConnections.insert(cc); - } +/** + * Class which tracks ClientConnections (the client connection pool) for each incoming + * connection, allowing stats access. + */ +class ActiveClientConnections { +public: + void add(const ClientConnections* cc) { + stdx::lock_guard<stdx::mutex> lock(_mutex); + _clientConnections.insert(cc); + } - void remove(const ClientConnections* cc) { - stdx::lock_guard<stdx::mutex> lock(_mutex); - _clientConnections.erase(cc); - } + void remove(const ClientConnections* cc) { + stdx::lock_guard<stdx::mutex> lock(_mutex); + _clientConnections.erase(cc); + } - void appendInfo(BSONObjBuilder& b); + void appendInfo(BSONObjBuilder& b); - private: - stdx::mutex _mutex; - set<const ClientConnections*> _clientConnections; +private: + stdx::mutex _mutex; + set<const ClientConnections*> _clientConnections; - } activeClientConnections; +} activeClientConnections; - /** - * Command to allow access to the sharded conn pool information in mongos. - */ - class ShardedPoolStats : public Command { - public: - - ShardedPoolStats() : Command( "shardConnPoolStats" ) { } - virtual void help( stringstream &help ) const { help << "stats about the shard connection pool"; } - virtual bool isWriteCommandForConfigServer() const { return false; } - virtual bool slaveOk() const { return true; } - - // Same privs as connPoolStats - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::connPoolStats); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - - virtual bool run(OperationContext* txn, - const string& dbname, - mongo::BSONObj& cmdObj, - int options, - std::string& errmsg, - mongo::BSONObjBuilder& result) { +/** + * Command to allow access to the sharded conn pool information in mongos. + */ +class ShardedPoolStats : public Command { +public: + ShardedPoolStats() : Command("shardConnPoolStats") {} + virtual void help(stringstream& help) const { + help << "stats about the shard connection pool"; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual bool slaveOk() const { + return true; + } - // Base pool info - shardConnectionPool.appendInfo(result); + // Same privs as connPoolStats + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::connPoolStats); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } - // Thread connection info - activeClientConnections.appendInfo(result); + virtual bool run(OperationContext* txn, + const string& dbname, + mongo::BSONObj& cmdObj, + int options, + std::string& errmsg, + mongo::BSONObjBuilder& result) { + // Base pool info + shardConnectionPool.appendInfo(result); - return true; - } + // Thread connection info + activeClientConnections.appendInfo(result); - } shardedPoolStatsCmd; + return true; + } - /** - * holds all the actual db connections for a client to various servers 1 per thread, so - * doesn't have to be thread safe. - */ - class ClientConnections { - MONGO_DISALLOW_COPYING(ClientConnections); - public: +} shardedPoolStatsCmd; - struct Status { - Status() : created(0), avail(0) { } +/** + * holds all the actual db connections for a client to various servers 1 per thread, so + * doesn't have to be thread safe. + */ +class ClientConnections { + MONGO_DISALLOW_COPYING(ClientConnections); - // May be read concurrently, but only written from - // this thread. - long long created; - DBClientBase* avail; - }; +public: + struct Status { + Status() : created(0), avail(0) {} - // Gets or creates the status object for the host - Status* _getStatus(const string& addr) { - scoped_spinlock lock(_lock); - Status* &temp = _hosts[addr]; - if (!temp) { - temp = new Status(); - } + // May be read concurrently, but only written from + // this thread. + long long created; + DBClientBase* avail; + }; - return temp; + // Gets or creates the status object for the host + Status* _getStatus(const string& addr) { + scoped_spinlock lock(_lock); + Status*& temp = _hosts[addr]; + if (!temp) { + temp = new Status(); } - ClientConnections() { - // Start tracking client connections - activeClientConnections.add(this); - } + return temp; + } - ~ClientConnections() { - // Stop tracking these client connections - activeClientConnections.remove(this); + ClientConnections() { + // Start tracking client connections + activeClientConnections.add(this); + } - releaseAll(true); - } + ~ClientConnections() { + // Stop tracking these client connections + activeClientConnections.remove(this); - void releaseAll(bool fromDestructor = false) { - // Don't need spinlock protection because if not in the destructor, we don't modify - // _hosts, and if in the destructor we are not accessible to external threads. - for (HostMap::iterator i = _hosts.begin(); i != _hosts.end(); ++i) { - const string addr = i->first; - Status* ss = i->second; - invariant(ss); - - if (ss->avail) { - // If we're shutting down, don't want to initiate release mechanism as it is - // slow, and isn't needed since all connections will be closed anyway. - if (inShutdown()) { - if (versionManager.isVersionableCB(ss->avail)) { - versionManager.resetShardVersionCB(ss->avail); - } - - delete ss->avail; - } - else { - release(addr, ss->avail); + releaseAll(true); + } + + void releaseAll(bool fromDestructor = false) { + // Don't need spinlock protection because if not in the destructor, we don't modify + // _hosts, and if in the destructor we are not accessible to external threads. + for (HostMap::iterator i = _hosts.begin(); i != _hosts.end(); ++i) { + const string addr = i->first; + Status* ss = i->second; + invariant(ss); + + if (ss->avail) { + // If we're shutting down, don't want to initiate release mechanism as it is + // slow, and isn't needed since all connections will be closed anyway. + if (inShutdown()) { + if (versionManager.isVersionableCB(ss->avail)) { + versionManager.resetShardVersionCB(ss->avail); } - ss->avail = 0; + delete ss->avail; + } else { + release(addr, ss->avail); } - if (fromDestructor) { - delete ss; - } + ss->avail = 0; } if (fromDestructor) { - _hosts.clear(); + delete ss; } } - DBClientBase * get(const string& addr, const string& ns) { - - { - // We want to report ns stats - scoped_spinlock lock(_lock); - if (ns.size() > 0) - _seenNS.insert(ns); - } + if (fromDestructor) { + _hosts.clear(); + } + } - Status* s = _getStatus(addr); + DBClientBase* get(const string& addr, const string& ns) { + { + // We want to report ns stats + scoped_spinlock lock(_lock); + if (ns.size() > 0) + _seenNS.insert(ns); + } - unique_ptr<DBClientBase> c; - if (s->avail) { - c.reset(s->avail); - s->avail = 0; + Status* s = _getStatus(addr); - // May throw an exception - shardConnectionPool.onHandedOut(c.get()); - } - else { - c.reset(shardConnectionPool.get(addr)); + unique_ptr<DBClientBase> c; + if (s->avail) { + c.reset(s->avail); + s->avail = 0; - // After, so failed creation doesn't get counted - s->created++; - } + // May throw an exception + shardConnectionPool.onHandedOut(c.get()); + } else { + c.reset(shardConnectionPool.get(addr)); - return c.release(); + // After, so failed creation doesn't get counted + s->created++; } - void done( const string& addr , DBClientBase* conn ) { - Status* s = _hosts[addr]; - verify( s ); - - const bool isConnGood = shardConnectionPool.isConnectionGood(addr, conn); + return c.release(); + } - if (s->avail != NULL) { - warning() << "Detected additional sharded connection in the " - << "thread local pool for " << addr; + void done(const string& addr, DBClientBase* conn) { + Status* s = _hosts[addr]; + verify(s); - if (DBException::traceExceptions) { - // There shouldn't be more than one connection checked out to the same - // host on the same thread. - printStackTrace(); - } + const bool isConnGood = shardConnectionPool.isConnectionGood(addr, conn); - if (!isConnGood) { - delete s->avail; - s->avail = NULL; - } + if (s->avail != NULL) { + warning() << "Detected additional sharded connection in the " + << "thread local pool for " << addr; - // Let the internal pool handle the bad connection, this can also - // update the lower bounds for the known good socket creation time - // for this host. - release(addr, conn); - return; + if (DBException::traceExceptions) { + // There shouldn't be more than one connection checked out to the same + // host on the same thread. + printStackTrace(); } if (!isConnGood) { - // Let the internal pool handle the bad connection. - release(addr, conn); - return; + delete s->avail; + s->avail = NULL; } - // Note: Although we try our best to clear bad connections as much as possible, - // some of them can still slip through because of how ClientConnections are being - // used - as thread local variables. This means that threads won't be able to - // see the s->avail connection of other threads. - - s->avail = conn; + // Let the internal pool handle the bad connection, this can also + // update the lower bounds for the known good socket creation time + // for this host. + release(addr, conn); + return; } - void sync() { - for ( HostMap::iterator i=_hosts.begin(); i!=_hosts.end(); ++i ) { - string addr = i->first; - Status* ss = i->second; - if ( ss->avail ) - ss->avail->getLastError(); - } + if (!isConnGood) { + // Let the internal pool handle the bad connection. + release(addr, conn); + return; } - void checkVersions( const string& ns ) { - - vector<ShardId> all; - grid.shardRegistry()->getAllShardIds(&all); + // Note: Although we try our best to clear bad connections as much as possible, + // some of them can still slip through because of how ClientConnections are being + // used - as thread local variables. This means that threads won't be able to + // see the s->avail connection of other threads. - // Don't report exceptions here as errors in GetLastError - LastError::Disabled ignoreForGLE(&LastError::get(cc())); + s->avail = conn; + } - // Now only check top-level shard connections - for (const ShardId& shardId : all) { - try { - const auto shard = grid.shardRegistry()->getShard(shardId); - if (!shard) { - continue; - } + void sync() { + for (HostMap::iterator i = _hosts.begin(); i != _hosts.end(); ++i) { + string addr = i->first; + Status* ss = i->second; + if (ss->avail) + ss->avail->getLastError(); + } + } - string sconnString = shard->getConnString().toString(); - Status* s = _getStatus( sconnString ); + void checkVersions(const string& ns) { + vector<ShardId> all; + grid.shardRegistry()->getAllShardIds(&all); - if( ! s->avail ) { - s->avail = shardConnectionPool.get( sconnString ); - s->created++; // After, so failed creation doesn't get counted - } + // Don't report exceptions here as errors in GetLastError + LastError::Disabled ignoreForGLE(&LastError::get(cc())); - versionManager.checkShardVersionCB( s->avail, ns, false, 1 ); + // Now only check top-level shard connections + for (const ShardId& shardId : all) { + try { + const auto shard = grid.shardRegistry()->getShard(shardId); + if (!shard) { + continue; } - catch ( const DBException& ex ) { - warning() << "problem while initially checking shard versions on" << " " - << shardId << causedBy(ex); + string sconnString = shard->getConnString().toString(); + Status* s = _getStatus(sconnString); - // NOTE: This is only a heuristic, to avoid multiple stale version retries - // across multiple shards, and does not affect correctness. + if (!s->avail) { + s->avail = shardConnectionPool.get(sconnString); + s->created++; // After, so failed creation doesn't get counted } + + versionManager.checkShardVersionCB(s->avail, ns, false, 1); + } catch (const DBException& ex) { + warning() << "problem while initially checking shard versions on" + << " " << shardId << causedBy(ex); + + // NOTE: This is only a heuristic, to avoid multiple stale version retries + // across multiple shards, and does not affect correctness. } } + } - void release( const string& addr , DBClientBase * conn ) { - shardConnectionPool.release( addr , conn ); + void release(const string& addr, DBClientBase* conn) { + shardConnectionPool.release(addr, conn); + } + + /** + * Appends info about the client connection pool to a BOBuilder + * Safe to call with activeClientConnections lock + */ + void appendInfo(BSONObjBuilder& b) const { + scoped_spinlock lock(_lock); + + BSONArrayBuilder hostsArrB(b.subarrayStart("hosts")); + for (HostMap::const_iterator i = _hosts.begin(); i != _hosts.end(); ++i) { + BSONObjBuilder bb(hostsArrB.subobjStart()); + bb.append("host", i->first); + bb.append("created", i->second->created); + bb.appendBool("avail", static_cast<bool>(i->second->avail)); + bb.done(); } + hostsArrB.done(); - /** - * Appends info about the client connection pool to a BOBuilder - * Safe to call with activeClientConnections lock - */ - void appendInfo( BSONObjBuilder& b ) const { - - scoped_spinlock lock( _lock ); - - BSONArrayBuilder hostsArrB( b.subarrayStart( "hosts" ) ); - for ( HostMap::const_iterator i = _hosts.begin(); i != _hosts.end(); ++i ) { - BSONObjBuilder bb( hostsArrB.subobjStart() ); - bb.append( "host", i->first ); - bb.append( "created", i->second->created ); - bb.appendBool( "avail", static_cast<bool>( i->second->avail ) ); - bb.done(); - } - hostsArrB.done(); + BSONArrayBuilder nsArrB(b.subarrayStart("seenNS")); + for (set<string>::const_iterator i = _seenNS.begin(); i != _seenNS.end(); ++i) { + nsArrB.append(*i); + } + nsArrB.done(); + } - BSONArrayBuilder nsArrB( b.subarrayStart( "seenNS" ) ); - for ( set<string>::const_iterator i = _seenNS.begin(); i != _seenNS.end(); ++i ) { - nsArrB.append(*i); + // Protects only the creation of new entries in the _hosts and _seenNS map + // from external threads. Reading _hosts / _seenNS in this thread doesn't + // need protection. + mutable SpinLock _lock; + typedef map<string, Status*, DBConnectionPool::serverNameCompare> HostMap; + HostMap _hosts; + set<string> _seenNS; + + /** + * Clears the connections kept by this pool (ie, not including the global pool) + */ + void clearPool() { + for (HostMap::iterator iter = _hosts.begin(); iter != _hosts.end(); ++iter) { + if (iter->second->avail != NULL) { + delete iter->second->avail; } - nsArrB.done(); + delete iter->second; } - - // Protects only the creation of new entries in the _hosts and _seenNS map - // from external threads. Reading _hosts / _seenNS in this thread doesn't - // need protection. - mutable SpinLock _lock; - typedef map<string,Status*,DBConnectionPool::serverNameCompare> HostMap; - HostMap _hosts; - set<string> _seenNS; - - /** - * Clears the connections kept by this pool (ie, not including the global pool) - */ - void clearPool() { - for(HostMap::iterator iter = _hosts.begin(); iter != _hosts.end(); ++iter) { - if (iter->second->avail != NULL) { - delete iter->second->avail; - } - delete iter->second; - } - _hosts.clear(); - } + _hosts.clear(); + } - void forgetNS( const string& ns ) { - scoped_spinlock lock( _lock ); - _seenNS.erase( ns ); - } + void forgetNS(const string& ns) { + scoped_spinlock lock(_lock); + _seenNS.erase(ns); + } - // ----- + // ----- - static thread_specific_ptr<ClientConnections> _perThread; + static thread_specific_ptr<ClientConnections> _perThread; - static ClientConnections* threadInstance() { - ClientConnections* cc = _perThread.get(); - if ( ! cc ) { - cc = new ClientConnections(); - _perThread.reset( cc ); - } - return cc; + static ClientConnections* threadInstance() { + ClientConnections* cc = _perThread.get(); + if (!cc) { + cc = new ClientConnections(); + _perThread.reset(cc); } - }; + return cc; + } +}; - void ActiveClientConnections::appendInfo(BSONObjBuilder& b) { - BSONArrayBuilder arr(64 * 1024); // There may be quite a few threads +void ActiveClientConnections::appendInfo(BSONObjBuilder& b) { + BSONArrayBuilder arr(64 * 1024); // There may be quite a few threads - { - stdx::lock_guard<stdx::mutex> lock(_mutex); - for (set<const ClientConnections*>::const_iterator i = _clientConnections.begin(); - i != _clientConnections.end(); - ++i) { - - BSONObjBuilder bb(arr.subobjStart()); - (*i)->appendInfo(bb); - bb.done(); - } + { + stdx::lock_guard<stdx::mutex> lock(_mutex); + for (set<const ClientConnections*>::const_iterator i = _clientConnections.begin(); + i != _clientConnections.end(); + ++i) { + BSONObjBuilder bb(arr.subobjStart()); + (*i)->appendInfo(bb); + bb.done(); } - - b.appendArray("threads", arr.obj()); } - thread_specific_ptr<ClientConnections> ClientConnections::_perThread; - -} // namespace + b.appendArray("threads", arr.obj()); +} - // The global connection pool - DBConnectionPool shardConnectionPool; +thread_specific_ptr<ClientConnections> ClientConnections::_perThread; - // Different between mongos and mongod - void usingAShardConnection(const string& addr); +} // namespace +// The global connection pool +DBConnectionPool shardConnectionPool; - ShardConnection::ShardConnection(const ConnectionString& connectionString, - const string& ns, - std::shared_ptr<ChunkManager> manager) - : _cs(connectionString), - _ns(ns), - _manager(manager) { +// Different between mongos and mongod +void usingAShardConnection(const string& addr); - _init(); - } - ShardConnection::~ShardConnection() { - if (_conn) { - if (_conn->isFailed()) { - if (_conn->getSockCreationMicroSec() == DBClientBase::INVALID_SOCK_CREATION_TIME) { - kill(); - } - else { - // The pool takes care of deleting the failed connection - this - // will also trigger disposal of older connections in the pool - done(); - } - } - else { - // see done() comments above for why we log this line - log() << "sharded connection to " << _conn->getServerAddress() - << " not being returned to the pool"; +ShardConnection::ShardConnection(const ConnectionString& connectionString, + const string& ns, + std::shared_ptr<ChunkManager> manager) + : _cs(connectionString), _ns(ns), _manager(manager) { + _init(); +} +ShardConnection::~ShardConnection() { + if (_conn) { + if (_conn->isFailed()) { + if (_conn->getSockCreationMicroSec() == DBClientBase::INVALID_SOCK_CREATION_TIME) { kill(); + } else { + // The pool takes care of deleting the failed connection - this + // will also trigger disposal of older connections in the pool + done(); } + } else { + // see done() comments above for why we log this line + log() << "sharded connection to " << _conn->getServerAddress() + << " not being returned to the pool"; + + kill(); } } +} + +void ShardConnection::_init() { + invariant(_cs.isValid()); + _conn = ClientConnections::threadInstance()->get(_cs.toString(), _ns); + _finishedInit = false; + usingAShardConnection(_cs.toString()); +} - void ShardConnection::_init() { - invariant(_cs.isValid()); - _conn = ClientConnections::threadInstance()->get(_cs.toString(), _ns); - _finishedInit = false; - usingAShardConnection(_cs.toString()); +void ShardConnection::_finishInit() { + if (_finishedInit) + return; + _finishedInit = true; + + if (versionManager.isVersionableCB(_conn)) { + // Make sure we specified a manager for the correct namespace + if (_ns.size() && _manager) + verify(_manager->getns() == _ns); + _setVersion = versionManager.checkShardVersionCB(this, false, 1); + } else { + // Make sure we didn't specify a manager for a non-versionable connection (i.e. config) + verify(!_manager); + _setVersion = false; } +} - void ShardConnection::_finishInit() { - if ( _finishedInit ) - return; +void ShardConnection::done() { + if (_conn) { + ClientConnections::threadInstance()->done(_cs.toString(), _conn); + _conn = 0; _finishedInit = true; + } +} +void ShardConnection::kill() { + if (_conn) { if (versionManager.isVersionableCB(_conn)) { - // Make sure we specified a manager for the correct namespace - if (_ns.size() && _manager) - verify(_manager->getns() == _ns); - _setVersion = versionManager.checkShardVersionCB( this , false , 1 ); - } - else { - // Make sure we didn't specify a manager for a non-versionable connection (i.e. config) - verify(!_manager); - _setVersion = false; + versionManager.resetShardVersionCB(_conn); } - } - void ShardConnection::done() { - if ( _conn ) { + if (_conn->isFailed()) { + // Let the pool know about the bad connection and also delegate disposal to it. ClientConnections::threadInstance()->done(_cs.toString(), _conn); - _conn = 0; - _finishedInit = true; + } else { + delete _conn; } + + _conn = 0; + _finishedInit = true; } +} - void ShardConnection::kill() { - if ( _conn ) { - if (versionManager.isVersionableCB(_conn)) { - versionManager.resetShardVersionCB(_conn); - } +void ShardConnection::sync() { + ClientConnections::threadInstance()->sync(); +} - if (_conn->isFailed()) { - // Let the pool know about the bad connection and also delegate disposal to it. - ClientConnections::threadInstance()->done(_cs.toString(), _conn); - } - else { - delete _conn; - } +void ShardConnection::checkMyConnectionVersions(const string& ns) { + ClientConnections::threadInstance()->checkVersions(ns); +} - _conn = 0; - _finishedInit = true; - } - } +void ShardConnection::releaseMyConnections() { + ClientConnections::threadInstance()->releaseAll(); +} - void ShardConnection::sync() { - ClientConnections::threadInstance()->sync(); - } +void ShardConnection::clearPool() { + shardConnectionPool.clear(); + ClientConnections::threadInstance()->clearPool(); +} - void ShardConnection::checkMyConnectionVersions( const string & ns ) { - ClientConnections::threadInstance()->checkVersions( ns ); - } +void ShardConnection::forgetNS(const string& ns) { + ClientConnections::threadInstance()->forgetNS(ns); +} - void ShardConnection::releaseMyConnections() { - ClientConnections::threadInstance()->releaseAll(); - } - void ShardConnection::clearPool() { - shardConnectionPool.clear(); - ClientConnections::threadInstance()->clearPool(); +bool setShardVersion(DBClientBase& conn, + const string& ns, + const string& configServerPrimary, + ChunkVersion version, + ChunkManager* manager, + bool authoritative, + BSONObj& result) { + BSONObjBuilder cmdBuilder; + cmdBuilder.append("setShardVersion", ns); + cmdBuilder.append("configdb", configServerPrimary); + + ShardId shardId; + { + const auto shard = grid.shardRegistry()->getShard(conn.getServerAddress()); + shardId = shard->getId(); + cmdBuilder.append("shard", shardId); + cmdBuilder.append("shardHost", shard->getConnString().toString()); } - void ShardConnection::forgetNS( const string& ns ) { - ClientConnections::threadInstance()->forgetNS( ns ); + if (ns.size() > 0) { + version.addToBSON(cmdBuilder); + } else { + cmdBuilder.append("init", true); } + if (authoritative) { + cmdBuilder.appendBool("authoritative", 1); + } - bool setShardVersion(DBClientBase& conn, - const string& ns, - const string& configServerPrimary, - ChunkVersion version, - ChunkManager* manager, - bool authoritative, - BSONObj& result) { - - BSONObjBuilder cmdBuilder; - cmdBuilder.append("setShardVersion", ns); - cmdBuilder.append("configdb", configServerPrimary); - - ShardId shardId; - { - const auto shard = grid.shardRegistry()->getShard(conn.getServerAddress()); - shardId = shard->getId(); - cmdBuilder.append("shard", shardId); - cmdBuilder.append("shardHost", shard->getConnString().toString()); - } - - if (ns.size() > 0) { - version.addToBSON(cmdBuilder); - } - else { - cmdBuilder.append("init", true); - } - - if (authoritative) { - cmdBuilder.appendBool("authoritative", 1); - } - - BSONObj cmd = cmdBuilder.obj(); + BSONObj cmd = cmdBuilder.obj(); - LOG(1) << " setShardVersion " << shardId << " " << conn.getServerAddress() - << " " << ns << " " << cmd - << (manager ? string(str::stream() << " " << manager->getSequenceNumber()) : ""); + LOG(1) << " setShardVersion " << shardId << " " << conn.getServerAddress() << " " << ns + << " " << cmd + << (manager ? string(str::stream() << " " << manager->getSequenceNumber()) : ""); - return conn.runCommand("admin", cmd, result, 0); - } + return conn.runCommand("admin", cmd, result, 0); +} } diff --git a/src/mongo/s/client/shard_connection.h b/src/mongo/s/client/shard_connection.h index 4fc7f0f01a8..42040a3ab79 100644 --- a/src/mongo/s/client/shard_connection.h +++ b/src/mongo/s/client/shard_connection.h @@ -35,124 +35,126 @@ namespace mongo { - class ChunkManager; - typedef std::shared_ptr<ChunkManager> ChunkManagerPtr; - - - class ShardConnection : public AScopedConnection { - public: - ShardConnection(const ConnectionString& connectionString, - const std::string& ns, - std::shared_ptr<ChunkManager> manager = nullptr); - - ~ShardConnection(); - - void done(); - void kill(); - - DBClientBase& conn() { - _finishInit(); - verify( _conn ); - return *_conn; - } - - DBClientBase* operator->() { - _finishInit(); - verify( _conn ); - return _conn; - } - - DBClientBase* get() { - _finishInit(); - verify( _conn ); - return _conn; - } - - /** - * @return the connection object underneath without setting the shard version. - * @throws AssertionException if _conn is uninitialized. - */ - DBClientBase* getRawConn() const { - verify( _conn ); - return _conn; - } - - std::string getHost() const { - return _cs.toString(); - } - - std::string getNS() const { - return _ns; - } - - std::shared_ptr<ChunkManager> getManager() const { - return _manager; - } - - bool setVersion() { - _finishInit(); - return _setVersion; - } - - static void sync(); - - void donotCheckVersion() { - _setVersion = false; - _finishedInit = true; - } - - bool ok() const { return _conn != NULL; } - - /** checks all of my thread local connections for the version of this ns */ - static void checkMyConnectionVersions( const std::string & ns ); - - /** - * Returns all the current sharded connections to the pool. - * Note: This is *dangerous* if we have GLE state. - */ - static void releaseMyConnections(); - - /** - * Clears all connections in the sharded pool, including connections in the - * thread local storage pool of the current thread. - */ - static void clearPool(); - - /** - * Forgets a namespace to prevent future versioning. - */ - static void forgetNS( const std::string& ns ); - - private: - void _init(); - void _finishInit(); - - const ConnectionString _cs; - const std::string _ns; - - std::shared_ptr<ChunkManager> _manager; - - bool _finishedInit; - - DBClientBase* _conn; - bool _setVersion; - }; +class ChunkManager; +typedef std::shared_ptr<ChunkManager> ChunkManagerPtr; +class ShardConnection : public AScopedConnection { +public: + ShardConnection(const ConnectionString& connectionString, + const std::string& ns, + std::shared_ptr<ChunkManager> manager = nullptr); + + ~ShardConnection(); + + void done(); + void kill(); + + DBClientBase& conn() { + _finishInit(); + verify(_conn); + return *_conn; + } + + DBClientBase* operator->() { + _finishInit(); + verify(_conn); + return _conn; + } + + DBClientBase* get() { + _finishInit(); + verify(_conn); + return _conn; + } + + /** + * @return the connection object underneath without setting the shard version. + * @throws AssertionException if _conn is uninitialized. + */ + DBClientBase* getRawConn() const { + verify(_conn); + return _conn; + } + + std::string getHost() const { + return _cs.toString(); + } + + std::string getNS() const { + return _ns; + } + + std::shared_ptr<ChunkManager> getManager() const { + return _manager; + } + + bool setVersion() { + _finishInit(); + return _setVersion; + } + + static void sync(); + + void donotCheckVersion() { + _setVersion = false; + _finishedInit = true; + } + + bool ok() const { + return _conn != NULL; + } + + /** checks all of my thread local connections for the version of this ns */ + static void checkMyConnectionVersions(const std::string& ns); + + /** + * Returns all the current sharded connections to the pool. + * Note: This is *dangerous* if we have GLE state. + */ + static void releaseMyConnections(); + /** - * Sends the setShardVersion command on the specified connection. + * Clears all connections in the sharded pool, including connections in the + * thread local storage pool of the current thread. */ - bool setShardVersion(DBClientBase & conn, - const std::string& ns, - const std::string& configServerPrimary, - ChunkVersion version, - ChunkManager* manager, - bool authoritative, - BSONObj& result); + static void clearPool(); + + /** + * Forgets a namespace to prevent future versioning. + */ + static void forgetNS(const std::string& ns); + +private: + void _init(); + void _finishInit(); + + const ConnectionString _cs; + const std::string _ns; + + std::shared_ptr<ChunkManager> _manager; + + bool _finishedInit; + + DBClientBase* _conn; + bool _setVersion; +}; + + +/** + * Sends the setShardVersion command on the specified connection. + */ +bool setShardVersion(DBClientBase& conn, + const std::string& ns, + const std::string& configServerPrimary, + ChunkVersion version, + ChunkManager* manager, + bool authoritative, + BSONObj& result); - typedef std::shared_ptr<ShardConnection> ShardConnectionPtr; +typedef std::shared_ptr<ShardConnection> ShardConnectionPtr; - extern DBConnectionPool shardConnectionPool; +extern DBConnectionPool shardConnectionPool; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/client/shard_connection_test.cpp b/src/mongo/s/client/shard_connection_test.cpp index 1751dd1c84e..5f14a3872f7 100644 --- a/src/mongo/s/client/shard_connection_test.cpp +++ b/src/mongo/s/client/shard_connection_test.cpp @@ -52,263 +52,261 @@ namespace mongo { namespace { - using std::string; - using std::vector; +using std::string; +using std::vector; - const string TARGET_HOST = "$dummy:27017"; +const string TARGET_HOST = "$dummy:27017"; - /** - * Warning: cannot run in parallel - */ - class ShardConnFixture: public mongo::unittest::Test { - public: - void setUp() { - if (!haveClient()) { - Client::initThread("ShardConnFixture", getGlobalServiceContext(), NULL); - } - _maxPoolSizePerHost = mongo::shardConnectionPool.getMaxPoolSize(); - - mongo::ConnectionString::setConnectionHook( - mongo::MockConnRegistry::get()->getConnStrHook()); - _dummyServer = new MockRemoteDBServer(TARGET_HOST); - mongo::MockConnRegistry::get()->addServer(_dummyServer); +/** + * Warning: cannot run in parallel + */ +class ShardConnFixture : public mongo::unittest::Test { +public: + void setUp() { + if (!haveClient()) { + Client::initThread("ShardConnFixture", getGlobalServiceContext(), NULL); } + _maxPoolSizePerHost = mongo::shardConnectionPool.getMaxPoolSize(); - void tearDown() { - ShardConnection::clearPool(); + mongo::ConnectionString::setConnectionHook( + mongo::MockConnRegistry::get()->getConnStrHook()); + _dummyServer = new MockRemoteDBServer(TARGET_HOST); + mongo::MockConnRegistry::get()->addServer(_dummyServer); + } - mongo::MockConnRegistry::get()->removeServer(_dummyServer->getServerAddress()); - delete _dummyServer; + void tearDown() { + ShardConnection::clearPool(); - mongo::shardConnectionPool.setMaxPoolSize(_maxPoolSizePerHost); - } + mongo::MockConnRegistry::get()->removeServer(_dummyServer->getServerAddress()); + delete _dummyServer; - void killServer() { - _dummyServer->shutdown(); - } + mongo::shardConnectionPool.setMaxPoolSize(_maxPoolSizePerHost); + } - void restartServer() { - _dummyServer->reboot(); - } + void killServer() { + _dummyServer->shutdown(); + } - protected: - static void assertGreaterThan(uint64_t a, uint64_t b) { - ASSERT_GREATER_THAN(a, b); - } + void restartServer() { + _dummyServer->reboot(); + } - static void assertNotEqual(uint64_t a, uint64_t b) { - ASSERT_NOT_EQUALS(a, b); - } +protected: + static void assertGreaterThan(uint64_t a, uint64_t b) { + ASSERT_GREATER_THAN(a, b); + } - /** - * Tries to grab a series of connections from the pool, perform checks on - * them, then put them back into the pool. After that, it checks these - * connections can be retrieved again from the pool. - * - * @param checkFunc method for comparing new connections and arg2. - * @param arg2 the value to pass as the 2nd parameter of checkFunc. - * @param newConnsToCreate the number of new connections to make. - */ - void checkNewConns(void (*checkFunc)(uint64_t, uint64_t), uint64_t arg2, - size_t newConnsToCreate) { - vector<ShardConnection*> newConnList; - for (size_t x = 0; x < newConnsToCreate; x++) { - ShardConnection* newConn = new ShardConnection(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - checkFunc(newConn->get()->getSockCreationMicroSec(), arg2); - newConnList.push_back(newConn); - } - - const uint64_t oldCreationTime = mongo::curTimeMicros64(); - - for (vector<ShardConnection*>::iterator iter = newConnList.begin(); - iter != newConnList.end(); ++iter) { - (*iter)->done(); - delete *iter; - } - - newConnList.clear(); - - // Check that connections created after the purge was put back to the pool. - for (size_t x = 0; x < newConnsToCreate; x++) { - ShardConnection* newConn = new ShardConnection(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ASSERT_LESS_THAN(newConn->get()->getSockCreationMicroSec(), oldCreationTime); - newConnList.push_back(newConn); - } - - for (vector<ShardConnection*>::iterator iter = newConnList.begin(); - iter != newConnList.end(); ++iter) { - (*iter)->done(); - delete *iter; - } + static void assertNotEqual(uint64_t a, uint64_t b) { + ASSERT_NOT_EQUALS(a, b); + } + + /** + * Tries to grab a series of connections from the pool, perform checks on + * them, then put them back into the pool. After that, it checks these + * connections can be retrieved again from the pool. + * + * @param checkFunc method for comparing new connections and arg2. + * @param arg2 the value to pass as the 2nd parameter of checkFunc. + * @param newConnsToCreate the number of new connections to make. + */ + void checkNewConns(void (*checkFunc)(uint64_t, uint64_t), + uint64_t arg2, + size_t newConnsToCreate) { + vector<ShardConnection*> newConnList; + for (size_t x = 0; x < newConnsToCreate; x++) { + ShardConnection* newConn = + new ShardConnection(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + checkFunc(newConn->get()->getSockCreationMicroSec(), arg2); + newConnList.push_back(newConn); } - private: - MockRemoteDBServer* _dummyServer; - uint32_t _maxPoolSizePerHost; - }; + const uint64_t oldCreationTime = mongo::curTimeMicros64(); - TEST_F(ShardConnFixture, BasicShardConnection) { - ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + for (vector<ShardConnection*>::iterator iter = newConnList.begin(); + iter != newConnList.end(); + ++iter) { + (*iter)->done(); + delete *iter; + } - DBClientBase* conn1Ptr = conn1.get(); - conn1.done(); + newConnList.clear(); - ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ASSERT_EQUALS(conn1Ptr, conn3.get()); + // Check that connections created after the purge was put back to the pool. + for (size_t x = 0; x < newConnsToCreate; x++) { + ShardConnection* newConn = + new ShardConnection(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ASSERT_LESS_THAN(newConn->get()->getSockCreationMicroSec(), oldCreationTime); + newConnList.push_back(newConn); + } - conn2.done(); - conn3.done(); + for (vector<ShardConnection*>::iterator iter = newConnList.begin(); + iter != newConnList.end(); + ++iter) { + (*iter)->done(); + delete *iter; + } } - TEST_F(ShardConnFixture, InvalidateBadConnInPool) { - ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); +private: + MockRemoteDBServer* _dummyServer; + uint32_t _maxPoolSizePerHost; +}; - conn1.done(); - conn3.done(); +TEST_F(ShardConnFixture, BasicShardConnection) { + ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - const uint64_t badCreationTime = mongo::curTimeMicros64(); - killServer(); + DBClientBase* conn1Ptr = conn1.get(); + conn1.done(); - try { - conn2.get()->query("test.user", mongo::Query()); - } - catch (const mongo::SocketException&) { - } + ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ASSERT_EQUALS(conn1Ptr, conn3.get()); - conn2.done(); + conn2.done(); + conn3.done(); +} - restartServer(); - checkNewConns(assertGreaterThan, badCreationTime, 10); - } +TEST_F(ShardConnFixture, InvalidateBadConnInPool) { + ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - TEST_F(ShardConnFixture, DontReturnKnownBadConnToPool) { - ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + conn1.done(); + conn3.done(); - conn1.done(); - killServer(); + const uint64_t badCreationTime = mongo::curTimeMicros64(); + killServer(); - try { - conn3.get()->query("test.user", mongo::Query()); - } - catch (const mongo::SocketException&) { - } + try { + conn2.get()->query("test.user", mongo::Query()); + } catch (const mongo::SocketException&) { + } + + conn2.done(); - restartServer(); + restartServer(); + checkNewConns(assertGreaterThan, badCreationTime, 10); +} - const uint64_t badCreationTime = conn3.get()->getSockCreationMicroSec(); - conn3.done(); - // attempting to put a 'bad' connection back to the pool - conn2.done(); +TEST_F(ShardConnFixture, DontReturnKnownBadConnToPool) { + ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - checkNewConns(assertGreaterThan, badCreationTime, 10); + conn1.done(); + killServer(); + + try { + conn3.get()->query("test.user", mongo::Query()); + } catch (const mongo::SocketException&) { } - TEST_F(ShardConnFixture, BadConnClearsPoolWhenKilled) { - ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + restartServer(); - conn1.done(); - killServer(); + const uint64_t badCreationTime = conn3.get()->getSockCreationMicroSec(); + conn3.done(); + // attempting to put a 'bad' connection back to the pool + conn2.done(); - try { - conn3.get()->query("test.user", mongo::Query()); - } - catch (const mongo::SocketException&) { - } + checkNewConns(assertGreaterThan, badCreationTime, 10); +} - restartServer(); +TEST_F(ShardConnFixture, BadConnClearsPoolWhenKilled) { + ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - const uint64_t badCreationTime = conn3.get()->getSockCreationMicroSec(); - conn3.kill(); - // attempting to put a 'bad' connection back to the pool - conn2.done(); + conn1.done(); + killServer(); - checkNewConns(assertGreaterThan, badCreationTime, 10); + try { + conn3.get()->query("test.user", mongo::Query()); + } catch (const mongo::SocketException&) { } - TEST_F(ShardConnFixture, KilledGoodConnShouldNotClearPool) { - ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + restartServer(); - const uint64_t upperBoundCreationTime = - conn3.get()->getSockCreationMicroSec(); - conn3.done(); + const uint64_t badCreationTime = conn3.get()->getSockCreationMicroSec(); + conn3.kill(); + // attempting to put a 'bad' connection back to the pool + conn2.done(); - const uint64_t badCreationTime = conn1.get()->getSockCreationMicroSec(); - conn1.kill(); + checkNewConns(assertGreaterThan, badCreationTime, 10); +} - conn2.done(); +TEST_F(ShardConnFixture, KilledGoodConnShouldNotClearPool) { + ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn4(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn5(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + const uint64_t upperBoundCreationTime = conn3.get()->getSockCreationMicroSec(); + conn3.done(); - ASSERT_GREATER_THAN(conn4.get()->getSockCreationMicroSec(), badCreationTime); - ASSERT_LESS_THAN_OR_EQUALS(conn4.get()->getSockCreationMicroSec(), - upperBoundCreationTime); + const uint64_t badCreationTime = conn1.get()->getSockCreationMicroSec(); + conn1.kill(); - ASSERT_GREATER_THAN(conn5.get()->getSockCreationMicroSec(), badCreationTime); - ASSERT_LESS_THAN_OR_EQUALS(conn5.get()->getSockCreationMicroSec(), - upperBoundCreationTime); + conn2.done(); - checkNewConns(assertGreaterThan, upperBoundCreationTime, 10); - } + ShardConnection conn4(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn5(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - TEST_F(ShardConnFixture, InvalidateBadConnEvenWhenPoolIsFull) { - mongo::shardConnectionPool.setMaxPoolSize(2); + ASSERT_GREATER_THAN(conn4.get()->getSockCreationMicroSec(), badCreationTime); + ASSERT_LESS_THAN_OR_EQUALS(conn4.get()->getSockCreationMicroSec(), upperBoundCreationTime); - ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ASSERT_GREATER_THAN(conn5.get()->getSockCreationMicroSec(), badCreationTime); + ASSERT_LESS_THAN_OR_EQUALS(conn5.get()->getSockCreationMicroSec(), upperBoundCreationTime); - conn1.done(); - conn3.done(); + checkNewConns(assertGreaterThan, upperBoundCreationTime, 10); +} - const uint64_t badCreationTime = mongo::curTimeMicros64(); - killServer(); +TEST_F(ShardConnFixture, InvalidateBadConnEvenWhenPoolIsFull) { + mongo::shardConnectionPool.setMaxPoolSize(2); - try { - conn2.get()->query("test.user", mongo::Query()); - } - catch (const mongo::SocketException&) { - } + ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ShardConnection conn3(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + + conn1.done(); + conn3.done(); - conn2.done(); + const uint64_t badCreationTime = mongo::curTimeMicros64(); + killServer(); - restartServer(); - checkNewConns(assertGreaterThan, badCreationTime, 2); + try { + conn2.get()->query("test.user", mongo::Query()); + } catch (const mongo::SocketException&) { } - TEST_F(ShardConnFixture, DontReturnConnGoneBadToPool) { - ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - const uint64_t conn1CreationTime = conn1.get()->getSockCreationMicroSec(); + conn2.done(); - uint64_t conn2CreationTime = 0; + restartServer(); + checkNewConns(assertGreaterThan, badCreationTime, 2); +} - { - ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - conn2CreationTime = conn2.get()->getSockCreationMicroSec(); +TEST_F(ShardConnFixture, DontReturnConnGoneBadToPool) { + ShardConnection conn1(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + const uint64_t conn1CreationTime = conn1.get()->getSockCreationMicroSec(); - conn1.done(); - // conn2 gets out of scope without calling done() - } + uint64_t conn2CreationTime = 0; - // conn2 should not have been put back into the pool but it should - // also not invalidate older connections since it didn't encounter - // a socket exception. - - ShardConnection conn1Again(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); - ASSERT_EQUALS(conn1CreationTime, conn1Again.get()->getSockCreationMicroSec()); + { + ShardConnection conn2(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + conn2CreationTime = conn2.get()->getSockCreationMicroSec(); - checkNewConns(assertNotEqual, conn2CreationTime, 10); - conn1Again.done(); + conn1.done(); + // conn2 gets out of scope without calling done() } -} // namespace -} // namespace mongo + // conn2 should not have been put back into the pool but it should + // also not invalidate older connections since it didn't encounter + // a socket exception. + + ShardConnection conn1Again(ConnectionString(HostAndPort(TARGET_HOST)), "test.user"); + ASSERT_EQUALS(conn1CreationTime, conn1Again.get()->getSockCreationMicroSec()); + + checkNewConns(assertNotEqual, conn2CreationTime, 10); + conn1Again.done(); +} + +} // namespace +} // namespace mongo diff --git a/src/mongo/s/client/shard_registry.cpp b/src/mongo/s/client/shard_registry.cpp index 01efc45aaa4..ad33f35c661 100644 --- a/src/mongo/s/client/shard_registry.cpp +++ b/src/mongo/s/client/shard_registry.cpp @@ -48,281 +48,271 @@ namespace mongo { - using std::shared_ptr; - using std::string; - using std::unique_ptr; - using std::vector; +using std::shared_ptr; +using std::string; +using std::unique_ptr; +using std::vector; - using executor::TaskExecutor; - using RemoteCommandCallbackArgs = TaskExecutor::RemoteCommandCallbackArgs; +using executor::TaskExecutor; +using RemoteCommandCallbackArgs = TaskExecutor::RemoteCommandCallbackArgs; namespace { - const Seconds kConfigCommandTimeout{30}; -} // unnamed namespace - - ShardRegistry::ShardRegistry(std::unique_ptr<RemoteCommandTargeterFactory> targeterFactory, - std::unique_ptr<RemoteCommandRunner> commandRunner, - std::unique_ptr<executor::TaskExecutor> executor, - CatalogManager* catalogManager) - : _targeterFactory(std::move(targeterFactory)), - _commandRunner(std::move(commandRunner)), - _executor(std::move(executor)), - _catalogManager(catalogManager) { - - // add config shard registry entry so know it's always there - std::lock_guard<std::mutex> lk(_mutex); - _addConfigShard_inlock(); +const Seconds kConfigCommandTimeout{30}; +} // unnamed namespace - } +ShardRegistry::ShardRegistry(std::unique_ptr<RemoteCommandTargeterFactory> targeterFactory, + std::unique_ptr<RemoteCommandRunner> commandRunner, + std::unique_ptr<executor::TaskExecutor> executor, + CatalogManager* catalogManager) + : _targeterFactory(std::move(targeterFactory)), + _commandRunner(std::move(commandRunner)), + _executor(std::move(executor)), + _catalogManager(catalogManager) { + // add config shard registry entry so know it's always there + std::lock_guard<std::mutex> lk(_mutex); + _addConfigShard_inlock(); +} - ShardRegistry::~ShardRegistry() = default; +ShardRegistry::~ShardRegistry() = default; - void ShardRegistry::reload() { - vector<ShardType> shards; - Status status = _catalogManager->getAllShards(&shards); - massert(13632, "couldn't get updated shard list from config server", status.isOK()); +void ShardRegistry::reload() { + vector<ShardType> shards; + Status status = _catalogManager->getAllShards(&shards); + massert(13632, "couldn't get updated shard list from config server", status.isOK()); - int numShards = shards.size(); + int numShards = shards.size(); - LOG(1) << "found " << numShards << " shards listed on config server(s)"; + LOG(1) << "found " << numShards << " shards listed on config server(s)"; - std::lock_guard<std::mutex> lk(_mutex); + std::lock_guard<std::mutex> lk(_mutex); - _lookup.clear(); - _rsLookup.clear(); + _lookup.clear(); + _rsLookup.clear(); - _addConfigShard_inlock(); + _addConfigShard_inlock(); - for (const ShardType& shardType : shards) { - uassertStatusOK(shardType.validate()); + for (const ShardType& shardType : shards) { + uassertStatusOK(shardType.validate()); - // Skip the config host even if there is one left over from legacy installations. The - // config host is installed manually from the catalog manager data. - if (shardType.getName() == "config") { - continue; - } - - _addShard_inlock(shardType); + // Skip the config host even if there is one left over from legacy installations. The + // config host is installed manually from the catalog manager data. + if (shardType.getName() == "config") { + continue; } + + _addShard_inlock(shardType); } +} - shared_ptr<Shard> ShardRegistry::getShard(const ShardId& shardId) { - shared_ptr<Shard> shard = _findUsingLookUp(shardId); - if (shard) { - return shard; - } +shared_ptr<Shard> ShardRegistry::getShard(const ShardId& shardId) { + shared_ptr<Shard> shard = _findUsingLookUp(shardId); + if (shard) { + return shard; + } - // If we can't find the shard, we might just need to reload the cache - reload(); + // If we can't find the shard, we might just need to reload the cache + reload(); - return _findUsingLookUp(shardId); - } + return _findUsingLookUp(shardId); +} - shared_ptr<Shard> ShardRegistry::lookupRSName(const string& name) const { - std::lock_guard<std::mutex> lk(_mutex); - ShardMap::const_iterator i = _rsLookup.find(name); +shared_ptr<Shard> ShardRegistry::lookupRSName(const string& name) const { + std::lock_guard<std::mutex> lk(_mutex); + ShardMap::const_iterator i = _rsLookup.find(name); - return (i == _rsLookup.end()) ? nullptr : i->second; - } + return (i == _rsLookup.end()) ? nullptr : i->second; +} - void ShardRegistry::remove(const ShardId& id) { - std::lock_guard<std::mutex> lk(_mutex); +void ShardRegistry::remove(const ShardId& id) { + std::lock_guard<std::mutex> lk(_mutex); - for (ShardMap::iterator i = _lookup.begin(); i != _lookup.end();) { - shared_ptr<Shard> s = i->second; - if (s->getId() == id) { - _lookup.erase(i++); - } - else { - ++i; - } + for (ShardMap::iterator i = _lookup.begin(); i != _lookup.end();) { + shared_ptr<Shard> s = i->second; + if (s->getId() == id) { + _lookup.erase(i++); + } else { + ++i; } + } - for (ShardMap::iterator i = _rsLookup.begin(); i != _rsLookup.end();) { - shared_ptr<Shard> s = i->second; - if (s->getId() == id) { - _rsLookup.erase(i++); - } - else { - ++i; - } + for (ShardMap::iterator i = _rsLookup.begin(); i != _rsLookup.end();) { + shared_ptr<Shard> s = i->second; + if (s->getId() == id) { + _rsLookup.erase(i++); + } else { + ++i; } } +} - void ShardRegistry::getAllShardIds(vector<ShardId>* all) const { - std::set<string> seen; - - { - std::lock_guard<std::mutex> lk(_mutex); - for (ShardMap::const_iterator i = _lookup.begin(); i != _lookup.end(); ++i) { - const shared_ptr<Shard>& s = i->second; - if (s->getId() == "config") { - continue; - } - - if (seen.count(s->getId())) { - continue; - } - seen.insert(s->getId()); +void ShardRegistry::getAllShardIds(vector<ShardId>* all) const { + std::set<string> seen; + + { + std::lock_guard<std::mutex> lk(_mutex); + for (ShardMap::const_iterator i = _lookup.begin(); i != _lookup.end(); ++i) { + const shared_ptr<Shard>& s = i->second; + if (s->getId() == "config") { + continue; } - } - all->assign(seen.begin(), seen.end()); + if (seen.count(s->getId())) { + continue; + } + seen.insert(s->getId()); + } } - void ShardRegistry::toBSON(BSONObjBuilder* result) { - BSONObjBuilder b(_lookup.size() + 50); + all->assign(seen.begin(), seen.end()); +} - std::lock_guard<std::mutex> lk(_mutex); +void ShardRegistry::toBSON(BSONObjBuilder* result) { + BSONObjBuilder b(_lookup.size() + 50); - for (ShardMap::const_iterator i = _lookup.begin(); i != _lookup.end(); ++i) { - b.append(i->first, i->second->getConnString().toString()); - } + std::lock_guard<std::mutex> lk(_mutex); - result->append("map", b.obj()); + for (ShardMap::const_iterator i = _lookup.begin(); i != _lookup.end(); ++i) { + b.append(i->first, i->second->getConnString().toString()); } - void ShardRegistry::_addConfigShard_inlock() { - ShardType configServerShard; - configServerShard.setName("config"); - configServerShard.setHost(_catalogManager->connectionString().toString()); - _addShard_inlock(configServerShard); + result->append("map", b.obj()); +} + +void ShardRegistry::_addConfigShard_inlock() { + ShardType configServerShard; + configServerShard.setName("config"); + configServerShard.setHost(_catalogManager->connectionString().toString()); + _addShard_inlock(configServerShard); +} + +void ShardRegistry::_addShard_inlock(const ShardType& shardType) { + // This validation should ideally go inside the ShardType::validate call. However, doing + // it there would prevent us from loading previously faulty shard hosts, which might have + // been stored (i.e., the entire getAllShards call would fail). + auto shardHostStatus = ConnectionString::parse(shardType.getHost()); + if (!shardHostStatus.isOK()) { + warning() << "Unable to parse shard host " << shardHostStatus.getStatus().toString(); } - void ShardRegistry::_addShard_inlock(const ShardType& shardType) { - // This validation should ideally go inside the ShardType::validate call. However, doing - // it there would prevent us from loading previously faulty shard hosts, which might have - // been stored (i.e., the entire getAllShards call would fail). - auto shardHostStatus = ConnectionString::parse(shardType.getHost()); - if (!shardHostStatus.isOK()) { - warning() << "Unable to parse shard host " - << shardHostStatus.getStatus().toString(); - } - - const ConnectionString& shardHost(shardHostStatus.getValue()); - - // Sync cluster connections (legacy config server) do not go through the normal targeting - // mechanism and must only be reachable through CatalogManagerLegacy or legacy-style - // queries and inserts. Do not create targeter for these connections. This code should go - // away after 3.2 is released. - if (shardHost.type() == ConnectionString::SYNC) { - _lookup[shardType.getName()] = - std::make_shared<Shard>(shardType.getName(), shardHost, nullptr); - return; - } + const ConnectionString& shardHost(shardHostStatus.getValue()); - // Non-SYNC shards - shared_ptr<Shard> shard = - std::make_shared<Shard>(shardType.getName(), - shardHost, - std::move(_targeterFactory->create(shardHost))); + // Sync cluster connections (legacy config server) do not go through the normal targeting + // mechanism and must only be reachable through CatalogManagerLegacy or legacy-style + // queries and inserts. Do not create targeter for these connections. This code should go + // away after 3.2 is released. + if (shardHost.type() == ConnectionString::SYNC) { + _lookup[shardType.getName()] = + std::make_shared<Shard>(shardType.getName(), shardHost, nullptr); + return; + } - _lookup[shardType.getName()] = shard; + // Non-SYNC shards + shared_ptr<Shard> shard = std::make_shared<Shard>( + shardType.getName(), shardHost, std::move(_targeterFactory->create(shardHost))); - // TODO: The only reason to have the shard host names in the lookup table is for the - // setShardVersion call, which resolves the shard id from the shard address. This is - // error-prone and will go away eventually when we switch all communications to go through - // the remote command runner. - _lookup[shardType.getHost()] = shard; + _lookup[shardType.getName()] = shard; - for (const HostAndPort& hostAndPort : shardHost.getServers()) { - _lookup[hostAndPort.toString()] = shard; + // TODO: The only reason to have the shard host names in the lookup table is for the + // setShardVersion call, which resolves the shard id from the shard address. This is + // error-prone and will go away eventually when we switch all communications to go through + // the remote command runner. + _lookup[shardType.getHost()] = shard; - // Maintain a mapping from host to shard it belongs to for the case where we need to - // update the shard connection string on reconfigurations. - if (shardHost.type() == ConnectionString::SET) { - _rsLookup[hostAndPort.toString()] = shard; - } - } + for (const HostAndPort& hostAndPort : shardHost.getServers()) { + _lookup[hostAndPort.toString()] = shard; + // Maintain a mapping from host to shard it belongs to for the case where we need to + // update the shard connection string on reconfigurations. if (shardHost.type() == ConnectionString::SET) { - _rsLookup[shardHost.getSetName()] = shard; + _rsLookup[hostAndPort.toString()] = shard; } } - shared_ptr<Shard> ShardRegistry::_findUsingLookUp(const ShardId& shardId) { - std::lock_guard<std::mutex> lk(_mutex); - ShardMap::iterator it = _lookup.find(shardId); - if (it != _lookup.end()) { - return it->second; - } - - return nullptr; + if (shardHost.type() == ConnectionString::SET) { + _rsLookup[shardHost.getSetName()] = shard; } +} - StatusWith<std::vector<BSONObj>> ShardRegistry::exhaustiveFind(const HostAndPort& host, - const NamespaceString& nss, - const BSONObj& query, - boost::optional<int> limit) { - // If for some reason the callback never gets invoked, we will return this status - Status status = Status(ErrorCodes::InternalError, "Internal error running find command"); - vector<BSONObj> results; - - auto fetcherCallback = [&status, &results](const Fetcher::QueryResponseStatus& dataStatus, - Fetcher::NextAction* nextAction) { - - // Throw out any accumulated results on error - if (!dataStatus.isOK()) { - status = dataStatus.getStatus(); - results.clear(); - return; - } +shared_ptr<Shard> ShardRegistry::_findUsingLookUp(const ShardId& shardId) { + std::lock_guard<std::mutex> lk(_mutex); + ShardMap::iterator it = _lookup.find(shardId); + if (it != _lookup.end()) { + return it->second; + } - auto& data = dataStatus.getValue(); - for (const BSONObj& doc : data.documents) { - results.push_back(std::move(doc.getOwned())); - } + return nullptr; +} - status = Status::OK(); - }; +StatusWith<std::vector<BSONObj>> ShardRegistry::exhaustiveFind(const HostAndPort& host, + const NamespaceString& nss, + const BSONObj& query, + boost::optional<int> limit) { + // If for some reason the callback never gets invoked, we will return this status + Status status = Status(ErrorCodes::InternalError, "Internal error running find command"); + vector<BSONObj> results; - unique_ptr<LiteParsedQuery> findCmd( - fassertStatusOK(28688, LiteParsedQuery::makeAsFindCmd(nss, query, std::move(limit)))); + auto fetcherCallback = [&status, &results](const Fetcher::QueryResponseStatus& dataStatus, + Fetcher::NextAction* nextAction) { - QueryFetcher fetcher(_executor.get(), - host, - nss, - findCmd->asFindCommand(), - fetcherCallback); + // Throw out any accumulated results on error + if (!dataStatus.isOK()) { + status = dataStatus.getStatus(); + results.clear(); + return; + } - Status scheduleStatus = fetcher.schedule(); - if (!scheduleStatus.isOK()) { - return scheduleStatus; + auto& data = dataStatus.getValue(); + for (const BSONObj& doc : data.documents) { + results.push_back(std::move(doc.getOwned())); } - fetcher.wait(); + status = Status::OK(); + }; - if (!status.isOK()) { - return status; - } + unique_ptr<LiteParsedQuery> findCmd( + fassertStatusOK(28688, LiteParsedQuery::makeAsFindCmd(nss, query, std::move(limit)))); + + QueryFetcher fetcher(_executor.get(), host, nss, findCmd->asFindCommand(), fetcherCallback); - return results; + Status scheduleStatus = fetcher.schedule(); + if (!scheduleStatus.isOK()) { + return scheduleStatus; } - StatusWith<BSONObj> ShardRegistry::runCommand(const HostAndPort& host, - const std::string& dbName, - const BSONObj& cmdObj) { - StatusWith<RemoteCommandResponse> responseStatus = - Status(ErrorCodes::InternalError, "Internal error running command"); + fetcher.wait(); - RemoteCommandRequest request(host, dbName, cmdObj, kConfigCommandTimeout); - auto callStatus = _executor->scheduleRemoteCommand(request, - [&responseStatus](const RemoteCommandCallbackArgs& args) { - responseStatus = args.response; - }); + if (!status.isOK()) { + return status; + } - if (!callStatus.isOK()) { - return callStatus.getStatus(); - } + return results; +} - // Block until the command is carried out - _executor->wait(callStatus.getValue()); +StatusWith<BSONObj> ShardRegistry::runCommand(const HostAndPort& host, + const std::string& dbName, + const BSONObj& cmdObj) { + StatusWith<RemoteCommandResponse> responseStatus = + Status(ErrorCodes::InternalError, "Internal error running command"); - if (!responseStatus.isOK()) { - return responseStatus.getStatus(); - } + RemoteCommandRequest request(host, dbName, cmdObj, kConfigCommandTimeout); + auto callStatus = + _executor->scheduleRemoteCommand(request, + [&responseStatus](const RemoteCommandCallbackArgs& args) { + responseStatus = args.response; + }); + + if (!callStatus.isOK()) { + return callStatus.getStatus(); + } + + // Block until the command is carried out + _executor->wait(callStatus.getValue()); - return responseStatus.getValue().data; + if (!responseStatus.isOK()) { + return responseStatus.getStatus(); } -} // namespace mongo + return responseStatus.getValue().data; +} + +} // namespace mongo diff --git a/src/mongo/s/client/shard_registry.h b/src/mongo/s/client/shard_registry.h index 57f6526d723..ce9898b23f2 100644 --- a/src/mongo/s/client/shard_registry.h +++ b/src/mongo/s/client/shard_registry.h @@ -38,127 +38,132 @@ namespace mongo { - class BSONObjBuilder; - class CatalogManager; - struct HostAndPort; - class NamespaceString; - class RemoteCommandRunner; - class RemoteCommandTargeterFactory; - class Shard; - class ShardType; - - template<typename T> class StatusWith; +class BSONObjBuilder; +class CatalogManager; +struct HostAndPort; +class NamespaceString; +class RemoteCommandRunner; +class RemoteCommandTargeterFactory; +class Shard; +class ShardType; + +template <typename T> +class StatusWith; namespace executor { - class TaskExecutor; +class TaskExecutor; + +} // namespace executor + +/** + * Maintains the set of all shards known to the instance and their connections. Manages polling + * the respective replica sets for membership changes. + */ +class ShardRegistry { +public: + /** + * Instantiates a new shard registry. + * + * @param targeterFactory Produces targeters for each shard's individual connection string + * @param commandRunner Command runner for executing commands against hosts + * @param executor Asynchronous task executor to use for making calls to shards. + * @param catalogManager Used to retrieve the list of registered shard. TODO: remove. + */ + ShardRegistry(std::unique_ptr<RemoteCommandTargeterFactory> targeterFactory, + std::unique_ptr<RemoteCommandRunner> commandRunner, + std::unique_ptr<executor::TaskExecutor> executor, + CatalogManager* catalogManager); + + ~ShardRegistry(); + + RemoteCommandRunner* getCommandRunner() const { + return _commandRunner.get(); + } + + executor::TaskExecutor* getExecutor() const { + return _executor.get(); + } -} // namespace executor + void reload(); /** - * Maintains the set of all shards known to the instance and their connections. Manages polling - * the respective replica sets for membership changes. + * Returns shared pointer to shard object with given shard id. */ - class ShardRegistry { - public: - /** - * Instantiates a new shard registry. - * - * @param targeterFactory Produces targeters for each shard's individual connection string - * @param commandRunner Command runner for executing commands against hosts - * @param executor Asynchronous task executor to use for making calls to shards. - * @param catalogManager Used to retrieve the list of registered shard. TODO: remove. - */ - ShardRegistry(std::unique_ptr<RemoteCommandTargeterFactory> targeterFactory, - std::unique_ptr<RemoteCommandRunner> commandRunner, - std::unique_ptr<executor::TaskExecutor> executor, - CatalogManager* catalogManager); - - ~ShardRegistry(); - - RemoteCommandRunner* getCommandRunner() const { return _commandRunner.get(); } - - executor::TaskExecutor* getExecutor() const { return _executor.get(); } - - void reload(); - - /** - * Returns shared pointer to shard object with given shard id. - */ - std::shared_ptr<Shard> getShard(const ShardId& shardId); - - /** - * Lookup shard by replica set name. Returns nullptr if the name can't be found. - * Note: this doesn't refresh the table if the name isn't found, so it's possible that a - * newly added shard/Replica Set may not be found. - */ - std::shared_ptr<Shard> lookupRSName(const std::string& name) const; - - void remove(const ShardId& id); - - void getAllShardIds(std::vector<ShardId>* all) const; - - void toBSON(BSONObjBuilder* result); - - /** - * Executes 'find' command against the specified host and fetches *all* the results that - * the host will return until there are no more or until an error is returned. - * - * Returns either the complete set of results or an error, never partial results. - * - * Note: should never be used outside of CatalogManagerReplicaSet or DistLockCatalogImpl. - */ - StatusWith<std::vector<BSONObj>> exhaustiveFind(const HostAndPort& host, - const NamespaceString& nss, - const BSONObj& query, - boost::optional<int> limit); - - /** - * Runs a command against the specified host and returns the result. - */ - StatusWith<BSONObj> runCommand(const HostAndPort& host, - const std::string& dbName, - const BSONObj& cmdObj); - - private: - typedef std::map<ShardId, std::shared_ptr<Shard>> ShardMap; - - /** - * Creates a shard based on the specified information and puts it into the lookup maps. - */ - void _addShard_inlock(const ShardType& shardType); - - /** - * Adds the "config" shard (representing the config server) to the shard registry. - */ - void _addConfigShard_inlock(); - - std::shared_ptr<Shard> _findUsingLookUp(const ShardId& shardId); - - // Factory to obtain remote command targeters for shards - const std::unique_ptr<RemoteCommandTargeterFactory> _targeterFactory; - - // API to run remote commands to shards in a synchronous manner - const std::unique_ptr<RemoteCommandRunner> _commandRunner; - - // Executor for scheduling work and remote commands to shards that run in an asynchronous - // manner. - const std::unique_ptr<executor::TaskExecutor> _executor; - - // Catalog manager from which to load the shard information. Not owned and must outlive - // the shard registry object. - CatalogManager* const _catalogManager; - - // Protects the maps below - mutable std::mutex _mutex; - - // Map of both shardName -> Shard and hostName -> Shard - ShardMap _lookup; - - // TODO: These should eventually disappear and become parts of Shard - - // Map from all hosts within a replica set to the shard representing this replica set - ShardMap _rsLookup; - }; - -} // namespace mongo + std::shared_ptr<Shard> getShard(const ShardId& shardId); + + /** + * Lookup shard by replica set name. Returns nullptr if the name can't be found. + * Note: this doesn't refresh the table if the name isn't found, so it's possible that a + * newly added shard/Replica Set may not be found. + */ + std::shared_ptr<Shard> lookupRSName(const std::string& name) const; + + void remove(const ShardId& id); + + void getAllShardIds(std::vector<ShardId>* all) const; + + void toBSON(BSONObjBuilder* result); + + /** + * Executes 'find' command against the specified host and fetches *all* the results that + * the host will return until there are no more or until an error is returned. + * + * Returns either the complete set of results or an error, never partial results. + * + * Note: should never be used outside of CatalogManagerReplicaSet or DistLockCatalogImpl. + */ + StatusWith<std::vector<BSONObj>> exhaustiveFind(const HostAndPort& host, + const NamespaceString& nss, + const BSONObj& query, + boost::optional<int> limit); + + /** + * Runs a command against the specified host and returns the result. + */ + StatusWith<BSONObj> runCommand(const HostAndPort& host, + const std::string& dbName, + const BSONObj& cmdObj); + +private: + typedef std::map<ShardId, std::shared_ptr<Shard>> ShardMap; + + /** + * Creates a shard based on the specified information and puts it into the lookup maps. + */ + void _addShard_inlock(const ShardType& shardType); + + /** + * Adds the "config" shard (representing the config server) to the shard registry. + */ + void _addConfigShard_inlock(); + + std::shared_ptr<Shard> _findUsingLookUp(const ShardId& shardId); + + // Factory to obtain remote command targeters for shards + const std::unique_ptr<RemoteCommandTargeterFactory> _targeterFactory; + + // API to run remote commands to shards in a synchronous manner + const std::unique_ptr<RemoteCommandRunner> _commandRunner; + + // Executor for scheduling work and remote commands to shards that run in an asynchronous + // manner. + const std::unique_ptr<executor::TaskExecutor> _executor; + + // Catalog manager from which to load the shard information. Not owned and must outlive + // the shard registry object. + CatalogManager* const _catalogManager; + + // Protects the maps below + mutable std::mutex _mutex; + + // Map of both shardName -> Shard and hostName -> Shard + ShardMap _lookup; + + // TODO: These should eventually disappear and become parts of Shard + + // Map from all hosts within a replica set to the shard representing this replica set + ShardMap _rsLookup; +}; + +} // namespace mongo diff --git a/src/mongo/s/client/sharding_connection_hook.cpp b/src/mongo/s/client/sharding_connection_hook.cpp index b1567441d83..80e924e1121 100644 --- a/src/mongo/s/client/sharding_connection_hook.cpp +++ b/src/mongo/s/client/sharding_connection_hook.cpp @@ -47,101 +47,97 @@ namespace mongo { - using std::string; +using std::string; namespace { - bool initWireVersion(DBClientBase* conn, std::string* errMsg) { - BSONObj response; - if (!conn->runCommand("admin", BSON("isMaster" << 1), response)) { - *errMsg = str::stream() << "Failed to determine wire version " - << "for internal connection: " << response; - return false; - } - - if (response.hasField("minWireVersion") && response.hasField("maxWireVersion")) { - int minWireVersion = response["minWireVersion"].numberInt(); - int maxWireVersion = response["maxWireVersion"].numberInt(); - conn->setWireVersions(minWireVersion, maxWireVersion); - } - - return true; +bool initWireVersion(DBClientBase* conn, std::string* errMsg) { + BSONObj response; + if (!conn->runCommand("admin", BSON("isMaster" << 1), response)) { + *errMsg = str::stream() << "Failed to determine wire version " + << "for internal connection: " << response; + return false; } -} // namespace + if (response.hasField("minWireVersion") && response.hasField("maxWireVersion")) { + int minWireVersion = response["minWireVersion"].numberInt(); + int maxWireVersion = response["maxWireVersion"].numberInt(); + conn->setWireVersions(minWireVersion, maxWireVersion); + } + return true; +} - ShardingConnectionHook::ShardingConnectionHook(bool shardedConnections) - : _shardedConnections(shardedConnections) { +} // namespace - } - void ShardingConnectionHook::onCreate(DBClientBase * conn) { +ShardingConnectionHook::ShardingConnectionHook(bool shardedConnections) + : _shardedConnections(shardedConnections) {} - // Authenticate as the first thing we do - // NOTE: Replica set authentication allows authentication against *any* online host - if (getGlobalAuthorizationManager()->isAuthEnabled()) { - LOG(2) << "calling onCreate auth for " << conn->toString(); +void ShardingConnectionHook::onCreate(DBClientBase* conn) { + // Authenticate as the first thing we do + // NOTE: Replica set authentication allows authentication against *any* online host + if (getGlobalAuthorizationManager()->isAuthEnabled()) { + LOG(2) << "calling onCreate auth for " << conn->toString(); - bool result = authenticateInternalUser(conn); + bool result = authenticateInternalUser(conn); - uassert(15847, str::stream() << "can't authenticate to server " - << conn->getServerAddress(), + uassert(15847, + str::stream() << "can't authenticate to server " << conn->getServerAddress(), result); - } - - // Initialize the wire version of single connections - if (conn->type() == ConnectionString::MASTER) { + } - LOG(2) << "checking wire version of new connection " << conn->toString(); + // Initialize the wire version of single connections + if (conn->type() == ConnectionString::MASTER) { + LOG(2) << "checking wire version of new connection " << conn->toString(); - // Initialize the wire protocol version of the connection to find out if we - // can send write commands to this connection. - string errMsg; - if (!initWireVersion(conn, &errMsg)) { - uasserted(17363, errMsg); - } + // Initialize the wire protocol version of the connection to find out if we + // can send write commands to this connection. + string errMsg; + if (!initWireVersion(conn, &errMsg)) { + uasserted(17363, errMsg); } + } - if (_shardedConnections) { - // For every DBClient created by mongos, add a hook that will capture the response from - // commands we pass along from the client, so that we can target the correct node when - // subsequent getLastError calls are made by mongos. - conn->setReplyMetadataReader([](const BSONObj& metadataObj, StringData hostString) + if (_shardedConnections) { + // For every DBClient created by mongos, add a hook that will capture the response from + // commands we pass along from the client, so that we can target the correct node when + // subsequent getLastError calls are made by mongos. + conn->setReplyMetadataReader([](const BSONObj& metadataObj, StringData hostString) -> Status { - saveGLEStats(metadataObj, hostString); - return Status::OK(); - }); - } - - // For every DBClient created by mongos, add a hook that will append impersonated users - // to the end of every runCommand. mongod uses this information to produce auditing - // records attributed to the proper authenticated user(s). - conn->setRequestMetadataWriter([](BSONObjBuilder* metadataBob) -> Status { - audit::writeImpersonatedUsersToMetadata(metadataBob); - return Status::OK(); - }); - - // For every SCC created, add a hook that will allow fastest-config-first config reads if - // the appropriate server options are set. - if (conn->type() == ConnectionString::SYNC) { - SyncClusterConnection* scc = dynamic_cast<SyncClusterConnection*>(conn); - if (scc) { - scc->attachQueryHandler(new SCCFastQueryHandler); - } - } + saveGLEStats(metadataObj, hostString); + return Status::OK(); + }); } - void ShardingConnectionHook::onDestroy(DBClientBase * conn) { - if (_shardedConnections && versionManager.isVersionableCB(conn)){ - versionManager.resetShardVersionCB(conn); + // For every DBClient created by mongos, add a hook that will append impersonated users + // to the end of every runCommand. mongod uses this information to produce auditing + // records attributed to the proper authenticated user(s). + conn->setRequestMetadataWriter([](BSONObjBuilder* metadataBob) -> Status { + audit::writeImpersonatedUsersToMetadata(metadataBob); + return Status::OK(); + }); + + // For every SCC created, add a hook that will allow fastest-config-first config reads if + // the appropriate server options are set. + if (conn->type() == ConnectionString::SYNC) { + SyncClusterConnection* scc = dynamic_cast<SyncClusterConnection*>(conn); + if (scc) { + scc->attachQueryHandler(new SCCFastQueryHandler); } } +} - void ShardingConnectionHook::onRelease(DBClientBase* conn) { - // This is currently for making the replica set connections release - // secondary connections to the pool. - conn->reset(); +void ShardingConnectionHook::onDestroy(DBClientBase* conn) { + if (_shardedConnections && versionManager.isVersionableCB(conn)) { + versionManager.resetShardVersionCB(conn); } +} + +void ShardingConnectionHook::onRelease(DBClientBase* conn) { + // This is currently for making the replica set connections release + // secondary connections to the pool. + conn->reset(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/client/sharding_connection_hook.h b/src/mongo/s/client/sharding_connection_hook.h index 661bb42f34c..a131f526472 100644 --- a/src/mongo/s/client/sharding_connection_hook.h +++ b/src/mongo/s/client/sharding_connection_hook.h @@ -32,22 +32,22 @@ namespace mongo { - class DBClientBase; +class DBClientBase; - /** - * Intercepts creation of sharded connections and transparently performs the internal - * authentication on them. - */ - class ShardingConnectionHook : public DBConnectionHook { - public: - ShardingConnectionHook(bool shardedConnections); +/** + * Intercepts creation of sharded connections and transparently performs the internal + * authentication on them. + */ +class ShardingConnectionHook : public DBConnectionHook { +public: + ShardingConnectionHook(bool shardedConnections); - virtual void onCreate(DBClientBase* conn); - virtual void onDestroy(DBClientBase* conn); - virtual void onRelease(DBClientBase* conn); + virtual void onCreate(DBClientBase* conn); + virtual void onDestroy(DBClientBase* conn); + virtual void onRelease(DBClientBase* conn); - private: - bool _shardedConnections; - }; +private: + bool _shardedConnections; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/cluster_explain.cpp b/src/mongo/s/cluster_explain.cpp index 1859b9b9a94..1c62c90e8fd 100644 --- a/src/mongo/s/cluster_explain.cpp +++ b/src/mongo/s/cluster_explain.cpp @@ -35,323 +35,311 @@ namespace mongo { - using std::vector; - - const char* ClusterExplain::kSingleShard = "SINGLE_SHARD"; - const char* ClusterExplain::kMergeFromShards = "SHARD_MERGE"; - const char* ClusterExplain::kMergeSortFromShards = "SHARD_MERGE_SORT"; - const char* ClusterExplain::kWriteOnShards = "SHARD_WRITE"; - - namespace { - - // - // BSON size limit management: these functions conditionally append to a - // BSON object or BSON array buidlder, depending on whether or not the - // maximum user size for a BSON object will be exceeded. - // - - bool appendIfRoom(BSONObjBuilder* bob, - const BSONObj& toAppend, - StringData fieldName) { - if ((bob->len() + toAppend.objsize()) < BSONObjMaxUserSize) { - bob->append(fieldName, toAppend); - return true; - } +using std::vector; - // Unless 'bob' has already exceeded the max BSON user size, add a warning indicating - // that data has been truncated. - if (bob->len() < BSONObjMaxUserSize) { - bob->append("warning", - "output truncated due to nearing BSON max user size"); - } +const char* ClusterExplain::kSingleShard = "SINGLE_SHARD"; +const char* ClusterExplain::kMergeFromShards = "SHARD_MERGE"; +const char* ClusterExplain::kMergeSortFromShards = "SHARD_MERGE_SORT"; +const char* ClusterExplain::kWriteOnShards = "SHARD_WRITE"; - return false; - } +namespace { - bool appendToArrayIfRoom(BSONArrayBuilder* arrayBuilder, - const BSONElement& toAppend) { - if ((arrayBuilder->len() + toAppend.size()) < BSONObjMaxUserSize) { - arrayBuilder->append(toAppend); - return true; - } +// +// BSON size limit management: these functions conditionally append to a +// BSON object or BSON array buidlder, depending on whether or not the +// maximum user size for a BSON object will be exceeded. +// - // Unless 'arrayBuilder' has already exceeded the max BSON user size, add a warning - // indicating that data has been truncated. - if (arrayBuilder->len() < BSONObjMaxUserSize) { - arrayBuilder->append(BSON("warning" << - "output truncated due to nearing BSON max user size")); - } +bool appendIfRoom(BSONObjBuilder* bob, const BSONObj& toAppend, StringData fieldName) { + if ((bob->len() + toAppend.objsize()) < BSONObjMaxUserSize) { + bob->append(fieldName, toAppend); + return true; + } - return false; - } + // Unless 'bob' has already exceeded the max BSON user size, add a warning indicating + // that data has been truncated. + if (bob->len() < BSONObjMaxUserSize) { + bob->append("warning", "output truncated due to nearing BSON max user size"); + } - bool appendElementsIfRoom(BSONObjBuilder* bob, - const BSONObj& toAppend) { - if ((bob->len() + toAppend.objsize()) < BSONObjMaxUserSize) { - bob->appendElements(toAppend); - return true; - } + return false; +} - // Unless 'bob' has already exceeded the max BSON user size, add a warning indicating - // that data has been truncated. - if (bob->len() < BSONObjMaxUserSize) { - bob->append("warning", - "output truncated due to nearing BSON max user size"); - } +bool appendToArrayIfRoom(BSONArrayBuilder* arrayBuilder, const BSONElement& toAppend) { + if ((arrayBuilder->len() + toAppend.size()) < BSONObjMaxUserSize) { + arrayBuilder->append(toAppend); + return true; + } - return false; - } + // Unless 'arrayBuilder' has already exceeded the max BSON user size, add a warning + // indicating that data has been truncated. + if (arrayBuilder->len() < BSONObjMaxUserSize) { + arrayBuilder->append(BSON("warning" + << "output truncated due to nearing BSON max user size")); + } - } // namespace + return false; +} - // static - void ClusterExplain::wrapAsExplain(const BSONObj& cmdObj, - ExplainCommon::Verbosity verbosity, - BSONObjBuilder* out) { - out->append("explain", cmdObj); - out->append("verbosity", ExplainCommon::verbosityString(verbosity)); +bool appendElementsIfRoom(BSONObjBuilder* bob, const BSONObj& toAppend) { + if ((bob->len() + toAppend.objsize()) < BSONObjMaxUserSize) { + bob->appendElements(toAppend); + return true; + } - // If the command has a readPreference, then pull it up to the top level. - if (cmdObj.hasField("$readPreference")) { - out->append("$queryOptions", cmdObj["$readPreference"].wrap()); - } + // Unless 'bob' has already exceeded the max BSON user size, add a warning indicating + // that data has been truncated. + if (bob->len() < BSONObjMaxUserSize) { + bob->append("warning", "output truncated due to nearing BSON max user size"); } - // static - Status ClusterExplain::validateShardResults( - const vector<Strategy::CommandResult>& shardResults) { - // Error we didn't get results from any shards. - if (shardResults.empty()) { - return Status(ErrorCodes::InternalError, "no shards found for explain"); - } + return false; +} - // Count up the number of shards that have execution stats and all plans - // execution level information. - size_t numShardsExecStats = 0; - size_t numShardsAllPlansStats = 0; - - // Check that the result from each shard has a true value for "ok" and has - // the expected "queryPlanner" field. - for (size_t i = 0; i < shardResults.size(); i++) { - if (!shardResults[i].result["ok"].trueValue()) { - // Try to pass up the error code from the shard. - ErrorCodes::Error error = ErrorCodes::OperationFailed; - if (shardResults[i].result["code"].isNumber()) { - error = ErrorCodes::fromInt(shardResults[i].result["code"].numberInt()); - } - - return Status(error, str::stream() - << "Explain command on shard " << shardResults[i].target.toString() - << " failed, caused by: " << shardResults[i].result); - } +} // namespace - if (Object != shardResults[i].result["queryPlanner"].type()) { - return Status(ErrorCodes::OperationFailed, str::stream() - << "Explain command on shard " << shardResults[i].target.toString() - << " failed, caused by: " << shardResults[i].result); - } +// static +void ClusterExplain::wrapAsExplain(const BSONObj& cmdObj, + ExplainCommon::Verbosity verbosity, + BSONObjBuilder* out) { + out->append("explain", cmdObj); + out->append("verbosity", ExplainCommon::verbosityString(verbosity)); + + // If the command has a readPreference, then pull it up to the top level. + if (cmdObj.hasField("$readPreference")) { + out->append("$queryOptions", cmdObj["$readPreference"].wrap()); + } +} + +// static +Status ClusterExplain::validateShardResults(const vector<Strategy::CommandResult>& shardResults) { + // Error we didn't get results from any shards. + if (shardResults.empty()) { + return Status(ErrorCodes::InternalError, "no shards found for explain"); + } - if (shardResults[i].result.hasField("executionStats")) { - numShardsExecStats++; - BSONObj execStats = shardResults[i].result["executionStats"].Obj(); - if (execStats.hasField("allPlansExecution")) { - numShardsAllPlansStats++; - } + // Count up the number of shards that have execution stats and all plans + // execution level information. + size_t numShardsExecStats = 0; + size_t numShardsAllPlansStats = 0; + + // Check that the result from each shard has a true value for "ok" and has + // the expected "queryPlanner" field. + for (size_t i = 0; i < shardResults.size(); i++) { + if (!shardResults[i].result["ok"].trueValue()) { + // Try to pass up the error code from the shard. + ErrorCodes::Error error = ErrorCodes::OperationFailed; + if (shardResults[i].result["code"].isNumber()) { + error = ErrorCodes::fromInt(shardResults[i].result["code"].numberInt()); } + + return Status(error, + str::stream() << "Explain command on shard " + << shardResults[i].target.toString() + << " failed, caused by: " << shardResults[i].result); } - // Either all shards should have execution stats info, or none should. - if (0 != numShardsExecStats && shardResults.size() != numShardsExecStats) { - return Status(ErrorCodes::InternalError, - str::stream() << "Only " << numShardsExecStats - << "/" << shardResults.size() - << " had executionStats explain information."); + if (Object != shardResults[i].result["queryPlanner"].type()) { + return Status(ErrorCodes::OperationFailed, + str::stream() << "Explain command on shard " + << shardResults[i].target.toString() + << " failed, caused by: " << shardResults[i].result); } - // Either all shards should have all plans execution stats, or none should. - if (0 != numShardsAllPlansStats && shardResults.size() != numShardsAllPlansStats) { - return Status(ErrorCodes::InternalError, - str::stream() << "Only " << numShardsAllPlansStats - << "/" << shardResults.size() - << " had allPlansExecution explain information."); + if (shardResults[i].result.hasField("executionStats")) { + numShardsExecStats++; + BSONObj execStats = shardResults[i].result["executionStats"].Obj(); + if (execStats.hasField("allPlansExecution")) { + numShardsAllPlansStats++; + } } + } - return Status::OK(); + // Either all shards should have execution stats info, or none should. + if (0 != numShardsExecStats && shardResults.size() != numShardsExecStats) { + return Status(ErrorCodes::InternalError, + str::stream() << "Only " << numShardsExecStats << "/" << shardResults.size() + << " had executionStats explain information."); } - // static - const char* ClusterExplain::getStageNameForReadOp( - const vector<Strategy::CommandResult>& shardResults, - const BSONObj& explainObj) { - if (shardResults.size() == 1) { - return kSingleShard; - } - else if (explainObj.hasField("sort")) { - return kMergeSortFromShards; - } - else { - return kMergeFromShards; - } + // Either all shards should have all plans execution stats, or none should. + if (0 != numShardsAllPlansStats && shardResults.size() != numShardsAllPlansStats) { + return Status(ErrorCodes::InternalError, + str::stream() << "Only " << numShardsAllPlansStats << "/" + << shardResults.size() + << " had allPlansExecution explain information."); } - // static - void ClusterExplain::buildPlannerInfo(const vector<Strategy::CommandResult>& shardResults, - const char* mongosStageName, - BSONObjBuilder* out) { - BSONObjBuilder queryPlannerBob(out->subobjStart("queryPlanner")); + return Status::OK(); +} + +// static +const char* ClusterExplain::getStageNameForReadOp( + const vector<Strategy::CommandResult>& shardResults, const BSONObj& explainObj) { + if (shardResults.size() == 1) { + return kSingleShard; + } else if (explainObj.hasField("sort")) { + return kMergeSortFromShards; + } else { + return kMergeFromShards; + } +} - queryPlannerBob.appendNumber("mongosPlannerVersion", 1); - BSONObjBuilder winningPlanBob(queryPlannerBob.subobjStart("winningPlan")); +// static +void ClusterExplain::buildPlannerInfo(const vector<Strategy::CommandResult>& shardResults, + const char* mongosStageName, + BSONObjBuilder* out) { + BSONObjBuilder queryPlannerBob(out->subobjStart("queryPlanner")); - winningPlanBob.append("stage", mongosStageName); - BSONArrayBuilder shardsBuilder(winningPlanBob.subarrayStart("shards")); - for (size_t i = 0; i < shardResults.size(); i++) { - BSONObjBuilder singleShardBob(shardsBuilder.subobjStart()); + queryPlannerBob.appendNumber("mongosPlannerVersion", 1); + BSONObjBuilder winningPlanBob(queryPlannerBob.subobjStart("winningPlan")); - BSONObj queryPlanner = shardResults[i].result["queryPlanner"].Obj(); - BSONObj serverInfo = shardResults[i].result["serverInfo"].Obj(); + winningPlanBob.append("stage", mongosStageName); + BSONArrayBuilder shardsBuilder(winningPlanBob.subarrayStart("shards")); + for (size_t i = 0; i < shardResults.size(); i++) { + BSONObjBuilder singleShardBob(shardsBuilder.subobjStart()); - singleShardBob.append("shardName", shardResults[i].shardTargetId); - { - const auto shard = - grid.shardRegistry()->getShard(shardResults[i].shardTargetId); - singleShardBob.append("connectionString", shard->getConnString().toString()); - } - appendIfRoom(&singleShardBob, serverInfo, "serverInfo"); - appendElementsIfRoom(&singleShardBob, queryPlanner); + BSONObj queryPlanner = shardResults[i].result["queryPlanner"].Obj(); + BSONObj serverInfo = shardResults[i].result["serverInfo"].Obj(); - singleShardBob.doneFast(); + singleShardBob.append("shardName", shardResults[i].shardTargetId); + { + const auto shard = grid.shardRegistry()->getShard(shardResults[i].shardTargetId); + singleShardBob.append("connectionString", shard->getConnString().toString()); } - shardsBuilder.doneFast(); - winningPlanBob.doneFast(); - queryPlannerBob.doneFast(); + appendIfRoom(&singleShardBob, serverInfo, "serverInfo"); + appendElementsIfRoom(&singleShardBob, queryPlanner); + + singleShardBob.doneFast(); + } + shardsBuilder.doneFast(); + winningPlanBob.doneFast(); + queryPlannerBob.doneFast(); +} + +// static +void ClusterExplain::buildExecStats(const vector<Strategy::CommandResult>& shardResults, + const char* mongosStageName, + long long millisElapsed, + BSONObjBuilder* out) { + if (!shardResults[0].result.hasField("executionStats")) { + // The shards don't have execution stats info. Bail out without adding anything + // to 'out'. + return; } - // static - void ClusterExplain::buildExecStats(const vector<Strategy::CommandResult>& shardResults, - const char* mongosStageName, - long long millisElapsed, - BSONObjBuilder* out) { - if (!shardResults[0].result.hasField("executionStats")) { - // The shards don't have execution stats info. Bail out without adding anything - // to 'out'. - return; + BSONObjBuilder executionStatsBob(out->subobjStart("executionStats")); + + // Collect summary stats from the shards. + long long nReturned = 0; + long long keysExamined = 0; + long long docsExamined = 0; + long long totalChildMillis = 0; + for (size_t i = 0; i < shardResults.size(); i++) { + BSONObj execStats = shardResults[i].result["executionStats"].Obj(); + if (execStats.hasField("nReturned")) { + nReturned += execStats["nReturned"].numberLong(); } - - BSONObjBuilder executionStatsBob(out->subobjStart("executionStats")); - - // Collect summary stats from the shards. - long long nReturned = 0; - long long keysExamined = 0; - long long docsExamined = 0; - long long totalChildMillis = 0; - for (size_t i = 0; i < shardResults.size(); i++) { - BSONObj execStats = shardResults[i].result["executionStats"].Obj(); - if (execStats.hasField("nReturned")) { - nReturned += execStats["nReturned"].numberLong(); - } - if (execStats.hasField("totalKeysExamined")) { - keysExamined += execStats["totalKeysExamined"].numberLong(); - } - if (execStats.hasField("totalDocsExamined")) { - docsExamined += execStats["totalDocsExamined"].numberLong(); - } - if (execStats.hasField("executionTimeMillis")) { - totalChildMillis += execStats["executionTimeMillis"].numberLong(); - } + if (execStats.hasField("totalKeysExamined")) { + keysExamined += execStats["totalKeysExamined"].numberLong(); } + if (execStats.hasField("totalDocsExamined")) { + docsExamined += execStats["totalDocsExamined"].numberLong(); + } + if (execStats.hasField("executionTimeMillis")) { + totalChildMillis += execStats["executionTimeMillis"].numberLong(); + } + } - // Fill in top-level stats. - executionStatsBob.appendNumber("nReturned", nReturned); - executionStatsBob.appendNumber("executionTimeMillis", millisElapsed); - executionStatsBob.appendNumber("totalKeysExamined", keysExamined); - executionStatsBob.appendNumber("totalDocsExamined", docsExamined); - - // Fill in the tree of stages. - BSONObjBuilder executionStagesBob( - executionStatsBob.subobjStart("executionStages")); - - // Info for the root mongos stage. - executionStagesBob.append("stage", mongosStageName); - executionStatsBob.appendNumber("nReturned", nReturned); - executionStatsBob.appendNumber("executionTimeMillis", millisElapsed); - executionStatsBob.appendNumber("totalKeysExamined", keysExamined); - executionStatsBob.appendNumber("totalDocsExamined", docsExamined); - executionStagesBob.append("totalChildMillis", totalChildMillis); + // Fill in top-level stats. + executionStatsBob.appendNumber("nReturned", nReturned); + executionStatsBob.appendNumber("executionTimeMillis", millisElapsed); + executionStatsBob.appendNumber("totalKeysExamined", keysExamined); + executionStatsBob.appendNumber("totalDocsExamined", docsExamined); - BSONArrayBuilder execShardsBuilder(executionStagesBob.subarrayStart("shards")); - for (size_t i = 0; i < shardResults.size(); i++) { - BSONObjBuilder singleShardBob(execShardsBuilder.subobjStart()); + // Fill in the tree of stages. + BSONObjBuilder executionStagesBob(executionStatsBob.subobjStart("executionStages")); - BSONObj execStats = shardResults[i].result["executionStats"].Obj(); - BSONObj execStages = execStats["executionStages"].Obj(); + // Info for the root mongos stage. + executionStagesBob.append("stage", mongosStageName); + executionStatsBob.appendNumber("nReturned", nReturned); + executionStatsBob.appendNumber("executionTimeMillis", millisElapsed); + executionStatsBob.appendNumber("totalKeysExamined", keysExamined); + executionStatsBob.appendNumber("totalDocsExamined", docsExamined); + executionStagesBob.append("totalChildMillis", totalChildMillis); - singleShardBob.append("shardName", shardResults[i].shardTargetId); + BSONArrayBuilder execShardsBuilder(executionStagesBob.subarrayStart("shards")); + for (size_t i = 0; i < shardResults.size(); i++) { + BSONObjBuilder singleShardBob(execShardsBuilder.subobjStart()); - // Append error-related fields, if present. - if (!execStats["executionSuccess"].eoo()) { - singleShardBob.append(execStats["executionSuccess"]); - } - if (!execStats["errorMessage"].eoo()) { - singleShardBob.append(execStats["errorMessage"]); - } - if (!execStats["errorCode"].eoo()) { - singleShardBob.append(execStats["errorCode"]); - } + BSONObj execStats = shardResults[i].result["executionStats"].Obj(); + BSONObj execStages = execStats["executionStages"].Obj(); - appendIfRoom(&singleShardBob, execStages, "executionStages"); + singleShardBob.append("shardName", shardResults[i].shardTargetId); - singleShardBob.doneFast(); + // Append error-related fields, if present. + if (!execStats["executionSuccess"].eoo()) { + singleShardBob.append(execStats["executionSuccess"]); + } + if (!execStats["errorMessage"].eoo()) { + singleShardBob.append(execStats["errorMessage"]); + } + if (!execStats["errorCode"].eoo()) { + singleShardBob.append(execStats["errorCode"]); } - execShardsBuilder.doneFast(); - executionStagesBob.doneFast(); + appendIfRoom(&singleShardBob, execStages, "executionStages"); - if (!shardResults[0].result["executionStats"].Obj().hasField("allPlansExecution")) { - // The shards don't have execution stats for all plans, so we're done. - executionStatsBob.doneFast(); - return; - } + singleShardBob.doneFast(); + } + execShardsBuilder.doneFast(); - // Add the allPlans stats from each shard. - BSONArrayBuilder allPlansExecBob( - executionStatsBob.subarrayStart("allPlansExecution")); - for (size_t i = 0; i < shardResults.size(); i++) { - BSONObjBuilder singleShardBob(execShardsBuilder.subobjStart()); + executionStagesBob.doneFast(); - singleShardBob.append("shardName", shardResults[i].shardTargetId); + if (!shardResults[0].result["executionStats"].Obj().hasField("allPlansExecution")) { + // The shards don't have execution stats for all plans, so we're done. + executionStatsBob.doneFast(); + return; + } - BSONObj execStats = shardResults[i].result["executionStats"].Obj(); - vector<BSONElement> allPlans = execStats["allPlansExecution"].Array(); + // Add the allPlans stats from each shard. + BSONArrayBuilder allPlansExecBob(executionStatsBob.subarrayStart("allPlansExecution")); + for (size_t i = 0; i < shardResults.size(); i++) { + BSONObjBuilder singleShardBob(execShardsBuilder.subobjStart()); - BSONArrayBuilder innerArrayBob(singleShardBob.subarrayStart("allPlans")); - for (size_t j = 0; j < allPlans.size(); j++) { - appendToArrayIfRoom(&innerArrayBob, allPlans[j]); - } - innerArrayBob.done(); + singleShardBob.append("shardName", shardResults[i].shardTargetId); - singleShardBob.doneFast(); + BSONObj execStats = shardResults[i].result["executionStats"].Obj(); + vector<BSONElement> allPlans = execStats["allPlansExecution"].Array(); + + BSONArrayBuilder innerArrayBob(singleShardBob.subarrayStart("allPlans")); + for (size_t j = 0; j < allPlans.size(); j++) { + appendToArrayIfRoom(&innerArrayBob, allPlans[j]); } - allPlansExecBob.doneFast(); + innerArrayBob.done(); - executionStatsBob.doneFast(); + singleShardBob.doneFast(); } + allPlansExecBob.doneFast(); - // static - Status ClusterExplain::buildExplainResult(const vector<Strategy::CommandResult>& shardResults, - const char* mongosStageName, - long long millisElapsed, - BSONObjBuilder* out) { - // Explain only succeeds if all shards support the explain command. - Status validateStatus = ClusterExplain::validateShardResults(shardResults); - if (!validateStatus.isOK()) { - return validateStatus; - } - - buildPlannerInfo(shardResults, mongosStageName, out); - buildExecStats(shardResults, mongosStageName, millisElapsed, out); + executionStatsBob.doneFast(); +} - return Status::OK(); +// static +Status ClusterExplain::buildExplainResult(const vector<Strategy::CommandResult>& shardResults, + const char* mongosStageName, + long long millisElapsed, + BSONObjBuilder* out) { + // Explain only succeeds if all shards support the explain command. + Status validateStatus = ClusterExplain::validateShardResults(shardResults); + if (!validateStatus.isOK()) { + return validateStatus; } -} // namespace mongo + buildPlannerInfo(shardResults, mongosStageName, out); + buildExecStats(shardResults, mongosStageName, millisElapsed, out); + + return Status::OK(); +} + +} // namespace mongo diff --git a/src/mongo/s/cluster_explain.h b/src/mongo/s/cluster_explain.h index ef86af7f76f..dad3b25c5e7 100644 --- a/src/mongo/s/cluster_explain.h +++ b/src/mongo/s/cluster_explain.h @@ -36,80 +36,80 @@ namespace mongo { +/** + * Namespace for the collection of static methods used by commands in the implementation of + * explain on mongos. + */ +class ClusterExplain { +public: /** - * Namespace for the collection of static methods used by commands in the implementation of - * explain on mongos. + * Given the BSON specification for a command, 'cmdObj', wraps the object in order to + * produce the BSON for an explain of that command, at the given verbosity level + * 'verbosity'. + * + * Adds the result to the BSONObj builder 'out'. */ - class ClusterExplain { - public: - /** - * Given the BSON specification for a command, 'cmdObj', wraps the object in order to - * produce the BSON for an explain of that command, at the given verbosity level - * 'verbosity'. - * - * Adds the result to the BSONObj builder 'out'. - */ - static void wrapAsExplain(const BSONObj& cmdObj, - ExplainCommon::Verbosity verbosity, - BSONObjBuilder* out); + static void wrapAsExplain(const BSONObj& cmdObj, + ExplainCommon::Verbosity verbosity, + BSONObjBuilder* out); - /** - * Determines the kind of "execution stage" that mongos would use in order to collect - * the results from the shards, assuming that the command being explained is a read - * operation such as find or count. - */ - static const char* getStageNameForReadOp(const std::vector<Strategy::CommandResult>& shardResults, - const BSONObj& explainObj); + /** + * Determines the kind of "execution stage" that mongos would use in order to collect + * the results from the shards, assuming that the command being explained is a read + * operation such as find or count. + */ + static const char* getStageNameForReadOp( + const std::vector<Strategy::CommandResult>& shardResults, const BSONObj& explainObj); - /** - * Command implementations on mongos use this method to construct the sharded explain - * output format based on the results from the shards in 'shardResults'. - * - * On success, the output is added to the BSONObj builder 'out'. - */ - static Status buildExplainResult(const std::vector<Strategy::CommandResult>& shardResults, - const char* mongosStageName, - long long millisElapsed, - BSONObjBuilder* out); + /** + * Command implementations on mongos use this method to construct the sharded explain + * output format based on the results from the shards in 'shardResults'. + * + * On success, the output is added to the BSONObj builder 'out'. + */ + static Status buildExplainResult(const std::vector<Strategy::CommandResult>& shardResults, + const char* mongosStageName, + long long millisElapsed, + BSONObjBuilder* out); - // - // Names of mock mongos execution stages. - // + // + // Names of mock mongos execution stages. + // - static const char* kSingleShard; - static const char* kMergeFromShards; - static const char* kMergeSortFromShards; - static const char* kWriteOnShards; + static const char* kSingleShard; + static const char* kMergeFromShards; + static const char* kMergeSortFromShards; + static const char* kWriteOnShards; - private: - /** - * Returns an OK status if all shards support the explain command and returned sensible - * results. Otherwise, returns a non-OK status and the entire explain should fail. - */ - static Status validateShardResults(const std::vector<Strategy::CommandResult>& shardResults); +private: + /** + * Returns an OK status if all shards support the explain command and returned sensible + * results. Otherwise, returns a non-OK status and the entire explain should fail. + */ + static Status validateShardResults(const std::vector<Strategy::CommandResult>& shardResults); - /** - * Populates the BSONObj builder 'out' with query planner explain information, based on - * the results from the shards contained in 'shardResults'. - * - * The planner info will display 'mongosStageName' as the name of the execution stage - * performed by mongos after gathering results from the shards. - */ - static void buildPlannerInfo(const std::vector<Strategy::CommandResult>& shardResults, - const char* mongosStageName, - BSONObjBuilder* out); + /** + * Populates the BSONObj builder 'out' with query planner explain information, based on + * the results from the shards contained in 'shardResults'. + * + * The planner info will display 'mongosStageName' as the name of the execution stage + * performed by mongos after gathering results from the shards. + */ + static void buildPlannerInfo(const std::vector<Strategy::CommandResult>& shardResults, + const char* mongosStageName, + BSONObjBuilder* out); - /** - * Populates the BSONObj builder 'out' with execution stats explain information, - * if the results from the shards in 'shardsResults' contain this info. - * - * Will display 'mongosStageName' as the name of the execution stage performed by mongos, - * and 'millisElapsed' as the execution time of the mongos stage. - */ - static void buildExecStats(const std::vector<Strategy::CommandResult>& shardResults, - const char* mongosStageName, - long long millisElapsed, - BSONObjBuilder* out); - }; + /** + * Populates the BSONObj builder 'out' with execution stats explain information, + * if the results from the shards in 'shardsResults' contain this info. + * + * Will display 'mongosStageName' as the name of the execution stage performed by mongos, + * and 'millisElapsed' as the execution time of the mongos stage. + */ + static void buildExecStats(const std::vector<Strategy::CommandResult>& shardResults, + const char* mongosStageName, + long long millisElapsed, + BSONObjBuilder* out); +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/cluster_last_error_info.cpp b/src/mongo/s/cluster_last_error_info.cpp index 30c79001be3..18124138825 100644 --- a/src/mongo/s/cluster_last_error_info.cpp +++ b/src/mongo/s/cluster_last_error_info.cpp @@ -43,72 +43,69 @@ namespace mongo { - const ClientBasic::Decoration<ClusterLastErrorInfo> ClusterLastErrorInfo::get = - ClientBasic::declareDecoration<ClusterLastErrorInfo>(); +const ClientBasic::Decoration<ClusterLastErrorInfo> ClusterLastErrorInfo::get = + ClientBasic::declareDecoration<ClusterLastErrorInfo>(); - void ClusterLastErrorInfo::addShardHost(const std::string& shardHost) { - _cur->shardHostsWritten.insert( shardHost ); - } +void ClusterLastErrorInfo::addShardHost(const std::string& shardHost) { + _cur->shardHostsWritten.insert(shardHost); +} - void ClusterLastErrorInfo::addHostOpTime(ConnectionString connStr, HostOpTime stat) { - _cur->hostOpTimes[connStr] = stat; - } +void ClusterLastErrorInfo::addHostOpTime(ConnectionString connStr, HostOpTime stat) { + _cur->hostOpTimes[connStr] = stat; +} - void ClusterLastErrorInfo::addHostOpTimes( const HostOpTimeMap& hostOpTimes ) { - for ( HostOpTimeMap::const_iterator it = hostOpTimes.begin(); - it != hostOpTimes.end(); ++it ) { - addHostOpTime(it->first, it->second); - } +void ClusterLastErrorInfo::addHostOpTimes(const HostOpTimeMap& hostOpTimes) { + for (HostOpTimeMap::const_iterator it = hostOpTimes.begin(); it != hostOpTimes.end(); ++it) { + addHostOpTime(it->first, it->second); + } +} + +void ClusterLastErrorInfo::newRequest() { + std::swap(_cur, _prev); + _cur->clear(); +} + +void ClusterLastErrorInfo::disableForCommand() { + RequestInfo* temp = _cur; + _cur = _prev; + _prev = temp; +} + +static TimerStats gleWtimeStats; +static ServerStatusMetricField<TimerStats> displayGleLatency("getLastError.wtime", &gleWtimeStats); + +void saveGLEStats(const BSONObj& metadata, StringData hostString) { + if (!haveClient()) { + // TODO: how can this happen? + return; } - void ClusterLastErrorInfo::newRequest() { - std::swap(_cur, _prev); - _cur->clear(); + auto swShardingMetadata = rpc::ShardingMetadata::readFromMetadata(metadata); + if (swShardingMetadata.getStatus() == ErrorCodes::NoSuchKey) { + return; + } else if (!swShardingMetadata.isOK()) { + warning() << "Got invalid sharding metadata " << swShardingMetadata.getStatus() + << " metadata object was '" << metadata << "'"; + return; } - void ClusterLastErrorInfo::disableForCommand() { - RequestInfo* temp = _cur; - _cur = _prev; - _prev = temp; + auto shardConn = ConnectionString::parse(hostString.toString()); + // If we got the reply from this host, we expect that its 'hostString' must be valid. + if (!shardConn.isOK()) { + severe() << "got bad host string in saveGLEStats: " << hostString; } + invariantOK(shardConn.getStatus()); - static TimerStats gleWtimeStats; - static ServerStatusMetricField<TimerStats> displayGleLatency("getLastError.wtime", - &gleWtimeStats); - - void saveGLEStats(const BSONObj& metadata, StringData hostString) { - if (!haveClient()) { - // TODO: how can this happen? - return; - } - - auto swShardingMetadata = rpc::ShardingMetadata::readFromMetadata(metadata); - if (swShardingMetadata.getStatus() == ErrorCodes::NoSuchKey) { - return; - } - else if (!swShardingMetadata.isOK()) { - warning() << "Got invalid sharding metadata " << swShardingMetadata.getStatus() - << " metadata object was '" << metadata << "'"; - return; - } - - auto shardConn = ConnectionString::parse(hostString.toString()); - // If we got the reply from this host, we expect that its 'hostString' must be valid. - if (!shardConn.isOK()) { - severe() << "got bad host string in saveGLEStats: " << hostString; - } - invariantOK(shardConn.getStatus()); - - auto shardingMetadata = std::move(swShardingMetadata.getValue()); - - auto& clientInfo = cc(); - LOG(4) << "saveGLEStats lastOpTime:" << shardingMetadata.getLastOpTime() - << " electionId:" << shardingMetadata.getLastElectionId(); - - ClusterLastErrorInfo::get(clientInfo).addHostOpTime( + auto shardingMetadata = std::move(swShardingMetadata.getValue()); + + auto& clientInfo = cc(); + LOG(4) << "saveGLEStats lastOpTime:" << shardingMetadata.getLastOpTime() + << " electionId:" << shardingMetadata.getLastElectionId(); + + ClusterLastErrorInfo::get(clientInfo) + .addHostOpTime( shardConn.getValue(), - HostOpTime(shardingMetadata.getLastOpTime(), - shardingMetadata.getLastElectionId())); - } + HostOpTime(shardingMetadata.getLastOpTime(), shardingMetadata.getLastElectionId())); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/cluster_last_error_info.h b/src/mongo/s/cluster_last_error_info.h index a3110e3d40b..d6104518957 100644 --- a/src/mongo/s/cluster_last_error_info.h +++ b/src/mongo/s/cluster_last_error_info.h @@ -34,76 +34,79 @@ namespace mongo { +/** + * Client decoration that holds information needed to by mongos to process + * getLastError commands. + */ +class ClusterLastErrorInfo { +public: + static const ClientBasic::Decoration<ClusterLastErrorInfo> get; + + /** new request not associated (yet or ever) with a client */ + void newRequest(); + /** - * Client decoration that holds information needed to by mongos to process - * getLastError commands. + * notes that this client use this shard + * keeps track of all shards accessed this request */ - class ClusterLastErrorInfo { - public: - static const ClientBasic::Decoration<ClusterLastErrorInfo> get; - - /** new request not associated (yet or ever) with a client */ - void newRequest(); - - /** - * notes that this client use this shard - * keeps track of all shards accessed this request - */ - void addShardHost( const std::string& shardHost ); - - /** - * Notes that this client wrote to these particular hosts with write commands. - */ - void addHostOpTime(ConnectionString connstr, HostOpTime stat); - void addHostOpTimes( const HostOpTimeMap& hostOpTimes ); - - /** - * gets shards used on the previous request - */ - std::set<std::string>* getPrevShardHosts() const { return &_prev->shardHostsWritten; } - - /** - * Gets the shards, hosts, and opTimes the client last wrote to with write commands. - */ - const HostOpTimeMap& getPrevHostOpTimes() const { - return _prev->hostOpTimes; - } + void addShardHost(const std::string& shardHost); - /** - * resets the information stored for the current request - */ - void clearRequestInfo(){ _cur->clear(); } + /** + * Notes that this client wrote to these particular hosts with write commands. + */ + void addHostOpTime(ConnectionString connstr, HostOpTime stat); + void addHostOpTimes(const HostOpTimeMap& hostOpTimes); - void disableForCommand(); + /** + * gets shards used on the previous request + */ + std::set<std::string>* getPrevShardHosts() const { + return &_prev->shardHostsWritten; + } - private: - struct RequestInfo { + /** + * Gets the shards, hosts, and opTimes the client last wrote to with write commands. + */ + const HostOpTimeMap& getPrevHostOpTimes() const { + return _prev->hostOpTimes; + } - void clear() { - shardHostsWritten.clear(); - hostOpTimes.clear(); - } + /** + * resets the information stored for the current request + */ + void clearRequestInfo() { + _cur->clear(); + } - std::set<std::string> shardHostsWritten; - HostOpTimeMap hostOpTimes; - }; + void disableForCommand(); - // We use 2 so we can flip for getLastError type operations. - RequestInfo _infos[2]; - RequestInfo* _cur = &_infos[0]; - RequestInfo* _prev = &_infos[1]; +private: + struct RequestInfo { + void clear() { + shardHostsWritten.clear(); + hostOpTimes.clear(); + } + + std::set<std::string> shardHostsWritten; + HostOpTimeMap hostOpTimes; }; - /** - * Looks for $gleStats in a command's reply metadata, and fills in the ClusterLastErrorInfo - * for this thread's associated Client with the data, if found. - * - * This data will be used by subsequent GLE calls, to ensure we look for the correct - * write on the correct PRIMARY. - * result: the result from calling runCommand - * conn: the std::string name of the hostAndPort where the command ran. This can be a replica - * set seed list. - */ - void saveGLEStats(const BSONObj& metadataObj, StringData conn); + // We use 2 so we can flip for getLastError type operations. + RequestInfo _infos[2]; + RequestInfo* _cur = &_infos[0]; + RequestInfo* _prev = &_infos[1]; +}; + +/** + * Looks for $gleStats in a command's reply metadata, and fills in the ClusterLastErrorInfo + * for this thread's associated Client with the data, if found. + * + * This data will be used by subsequent GLE calls, to ensure we look for the correct + * write on the correct PRIMARY. + * result: the result from calling runCommand + * conn: the std::string name of the hostAndPort where the command ran. This can be a replica + * set seed list. + */ +void saveGLEStats(const BSONObj& metadataObj, StringData conn); } // namespace mongo diff --git a/src/mongo/s/cluster_write.cpp b/src/mongo/s/cluster_write.cpp index 4d6e5b878b5..45b3b0abb52 100644 --- a/src/mongo/s/cluster_write.cpp +++ b/src/mongo/s/cluster_write.cpp @@ -51,250 +51,239 @@ namespace mongo { - using std::shared_ptr; - using std::unique_ptr; - using std::vector; - using std::map; - using std::string; - using std::stringstream; - - const int ConfigOpTimeoutMillis = 30 * 1000; - - namespace { - - /** - * Constructs the BSON specification document for the given namespace, index key - * and options. - */ - BSONObj createIndexDoc(const string& ns, const BSONObj& keys, bool unique) { - BSONObjBuilder indexDoc; - indexDoc.append("ns", ns); - indexDoc.append("key", keys); - - stringstream indexName; - - bool isFirstKey = true; - for (BSONObjIterator keyIter(keys); keyIter.more();) { - BSONElement currentKey = keyIter.next(); - - if (isFirstKey) { - isFirstKey = false; - } - else { - indexName << "_"; - } - - indexName << currentKey.fieldName() << "_"; - if (currentKey.isNumber()) { - indexName << currentKey.numberInt(); - } - else { - indexName << currentKey.str(); //this should match up with shell command - } - } - - indexDoc.append("name", indexName.str()); - - if (unique) { - indexDoc.appendBool("unique", unique); - } - - return indexDoc.obj(); - } +using std::shared_ptr; +using std::unique_ptr; +using std::vector; +using std::map; +using std::string; +using std::stringstream; + +const int ConfigOpTimeoutMillis = 30 * 1000; + +namespace { + +/** + * Constructs the BSON specification document for the given namespace, index key + * and options. + */ +BSONObj createIndexDoc(const string& ns, const BSONObj& keys, bool unique) { + BSONObjBuilder indexDoc; + indexDoc.append("ns", ns); + indexDoc.append("key", keys); + + stringstream indexName; - void toBatchError(const Status& status, BatchedCommandResponse* response) { - response->clear(); - response->setErrCode(status.code()); - response->setErrMessage(status.reason()); - response->setOk(false); - dassert(response->isValid(NULL)); + bool isFirstKey = true; + for (BSONObjIterator keyIter(keys); keyIter.more();) { + BSONElement currentKey = keyIter.next(); + + if (isFirstKey) { + isFirstKey = false; + } else { + indexName << "_"; } - /** - * Splits the chunks touched based from the targeter stats if needed. - */ - void splitIfNeeded(const NamespaceString& nss, const TargeterStats& stats) { - if (!Chunk::ShouldAutoSplit) { - return; - } - - auto status = grid.catalogCache()->getDatabase(nss.db().toString()); - if (!status.isOK()) { - warning() << "failed to get database config for " << nss - << " while checking for auto-split: " << status.getStatus(); - return; - } - - shared_ptr<DBConfig> config = status.getValue(); - - ChunkManagerPtr chunkManager; - ShardPtr dummyShard; - config->getChunkManagerOrPrimary(nss, chunkManager, dummyShard); - - if (!chunkManager) { - return; - } - - for (map<BSONObj, int>::const_iterator it = stats.chunkSizeDelta.begin(); - it != stats.chunkSizeDelta.end(); - ++it) { - - ChunkPtr chunk; - try { - chunk = chunkManager->findIntersectingChunk(it->first); - } - catch (const AssertionException& ex) { - warning() << "could not find chunk while checking for auto-split: " - << causedBy(ex); - return; - } - - chunk->splitIfShould(it->second); - } + indexName << currentKey.fieldName() << "_"; + if (currentKey.isNumber()) { + indexName << currentKey.numberInt(); + } else { + indexName << currentKey.str(); // this should match up with shell command } + } - } // namespace + indexDoc.append("name", indexName.str()); - Status clusterCreateIndex( const string& ns, - BSONObj keys, - bool unique, - BatchedCommandResponse* response ) { + if (unique) { + indexDoc.appendBool("unique", unique); + } - const NamespaceString nss(ns); - const std::string dbName = nss.db().toString(); + return indexDoc.obj(); +} - BSONObj indexDoc = createIndexDoc(ns, keys, unique); +void toBatchError(const Status& status, BatchedCommandResponse* response) { + response->clear(); + response->setErrCode(status.code()); + response->setErrMessage(status.reason()); + response->setOk(false); + dassert(response->isValid(NULL)); +} - // Go through the shard insert path - std::unique_ptr<BatchedInsertRequest> insert(new BatchedInsertRequest()); - insert->addToDocuments(indexDoc); +/** + * Splits the chunks touched based from the targeter stats if needed. + */ +void splitIfNeeded(const NamespaceString& nss, const TargeterStats& stats) { + if (!Chunk::ShouldAutoSplit) { + return; + } - BatchedCommandRequest request(insert.release()); - request.setNS(nss.getSystemIndexesCollection()); - request.setWriteConcern(WriteConcernOptions::Acknowledged); + auto status = grid.catalogCache()->getDatabase(nss.db().toString()); + if (!status.isOK()) { + warning() << "failed to get database config for " << nss + << " while checking for auto-split: " << status.getStatus(); + return; + } - BatchedCommandResponse dummyResponse; - if (response == NULL) { - response = &dummyResponse; - } + shared_ptr<DBConfig> config = status.getValue(); + + ChunkManagerPtr chunkManager; + ShardPtr dummyShard; + config->getChunkManagerOrPrimary(nss, chunkManager, dummyShard); - ClusterWriter writer(false, 0); - writer.write(request, response); + if (!chunkManager) { + return; + } - if (response->getOk() != 1) { - return Status(static_cast<ErrorCodes::Error>(response->getErrCode()), - response->getErrMessage()); + for (map<BSONObj, int>::const_iterator it = stats.chunkSizeDelta.begin(); + it != stats.chunkSizeDelta.end(); + ++it) { + ChunkPtr chunk; + try { + chunk = chunkManager->findIntersectingChunk(it->first); + } catch (const AssertionException& ex) { + warning() << "could not find chunk while checking for auto-split: " << causedBy(ex); + return; } - if (response->isErrDetailsSet()) { - const WriteErrorDetail* errDetail = response->getErrDetails().front(); + chunk->splitIfShould(it->second); + } +} + +} // namespace - return Status(static_cast<ErrorCodes::Error>(errDetail->getErrCode()), - errDetail->getErrMessage()); - } +Status clusterCreateIndex(const string& ns, + BSONObj keys, + bool unique, + BatchedCommandResponse* response) { + const NamespaceString nss(ns); + const std::string dbName = nss.db().toString(); - if (response->isWriteConcernErrorSet()) { - const WCErrorDetail* errDetail = response->getWriteConcernError(); + BSONObj indexDoc = createIndexDoc(ns, keys, unique); - return Status(static_cast<ErrorCodes::Error>(errDetail->getErrCode()), - errDetail->getErrMessage()); - } + // Go through the shard insert path + std::unique_ptr<BatchedInsertRequest> insert(new BatchedInsertRequest()); + insert->addToDocuments(indexDoc); - return Status::OK(); + BatchedCommandRequest request(insert.release()); + request.setNS(nss.getSystemIndexesCollection()); + request.setWriteConcern(WriteConcernOptions::Acknowledged); + + BatchedCommandResponse dummyResponse; + if (response == NULL) { + response = &dummyResponse; } + ClusterWriter writer(false, 0); + writer.write(request, response); - void ClusterWriter::write( const BatchedCommandRequest& origRequest, - BatchedCommandResponse* response ) { + if (response->getOk() != 1) { + return Status(static_cast<ErrorCodes::Error>(response->getErrCode()), + response->getErrMessage()); + } - // Add _ids to insert request if req'd - unique_ptr<BatchedCommandRequest> idRequest(BatchedCommandRequest::cloneWithIds(origRequest)); - const BatchedCommandRequest& request = NULL != idRequest.get() ? *idRequest : origRequest; + if (response->isErrDetailsSet()) { + const WriteErrorDetail* errDetail = response->getErrDetails().front(); - const NamespaceString& nss = request.getNSS(); - if ( !nss.isValid() ) { - toBatchError( Status( ErrorCodes::InvalidNamespace, - nss.ns() + " is not a valid namespace" ), - response ); - return; - } + return Status(static_cast<ErrorCodes::Error>(errDetail->getErrCode()), + errDetail->getErrMessage()); + } - if ( !NamespaceString::validCollectionName( nss.coll() ) ) { - toBatchError( Status( ErrorCodes::BadValue, - str::stream() << "invalid collection name " << nss.coll() ), - response ); - return; - } + if (response->isWriteConcernErrorSet()) { + const WCErrorDetail* errDetail = response->getWriteConcernError(); - if ( request.sizeWriteOps() == 0u ) { - toBatchError( Status( ErrorCodes::InvalidLength, - "no write ops were included in the batch" ), - response ); - return; - } + return Status(static_cast<ErrorCodes::Error>(errDetail->getErrCode()), + errDetail->getErrMessage()); + } - if ( request.sizeWriteOps() > BatchedCommandRequest::kMaxWriteBatchSize ) { - toBatchError( Status( ErrorCodes::InvalidLength, - str::stream() << "exceeded maximum write batch size of " - << BatchedCommandRequest::kMaxWriteBatchSize ), - response ); - return; - } + return Status::OK(); +} - string errMsg; - if ( request.isInsertIndexRequest() && !request.isValidIndexRequest( &errMsg ) ) { - toBatchError( Status( ErrorCodes::InvalidOptions, errMsg ), response ); - return; - } - // Config writes and shard writes are done differently - const string dbName = nss.db().toString(); +void ClusterWriter::write(const BatchedCommandRequest& origRequest, + BatchedCommandResponse* response) { + // Add _ids to insert request if req'd + unique_ptr<BatchedCommandRequest> idRequest(BatchedCommandRequest::cloneWithIds(origRequest)); + const BatchedCommandRequest& request = NULL != idRequest.get() ? *idRequest : origRequest; - if (dbName == "config" || dbName == "admin") { - grid.catalogManager()->writeConfigServerDirect(request, response); - } - else { - ChunkManagerTargeter targeter(request.getTargetingNSS()); - Status targetInitStatus = targeter.init(); - - if (!targetInitStatus.isOK()) { - // Errors will be reported in response if we are unable to target - warning() << "could not initialize targeter for" - << (request.isInsertIndexRequest() ? " index" : "") - << " write op in collection " << request.getTargetingNS(); - } - - DBClientShardResolver resolver; - DBClientMultiCommand dispatcher; - BatchWriteExec exec(&targeter, &resolver, &dispatcher); - exec.executeBatch(request, response); - - if (_autoSplit) { - splitIfNeeded(request.getNSS(), *targeter.getStats()); - } - - _stats->setShardStats(exec.releaseStats()); - } + const NamespaceString& nss = request.getNSS(); + if (!nss.isValid()) { + toBatchError(Status(ErrorCodes::InvalidNamespace, nss.ns() + " is not a valid namespace"), + response); + return; } - ClusterWriter::ClusterWriter( bool autoSplit, int timeoutMillis ) : - _autoSplit( autoSplit ), _timeoutMillis( timeoutMillis ), _stats( new ClusterWriterStats ) { + if (!NamespaceString::validCollectionName(nss.coll())) { + toBatchError( + Status(ErrorCodes::BadValue, str::stream() << "invalid collection name " << nss.coll()), + response); + return; } - const ClusterWriterStats& ClusterWriter::getStats() { - return *_stats; + if (request.sizeWriteOps() == 0u) { + toBatchError(Status(ErrorCodes::InvalidLength, "no write ops were included in the batch"), + response); + return; } - void ClusterWriterStats::setShardStats( BatchWriteExecStats* shardStats ) { - _shardStats.reset( shardStats ); + if (request.sizeWriteOps() > BatchedCommandRequest::kMaxWriteBatchSize) { + toBatchError(Status(ErrorCodes::InvalidLength, + str::stream() << "exceeded maximum write batch size of " + << BatchedCommandRequest::kMaxWriteBatchSize), + response); + return; } - bool ClusterWriterStats::hasShardStats() const { - return NULL != _shardStats.get(); + string errMsg; + if (request.isInsertIndexRequest() && !request.isValidIndexRequest(&errMsg)) { + toBatchError(Status(ErrorCodes::InvalidOptions, errMsg), response); + return; } - const BatchWriteExecStats& ClusterWriterStats::getShardStats() const { - return *_shardStats; + // Config writes and shard writes are done differently + const string dbName = nss.db().toString(); + + if (dbName == "config" || dbName == "admin") { + grid.catalogManager()->writeConfigServerDirect(request, response); + } else { + ChunkManagerTargeter targeter(request.getTargetingNSS()); + Status targetInitStatus = targeter.init(); + + if (!targetInitStatus.isOK()) { + // Errors will be reported in response if we are unable to target + warning() << "could not initialize targeter for" + << (request.isInsertIndexRequest() ? " index" : "") + << " write op in collection " << request.getTargetingNS(); + } + + DBClientShardResolver resolver; + DBClientMultiCommand dispatcher; + BatchWriteExec exec(&targeter, &resolver, &dispatcher); + exec.executeBatch(request, response); + + if (_autoSplit) { + splitIfNeeded(request.getNSS(), *targeter.getStats()); + } + + _stats->setShardStats(exec.releaseStats()); } +} + +ClusterWriter::ClusterWriter(bool autoSplit, int timeoutMillis) + : _autoSplit(autoSplit), _timeoutMillis(timeoutMillis), _stats(new ClusterWriterStats) {} + +const ClusterWriterStats& ClusterWriter::getStats() { + return *_stats; +} + +void ClusterWriterStats::setShardStats(BatchWriteExecStats* shardStats) { + _shardStats.reset(shardStats); +} + +bool ClusterWriterStats::hasShardStats() const { + return NULL != _shardStats.get(); +} + +const BatchWriteExecStats& ClusterWriterStats::getShardStats() const { + return *_shardStats; +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/cluster_write.h b/src/mongo/s/cluster_write.h index fce2f26564d..9fd177f756a 100644 --- a/src/mongo/s/cluster_write.h +++ b/src/mongo/s/cluster_write.h @@ -35,50 +35,47 @@ namespace mongo { - class ClusterWriterStats; - class BatchWriteExecStats; +class ClusterWriterStats; +class BatchWriteExecStats; - class ClusterWriter { - public: +class ClusterWriter { +public: + ClusterWriter(bool autoSplit, int timeoutMillis); - ClusterWriter( bool autoSplit, int timeoutMillis ); + void write(const BatchedCommandRequest& request, BatchedCommandResponse* response); - void write( const BatchedCommandRequest& request, BatchedCommandResponse* response ); + const ClusterWriterStats& getStats(); - const ClusterWriterStats& getStats(); +private: + const bool _autoSplit; + const int _timeoutMillis; - private: - const bool _autoSplit; - const int _timeoutMillis; + std::unique_ptr<ClusterWriterStats> _stats; +}; - std::unique_ptr<ClusterWriterStats> _stats; - }; +class ClusterWriterStats { +public: + // Transfers ownership to the cluster write stats + void setShardStats(BatchWriteExecStats* _shardStats); - class ClusterWriterStats { - public: + bool hasShardStats() const; - // Transfers ownership to the cluster write stats - void setShardStats( BatchWriteExecStats* _shardStats ); + const BatchWriteExecStats& getShardStats() const; - bool hasShardStats() const; + // TODO: When we have ConfigCoordinator stats, put these here too. - const BatchWriteExecStats& getShardStats() const; +private: + std::unique_ptr<BatchWriteExecStats> _shardStats; +}; - // TODO: When we have ConfigCoordinator stats, put these here too. - - private: - - std::unique_ptr<BatchWriteExecStats> _shardStats; - }; - - /** - * Used only for writes to the config server, config and admin databases. - * - * Note: response can be NULL if you don't care about the write statistics. - */ - Status clusterCreateIndex( const std::string& ns, - BSONObj keys, - bool unique, - BatchedCommandResponse* response ); +/** + * Used only for writes to the config server, config and admin databases. + * + * Note: response can be NULL if you don't care about the write statistics. + */ +Status clusterCreateIndex(const std::string& ns, + BSONObj keys, + bool unique, + BatchedCommandResponse* response); -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/collection_metadata.cpp b/src/mongo/s/collection_metadata.cpp index 98f62c90a01..05d448dc239 100644 --- a/src/mongo/s/collection_metadata.cpp +++ b/src/mongo/s/collection_metadata.cpp @@ -38,431 +38,398 @@ namespace mongo { - using std::unique_ptr; - using std::endl; - using std::make_pair; - using std::string; - using std::vector; +using std::unique_ptr; +using std::endl; +using std::make_pair; +using std::string; +using std::vector; - using mongoutils::str::stream; +using mongoutils::str::stream; - CollectionMetadata::CollectionMetadata() { } +CollectionMetadata::CollectionMetadata() {} - CollectionMetadata::~CollectionMetadata() { } +CollectionMetadata::~CollectionMetadata() {} - CollectionMetadata* CollectionMetadata::cloneMigrate( const ChunkType& chunk, - const ChunkVersion& newShardVersion, - string* errMsg ) const { - // The error message string is optional. - string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } +CollectionMetadata* CollectionMetadata::cloneMigrate(const ChunkType& chunk, + const ChunkVersion& newShardVersion, + string* errMsg) const { + // The error message string is optional. + string dummy; + if (errMsg == NULL) { + errMsg = &dummy; + } - // Check that we have the exact chunk that will be subtracted. - if ( !rangeMapContains( _chunksMap, chunk.getMin(), chunk.getMax() ) ) { + // Check that we have the exact chunk that will be subtracted. + if (!rangeMapContains(_chunksMap, chunk.getMin(), chunk.getMax())) { + *errMsg = stream() << "cannot remove chunk " + << rangeToString(chunk.getMin(), chunk.getMax()) + << ", this shard does not contain the chunk"; - *errMsg = stream() << "cannot remove chunk " - << rangeToString( chunk.getMin(), chunk.getMax() ) - << ", this shard does not contain the chunk"; + if (rangeMapOverlaps(_chunksMap, chunk.getMin(), chunk.getMax())) { + RangeVector overlap; + getRangeMapOverlap(_chunksMap, chunk.getMin(), chunk.getMax(), &overlap); - if ( rangeMapOverlaps( _chunksMap, chunk.getMin(), chunk.getMax() ) ) { + *errMsg += stream() << " and it overlaps " << overlapToString(overlap); + } - RangeVector overlap; - getRangeMapOverlap( _chunksMap, chunk.getMin(), chunk.getMax(), &overlap ); + warning() << *errMsg << endl; + return NULL; + } - *errMsg += stream() << " and it overlaps " << overlapToString( overlap ); - } + // If left with no chunks, check that the version is zero. + if (_chunksMap.size() == 1) { + if (newShardVersion.isSet()) { + *errMsg = stream() << "cannot set shard version to non-zero value " + << newShardVersion.toString() << " when removing last chunk " + << rangeToString(chunk.getMin(), chunk.getMax()); warning() << *errMsg << endl; return NULL; } + } + // Can't move version backwards when subtracting chunks. This is what guarantees that + // no read or write would be taken once we subtract data from the current shard. + else if (newShardVersion <= _shardVersion) { + *errMsg = stream() << "cannot remove chunk " + << rangeToString(chunk.getMin(), chunk.getMax()) + << " because the new shard version " << newShardVersion.toString() + << " is not greater than the current shard version " + << _shardVersion.toString(); + + warning() << *errMsg << endl; + return NULL; + } - // If left with no chunks, check that the version is zero. - if (_chunksMap.size() == 1) { - if (newShardVersion.isSet()) { - - *errMsg = stream() << "cannot set shard version to non-zero value " - << newShardVersion.toString() << " when removing last chunk " - << rangeToString( chunk.getMin(), chunk.getMax() ); - - warning() << *errMsg << endl; - return NULL; - } - } - // Can't move version backwards when subtracting chunks. This is what guarantees that - // no read or write would be taken once we subtract data from the current shard. - else if (newShardVersion <= _shardVersion) { + unique_ptr<CollectionMetadata> metadata(new CollectionMetadata); + metadata->_keyPattern = this->_keyPattern; + metadata->_keyPattern.getOwned(); + metadata->fillKeyPatternFields(); + metadata->_pendingMap = this->_pendingMap; + metadata->_chunksMap = this->_chunksMap; + metadata->_chunksMap.erase(chunk.getMin()); + metadata->_shardVersion = newShardVersion; + metadata->_collVersion = newShardVersion > _collVersion ? newShardVersion : this->_collVersion; + metadata->fillRanges(); + + invariant(metadata->isValid()); + return metadata.release(); +} + +CollectionMetadata* CollectionMetadata::clonePlusChunk(const ChunkType& chunk, + const ChunkVersion& newShardVersion, + string* errMsg) const { + // The error message string is optional. + string dummy; + if (errMsg == NULL) { + errMsg = &dummy; + } - *errMsg = stream() << "cannot remove chunk " - << rangeToString( chunk.getMin(), chunk.getMax() ) - << " because the new shard version " << newShardVersion.toString() - << " is not greater than the current shard version " - << _shardVersion.toString(); + // It is acceptable to move version backwards (e.g., undoing a migration that went bad + // during commit) but only cloning away the last chunk may reset the version to 0. + if (!newShardVersion.isSet()) { + *errMsg = stream() << "cannot add chunk " << rangeToString(chunk.getMin(), chunk.getMax()) + << " with zero shard version"; - warning() << *errMsg << endl; - return NULL; - } + warning() << *errMsg << endl; + return NULL; + } - unique_ptr<CollectionMetadata> metadata( new CollectionMetadata ); - metadata->_keyPattern = this->_keyPattern; - metadata->_keyPattern.getOwned(); - metadata->fillKeyPatternFields(); - metadata->_pendingMap = this->_pendingMap; - metadata->_chunksMap = this->_chunksMap; - metadata->_chunksMap.erase( chunk.getMin() ); - metadata->_shardVersion = newShardVersion; - metadata->_collVersion = - newShardVersion > _collVersion ? newShardVersion : this->_collVersion; - metadata->fillRanges(); - - invariant(metadata->isValid()); - return metadata.release(); - } - - CollectionMetadata* CollectionMetadata::clonePlusChunk( const ChunkType& chunk, - const ChunkVersion& newShardVersion, - string* errMsg ) const { - // The error message string is optional. - string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } + invariant(chunk.getMin().woCompare(chunk.getMax()) < 0); - // It is acceptable to move version backwards (e.g., undoing a migration that went bad - // during commit) but only cloning away the last chunk may reset the version to 0. - if (!newShardVersion.isSet()) { + // Check that there isn't any chunk on the interval to be added. + if (rangeMapOverlaps(_chunksMap, chunk.getMin(), chunk.getMax())) { + RangeVector overlap; + getRangeMapOverlap(_chunksMap, chunk.getMin(), chunk.getMax(), &overlap); - *errMsg = stream() << "cannot add chunk " - << rangeToString( chunk.getMin(), chunk.getMax() ) - << " with zero shard version"; + *errMsg = stream() << "cannot add chunk " << rangeToString(chunk.getMin(), chunk.getMax()) + << " because the chunk overlaps " << overlapToString(overlap); - warning() << *errMsg << endl; - return NULL; - } + warning() << *errMsg << endl; + return NULL; + } - invariant( chunk.getMin().woCompare(chunk.getMax()) < 0 ); + unique_ptr<CollectionMetadata> metadata(new CollectionMetadata); + metadata->_keyPattern = this->_keyPattern; + metadata->_keyPattern.getOwned(); + metadata->fillKeyPatternFields(); + metadata->_pendingMap = this->_pendingMap; + metadata->_chunksMap = this->_chunksMap; + metadata->_chunksMap.insert(make_pair(chunk.getMin().getOwned(), chunk.getMax().getOwned())); + metadata->_shardVersion = newShardVersion; + metadata->_collVersion = newShardVersion > _collVersion ? newShardVersion : this->_collVersion; + metadata->fillRanges(); + + invariant(metadata->isValid()); + return metadata.release(); +} + +CollectionMetadata* CollectionMetadata::cloneMinusPending(const ChunkType& pending, + string* errMsg) const { + // The error message string is optional. + string dummy; + if (errMsg == NULL) { + errMsg = &dummy; + } - // Check that there isn't any chunk on the interval to be added. - if ( rangeMapOverlaps( _chunksMap, chunk.getMin(), chunk.getMax() ) ) { + // Check that we have the exact chunk that will be subtracted. + if (!rangeMapContains(_pendingMap, pending.getMin(), pending.getMax())) { + *errMsg = stream() << "cannot remove pending chunk " + << rangeToString(pending.getMin(), pending.getMax()) + << ", this shard does not contain the chunk"; + if (rangeMapOverlaps(_pendingMap, pending.getMin(), pending.getMax())) { RangeVector overlap; - getRangeMapOverlap( _chunksMap, chunk.getMin(), chunk.getMax(), &overlap ); - - *errMsg = stream() << "cannot add chunk " - << rangeToString( chunk.getMin(), chunk.getMax() ) - << " because the chunk overlaps " << overlapToString( overlap ); + getRangeMapOverlap(_pendingMap, pending.getMin(), pending.getMax(), &overlap); - warning() << *errMsg << endl; - return NULL; + *errMsg += stream() << " and it overlaps " << overlapToString(overlap); } - unique_ptr<CollectionMetadata> metadata( new CollectionMetadata ); - metadata->_keyPattern = this->_keyPattern; - metadata->_keyPattern.getOwned(); - metadata->fillKeyPatternFields(); - metadata->_pendingMap = this->_pendingMap; - metadata->_chunksMap = this->_chunksMap; - metadata->_chunksMap.insert( make_pair( chunk.getMin().getOwned(), - chunk.getMax().getOwned() ) ); - metadata->_shardVersion = newShardVersion; - metadata->_collVersion = - newShardVersion > _collVersion ? newShardVersion : this->_collVersion; - metadata->fillRanges(); - - invariant(metadata->isValid()); - return metadata.release(); - } - - CollectionMetadata* CollectionMetadata::cloneMinusPending( const ChunkType& pending, - string* errMsg ) const { - // The error message string is optional. - string dummy; - if ( errMsg == NULL ) { - errMsg = &dummy; - } - - // Check that we have the exact chunk that will be subtracted. - if ( !rangeMapContains( _pendingMap, pending.getMin(), pending.getMax() ) ) { - - *errMsg = stream() << "cannot remove pending chunk " - << rangeToString( pending.getMin(), pending.getMax() ) - << ", this shard does not contain the chunk"; - - if ( rangeMapOverlaps( _pendingMap, pending.getMin(), pending.getMax() ) ) { - - RangeVector overlap; - getRangeMapOverlap( _pendingMap, pending.getMin(), pending.getMax(), &overlap ); - - *errMsg += stream() << " and it overlaps " << overlapToString( overlap ); - } - - warning() << *errMsg << endl; - return NULL; - } + warning() << *errMsg << endl; + return NULL; + } - unique_ptr<CollectionMetadata> metadata( new CollectionMetadata ); - metadata->_keyPattern = this->_keyPattern; - metadata->_keyPattern.getOwned(); - metadata->fillKeyPatternFields(); - metadata->_pendingMap = this->_pendingMap; - metadata->_pendingMap.erase( pending.getMin() ); - metadata->_chunksMap = this->_chunksMap; - metadata->_rangesMap = this->_rangesMap; - metadata->_shardVersion = _shardVersion; - metadata->_collVersion = _collVersion; - - invariant(metadata->isValid()); - return metadata.release(); - } - - CollectionMetadata* CollectionMetadata::clonePlusPending( const ChunkType& pending, - string* errMsg ) const { - // The error message string is optional. - string dummy; - if ( errMsg == NULL ) { - errMsg = &dummy; - } + unique_ptr<CollectionMetadata> metadata(new CollectionMetadata); + metadata->_keyPattern = this->_keyPattern; + metadata->_keyPattern.getOwned(); + metadata->fillKeyPatternFields(); + metadata->_pendingMap = this->_pendingMap; + metadata->_pendingMap.erase(pending.getMin()); + metadata->_chunksMap = this->_chunksMap; + metadata->_rangesMap = this->_rangesMap; + metadata->_shardVersion = _shardVersion; + metadata->_collVersion = _collVersion; + + invariant(metadata->isValid()); + return metadata.release(); +} + +CollectionMetadata* CollectionMetadata::clonePlusPending(const ChunkType& pending, + string* errMsg) const { + // The error message string is optional. + string dummy; + if (errMsg == NULL) { + errMsg = &dummy; + } - if ( rangeMapOverlaps( _chunksMap, pending.getMin(), pending.getMax() ) ) { + if (rangeMapOverlaps(_chunksMap, pending.getMin(), pending.getMax())) { + RangeVector overlap; + getRangeMapOverlap(_chunksMap, pending.getMin(), pending.getMax(), &overlap); - RangeVector overlap; - getRangeMapOverlap( _chunksMap, pending.getMin(), pending.getMax(), &overlap ); + *errMsg = stream() << "cannot add pending chunk " + << rangeToString(pending.getMin(), pending.getMax()) + << " because the chunk overlaps " << overlapToString(overlap); - *errMsg = stream() << "cannot add pending chunk " - << rangeToString( pending.getMin(), pending.getMax() ) - << " because the chunk overlaps " << overlapToString( overlap ); + warning() << *errMsg << endl; + return NULL; + } - warning() << *errMsg << endl; - return NULL; + unique_ptr<CollectionMetadata> metadata(new CollectionMetadata); + metadata->_keyPattern = this->_keyPattern; + metadata->_keyPattern.getOwned(); + metadata->fillKeyPatternFields(); + metadata->_pendingMap = this->_pendingMap; + metadata->_chunksMap = this->_chunksMap; + metadata->_rangesMap = this->_rangesMap; + metadata->_shardVersion = _shardVersion; + metadata->_collVersion = _collVersion; + + // If there are any pending chunks on the interval to be added this is ok, since pending + // chunks aren't officially tracked yet and something may have changed on servers we do not + // see yet. + // We remove any chunks we overlap, the remote request starting a chunk migration must have + // been authoritative. + + if (rangeMapOverlaps(_pendingMap, pending.getMin(), pending.getMax())) { + RangeVector pendingOverlap; + getRangeMapOverlap(_pendingMap, pending.getMin(), pending.getMax(), &pendingOverlap); + + warning() << "new pending chunk " << rangeToString(pending.getMin(), pending.getMax()) + << " overlaps existing pending chunks " << overlapToString(pendingOverlap) + << ", a migration may not have completed" << endl; + + for (RangeVector::iterator it = pendingOverlap.begin(); it != pendingOverlap.end(); ++it) { + metadata->_pendingMap.erase(it->first); } + } - unique_ptr<CollectionMetadata> metadata( new CollectionMetadata ); - metadata->_keyPattern = this->_keyPattern; - metadata->_keyPattern.getOwned(); - metadata->fillKeyPatternFields(); - metadata->_pendingMap = this->_pendingMap; - metadata->_chunksMap = this->_chunksMap; - metadata->_rangesMap = this->_rangesMap; - metadata->_shardVersion = _shardVersion; - metadata->_collVersion = _collVersion; - - // If there are any pending chunks on the interval to be added this is ok, since pending - // chunks aren't officially tracked yet and something may have changed on servers we do not - // see yet. - // We remove any chunks we overlap, the remote request starting a chunk migration must have - // been authoritative. - - if ( rangeMapOverlaps( _pendingMap, pending.getMin(), pending.getMax() ) ) { - - RangeVector pendingOverlap; - getRangeMapOverlap( _pendingMap, pending.getMin(), pending.getMax(), &pendingOverlap ); - - warning() << "new pending chunk " << rangeToString( pending.getMin(), pending.getMax() ) - << " overlaps existing pending chunks " << overlapToString( pendingOverlap ) - << ", a migration may not have completed" << endl; - - for ( RangeVector::iterator it = pendingOverlap.begin(); it != pendingOverlap.end(); - ++it ) { - metadata->_pendingMap.erase( it->first ); - } - } + metadata->_pendingMap.insert(make_pair(pending.getMin(), pending.getMax())); - metadata->_pendingMap.insert( make_pair( pending.getMin(), pending.getMax() ) ); + invariant(metadata->isValid()); + return metadata.release(); +} - invariant(metadata->isValid()); - return metadata.release(); +CollectionMetadata* CollectionMetadata::cloneSplit(const ChunkType& chunk, + const vector<BSONObj>& splitKeys, + const ChunkVersion& newShardVersion, + string* errMsg) const { + // The error message string is optional. + string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - CollectionMetadata* CollectionMetadata::cloneSplit( const ChunkType& chunk, - const vector<BSONObj>& splitKeys, - const ChunkVersion& newShardVersion, - string* errMsg ) const { - // The error message string is optional. - string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } + // The version required in both resulting chunks could be simply an increment in the + // minor portion of the current version. However, we are enforcing uniqueness over the + // attributes <ns, version> of the configdb collection 'chunks'. So in practice, a + // migrate somewhere may force this split to pick up a version that has the major + // portion higher than the one that this shard has been using. + // + // TODO drop the uniqueness constraint and tighten the check below so that only the + // minor portion of version changes + if (newShardVersion <= _shardVersion) { + *errMsg = stream() << "cannot split chunk " << rangeToString(chunk.getMin(), chunk.getMax()) + << ", new shard version " << newShardVersion.toString() + << " is not greater than current version " << _shardVersion.toString(); + + warning() << *errMsg << endl; + return NULL; + } - // The version required in both resulting chunks could be simply an increment in the - // minor portion of the current version. However, we are enforcing uniqueness over the - // attributes <ns, version> of the configdb collection 'chunks'. So in practice, a - // migrate somewhere may force this split to pick up a version that has the major - // portion higher than the one that this shard has been using. - // - // TODO drop the uniqueness constraint and tighten the check below so that only the - // minor portion of version changes - if (newShardVersion <= _shardVersion) { + // Check that we have the exact chunk that will be subtracted. + if (!rangeMapContains(_chunksMap, chunk.getMin(), chunk.getMax())) { + *errMsg = stream() << "cannot split chunk " << rangeToString(chunk.getMin(), chunk.getMax()) + << ", this shard does not contain the chunk"; - *errMsg = stream() << "cannot split chunk " - << rangeToString( chunk.getMin(), chunk.getMax() ) - << ", new shard version " - << newShardVersion.toString() - << " is not greater than current version " - << _shardVersion.toString(); + if (rangeMapOverlaps(_chunksMap, chunk.getMin(), chunk.getMax())) { + RangeVector overlap; + getRangeMapOverlap(_chunksMap, chunk.getMin(), chunk.getMax(), &overlap); - warning() << *errMsg << endl; - return NULL; + *errMsg += stream() << " and it overlaps " << overlapToString(overlap); } - // Check that we have the exact chunk that will be subtracted. - if ( !rangeMapContains( _chunksMap, chunk.getMin(), chunk.getMax() ) ) { + warning() << *errMsg << endl; + return NULL; + } + // Check that the split key is valid + for (vector<BSONObj>::const_iterator it = splitKeys.begin(); it != splitKeys.end(); ++it) { + if (!rangeContains(chunk.getMin(), chunk.getMax(), *it)) { *errMsg = stream() << "cannot split chunk " - << rangeToString( chunk.getMin(), chunk.getMax() ) - << ", this shard does not contain the chunk"; - - if ( rangeMapOverlaps( _chunksMap, chunk.getMin(), chunk.getMax() ) ) { - - RangeVector overlap; - getRangeMapOverlap( _chunksMap, chunk.getMin(), chunk.getMax(), &overlap ); - - *errMsg += stream() << " and it overlaps " << overlapToString( overlap ); - } + << rangeToString(chunk.getMin(), chunk.getMax()) << " at key " + << *it; warning() << *errMsg << endl; return NULL; } - - // Check that the split key is valid - for ( vector<BSONObj>::const_iterator it = splitKeys.begin(); it != splitKeys.end(); ++it ) - { - if (!rangeContains(chunk.getMin(), chunk.getMax(), *it)) { - - *errMsg = stream() << "cannot split chunk " - << rangeToString( chunk.getMin(), chunk.getMax() ) << " at key " - << *it; - - warning() << *errMsg << endl; - return NULL; - } - } - - unique_ptr<CollectionMetadata> metadata(new CollectionMetadata); - metadata->_keyPattern = this->_keyPattern; - metadata->_keyPattern.getOwned(); - metadata->fillKeyPatternFields(); - metadata->_pendingMap = this->_pendingMap; - metadata->_chunksMap = this->_chunksMap; - metadata->_shardVersion = newShardVersion; // will increment 2nd, 3rd,... chunks below - - BSONObj startKey = chunk.getMin(); - for ( vector<BSONObj>::const_iterator it = splitKeys.begin(); it != splitKeys.end(); - ++it ) { - BSONObj split = *it; - invariant(split.woCompare(startKey) > 0); - metadata->_chunksMap[startKey] = split.getOwned(); - metadata->_chunksMap.insert( make_pair( split.getOwned(), chunk.getMax().getOwned() )); - metadata->_shardVersion.incMinor(); - startKey = split; - } - - metadata->_collVersion = - metadata->_shardVersion > _collVersion ? metadata->_shardVersion : _collVersion; - metadata->fillRanges(); - - invariant(metadata->isValid()); - return metadata.release(); } - CollectionMetadata* CollectionMetadata::cloneMerge( const BSONObj& minKey, - const BSONObj& maxKey, - const ChunkVersion& newShardVersion, - string* errMsg ) const { - - if (newShardVersion <= _shardVersion) { - - *errMsg = stream() << "cannot merge range " << rangeToString( minKey, maxKey ) - << ", new shard version " << newShardVersion.toString() - << " is not greater than current version " - << _shardVersion.toString(); - - warning() << *errMsg << endl; - return NULL; - } + unique_ptr<CollectionMetadata> metadata(new CollectionMetadata); + metadata->_keyPattern = this->_keyPattern; + metadata->_keyPattern.getOwned(); + metadata->fillKeyPatternFields(); + metadata->_pendingMap = this->_pendingMap; + metadata->_chunksMap = this->_chunksMap; + metadata->_shardVersion = newShardVersion; // will increment 2nd, 3rd,... chunks below + + BSONObj startKey = chunk.getMin(); + for (vector<BSONObj>::const_iterator it = splitKeys.begin(); it != splitKeys.end(); ++it) { + BSONObj split = *it; + invariant(split.woCompare(startKey) > 0); + metadata->_chunksMap[startKey] = split.getOwned(); + metadata->_chunksMap.insert(make_pair(split.getOwned(), chunk.getMax().getOwned())); + metadata->_shardVersion.incMinor(); + startKey = split; + } - RangeVector overlap; - getRangeMapOverlap( _chunksMap, minKey, maxKey, &overlap ); + metadata->_collVersion = + metadata->_shardVersion > _collVersion ? metadata->_shardVersion : _collVersion; + metadata->fillRanges(); + + invariant(metadata->isValid()); + return metadata.release(); +} + +CollectionMetadata* CollectionMetadata::cloneMerge(const BSONObj& minKey, + const BSONObj& maxKey, + const ChunkVersion& newShardVersion, + string* errMsg) const { + if (newShardVersion <= _shardVersion) { + *errMsg = stream() << "cannot merge range " << rangeToString(minKey, maxKey) + << ", new shard version " << newShardVersion.toString() + << " is not greater than current version " << _shardVersion.toString(); + + warning() << *errMsg << endl; + return NULL; + } - if ( overlap.empty() || overlap.size() == 1 ) { + RangeVector overlap; + getRangeMapOverlap(_chunksMap, minKey, maxKey, &overlap); - *errMsg = stream() << "cannot merge range " << rangeToString( minKey, maxKey ) - << ( overlap.empty() ? ", no chunks found in this range" : - ", only one chunk found in this range" ); + if (overlap.empty() || overlap.size() == 1) { + *errMsg = stream() << "cannot merge range " << rangeToString(minKey, maxKey) + << (overlap.empty() ? ", no chunks found in this range" + : ", only one chunk found in this range"); - warning() << *errMsg << endl; - return NULL; - } + warning() << *errMsg << endl; + return NULL; + } - bool validStartEnd = true; - bool validNoHoles = true; - if ( overlap.begin()->first.woCompare( minKey ) != 0 ) { - // First chunk doesn't start with minKey - validStartEnd = false; - } - else if ( overlap.rbegin()->second.woCompare( maxKey ) != 0 ) { - // Last chunk doesn't end with maxKey - validStartEnd = false; - } - else { - // Check that there are no holes - BSONObj prevMaxKey = minKey; - for ( RangeVector::iterator it = overlap.begin(); it != overlap.end(); ++it ) { - if ( it->first.woCompare( prevMaxKey ) != 0 ) { - validNoHoles = false; - break; - } - prevMaxKey = it->second; + bool validStartEnd = true; + bool validNoHoles = true; + if (overlap.begin()->first.woCompare(minKey) != 0) { + // First chunk doesn't start with minKey + validStartEnd = false; + } else if (overlap.rbegin()->second.woCompare(maxKey) != 0) { + // Last chunk doesn't end with maxKey + validStartEnd = false; + } else { + // Check that there are no holes + BSONObj prevMaxKey = minKey; + for (RangeVector::iterator it = overlap.begin(); it != overlap.end(); ++it) { + if (it->first.woCompare(prevMaxKey) != 0) { + validNoHoles = false; + break; } + prevMaxKey = it->second; } + } - if ( !validStartEnd || !validNoHoles ) { + if (!validStartEnd || !validNoHoles) { + *errMsg = stream() << "cannot merge range " << rangeToString(minKey, maxKey) + << ", overlapping chunks " << overlapToString(overlap) + << (!validStartEnd ? " do not have the same min and max key" + : " are not all adjacent"); - *errMsg = stream() << "cannot merge range " << rangeToString( minKey, maxKey ) - << ", overlapping chunks " << overlapToString( overlap ) - << ( !validStartEnd ? " do not have the same min and max key" : - " are not all adjacent" ); + warning() << *errMsg << endl; + return NULL; + } - warning() << *errMsg << endl; - return NULL; - } + unique_ptr<CollectionMetadata> metadata(new CollectionMetadata); + metadata->_keyPattern = this->_keyPattern; + metadata->_keyPattern.getOwned(); + metadata->fillKeyPatternFields(); + metadata->_pendingMap = this->_pendingMap; + metadata->_chunksMap = this->_chunksMap; + metadata->_rangesMap = this->_rangesMap; + metadata->_shardVersion = newShardVersion; + metadata->_collVersion = newShardVersion > _collVersion ? newShardVersion : this->_collVersion; + + for (RangeVector::iterator it = overlap.begin(); it != overlap.end(); ++it) { + metadata->_chunksMap.erase(it->first); + } - unique_ptr<CollectionMetadata> metadata( new CollectionMetadata ); - metadata->_keyPattern = this->_keyPattern; - metadata->_keyPattern.getOwned(); - metadata->fillKeyPatternFields(); - metadata->_pendingMap = this->_pendingMap; - metadata->_chunksMap = this->_chunksMap; - metadata->_rangesMap = this->_rangesMap; - metadata->_shardVersion = newShardVersion; - metadata->_collVersion = - newShardVersion > _collVersion ? newShardVersion : this->_collVersion; - - for ( RangeVector::iterator it = overlap.begin(); it != overlap.end(); ++it ) { - metadata->_chunksMap.erase( it->first ); - } + metadata->_chunksMap.insert(make_pair(minKey, maxKey)); - metadata->_chunksMap.insert( make_pair( minKey, maxKey ) ); + invariant(metadata->isValid()); + return metadata.release(); +} - invariant(metadata->isValid()); - return metadata.release(); +bool CollectionMetadata::keyBelongsToMe(const BSONObj& key) const { + // For now, collections don't move. So if the collection is not sharded, assume + // the document with the given key can be accessed. + if (_keyPattern.isEmpty()) { + return true; } - bool CollectionMetadata::keyBelongsToMe( const BSONObj& key ) const { - // For now, collections don't move. So if the collection is not sharded, assume - // the document with the given key can be accessed. - if ( _keyPattern.isEmpty() ) { - return true; - } - - if ( _rangesMap.size() <= 0 ) { - return false; - } + if (_rangesMap.size() <= 0) { + return false; + } - RangeMap::const_iterator it = _rangesMap.upper_bound( key ); - if ( it != _rangesMap.begin() ) it--; + RangeMap::const_iterator it = _rangesMap.upper_bound(key); + if (it != _rangesMap.begin()) + it--; - bool good = rangeContains( it->first, it->second, key ); + bool good = rangeContains(it->first, it->second, key); #if 0 // DISABLED because of SERVER-11175 - huge amount of logging @@ -477,286 +444,285 @@ namespace mongo { } #endif - return good; - } - - bool CollectionMetadata::keyIsPending( const BSONObj& key ) const { - // If we aren't sharded, then the key is never pending (though it belongs-to-me) - if ( _keyPattern.isEmpty() ) { - return false; - } - - if ( _pendingMap.size() <= 0 ) { - return false; - } - - RangeMap::const_iterator it = _pendingMap.upper_bound( key ); - if ( it != _pendingMap.begin() ) it--; + return good; +} - bool isPending = rangeContains( it->first, it->second, key ); - return isPending; +bool CollectionMetadata::keyIsPending(const BSONObj& key) const { + // If we aren't sharded, then the key is never pending (though it belongs-to-me) + if (_keyPattern.isEmpty()) { + return false; } - bool CollectionMetadata::getNextChunk( const BSONObj& lookupKey, ChunkType* chunk ) const { - RangeMap::const_iterator upperChunkIt = _chunksMap.upper_bound(lookupKey); - RangeMap::const_iterator lowerChunkIt = upperChunkIt; + if (_pendingMap.size() <= 0) { + return false; + } - if (upperChunkIt != _chunksMap.begin()) { - --lowerChunkIt; - } - else { - lowerChunkIt = _chunksMap.end(); - } + RangeMap::const_iterator it = _pendingMap.upper_bound(key); + if (it != _pendingMap.begin()) + it--; - if (lowerChunkIt != _chunksMap.end() && lowerChunkIt->second.woCompare(lookupKey) > 0) { - chunk->setMin(lowerChunkIt->first); - chunk->setMax(lowerChunkIt->second); - return true; - } + bool isPending = rangeContains(it->first, it->second, key); + return isPending; +} - if (upperChunkIt != _chunksMap.end()) { - chunk->setMin(upperChunkIt->first); - chunk->setMax(upperChunkIt->second); - return true; - } +bool CollectionMetadata::getNextChunk(const BSONObj& lookupKey, ChunkType* chunk) const { + RangeMap::const_iterator upperChunkIt = _chunksMap.upper_bound(lookupKey); + RangeMap::const_iterator lowerChunkIt = upperChunkIt; - return false; + if (upperChunkIt != _chunksMap.begin()) { + --lowerChunkIt; + } else { + lowerChunkIt = _chunksMap.end(); } - BSONObj CollectionMetadata::toBSON() const { - BSONObjBuilder bb; - toBSON( bb ); - return bb.obj(); + if (lowerChunkIt != _chunksMap.end() && lowerChunkIt->second.woCompare(lookupKey) > 0) { + chunk->setMin(lowerChunkIt->first); + chunk->setMax(lowerChunkIt->second); + return true; } - void CollectionMetadata::toBSONChunks( BSONArrayBuilder& bb ) const { - - if ( _chunksMap.empty() ) return; - - for (RangeMap::const_iterator it = _chunksMap.begin(); it != _chunksMap.end(); ++it ) { - BSONArrayBuilder chunkBB( bb.subarrayStart() ); - chunkBB.append( it->first ); - chunkBB.append( it->second ); - chunkBB.done(); - } + if (upperChunkIt != _chunksMap.end()) { + chunk->setMin(upperChunkIt->first); + chunk->setMax(upperChunkIt->second); + return true; } - void CollectionMetadata::toBSONPending( BSONArrayBuilder& bb ) const { + return false; +} - if ( _pendingMap.empty() ) return; - - for (RangeMap::const_iterator it = _pendingMap.begin(); it != _pendingMap.end(); ++it ) { - BSONArrayBuilder pendingBB( bb.subarrayStart() ); - pendingBB.append( it->first ); - pendingBB.append( it->second ); - pendingBB.done(); - } - } +BSONObj CollectionMetadata::toBSON() const { + BSONObjBuilder bb; + toBSON(bb); + return bb.obj(); +} - void CollectionMetadata::toBSON( BSONObjBuilder& bb ) const { +void CollectionMetadata::toBSONChunks(BSONArrayBuilder& bb) const { + if (_chunksMap.empty()) + return; - _collVersion.addToBSON( bb, "collVersion" ); - _shardVersion.addToBSON( bb, "shardVersion" ); - bb.append( "keyPattern", _keyPattern ); + for (RangeMap::const_iterator it = _chunksMap.begin(); it != _chunksMap.end(); ++it) { + BSONArrayBuilder chunkBB(bb.subarrayStart()); + chunkBB.append(it->first); + chunkBB.append(it->second); + chunkBB.done(); + } +} - BSONArrayBuilder chunksBB( bb.subarrayStart( "chunks" ) ); - toBSONChunks( chunksBB ); - chunksBB.done(); +void CollectionMetadata::toBSONPending(BSONArrayBuilder& bb) const { + if (_pendingMap.empty()) + return; - BSONArrayBuilder pendingBB( bb.subarrayStart( "pending" ) ); - toBSONPending( pendingBB ); + for (RangeMap::const_iterator it = _pendingMap.begin(); it != _pendingMap.end(); ++it) { + BSONArrayBuilder pendingBB(bb.subarrayStart()); + pendingBB.append(it->first); + pendingBB.append(it->second); pendingBB.done(); } +} - bool CollectionMetadata::getNextOrphanRange( const BSONObj& origLookupKey, - KeyRange* range ) const { +void CollectionMetadata::toBSON(BSONObjBuilder& bb) const { + _collVersion.addToBSON(bb, "collVersion"); + _shardVersion.addToBSON(bb, "shardVersion"); + bb.append("keyPattern", _keyPattern); - if ( _keyPattern.isEmpty() ) return false; + BSONArrayBuilder chunksBB(bb.subarrayStart("chunks")); + toBSONChunks(chunksBB); + chunksBB.done(); - BSONObj lookupKey = origLookupKey; - BSONObj maxKey = getMaxKey(); // so we don't keep rebuilding - while ( lookupKey.woCompare( maxKey ) < 0 ) { + BSONArrayBuilder pendingBB(bb.subarrayStart("pending")); + toBSONPending(pendingBB); + pendingBB.done(); +} - RangeMap::const_iterator lowerChunkIt = _chunksMap.end(); - RangeMap::const_iterator upperChunkIt = _chunksMap.end(); +bool CollectionMetadata::getNextOrphanRange(const BSONObj& origLookupKey, KeyRange* range) const { + if (_keyPattern.isEmpty()) + return false; - if ( !_chunksMap.empty() ) { - upperChunkIt = _chunksMap.upper_bound( lookupKey ); - lowerChunkIt = upperChunkIt; - if ( upperChunkIt != _chunksMap.begin() ) --lowerChunkIt; - else lowerChunkIt = _chunksMap.end(); - } + BSONObj lookupKey = origLookupKey; + BSONObj maxKey = getMaxKey(); // so we don't keep rebuilding + while (lookupKey.woCompare(maxKey) < 0) { + RangeMap::const_iterator lowerChunkIt = _chunksMap.end(); + RangeMap::const_iterator upperChunkIt = _chunksMap.end(); - // If we overlap, continue after the overlap - // TODO: Could optimize slightly by finding next non-contiguous chunk - if ( lowerChunkIt != _chunksMap.end() - && lowerChunkIt->second.woCompare( lookupKey ) > 0 ) { - lookupKey = lowerChunkIt->second; - continue; - } - - RangeMap::const_iterator lowerPendingIt = _pendingMap.end(); - RangeMap::const_iterator upperPendingIt = _pendingMap.end(); + if (!_chunksMap.empty()) { + upperChunkIt = _chunksMap.upper_bound(lookupKey); + lowerChunkIt = upperChunkIt; + if (upperChunkIt != _chunksMap.begin()) + --lowerChunkIt; + else + lowerChunkIt = _chunksMap.end(); + } - if ( !_pendingMap.empty() ) { + // If we overlap, continue after the overlap + // TODO: Could optimize slightly by finding next non-contiguous chunk + if (lowerChunkIt != _chunksMap.end() && lowerChunkIt->second.woCompare(lookupKey) > 0) { + lookupKey = lowerChunkIt->second; + continue; + } - upperPendingIt = _pendingMap.upper_bound( lookupKey ); - lowerPendingIt = upperPendingIt; - if ( upperPendingIt != _pendingMap.begin() ) --lowerPendingIt; - else lowerPendingIt = _pendingMap.end(); - } + RangeMap::const_iterator lowerPendingIt = _pendingMap.end(); + RangeMap::const_iterator upperPendingIt = _pendingMap.end(); - // If we overlap, continue after the overlap - // TODO: Could optimize slightly by finding next non-contiguous chunk - if ( lowerPendingIt != _pendingMap.end() - && lowerPendingIt->second.woCompare( lookupKey ) > 0 ) { - lookupKey = lowerPendingIt->second; - continue; - } + if (!_pendingMap.empty()) { + upperPendingIt = _pendingMap.upper_bound(lookupKey); + lowerPendingIt = upperPendingIt; + if (upperPendingIt != _pendingMap.begin()) + --lowerPendingIt; + else + lowerPendingIt = _pendingMap.end(); + } - // - // We know that the lookup key is not covered by a chunk or pending range, and where the - // previous chunk and pending chunks are. Now we fill in the bounds as the closest - // bounds of the surrounding ranges in both maps. - // + // If we overlap, continue after the overlap + // TODO: Could optimize slightly by finding next non-contiguous chunk + if (lowerPendingIt != _pendingMap.end() && + lowerPendingIt->second.woCompare(lookupKey) > 0) { + lookupKey = lowerPendingIt->second; + continue; + } - range->keyPattern = _keyPattern; - range->minKey = getMinKey(); - range->maxKey = maxKey; + // + // We know that the lookup key is not covered by a chunk or pending range, and where the + // previous chunk and pending chunks are. Now we fill in the bounds as the closest + // bounds of the surrounding ranges in both maps. + // - if ( lowerChunkIt != _chunksMap.end() - && lowerChunkIt->second.woCompare( range->minKey ) > 0 ) { - range->minKey = lowerChunkIt->second; - } + range->keyPattern = _keyPattern; + range->minKey = getMinKey(); + range->maxKey = maxKey; - if ( upperChunkIt != _chunksMap.end() - && upperChunkIt->first.woCompare( range->maxKey ) < 0 ) { - range->maxKey = upperChunkIt->first; - } + if (lowerChunkIt != _chunksMap.end() && lowerChunkIt->second.woCompare(range->minKey) > 0) { + range->minKey = lowerChunkIt->second; + } - if ( lowerPendingIt != _pendingMap.end() - && lowerPendingIt->second.woCompare( range->minKey ) > 0 ) { - range->minKey = lowerPendingIt->second; - } + if (upperChunkIt != _chunksMap.end() && upperChunkIt->first.woCompare(range->maxKey) < 0) { + range->maxKey = upperChunkIt->first; + } - if ( upperPendingIt != _pendingMap.end() - && upperPendingIt->first.woCompare( range->maxKey ) < 0 ) { - range->maxKey = upperPendingIt->first; - } + if (lowerPendingIt != _pendingMap.end() && + lowerPendingIt->second.woCompare(range->minKey) > 0) { + range->minKey = lowerPendingIt->second; + } - return true; + if (upperPendingIt != _pendingMap.end() && + upperPendingIt->first.woCompare(range->maxKey) < 0) { + range->maxKey = upperPendingIt->first; } - return false; + return true; } - string CollectionMetadata::toString() const { - StringBuilder ss; - ss << " CollectionManager version: " << _shardVersion.toString() << " key: " << _keyPattern; - if (_rangesMap.empty()) { - return ss.str(); - } + return false; +} - RangeMap::const_iterator it = _rangesMap.begin(); - ss << it->first << " -> " << it->second; - while (it != _rangesMap.end()) { - ss << ", "<< it->first << " -> " << it->second; - } +string CollectionMetadata::toString() const { + StringBuilder ss; + ss << " CollectionManager version: " << _shardVersion.toString() << " key: " << _keyPattern; + if (_rangesMap.empty()) { return ss.str(); } - BSONObj CollectionMetadata::getMinKey() const { - BSONObjIterator it( _keyPattern ); - BSONObjBuilder minKeyB; - while ( it.more() ) minKeyB << it.next().fieldName() << MINKEY; - return minKeyB.obj(); - } - - BSONObj CollectionMetadata::getMaxKey() const { - BSONObjIterator it( _keyPattern ); - BSONObjBuilder maxKeyB; - while ( it.more() ) maxKeyB << it.next().fieldName() << MAXKEY; - return maxKeyB.obj(); + RangeMap::const_iterator it = _rangesMap.begin(); + ss << it->first << " -> " << it->second; + while (it != _rangesMap.end()) { + ss << ", " << it->first << " -> " << it->second; } + return ss.str(); +} + +BSONObj CollectionMetadata::getMinKey() const { + BSONObjIterator it(_keyPattern); + BSONObjBuilder minKeyB; + while (it.more()) + minKeyB << it.next().fieldName() << MINKEY; + return minKeyB.obj(); +} + +BSONObj CollectionMetadata::getMaxKey() const { + BSONObjIterator it(_keyPattern); + BSONObjBuilder maxKeyB; + while (it.more()) + maxKeyB << it.next().fieldName() << MAXKEY; + return maxKeyB.obj(); +} + +bool CollectionMetadata::isValid() const { + if (_shardVersion > _collVersion) + return false; + if (_collVersion.majorVersion() == 0) + return false; + if (_collVersion.epoch() != _shardVersion.epoch()) + return false; - bool CollectionMetadata::isValid() const { - if (_shardVersion > _collVersion) + if (_shardVersion.majorVersion() > 0) { + // Must be chunks + if (_rangesMap.size() == 0 || _chunksMap.size() == 0) return false; - if (_collVersion.majorVersion() == 0) + } else { + // No chunks + if (_shardVersion.minorVersion() > 0) return false; - if (_collVersion.epoch() != _shardVersion.epoch()) + if (_rangesMap.size() > 0 || _chunksMap.size() > 0) return false; - - if (_shardVersion.majorVersion() > 0) { - // Must be chunks - if (_rangesMap.size() == 0 || _chunksMap.size() == 0) - return false; - } - else { - // No chunks - if (_shardVersion.minorVersion() > 0) - return false; - if (_rangesMap.size() > 0 || _chunksMap.size() > 0) - return false; - } - - return true; } - bool CollectionMetadata::isValidKey( const BSONObj& key ) const { - BSONObjIterator it( _keyPattern ); - while ( it.more() ) { - BSONElement next = it.next(); - if ( !key.hasField( next.fieldName() ) ) return false; - } - return key.nFields() == _keyPattern.nFields(); - } - - void CollectionMetadata::fillRanges() { - if (_chunksMap.empty()) - return; - - // Load the chunk information, coallesceing their ranges. The version for this shard - // would be the highest version for any of the chunks. - RangeMap::const_iterator it = _chunksMap.begin(); - BSONObj min,max; - while (it != _chunksMap.end()) { - BSONObj currMin = it->first; - BSONObj currMax = it->second; - ++it; - - // coalesce the chunk's bounds in ranges if they are adjacent chunks - if (min.isEmpty()) { - min = currMin; - max = currMax; - continue; - } - if (max == currMin) { - max = currMax; - continue; - } - - _rangesMap.insert(make_pair(min, max)); + return true; +} +bool CollectionMetadata::isValidKey(const BSONObj& key) const { + BSONObjIterator it(_keyPattern); + while (it.more()) { + BSONElement next = it.next(); + if (!key.hasField(next.fieldName())) + return false; + } + return key.nFields() == _keyPattern.nFields(); +} + +void CollectionMetadata::fillRanges() { + if (_chunksMap.empty()) + return; + + // Load the chunk information, coallesceing their ranges. The version for this shard + // would be the highest version for any of the chunks. + RangeMap::const_iterator it = _chunksMap.begin(); + BSONObj min, max; + while (it != _chunksMap.end()) { + BSONObj currMin = it->first; + BSONObj currMax = it->second; + ++it; + + // coalesce the chunk's bounds in ranges if they are adjacent chunks + if (min.isEmpty()) { min = currMin; max = currMax; + continue; + } + if (max == currMin) { + max = currMax; + continue; } - dassert(!min.isEmpty()); _rangesMap.insert(make_pair(min, max)); + + min = currMin; + max = currMax; } + dassert(!min.isEmpty()); - void CollectionMetadata::fillKeyPatternFields() { - // Parse the shard keys into the states 'keys' and 'keySet' members. - BSONObjIterator patternIter = _keyPattern.begin(); - while (patternIter.more()) { - BSONElement current = patternIter.next(); + _rangesMap.insert(make_pair(min, max)); +} - _keyFields.mutableVector().push_back(new FieldRef); - FieldRef* const newFieldRef = _keyFields.mutableVector().back(); - newFieldRef->parse(current.fieldNameStringData()); - } +void CollectionMetadata::fillKeyPatternFields() { + // Parse the shard keys into the states 'keys' and 'keySet' members. + BSONObjIterator patternIter = _keyPattern.begin(); + while (patternIter.more()) { + BSONElement current = patternIter.next(); + + _keyFields.mutableVector().push_back(new FieldRef); + FieldRef* const newFieldRef = _keyFields.mutableVector().back(); + newFieldRef->parse(current.fieldNameStringData()); } +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/collection_metadata.h b/src/mongo/s/collection_metadata.h index b7f30c6a760..5f3ae419dbd 100644 --- a/src/mongo/s/collection_metadata.h +++ b/src/mongo/s/collection_metadata.h @@ -39,280 +39,280 @@ namespace mongo { - class MetadataLoader; - class CollectionMetadata; +class MetadataLoader; +class CollectionMetadata; - typedef std::shared_ptr<const CollectionMetadata> CollectionMetadataPtr; +typedef std::shared_ptr<const CollectionMetadata> CollectionMetadataPtr; + +/** + * The collection metadata has metadata information about a collection, in particular the + * sharding information. It's main goal in life is to be capable of answering if a certain + * document belongs to it or not. (In some scenarios such as chunk migration, a given + * document is in a shard but cannot be accessed.) + * + * To build a collection from config data, please check the MetadataLoader. The methods + * here allow building a new incarnation of a collection's metadata based on an existing + * one (e.g, we're splitting in a given collection.). + * + * This class is immutable once constructed. + */ +class CollectionMetadata { + MONGO_DISALLOW_COPYING(CollectionMetadata); + +public: + ~CollectionMetadata(); + + // + // cloning support + // + + /** + * Returns a new metadata's instance based on 'this's state by removing a 'pending' chunk. + * + * The shard and collection version of the new metadata are unaffected. The caller owns the + * new metadata. + * + * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was + * provided. + */ + CollectionMetadata* cloneMinusPending(const ChunkType& pending, std::string* errMsg) const; + + /** + * Returns a new metadata's instance based on 'this's state by adding a 'pending' chunk. + * + * The shard and collection version of the new metadata are unaffected. The caller owns the + * new metadata. + * + * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was + * provided. + */ + CollectionMetadata* clonePlusPending(const ChunkType& pending, std::string* errMsg) const; + + /** + * Returns a new metadata's instance based on 'this's state by removing 'chunk'. + * When cloning away the last chunk, 'newShardVersion' must be zero. In any case, + * the caller owns the new metadata when the cloning is successful. + * + * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was + * provided. + */ + CollectionMetadata* cloneMigrate(const ChunkType& chunk, + const ChunkVersion& newShardVersion, + std::string* errMsg) const; + + /** + * Returns a new metadata's instance by splitting an existing 'chunk' at the points + * described by 'splitKeys'. The first resulting chunk will have 'newShardVersion' and + * subsequent one would have that with the minor version incremented at each chunk. The + * caller owns the metadata. + * + * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was + * provided. + * + * Note: 'splitKeys' must be sorted in ascending order. + */ + CollectionMetadata* cloneSplit(const ChunkType& chunk, + const std::vector<BSONObj>& splitKeys, + const ChunkVersion& newShardVersion, + std::string* errMsg) const; /** - * The collection metadata has metadata information about a collection, in particular the - * sharding information. It's main goal in life is to be capable of answering if a certain - * document belongs to it or not. (In some scenarios such as chunk migration, a given - * document is in a shard but cannot be accessed.) + * Returns a new metadata instance by merging a key range which starts and ends at existing + * chunks into a single chunk. The range may not have holes. The resulting metadata will + * have the 'newShardVersion'. The caller owns the new metadata. * - * To build a collection from config data, please check the MetadataLoader. The methods - * here allow building a new incarnation of a collection's metadata based on an existing - * one (e.g, we're splitting in a given collection.). + * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was + * provided. + */ + CollectionMetadata* cloneMerge(const BSONObj& minKey, + const BSONObj& maxKey, + const ChunkVersion& newShardVersion, + std::string* errMsg) const; + + // + // verification logic + // + + /** + * Returns true if the document key 'key' is a valid instance of a shard key for this + * metadata. The 'key' must contain exactly the same fields as the shard key pattern. + */ + bool isValidKey(const BSONObj& key) const; + + /** + * Returns true if the document key 'key' belongs to this chunkset. Recall that documents of + * an in-flight chunk migration may be present and should not be considered part of the + * collection / chunkset yet. Key must be the full shard key. + */ + bool keyBelongsToMe(const BSONObj& key) const; + + /** + * Returns true if the document key 'key' is or has been migrated to this shard, and may + * belong to us after a subsequent config reload. Key must be the full shard key. + */ + bool keyIsPending(const BSONObj& key) const; + + /** + * Given a key 'lookupKey' in the shard key range, get the next chunk which overlaps or is + * greater than this key. Returns true if a chunk exists, false otherwise. + * + * Passing a key that is not a valid shard key for this range results in undefined behavior. + */ + bool getNextChunk(const BSONObj& lookupKey, ChunkType* chunk) const; + + /** + * Given a key in the shard key range, get the next range which overlaps or is greater than + * this key. + * + * This allows us to do the following to iterate over all orphan ranges: * - * This class is immutable once constructed. + * KeyRange range; + * BSONObj lookupKey = metadata->getMinKey(); + * while( metadata->getNextOrphanRange( lookupKey, &orphanRange ) ) { + * // Do stuff with range + * lookupKey = orphanRange.maxKey; + * } + * + * @param lookupKey passing a key that does not belong to this metadata is undefined. + * @param orphanRange the output range. Note that the NS is not set. */ - class CollectionMetadata { - MONGO_DISALLOW_COPYING(CollectionMetadata); - public: - - ~CollectionMetadata(); - - // - // cloning support - // - - /** - * Returns a new metadata's instance based on 'this's state by removing a 'pending' chunk. - * - * The shard and collection version of the new metadata are unaffected. The caller owns the - * new metadata. - * - * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was - * provided. - */ - CollectionMetadata* cloneMinusPending( const ChunkType& pending, std::string* errMsg ) const; - - /** - * Returns a new metadata's instance based on 'this's state by adding a 'pending' chunk. - * - * The shard and collection version of the new metadata are unaffected. The caller owns the - * new metadata. - * - * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was - * provided. - */ - CollectionMetadata* clonePlusPending( const ChunkType& pending, std::string* errMsg ) const; - - /** - * Returns a new metadata's instance based on 'this's state by removing 'chunk'. - * When cloning away the last chunk, 'newShardVersion' must be zero. In any case, - * the caller owns the new metadata when the cloning is successful. - * - * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was - * provided. - */ - CollectionMetadata* cloneMigrate( const ChunkType& chunk, - const ChunkVersion& newShardVersion, - std::string* errMsg ) const; - - /** - * Returns a new metadata's instance by splitting an existing 'chunk' at the points - * described by 'splitKeys'. The first resulting chunk will have 'newShardVersion' and - * subsequent one would have that with the minor version incremented at each chunk. The - * caller owns the metadata. - * - * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was - * provided. - * - * Note: 'splitKeys' must be sorted in ascending order. - */ - CollectionMetadata* cloneSplit( const ChunkType& chunk, - const std::vector<BSONObj>& splitKeys, - const ChunkVersion& newShardVersion, - std::string* errMsg ) const; - - /** - * Returns a new metadata instance by merging a key range which starts and ends at existing - * chunks into a single chunk. The range may not have holes. The resulting metadata will - * have the 'newShardVersion'. The caller owns the new metadata. - * - * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was - * provided. - */ - CollectionMetadata* cloneMerge( const BSONObj& minKey, - const BSONObj& maxKey, - const ChunkVersion& newShardVersion, - std::string* errMsg ) const; - - // - // verification logic - // - - /** - * Returns true if the document key 'key' is a valid instance of a shard key for this - * metadata. The 'key' must contain exactly the same fields as the shard key pattern. - */ - bool isValidKey( const BSONObj& key ) const; - - /** - * Returns true if the document key 'key' belongs to this chunkset. Recall that documents of - * an in-flight chunk migration may be present and should not be considered part of the - * collection / chunkset yet. Key must be the full shard key. - */ - bool keyBelongsToMe( const BSONObj& key ) const; - - /** - * Returns true if the document key 'key' is or has been migrated to this shard, and may - * belong to us after a subsequent config reload. Key must be the full shard key. - */ - bool keyIsPending( const BSONObj& key ) const; - - /** - * Given a key 'lookupKey' in the shard key range, get the next chunk which overlaps or is - * greater than this key. Returns true if a chunk exists, false otherwise. - * - * Passing a key that is not a valid shard key for this range results in undefined behavior. - */ - bool getNextChunk( const BSONObj& lookupKey, ChunkType* chunk ) const; - - /** - * Given a key in the shard key range, get the next range which overlaps or is greater than - * this key. - * - * This allows us to do the following to iterate over all orphan ranges: - * - * KeyRange range; - * BSONObj lookupKey = metadata->getMinKey(); - * while( metadata->getNextOrphanRange( lookupKey, &orphanRange ) ) { - * // Do stuff with range - * lookupKey = orphanRange.maxKey; - * } - * - * @param lookupKey passing a key that does not belong to this metadata is undefined. - * @param orphanRange the output range. Note that the NS is not set. - */ - bool getNextOrphanRange( const BSONObj& lookupKey, KeyRange* orphanRange ) const; - - // - // accessors - // - - ChunkVersion getCollVersion() const { - return _collVersion; - } - - ChunkVersion getShardVersion() const { - return _shardVersion; - } - - BSONObj getKeyPattern() const { - return _keyPattern; - } - - const std::vector<FieldRef*>& getKeyPatternFields() const { - return _keyFields.vector(); - } - - BSONObj getMinKey() const; - - BSONObj getMaxKey() const; - - std::size_t getNumChunks() const { - return _chunksMap.size(); - } - - std::size_t getNumPending() const { - return _pendingMap.size(); - } - - // - // reporting - // - - /** - * BSON output of the metadata information. - */ - BSONObj toBSON() const; - - /** - * BSON output of the metadata information, into a builder. - */ - void toBSON( BSONObjBuilder& bb ) const; - - /** - * BSON output of the chunks metadata into a BSONArray - */ - void toBSONChunks( BSONArrayBuilder& bb ) const; - - /** - * BSON output of the pending metadata into a BSONArray - */ - void toBSONPending( BSONArrayBuilder& bb ) const; - - /** - * std::string output of the metadata information. - */ - std::string toString() const; - - /** - * Use the MetadataLoader to fill the empty metadata from the config server, or use - * clone*() methods to use existing metadatas to build new ones. - * - * Unless you are the MetadataLoader or a test you should probably not be using this - * directly. - */ - CollectionMetadata(); - - /** - * TESTING ONLY - * - * Returns a new metadata's instance based on 'this's state by adding 'chunk'. The new - * metadata can never be zero, though (see cloneMinus). The caller owns the new metadata. - * - * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was - * provided. - */ - CollectionMetadata* clonePlusChunk( const ChunkType& chunk, - const ChunkVersion& newShardVersion, - std::string* errMsg ) const; - - private: - // Effectively, the MetadataLoader is this class's builder. So we open an exception - // and grant it friendship. - friend class MetadataLoader; - - // a version for this collection that identifies the collection incarnation (ie, a - // dropped and recreated collection with the same name would have a different version) - ChunkVersion _collVersion; - - // - // sharded state below, for when the collection gets sharded - // - - // highest ChunkVersion for which this metadata's information is accurate - ChunkVersion _shardVersion; - - // key pattern for chunks under this range - BSONObj _keyPattern; - - // A vector owning the FieldRefs parsed from the shard-key pattern of field names. - OwnedPointerVector<FieldRef> _keyFields; - - // - // RangeMaps represent chunks by mapping the min key to the chunk's max key, allowing - // efficient lookup and intersection. - // - - // Map of ranges of chunks that are migrating but have not been confirmed added yet - RangeMap _pendingMap; - - // Map of chunks tracked by this shard - RangeMap _chunksMap; - - // A second map from a min key into a range or contiguous chunks. The map is redundant - // w.r.t. _chunkMap but we expect high chunk contiguity, especially in small - // installations. - RangeMap _rangesMap; - - /** - * Returns true if this metadata was loaded with all necessary information. - */ - bool isValid() const; - - /** - * Try to find chunks that are adjacent and record these intervals in the _rangesMap - */ - void fillRanges(); - - /** - * Creates the _keyField* local data - */ - void fillKeyPatternFields(); - }; - -} // namespace mongo + bool getNextOrphanRange(const BSONObj& lookupKey, KeyRange* orphanRange) const; + + // + // accessors + // + + ChunkVersion getCollVersion() const { + return _collVersion; + } + + ChunkVersion getShardVersion() const { + return _shardVersion; + } + + BSONObj getKeyPattern() const { + return _keyPattern; + } + + const std::vector<FieldRef*>& getKeyPatternFields() const { + return _keyFields.vector(); + } + + BSONObj getMinKey() const; + + BSONObj getMaxKey() const; + + std::size_t getNumChunks() const { + return _chunksMap.size(); + } + + std::size_t getNumPending() const { + return _pendingMap.size(); + } + + // + // reporting + // + + /** + * BSON output of the metadata information. + */ + BSONObj toBSON() const; + + /** + * BSON output of the metadata information, into a builder. + */ + void toBSON(BSONObjBuilder& bb) const; + + /** + * BSON output of the chunks metadata into a BSONArray + */ + void toBSONChunks(BSONArrayBuilder& bb) const; + + /** + * BSON output of the pending metadata into a BSONArray + */ + void toBSONPending(BSONArrayBuilder& bb) const; + + /** + * std::string output of the metadata information. + */ + std::string toString() const; + + /** + * Use the MetadataLoader to fill the empty metadata from the config server, or use + * clone*() methods to use existing metadatas to build new ones. + * + * Unless you are the MetadataLoader or a test you should probably not be using this + * directly. + */ + CollectionMetadata(); + + /** + * TESTING ONLY + * + * Returns a new metadata's instance based on 'this's state by adding 'chunk'. The new + * metadata can never be zero, though (see cloneMinus). The caller owns the new metadata. + * + * If a new metadata can't be created, returns NULL and fills in 'errMsg', if it was + * provided. + */ + CollectionMetadata* clonePlusChunk(const ChunkType& chunk, + const ChunkVersion& newShardVersion, + std::string* errMsg) const; + +private: + // Effectively, the MetadataLoader is this class's builder. So we open an exception + // and grant it friendship. + friend class MetadataLoader; + + // a version for this collection that identifies the collection incarnation (ie, a + // dropped and recreated collection with the same name would have a different version) + ChunkVersion _collVersion; + + // + // sharded state below, for when the collection gets sharded + // + + // highest ChunkVersion for which this metadata's information is accurate + ChunkVersion _shardVersion; + + // key pattern for chunks under this range + BSONObj _keyPattern; + + // A vector owning the FieldRefs parsed from the shard-key pattern of field names. + OwnedPointerVector<FieldRef> _keyFields; + + // + // RangeMaps represent chunks by mapping the min key to the chunk's max key, allowing + // efficient lookup and intersection. + // + + // Map of ranges of chunks that are migrating but have not been confirmed added yet + RangeMap _pendingMap; + + // Map of chunks tracked by this shard + RangeMap _chunksMap; + + // A second map from a min key into a range or contiguous chunks. The map is redundant + // w.r.t. _chunkMap but we expect high chunk contiguity, especially in small + // installations. + RangeMap _rangesMap; + + /** + * Returns true if this metadata was loaded with all necessary information. + */ + bool isValid() const; + + /** + * Try to find chunks that are adjacent and record these intervals in the _rangesMap + */ + void fillRanges(); + + /** + * Creates the _keyField* local data + */ + void fillKeyPatternFields(); +}; + +} // namespace mongo diff --git a/src/mongo/s/collection_metadata_test.cpp b/src/mongo/s/collection_metadata_test.cpp index 8774d209e03..05a44fb5c56 100644 --- a/src/mongo/s/collection_metadata_test.cpp +++ b/src/mongo/s/collection_metadata_test.cpp @@ -45,1382 +45,1277 @@ namespace { - using namespace mongo; - - using std::make_pair; - using std::string; - using std::unique_ptr; - using std::vector; - - const std::string CONFIG_HOST_PORT = "$dummy_config:27017"; - - class NoChunkFixture : public mongo::unittest::Test { - protected: - void setUp() { - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); - - OID epoch = OID::gen(); - - CollectionType collType; - collType.setNs(NamespaceString{"test.foo"}); - collType.setKeyPattern( BSON("a" << 1) ); - collType.setUnique( false ); - collType.setUpdatedAt( Date_t::fromMillisSinceEpoch(1) ); - collType.setEpoch( epoch ); - ASSERT_OK(collType.validate()); - - _dummyConfig->insert( CollectionType::ConfigNS, collType.toBSON() ); - - // Need a chunk on another shard, otherwise the chunks are invalid in general and we - // can't load metadata - ChunkType chunkType; - chunkType.setNS(NamespaceString{"test.foo"}); - chunkType.setShard( "shard0001" ); - chunkType.setMin( BSON( "a" << MINKEY ) ); - chunkType.setMax( BSON( "a" << MAXKEY ) ); - chunkType.setVersion( ChunkVersion( 1, 0, epoch ) ); - chunkType.setName( OID::gen().toString() ); - ASSERT_OK(collType.validate()); - - _dummyConfig->insert( ChunkType::ConfigNS, chunkType.toBSON() ); - - ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); - ASSERT(configLoc.isValid()); - CatalogManagerLegacy catalogManager; - catalogManager.init(configLoc); - - MetadataLoader loader; - Status status = loader.makeCollectionMetadata(&catalogManager, - "test.foo", - "shard0000", - NULL, - &_metadata); - ASSERT_OK(status); - ASSERT_EQUALS( 0u, _metadata.getNumChunks() ); - } - - void tearDown() { - MockConnRegistry::get()->clear(); - } - - const CollectionMetadata& getCollMetadata() const { - return _metadata; - } - - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - CollectionMetadata _metadata; - }; - - TEST_F(NoChunkFixture, BasicBelongsToMe) { - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << MINKEY)) ); - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << 10)) ); +using namespace mongo; + +using std::make_pair; +using std::string; +using std::unique_ptr; +using std::vector; + +const std::string CONFIG_HOST_PORT = "$dummy_config:27017"; + +class NoChunkFixture : public mongo::unittest::Test { +protected: + void setUp() { + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); + + OID epoch = OID::gen(); + + CollectionType collType; + collType.setNs(NamespaceString{"test.foo"}); + collType.setKeyPattern(BSON("a" << 1)); + collType.setUnique(false); + collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + collType.setEpoch(epoch); + ASSERT_OK(collType.validate()); + + _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); + + // Need a chunk on another shard, otherwise the chunks are invalid in general and we + // can't load metadata + ChunkType chunkType; + chunkType.setNS(NamespaceString{"test.foo"}); + chunkType.setShard("shard0001"); + chunkType.setMin(BSON("a" << MINKEY)); + chunkType.setMax(BSON("a" << MAXKEY)); + chunkType.setVersion(ChunkVersion(1, 0, epoch)); + chunkType.setName(OID::gen().toString()); + ASSERT_OK(collType.validate()); + + _dummyConfig->insert(ChunkType::ConfigNS, chunkType.toBSON()); + + ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); + ASSERT(configLoc.isValid()); + CatalogManagerLegacy catalogManager; + catalogManager.init(configLoc); + + MetadataLoader loader; + Status status = loader.makeCollectionMetadata( + &catalogManager, "test.foo", "shard0000", NULL, &_metadata); + ASSERT_OK(status); + ASSERT_EQUALS(0u, _metadata.getNumChunks()); } - TEST_F(NoChunkFixture, CompoundKeyBelongsToMe) { - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << 1 << "b" << 2)) ); + void tearDown() { + MockConnRegistry::get()->clear(); } - TEST_F(NoChunkFixture, IsKeyValid) { - ASSERT_TRUE( getCollMetadata().isValidKey( BSON( "a" << "abcde" ) ) ); - ASSERT_TRUE( getCollMetadata().isValidKey( BSON( "a" << 3 ) ) ); - ASSERT_FALSE( getCollMetadata().isValidKey( BSON( "a" << "abcde" << "b" << 1 ) ) ); - ASSERT_FALSE( getCollMetadata().isValidKey( BSON( "c" << "abcde" ) ) ); + const CollectionMetadata& getCollMetadata() const { + return _metadata; } - TEST_F(NoChunkFixture, getNextFromEmpty) { - ChunkType nextChunk; - ASSERT( !getCollMetadata().getNextChunk( getCollMetadata().getMinKey(), &nextChunk ) ); - } +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; + CollectionMetadata _metadata; +}; + +TEST_F(NoChunkFixture, BasicBelongsToMe) { + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << MINKEY))); + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << 10))); +} + +TEST_F(NoChunkFixture, CompoundKeyBelongsToMe) { + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << 1 << "b" << 2))); +} + +TEST_F(NoChunkFixture, IsKeyValid) { + ASSERT_TRUE(getCollMetadata().isValidKey(BSON("a" + << "abcde"))); + ASSERT_TRUE(getCollMetadata().isValidKey(BSON("a" << 3))); + ASSERT_FALSE(getCollMetadata().isValidKey(BSON("a" + << "abcde" + << "b" << 1))); + ASSERT_FALSE(getCollMetadata().isValidKey(BSON("c" + << "abcde"))); +} + +TEST_F(NoChunkFixture, getNextFromEmpty) { + ChunkType nextChunk; + ASSERT(!getCollMetadata().getNextChunk(getCollMetadata().getMinKey(), &nextChunk)); +} - TEST_F(NoChunkFixture, FirstChunkClonePlus) { - ChunkType chunk; - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); - - string errMsg; - const ChunkVersion version( 99, 0, OID() ); - unique_ptr<CollectionMetadata> cloned( getCollMetadata().clonePlusChunk( chunk, - version, - &errMsg ) ); - - ASSERT( errMsg.empty() ); - ASSERT_EQUALS( 1u, cloned->getNumChunks() ); - ASSERT_EQUALS( cloned->getShardVersion().toLong(), version.toLong() ); - ASSERT_EQUALS( cloned->getCollVersion().toLong(), version.toLong() ); - ASSERT( cloned->keyBelongsToMe(BSON("a" << 15)) ); - } +TEST_F(NoChunkFixture, FirstChunkClonePlus) { + ChunkType chunk; + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); + + string errMsg; + const ChunkVersion version(99, 0, OID()); + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().clonePlusChunk(chunk, version, &errMsg)); + + ASSERT(errMsg.empty()); + ASSERT_EQUALS(1u, cloned->getNumChunks()); + ASSERT_EQUALS(cloned->getShardVersion().toLong(), version.toLong()); + ASSERT_EQUALS(cloned->getCollVersion().toLong(), version.toLong()); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 15))); +} - TEST_F(NoChunkFixture, MustHaveVersionForFirstChunk) { - ChunkType chunk; - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); +TEST_F(NoChunkFixture, MustHaveVersionForFirstChunk) { + ChunkType chunk; + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - string errMsg; - unique_ptr<CollectionMetadata> cloned( getCollMetadata() // br - .clonePlusChunk( chunk, ChunkVersion( 0, 0, OID() ), &errMsg ) ); + string errMsg; + unique_ptr<CollectionMetadata> cloned( + getCollMetadata() // br + .clonePlusChunk(chunk, ChunkVersion(0, 0, OID()), &errMsg)); - ASSERT( cloned == NULL ); - ASSERT_FALSE( errMsg.empty() ); - } + ASSERT(cloned == NULL); + ASSERT_FALSE(errMsg.empty()); +} - TEST_F(NoChunkFixture, NoPendingChunks) { - ASSERT( !getCollMetadata().keyIsPending(BSON( "a" << 15 )) ); - ASSERT( !getCollMetadata().keyIsPending(BSON( "a" << 25 )) ); - } +TEST_F(NoChunkFixture, NoPendingChunks) { + ASSERT(!getCollMetadata().keyIsPending(BSON("a" << 15))); + ASSERT(!getCollMetadata().keyIsPending(BSON("a" << 25))); +} - TEST_F(NoChunkFixture, FirstPendingChunk) { +TEST_F(NoChunkFixture, FirstPendingChunk) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + ASSERT(cloned->keyIsPending(BSON("a" << 15))); + ASSERT(!cloned->keyIsPending(BSON("a" << 25))); + ASSERT(cloned->keyIsPending(BSON("a" << 10))); + ASSERT(!cloned->keyIsPending(BSON("a" << 20))); +} - ASSERT( cloned->keyIsPending(BSON( "a" << 15 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 25 )) ); - ASSERT( cloned->keyIsPending(BSON( "a" << 10 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 20 )) ); - } +TEST_F(NoChunkFixture, EmptyMultiPendingChunk) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - TEST_F(NoChunkFixture, EmptyMultiPendingChunk) { + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + chunk.setMin(BSON("a" << 40)); + chunk.setMax(BSON("a" << 50)); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + cloned.reset(cloned->clonePlusPending(chunk, &errMsg)); - chunk.setMin( BSON("a" << 40) ); - chunk.setMax( BSON("a" << 50) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - cloned.reset( cloned->clonePlusPending( chunk, &errMsg ) ); + ASSERT(cloned->keyIsPending(BSON("a" << 15))); + ASSERT(!cloned->keyIsPending(BSON("a" << 25))); + ASSERT(cloned->keyIsPending(BSON("a" << 45))); + ASSERT(!cloned->keyIsPending(BSON("a" << 55))); +} - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); +TEST_F(NoChunkFixture, MinusPendingChunk) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - ASSERT( cloned->keyIsPending(BSON( "a" << 15 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 25 )) ); - ASSERT( cloned->keyIsPending(BSON( "a" << 45 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 55 )) ); - } + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - TEST_F(NoChunkFixture, MinusPendingChunk) { + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); + cloned.reset(cloned->cloneMinusPending(chunk, &errMsg)); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + ASSERT(!cloned->keyIsPending(BSON("a" << 15))); + ASSERT(!cloned->keyIsPending(BSON("a" << 25))); +} - cloned.reset( cloned->cloneMinusPending( chunk, &errMsg ) ); +TEST_F(NoChunkFixture, OverlappingPendingChunk) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 30)); - ASSERT( !cloned->keyIsPending(BSON( "a" << 15 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 25 )) ); - } + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); - TEST_F(NoChunkFixture, OverlappingPendingChunk) { + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 40)); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 30) ); + cloned.reset(cloned->clonePlusPending(chunk, &errMsg)); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + ASSERT(!cloned->keyIsPending(BSON("a" << 15))); + ASSERT(cloned->keyIsPending(BSON("a" << 25))); + ASSERT(cloned->keyIsPending(BSON("a" << 35))); + ASSERT(!cloned->keyIsPending(BSON("a" << 45))); +} - chunk.setMin( BSON("a" << 20) ); - chunk.setMax( BSON("a" << 40) ); +TEST_F(NoChunkFixture, OverlappingPendingChunks) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - cloned.reset( cloned->clonePlusPending( chunk, &errMsg ) ); + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 30)); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); - ASSERT( !cloned->keyIsPending(BSON( "a" << 15 )) ); - ASSERT( cloned->keyIsPending(BSON( "a" << 25 )) ); - ASSERT( cloned->keyIsPending(BSON( "a" << 35 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 45 )) ); - } + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - TEST_F(NoChunkFixture, OverlappingPendingChunks) { + chunk.setMin(BSON("a" << 30)); + chunk.setMax(BSON("a" << 50)); - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + cloned.reset(cloned->clonePlusPending(chunk, &errMsg)); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 30) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 40)); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + cloned.reset(cloned->clonePlusPending(chunk, &errMsg)); - chunk.setMin( BSON("a" << 30) ); - chunk.setMax( BSON("a" << 50) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - cloned.reset( cloned->clonePlusPending( chunk, &errMsg ) ); + ASSERT(!cloned->keyIsPending(BSON("a" << 15))); + ASSERT(cloned->keyIsPending(BSON("a" << 25))); + ASSERT(cloned->keyIsPending(BSON("a" << 35))); + ASSERT(!cloned->keyIsPending(BSON("a" << 45))); +} - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); +TEST_F(NoChunkFixture, MinusInvalidPendingChunk) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - chunk.setMin( BSON("a" << 20) ); - chunk.setMax( BSON("a" << 40) ); + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 30)); - cloned.reset( cloned->clonePlusPending( chunk, &errMsg ) ); + cloned.reset(getCollMetadata().cloneMinusPending(chunk, &errMsg)); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(cloned == NULL); +} - ASSERT( !cloned->keyIsPending(BSON( "a" << 15 )) ); - ASSERT( cloned->keyIsPending(BSON( "a" << 25 )) ); - ASSERT( cloned->keyIsPending(BSON( "a" << 35 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 45 )) ); - } +TEST_F(NoChunkFixture, MinusOverlappingPendingChunk) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - TEST_F(NoChunkFixture, MinusInvalidPendingChunk) { + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 30)); - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 30) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - cloned.reset( getCollMetadata().cloneMinusPending( chunk, &errMsg ) ); + chunk.setMin(BSON("a" << 15)); + chunk.setMax(BSON("a" << 35)); - ASSERT_NOT_EQUALS( errMsg, "" ); - ASSERT( cloned == NULL ); - } + cloned.reset(cloned->cloneMinusPending(chunk, &errMsg)); - TEST_F(NoChunkFixture, MinusOverlappingPendingChunk) { + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(cloned == NULL); +} - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; +TEST_F(NoChunkFixture, PlusChunkWithPending) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 30) ); + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - chunk.setMin( BSON("a" << 15) ); - chunk.setMax( BSON("a" << 35) ); + ASSERT(cloned->keyIsPending(BSON("a" << 15))); + ASSERT(!cloned->keyIsPending(BSON("a" << 25))); - cloned.reset( cloned->cloneMinusPending( chunk, &errMsg ) ); + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 30)); - ASSERT_NOT_EQUALS( errMsg, "" ); - ASSERT( cloned == NULL ); - } - - TEST_F(NoChunkFixture, PlusChunkWithPending) { - - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + cloned.reset(cloned->clonePlusChunk( + chunk, ChunkVersion(1, 0, cloned->getCollVersion().epoch()), &errMsg)); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + ASSERT(cloned->keyIsPending(BSON("a" << 15))); + ASSERT(!cloned->keyIsPending(BSON("a" << 25))); +} - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); +TEST_F(NoChunkFixture, MergeChunkEmpty) { + string errMsg; + unique_ptr<CollectionMetadata> cloned; - ASSERT( cloned->keyIsPending(BSON( "a" << 15 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 25 )) ); + cloned.reset(getCollMetadata().cloneMerge( + BSON("a" << 15), BSON("a" << 25), ChunkVersion(1, 0, OID::gen()), &errMsg)); - chunk.setMin( BSON("a" << 20) ); - chunk.setMax( BSON("a" << 30) ); + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(cloned == NULL); +} - cloned.reset( cloned->clonePlusChunk( chunk, - ChunkVersion( 1, - 0, - cloned->getCollVersion().epoch() ), - &errMsg ) ); +TEST_F(NoChunkFixture, OrphanedDataRangeBegin) { + const CollectionMetadata& metadata = getCollMetadata(); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + KeyRange keyRange; + BSONObj lookupKey = metadata.getMinKey(); + ASSERT(metadata.getNextOrphanRange(lookupKey, &keyRange)); - ASSERT( cloned->keyIsPending(BSON( "a" << 15 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 25 )) ); - } + ASSERT(keyRange.minKey.woCompare(metadata.getMinKey()) == 0); + ASSERT(keyRange.maxKey.woCompare(metadata.getMaxKey()) == 0); - TEST_F(NoChunkFixture, MergeChunkEmpty) { + // Make sure we don't have any more ranges + ASSERT(!metadata.getNextOrphanRange(keyRange.maxKey, &keyRange)); +} - string errMsg; - unique_ptr<CollectionMetadata> cloned; +TEST_F(NoChunkFixture, OrphanedDataRangeMiddle) { + const CollectionMetadata& metadata = getCollMetadata(); - cloned.reset( getCollMetadata().cloneMerge( BSON( "a" << 15 ), - BSON( "a" << 25 ), - ChunkVersion( 1, 0, OID::gen() ), - &errMsg ) ); + KeyRange keyRange; + BSONObj lookupKey = BSON("a" << 20); + ASSERT(metadata.getNextOrphanRange(lookupKey, &keyRange)); - ASSERT_NOT_EQUALS( errMsg, "" ); - ASSERT( cloned == NULL ); - } + ASSERT(keyRange.minKey.woCompare(metadata.getMinKey()) == 0); + ASSERT(keyRange.maxKey.woCompare(metadata.getMaxKey()) == 0); + ASSERT(keyRange.keyPattern.woCompare(metadata.getKeyPattern()) == 0); - TEST_F(NoChunkFixture, OrphanedDataRangeBegin) { + // Make sure we don't have any more ranges + ASSERT(!metadata.getNextOrphanRange(keyRange.maxKey, &keyRange)); +} - const CollectionMetadata& metadata = getCollMetadata(); +TEST_F(NoChunkFixture, OrphanedDataRangeEnd) { + const CollectionMetadata& metadata = getCollMetadata(); - KeyRange keyRange; - BSONObj lookupKey = metadata.getMinKey(); - ASSERT( metadata.getNextOrphanRange( lookupKey, &keyRange ) ); + KeyRange keyRange; + ASSERT(!metadata.getNextOrphanRange(metadata.getMaxKey(), &keyRange)); +} - ASSERT( keyRange.minKey.woCompare( metadata.getMinKey() ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( metadata.getMaxKey() ) == 0 ); +TEST_F(NoChunkFixture, PendingOrphanedDataRanges) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - // Make sure we don't have any more ranges - ASSERT( !metadata.getNextOrphanRange( keyRange.maxKey, &keyRange ) ); - } + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - TEST_F(NoChunkFixture, OrphanedDataRangeMiddle) { + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); + ASSERT_EQUALS(errMsg, string("")); + ASSERT(cloned != NULL); - const CollectionMetadata& metadata = getCollMetadata(); + KeyRange keyRange; + ASSERT(cloned->getNextOrphanRange(cloned->getMinKey(), &keyRange)); + ASSERT(keyRange.minKey.woCompare(cloned->getMinKey()) == 0); + ASSERT(keyRange.maxKey.woCompare(BSON("a" << 10)) == 0); + ASSERT(keyRange.keyPattern.woCompare(cloned->getKeyPattern()) == 0); - KeyRange keyRange; - BSONObj lookupKey = BSON( "a" << 20 ); - ASSERT( metadata.getNextOrphanRange( lookupKey, &keyRange ) ); + ASSERT(cloned->getNextOrphanRange(keyRange.maxKey, &keyRange)); + ASSERT(keyRange.minKey.woCompare(BSON("a" << 20)) == 0); + ASSERT(keyRange.maxKey.woCompare(cloned->getMaxKey()) == 0); + ASSERT(keyRange.keyPattern.woCompare(cloned->getKeyPattern()) == 0); - ASSERT( keyRange.minKey.woCompare( metadata.getMinKey() ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( metadata.getMaxKey() ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( metadata.getKeyPattern() ) == 0 ); + ASSERT(!cloned->getNextOrphanRange(keyRange.maxKey, &keyRange)); +} - // Make sure we don't have any more ranges - ASSERT( !metadata.getNextOrphanRange( keyRange.maxKey, &keyRange ) ); +/** + * Fixture with single chunk containing: + * [10->20) + */ +class SingleChunkFixture : public mongo::unittest::Test { +protected: + void setUp() { + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); + + OID epoch = OID::gen(); + ChunkVersion chunkVersion = ChunkVersion(1, 0, epoch); + + CollectionType collType; + collType.setNs(NamespaceString{"test.foo"}); + collType.setKeyPattern(BSON("a" << 1)); + collType.setUnique(false); + collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + collType.setEpoch(epoch); + _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); + + BSONObj fooSingle = BSON( + ChunkType::name("test.foo-a_10") + << ChunkType::ns("test.foo") << ChunkType::min(BSON("a" << 10)) + << ChunkType::max(BSON("a" << 20)) + << ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch(chunkVersion.toLong())) + << ChunkType::DEPRECATED_epoch(epoch) << ChunkType::shard("shard0000")); + _dummyConfig->insert(ChunkType::ConfigNS, fooSingle); + + ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); + ASSERT(configLoc.isValid()); + CatalogManagerLegacy catalogManager; + catalogManager.init(configLoc); + + MetadataLoader loader; + Status status = loader.makeCollectionMetadata( + &catalogManager, "test.foo", "shard0000", NULL, &_metadata); + ASSERT_OK(status); } - TEST_F(NoChunkFixture, OrphanedDataRangeEnd) { - - const CollectionMetadata& metadata = getCollMetadata(); - - KeyRange keyRange; - ASSERT( !metadata.getNextOrphanRange( metadata.getMaxKey(), &keyRange ) ); + void tearDown() { + MockConnRegistry::get()->clear(); } - TEST_F(NoChunkFixture, PendingOrphanedDataRanges) { - - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; - - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); - - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); - ASSERT_EQUALS( errMsg, string("") ); - ASSERT( cloned != NULL ); - - KeyRange keyRange; - ASSERT( cloned->getNextOrphanRange( cloned->getMinKey(), &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( cloned->getMinKey() ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( BSON( "a" << 10 ) ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( cloned->getKeyPattern() ) == 0 ); - - ASSERT( cloned->getNextOrphanRange( keyRange.maxKey, &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( BSON( "a" << 20 ) ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( cloned->getMaxKey() ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( cloned->getKeyPattern() ) == 0 ); - - ASSERT( !cloned->getNextOrphanRange( keyRange.maxKey, &keyRange ) ); + const CollectionMetadata& getCollMetadata() const { + return _metadata; } - /** - * Fixture with single chunk containing: - * [10->20) - */ - class SingleChunkFixture : public mongo::unittest::Test { - protected: - void setUp() { - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); - - OID epoch = OID::gen(); - ChunkVersion chunkVersion = ChunkVersion( 1, 0, epoch ); - - CollectionType collType; - collType.setNs(NamespaceString{"test.foo"}); - collType.setKeyPattern(BSON("a" << 1)); - collType.setUnique(false); - collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); - collType.setEpoch(epoch); - _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); - - BSONObj fooSingle = BSON(ChunkType::name("test.foo-a_10") << - ChunkType::ns("test.foo") << - ChunkType::min(BSON("a" << 10)) << - ChunkType::max(BSON("a" << 20)) << - ChunkType::DEPRECATED_lastmod( - Date_t::fromMillisSinceEpoch(chunkVersion.toLong())) << - ChunkType::DEPRECATED_epoch(epoch) << - ChunkType::shard("shard0000")); - _dummyConfig->insert( ChunkType::ConfigNS, fooSingle ); - - ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); - ASSERT(configLoc.isValid()); - CatalogManagerLegacy catalogManager; - catalogManager.init(configLoc); - - MetadataLoader loader; - Status status = loader.makeCollectionMetadata(&catalogManager, - "test.foo", - "shard0000", - NULL, - &_metadata); - ASSERT_OK(status); - } +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; + CollectionMetadata _metadata; +}; + +TEST_F(SingleChunkFixture, BasicBelongsToMe) { + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 10))); + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 15))); + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 19))); +} + +TEST_F(SingleChunkFixture, DoesntBelongsToMe) { + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << 0))); + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << 9))); + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << 20))); + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << 1234))); + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << MINKEY))); + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << MAXKEY))); +} + +TEST_F(SingleChunkFixture, CompoudKeyBelongsToMe) { + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 15 << "a" << 14))); +} + +TEST_F(SingleChunkFixture, getNextFromEmpty) { + ChunkType nextChunk; + ASSERT(getCollMetadata().getNextChunk(getCollMetadata().getMinKey(), &nextChunk)); + ASSERT_EQUALS(0, nextChunk.getMin().woCompare(BSON("a" << 10))); + ASSERT_EQUALS(0, nextChunk.getMax().woCompare(BSON("a" << 20))); +} + +TEST_F(SingleChunkFixture, GetLastChunkIsFalse) { + ChunkType nextChunk; + ASSERT(!getCollMetadata().getNextChunk(getCollMetadata().getMaxKey(), &nextChunk)); +} + +TEST_F(SingleChunkFixture, LastChunkCloneMinus) { + ChunkType chunk; + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); + + string errMsg; + const ChunkVersion zeroVersion(0, 0, getCollMetadata().getShardVersion().epoch()); + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().cloneMigrate(chunk, zeroVersion, &errMsg)); + + ASSERT(errMsg.empty()); + ASSERT_EQUALS(0u, cloned->getNumChunks()); + ASSERT_EQUALS(cloned->getShardVersion().toLong(), zeroVersion.toLong()); + ASSERT_EQUALS(cloned->getCollVersion().toLong(), getCollMetadata().getCollVersion().toLong()); + ASSERT_FALSE(cloned->keyBelongsToMe(BSON("a" << 15))); +} + +TEST_F(SingleChunkFixture, LastChunkMinusCantHaveNonZeroVersion) { + ChunkType chunk; + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); + + string errMsg; + ChunkVersion version(99, 0, OID()); + unique_ptr<CollectionMetadata> cloned(getCollMetadata().cloneMigrate(chunk, version, &errMsg)); + + ASSERT(cloned == NULL); + ASSERT_FALSE(errMsg.empty()); +} + +TEST_F(SingleChunkFixture, PlusPendingChunk) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; + + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 30)); + + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); + + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); + + ASSERT(cloned->keyBelongsToMe(BSON("a" << 15))); + ASSERT(!cloned->keyBelongsToMe(BSON("a" << 25))); + ASSERT(!cloned->keyIsPending(BSON("a" << 15))); + ASSERT(cloned->keyIsPending(BSON("a" << 25))); +} + +TEST_F(SingleChunkFixture, PlusOverlapPendingChunk) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; + + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); + + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); + + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(cloned == NULL); +} + +TEST_F(SingleChunkFixture, MinusChunkWithPending) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; + + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 30)); + + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); + + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); + + ASSERT(cloned->keyIsPending(BSON("a" << 25))); + ASSERT(!cloned->keyIsPending(BSON("a" << 35))); - void tearDown() { - MockConnRegistry::get()->clear(); - } + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - const CollectionMetadata& getCollMetadata() const { - return _metadata; - } + cloned.reset( + cloned->cloneMigrate(chunk, ChunkVersion(0, 0, cloned->getCollVersion().epoch()), &errMsg)); - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - CollectionMetadata _metadata; - }; + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - TEST_F(SingleChunkFixture, BasicBelongsToMe) { - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 10)) ); - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 15)) ); - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 19)) ); - } + ASSERT(cloned->keyIsPending(BSON("a" << 25))); + ASSERT(!cloned->keyIsPending(BSON("a" << 35))); +} - TEST_F(SingleChunkFixture, DoesntBelongsToMe) { - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << 0)) ); - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << 9)) ); - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << 20)) ); - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << 1234)) ); - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << MINKEY)) ); - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << MAXKEY)) ); - } +TEST_F(SingleChunkFixture, SingleSplit) { + ChunkVersion version; + getCollMetadata().getCollVersion().cloneTo(&version); + version.incMinor(); - TEST_F(SingleChunkFixture, CompoudKeyBelongsToMe) { - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 15 << "a" << 14)) ); - } + ChunkType chunk; + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - TEST_F(SingleChunkFixture, getNextFromEmpty) { - ChunkType nextChunk; - ASSERT( getCollMetadata().getNextChunk( getCollMetadata().getMinKey(), &nextChunk ) ); - ASSERT_EQUALS( 0, nextChunk.getMin().woCompare(BSON("a" << 10)) ); - ASSERT_EQUALS( 0, nextChunk.getMax().woCompare(BSON("a" << 20)) ); - } + vector<BSONObj> splitPoints; + splitPoints.push_back(BSON("a" << 14)); - TEST_F(SingleChunkFixture, GetLastChunkIsFalse) { - ChunkType nextChunk; - ASSERT( !getCollMetadata().getNextChunk( getCollMetadata().getMaxKey(), &nextChunk ) ); - } + string errMsg; + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().cloneSplit(chunk, splitPoints, version, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - TEST_F(SingleChunkFixture, LastChunkCloneMinus) { - ChunkType chunk; - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); - - string errMsg; - const ChunkVersion zeroVersion( 0, 0, getCollMetadata().getShardVersion().epoch() ); - unique_ptr<CollectionMetadata> cloned( getCollMetadata().cloneMigrate( chunk, - zeroVersion, - &errMsg ) ); - - ASSERT( errMsg.empty() ); - ASSERT_EQUALS( 0u, cloned->getNumChunks() ); - ASSERT_EQUALS( cloned->getShardVersion().toLong(), zeroVersion.toLong() ); - ASSERT_EQUALS( cloned->getCollVersion().toLong(), - getCollMetadata().getCollVersion().toLong() ); - ASSERT_FALSE( cloned->keyBelongsToMe(BSON("a" << 15)) ); - } + ChunkVersion newVersion(cloned->getCollVersion()); + ASSERT_EQUALS(version.epoch(), newVersion.epoch()); + ASSERT_EQUALS(version.majorVersion(), newVersion.majorVersion()); + ASSERT_EQUALS(version.minorVersion() + 1, newVersion.minorVersion()); - TEST_F(SingleChunkFixture, LastChunkMinusCantHaveNonZeroVersion) { - ChunkType chunk; - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); + ASSERT(cloned->getNextChunk(BSON("a" << MINKEY), &chunk)); + ASSERT(chunk.getMin().woCompare(BSON("a" << 10)) == 0); + ASSERT(chunk.getMax().woCompare(BSON("a" << 14)) == 0); - string errMsg; - ChunkVersion version( 99, 0, OID() ); - unique_ptr<CollectionMetadata> cloned( getCollMetadata().cloneMigrate( chunk, - version, - &errMsg ) ); + ASSERT(cloned->getNextChunk(BSON("a" << 14), &chunk)); + ASSERT(chunk.getMin().woCompare(BSON("a" << 14)) == 0); + ASSERT(chunk.getMax().woCompare(BSON("a" << 20)) == 0); - ASSERT( cloned == NULL ); - ASSERT_FALSE( errMsg.empty() ); - } + ASSERT_FALSE(cloned->getNextChunk(BSON("a" << 20), &chunk)); +} - TEST_F(SingleChunkFixture, PlusPendingChunk) { +TEST_F(SingleChunkFixture, MultiSplit) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - chunk.setMin( BSON("a" << 20) ); - chunk.setMax( BSON("a" << 30) ); + vector<BSONObj> splitPoints; + splitPoints.push_back(BSON("a" << 14)); + splitPoints.push_back(BSON("a" << 16)); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + ChunkVersion version; + getCollMetadata().getCollVersion().cloneTo(&version); + version.incMinor(); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + cloned.reset(getCollMetadata().cloneSplit(chunk, splitPoints, version, &errMsg)); - ASSERT( cloned->keyBelongsToMe(BSON("a" << 15)) ); - ASSERT( !cloned->keyBelongsToMe(BSON("a" << 25)) ); - ASSERT( !cloned->keyIsPending(BSON("a" << 15)) ); - ASSERT( cloned->keyIsPending(BSON("a" << 25)) ); - } + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - TEST_F(SingleChunkFixture, PlusOverlapPendingChunk) { + ChunkVersion newVersion(cloned->getCollVersion()); + ASSERT_EQUALS(version.epoch(), newVersion.epoch()); + ASSERT_EQUALS(version.majorVersion(), newVersion.majorVersion()); + ASSERT_EQUALS(version.minorVersion() + 2, newVersion.minorVersion()); - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + ASSERT(cloned->getNextChunk(BSON("a" << MINKEY), &chunk)); + ASSERT(chunk.getMin().woCompare(BSON("a" << 10)) == 0); + ASSERT(chunk.getMax().woCompare(BSON("a" << 14)) == 0); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); + ASSERT(cloned->getNextChunk(BSON("a" << 14), &chunk)); + ASSERT(chunk.getMin().woCompare(BSON("a" << 14)) == 0); + ASSERT(chunk.getMax().woCompare(BSON("a" << 16)) == 0); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + ASSERT(cloned->getNextChunk(BSON("a" << 16), &chunk)); + ASSERT(chunk.getMin().woCompare(BSON("a" << 16)) == 0); + ASSERT(chunk.getMax().woCompare(BSON("a" << 20)) == 0); - ASSERT_NOT_EQUALS( errMsg, "" ); - ASSERT( cloned == NULL ); - } + ASSERT_FALSE(cloned->getNextChunk(BSON("a" << 20), &chunk)); +} - TEST_F(SingleChunkFixture, MinusChunkWithPending) { +TEST_F(SingleChunkFixture, SplitChunkWithPending) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 30)); - chunk.setMin( BSON("a" << 20) ); - chunk.setMax( BSON("a" << 30) ); + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + ASSERT(cloned->keyIsPending(BSON("a" << 25))); + ASSERT(!cloned->keyIsPending(BSON("a" << 35))); - ASSERT( cloned->keyIsPending(BSON( "a" << 25 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 35 )) ); + chunk.setMin(BSON("a" << 10)); + chunk.setMax(BSON("a" << 20)); - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); + vector<BSONObj> splitPoints; + splitPoints.push_back(BSON("a" << 14)); + splitPoints.push_back(BSON("a" << 16)); - cloned.reset( cloned->cloneMigrate( chunk, - ChunkVersion( 0, 0, cloned->getCollVersion().epoch() ), - &errMsg ) ); + cloned.reset(cloned->cloneSplit(chunk, + splitPoints, + ChunkVersion(cloned->getCollVersion().majorVersion() + 1, + 0, + cloned->getCollVersion().epoch()), + &errMsg)); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); - ASSERT( cloned->keyIsPending(BSON( "a" << 25 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 35 )) ); - } + ASSERT(cloned->keyIsPending(BSON("a" << 25))); + ASSERT(!cloned->keyIsPending(BSON("a" << 35))); +} - TEST_F(SingleChunkFixture, SingleSplit) { - ChunkVersion version; - getCollMetadata().getCollVersion().cloneTo(&version); - version.incMinor(); - ChunkType chunk; - chunk.setMin(BSON("a" << 10)); - chunk.setMax(BSON("a" << 20)); +TEST_F(SingleChunkFixture, MergeChunkSingle) { + string errMsg; + unique_ptr<CollectionMetadata> cloned; - vector<BSONObj> splitPoints; - splitPoints.push_back(BSON("a" << 14)); + cloned.reset(getCollMetadata().cloneMerge( + BSON("a" << 10), BSON("a" << 20), ChunkVersion(2, 0, OID::gen()), &errMsg)); - string errMsg; - unique_ptr<CollectionMetadata> cloned(getCollMetadata().cloneSplit(chunk, - splitPoints, - version, - &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT(cloned != NULL); + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(cloned == NULL); +} - ChunkVersion newVersion( cloned->getCollVersion() ); - ASSERT_EQUALS( version.epoch(), newVersion.epoch() ); - ASSERT_EQUALS( version.majorVersion(), newVersion.majorVersion() ); - ASSERT_EQUALS( version.minorVersion() + 1, newVersion.minorVersion() ); +TEST_F(SingleChunkFixture, ChunkOrphanedDataRanges) { + KeyRange keyRange; + ASSERT(getCollMetadata().getNextOrphanRange(getCollMetadata().getMinKey(), &keyRange)); + ASSERT(keyRange.minKey.woCompare(getCollMetadata().getMinKey()) == 0); + ASSERT(keyRange.maxKey.woCompare(BSON("a" << 10)) == 0); + ASSERT(keyRange.keyPattern.woCompare(getCollMetadata().getKeyPattern()) == 0); - ASSERT( cloned->getNextChunk(BSON("a" << MINKEY), &chunk) ); - ASSERT( chunk.getMin().woCompare( BSON("a" << 10) ) == 0 ); - ASSERT( chunk.getMax().woCompare( BSON("a" << 14) ) == 0 ); + ASSERT(getCollMetadata().getNextOrphanRange(keyRange.maxKey, &keyRange)); + ASSERT(keyRange.minKey.woCompare(BSON("a" << 20)) == 0); + ASSERT(keyRange.maxKey.woCompare(getCollMetadata().getMaxKey()) == 0); + ASSERT(keyRange.keyPattern.woCompare(getCollMetadata().getKeyPattern()) == 0); - ASSERT( cloned->getNextChunk(BSON("a" << 14), &chunk) ); - ASSERT( chunk.getMin().woCompare( BSON("a" << 14) ) == 0 ); - ASSERT( chunk.getMax().woCompare( BSON("a" << 20) ) == 0 ); + ASSERT(!getCollMetadata().getNextOrphanRange(keyRange.maxKey, &keyRange)); +} - ASSERT_FALSE( cloned->getNextChunk(BSON("a" << 20), &chunk) ); +/** + * Fixture with single chunk containing: + * [(min, min)->(max, max)) + */ +class SingleChunkMinMaxCompoundKeyFixture : public mongo::unittest::Test { +protected: + void setUp() { + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); + + OID epoch = OID::gen(); + ChunkVersion chunkVersion = ChunkVersion(1, 0, epoch); + + CollectionType collType; + collType.setNs(NamespaceString{"test.foo"}); + collType.setKeyPattern(BSON("a" << 1)); + collType.setUnique(false); + collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + collType.setEpoch(epoch); + _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); + + BSONObj fooSingle = BSON( + ChunkType::name("test.foo-a_MinKey") + << ChunkType::ns("test.foo") << ChunkType::min(BSON("a" << MINKEY << "b" << MINKEY)) + << ChunkType::max(BSON("a" << MAXKEY << "b" << MAXKEY)) + << ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch(chunkVersion.toLong())) + << ChunkType::DEPRECATED_epoch(epoch) << ChunkType::shard("shard0000")); + _dummyConfig->insert(ChunkType::ConfigNS, fooSingle); + + ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); + ASSERT(configLoc.isValid()); + CatalogManagerLegacy catalogManager; + catalogManager.init(configLoc); + + MetadataLoader loader; + Status status = loader.makeCollectionMetadata( + &catalogManager, "test.foo", "shard0000", NULL, &_metadata); + ASSERT_OK(status); } - TEST_F(SingleChunkFixture, MultiSplit) { - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; - - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); - - vector<BSONObj> splitPoints; - splitPoints.push_back( BSON("a" << 14) ); - splitPoints.push_back( BSON("a" << 16) ); - - ChunkVersion version; - getCollMetadata().getCollVersion().cloneTo( &version ); - version.incMinor(); - - cloned.reset( getCollMetadata().cloneSplit( chunk, - splitPoints, - version, - &errMsg ) ); - - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); - - ChunkVersion newVersion( cloned->getCollVersion() ); - ASSERT_EQUALS( version.epoch(), newVersion.epoch() ); - ASSERT_EQUALS( version.majorVersion(), newVersion.majorVersion() ); - ASSERT_EQUALS( version.minorVersion() + 2, newVersion.minorVersion() ); - - ASSERT( cloned->getNextChunk(BSON("a" << MINKEY), &chunk) ); - ASSERT( chunk.getMin().woCompare( BSON("a" << 10) ) == 0 ); - ASSERT( chunk.getMax().woCompare( BSON("a" << 14) ) == 0 ); - - ASSERT( cloned->getNextChunk(BSON("a" << 14), &chunk) ); - ASSERT( chunk.getMin().woCompare( BSON("a" << 14) ) == 0 ); - ASSERT( chunk.getMax().woCompare( BSON("a" << 16) ) == 0 ); - - ASSERT( cloned->getNextChunk(BSON("a" << 16), &chunk) ); - ASSERT( chunk.getMin().woCompare( BSON("a" << 16) ) == 0 ); - ASSERT( chunk.getMax().woCompare( BSON("a" << 20) ) == 0 ); - - ASSERT_FALSE( cloned->getNextChunk(BSON("a" << 20), &chunk) ); + void tearDown() { + MockConnRegistry::get()->clear(); } - TEST_F(SingleChunkFixture, SplitChunkWithPending) { - - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; - - chunk.setMin( BSON("a" << 20) ); - chunk.setMax( BSON("a" << 30) ); - - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); - - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); - - ASSERT( cloned->keyIsPending(BSON( "a" << 25 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 35 )) ); - - chunk.setMin( BSON("a" << 10) ); - chunk.setMax( BSON("a" << 20) ); - - vector<BSONObj> splitPoints; - splitPoints.push_back( BSON("a" << 14) ); - splitPoints.push_back( BSON("a" << 16) ); - - cloned.reset( cloned->cloneSplit( chunk, - splitPoints, - ChunkVersion( cloned->getCollVersion().majorVersion() + 1, - 0, - cloned->getCollVersion().epoch() ), - &errMsg ) ); - - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); - - ASSERT( cloned->keyIsPending(BSON( "a" << 25 )) ); - ASSERT( !cloned->keyIsPending(BSON( "a" << 35 )) ); + const CollectionMetadata& getCollMetadata() const { + return _metadata; } +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; + CollectionMetadata _metadata; +}; - TEST_F(SingleChunkFixture, MergeChunkSingle) { - - string errMsg; - unique_ptr<CollectionMetadata> cloned; +// Note: no tests for single key belongsToMe because they are not allowed +// if shard key is compound. - cloned.reset( getCollMetadata().cloneMerge( BSON( "a" << 10 ), - BSON( "a" << 20 ), - ChunkVersion( 2, 0, OID::gen() ), - &errMsg ) ); +TEST_F(SingleChunkMinMaxCompoundKeyFixture, CompoudKeyBelongsToMe) { + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << MINKEY << "b" << MINKEY))); + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << MAXKEY << "b" << MAXKEY))); + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << MINKEY << "b" << 10))); + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 10 << "b" << 20))); +} - ASSERT_NOT_EQUALS( errMsg, "" ); - ASSERT( cloned == NULL ); +/** + * Fixture with chunks: + * [(10, 0)->(20, 0)), [(30, 0)->(40, 0)) + */ +class TwoChunksWithGapCompoundKeyFixture : public mongo::unittest::Test { +protected: + void setUp() { + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); + + OID epoch = OID::gen(); + ChunkVersion chunkVersion = ChunkVersion(1, 0, epoch); + + CollectionType collType; + collType.setNs(NamespaceString{"test.foo"}); + collType.setKeyPattern(BSON("a" << 1)); + collType.setUnique(false); + collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + collType.setEpoch(epoch); + _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); + + _dummyConfig->insert( + ChunkType::ConfigNS, + BSON(ChunkType::name("test.foo-a_10") + << ChunkType::ns("test.foo") << ChunkType::min(BSON("a" << 10 << "b" << 0)) + << ChunkType::max(BSON("a" << 20 << "b" << 0)) + << ChunkType::DEPRECATED_lastmod( + Date_t::fromMillisSinceEpoch(chunkVersion.toLong())) + << ChunkType::DEPRECATED_epoch(epoch) << ChunkType::shard("shard0000"))); + + _dummyConfig->insert( + ChunkType::ConfigNS, + BSON(ChunkType::name("test.foo-a_10") + << ChunkType::ns("test.foo") << ChunkType::min(BSON("a" << 30 << "b" << 0)) + << ChunkType::max(BSON("a" << 40 << "b" << 0)) + << ChunkType::DEPRECATED_lastmod( + Date_t::fromMillisSinceEpoch(chunkVersion.toLong())) + << ChunkType::DEPRECATED_epoch(epoch) << ChunkType::shard("shard0000"))); + + ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); + ASSERT(configLoc.isValid()); + CatalogManagerLegacy catalogManager; + catalogManager.init(configLoc); + + MetadataLoader loader; + Status status = loader.makeCollectionMetadata( + &catalogManager, "test.foo", "shard0000", NULL, &_metadata); + ASSERT_OK(status); } - TEST_F(SingleChunkFixture, ChunkOrphanedDataRanges) { - - KeyRange keyRange; - ASSERT( getCollMetadata().getNextOrphanRange( getCollMetadata().getMinKey(), &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( getCollMetadata().getMinKey() ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( BSON( "a" << 10 ) ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( getCollMetadata().getKeyPattern() ) == 0 ); - - ASSERT( getCollMetadata().getNextOrphanRange( keyRange.maxKey, &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( BSON( "a" << 20 ) ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( getCollMetadata().getMaxKey() ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( getCollMetadata().getKeyPattern() ) == 0 ); - - ASSERT( !getCollMetadata().getNextOrphanRange( keyRange.maxKey, &keyRange ) ); + void tearDown() { + MockConnRegistry::get()->clear(); } - /** - * Fixture with single chunk containing: - * [(min, min)->(max, max)) - */ - class SingleChunkMinMaxCompoundKeyFixture : public mongo::unittest::Test { - protected: - void setUp() { - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); - - OID epoch = OID::gen(); - ChunkVersion chunkVersion = ChunkVersion( 1, 0, epoch ); - - CollectionType collType; - collType.setNs(NamespaceString{"test.foo"}); - collType.setKeyPattern(BSON("a" << 1)); - collType.setUnique(false); - collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); - collType.setEpoch(epoch); - _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); - - BSONObj fooSingle = BSON( - ChunkType::name("test.foo-a_MinKey") << - ChunkType::ns("test.foo") << - ChunkType::min(BSON("a" << MINKEY << "b" << MINKEY)) << - ChunkType::max(BSON("a" << MAXKEY << "b" << MAXKEY)) << - ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch( - chunkVersion.toLong())) << - ChunkType::DEPRECATED_epoch(epoch) << - ChunkType::shard("shard0000")); - _dummyConfig->insert( ChunkType::ConfigNS, fooSingle ); - - ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); - ASSERT(configLoc.isValid()); - CatalogManagerLegacy catalogManager; - catalogManager.init(configLoc); - - MetadataLoader loader; - Status status = loader.makeCollectionMetadata(&catalogManager, - "test.foo", - "shard0000", - NULL, - &_metadata); - ASSERT_OK(status); - } - - void tearDown() { - MockConnRegistry::get()->clear(); - } - - const CollectionMetadata& getCollMetadata() const { - return _metadata; - } - - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - CollectionMetadata _metadata; - }; - - // Note: no tests for single key belongsToMe because they are not allowed - // if shard key is compound. - - TEST_F(SingleChunkMinMaxCompoundKeyFixture, CompoudKeyBelongsToMe) { - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << MINKEY << "b" << MINKEY)) ); - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << MAXKEY << "b" << MAXKEY)) ); - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << MINKEY << "b" << 10)) ); - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 10 << "b" << 20)) ); + const CollectionMetadata& getCollMetadata() const { + return _metadata; } - /** - * Fixture with chunks: - * [(10, 0)->(20, 0)), [(30, 0)->(40, 0)) - */ - class TwoChunksWithGapCompoundKeyFixture : public mongo::unittest::Test { - protected: - void setUp() { - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; + CollectionMetadata _metadata; +}; + +TEST_F(TwoChunksWithGapCompoundKeyFixture, ClonePlusBasic) { + ChunkType chunk; + chunk.setMin(BSON("a" << 40 << "b" << 0)); + chunk.setMax(BSON("a" << 50 << "b" << 0)); + + string errMsg; + ChunkVersion version(1, 0, getCollMetadata().getShardVersion().epoch()); + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().clonePlusChunk(chunk, version, &errMsg)); + + ASSERT(errMsg.empty()); + ASSERT_EQUALS(2u, getCollMetadata().getNumChunks()); + ASSERT_EQUALS(3u, cloned->getNumChunks()); + + // TODO: test maxShardVersion, maxCollVersion + + ASSERT_FALSE(cloned->keyBelongsToMe(BSON("a" << 25 << "b" << 0))); + ASSERT_FALSE(cloned->keyBelongsToMe(BSON("a" << 29 << "b" << 0))); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 30 << "b" << 0))); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 45 << "b" << 0))); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 49 << "b" << 0))); + ASSERT_FALSE(cloned->keyBelongsToMe(BSON("a" << 50 << "b" << 0))); +} + +TEST_F(TwoChunksWithGapCompoundKeyFixture, ClonePlusOverlappingRange) { + ChunkType chunk; + chunk.setMin(BSON("a" << 15 << "b" << 0)); + chunk.setMax(BSON("a" << 25 << "b" << 0)); + + string errMsg; + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().clonePlusChunk(chunk, ChunkVersion(1, 0, OID()), &errMsg)); + ASSERT(cloned == NULL); + ASSERT_FALSE(errMsg.empty()); + ASSERT_EQUALS(2u, getCollMetadata().getNumChunks()); +} + +TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneMinusBasic) { + ChunkType chunk; + chunk.setMin(BSON("a" << 10 << "b" << 0)); + chunk.setMax(BSON("a" << 20 << "b" << 0)); + + string errMsg; + ChunkVersion version(2, 0, OID()); + unique_ptr<CollectionMetadata> cloned(getCollMetadata().cloneMigrate(chunk, version, &errMsg)); + + ASSERT(errMsg.empty()); + ASSERT_EQUALS(2u, getCollMetadata().getNumChunks()); + ASSERT_EQUALS(1u, cloned->getNumChunks()); + + // TODO: test maxShardVersion, maxCollVersion + + ASSERT_FALSE(cloned->keyBelongsToMe(BSON("a" << 5 << "b" << 0))); + ASSERT_FALSE(cloned->keyBelongsToMe(BSON("a" << 15 << "b" << 0))); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 30 << "b" << 0))); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 35 << "b" << 0))); + ASSERT_FALSE(cloned->keyBelongsToMe(BSON("a" << 40 << "b" << 0))); +} + +TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneMinusNonExisting) { + ChunkType chunk; + chunk.setMin(BSON("a" << 25 << "b" << 0)); + chunk.setMax(BSON("a" << 28 << "b" << 0)); + + string errMsg; + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().cloneMigrate(chunk, ChunkVersion(1, 0, OID()), &errMsg)); + ASSERT(cloned == NULL); + ASSERT_FALSE(errMsg.empty()); + ASSERT_EQUALS(2u, getCollMetadata().getNumChunks()); +} + +TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneSplitBasic) { + const BSONObj min(BSON("a" << 10 << "b" << 0)); + const BSONObj max(BSON("a" << 20 << "b" << 0)); + + ChunkType chunk; + chunk.setMin(min); + chunk.setMax(max); + + const BSONObj split1(BSON("a" << 15 << "b" << 0)); + const BSONObj split2(BSON("a" << 18 << "b" << 0)); + vector<BSONObj> splitKeys; + splitKeys.push_back(split1); + splitKeys.push_back(split2); + ChunkVersion version(1, 99, OID()); // first chunk 1|99 , second 1|100 + + string errMsg; + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().cloneSplit(chunk, splitKeys, version, &errMsg)); + + version.incMinor(); /* second chunk 1|100, first split point */ + version.incMinor(); /* third chunk 1|101, second split point */ + ASSERT_EQUALS(cloned->getShardVersion().toLong(), version.toLong() /* 1|101 */); + ASSERT_EQUALS(cloned->getCollVersion().toLong(), version.toLong()); + ASSERT_EQUALS(getCollMetadata().getNumChunks(), 2u); + ASSERT_EQUALS(cloned->getNumChunks(), 4u); + ASSERT(cloned->keyBelongsToMe(min)); + ASSERT(cloned->keyBelongsToMe(split1)); + ASSERT(cloned->keyBelongsToMe(split2)); + ASSERT(!cloned->keyBelongsToMe(max)); +} + +TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneSplitOutOfRangeSplitPoint) { + ChunkType chunk; + chunk.setMin(BSON("a" << 10 << "b" << 0)); + chunk.setMax(BSON("a" << 20 << "b" << 0)); + + vector<BSONObj> splitKeys; + splitKeys.push_back(BSON("a" << 5 << "b" << 0)); + + string errMsg; + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().cloneSplit(chunk, splitKeys, ChunkVersion(1, 0, OID()), &errMsg)); + + ASSERT(cloned == NULL); + ASSERT_FALSE(errMsg.empty()); + ASSERT_EQUALS(2u, getCollMetadata().getNumChunks()); +} + +TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneSplitBadChunkRange) { + const BSONObj min(BSON("a" << 10 << "b" << 0)); + const BSONObj max(BSON("a" << 25 << "b" << 0)); + + ChunkType chunk; + chunk.setMin(BSON("a" << 10 << "b" << 0)); + chunk.setMax(BSON("a" << 25 << "b" << 0)); + + vector<BSONObj> splitKeys; + splitKeys.push_back(BSON("a" << 15 << "b" << 0)); + + string errMsg; + unique_ptr<CollectionMetadata> cloned( + getCollMetadata().cloneSplit(chunk, splitKeys, ChunkVersion(1, 0, OID()), &errMsg)); + + ASSERT(cloned == NULL); + ASSERT_FALSE(errMsg.empty()); + ASSERT_EQUALS(2u, getCollMetadata().getNumChunks()); +} + +TEST_F(TwoChunksWithGapCompoundKeyFixture, ChunkGapOrphanedDataRanges) { + KeyRange keyRange; + ASSERT(getCollMetadata().getNextOrphanRange(getCollMetadata().getMinKey(), &keyRange)); + ASSERT(keyRange.minKey.woCompare(getCollMetadata().getMinKey()) == 0); + ASSERT(keyRange.maxKey.woCompare(BSON("a" << 10 << "b" << 0)) == 0); + ASSERT(keyRange.keyPattern.woCompare(getCollMetadata().getKeyPattern()) == 0); + + ASSERT(getCollMetadata().getNextOrphanRange(keyRange.maxKey, &keyRange)); + ASSERT(keyRange.minKey.woCompare(BSON("a" << 20 << "b" << 0)) == 0); + ASSERT(keyRange.maxKey.woCompare(BSON("a" << 30 << "b" << 0)) == 0); + ASSERT(keyRange.keyPattern.woCompare(getCollMetadata().getKeyPattern()) == 0); + + ASSERT(getCollMetadata().getNextOrphanRange(keyRange.maxKey, &keyRange)); + ASSERT(keyRange.minKey.woCompare(BSON("a" << 40 << "b" << 0)) == 0); + ASSERT(keyRange.maxKey.woCompare(getCollMetadata().getMaxKey()) == 0); + ASSERT(keyRange.keyPattern.woCompare(getCollMetadata().getKeyPattern()) == 0); + + ASSERT(!getCollMetadata().getNextOrphanRange(keyRange.maxKey, &keyRange)); +} + +TEST_F(TwoChunksWithGapCompoundKeyFixture, ChunkGapAndPendingOrphanedDataRanges) { + string errMsg; + ChunkType chunk; + unique_ptr<CollectionMetadata> cloned; + + chunk.setMin(BSON("a" << 20 << "b" << 0)); + chunk.setMax(BSON("a" << 30 << "b" << 0)); + + cloned.reset(getCollMetadata().clonePlusPending(chunk, &errMsg)); + ASSERT_EQUALS(errMsg, string("")); + ASSERT(cloned != NULL); + + KeyRange keyRange; + ASSERT(cloned->getNextOrphanRange(cloned->getMinKey(), &keyRange)); + ASSERT(keyRange.minKey.woCompare(cloned->getMinKey()) == 0); + ASSERT(keyRange.maxKey.woCompare(BSON("a" << 10 << "b" << 0)) == 0); + ASSERT(keyRange.keyPattern.woCompare(cloned->getKeyPattern()) == 0); + + ASSERT(cloned->getNextOrphanRange(keyRange.maxKey, &keyRange)); + ASSERT(keyRange.minKey.woCompare(BSON("a" << 40 << "b" << 0)) == 0); + ASSERT(keyRange.maxKey.woCompare(cloned->getMaxKey()) == 0); + ASSERT(keyRange.keyPattern.woCompare(cloned->getKeyPattern()) == 0); + + ASSERT(!cloned->getNextOrphanRange(keyRange.maxKey, &keyRange)); +} - OID epoch = OID::gen(); - ChunkVersion chunkVersion = ChunkVersion( 1, 0, epoch ); +/** + * Fixture with chunk containing: + * [min->10) , [10->20) , <gap> , [30->max) + */ +class ThreeChunkWithRangeGapFixture : public mongo::unittest::Test { +protected: + void setUp() { + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); + OID epoch(OID::gen()); + + { CollectionType collType; - collType.setNs(NamespaceString{"test.foo"}); + collType.setNs(NamespaceString{"x.y"}); collType.setKeyPattern(BSON("a" << 1)); collType.setUnique(false); collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); collType.setEpoch(epoch); _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); - - _dummyConfig->insert( - ChunkType::ConfigNS, - BSON(ChunkType::name("test.foo-a_10") << - ChunkType::ns("test.foo") << - ChunkType::min(BSON("a" << 10 << "b" << 0)) << - ChunkType::max(BSON("a" << 20 << "b" << 0)) << - ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch( - chunkVersion.toLong())) << - ChunkType::DEPRECATED_epoch(epoch) << - ChunkType::shard("shard0000")) ); - - _dummyConfig->insert( - ChunkType::ConfigNS, - BSON(ChunkType::name("test.foo-a_10") << - ChunkType::ns("test.foo") << - ChunkType::min(BSON("a" << 30 << "b" << 0)) << - ChunkType::max(BSON("a" << 40 << "b" << 0)) << - ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch( - chunkVersion.toLong())) << - ChunkType::DEPRECATED_epoch(epoch) << - ChunkType::shard("shard0000")) ); - - ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); - ASSERT(configLoc.isValid()); - CatalogManagerLegacy catalogManager; - catalogManager.init(configLoc); - - MetadataLoader loader; - Status status = loader.makeCollectionMetadata(&catalogManager, - "test.foo", - "shard0000", - NULL, - &_metadata); - ASSERT_OK(status); } - void tearDown() { - MockConnRegistry::get()->clear(); - } - - const CollectionMetadata& getCollMetadata() const { - return _metadata; - } - - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - CollectionMetadata _metadata; - }; - - TEST_F(TwoChunksWithGapCompoundKeyFixture, ClonePlusBasic) { - ChunkType chunk; - chunk.setMin( BSON("a" << 40 << "b" << 0) ); - chunk.setMax( BSON("a" << 50 << "b" << 0) ); - - string errMsg; - ChunkVersion version( 1, 0, getCollMetadata().getShardVersion().epoch() ); - unique_ptr<CollectionMetadata> cloned( getCollMetadata().clonePlusChunk( chunk, - version, - &errMsg ) ); - - ASSERT( errMsg.empty() ); - ASSERT_EQUALS( 2u, getCollMetadata().getNumChunks() ); - ASSERT_EQUALS( 3u, cloned->getNumChunks() ); - - // TODO: test maxShardVersion, maxCollVersion - - ASSERT_FALSE( cloned->keyBelongsToMe(BSON("a" << 25 << "b" << 0)) ); - ASSERT_FALSE( cloned->keyBelongsToMe(BSON("a" << 29 << "b" << 0)) ); - ASSERT( cloned->keyBelongsToMe(BSON("a" << 30 << "b" << 0)) ); - ASSERT( cloned->keyBelongsToMe(BSON("a" << 45 << "b" << 0)) ); - ASSERT( cloned->keyBelongsToMe(BSON("a" << 49 << "b" << 0)) ); - ASSERT_FALSE( cloned->keyBelongsToMe(BSON("a" << 50 << "b" << 0)) ); - } - - TEST_F(TwoChunksWithGapCompoundKeyFixture, ClonePlusOverlappingRange) { - ChunkType chunk; - chunk.setMin( BSON("a" << 15 << "b" << 0) ); - chunk.setMax( BSON("a" << 25 << "b" << 0) ); - - string errMsg; - unique_ptr<CollectionMetadata> cloned( getCollMetadata().clonePlusChunk( chunk, - ChunkVersion( 1, - 0, - OID() ), - &errMsg ) ); - ASSERT( cloned == NULL ); - ASSERT_FALSE( errMsg.empty() ); - ASSERT_EQUALS( 2u, getCollMetadata().getNumChunks() ); - } - - TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneMinusBasic) { - ChunkType chunk; - chunk.setMin( BSON("a" << 10 << "b" << 0) ); - chunk.setMax( BSON("a" << 20 << "b" << 0) ); - - string errMsg; - ChunkVersion version( 2, 0, OID() ); - unique_ptr<CollectionMetadata> cloned( getCollMetadata().cloneMigrate( chunk, - version, - &errMsg ) ); - - ASSERT( errMsg.empty() ); - ASSERT_EQUALS( 2u, getCollMetadata().getNumChunks() ); - ASSERT_EQUALS( 1u, cloned->getNumChunks() ); - - // TODO: test maxShardVersion, maxCollVersion - - ASSERT_FALSE( cloned->keyBelongsToMe(BSON("a" << 5 << "b" << 0)) ); - ASSERT_FALSE( cloned->keyBelongsToMe(BSON("a" << 15 << "b" << 0)) ); - ASSERT( cloned->keyBelongsToMe(BSON("a" << 30 << "b" << 0)) ); - ASSERT( cloned->keyBelongsToMe(BSON("a" << 35 << "b" << 0)) ); - ASSERT_FALSE( cloned->keyBelongsToMe(BSON("a" << 40 << "b" << 0)) ); - } - - TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneMinusNonExisting) { - ChunkType chunk; - chunk.setMin( BSON("a" << 25 << "b" << 0) ); - chunk.setMax( BSON("a" << 28 << "b" << 0) ); - - string errMsg; - unique_ptr<CollectionMetadata> cloned( getCollMetadata().cloneMigrate( chunk, - ChunkVersion( 1, - 0, - OID() ), - &errMsg ) ); - ASSERT( cloned == NULL ); - ASSERT_FALSE( errMsg.empty() ); - ASSERT_EQUALS( 2u, getCollMetadata().getNumChunks() ); - } - - TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneSplitBasic) { - const BSONObj min( BSON("a" << 10 << "b" << 0) ); - const BSONObj max( BSON("a" << 20 << "b" << 0) ); - - ChunkType chunk; - chunk.setMin( min ); - chunk.setMax( max ); - - const BSONObj split1( BSON("a" << 15 << "b" << 0) ); - const BSONObj split2( BSON("a" << 18 << "b" << 0) ); - vector<BSONObj> splitKeys; - splitKeys.push_back( split1 ); - splitKeys.push_back( split2 ); - ChunkVersion version( 1, 99, OID() ); // first chunk 1|99 , second 1|100 - - string errMsg; - unique_ptr<CollectionMetadata> cloned( getCollMetadata().cloneSplit( chunk, - splitKeys, - version, - &errMsg ) ); - - version.incMinor(); /* second chunk 1|100, first split point */ - version.incMinor(); /* third chunk 1|101, second split point */ - ASSERT_EQUALS( cloned->getShardVersion().toLong(), version.toLong() /* 1|101 */); - ASSERT_EQUALS( cloned->getCollVersion().toLong(), version.toLong() ); - ASSERT_EQUALS( getCollMetadata().getNumChunks(), 2u ); - ASSERT_EQUALS( cloned->getNumChunks(), 4u ); - ASSERT( cloned->keyBelongsToMe( min ) ); - ASSERT( cloned->keyBelongsToMe( split1 ) ); - ASSERT( cloned->keyBelongsToMe( split2 ) ); - ASSERT( !cloned->keyBelongsToMe( max ) ); - } - - TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneSplitOutOfRangeSplitPoint) { - ChunkType chunk; - chunk.setMin( BSON("a" << 10 << "b" << 0) ); - chunk.setMax( BSON("a" << 20 << "b" << 0) ); - - vector<BSONObj> splitKeys; - splitKeys.push_back( BSON("a" << 5 << "b" << 0) ); - - string errMsg; - unique_ptr<CollectionMetadata> cloned( getCollMetadata().cloneSplit( chunk, - splitKeys, - ChunkVersion( 1, - 0, - OID() ), - &errMsg ) ); - - ASSERT( cloned == NULL ); - ASSERT_FALSE( errMsg.empty() ); - ASSERT_EQUALS( 2u, getCollMetadata().getNumChunks() ); - } - - TEST_F(TwoChunksWithGapCompoundKeyFixture, CloneSplitBadChunkRange) { - const BSONObj min( BSON("a" << 10 << "b" << 0) ); - const BSONObj max( BSON("a" << 25 << "b" << 0) ); - - ChunkType chunk; - chunk.setMin( BSON("a" << 10 << "b" << 0) ); - chunk.setMax( BSON("a" << 25 << "b" << 0) ); - - vector<BSONObj> splitKeys; - splitKeys.push_back( BSON("a" << 15 << "b" << 0) ); - - string errMsg; - unique_ptr<CollectionMetadata> cloned( getCollMetadata().cloneSplit( chunk, - splitKeys, - ChunkVersion( 1, - 0, - OID() ), - &errMsg ) ); - - ASSERT( cloned == NULL ); - ASSERT_FALSE( errMsg.empty() ); - ASSERT_EQUALS( 2u, getCollMetadata().getNumChunks() ); - } - - TEST_F(TwoChunksWithGapCompoundKeyFixture, ChunkGapOrphanedDataRanges) { - - KeyRange keyRange; - ASSERT( getCollMetadata().getNextOrphanRange( getCollMetadata().getMinKey(), &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( getCollMetadata().getMinKey() ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( BSON( "a" << 10 << "b" << 0 ) ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( getCollMetadata().getKeyPattern() ) == 0 ); - - ASSERT( getCollMetadata().getNextOrphanRange( keyRange.maxKey, &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( BSON( "a" << 20 << "b" << 0 ) ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( BSON( "a" << 30 << "b" << 0 ) ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( getCollMetadata().getKeyPattern() ) == 0 ); - - ASSERT( getCollMetadata().getNextOrphanRange( keyRange.maxKey, &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( BSON( "a" << 40 << "b" << 0 ) ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( getCollMetadata().getMaxKey() ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( getCollMetadata().getKeyPattern() ) == 0 ); - - ASSERT( !getCollMetadata().getNextOrphanRange( keyRange.maxKey, &keyRange ) ); - } - - TEST_F(TwoChunksWithGapCompoundKeyFixture, ChunkGapAndPendingOrphanedDataRanges) { - - string errMsg; - ChunkType chunk; - unique_ptr<CollectionMetadata> cloned; - - chunk.setMin( BSON( "a" << 20 << "b" << 0 ) ); - chunk.setMax( BSON( "a" << 30 << "b" << 0 ) ); - - cloned.reset( getCollMetadata().clonePlusPending( chunk, &errMsg ) ); - ASSERT_EQUALS( errMsg, string("") ); - ASSERT( cloned != NULL ); - - KeyRange keyRange; - ASSERT( cloned->getNextOrphanRange( cloned->getMinKey(), &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( cloned->getMinKey() ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( BSON( "a" << 10 << "b" << 0 ) ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( cloned->getKeyPattern() ) == 0 ); - - ASSERT( cloned->getNextOrphanRange( keyRange.maxKey, &keyRange ) ); - ASSERT( keyRange.minKey.woCompare( BSON( "a" << 40 << "b" << 0 ) ) == 0 ); - ASSERT( keyRange.maxKey.woCompare( cloned->getMaxKey() ) == 0 ); - ASSERT( keyRange.keyPattern.woCompare( cloned->getKeyPattern() ) == 0 ); - - ASSERT( !cloned->getNextOrphanRange( keyRange.maxKey, &keyRange ) ); - } - - /** - * Fixture with chunk containing: - * [min->10) , [10->20) , <gap> , [30->max) - */ - class ThreeChunkWithRangeGapFixture : public mongo::unittest::Test { - protected: - void setUp() { - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); - - OID epoch(OID::gen()); - - { - CollectionType collType; - collType.setNs(NamespaceString{"x.y"}); - collType.setKeyPattern(BSON("a" << 1)); - collType.setUnique(false); - collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); - collType.setEpoch(epoch); - _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); - } - - { - ChunkVersion version( 1, 1, epoch ); - _dummyConfig->insert( ChunkType::ConfigNS, BSON(ChunkType::name("x.y-a_MinKey") << - ChunkType::ns("x.y") << - ChunkType::min(BSON("a" << MINKEY)) << - ChunkType::max(BSON("a" << 10)) << - ChunkType::DEPRECATED_lastmod( - Date_t::fromMillisSinceEpoch(version.toLong())) << - ChunkType::DEPRECATED_epoch(version.epoch()) << - ChunkType::shard("shard0000")) ); - } - - { - ChunkVersion version( 1, 3, epoch ); - _dummyConfig->insert( ChunkType::ConfigNS, BSON(ChunkType::name("x.y-a_10") << - ChunkType::ns("x.y") << - ChunkType::min(BSON("a" << 10)) << - ChunkType::max(BSON("a" << 20)) << - ChunkType::DEPRECATED_lastmod( - Date_t::fromMillisSinceEpoch(version.toLong())) << - ChunkType::DEPRECATED_epoch(version.epoch()) << - ChunkType::shard("shard0000")) ); - } - - { - ChunkVersion version( 1, 2, epoch ); - _dummyConfig->insert( ChunkType::ConfigNS, BSON(ChunkType::name("x.y-a_30") << - ChunkType::ns("x.y") << - ChunkType::min(BSON("a" << 30)) << - ChunkType::max(BSON("a" << MAXKEY)) << - ChunkType::DEPRECATED_lastmod( - Date_t::fromMillisSinceEpoch(version.toLong())) << - ChunkType::DEPRECATED_epoch(version.epoch()) << - ChunkType::shard("shard0000")) ); - } - - ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); - ASSERT(configLoc.isValid()); - CatalogManagerLegacy catalogManager; - catalogManager.init(configLoc); - - MetadataLoader loader; - Status status = loader.makeCollectionMetadata(&catalogManager, - "test.foo", - "shard0000", - NULL, - &_metadata); - ASSERT_OK(status); + { + ChunkVersion version(1, 1, epoch); + _dummyConfig->insert(ChunkType::ConfigNS, + BSON(ChunkType::name("x.y-a_MinKey") + << ChunkType::ns("x.y") << ChunkType::min(BSON("a" << MINKEY)) + << ChunkType::max(BSON("a" << 10)) + << ChunkType::DEPRECATED_lastmod( + Date_t::fromMillisSinceEpoch(version.toLong())) + << ChunkType::DEPRECATED_epoch(version.epoch()) + << ChunkType::shard("shard0000"))); } - void tearDown() { - MockConnRegistry::get()->clear(); + { + ChunkVersion version(1, 3, epoch); + _dummyConfig->insert(ChunkType::ConfigNS, + BSON(ChunkType::name("x.y-a_10") + << ChunkType::ns("x.y") << ChunkType::min(BSON("a" << 10)) + << ChunkType::max(BSON("a" << 20)) + << ChunkType::DEPRECATED_lastmod( + Date_t::fromMillisSinceEpoch(version.toLong())) + << ChunkType::DEPRECATED_epoch(version.epoch()) + << ChunkType::shard("shard0000"))); } - const CollectionMetadata& getCollMetadata() const { - return _metadata; + { + ChunkVersion version(1, 2, epoch); + _dummyConfig->insert(ChunkType::ConfigNS, + BSON(ChunkType::name("x.y-a_30") + << ChunkType::ns("x.y") << ChunkType::min(BSON("a" << 30)) + << ChunkType::max(BSON("a" << MAXKEY)) + << ChunkType::DEPRECATED_lastmod( + Date_t::fromMillisSinceEpoch(version.toLong())) + << ChunkType::DEPRECATED_epoch(version.epoch()) + << ChunkType::shard("shard0000"))); } - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - CollectionMetadata _metadata; - }; + ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); + ASSERT(configLoc.isValid()); + CatalogManagerLegacy catalogManager; + catalogManager.init(configLoc); - TEST_F(ThreeChunkWithRangeGapFixture, ShardOwnsDoc) { - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 5)) ); - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 10)) ); - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 30)) ); - ASSERT( getCollMetadata().keyBelongsToMe(BSON("a" << 40)) ); + MetadataLoader loader; + Status status = loader.makeCollectionMetadata( + &catalogManager, "test.foo", "shard0000", NULL, &_metadata); + ASSERT_OK(status); } - TEST_F(ThreeChunkWithRangeGapFixture, ShardDoesntOwnDoc) { - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << 25)) ); - ASSERT_FALSE( getCollMetadata().keyBelongsToMe(BSON("a" << MAXKEY)) ); + void tearDown() { + MockConnRegistry::get()->clear(); } - TEST_F(ThreeChunkWithRangeGapFixture, GetNextFromEmpty) { - ChunkType nextChunk; - ASSERT( getCollMetadata().getNextChunk( getCollMetadata().getMinKey(), &nextChunk ) ); - ASSERT_EQUALS( 0, nextChunk.getMin().woCompare(BSON("a" << MINKEY)) ); - ASSERT_EQUALS( 0, nextChunk.getMax().woCompare(BSON("a" << 10)) ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, GetNextFromMiddle) { - ChunkType nextChunk; - ASSERT( getCollMetadata().getNextChunk(BSON("a" << 20), &nextChunk) ); - ASSERT_EQUALS( 0, nextChunk.getMin().woCompare(BSON("a" << 30)) ); - ASSERT_EQUALS( 0, nextChunk.getMax().woCompare(BSON("a" << MAXKEY)) ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, GetNextFromLast) { - ChunkType nextChunk; - ASSERT( getCollMetadata().getNextChunk(BSON("a" << 30), &nextChunk) ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkHoleInRange) { - - string errMsg; - unique_ptr<CollectionMetadata> cloned; - - // Try to merge with hole in range - ChunkVersion newShardVersion( 5, 0, getCollMetadata().getShardVersion().epoch() ); - cloned.reset( getCollMetadata().cloneMerge( BSON( "a" << 10 ), - BSON( "a" << MAXKEY ), - newShardVersion, - &errMsg ) ); - - ASSERT_NOT_EQUALS( errMsg, "" ); - ASSERT( cloned == NULL ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkDiffEndKey) { - - string errMsg; - unique_ptr<CollectionMetadata> cloned; - - // Try to merge with different end key - ChunkVersion newShardVersion( 5, 0, getCollMetadata().getShardVersion().epoch() ); - cloned.reset( getCollMetadata().cloneMerge( BSON( "a" << MINKEY ), - BSON( "a" << 19 ), - newShardVersion, - &errMsg ) ); - - ASSERT_NOT_EQUALS( errMsg, "" ); - ASSERT( cloned == NULL ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkMinKey) { - - string errMsg; - unique_ptr<CollectionMetadata> cloned; - - ASSERT_EQUALS( getCollMetadata().getNumChunks(), 3u ); - - // Try to merge lowest chunks together - ChunkVersion newShardVersion( 5, 0, getCollMetadata().getShardVersion().epoch() ); - cloned.reset( getCollMetadata().cloneMerge( BSON( "a" << MINKEY ), - BSON( "a" << 20 ), - newShardVersion, - &errMsg ) ); - - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); - ASSERT( cloned->keyBelongsToMe( BSON( "a" << 10 ) ) ); - ASSERT_EQUALS( cloned->getNumChunks(), 2u ); - ASSERT_EQUALS( cloned->getShardVersion().majorVersion(), 5 ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkMaxKey) { - - string errMsg; - unique_ptr<CollectionMetadata> cloned; - ChunkVersion newShardVersion( 5, 0, getCollMetadata().getShardVersion().epoch() ); - - // Add one chunk to complete the range - ChunkType chunk; - chunk.setMin( BSON( "a" << 20 ) ); - chunk.setMax( BSON( "a" << 30 ) ); - cloned.reset( getCollMetadata().clonePlusChunk( chunk, - newShardVersion, - &errMsg ) ); - ASSERT_EQUALS( errMsg, "" ); - ASSERT_EQUALS( cloned->getNumChunks(), 4u ); - ASSERT( cloned != NULL ); - - // Try to merge highest chunks together - newShardVersion.incMajor(); - cloned.reset( cloned->cloneMerge( BSON( "a" << 20 ), - BSON( "a" << MAXKEY ), - newShardVersion, - &errMsg ) ); - - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); - ASSERT( cloned->keyBelongsToMe( BSON( "a" << 30 ) ) ); - ASSERT_EQUALS( cloned->getNumChunks(), 3u ); - ASSERT_EQUALS( cloned->getShardVersion().majorVersion(), 6 ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkFullRange) { - - string errMsg; - unique_ptr<CollectionMetadata> cloned; - ChunkVersion newShardVersion( 5, 0, getCollMetadata().getShardVersion().epoch() ); - - // Add one chunk to complete the range - ChunkType chunk; - chunk.setMin( BSON( "a" << 20 ) ); - chunk.setMax( BSON( "a" << 30 ) ); - cloned.reset( getCollMetadata().clonePlusChunk( chunk, - newShardVersion, - &errMsg ) ); - ASSERT_EQUALS( errMsg, "" ); - ASSERT_EQUALS( cloned->getNumChunks(), 4u ); - ASSERT( cloned != NULL ); - - // Try to merge all chunks together - newShardVersion.incMajor(); - cloned.reset( cloned->cloneMerge( BSON( "a" << MINKEY ), - BSON( "a" << MAXKEY ), - newShardVersion, - &errMsg ) ); - - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); - ASSERT( cloned->keyBelongsToMe( BSON( "a" << 10 ) ) ); - ASSERT( cloned->keyBelongsToMe( BSON( "a" << 30 ) ) ); - ASSERT_EQUALS( cloned->getNumChunks(), 1u ); - ASSERT_EQUALS( cloned->getShardVersion().majorVersion(), 6 ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkMiddleRange) { - - string errMsg; - unique_ptr<CollectionMetadata> cloned; - ChunkVersion newShardVersion( 5, 0, getCollMetadata().getShardVersion().epoch() ); - - // Add one chunk to complete the range - ChunkType chunk; - chunk.setMin( BSON( "a" << 20 ) ); - chunk.setMax( BSON( "a" << 30 ) ); - cloned.reset( getCollMetadata().clonePlusChunk( chunk, - newShardVersion, - &errMsg ) ); - ASSERT_EQUALS( errMsg, "" ); - ASSERT_EQUALS( cloned->getNumChunks(), 4u ); - ASSERT( cloned != NULL ); - - // Try to merge middle two chunks - newShardVersion.incMajor(); - cloned.reset( cloned->cloneMerge( BSON( "a" << 10 ), - BSON( "a" << 30 ), - newShardVersion, - &errMsg ) ); - - ASSERT_EQUALS( errMsg, "" ); - ASSERT( cloned != NULL ); - ASSERT( cloned->keyBelongsToMe( BSON( "a" << 20 ) ) ); - ASSERT_EQUALS( cloned->getNumChunks(), 3u ); - ASSERT_EQUALS( cloned->getShardVersion().majorVersion(), 6 ); - } - - TEST_F(ThreeChunkWithRangeGapFixture, CannotMergeWithHole) { - - string errMsg; - ChunkVersion newShardVersion( 5, 0, getCollMetadata().getShardVersion().epoch() ); - - // Try to merge middle two chunks with a hole in the middle. - newShardVersion.incMajor(); - CollectionMetadata* result = getCollMetadata().cloneMerge( BSON( "a" << 10 ), - BSON( "a" << 30 ), - newShardVersion, - &errMsg ); - ASSERT( result == NULL ); - ASSERT( !errMsg.empty() ); + const CollectionMetadata& getCollMetadata() const { + return _metadata; } -} // unnamed namespace +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; + CollectionMetadata _metadata; +}; + +TEST_F(ThreeChunkWithRangeGapFixture, ShardOwnsDoc) { + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 5))); + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 10))); + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 30))); + ASSERT(getCollMetadata().keyBelongsToMe(BSON("a" << 40))); +} + +TEST_F(ThreeChunkWithRangeGapFixture, ShardDoesntOwnDoc) { + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << 25))); + ASSERT_FALSE(getCollMetadata().keyBelongsToMe(BSON("a" << MAXKEY))); +} + +TEST_F(ThreeChunkWithRangeGapFixture, GetNextFromEmpty) { + ChunkType nextChunk; + ASSERT(getCollMetadata().getNextChunk(getCollMetadata().getMinKey(), &nextChunk)); + ASSERT_EQUALS(0, nextChunk.getMin().woCompare(BSON("a" << MINKEY))); + ASSERT_EQUALS(0, nextChunk.getMax().woCompare(BSON("a" << 10))); +} + +TEST_F(ThreeChunkWithRangeGapFixture, GetNextFromMiddle) { + ChunkType nextChunk; + ASSERT(getCollMetadata().getNextChunk(BSON("a" << 20), &nextChunk)); + ASSERT_EQUALS(0, nextChunk.getMin().woCompare(BSON("a" << 30))); + ASSERT_EQUALS(0, nextChunk.getMax().woCompare(BSON("a" << MAXKEY))); +} + +TEST_F(ThreeChunkWithRangeGapFixture, GetNextFromLast) { + ChunkType nextChunk; + ASSERT(getCollMetadata().getNextChunk(BSON("a" << 30), &nextChunk)); +} + +TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkHoleInRange) { + string errMsg; + unique_ptr<CollectionMetadata> cloned; + + // Try to merge with hole in range + ChunkVersion newShardVersion(5, 0, getCollMetadata().getShardVersion().epoch()); + cloned.reset(getCollMetadata().cloneMerge( + BSON("a" << 10), BSON("a" << MAXKEY), newShardVersion, &errMsg)); + + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(cloned == NULL); +} + +TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkDiffEndKey) { + string errMsg; + unique_ptr<CollectionMetadata> cloned; + + // Try to merge with different end key + ChunkVersion newShardVersion(5, 0, getCollMetadata().getShardVersion().epoch()); + cloned.reset(getCollMetadata().cloneMerge( + BSON("a" << MINKEY), BSON("a" << 19), newShardVersion, &errMsg)); + + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(cloned == NULL); +} + +TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkMinKey) { + string errMsg; + unique_ptr<CollectionMetadata> cloned; + + ASSERT_EQUALS(getCollMetadata().getNumChunks(), 3u); + + // Try to merge lowest chunks together + ChunkVersion newShardVersion(5, 0, getCollMetadata().getShardVersion().epoch()); + cloned.reset(getCollMetadata().cloneMerge( + BSON("a" << MINKEY), BSON("a" << 20), newShardVersion, &errMsg)); + + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 10))); + ASSERT_EQUALS(cloned->getNumChunks(), 2u); + ASSERT_EQUALS(cloned->getShardVersion().majorVersion(), 5); +} + +TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkMaxKey) { + string errMsg; + unique_ptr<CollectionMetadata> cloned; + ChunkVersion newShardVersion(5, 0, getCollMetadata().getShardVersion().epoch()); + + // Add one chunk to complete the range + ChunkType chunk; + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 30)); + cloned.reset(getCollMetadata().clonePlusChunk(chunk, newShardVersion, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_EQUALS(cloned->getNumChunks(), 4u); + ASSERT(cloned != NULL); + + // Try to merge highest chunks together + newShardVersion.incMajor(); + cloned.reset( + cloned->cloneMerge(BSON("a" << 20), BSON("a" << MAXKEY), newShardVersion, &errMsg)); + + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 30))); + ASSERT_EQUALS(cloned->getNumChunks(), 3u); + ASSERT_EQUALS(cloned->getShardVersion().majorVersion(), 6); +} + +TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkFullRange) { + string errMsg; + unique_ptr<CollectionMetadata> cloned; + ChunkVersion newShardVersion(5, 0, getCollMetadata().getShardVersion().epoch()); + + // Add one chunk to complete the range + ChunkType chunk; + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 30)); + cloned.reset(getCollMetadata().clonePlusChunk(chunk, newShardVersion, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_EQUALS(cloned->getNumChunks(), 4u); + ASSERT(cloned != NULL); + + // Try to merge all chunks together + newShardVersion.incMajor(); + cloned.reset( + cloned->cloneMerge(BSON("a" << MINKEY), BSON("a" << MAXKEY), newShardVersion, &errMsg)); + + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 10))); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 30))); + ASSERT_EQUALS(cloned->getNumChunks(), 1u); + ASSERT_EQUALS(cloned->getShardVersion().majorVersion(), 6); +} + +TEST_F(ThreeChunkWithRangeGapFixture, MergeChunkMiddleRange) { + string errMsg; + unique_ptr<CollectionMetadata> cloned; + ChunkVersion newShardVersion(5, 0, getCollMetadata().getShardVersion().epoch()); + + // Add one chunk to complete the range + ChunkType chunk; + chunk.setMin(BSON("a" << 20)); + chunk.setMax(BSON("a" << 30)); + cloned.reset(getCollMetadata().clonePlusChunk(chunk, newShardVersion, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_EQUALS(cloned->getNumChunks(), 4u); + ASSERT(cloned != NULL); + + // Try to merge middle two chunks + newShardVersion.incMajor(); + cloned.reset(cloned->cloneMerge(BSON("a" << 10), BSON("a" << 30), newShardVersion, &errMsg)); + + ASSERT_EQUALS(errMsg, ""); + ASSERT(cloned != NULL); + ASSERT(cloned->keyBelongsToMe(BSON("a" << 20))); + ASSERT_EQUALS(cloned->getNumChunks(), 3u); + ASSERT_EQUALS(cloned->getShardVersion().majorVersion(), 6); +} + +TEST_F(ThreeChunkWithRangeGapFixture, CannotMergeWithHole) { + string errMsg; + ChunkVersion newShardVersion(5, 0, getCollMetadata().getShardVersion().epoch()); + + // Try to merge middle two chunks with a hole in the middle. + newShardVersion.incMajor(); + CollectionMetadata* result = + getCollMetadata().cloneMerge(BSON("a" << 10), BSON("a" << 30), newShardVersion, &errMsg); + ASSERT(result == NULL); + ASSERT(!errMsg.empty()); +} + +} // unnamed namespace diff --git a/src/mongo/s/commands/cluster_add_shard_cmd.cpp b/src/mongo/s/commands/cluster_add_shard_cmd.cpp index 8fdeb3f5b4d..78c2a693846 100644 --- a/src/mongo/s/commands/cluster_add_shard_cmd.cpp +++ b/src/mongo/s/commands/cluster_add_shard_cmd.cpp @@ -41,105 +41,102 @@ namespace mongo { - using std::string; +using std::string; namespace { - class AddShardCmd : public Command { - public: - AddShardCmd() : Command("addShard", false, "addshard") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; - } - - virtual bool isWriteCommandForConfigServer() const { +class AddShardCmd : public Command { +public: + AddShardCmd() : Command("addShard", false, "addshard") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "add a new shard to the system"; + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::addShard); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + // get replica set component hosts + ConnectionString servers = + ConnectionString::parse(cmdObj.firstElement().valuestrsafe(), errmsg); + if (!errmsg.empty()) { + log() << "addshard request " << cmdObj << " failed: " << errmsg; return false; } - virtual void help(std::stringstream& help) const { - help << "add a new shard to the system"; - } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - - ActionSet actions; - actions.addAction(ActionType::addShard); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - // get replica set component hosts - ConnectionString servers = ConnectionString::parse( - cmdObj.firstElement().valuestrsafe(), errmsg); - if (!errmsg.empty()) { - log() << "addshard request " << cmdObj << " failed: " << errmsg; + // using localhost in server names implies every other process must use localhost addresses too + std::vector<HostAndPort> serverAddrs = servers.getServers(); + for (size_t i = 0; i < serverAddrs.size(); i++) { + if (serverAddrs[i].isLocalHost() != grid.allowLocalHost()) { + errmsg = str::stream() + << "Can't use localhost as a shard since all shards need to" + << " communicate. Either use all shards and configdbs in localhost" + << " or all in actual IPs. host: " << serverAddrs[i].toString() + << " isLocalHost:" << serverAddrs[i].isLocalHost(); + + log() << "addshard request " << cmdObj + << " failed: attempt to mix localhosts and IPs"; return false; } - // using localhost in server names implies every other process must use localhost addresses too - std::vector<HostAndPort> serverAddrs = servers.getServers(); - for (size_t i = 0; i < serverAddrs.size(); i++) { - if (serverAddrs[i].isLocalHost() != grid.allowLocalHost()) { - errmsg = str::stream() << - "Can't use localhost as a shard since all shards need to" << - " communicate. Either use all shards and configdbs in localhost" << - " or all in actual IPs. host: " << serverAddrs[i].toString() << - " isLocalHost:" << serverAddrs[i].isLocalHost(); - - log() << "addshard request " << cmdObj - << " failed: attempt to mix localhosts and IPs"; - return false; - } - - // it's fine if mongods of a set all use default port - if (!serverAddrs[i].hasPort()) { - serverAddrs[i] = HostAndPort(serverAddrs[i].host(), - ServerGlobalParams::ShardServerPort); - } + // it's fine if mongods of a set all use default port + if (!serverAddrs[i].hasPort()) { + serverAddrs[i] = + HostAndPort(serverAddrs[i].host(), ServerGlobalParams::ShardServerPort); } + } - // name is optional; addShard will provide one if needed - string name = ""; - if (cmdObj["name"].type() == String) { - name = cmdObj["name"].valuestrsafe(); - } + // name is optional; addShard will provide one if needed + string name = ""; + if (cmdObj["name"].type() == String) { + name = cmdObj["name"].valuestrsafe(); + } - // maxSize is the space usage cap in a shard in MBs - long long maxSize = 0; - if (cmdObj[ShardType::maxSizeMB()].isNumber()) { - maxSize = cmdObj[ShardType::maxSizeMB()].numberLong(); - } + // maxSize is the space usage cap in a shard in MBs + long long maxSize = 0; + if (cmdObj[ShardType::maxSizeMB()].isNumber()) { + maxSize = cmdObj[ShardType::maxSizeMB()].numberLong(); + } - audit::logAddShard(ClientBasic::getCurrent(), name, servers.toString(), maxSize); + audit::logAddShard(ClientBasic::getCurrent(), name, servers.toString(), maxSize); - StatusWith<string> addShardResult = - grid.catalogManager()->addShard(name, servers, maxSize); - if (!addShardResult.isOK()) { - log() << "addShard request '" << cmdObj << "'" - << " failed: " << addShardResult.getStatus().reason(); - return appendCommandStatus(result, addShardResult.getStatus()); - } + StatusWith<string> addShardResult = grid.catalogManager()->addShard(name, servers, maxSize); + if (!addShardResult.isOK()) { + log() << "addShard request '" << cmdObj << "'" + << " failed: " << addShardResult.getStatus().reason(); + return appendCommandStatus(result, addShardResult.getStatus()); + } - result << "shardAdded" << addShardResult.getValue(); + result << "shardAdded" << addShardResult.getValue(); - return true; - } + return true; + } - } addShard; +} addShard; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_commands_common.cpp b/src/mongo/s/commands/cluster_commands_common.cpp index 52afad182f5..f390a17d992 100644 --- a/src/mongo/s/commands/cluster_commands_common.cpp +++ b/src/mongo/s/commands/cluster_commands_common.cpp @@ -36,79 +36,76 @@ namespace mongo { - int getUniqueCodeFromCommandResults(const std::vector<Strategy::CommandResult>& results) { - int commonErrCode = -1; - for (std::vector<Strategy::CommandResult>::const_iterator it = results.begin(); - it != results.end(); - ++it) { - - // Only look at shards with errors. - if (!it->result["ok"].trueValue()) { - int errCode = it->result["code"].numberInt(); - - if (commonErrCode == -1) { - commonErrCode = errCode; - } - else if (commonErrCode != errCode) { - // At least two shards with errors disagree on the error code - commonErrCode = 0; - } +int getUniqueCodeFromCommandResults(const std::vector<Strategy::CommandResult>& results) { + int commonErrCode = -1; + for (std::vector<Strategy::CommandResult>::const_iterator it = results.begin(); + it != results.end(); + ++it) { + // Only look at shards with errors. + if (!it->result["ok"].trueValue()) { + int errCode = it->result["code"].numberInt(); + + if (commonErrCode == -1) { + commonErrCode = errCode; + } else if (commonErrCode != errCode) { + // At least two shards with errors disagree on the error code + commonErrCode = 0; } } + } - // If no error encountered or shards with errors disagree on the error code, return 0 - if (commonErrCode == -1 || commonErrCode == 0) { - return 0; - } - - // Otherwise, shards with errors agree on the error code; return that code - return commonErrCode; + // If no error encountered or shards with errors disagree on the error code, return 0 + if (commonErrCode == -1 || commonErrCode == 0) { + return 0; } - bool appendEmptyResultSet(BSONObjBuilder& result, Status status, const std::string& ns) { - invariant(!status.isOK()); + // Otherwise, shards with errors agree on the error code; return that code + return commonErrCode; +} - if (status == ErrorCodes::DatabaseNotFound) { - // Old style reply - result << "result" << BSONArray(); +bool appendEmptyResultSet(BSONObjBuilder& result, Status status, const std::string& ns) { + invariant(!status.isOK()); - // New (command) style reply - appendCursorResponseObject(0LL, ns, BSONArray(), &result); + if (status == ErrorCodes::DatabaseNotFound) { + // Old style reply + result << "result" << BSONArray(); - return true; - } + // New (command) style reply + appendCursorResponseObject(0LL, ns, BSONArray(), &result); - return Command::appendCommandStatus(result, status); + return true; } - Status storePossibleCursor(const std::string& server, const BSONObj& cmdResult) { - if (cmdResult["ok"].trueValue() && cmdResult.hasField("cursor")) { - BSONElement cursorIdElt = cmdResult.getFieldDotted("cursor.id"); + return Command::appendCommandStatus(result, status); +} + +Status storePossibleCursor(const std::string& server, const BSONObj& cmdResult) { + if (cmdResult["ok"].trueValue() && cmdResult.hasField("cursor")) { + BSONElement cursorIdElt = cmdResult.getFieldDotted("cursor.id"); - if (cursorIdElt.type() != mongo::NumberLong) { + if (cursorIdElt.type() != mongo::NumberLong) { + return Status(ErrorCodes::TypeMismatch, + str::stream() << "expected \"cursor.id\" field from shard " + << "response to have NumberLong type, instead " + << "got: " << typeName(cursorIdElt.type())); + } + + const long long cursorId = cursorIdElt.Long(); + if (cursorId != 0) { + BSONElement cursorNsElt = cmdResult.getFieldDotted("cursor.ns"); + if (cursorNsElt.type() != mongo::String) { return Status(ErrorCodes::TypeMismatch, - str::stream() << "expected \"cursor.id\" field from shard " - << "response to have NumberLong type, instead " - << "got: " << typeName(cursorIdElt.type())); + str::stream() << "expected \"cursor.ns\" field from " + << "shard response to have String type, " + << "instead got: " << typeName(cursorNsElt.type())); } - const long long cursorId = cursorIdElt.Long(); - if (cursorId != 0) { - BSONElement cursorNsElt = cmdResult.getFieldDotted("cursor.ns"); - if (cursorNsElt.type() != mongo::String) { - return Status(ErrorCodes::TypeMismatch, - str::stream() << "expected \"cursor.ns\" field from " - << "shard response to have String type, " - << "instead got: " - << typeName(cursorNsElt.type())); - } - - const std::string cursorNs = cursorNsElt.String(); - cursorCache.storeRef(server, cursorId, cursorNs); - } + const std::string cursorNs = cursorNsElt.String(); + cursorCache.storeRef(server, cursorId, cursorNs); } - - return Status::OK(); } -} // namespace mongo + return Status::OK(); +} + +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_commands_common.h b/src/mongo/s/commands/cluster_commands_common.h index bda035f84c6..d06e13c63f2 100644 --- a/src/mongo/s/commands/cluster_commands_common.h +++ b/src/mongo/s/commands/cluster_commands_common.h @@ -35,27 +35,27 @@ namespace mongo { - class BSONObj; - - /** - * Utility function to compute a single error code from a vector of command results. - * - * @return If there is an error code common to all of the error results, returns that error - * code; otherwise, returns 0. - */ - int getUniqueCodeFromCommandResults(const std::vector<Strategy::CommandResult>& results); - - /** - * Utility function to return an empty result set from a command. - */ - bool appendEmptyResultSet(BSONObjBuilder& result, Status status, const std::string& ns); - - /** - * Utility function to parse a cursor command response and save the cursor in the CursorCache - * "refs" container. Returns Status::OK() if the cursor was successfully saved or no cursor - * was specified in the command response, and returns an error Status if a parsing error was - * encountered. - */ - Status storePossibleCursor(const std::string& server, const BSONObj& cmdResult); - -} // namespace mongo +class BSONObj; + +/** + * Utility function to compute a single error code from a vector of command results. + * + * @return If there is an error code common to all of the error results, returns that error + * code; otherwise, returns 0. + */ +int getUniqueCodeFromCommandResults(const std::vector<Strategy::CommandResult>& results); + +/** + * Utility function to return an empty result set from a command. + */ +bool appendEmptyResultSet(BSONObjBuilder& result, Status status, const std::string& ns); + +/** + * Utility function to parse a cursor command response and save the cursor in the CursorCache + * "refs" container. Returns Status::OK() if the cursor was successfully saved or no cursor + * was specified in the command response, and returns an error Status if a parsing error was + * encountered. + */ +Status storePossibleCursor(const std::string& server, const BSONObj& cmdResult); + +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_count_cmd.cpp b/src/mongo/s/commands/cluster_count_cmd.cpp index 6d7d3f22186..9db1762b16a 100644 --- a/src/mongo/s/commands/cluster_count_cmd.cpp +++ b/src/mongo/s/commands/cluster_count_cmd.cpp @@ -38,210 +38,191 @@ namespace mongo { - using std::string; - using std::vector; +using std::string; +using std::vector; namespace { - long long applySkipLimit(long long num, const BSONObj& cmd) { - BSONElement s = cmd["skip"]; - BSONElement l = cmd["limit"]; +long long applySkipLimit(long long num, const BSONObj& cmd) { + BSONElement s = cmd["skip"]; + BSONElement l = cmd["limit"]; - if (s.isNumber()) { - num = num - s.numberLong(); - if (num < 0) { - num = 0; - } + if (s.isNumber()) { + num = num - s.numberLong(); + if (num < 0) { + num = 0; } + } - if (l.isNumber()) { - long long limit = l.numberLong(); - if (limit < 0){ - limit = -limit; - } - - // 0 limit means no limit - if (limit < num && limit != 0) { - num = limit; - } + if (l.isNumber()) { + long long limit = l.numberLong(); + if (limit < 0) { + limit = -limit; } - return num; + // 0 limit means no limit + if (limit < num && limit != 0) { + num = limit; + } } + return num; +} - class ClusterCountCmd : public Command { - public: - ClusterCountCmd() : Command("count", false) { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return false; - } - virtual bool isWriteCommandForConfigServer() const { - return false; - } +class ClusterCountCmd : public Command { +public: + ClusterCountCmd() : Command("count", false) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + virtual bool slaveOk() const { + return true; + } - ActionSet actions; - actions.addAction(ActionType::find); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } + virtual bool adminOnly() const { + return false; + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + virtual bool isWriteCommandForConfigServer() const { + return false; + } - long long skip = 0; + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::find); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } - if (cmdObj["skip"].isNumber()) { - skip = cmdObj["skip"].numberLong(); - if (skip < 0) { - errmsg = "skip value is negative in count query"; - return false; - } - } - else if (cmdObj["skip"].ok()) { - errmsg = "skip value is not a valid number"; + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + long long skip = 0; + + if (cmdObj["skip"].isNumber()) { + skip = cmdObj["skip"].numberLong(); + if (skip < 0) { + errmsg = "skip value is negative in count query"; return false; } + } else if (cmdObj["skip"].ok()) { + errmsg = "skip value is not a valid number"; + return false; + } - const string collection = cmdObj.firstElement().valuestrsafe(); - const string fullns = dbname + "." + collection; - - BSONObjBuilder countCmdBuilder; - countCmdBuilder.append("count", collection); + const string collection = cmdObj.firstElement().valuestrsafe(); + const string fullns = dbname + "." + collection; - BSONObj filter; - if (cmdObj["query"].isABSONObj()) { - countCmdBuilder.append("query", cmdObj["query"].Obj()); - filter = cmdObj["query"].Obj(); - } + BSONObjBuilder countCmdBuilder; + countCmdBuilder.append("count", collection); - if (cmdObj["limit"].isNumber()) { - long long limit = cmdObj["limit"].numberLong(); - - // We only need to factor in the skip value when sending to the shards if we - // have a value for limit, otherwise, we apply it only once we have collected all - // counts. - if (limit != 0 && cmdObj["skip"].isNumber()) { - if (limit > 0) - limit += skip; - else - limit -= skip; - } + BSONObj filter; + if (cmdObj["query"].isABSONObj()) { + countCmdBuilder.append("query", cmdObj["query"].Obj()); + filter = cmdObj["query"].Obj(); + } - countCmdBuilder.append("limit", limit); + if (cmdObj["limit"].isNumber()) { + long long limit = cmdObj["limit"].numberLong(); + + // We only need to factor in the skip value when sending to the shards if we + // have a value for limit, otherwise, we apply it only once we have collected all + // counts. + if (limit != 0 && cmdObj["skip"].isNumber()) { + if (limit > 0) + limit += skip; + else + limit -= skip; } - if (cmdObj.hasField("hint")) { - countCmdBuilder.append(cmdObj["hint"]); - } + countCmdBuilder.append("limit", limit); + } - if (cmdObj.hasField("$queryOptions")) { - countCmdBuilder.append(cmdObj["$queryOptions"]); - } + if (cmdObj.hasField("hint")) { + countCmdBuilder.append(cmdObj["hint"]); + } - if (cmdObj.hasField(LiteParsedQuery::cmdOptionMaxTimeMS)) { - countCmdBuilder.append(cmdObj[LiteParsedQuery::cmdOptionMaxTimeMS]); - } + if (cmdObj.hasField("$queryOptions")) { + countCmdBuilder.append(cmdObj["$queryOptions"]); + } + + if (cmdObj.hasField(LiteParsedQuery::cmdOptionMaxTimeMS)) { + countCmdBuilder.append(cmdObj[LiteParsedQuery::cmdOptionMaxTimeMS]); + } - vector<Strategy::CommandResult> countResult; - Strategy::commandOp(dbname, - countCmdBuilder.done(), - options, - fullns, - filter, - &countResult); + vector<Strategy::CommandResult> countResult; + Strategy::commandOp(dbname, countCmdBuilder.done(), options, fullns, filter, &countResult); - long long total = 0; - BSONObjBuilder shardSubTotal(result.subobjStart("shards")); + long long total = 0; + BSONObjBuilder shardSubTotal(result.subobjStart("shards")); - for (vector<Strategy::CommandResult>::const_iterator iter = countResult.begin(); - iter != countResult.end(); - ++iter) { + for (vector<Strategy::CommandResult>::const_iterator iter = countResult.begin(); + iter != countResult.end(); + ++iter) { + const string& shardName = iter->shardTargetId; - const string& shardName = iter->shardTargetId; + if (iter->result["ok"].trueValue()) { + long long shardCount = iter->result["n"].numberLong(); - if (iter->result["ok"].trueValue()) { - long long shardCount = iter->result["n"].numberLong(); + shardSubTotal.appendNumber(shardName, shardCount); + total += shardCount; + } else { + shardSubTotal.doneFast(); + errmsg = "failed on : " + shardName; + result.append("cause", iter->result); - shardSubTotal.appendNumber(shardName, shardCount); - total += shardCount; - } - else { - shardSubTotal.doneFast(); - errmsg = "failed on : " + shardName; - result.append("cause", iter->result); - - // Add "code" to the top-level response, if the failure of the sharded command - // can be accounted to a single error - int code = getUniqueCodeFromCommandResults(countResult); - if (code != 0) { - result.append("code", code); - } - - return false; + // Add "code" to the top-level response, if the failure of the sharded command + // can be accounted to a single error + int code = getUniqueCodeFromCommandResults(countResult); + if (code != 0) { + result.append("code", code); } - } - shardSubTotal.doneFast(); - total = applySkipLimit(total, cmdObj); - result.appendNumber("n", total); - - return true; + return false; + } } - virtual Status explain(OperationContext* txn, - const std::string& dbname, - const BSONObj& cmdObj, - ExplainCommon::Verbosity verbosity, - BSONObjBuilder* out) const { + shardSubTotal.doneFast(); + total = applySkipLimit(total, cmdObj); + result.appendNumber("n", total); - const string fullns = parseNs(dbname, cmdObj); + return true; + } - // Extract the targeting query. - BSONObj targetingQuery; - if (Object == cmdObj["query"].type()) { - targetingQuery = cmdObj["query"].Obj(); - } + virtual Status explain(OperationContext* txn, + const std::string& dbname, + const BSONObj& cmdObj, + ExplainCommon::Verbosity verbosity, + BSONObjBuilder* out) const { + const string fullns = parseNs(dbname, cmdObj); + + // Extract the targeting query. + BSONObj targetingQuery; + if (Object == cmdObj["query"].type()) { + targetingQuery = cmdObj["query"].Obj(); + } - BSONObjBuilder explainCmdBob; - ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmdBob); + BSONObjBuilder explainCmdBob; + ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmdBob); - // We will time how long it takes to run the commands on the shards - Timer timer; + // We will time how long it takes to run the commands on the shards + Timer timer; - vector<Strategy::CommandResult> shardResults; - Strategy::commandOp(dbname, - explainCmdBob.obj(), - 0, - fullns, - targetingQuery, - &shardResults); + vector<Strategy::CommandResult> shardResults; + Strategy::commandOp(dbname, explainCmdBob.obj(), 0, fullns, targetingQuery, &shardResults); - long long millisElapsed = timer.millis(); + long long millisElapsed = timer.millis(); - const char* mongosStageName = ClusterExplain::getStageNameForReadOp(shardResults, - cmdObj); + const char* mongosStageName = ClusterExplain::getStageNameForReadOp(shardResults, cmdObj); - return ClusterExplain::buildExplainResult(shardResults, - mongosStageName, - millisElapsed, - out); - } + return ClusterExplain::buildExplainResult( + shardResults, mongosStageName, millisElapsed, out); + } - } clusterCountCmd; +} clusterCountCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_current_op.cpp b/src/mongo/s/commands/cluster_current_op.cpp index c4de4dcee82..28f1311ca40 100644 --- a/src/mongo/s/commands/cluster_current_op.cpp +++ b/src/mongo/s/commands/cluster_current_op.cpp @@ -46,181 +46,165 @@ namespace mongo { namespace { - const char kInprogFieldName[] = "inprog"; - const char kOpIdFieldName[] = "opid"; - const char kClientFieldName[] = "client"; - // awkward underscores used to make this visually distinct from kClientFieldName - const char kClient_S_FieldName[] = "client_s"; - const char kLegacyInprogCollection[] = "$cmd.sys.inprog"; - - const char kCommandName[] = "currentOp"; - - class ClusterCurrentOpCommand : public RunOnAllShardsCommand { - public: - - ClusterCurrentOpCommand() : RunOnAllShardsCommand(kCommandName) { } - - bool adminOnly() const final { return true; } - - Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) final { - - - bool isAuthorized = AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forClusterResource(), - ActionType::inprog); - - return isAuthorized ? Status::OK() : Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - - // TODO remove after 3.2 - BSONObj specialErrorHandler(const std::string& server, - const std::string& db, - const BSONObj& cmdObj, - const BSONObj& originalResult) const final { - - // it is unfortunate that this logic needs to be duplicated from - // DBClientWithCommands::runPseudoCommand - // but I don't see a better way to do it without performing heart surgery on - // Future/CommandResponse. - - auto status = getStatusFromCommandResult(originalResult); - invariant(!status.isOK()); - - uassert(28629, - str::stream() << "Received bad " - << kCommandName - << " response from server " << server - << " got: " << originalResult, - status != ErrorCodes::CommandResultSchemaViolation); - - // getStatusFromCommandResult handles cooercing "no such command" into the right - // Status type - if (status == ErrorCodes::CommandNotFound) { - // fall back to the old inprog pseudo-command - NamespaceString pseudoCommandNss("admin", kLegacyInprogCollection); - BSONObj legacyResult; - - BSONObjBuilder legacyCommandBob; - - // need to exclude {currentOp: 1} - for (auto&& cmdElem : cmdObj) { - if (cmdElem.fieldNameStringData() != kCommandName) { - legacyCommandBob.append(cmdElem); - } - } - auto legacyCommand = legacyCommandBob.done(); - - try { - ScopedDbConnection conn(server); - legacyResult = - conn->findOne(pseudoCommandNss.ns(), legacyCommand); - - } - catch (const DBException& ex) { - // If there is a non-DBException exception the entire operation will be - // terminated, as that would be a programmer error. - - // We convert the exception to a BSONObj so that the ordinary - // failure path for RunOnAllShardsCommand will handle the failure - - // TODO: consider adding an exceptionToBSONObj utility? - BSONObjBuilder b; - b.append("errmsg", ex.toString()); - b.append("code", ex.getCode()); - return b.obj(); +const char kInprogFieldName[] = "inprog"; +const char kOpIdFieldName[] = "opid"; +const char kClientFieldName[] = "client"; +// awkward underscores used to make this visually distinct from kClientFieldName +const char kClient_S_FieldName[] = "client_s"; +const char kLegacyInprogCollection[] = "$cmd.sys.inprog"; + +const char kCommandName[] = "currentOp"; + +class ClusterCurrentOpCommand : public RunOnAllShardsCommand { +public: + ClusterCurrentOpCommand() : RunOnAllShardsCommand(kCommandName) {} + + bool adminOnly() const final { + return true; + } + + Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) final { + bool isAuthorized = AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forClusterResource(), ActionType::inprog); + + return isAuthorized ? Status::OK() : Status(ErrorCodes::Unauthorized, "Unauthorized"); + } + + // TODO remove after 3.2 + BSONObj specialErrorHandler(const std::string& server, + const std::string& db, + const BSONObj& cmdObj, + const BSONObj& originalResult) const final { + // it is unfortunate that this logic needs to be duplicated from + // DBClientWithCommands::runPseudoCommand + // but I don't see a better way to do it without performing heart surgery on + // Future/CommandResponse. + + auto status = getStatusFromCommandResult(originalResult); + invariant(!status.isOK()); + + uassert(28629, + str::stream() << "Received bad " << kCommandName << " response from server " + << server << " got: " << originalResult, + status != ErrorCodes::CommandResultSchemaViolation); + + // getStatusFromCommandResult handles cooercing "no such command" into the right + // Status type + if (status == ErrorCodes::CommandNotFound) { + // fall back to the old inprog pseudo-command + NamespaceString pseudoCommandNss("admin", kLegacyInprogCollection); + BSONObj legacyResult; + + BSONObjBuilder legacyCommandBob; + + // need to exclude {currentOp: 1} + for (auto&& cmdElem : cmdObj) { + if (cmdElem.fieldNameStringData() != kCommandName) { + legacyCommandBob.append(cmdElem); } - return legacyResult; } - // if the command failed for another reason then we don't retry it. - return originalResult; - } - - void aggregateResults(const std::vector<ShardAndReply>& results, - BSONObjBuilder& output) final { - // Each shard responds with a document containing an array of subdocuments. - // Each subdocument represents an operation running on that shard. - // We merge the responses into a single document containg an array - // of the operations from all shards. - - // There are two modifications we make. - // 1) we prepend the shardid (with a colon separator) to the opid of each operation. - // This allows users to pass the value of the opid field directly to killOp. + auto legacyCommand = legacyCommandBob.done(); - // 2) we change the field name of "client" to "client_s". This is because each - // client is actually a mongos. + try { + ScopedDbConnection conn(server); + legacyResult = conn->findOne(pseudoCommandNss.ns(), legacyCommand); - // TODO: failpoint for a shard response being invalid. + } catch (const DBException& ex) { + // If there is a non-DBException exception the entire operation will be + // terminated, as that would be a programmer error. - // Error handling - we maintain the same behavior as legacy currentOp/inprog - // that is, if any shard replies with an invalid response (i.e. it does not - // contain a field 'inprog' that is an array), we ignore it. - // - // If there is a lower level error (i.e. the command fails, network error, etc) - // RunOnAllShardsCommand will handle returning an error to the user. - BSONArrayBuilder aggregatedOpsBab(output.subarrayStart(kInprogFieldName)); + // We convert the exception to a BSONObj so that the ordinary + // failure path for RunOnAllShardsCommand will handle the failure - for (auto&& shardResponse : results) { - - StringData shardName; - BSONObj shardResponseObj; - std::tie(shardName, shardResponseObj) = shardResponse; + // TODO: consider adding an exceptionToBSONObj utility? + BSONObjBuilder b; + b.append("errmsg", ex.toString()); + b.append("code", ex.getCode()); + return b.obj(); + } + return legacyResult; + } + // if the command failed for another reason then we don't retry it. + return originalResult; + } + + void aggregateResults(const std::vector<ShardAndReply>& results, BSONObjBuilder& output) final { + // Each shard responds with a document containing an array of subdocuments. + // Each subdocument represents an operation running on that shard. + // We merge the responses into a single document containg an array + // of the operations from all shards. + + // There are two modifications we make. + // 1) we prepend the shardid (with a colon separator) to the opid of each operation. + // This allows users to pass the value of the opid field directly to killOp. + + // 2) we change the field name of "client" to "client_s". This is because each + // client is actually a mongos. + + // TODO: failpoint for a shard response being invalid. + + // Error handling - we maintain the same behavior as legacy currentOp/inprog + // that is, if any shard replies with an invalid response (i.e. it does not + // contain a field 'inprog' that is an array), we ignore it. + // + // If there is a lower level error (i.e. the command fails, network error, etc) + // RunOnAllShardsCommand will handle returning an error to the user. + BSONArrayBuilder aggregatedOpsBab(output.subarrayStart(kInprogFieldName)); + + for (auto&& shardResponse : results) { + StringData shardName; + BSONObj shardResponseObj; + std::tie(shardName, shardResponseObj) = shardResponse; + + auto shardOps = shardResponseObj[kInprogFieldName]; + + // legacy behavior + if (!shardOps.isABSONObj()) { + warning() << "invalid currentOp response from shard " << shardName + << ", got: " << shardOps; + continue; + } - auto shardOps = shardResponseObj[kInprogFieldName]; + for (auto&& shardOp : shardOps.Obj()) { + BSONObjBuilder modifiedShardOpBob; - // legacy behavior - if (!shardOps.isABSONObj()) { - warning() << "invalid currentOp response from shard " - << shardName - << ", got: " - << shardOps; + // maintain legacy behavior + // but log it first + if (!shardOp.isABSONObj()) { + warning() << "invalid currentOp response from shard " << shardName + << ", got: " << shardOp; continue; } - for (auto&& shardOp : shardOps.Obj()) { - BSONObjBuilder modifiedShardOpBob; - - // maintain legacy behavior - // but log it first - if (!shardOp.isABSONObj()) { - warning() << "invalid currentOp response from shard " - << shardName - << ", got: " - << shardOp; - continue; - } - - for (auto&& shardOpElement : shardOp.Obj()) { - auto fieldName = shardOpElement.fieldNameStringData(); - if (fieldName == kOpIdFieldName) { - uassert(28630, - str::stream() << "expected numeric opid from currentOp response" - << " from shard " << shardName - << ", got: " << shardOpElement, - shardOpElement.isNumber()); - - modifiedShardOpBob.append(kOpIdFieldName, - str::stream() << shardName - << ":" - << shardOpElement.numberInt()); - } - else if (fieldName == kClientFieldName) { - modifiedShardOpBob.appendAs(shardOpElement, kClient_S_FieldName); - } - else { - modifiedShardOpBob.append(shardOpElement); - } + for (auto&& shardOpElement : shardOp.Obj()) { + auto fieldName = shardOpElement.fieldNameStringData(); + if (fieldName == kOpIdFieldName) { + uassert(28630, + str::stream() << "expected numeric opid from currentOp response" + << " from shard " << shardName + << ", got: " << shardOpElement, + shardOpElement.isNumber()); + + modifiedShardOpBob.append(kOpIdFieldName, + str::stream() << shardName << ":" + << shardOpElement.numberInt()); + } else if (fieldName == kClientFieldName) { + modifiedShardOpBob.appendAs(shardOpElement, kClient_S_FieldName); + } else { + modifiedShardOpBob.append(shardOpElement); } - modifiedShardOpBob.done(); - // append the modified document to the output array - aggregatedOpsBab.append(modifiedShardOpBob.obj()); } + modifiedShardOpBob.done(); + // append the modified document to the output array + aggregatedOpsBab.append(modifiedShardOpBob.obj()); } - aggregatedOpsBab.done(); } + aggregatedOpsBab.done(); + } - } clusterCurrentOpCmd; +} clusterCurrentOpCmd; } // namespace } // namespace mongo diff --git a/src/mongo/s/commands/cluster_db_stats_cmd.cpp b/src/mongo/s/commands/cluster_db_stats_cmd.cpp index 33f804175aa..a5a50c867cf 100644 --- a/src/mongo/s/commands/cluster_db_stats_cmd.cpp +++ b/src/mongo/s/commands/cluster_db_stats_cmd.cpp @@ -35,74 +35,72 @@ namespace mongo { namespace { - using std::vector; - - class DBStatsCmd : public RunOnAllShardsCommand { - public: - DBStatsCmd() : RunOnAllShardsCommand("dbStats", "dbstats") {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::dbStats); - out->push_back(Privilege(ResourcePattern::forDatabaseName(dbname), actions)); - } - - virtual void aggregateResults(const vector<ShardAndReply>& results, - BSONObjBuilder& output) { - long long objects = 0; - long long unscaledDataSize = 0; - long long dataSize = 0; - long long storageSize = 0; - long long numExtents = 0; - long long indexes = 0; - long long indexSize = 0; - long long fileSize = 0; - - long long freeListNum = 0; - long long freeListSize = 0; - - for (const ShardAndReply& shardAndReply : results) { - const BSONObj& b = std::get<1>(shardAndReply); - - objects += b["objects"].numberLong(); - unscaledDataSize += b["avgObjSize"].numberLong() * b["objects"].numberLong(); - dataSize += b["dataSize"].numberLong(); - storageSize += b["storageSize"].numberLong(); - numExtents += b["numExtents"].numberLong(); - indexes += b["indexes"].numberLong(); - indexSize += b["indexSize"].numberLong(); - fileSize += b["fileSize"].numberLong(); - - if (b["extentFreeList"].isABSONObj()) { - freeListNum += b["extentFreeList"].Obj()["num"].numberLong(); - freeListSize += b["extentFreeList"].Obj()["totalSize"].numberLong(); - } +using std::vector; + +class DBStatsCmd : public RunOnAllShardsCommand { +public: + DBStatsCmd() : RunOnAllShardsCommand("dbStats", "dbstats") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::dbStats); + out->push_back(Privilege(ResourcePattern::forDatabaseName(dbname), actions)); + } + + virtual void aggregateResults(const vector<ShardAndReply>& results, BSONObjBuilder& output) { + long long objects = 0; + long long unscaledDataSize = 0; + long long dataSize = 0; + long long storageSize = 0; + long long numExtents = 0; + long long indexes = 0; + long long indexSize = 0; + long long fileSize = 0; + + long long freeListNum = 0; + long long freeListSize = 0; + + for (const ShardAndReply& shardAndReply : results) { + const BSONObj& b = std::get<1>(shardAndReply); + + objects += b["objects"].numberLong(); + unscaledDataSize += b["avgObjSize"].numberLong() * b["objects"].numberLong(); + dataSize += b["dataSize"].numberLong(); + storageSize += b["storageSize"].numberLong(); + numExtents += b["numExtents"].numberLong(); + indexes += b["indexes"].numberLong(); + indexSize += b["indexSize"].numberLong(); + fileSize += b["fileSize"].numberLong(); + + if (b["extentFreeList"].isABSONObj()) { + freeListNum += b["extentFreeList"].Obj()["num"].numberLong(); + freeListSize += b["extentFreeList"].Obj()["totalSize"].numberLong(); } + } - // result.appendNumber( "collections" , ncollections ); //TODO: need to find a good way to get this - output.appendNumber("objects", objects); - - // avgObjSize on mongod is not scaled based on the argument to db.stats(), so we use - // unscaledDataSize here for consistency. See SERVER-7347. - output.append("avgObjSize", objects == 0 ? 0 : - double(unscaledDataSize) / double(objects)); - output.appendNumber("dataSize", dataSize); - output.appendNumber("storageSize", storageSize); - output.appendNumber("numExtents", numExtents); - output.appendNumber("indexes", indexes); - output.appendNumber("indexSize", indexSize); - output.appendNumber("fileSize", fileSize); - - { - BSONObjBuilder extentFreeList(output.subobjStart("extentFreeList")); - extentFreeList.appendNumber("num", freeListNum); - extentFreeList.appendNumber("totalSize", freeListSize); - extentFreeList.done(); - } + // result.appendNumber( "collections" , ncollections ); //TODO: need to find a good way to get this + output.appendNumber("objects", objects); + + // avgObjSize on mongod is not scaled based on the argument to db.stats(), so we use + // unscaledDataSize here for consistency. See SERVER-7347. + output.append("avgObjSize", objects == 0 ? 0 : double(unscaledDataSize) / double(objects)); + output.appendNumber("dataSize", dataSize); + output.appendNumber("storageSize", storageSize); + output.appendNumber("numExtents", numExtents); + output.appendNumber("indexes", indexes); + output.appendNumber("indexSize", indexSize); + output.appendNumber("fileSize", fileSize); + + { + BSONObjBuilder extentFreeList(output.subobjStart("extentFreeList")); + extentFreeList.appendNumber("num", freeListNum); + extentFreeList.appendNumber("totalSize", freeListSize); + extentFreeList.done(); } + } - } clusterDBStatsCmd; +} clusterDBStatsCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_drop_database_cmd.cpp b/src/mongo/s/commands/cluster_drop_database_cmd.cpp index 0cf6eec2f20..8ac6a6351d2 100644 --- a/src/mongo/s/commands/cluster_drop_database_cmd.cpp +++ b/src/mongo/s/commands/cluster_drop_database_cmd.cpp @@ -40,85 +40,83 @@ namespace mongo { - using std::shared_ptr; +using std::shared_ptr; namespace { - class DropDatabaseCmd : public Command { - public: - DropDatabaseCmd() : Command("dropDatabase") { } - - virtual bool slaveOk() const { - return true; +class DropDatabaseCmd : public Command { +public: + DropDatabaseCmd() : Command("dropDatabase") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::dropDatabase); + out->push_back(Privilege(ResourcePattern::forDatabaseName(dbname), actions)); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + // Disallow dropping the config database from mongos + if (dbname == "config") { + return appendCommandStatus( + result, Status(ErrorCodes::IllegalOperation, "Cannot drop the config database")); } - virtual bool adminOnly() const { - return false; - } + BSONElement e = cmdObj.firstElement(); - virtual bool isWriteCommandForConfigServer() const { - return false; + if (!e.isNumber() || e.number() != 1) { + errmsg = "invalid params"; + return 0; } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - - ActionSet actions; - actions.addAction(ActionType::dropDatabase); - out->push_back(Privilege(ResourcePattern::forDatabaseName(dbname), actions)); - } - - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - // Disallow dropping the config database from mongos - if (dbname == "config") { - return appendCommandStatus(result, Status(ErrorCodes::IllegalOperation, - "Cannot drop the config database")); - } - - BSONElement e = cmdObj.firstElement(); + // Refresh the database metadata + grid.catalogCache()->invalidate(dbname); - if (!e.isNumber() || e.number() != 1) { - errmsg = "invalid params"; - return 0; + auto status = grid.catalogCache()->getDatabase(dbname); + if (!status.isOK()) { + if (status == ErrorCodes::DatabaseNotFound) { + result.append("info", "database does not exist"); + return true; } - // Refresh the database metadata - grid.catalogCache()->invalidate(dbname); - - auto status = grid.catalogCache()->getDatabase(dbname); - if (!status.isOK()) { - if (status == ErrorCodes::DatabaseNotFound) { - result.append("info", "database does not exist"); - return true; - } - - return appendCommandStatus(result, status.getStatus()); - } - - log() << "DROP DATABASE: " << dbname; + return appendCommandStatus(result, status.getStatus()); + } - shared_ptr<DBConfig> conf = status.getValue(); + log() << "DROP DATABASE: " << dbname; - // TODO: Make dropping logic saner and more tolerant of partial drops. This is - // particularly important since a database drop can be aborted by *any* collection - // with a distributed namespace lock taken (migrates/splits) + shared_ptr<DBConfig> conf = status.getValue(); - if (!conf->dropDatabase(errmsg)) { - return false; - } + // TODO: Make dropping logic saner and more tolerant of partial drops. This is + // particularly important since a database drop can be aborted by *any* collection + // with a distributed namespace lock taken (migrates/splits) - result.append("dropped", dbname); - return true; + if (!conf->dropDatabase(errmsg)) { + return false; } - } clusterDropDatabaseCmd; + result.append("dropped", dbname); + return true; + } + +} clusterDropDatabaseCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_enable_sharding_cmd.cpp b/src/mongo/s/commands/cluster_enable_sharding_cmd.cpp index 3c17bc7433e..62749bfbab2 100644 --- a/src/mongo/s/commands/cluster_enable_sharding_cmd.cpp +++ b/src/mongo/s/commands/cluster_enable_sharding_cmd.cpp @@ -48,77 +48,74 @@ namespace mongo { namespace { - class EnableShardingCmd : public Command { - public: - EnableShardingCmd() : Command("enableSharding", false, "enablesharding") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; - } - - virtual bool isWriteCommandForConfigServer() const { - return false; +class EnableShardingCmd : public Command { +public: + EnableShardingCmd() : Command("enableSharding", false, "enablesharding") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "Enable sharding for a database. " + << "(Use 'shardcollection' command afterwards.)\n" + << " { enablesharding : \"<dbname>\" }\n"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forDatabaseName(parseNs(dbname, cmdObj)), + ActionType::enableSharding)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); } - virtual void help(std::stringstream& help) const { - help << "Enable sharding for a database. " - << "(Use 'shardcollection' command afterwards.)\n" - << " { enablesharding : \"<dbname>\" }\n"; - } + return Status::OK(); + } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return cmdObj.firstElement().str(); + } - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forDatabaseName( - parseNs(dbname, cmdObj)), - ActionType::enableSharding)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } + virtual bool run(OperationContext* txn, + const std::string& dbname_unused, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + const std::string dbname = parseNs("", cmdObj); - return Status::OK(); + if (dbname.empty() || !nsIsDbOnly(dbname)) { + errmsg = "invalid db name specified: " + dbname; + return false; } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return cmdObj.firstElement().str(); + if (dbname == "admin" || dbname == "config" || dbname == "local") { + errmsg = "can't shard " + dbname + " database"; + return false; } - virtual bool run(OperationContext* txn, - const std::string& dbname_unused, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - const std::string dbname = parseNs("", cmdObj); - - if (dbname.empty() || !nsIsDbOnly(dbname)) { - errmsg = "invalid db name specified: " + dbname; - return false; - } - - if (dbname == "admin" || dbname == "config" || dbname == "local") { - errmsg = "can't shard " + dbname + " database"; - return false; - } - - Status status = grid.catalogManager()->enableSharding(dbname); - if (status.isOK()) { - audit::logEnableSharding(ClientBasic::getCurrent(), dbname); - } + Status status = grid.catalogManager()->enableSharding(dbname); + if (status.isOK()) { + audit::logEnableSharding(ClientBasic::getCurrent(), dbname); + } - // Make sure to force update of any stale metadata - grid.catalogCache()->invalidate(dbname); + // Make sure to force update of any stale metadata + grid.catalogCache()->invalidate(dbname); - return appendCommandStatus(result, status); - } + return appendCommandStatus(result, status); + } - } enableShardingCmd; +} enableShardingCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_explain_cmd.cpp b/src/mongo/s/commands/cluster_explain_cmd.cpp index 4fa094affed..9cb8bae6b9c 100644 --- a/src/mongo/s/commands/cluster_explain_cmd.cpp +++ b/src/mongo/s/commands/cluster_explain_cmd.cpp @@ -33,98 +33,107 @@ namespace mongo { +/** + * Implements the explain command on mongos. + * + * "Old-style" explains (i.e. queries which have the $explain flag set), do not run + * through this path. Such explains will be supported for backwards compatibility, + * and must succeed in multiversion clusters. + * + * "New-style" explains use the explain command. When the explain command is routed + * through mongos, it is forwarded to all relevant shards. If *any* shard does not + * support a new-style explain, then the entire explain will fail (i.e. new-style + * explains cannot be used in multiversion clusters). + */ +class ClusterExplainCmd : public Command { + MONGO_DISALLOW_COPYING(ClusterExplainCmd); + +public: + ClusterExplainCmd() : Command("explain") {} + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + /** - * Implements the explain command on mongos. - * - * "Old-style" explains (i.e. queries which have the $explain flag set), do not run - * through this path. Such explains will be supported for backwards compatibility, - * and must succeed in multiversion clusters. - * - * "New-style" explains use the explain command. When the explain command is routed - * through mongos, it is forwarded to all relevant shards. If *any* shard does not - * support a new-style explain, then the entire explain will fail (i.e. new-style - * explains cannot be used in multiversion clusters). + * Running an explain on a secondary requires explicitly setting slaveOk. */ - class ClusterExplainCmd : public Command { - MONGO_DISALLOW_COPYING(ClusterExplainCmd); - public: - ClusterExplainCmd() : Command("explain") { } + virtual bool slaveOk() const { + return false; + } + virtual bool slaveOverrideOk() const { + return true; + } - virtual bool isWriteCommandForConfigServer() const { return false; } + virtual bool maintenanceOk() const { + return false; + } - /** - * Running an explain on a secondary requires explicitly setting slaveOk. - */ - virtual bool slaveOk() const { return false; } - virtual bool slaveOverrideOk() const { return true; } + virtual bool adminOnly() const { + return false; + } - virtual bool maintenanceOk() const { return false; } + virtual void help(std::stringstream& help) const { + help << "explain database reads and writes"; + } - virtual bool adminOnly() const { return false; } + /** + * You are authorized to run an explain if you are authorized to run + * the command that you are explaining. The auth check is performed recursively + * on the nested command. + */ + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (Object != cmdObj.firstElement().type()) { + return Status(ErrorCodes::BadValue, "explain command requires a nested object"); + } - virtual void help(std::stringstream& help) const { - help << "explain database reads and writes"; + BSONObj explainObj = cmdObj.firstElement().Obj(); + + Command* commToExplain = Command::findCommand(explainObj.firstElementFieldName()); + if (NULL == commToExplain) { + mongoutils::str::stream ss; + ss << "unknown command: " << explainObj.firstElementFieldName(); + return Status(ErrorCodes::CommandNotFound, ss); } - /** - * You are authorized to run an explain if you are authorized to run - * the command that you are explaining. The auth check is performed recursively - * on the nested command. - */ - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - - if (Object != cmdObj.firstElement().type()) { - return Status(ErrorCodes::BadValue, "explain command requires a nested object"); - } - - BSONObj explainObj = cmdObj.firstElement().Obj(); - - Command* commToExplain = Command::findCommand(explainObj.firstElementFieldName()); - if (NULL == commToExplain) { - mongoutils::str::stream ss; - ss << "unknown command: " << explainObj.firstElementFieldName(); - return Status(ErrorCodes::CommandNotFound, ss); - } - - return commToExplain->checkAuthForCommand(client, dbname, explainObj); + return commToExplain->checkAuthForCommand(client, dbname, explainObj); + } + + virtual bool run(OperationContext* txn, + const std::string& dbName, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + ExplainCommon::Verbosity verbosity; + Status parseStatus = ExplainCommon::parseCmdBSON(cmdObj, &verbosity); + if (!parseStatus.isOK()) { + return appendCommandStatus(result, parseStatus); } - virtual bool run(OperationContext* txn, - const std::string& dbName, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - ExplainCommon::Verbosity verbosity; - Status parseStatus = ExplainCommon::parseCmdBSON(cmdObj, &verbosity); - if (!parseStatus.isOK()) { - return appendCommandStatus(result, parseStatus); - } - - // This is the nested command which we are explaining. - BSONObj explainObj = cmdObj.firstElement().Obj(); - - const std::string cmdName = explainObj.firstElementFieldName(); - Command* commToExplain = Command::findCommand(cmdName); - if (NULL == commToExplain) { - mongoutils::str::stream ss; - ss << "Explain failed due to unknown command: " << cmdName; - Status explainStatus(ErrorCodes::CommandNotFound, ss); - return appendCommandStatus(result, explainStatus); - } - - // Actually call the nested command's explain(...) method. - Status explainStatus = commToExplain->explain(txn, dbName, explainObj, verbosity, &result); - if (!explainStatus.isOK()) { - return appendCommandStatus(result, explainStatus); - } - - return true; + // This is the nested command which we are explaining. + BSONObj explainObj = cmdObj.firstElement().Obj(); + + const std::string cmdName = explainObj.firstElementFieldName(); + Command* commToExplain = Command::findCommand(cmdName); + if (NULL == commToExplain) { + mongoutils::str::stream ss; + ss << "Explain failed due to unknown command: " << cmdName; + Status explainStatus(ErrorCodes::CommandNotFound, ss); + return appendCommandStatus(result, explainStatus); } - } cmdExplainCluster; + // Actually call the nested command's explain(...) method. + Status explainStatus = commToExplain->explain(txn, dbName, explainObj, verbosity, &result); + if (!explainStatus.isOK()) { + return appendCommandStatus(result, explainStatus); + } + + return true; + } + +} cmdExplainCluster; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_find_and_modify_cmd.cpp b/src/mongo/s/commands/cluster_find_and_modify_cmd.cpp index 8fcff46f1cd..4f14f66fe2c 100644 --- a/src/mongo/s/commands/cluster_find_and_modify_cmd.cpp +++ b/src/mongo/s/commands/cluster_find_and_modify_cmd.cpp @@ -49,192 +49,181 @@ namespace mongo { namespace { - using std::shared_ptr; - using std::string; - using std::vector; +using std::shared_ptr; +using std::string; +using std::vector; - class FindAndModifyCmd : public Command { - public: - FindAndModifyCmd() : Command("findAndModify", false, "findandmodify") { } +class FindAndModifyCmd : public Command { +public: + FindAndModifyCmd() : Command("findAndModify", false, "findandmodify") {} - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return false; - } - - virtual bool isWriteCommandForConfigServer() const { - return false; - } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - - find_and_modify::addPrivilegesRequiredForFindAndModify(this, dbname, cmdObj, out); - } - - virtual Status explain(OperationContext* txn, - const std::string& dbName, - const BSONObj& cmdObj, - ExplainCommon::Verbosity verbosity, - BSONObjBuilder* out) const { - - const string ns = parseNsCollectionRequired(dbName, cmdObj); - - auto status = grid.catalogCache()->getDatabase(dbName); - uassertStatusOK(status); + virtual bool slaveOk() const { + return true; + } - shared_ptr<DBConfig> conf = status.getValue(); + virtual bool adminOnly() const { + return false; + } - shared_ptr<Shard> shard; + virtual bool isWriteCommandForConfigServer() const { + return false; + } - if (!conf->isShardingEnabled() || !conf->isSharded(ns)) { - shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); - } - else { - shared_ptr<ChunkManager> chunkMgr = _getChunkManager(conf, ns); - - const BSONObj query = cmdObj.getObjectField("query"); - - StatusWith<BSONObj> status = _getShardKey(chunkMgr, query); - if (!status.isOK()) { - return status.getStatus(); - } - - BSONObj shardKey = status.getValue(); - ChunkPtr chunk = chunkMgr->findIntersectingChunk(shardKey); - - shard = grid.shardRegistry()->getShard(chunk->getShardId()); - } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + find_and_modify::addPrivilegesRequiredForFindAndModify(this, dbname, cmdObj, out); + } - BSONObjBuilder explainCmd; - ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmd); + virtual Status explain(OperationContext* txn, + const std::string& dbName, + const BSONObj& cmdObj, + ExplainCommon::Verbosity verbosity, + BSONObjBuilder* out) const { + const string ns = parseNsCollectionRequired(dbName, cmdObj); - // Time how long it takes to run the explain command on the shard. - Timer timer; + auto status = grid.catalogCache()->getDatabase(dbName); + uassertStatusOK(status); - BSONObjBuilder result; - bool ok = _runCommand(conf, shard->getId(), ns, explainCmd.obj(), result); - long long millisElapsed = timer.millis(); + shared_ptr<DBConfig> conf = status.getValue(); - if (!ok) { - BSONObj res = result.obj(); - return Status(ErrorCodes::OperationFailed, - str::stream() << "Explain for findAndModify failed: " << res); - } - - Strategy::CommandResult cmdResult; - cmdResult.shardTargetId = shard->getId(); - cmdResult.target = shard->getConnString(); - cmdResult.result = result.obj(); - - vector<Strategy::CommandResult> shardResults; - shardResults.push_back(cmdResult); - - return ClusterExplain::buildExplainResult(shardResults, - ClusterExplain::kSingleShard, - millisElapsed, - out); - } - - virtual bool run(OperationContext* txn, - const std::string& dbName, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - const string ns = parseNsCollectionRequired(dbName, cmdObj); - - // findAndModify should only be creating database if upsert is true, but this would - // require that the parsing be pulled into this function. - auto conf = uassertStatusOK(grid.implicitCreateDb(dbName)); - if (!conf->isShardingEnabled() || !conf->isSharded(ns)) { - return _runCommand(conf, conf->getPrimaryId(), ns, cmdObj, result); - } + shared_ptr<Shard> shard; + if (!conf->isShardingEnabled() || !conf->isSharded(ns)) { + shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); + } else { shared_ptr<ChunkManager> chunkMgr = _getChunkManager(conf, ns); const BSONObj query = cmdObj.getObjectField("query"); StatusWith<BSONObj> status = _getShardKey(chunkMgr, query); if (!status.isOK()) { - // Bad query - return appendCommandStatus(result, status.getStatus()); + return status.getStatus(); } BSONObj shardKey = status.getValue(); ChunkPtr chunk = chunkMgr->findIntersectingChunk(shardKey); - bool ok = _runCommand(conf, chunk->getShardId(), ns, cmdObj, result); - if (ok) { - // check whether split is necessary (using update object for size heuristic) - if (Chunk::ShouldAutoSplit) { - chunk->splitIfShould(cmdObj.getObjectField("update").objsize()); - } - } - - return ok; + shard = grid.shardRegistry()->getShard(chunk->getShardId()); } - private: - shared_ptr<ChunkManager> _getChunkManager(shared_ptr<DBConfig> conf, - const string& ns) const { + BSONObjBuilder explainCmd; + ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmd); - shared_ptr<ChunkManager> chunkMgr = conf->getChunkManager(ns); - massert(13002, "shard internal error chunk manager should never be null", chunkMgr); + // Time how long it takes to run the explain command on the shard. + Timer timer; - return chunkMgr; + BSONObjBuilder result; + bool ok = _runCommand(conf, shard->getId(), ns, explainCmd.obj(), result); + long long millisElapsed = timer.millis(); + + if (!ok) { + BSONObj res = result.obj(); + return Status(ErrorCodes::OperationFailed, + str::stream() << "Explain for findAndModify failed: " << res); } - StatusWith<BSONObj> _getShardKey(shared_ptr<ChunkManager> chunkMgr, - const BSONObj& query) const { + Strategy::CommandResult cmdResult; + cmdResult.shardTargetId = shard->getId(); + cmdResult.target = shard->getConnString(); + cmdResult.result = result.obj(); + + vector<Strategy::CommandResult> shardResults; + shardResults.push_back(cmdResult); + + return ClusterExplain::buildExplainResult( + shardResults, ClusterExplain::kSingleShard, millisElapsed, out); + } + + virtual bool run(OperationContext* txn, + const std::string& dbName, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + const string ns = parseNsCollectionRequired(dbName, cmdObj); + + // findAndModify should only be creating database if upsert is true, but this would + // require that the parsing be pulled into this function. + auto conf = uassertStatusOK(grid.implicitCreateDb(dbName)); + if (!conf->isShardingEnabled() || !conf->isSharded(ns)) { + return _runCommand(conf, conf->getPrimaryId(), ns, cmdObj, result); + } - // Verify that the query has an equality predicate using the shard key - StatusWith<BSONObj> status = - chunkMgr->getShardKeyPattern().extractShardKeyFromQuery(query); + shared_ptr<ChunkManager> chunkMgr = _getChunkManager(conf, ns); - if (!status.isOK()) { - return status; - } + const BSONObj query = cmdObj.getObjectField("query"); - BSONObj shardKey = status.getValue(); + StatusWith<BSONObj> status = _getShardKey(chunkMgr, query); + if (!status.isOK()) { + // Bad query + return appendCommandStatus(result, status.getStatus()); + } - if (shardKey.isEmpty()) { - return Status(ErrorCodes::ShardKeyNotFound, - "query for sharded findAndModify must have shardkey"); - } + BSONObj shardKey = status.getValue(); + ChunkPtr chunk = chunkMgr->findIntersectingChunk(shardKey); - return shardKey; + bool ok = _runCommand(conf, chunk->getShardId(), ns, cmdObj, result); + if (ok) { + // check whether split is necessary (using update object for size heuristic) + if (Chunk::ShouldAutoSplit) { + chunk->splitIfShould(cmdObj.getObjectField("update").objsize()); + } } - bool _runCommand(DBConfigPtr conf, - const ShardId& shardId, - const string& ns, - const BSONObj& cmdObj, - BSONObjBuilder& result) const { + return ok; + } - BSONObj res; +private: + shared_ptr<ChunkManager> _getChunkManager(shared_ptr<DBConfig> conf, const string& ns) const { + shared_ptr<ChunkManager> chunkMgr = conf->getChunkManager(ns); + massert(13002, "shard internal error chunk manager should never be null", chunkMgr); - const auto shard = grid.shardRegistry()->getShard(shardId); - ShardConnection conn(shard->getConnString(), ns); - bool ok = conn->runCommand(conf->name(), cmdObj, res); - conn.done(); + return chunkMgr; + } - // RecvStaleConfigCode is the code for RecvStaleConfigException. - if (!ok && res.getIntField("code") == RecvStaleConfigCode) { - // Command code traps this exception and re-runs - throw RecvStaleConfigException("FindAndModify", res); - } + StatusWith<BSONObj> _getShardKey(shared_ptr<ChunkManager> chunkMgr, + const BSONObj& query) const { + // Verify that the query has an equality predicate using the shard key + StatusWith<BSONObj> status = chunkMgr->getShardKeyPattern().extractShardKeyFromQuery(query); + + if (!status.isOK()) { + return status; + } + + BSONObj shardKey = status.getValue(); - result.appendElements(res); - return ok; + if (shardKey.isEmpty()) { + return Status(ErrorCodes::ShardKeyNotFound, + "query for sharded findAndModify must have shardkey"); } - } findAndModifyCmd; + return shardKey; + } + + bool _runCommand(DBConfigPtr conf, + const ShardId& shardId, + const string& ns, + const BSONObj& cmdObj, + BSONObjBuilder& result) const { + BSONObj res; + + const auto shard = grid.shardRegistry()->getShard(shardId); + ShardConnection conn(shard->getConnString(), ns); + bool ok = conn->runCommand(conf->name(), cmdObj, res); + conn.done(); + + // RecvStaleConfigCode is the code for RecvStaleConfigException. + if (!ok && res.getIntField("code") == RecvStaleConfigCode) { + // Command code traps this exception and re-runs + throw RecvStaleConfigException("FindAndModify", res); + } + + result.appendElements(res); + return ok; + } + +} findAndModifyCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_find_cmd.cpp b/src/mongo/s/commands/cluster_find_cmd.cpp index f85b65fe39e..b40e919ca3c 100644 --- a/src/mongo/s/commands/cluster_find_cmd.cpp +++ b/src/mongo/s/commands/cluster_find_cmd.cpp @@ -37,111 +37,118 @@ namespace mongo { namespace { - using std::unique_ptr; - using std::string; - using std::vector; +using std::unique_ptr; +using std::string; +using std::vector; - /** - * Implements the find command on mongos. - * - * TODO: this is just a placeholder. It needs to be implemented for real under SERVER-15176. - */ - class ClusterFindCmd : public Command { - MONGO_DISALLOW_COPYING(ClusterFindCmd); - public: - ClusterFindCmd() : Command("find") { } - - virtual bool isWriteCommandForConfigServer() const { return false; } +/** + * Implements the find command on mongos. + * + * TODO: this is just a placeholder. It needs to be implemented for real under SERVER-15176. + */ +class ClusterFindCmd : public Command { + MONGO_DISALLOW_COPYING(ClusterFindCmd); - virtual bool slaveOk() const { return false; } +public: + ClusterFindCmd() : Command("find") {} - virtual bool slaveOverrideOk() const { return true; } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual bool maintenanceOk() const { return false; } + virtual bool slaveOk() const { + return false; + } - virtual bool adminOnly() const { return false; } + virtual bool slaveOverrideOk() const { + return true; + } - virtual void help(std::stringstream& help) const { - help << "query for documents"; - } + virtual bool maintenanceOk() const { + return false; + } - /** - * In order to run the find command, you must be authorized for the "find" action - * type on the collection. - */ - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { + virtual bool adminOnly() const { + return false; + } - AuthorizationSession* authzSession = AuthorizationSession::get(client); - ResourcePattern pattern = parseResourcePattern(dbname, cmdObj); + virtual void help(std::stringstream& help) const { + help << "query for documents"; + } - if (authzSession->isAuthorizedForActionsOnResource(pattern, ActionType::find)) { - return Status::OK(); - } + /** + * In order to run the find command, you must be authorized for the "find" action + * type on the collection. + */ + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + AuthorizationSession* authzSession = AuthorizationSession::get(client); + ResourcePattern pattern = parseResourcePattern(dbname, cmdObj); + + if (authzSession->isAuthorizedForActionsOnResource(pattern, ActionType::find)) { + return Status::OK(); + } - return Status(ErrorCodes::Unauthorized, "unauthorized"); + return Status(ErrorCodes::Unauthorized, "unauthorized"); + } + + virtual Status explain(OperationContext* txn, + const std::string& dbname, + const BSONObj& cmdObj, + ExplainCommon::Verbosity verbosity, + BSONObjBuilder* out) const { + const string fullns = parseNs(dbname, cmdObj); + const NamespaceString nss(fullns); + if (!nss.isValid()) { + return {ErrorCodes::InvalidNamespace, + str::stream() << "Invalid collection name: " << nss.ns()}; } - virtual Status explain(OperationContext* txn, - const std::string& dbname, - const BSONObj& cmdObj, - ExplainCommon::Verbosity verbosity, - BSONObjBuilder* out) const { - - const string fullns = parseNs(dbname, cmdObj); - const NamespaceString nss(fullns); - if (!nss.isValid()) { - return {ErrorCodes::InvalidNamespace, - str::stream() << "Invalid collection name: " << nss.ns()}; - } - - // Parse the command BSON to a LiteParsedQuery. - bool isExplain = true; - auto lpqStatus = LiteParsedQuery::makeFromFindCommand(nss, cmdObj, isExplain); - if (!lpqStatus.isOK()) { - return lpqStatus.getStatus(); - } - - auto& lpq = lpqStatus.getValue(); - - BSONObjBuilder explainCmdBob; - ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmdBob); - - // We will time how long it takes to run the commands on the shards. - Timer timer; - - vector<Strategy::CommandResult> shardResults; - Strategy::commandOp(dbname, - explainCmdBob.obj(), - lpq->getOptions(), - fullns, - lpq->getFilter(), - &shardResults); - - long long millisElapsed = timer.millis(); - - const char* mongosStageName = ClusterExplain::getStageNameForReadOp(shardResults, cmdObj); - - return ClusterExplain::buildExplainResult(shardResults, - mongosStageName, - millisElapsed, - out); + // Parse the command BSON to a LiteParsedQuery. + bool isExplain = true; + auto lpqStatus = LiteParsedQuery::makeFromFindCommand(nss, cmdObj, isExplain); + if (!lpqStatus.isOK()) { + return lpqStatus.getStatus(); } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, int options, - std::string& errmsg, - BSONObjBuilder& result) { + auto& lpq = lpqStatus.getValue(); - // Currently only explains of finds run through the find command. Queries that are not - // explained use the legacy OP_QUERY path. - errmsg = "find command not yet implemented"; - return false; - } + BSONObjBuilder explainCmdBob; + ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmdBob); + + // We will time how long it takes to run the commands on the shards. + Timer timer; + + vector<Strategy::CommandResult> shardResults; + Strategy::commandOp(dbname, + explainCmdBob.obj(), + lpq->getOptions(), + fullns, + lpq->getFilter(), + &shardResults); + + long long millisElapsed = timer.millis(); + + const char* mongosStageName = ClusterExplain::getStageNameForReadOp(shardResults, cmdObj); + + return ClusterExplain::buildExplainResult( + shardResults, mongosStageName, millisElapsed, out); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + // Currently only explains of finds run through the find command. Queries that are not + // explained use the legacy OP_QUERY path. + errmsg = "find command not yet implemented"; + return false; + } - } cmdFindCluster; +} cmdFindCluster; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_flush_router_config_cmd.cpp b/src/mongo/s/commands/cluster_flush_router_config_cmd.cpp index 19477622b64..dbb52870602 100644 --- a/src/mongo/s/commands/cluster_flush_router_config_cmd.cpp +++ b/src/mongo/s/commands/cluster_flush_router_config_cmd.cpp @@ -36,48 +36,47 @@ namespace mongo { namespace { - class FlushRouterConfigCmd : public Command { - public: - FlushRouterConfigCmd() : Command("flushRouterConfig", false, "flushrouterconfig") { } +class FlushRouterConfigCmd : public Command { +public: + FlushRouterConfigCmd() : Command("flushRouterConfig", false, "flushrouterconfig") {} - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual bool adminOnly() const { - return true; - } + virtual bool adminOnly() const { + return true; + } - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual void help(std::stringstream& help) const { - help << "flush all router config"; - } + virtual void help(std::stringstream& help) const { + help << "flush all router config"; + } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::flushRouterConfig); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::flushRouterConfig); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + grid.catalogCache()->invalidateAll(); - grid.catalogCache()->invalidateAll(); + result.appendBool("flushed", true); + return true; + } - result.appendBool("flushed", true); - return true; - } +} flushRouterConfigCmd; - } flushRouterConfigCmd; - -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_fsync_cmd.cpp b/src/mongo/s/commands/cluster_fsync_cmd.cpp index 99ae1cf8fea..f9dc78ee9d7 100644 --- a/src/mongo/s/commands/cluster_fsync_cmd.cpp +++ b/src/mongo/s/commands/cluster_fsync_cmd.cpp @@ -36,77 +36,76 @@ namespace mongo { namespace { - class FsyncCommand : public Command { - public: - FsyncCommand() : Command("fsync", false, "fsync") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; - } - - virtual bool isWriteCommandForConfigServer() const { +class FsyncCommand : public Command { +public: + FsyncCommand() : Command("fsync", false, "fsync") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "invoke fsync on all shards belonging to the cluster"; + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::fsync); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + if (cmdObj["lock"].trueValue()) { + errmsg = "can't do lock through mongos"; return false; } - virtual void help(std::stringstream& help) const { - help << "invoke fsync on all shards belonging to the cluster"; - } + BSONObjBuilder sub; - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::fsync); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } + bool ok = true; + int numFiles = 0; - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + std::vector<ShardId> shardIds; + grid.shardRegistry()->getAllShardIds(&shardIds); - if (cmdObj["lock"].trueValue()) { - errmsg = "can't do lock through mongos"; - return false; + for (const ShardId& shardId : shardIds) { + const auto s = grid.shardRegistry()->getShard(shardId); + if (!s) { + continue; } - BSONObjBuilder sub; + BSONObj x = s->runCommand("admin", "fsync"); + sub.append(s->getId(), x); - bool ok = true; - int numFiles = 0; - - std::vector<ShardId> shardIds; - grid.shardRegistry()->getAllShardIds(&shardIds); - - for (const ShardId& shardId : shardIds) { - const auto s = grid.shardRegistry()->getShard(shardId); - if (!s) { - continue; - } - - BSONObj x = s->runCommand("admin", "fsync"); - sub.append(s->getId(), x); - - if (!x["ok"].trueValue()) { - ok = false; - errmsg = x["errmsg"].String(); - } - - numFiles += x["numFiles"].numberInt(); + if (!x["ok"].trueValue()) { + ok = false; + errmsg = x["errmsg"].String(); } - result.append("numFiles", numFiles); - result.append("all", sub.obj()); - return ok; + numFiles += x["numFiles"].numberInt(); } - } fsyncCmd; + result.append("numFiles", numFiles); + result.append("all", sub.obj()); + return ok; + } + +} fsyncCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_get_last_error_cmd.cpp b/src/mongo/s/commands/cluster_get_last_error_cmd.cpp index 45a1cc9792e..1e219d77114 100644 --- a/src/mongo/s/commands/cluster_get_last_error_cmd.cpp +++ b/src/mongo/s/commands/cluster_get_last_error_cmd.cpp @@ -41,174 +41,163 @@ namespace mongo { namespace { - class GetLastErrorCmd : public Command { - public: - GetLastErrorCmd() : Command("getLastError", false, "getlasterror") { } - - virtual bool isWriteCommandForConfigServer() const { - return false; - } - - virtual bool slaveOk() const { - return true; - } - - virtual void help(std::stringstream& help) const { - help << "check for an error on the last command executed"; - } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - - // No auth required for getlasterror +class GetLastErrorCmd : public Command { +public: + GetLastErrorCmd() : Command("getLastError", false, "getlasterror") {} + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual bool slaveOk() const { + return true; + } + + virtual void help(std::stringstream& help) const { + help << "check for an error on the last command executed"; + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + // No auth required for getlasterror + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + // Mongos GLE - finicky. + // + // To emulate mongod, we first append any write errors we had, then try to append + // write concern error if there was no write error. We need to contact the previous + // shards regardless to maintain 2.4 behavior. + // + // If there are any unexpected or connectivity errors when calling GLE, fail the + // command. + // + // Finally, report the write concern errors IF we don't already have an error. + // If we only get one write concern error back, report that, otherwise report an + // aggregated error. + // + // TODO: Do we need to contact the prev shards regardless - do we care that much + // about 2.4 behavior? + // + + LastError* le = &LastError::get(cc()); + le->disable(); + + + // Write commands always have the error stored in the mongos last error + bool errorOccurred = false; + if (le->getNPrev() == 1) { + errorOccurred = le->appendSelf(result, false); } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - // Mongos GLE - finicky. - // - // To emulate mongod, we first append any write errors we had, then try to append - // write concern error if there was no write error. We need to contact the previous - // shards regardless to maintain 2.4 behavior. - // - // If there are any unexpected or connectivity errors when calling GLE, fail the - // command. - // - // Finally, report the write concern errors IF we don't already have an error. - // If we only get one write concern error back, report that, otherwise report an - // aggregated error. - // - // TODO: Do we need to contact the prev shards regardless - do we care that much - // about 2.4 behavior? - // - - LastError *le = &LastError::get(cc()); - le->disable(); - - - // Write commands always have the error stored in the mongos last error - bool errorOccurred = false; - if (le->getNPrev() == 1) { - errorOccurred = le->appendSelf(result, false); - } - - // For compatibility with 2.4 sharded GLE, we always enforce the write concern - // across all shards. - const HostOpTimeMap hostOpTimes(ClusterLastErrorInfo::get(cc()).getPrevHostOpTimes()); - HostOpTimeMap resolvedHostOpTimes; - - Status status(Status::OK()); - for (HostOpTimeMap::const_iterator it = hostOpTimes.begin(); - it != hostOpTimes.end(); - ++it) { - - const ConnectionString& shardEndpoint = it->first; - const HostOpTime& hot = it->second; + // For compatibility with 2.4 sharded GLE, we always enforce the write concern + // across all shards. + const HostOpTimeMap hostOpTimes(ClusterLastErrorInfo::get(cc()).getPrevHostOpTimes()); + HostOpTimeMap resolvedHostOpTimes; - ConnectionString resolvedHost; - status = DBClientShardResolver::findMaster(shardEndpoint, &resolvedHost); - if (!status.isOK()) { - break; - } - - resolvedHostOpTimes[resolvedHost] = hot; - } - - DBClientMultiCommand dispatcher; - std::vector<LegacyWCResponse> wcResponses; - if (status.isOK()) { - status = enforceLegacyWriteConcern(&dispatcher, - dbname, - cmdObj, - resolvedHostOpTimes, - &wcResponses); - } - - // Don't forget about our last hosts, reset the client info - ClusterLastErrorInfo::get(cc()).disableForCommand(); - - // We're now done contacting all remote servers, just report results + Status status(Status::OK()); + for (HostOpTimeMap::const_iterator it = hostOpTimes.begin(); it != hostOpTimes.end(); + ++it) { + const ConnectionString& shardEndpoint = it->first; + const HostOpTime& hot = it->second; + ConnectionString resolvedHost; + status = DBClientShardResolver::findMaster(shardEndpoint, &resolvedHost); if (!status.isOK()) { - // Return immediately if we failed to contact a shard, unexpected GLE issue - // Can't return code, since it may have been set above (2.4 compatibility) - result.append("errmsg", status.reason()); - return false; + break; } - // Go through all the write concern responses and find errors - BSONArrayBuilder shards; - BSONObjBuilder shardRawGLE; - BSONArrayBuilder errors; - BSONArrayBuilder errorRawGLE; + resolvedHostOpTimes[resolvedHost] = hot; + } - int numWCErrors = 0; - const LegacyWCResponse* lastErrResponse = NULL; + DBClientMultiCommand dispatcher; + std::vector<LegacyWCResponse> wcResponses; + if (status.isOK()) { + status = enforceLegacyWriteConcern( + &dispatcher, dbname, cmdObj, resolvedHostOpTimes, &wcResponses); + } - for (std::vector<LegacyWCResponse>::const_iterator it = wcResponses.begin(); - it != wcResponses.end(); - ++it) { + // Don't forget about our last hosts, reset the client info + ClusterLastErrorInfo::get(cc()).disableForCommand(); - const LegacyWCResponse& wcResponse = *it; + // We're now done contacting all remote servers, just report results - shards.append(wcResponse.shardHost); - shardRawGLE.append(wcResponse.shardHost, wcResponse.gleResponse); + if (!status.isOK()) { + // Return immediately if we failed to contact a shard, unexpected GLE issue + // Can't return code, since it may have been set above (2.4 compatibility) + result.append("errmsg", status.reason()); + return false; + } - if (!wcResponse.errToReport.empty()) { - numWCErrors++; - lastErrResponse = &wcResponse; - errors.append(wcResponse.errToReport); - errorRawGLE.append(wcResponse.gleResponse); - } - } + // Go through all the write concern responses and find errors + BSONArrayBuilder shards; + BSONObjBuilder shardRawGLE; + BSONArrayBuilder errors; + BSONArrayBuilder errorRawGLE; - // Always report what we found to match 2.4 behavior and for debugging - if (wcResponses.size() == 1u) { - result.append("singleShard", wcResponses.front().shardHost); - } - else { - result.append("shards", shards.arr()); - result.append("shardRawGLE", shardRawGLE.obj()); - } + int numWCErrors = 0; + const LegacyWCResponse* lastErrResponse = NULL; - // Suppress write concern errors if a write error occurred, to match mongod behavior - if (errorOccurred || numWCErrors == 0) { - // Still need to return err - if (!errorOccurred) { - result.appendNull("err"); - } + for (std::vector<LegacyWCResponse>::const_iterator it = wcResponses.begin(); + it != wcResponses.end(); + ++it) { + const LegacyWCResponse& wcResponse = *it; - return true; - } + shards.append(wcResponse.shardHost); + shardRawGLE.append(wcResponse.shardHost, wcResponse.gleResponse); - if (numWCErrors == 1) { - // Return the single write concern error we found, err should be set or not - // from gle response - result.appendElements(lastErrResponse->gleResponse); - return lastErrResponse->gleResponse["ok"].trueValue(); + if (!wcResponse.errToReport.empty()) { + numWCErrors++; + lastErrResponse = &wcResponse; + errors.append(wcResponse.errToReport); + errorRawGLE.append(wcResponse.gleResponse); } - else { + } - // Return a generic combined WC error message - result.append("errs", errors.arr()); - result.append("errObjects", errorRawGLE.arr()); + // Always report what we found to match 2.4 behavior and for debugging + if (wcResponses.size() == 1u) { + result.append("singleShard", wcResponses.front().shardHost); + } else { + result.append("shards", shards.arr()); + result.append("shardRawGLE", shardRawGLE.obj()); + } - // Need to always return err + // Suppress write concern errors if a write error occurred, to match mongod behavior + if (errorOccurred || numWCErrors == 0) { + // Still need to return err + if (!errorOccurred) { result.appendNull("err"); - - return appendCommandStatus(result, - Status(ErrorCodes::WriteConcernFailed, - "multiple write concern errors occurred")); } + + return true; + } + + if (numWCErrors == 1) { + // Return the single write concern error we found, err should be set or not + // from gle response + result.appendElements(lastErrResponse->gleResponse); + return lastErrResponse->gleResponse["ok"].trueValue(); + } else { + // Return a generic combined WC error message + result.append("errs", errors.arr()); + result.append("errObjects", errorRawGLE.arr()); + + // Need to always return err + result.appendNull("err"); + + return appendCommandStatus( + result, + Status(ErrorCodes::WriteConcernFailed, "multiple write concern errors occurred")); } + } - } cmdGetLastError; +} cmdGetLastError; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_get_prev_error_cmd.cpp b/src/mongo/s/commands/cluster_get_prev_error_cmd.cpp index 5e04c0471ea..c0dcdfc618c 100644 --- a/src/mongo/s/commands/cluster_get_prev_error_cmd.cpp +++ b/src/mongo/s/commands/cluster_get_prev_error_cmd.cpp @@ -38,41 +38,39 @@ namespace mongo { namespace { - class GetPrevErrorCmd : public Command { - public: - GetPrevErrorCmd() : Command("getPrevError", false, "getpreverror") { } +class GetPrevErrorCmd : public Command { +public: + GetPrevErrorCmd() : Command("getPrevError", false, "getpreverror") {} - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual void help(std::stringstream& help) const { - help << "get previous error (since last reseterror command)"; - } + virtual void help(std::stringstream& help) const { + help << "get previous error (since last reseterror command)"; + } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + // No auth required + } - // No auth required - } + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + errmsg += "getpreverror not supported for sharded environments"; + return false; + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { +} cmdGetPrevError; - errmsg += "getpreverror not supported for sharded environments"; - return false; - } - - } cmdGetPrevError; - -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_get_shard_map_cmd.cpp b/src/mongo/s/commands/cluster_get_shard_map_cmd.cpp index cd5ee8608fe..0fcdc77d6fc 100644 --- a/src/mongo/s/commands/cluster_get_shard_map_cmd.cpp +++ b/src/mongo/s/commands/cluster_get_shard_map_cmd.cpp @@ -38,50 +38,51 @@ namespace mongo { namespace { - class CmdGetShardMap : public Command { - public: - CmdGetShardMap() : Command("getShardMap") { } +class CmdGetShardMap : public Command { +public: + CmdGetShardMap() : Command("getShardMap") {} - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual void help(std::stringstream& help) const { - help << "lists the set of shards known to this instance"; - } + virtual void help(std::stringstream& help) const { + help << "lists the set of shards known to this instance"; + } - virtual bool adminOnly() const { return true; } + virtual bool adminOnly() const { + return true; + } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::getShardMap); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::getShardMap); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - // MongoD instances do not know that they are part of a sharded cluster until they - // receive a setShardVersion command and that's when the catalog manager and the shard - // registry get initialized. - if (grid.shardRegistry()) { - grid.shardRegistry()->toBSON(&result); - } - - return true; + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + // MongoD instances do not know that they are part of a sharded cluster until they + // receive a setShardVersion command and that's when the catalog manager and the shard + // registry get initialized. + if (grid.shardRegistry()) { + grid.shardRegistry()->toBSON(&result); } - } getShardMapCmd; + return true; + } + +} getShardMapCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_get_shard_version_cmd.cpp b/src/mongo/s/commands/cluster_get_shard_version_cmd.cpp index 486906aaf4a..54d88f7d6bb 100644 --- a/src/mongo/s/commands/cluster_get_shard_version_cmd.cpp +++ b/src/mongo/s/commands/cluster_get_shard_version_cmd.cpp @@ -42,86 +42,83 @@ namespace mongo { - using std::shared_ptr; +using std::shared_ptr; namespace { - class GetShardVersion : public Command { - public: - GetShardVersion() : Command("getShardVersion", false, "getshardversion") { } - - virtual bool slaveOk() const { - return true; +class GetShardVersion : public Command { +public: + GetShardVersion() : Command("getShardVersion", false, "getshardversion") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << " example: { getShardVersion : 'alleyinsider.foo' } "; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::getShardVersion)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); } - virtual bool adminOnly() const { - return true; + return Status::OK(); + } + + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + const NamespaceString nss(parseNs(dbname, cmdObj)); + if (nss.size() == 0) { + return appendCommandStatus( + result, Status(ErrorCodes::InvalidNamespace, "no namespace specified")); } - virtual bool isWriteCommandForConfigServer() const { - return false; + auto status = grid.catalogCache()->getDatabase(nss.db().toString()); + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); } - virtual void help(std::stringstream& help) const { - help << " example: { getShardVersion : 'alleyinsider.foo' } "; + std::shared_ptr<DBConfig> config = status.getValue(); + if (!config->isSharded(nss.ns())) { + return appendCommandStatus( + result, + Status(ErrorCodes::NamespaceNotSharded, "ns [" + nss.ns() + " is not sharded.")); } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace( - NamespaceString(parseNs(dbname, - cmdObj))), - ActionType::getShardVersion)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - - return Status::OK(); + ChunkManagerPtr cm = config->getChunkManagerIfExists(nss.ns()); + if (!cm) { + errmsg = "no chunk manager?"; + return false; } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); - } + cm->_printChunks(); + cm->getVersion().addToBSON(result); - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - const NamespaceString nss(parseNs(dbname, cmdObj)); - if (nss.size() == 0) { - return appendCommandStatus(result, Status(ErrorCodes::InvalidNamespace, - "no namespace specified")); - } - - auto status = grid.catalogCache()->getDatabase(nss.db().toString()); - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } - - std::shared_ptr<DBConfig> config = status.getValue(); - if (!config->isSharded(nss.ns())) { - return appendCommandStatus(result, Status(ErrorCodes::NamespaceNotSharded, - "ns [" + nss.ns() + " is not sharded.")); - } - - ChunkManagerPtr cm = config->getChunkManagerIfExists(nss.ns()); - if (!cm) { - errmsg = "no chunk manager?"; - return false; - } - - cm->_printChunks(); - cm->getVersion().addToBSON(result); - - return true; - } + return true; + } - } getShardVersionCmd; +} getShardVersionCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_index_filter_cmd.cpp b/src/mongo/s/commands/cluster_index_filter_cmd.cpp index 323ad646a33..d3465a4bb48 100644 --- a/src/mongo/s/commands/cluster_index_filter_cmd.cpp +++ b/src/mongo/s/commands/cluster_index_filter_cmd.cpp @@ -38,137 +38,135 @@ namespace mongo { - using std::string; - using std::stringstream; - using std::vector; +using std::string; +using std::stringstream; +using std::vector; - /** - * Base class for mongos index filter commands. - * Cluster index filter commands don't do much more than - * forwarding the commands to all shards and combining the results. - */ - class ClusterIndexFilterCmd : public Command { +/** + * Base class for mongos index filter commands. + * Cluster index filter commands don't do much more than + * forwarding the commands to all shards and combining the results. + */ +class ClusterIndexFilterCmd : public Command { MONGO_DISALLOW_COPYING(ClusterIndexFilterCmd); - public: - virtual ~ClusterIndexFilterCmd() { - } +public: + virtual ~ClusterIndexFilterCmd() {} - bool slaveOk() const { - return false; - } + bool slaveOk() const { + return false; + } - bool slaveOverrideOk() const { - return true; - } + bool slaveOverrideOk() const { + return true; + } - virtual bool isWriteCommandForConfigServer() const { return false; } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - void help(stringstream& ss) const { - ss << _helpText; - } + void help(stringstream& ss) const { + ss << _helpText; + } - Status checkAuthForCommand( ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj ) { - AuthorizationSession* authzSession = AuthorizationSession::get(client); - ResourcePattern pattern = parseResourcePattern(dbname, cmdObj); - - if (authzSession->isAuthorizedForActionsOnResource(pattern, - ActionType::planCacheIndexFilter)) { - return Status::OK(); - } - - return Status(ErrorCodes::Unauthorized, "unauthorized"); - } + Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + AuthorizationSession* authzSession = AuthorizationSession::get(client); + ResourcePattern pattern = parseResourcePattern(dbname, cmdObj); - // Cluster plan cache command entry point. - bool run(OperationContext* txn, const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result); - - public: - - /** - * Instantiates a command that can be invoked by "name", which will be described by - * "helpText", and will require privilege "actionType" to run. - */ - ClusterIndexFilterCmd( const std::string& name, const std::string& helpText) : - Command( name ), _helpText( helpText ) { + if (authzSession->isAuthorizedForActionsOnResource(pattern, + ActionType::planCacheIndexFilter)) { + return Status::OK(); } - private: - - std::string _helpText; - }; - - // - // Cluster index filter command implementation(s) below - // - - bool ClusterIndexFilterCmd::run(OperationContext* txn, const std::string& dbName, - BSONObj& cmdObj, - int options, - std::string& errMsg, - BSONObjBuilder& result) { - const std::string fullns = parseNs(dbName, cmdObj); - NamespaceString nss(fullns); - - // Dispatch command to all the shards. - // Targeted shard commands are generally data-dependent but index filter - // commands are tied to query shape (data has no effect on query shape). - vector<Strategy::CommandResult> results; - Strategy::commandOp(dbName, cmdObj, options, nss.ns(), BSONObj(), &results); - - // Set value of first shard result's "ok" field. - bool clusterCmdResult = true; - - for (vector<Strategy::CommandResult>::const_iterator i = results.begin(); - i != results.end(); ++i) { - const Strategy::CommandResult& cmdResult = *i; - - // XXX: In absence of sensible aggregation strategy, - // promote first shard's result to top level. - if (i == results.begin()) { - result.appendElements(cmdResult.result); - clusterCmdResult = cmdResult.result["ok"].trueValue(); - } - - // Append shard result as a sub object. - // Name the field after the shard. - result.append(cmdResult.shardTargetId, cmdResult.result); + return Status(ErrorCodes::Unauthorized, "unauthorized"); + } + + // Cluster plan cache command entry point. + bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result); + +public: + /** + * Instantiates a command that can be invoked by "name", which will be described by + * "helpText", and will require privilege "actionType" to run. + */ + ClusterIndexFilterCmd(const std::string& name, const std::string& helpText) + : Command(name), _helpText(helpText) {} + +private: + std::string _helpText; +}; + +// +// Cluster index filter command implementation(s) below +// + +bool ClusterIndexFilterCmd::run(OperationContext* txn, + const std::string& dbName, + BSONObj& cmdObj, + int options, + std::string& errMsg, + BSONObjBuilder& result) { + const std::string fullns = parseNs(dbName, cmdObj); + NamespaceString nss(fullns); + + // Dispatch command to all the shards. + // Targeted shard commands are generally data-dependent but index filter + // commands are tied to query shape (data has no effect on query shape). + vector<Strategy::CommandResult> results; + Strategy::commandOp(dbName, cmdObj, options, nss.ns(), BSONObj(), &results); + + // Set value of first shard result's "ok" field. + bool clusterCmdResult = true; + + for (vector<Strategy::CommandResult>::const_iterator i = results.begin(); i != results.end(); + ++i) { + const Strategy::CommandResult& cmdResult = *i; + + // XXX: In absence of sensible aggregation strategy, + // promote first shard's result to top level. + if (i == results.begin()) { + result.appendElements(cmdResult.result); + clusterCmdResult = cmdResult.result["ok"].trueValue(); } - return clusterCmdResult; + // Append shard result as a sub object. + // Name the field after the shard. + result.append(cmdResult.shardTargetId, cmdResult.result); } - // - // Register index filter commands at startup - // + return clusterCmdResult; +} - namespace { +// +// Register index filter commands at startup +// - MONGO_INITIALIZER(RegisterIndexFilterCommands)(InitializerContext* context) { - // Leaked intentionally: a Command registers itself when constructed. +namespace { - new ClusterIndexFilterCmd( - "planCacheListFilters", - "Displays index filters for all query shapes in a collection." ); +MONGO_INITIALIZER(RegisterIndexFilterCommands)(InitializerContext* context) { + // Leaked intentionally: a Command registers itself when constructed. - new ClusterIndexFilterCmd( - "planCacheClearFilters", - "Clears index filter for a single query shape or, " - "if the query shape is omitted, all filters for the collection." ); + new ClusterIndexFilterCmd("planCacheListFilters", + "Displays index filters for all query shapes in a collection."); - new ClusterIndexFilterCmd( - "planCacheSetFilter", - "Sets index filter for a query shape. Overrides existing index filter." ); + new ClusterIndexFilterCmd("planCacheClearFilters", + "Clears index filter for a single query shape or, " + "if the query shape is omitted, all filters for the collection."); - return Status::OK(); - } + new ClusterIndexFilterCmd( + "planCacheSetFilter", + "Sets index filter for a query shape. Overrides existing index filter."); + + return Status::OK(); +} - } // namespace +} // namespace -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_is_db_grid_cmd.cpp b/src/mongo/s/commands/cluster_is_db_grid_cmd.cpp index d4c958582ac..3c16419807d 100644 --- a/src/mongo/s/commands/cluster_is_db_grid_cmd.cpp +++ b/src/mongo/s/commands/cluster_is_db_grid_cmd.cpp @@ -34,38 +34,36 @@ namespace mongo { namespace { - class IsDbGridCmd : public Command { - public: - IsDbGridCmd() : Command("isdbgrid") { } +class IsDbGridCmd : public Command { +public: + IsDbGridCmd() : Command("isdbgrid") {} - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + // No auth required + } - // No auth required - } + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + result.append("isdbgrid", 1); + result.append("hostname", getHostNameCached()); + return true; + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { +} isdbGrid; - result.append("isdbgrid", 1); - result.append("hostname", getHostNameCached()); - return true; - } - - } isdbGrid; - -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_is_master_cmd.cpp b/src/mongo/s/commands/cluster_is_master_cmd.cpp index 5f926811a27..75dde8f15f9 100644 --- a/src/mongo/s/commands/cluster_is_master_cmd.cpp +++ b/src/mongo/s/commands/cluster_is_master_cmd.cpp @@ -35,52 +35,50 @@ namespace mongo { namespace { - class CmdIsMaster : public Command { - public: - CmdIsMaster() : Command("isMaster", false, "ismaster") { } +class CmdIsMaster : public Command { +public: + CmdIsMaster() : Command("isMaster", false, "ismaster") {} - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual void help(std::stringstream& help) const { - help << "test if this is master half of a replica pair"; - } + virtual void help(std::stringstream& help) const { + help << "test if this is master half of a replica pair"; + } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + // No auth required + } - // No auth required - } + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + result.appendBool("ismaster", true); + result.append("msg", "isdbgrid"); + result.appendNumber("maxBsonObjectSize", BSONObjMaxUserSize); + result.appendNumber("maxMessageSizeBytes", MaxMessageSizeBytes); + result.appendNumber("maxWriteBatchSize", BatchedCommandRequest::kMaxWriteBatchSize); + result.appendDate("localTime", jsTime()); - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + // Mongos tries to keep exactly the same version range of the server for which + // it is compiled. + result.append("maxWireVersion", maxWireVersion); + result.append("minWireVersion", minWireVersion); - result.appendBool("ismaster", true); - result.append("msg", "isdbgrid"); - result.appendNumber("maxBsonObjectSize", BSONObjMaxUserSize); - result.appendNumber("maxMessageSizeBytes", MaxMessageSizeBytes); - result.appendNumber("maxWriteBatchSize", BatchedCommandRequest::kMaxWriteBatchSize); - result.appendDate("localTime", jsTime()); + return true; + } - // Mongos tries to keep exactly the same version range of the server for which - // it is compiled. - result.append("maxWireVersion", maxWireVersion); - result.append("minWireVersion", minWireVersion); +} isMaster; - return true; - } - - } isMaster; - -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_kill_op.cpp b/src/mongo/s/commands/cluster_kill_op.cpp index e6f3a127b56..056f4b7ff60 100644 --- a/src/mongo/s/commands/cluster_kill_op.cpp +++ b/src/mongo/s/commands/cluster_kill_op.cpp @@ -50,81 +50,84 @@ namespace mongo { namespace { - class ClusterKillOpCommand : public Command { - public: - ClusterKillOpCommand() : Command("killOp") {} - - bool isWriteCommandForConfigServer() const final { return false; } +class ClusterKillOpCommand : public Command { +public: + ClusterKillOpCommand() : Command("killOp") {} + + bool isWriteCommandForConfigServer() const final { + return false; + } + + bool slaveOk() const final { + return true; + } + + bool adminOnly() const final { + return true; + } + + Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) final { + bool isAuthorized = AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forClusterResource(), ActionType::killop); + return isAuthorized ? Status::OK() : Status(ErrorCodes::Unauthorized, "Unauthorized"); + } + + bool run(OperationContext* txn, + const std::string& db, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) final { + // The format of op is shardid:opid + // This is different than the format passed to the mongod killOp command. + std::string opToKill; + uassertStatusOK(bsonExtractStringField(cmdObj, "op", &opToKill)); + + const auto opSepPos = opToKill.find(':'); + + uassert(28625, + str::stream() << "The op argument to killOp must be of the format shardid:opid" + << " but found \"" << opToKill << '"', + (opToKill.size() >= 3) && // must have at least N:N + (opSepPos != std::string::npos) && // must have ':' as separator + (opSepPos != 0) && // can't be :NN + (opSepPos != (opToKill.size() - 1))); // can't be NN: + + auto shardIdent = opToKill.substr(0, opSepPos); + log() << "want to kill op: " << opToKill; + + // Will throw if shard id is not found + auto shard = grid.shardRegistry()->getShard(shardIdent); + if (!shard) { + return appendCommandStatus( + result, + Status(ErrorCodes::ShardNotFound, + str::stream() << "shard " << shardIdent << " does not exist")); + } - bool slaveOk() const final { return true; } + auto opId = std::stoi(opToKill.substr(opSepPos + 1)); - bool adminOnly() const final { return true; } + // shardid is actually the opid - keeping for backwards compatibility. + result.append("shard", shardIdent); + result.append("shardid", opId); - Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) final { + ScopedDbConnection conn(shard->getConnString()); + BSONObj cmdRes; + BSONObjBuilder argsBob; + argsBob.append("op", opId); + auto args = argsBob.done(); + // intentionally ignore return value - that is how legacy killOp worked. + conn->runPseudoCommand("admin", "killOp", "$cmd.sys.killop", args, cmdRes); + conn.done(); - bool isAuthorized = AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forClusterResource(), - ActionType::killop); - return isAuthorized ? Status::OK() : Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - - bool run(OperationContext* txn, - const std::string& db, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) final { - - // The format of op is shardid:opid - // This is different than the format passed to the mongod killOp command. - std::string opToKill; - uassertStatusOK(bsonExtractStringField(cmdObj, "op", &opToKill)); - - const auto opSepPos = opToKill.find(':'); - - uassert(28625, - str::stream() << "The op argument to killOp must be of the format shardid:opid" - << " but found \"" << opToKill << '"', - (opToKill.size() >= 3) && // must have at least N:N - (opSepPos != std::string::npos) && // must have ':' as separator - (opSepPos != 0) && // can't be :NN - (opSepPos != (opToKill.size() - 1))); // can't be NN: - - auto shardIdent = opToKill.substr(0, opSepPos); - log() << "want to kill op: " << opToKill; - - // Will throw if shard id is not found - auto shard = grid.shardRegistry()->getShard(shardIdent); - if (!shard) { - return appendCommandStatus(result, - Status(ErrorCodes::ShardNotFound, - str::stream() << "shard " << shardIdent - << " does not exist")); - } - - auto opId = std::stoi(opToKill.substr(opSepPos + 1)); - - // shardid is actually the opid - keeping for backwards compatibility. - result.append("shard", shardIdent); - result.append("shardid", opId); - - ScopedDbConnection conn(shard->getConnString()); - BSONObj cmdRes; - BSONObjBuilder argsBob; - argsBob.append("op", opId); - auto args = argsBob.done(); - // intentionally ignore return value - that is how legacy killOp worked. - conn->runPseudoCommand("admin", "killOp", "$cmd.sys.killop", args, cmdRes); - conn.done(); - - // The original behavior of killOp on mongos is to always return success, regardless of - // whether the shard reported success or not. - return true; - } + // The original behavior of killOp on mongos is to always return success, regardless of + // whether the shard reported success or not. + return true; + } - } clusterKillOpCommand; +} clusterKillOpCommand; } // namespace } // namespace mongo diff --git a/src/mongo/s/commands/cluster_list_databases_cmd.cpp b/src/mongo/s/commands/cluster_list_databases_cmd.cpp index 1fc52781cce..248fb79f488 100644 --- a/src/mongo/s/commands/cluster_list_databases_cmd.cpp +++ b/src/mongo/s/commands/cluster_list_databases_cmd.cpp @@ -41,173 +41,169 @@ namespace mongo { - using std::unique_ptr; - using std::map; - using std::string; - using std::vector; +using std::unique_ptr; +using std::map; +using std::string; +using std::vector; namespace { - class ListDatabasesCmd : public Command { - public: - ListDatabasesCmd() : Command("listDatabases", true, "listdatabases") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool slaveOverrideOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; - } - - virtual bool isWriteCommandForConfigServer() const { - return false; - } - - virtual void help(std::stringstream& help) const { - help << "list databases in a cluster"; - } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::listDatabases); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - - virtual bool run(OperationContext* txn, - const std::string& dbname_unused, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - map<string, long long> sizes; - map<string, unique_ptr<BSONObjBuilder> > dbShardInfo; - - vector<ShardId> shardIds; - grid.shardRegistry()->getAllShardIds(&shardIds); - - for (const ShardId& shardId : shardIds) { - const auto s = grid.shardRegistry()->getShard(shardId); - if (!s) { - continue; - } - - BSONObj x = s->runCommand("admin", "listDatabases"); +class ListDatabasesCmd : public Command { +public: + ListDatabasesCmd() : Command("listDatabases", true, "listdatabases") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool slaveOverrideOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "list databases in a cluster"; + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::listDatabases); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname_unused, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + map<string, long long> sizes; + map<string, unique_ptr<BSONObjBuilder>> dbShardInfo; + + vector<ShardId> shardIds; + grid.shardRegistry()->getAllShardIds(&shardIds); + + for (const ShardId& shardId : shardIds) { + const auto s = grid.shardRegistry()->getShard(shardId); + if (!s) { + continue; + } - BSONObjIterator j(x["databases"].Obj()); - while (j.more()) { - BSONObj dbObj = j.next().Obj(); + BSONObj x = s->runCommand("admin", "listDatabases"); - const string name = dbObj["name"].String(); - const long long size = dbObj["sizeOnDisk"].numberLong(); + BSONObjIterator j(x["databases"].Obj()); + while (j.more()) { + BSONObj dbObj = j.next().Obj(); - long long& totalSize = sizes[name]; - if (size == 1) { - if (totalSize <= 1) { - totalSize = 1; - } - } - else { - totalSize += size; - } + const string name = dbObj["name"].String(); + const long long size = dbObj["sizeOnDisk"].numberLong(); - unique_ptr<BSONObjBuilder>& bb = dbShardInfo[name]; - if (!bb.get()) { - bb.reset(new BSONObjBuilder()); + long long& totalSize = sizes[name]; + if (size == 1) { + if (totalSize <= 1) { + totalSize = 1; } + } else { + totalSize += size; + } - bb->appendNumber(s->getId(), size); + unique_ptr<BSONObjBuilder>& bb = dbShardInfo[name]; + if (!bb.get()) { + bb.reset(new BSONObjBuilder()); } + bb->appendNumber(s->getId(), size); } + } - long long totalSize = 0; + long long totalSize = 0; - BSONArrayBuilder bb(result.subarrayStart("databases")); - for (map<string, long long>::iterator i = sizes.begin(); i != sizes.end(); ++i) { - const string name = i->first; + BSONArrayBuilder bb(result.subarrayStart("databases")); + for (map<string, long long>::iterator i = sizes.begin(); i != sizes.end(); ++i) { + const string name = i->first; - if (name == "local") { - // We don't return local, since all shards have their own independent local - continue; - } + if (name == "local") { + // We don't return local, since all shards have their own independent local + continue; + } - if (name == "config" || name == "admin") { - // Always get this from the config servers - continue; - } + if (name == "config" || name == "admin") { + // Always get this from the config servers + continue; + } - long long size = i->second; - totalSize += size; + long long size = i->second; + totalSize += size; - BSONObjBuilder temp; - temp.append("name", name); - temp.appendNumber("sizeOnDisk", size); - temp.appendBool("empty", size == 1); - temp.append("shards", dbShardInfo[name]->obj()); + BSONObjBuilder temp; + temp.append("name", name); + temp.appendNumber("sizeOnDisk", size); + temp.appendBool("empty", size == 1); + temp.append("shards", dbShardInfo[name]->obj()); - bb.append(temp.obj()); - } + bb.append(temp.obj()); + } - // Obtain the cached config shard - const auto configShard = grid.shardRegistry()->getShard("config"); - - { - // get config db from the config servers (first one) - BSONObj x; - if (configShard->runCommand("config", "dbstats", x)) { - BSONObjBuilder b; - b.append("name", "config"); - b.appendBool("empty", false); - if (x["fileSize"].type()) - b.appendAs(x["fileSize"], "sizeOnDisk"); - else - b.append("sizeOnDisk", 1); - bb.append(b.obj()); - } - else { - bb.append(BSON("name" << "config")); - } + // Obtain the cached config shard + const auto configShard = grid.shardRegistry()->getShard("config"); + + { + // get config db from the config servers (first one) + BSONObj x; + if (configShard->runCommand("config", "dbstats", x)) { + BSONObjBuilder b; + b.append("name", "config"); + b.appendBool("empty", false); + if (x["fileSize"].type()) + b.appendAs(x["fileSize"], "sizeOnDisk"); + else + b.append("sizeOnDisk", 1); + bb.append(b.obj()); + } else { + bb.append(BSON("name" + << "config")); } + } - { - // get admin db from the config servers (first one) - BSONObj x; - if (configShard->runCommand("admin", "dbstats", x)) { - BSONObjBuilder b; - b.append("name", "admin"); - b.appendBool("empty", false); - - if (x["fileSize"].type()) { - b.appendAs(x["fileSize"], "sizeOnDisk"); - } - else { - b.append("sizeOnDisk", 1); - } - - bb.append(b.obj()); - } - else { - bb.append(BSON("name" << "admin")); + { + // get admin db from the config servers (first one) + BSONObj x; + if (configShard->runCommand("admin", "dbstats", x)) { + BSONObjBuilder b; + b.append("name", "admin"); + b.appendBool("empty", false); + + if (x["fileSize"].type()) { + b.appendAs(x["fileSize"], "sizeOnDisk"); + } else { + b.append("sizeOnDisk", 1); } + + bb.append(b.obj()); + } else { + bb.append(BSON("name" + << "admin")); } + } - bb.done(); + bb.done(); - result.appendNumber("totalSize", totalSize); - result.appendNumber("totalSizeMb", totalSize / (1024 * 1024)); + result.appendNumber("totalSize", totalSize); + result.appendNumber("totalSizeMb", totalSize / (1024 * 1024)); - return 1; - } + return 1; + } - } cmdListDatabases; +} cmdListDatabases; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_list_shards_cmd.cpp b/src/mongo/s/commands/cluster_list_shards_cmd.cpp index 3cb365710e5..16eccd3c8e1 100644 --- a/src/mongo/s/commands/cluster_list_shards_cmd.cpp +++ b/src/mongo/s/commands/cluster_list_shards_cmd.cpp @@ -39,59 +39,55 @@ namespace mongo { namespace { - class ListShardsCmd : public Command { - public: - ListShardsCmd() : Command("listShards", false, "listshards") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; - } - - virtual bool isWriteCommandForConfigServer() const { - return false; - } - - virtual void help(std::stringstream& help) const { - help << "list all shards of the system"; +class ListShardsCmd : public Command { +public: + ListShardsCmd() : Command("listShards", false, "listshards") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "list all shards of the system"; + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::listShards); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + std::vector<ShardType> shards; + Status status = grid.catalogManager()->getAllShards(&shards); + if (!status.isOK()) { + return appendCommandStatus(result, status); } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - - ActionSet actions; - actions.addAction(ActionType::listShards); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + std::vector<BSONObj> shardsObj; + for (std::vector<ShardType>::const_iterator it = shards.begin(); it != shards.end(); it++) { + shardsObj.push_back(it->toBSON()); } + result.append("shards", shardsObj); - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - std::vector<ShardType> shards; - Status status = grid.catalogManager()->getAllShards(&shards); - if (!status.isOK()) { - return appendCommandStatus(result, status); - } - std::vector<BSONObj> shardsObj; - for (std::vector<ShardType>::const_iterator it = shards.begin(); - it != shards.end(); - it++) { - shardsObj.push_back(it->toBSON()); - } - result.append("shards", shardsObj); - - return true; - } + return true; + } - } listShards; +} listShards; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_map_reduce_cmd.cpp b/src/mongo/s/commands/cluster_map_reduce_cmd.cpp index f4a0347de5c..78ecd3ae927 100644 --- a/src/mongo/s/commands/cluster_map_reduce_cmd.cpp +++ b/src/mongo/s/commands/cluster_map_reduce_cmd.cpp @@ -55,564 +55,536 @@ namespace mongo { - using std::shared_ptr; - using std::map; - using std::set; - using std::string; - using std::vector; +using std::shared_ptr; +using std::map; +using std::set; +using std::string; +using std::vector; namespace { - AtomicUInt32 JOB_NUMBER; +AtomicUInt32 JOB_NUMBER; - /** - * Generates a unique name for the temporary M/R output collection. - */ - string getTmpName(const string& coll) { - StringBuilder sb; - sb << "tmp.mrs." << coll << "_" << time(0) << "_" << JOB_NUMBER.fetchAndAdd(1); - return sb.str(); - } +/** + * Generates a unique name for the temporary M/R output collection. + */ +string getTmpName(const string& coll) { + StringBuilder sb; + sb << "tmp.mrs." << coll << "_" << time(0) << "_" << JOB_NUMBER.fetchAndAdd(1); + return sb.str(); +} - /** - * Given an input map/reduce command, this call generates the matching command which should - * be sent to the shards as part of the first phase of map/reduce. - */ - BSONObj fixForShards(const BSONObj& orig, - const string& output, - string& badShardedField, - int maxChunkSizeBytes) { - - BSONObjBuilder b; - BSONObjIterator i(orig); - while (i.more()) { - BSONElement e = i.next(); - const string fn = e.fieldName(); - - if (fn == bypassDocumentValidationCommandOption() || - fn == "map" || - fn == "mapreduce" || - fn == "mapReduce" || - fn == "mapparams" || - fn == "reduce" || - fn == "query" || - fn == "sort" || - fn == "scope" || - fn == "verbose" || - fn == "$queryOptions" || - fn == LiteParsedQuery::cmdOptionMaxTimeMS) { - - b.append(e); - } - else if (fn == "out" || fn == "finalize") { - // We don't want to copy these - } - else { - badShardedField = fn; - return BSONObj(); - } +/** + * Given an input map/reduce command, this call generates the matching command which should + * be sent to the shards as part of the first phase of map/reduce. + */ +BSONObj fixForShards(const BSONObj& orig, + const string& output, + string& badShardedField, + int maxChunkSizeBytes) { + BSONObjBuilder b; + BSONObjIterator i(orig); + while (i.more()) { + BSONElement e = i.next(); + const string fn = e.fieldName(); + + if (fn == bypassDocumentValidationCommandOption() || fn == "map" || fn == "mapreduce" || + fn == "mapReduce" || fn == "mapparams" || fn == "reduce" || fn == "query" || + fn == "sort" || fn == "scope" || fn == "verbose" || fn == "$queryOptions" || + fn == LiteParsedQuery::cmdOptionMaxTimeMS) { + b.append(e); + } else if (fn == "out" || fn == "finalize") { + // We don't want to copy these + } else { + badShardedField = fn; + return BSONObj(); } + } - b.append("out", output); - b.append("shardedFirstPass", true); + b.append("out", output); + b.append("shardedFirstPass", true); - if (maxChunkSizeBytes > 0) { - // Will need to figure out chunks, ask shards for points - b.append("splitInfo", maxChunkSizeBytes); - } + if (maxChunkSizeBytes > 0) { + // Will need to figure out chunks, ask shards for points + b.append("splitInfo", maxChunkSizeBytes); + } + + return b.obj(); +} + + +/** + * Outline for sharded map reduce for sharded output, $out replace: + * + * ============= mongos ============= + * 1. Send map reduce command to all relevant shards with some extra info like the value for + * the chunkSize and the name of the temporary output collection. + * + * ============= shard ============= + * 2. Does normal map reduce. + * + * 3. Calls splitVector on itself against the output collection and puts the results into the + * response object. + * + * ============= mongos ============= + * 4. If the output collection is *not* sharded, uses the information from splitVector to + * create a pre-split sharded collection. + * + * 5. Grabs the distributed lock for the final output collection. + * + * 6. Sends mapReduce.shardedfinish. + * + * ============= shard ============= + * 7. Extracts the list of shards from the mapReduce.shardedfinish and performs a broadcast + * query against all of them to obtain all documents that this shard owns. + * + * 8. Performs the reduce operation against every document from step #7 and outputs them to + * another temporary collection. Also keeps track of the BSONObject size of every "reduced" + * document for each chunk range. + * + * 9. Atomically drops the old output collection and renames the temporary collection to the + * output collection. + * + * ============= mongos ============= + * 10. Releases the distributed lock acquired at step #5. + * + * 11. Inspects the BSONObject size from step #8 and determines if it needs to split. + */ +class MRCmd : public Command { +public: + MRCmd() : Command("mapReduce", false, "mapreduce") {} - return b.obj(); + virtual bool slaveOk() const { + return true; } + virtual bool adminOnly() const { + return false; + } - /** - * Outline for sharded map reduce for sharded output, $out replace: - * - * ============= mongos ============= - * 1. Send map reduce command to all relevant shards with some extra info like the value for - * the chunkSize and the name of the temporary output collection. - * - * ============= shard ============= - * 2. Does normal map reduce. - * - * 3. Calls splitVector on itself against the output collection and puts the results into the - * response object. - * - * ============= mongos ============= - * 4. If the output collection is *not* sharded, uses the information from splitVector to - * create a pre-split sharded collection. - * - * 5. Grabs the distributed lock for the final output collection. - * - * 6. Sends mapReduce.shardedfinish. - * - * ============= shard ============= - * 7. Extracts the list of shards from the mapReduce.shardedfinish and performs a broadcast - * query against all of them to obtain all documents that this shard owns. - * - * 8. Performs the reduce operation against every document from step #7 and outputs them to - * another temporary collection. Also keeps track of the BSONObject size of every "reduced" - * document for each chunk range. - * - * 9. Atomically drops the old output collection and renames the temporary collection to the - * output collection. - * - * ============= mongos ============= - * 10. Releases the distributed lock acquired at step #5. - * - * 11. Inspects the BSONObject size from step #8 and determines if it needs to split. - */ - class MRCmd : public Command { - public: - MRCmd() : Command("mapReduce", false, "mapreduce") { } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual bool slaveOk() const { - return true; - } + virtual void help(std::stringstream& help) const { + help << "Runs the sharded map/reduce command"; + } - virtual bool adminOnly() const { - return false; - } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + mr::addPrivilegesRequiredForMapReduce(this, dbname, cmdObj, out); + } - virtual bool isWriteCommandForConfigServer() const { - return false; + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + Timer t; + + const string collection = cmdObj.firstElement().valuestrsafe(); + const string fullns = dbname + "." + collection; + const string shardResultCollection = getTmpName(collection); + + BSONObj customOut; + string finalColShort; + string finalColLong; + bool customOutDB = false; + + string outDB = dbname; + + BSONElement outElmt = cmdObj.getField("out"); + if (outElmt.type() == Object) { + // Check if there is a custom output + BSONObj out = outElmt.embeddedObject(); + customOut = out; + + // Mode must be 1st element + finalColShort = out.firstElement().str(); + if (customOut.hasField("db")) { + customOutDB = true; + outDB = customOut.getField("db").str(); + } + + finalColLong = outDB + "." + finalColShort; } - virtual void help(std::stringstream& help) const { - help << "Runs the sharded map/reduce command"; + // Ensure the input database exists + auto status = grid.catalogCache()->getDatabase(dbname); + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + shared_ptr<DBConfig> confIn = status.getValue(); - mr::addPrivilegesRequiredForMapReduce(this, dbname, cmdObj, out); + shared_ptr<DBConfig> confOut; + if (customOutDB) { + // Create the output database implicitly, since we have a custom output requested + confOut = uassertStatusOK(grid.implicitCreateDb(outDB)); + } else { + confOut = confIn; } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - Timer t; - - const string collection = cmdObj.firstElement().valuestrsafe(); - const string fullns = dbname + "." + collection; - const string shardResultCollection = getTmpName(collection); - - BSONObj customOut; - string finalColShort; - string finalColLong; - bool customOutDB = false; - - string outDB = dbname; - - BSONElement outElmt = cmdObj.getField("out"); - if (outElmt.type() == Object) { - // Check if there is a custom output - BSONObj out = outElmt.embeddedObject(); - customOut = out; - - // Mode must be 1st element - finalColShort = out.firstElement().str(); - if (customOut.hasField("db")) { - customOutDB = true; - outDB = customOut.getField("db").str(); - } + const bool shardedInput = + confIn && confIn->isShardingEnabled() && confIn->isSharded(fullns); + const bool shardedOutput = customOut.getBoolField("sharded"); - finalColLong = outDB + "." + finalColShort; - } - - // Ensure the input database exists - auto status = grid.catalogCache()->getDatabase(dbname); - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } + if (!shardedOutput) { + uassert(15920, + "Cannot output to a non-sharded collection because " + "sharded collection exists already", + !confOut->isSharded(finalColLong)); - shared_ptr<DBConfig> confIn = status.getValue(); + // TODO: Should we also prevent going from non-sharded to sharded? During the + // transition client may see partial data. + } - shared_ptr<DBConfig> confOut; - if (customOutDB) { - // Create the output database implicitly, since we have a custom output requested - confOut = uassertStatusOK(grid.implicitCreateDb(outDB)); - } - else { - confOut = confIn; + int64_t maxChunkSizeBytes = 0; + if (shardedOutput) { + // Will need to figure out chunks, ask shards for points + maxChunkSizeBytes = cmdObj["maxChunkSizeBytes"].numberLong(); + if (maxChunkSizeBytes == 0) { + maxChunkSizeBytes = Chunk::MaxChunkSize; } - const bool shardedInput = confIn && - confIn->isShardingEnabled() && - confIn->isSharded(fullns); - const bool shardedOutput = customOut.getBoolField("sharded"); + // maxChunkSizeBytes is sent as int BSON field + invariant(maxChunkSizeBytes < std::numeric_limits<int>::max()); + } - if (!shardedOutput) { - uassert(15920, - "Cannot output to a non-sharded collection because " - "sharded collection exists already", - !confOut->isSharded(finalColLong)); + if (customOut.hasField("inline") && shardedOutput) { + errmsg = "cannot specify inline and sharded output at the same time"; + return false; + } - // TODO: Should we also prevent going from non-sharded to sharded? During the - // transition client may see partial data. - } + // modify command to run on shards with output to tmp collection + string badShardedField; + BSONObj shardedCommand = + fixForShards(cmdObj, shardResultCollection, badShardedField, maxChunkSizeBytes); - int64_t maxChunkSizeBytes = 0; - if (shardedOutput) { - // Will need to figure out chunks, ask shards for points - maxChunkSizeBytes = cmdObj["maxChunkSizeBytes"].numberLong(); - if (maxChunkSizeBytes == 0) { - maxChunkSizeBytes = Chunk::MaxChunkSize; - } + if (!shardedInput && !shardedOutput && !customOutDB) { + LOG(1) << "simple MR, just passthrough"; - // maxChunkSizeBytes is sent as int BSON field - invariant(maxChunkSizeBytes < std::numeric_limits<int>::max()); - } + const auto shard = grid.shardRegistry()->getShard(confIn->getPrimaryId()); + ShardConnection conn(shard->getConnString(), ""); - if (customOut.hasField("inline") && shardedOutput) { - errmsg = "cannot specify inline and sharded output at the same time"; - return false; - } + BSONObj res; + bool ok = conn->runCommand(dbname, cmdObj, res); + conn.done(); - // modify command to run on shards with output to tmp collection - string badShardedField; - BSONObj shardedCommand = fixForShards(cmdObj, - shardResultCollection, - badShardedField, - maxChunkSizeBytes); + result.appendElements(res); + return ok; + } - if (!shardedInput && !shardedOutput && !customOutDB) { - LOG(1) << "simple MR, just passthrough"; + if (badShardedField.size()) { + errmsg = str::stream() << "unknown m/r field for sharding: " << badShardedField; + return false; + } - const auto shard = grid.shardRegistry()->getShard(confIn->getPrimaryId()); - ShardConnection conn(shard->getConnString(), ""); + BSONObj q; + if (cmdObj["query"].type() == Object) { + q = cmdObj["query"].embeddedObjectUserCheck(); + } - BSONObj res; - bool ok = conn->runCommand(dbname, cmdObj, res); - conn.done(); + set<string> servers; + vector<Strategy::CommandResult> mrCommandResults; - result.appendElements(res); - return ok; - } + BSONObjBuilder shardResultsB; + BSONObjBuilder shardCountsB; + map<string, int64_t> countsMap; + set<BSONObj> splitPts; - if (badShardedField.size()) { - errmsg = str::stream() << "unknown m/r field for sharding: " << badShardedField; - return false; - } + { + bool ok = true; - BSONObj q; - if (cmdObj["query"].type() == Object) { - q = cmdObj["query"].embeddedObjectUserCheck(); - } + // TODO: take distributed lock to prevent split / migration? - set<string> servers; - vector<Strategy::CommandResult> mrCommandResults; + try { + Strategy::commandOp(dbname, shardedCommand, 0, fullns, q, &mrCommandResults); + } catch (DBException& e) { + e.addContext(str::stream() << "could not run map command on all shards for ns " + << fullns << " and query " << q); + throw; + } - BSONObjBuilder shardResultsB; - BSONObjBuilder shardCountsB; - map<string, int64_t> countsMap; - set< BSONObj > splitPts; + for (const auto& mrResult : mrCommandResults) { + // Need to gather list of all servers even if an error happened + string server; + { + const auto shard = grid.shardRegistry()->getShard(mrResult.shardTargetId); + server = shard->getConnString().toString(); + } + servers.insert(server); - { - bool ok = true; + if (!ok) { + continue; + } - // TODO: take distributed lock to prevent split / migration? + BSONObj singleResult = mrResult.result; + ok = singleResult["ok"].trueValue(); - try { - Strategy::commandOp(dbname, shardedCommand, 0, fullns, q, &mrCommandResults); - } - catch (DBException& e){ - e.addContext(str::stream() << "could not run map command on all shards for ns " - << fullns << " and query " << q); - throw; + if (!ok) { + // At this point we will return + errmsg = str::stream() + << "MR parallel processing failed: " << singleResult.toString(); + continue; } - for (const auto& mrResult : mrCommandResults) { - // Need to gather list of all servers even if an error happened - string server; - { - const auto shard = - grid.shardRegistry()->getShard(mrResult.shardTargetId); - server = shard->getConnString().toString(); - } - servers.insert(server); + shardResultsB.append(server, singleResult); - if (!ok) { - continue; - } + BSONObj counts = singleResult["counts"].embeddedObjectUserCheck(); + shardCountsB.append(server, counts); - BSONObj singleResult = mrResult.result; - ok = singleResult["ok"].trueValue(); + // Add up the counts for each shard. Some of them will be fixed later like + // output and reduce. + BSONObjIterator j(counts); + while (j.more()) { + BSONElement temp = j.next(); + countsMap[temp.fieldName()] += temp.numberLong(); + } - if (!ok) { - // At this point we will return - errmsg = str::stream() << "MR parallel processing failed: " - << singleResult.toString(); - continue; + if (singleResult.hasField("splitKeys")) { + BSONElement splitKeys = singleResult.getField("splitKeys"); + vector<BSONElement> pts = splitKeys.Array(); + for (vector<BSONElement>::iterator it = pts.begin(); it != pts.end(); ++it) { + splitPts.insert(it->Obj().getOwned()); } + } + } - shardResultsB.append(server, singleResult); + if (!ok) { + _cleanUp(servers, dbname, shardResultCollection); - BSONObj counts = singleResult["counts"].embeddedObjectUserCheck(); - shardCountsB.append(server, counts); + // Add "code" to the top-level response, if the failure of the sharded command + // can be accounted to a single error. + int code = getUniqueCodeFromCommandResults(mrCommandResults); + if (code != 0) { + result.append("code", code); + } - // Add up the counts for each shard. Some of them will be fixed later like - // output and reduce. - BSONObjIterator j(counts); - while (j.more()) { - BSONElement temp = j.next(); - countsMap[temp.fieldName()] += temp.numberLong(); - } + return false; + } + } - if (singleResult.hasField("splitKeys")) { - BSONElement splitKeys = singleResult.getField("splitKeys"); - vector<BSONElement> pts = splitKeys.Array(); - for (vector<BSONElement>::iterator it = pts.begin(); it != pts.end(); ++it) { - splitPts.insert(it->Obj().getOwned()); - } - } - } + // Build the sharded finish command + BSONObjBuilder finalCmd; + finalCmd.append("mapreduce.shardedfinish", cmdObj); + finalCmd.append("inputDB", dbname); + finalCmd.append("shardedOutputCollection", shardResultCollection); + finalCmd.append("shards", shardResultsB.done()); - if (!ok) { - _cleanUp(servers, dbname, shardResultCollection); + BSONObj shardCounts = shardCountsB.done(); + finalCmd.append("shardCounts", shardCounts); - // Add "code" to the top-level response, if the failure of the sharded command - // can be accounted to a single error. - int code = getUniqueCodeFromCommandResults(mrCommandResults); - if (code != 0) { - result.append("code", code); - } + BSONObjBuilder timingBuilder; + timingBuilder.append("shardProcessing", t.millis()); - return false; - } - } + BSONObjBuilder aggCountsB; + for (const auto& countEntry : countsMap) { + aggCountsB.append(countEntry.first, static_cast<long long>(countEntry.second)); + } - // Build the sharded finish command - BSONObjBuilder finalCmd; - finalCmd.append("mapreduce.shardedfinish", cmdObj); - finalCmd.append("inputDB", dbname); - finalCmd.append("shardedOutputCollection", shardResultCollection); - finalCmd.append("shards", shardResultsB.done()); + BSONObj aggCounts = aggCountsB.done(); + finalCmd.append("counts", aggCounts); - BSONObj shardCounts = shardCountsB.done(); - finalCmd.append("shardCounts", shardCounts); + if (auto elem = cmdObj[LiteParsedQuery::cmdOptionMaxTimeMS]) + finalCmd.append(elem); + if (auto elem = cmdObj[bypassDocumentValidationCommandOption()]) + finalCmd.append(elem); - BSONObjBuilder timingBuilder; - timingBuilder.append("shardProcessing", t.millis()); + Timer t2; - BSONObjBuilder aggCountsB; - for (const auto& countEntry : countsMap) { - aggCountsB.append(countEntry.first, static_cast<long long>(countEntry.second)); - } + long long reduceCount = 0; + long long outputCount = 0; + BSONObjBuilder postCountsB; - BSONObj aggCounts = aggCountsB.done(); - finalCmd.append("counts", aggCounts); + bool ok = true; + BSONObj singleResult; - if (auto elem = cmdObj[LiteParsedQuery::cmdOptionMaxTimeMS]) finalCmd.append(elem); - if (auto elem = cmdObj[bypassDocumentValidationCommandOption()]) finalCmd.append(elem); + if (!shardedOutput) { + const auto shard = grid.shardRegistry()->getShard(confOut->getPrimaryId()); + LOG(1) << "MR with single shard output, NS=" << finalColLong + << " primary=" << shard->toString(); - Timer t2; + ShardConnection conn(shard->getConnString(), finalColLong); + ok = conn->runCommand(outDB, finalCmd.obj(), singleResult); - long long reduceCount = 0; - long long outputCount = 0; - BSONObjBuilder postCountsB; + BSONObj counts = singleResult.getObjectField("counts"); + postCountsB.append(conn->getServerAddress(), counts); + reduceCount = counts.getIntField("reduce"); + outputCount = counts.getIntField("output"); - bool ok = true; - BSONObj singleResult; + conn.done(); + } else { + LOG(1) << "MR with sharded output, NS=" << finalColLong; - if (!shardedOutput) { - const auto shard = grid.shardRegistry()->getShard(confOut->getPrimaryId()); - LOG(1) << "MR with single shard output, NS=" << finalColLong - << " primary=" << shard->toString(); + // Create the sharded collection if needed + if (!confOut->isSharded(finalColLong)) { + // Enable sharding on db + confOut->enableSharding(); - ShardConnection conn(shard->getConnString(), finalColLong); - ok = conn->runCommand(outDB, finalCmd.obj(), singleResult); + // Shard collection according to split points + vector<BSONObj> sortedSplitPts; - BSONObj counts = singleResult.getObjectField("counts"); - postCountsB.append(conn->getServerAddress(), counts); - reduceCount = counts.getIntField("reduce"); - outputCount = counts.getIntField("output"); + // Points will be properly sorted using the set + for (const auto& splitPt : splitPts) { + sortedSplitPts.push_back(splitPt); + } - conn.done(); + // Pre-split the collection onto all the shards for this database. Note that + // it's not completely safe to pre-split onto non-primary shards using the + // shardcollection method (a conflict may result if multiple map-reduces are + // writing to the same output collection, for instance). + // + // TODO: pre-split mapReduce output in a safer way. + + set<ShardId> outShardIds; + confOut->getAllShardIds(&outShardIds); + + BSONObj sortKey = BSON("_id" << 1); + ShardKeyPattern sortKeyPattern(sortKey); + Status status = grid.catalogManager()->shardCollection( + finalColLong, sortKeyPattern, true, &sortedSplitPts, &outShardIds); + if (!status.isOK()) { + return appendCommandStatus(result, status); + } } - else { - LOG(1) << "MR with sharded output, NS=" << finalColLong; - - // Create the sharded collection if needed - if (!confOut->isSharded(finalColLong)) { - // Enable sharding on db - confOut->enableSharding(); - // Shard collection according to split points - vector<BSONObj> sortedSplitPts; + map<BSONObj, int> chunkSizes; + { + // Take distributed lock to prevent split / migration. + auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock( + finalColLong, + "mr-post-process", + stdx::chrono::milliseconds(-1), // retry indefinitely + stdx::chrono::milliseconds(100)); + + if (!scopedDistLock.isOK()) { + return appendCommandStatus(result, scopedDistLock.getStatus()); + } - // Points will be properly sorted using the set - for (const auto& splitPt : splitPts) { - sortedSplitPts.push_back(splitPt); - } + BSONObj finalCmdObj = finalCmd.obj(); + mrCommandResults.clear(); - // Pre-split the collection onto all the shards for this database. Note that - // it's not completely safe to pre-split onto non-primary shards using the - // shardcollection method (a conflict may result if multiple map-reduces are - // writing to the same output collection, for instance). - // - // TODO: pre-split mapReduce output in a safer way. - - set<ShardId> outShardIds; - confOut->getAllShardIds(&outShardIds); - - BSONObj sortKey = BSON("_id" << 1); - ShardKeyPattern sortKeyPattern(sortKey); - Status status = grid.catalogManager()->shardCollection(finalColLong, - sortKeyPattern, - true, - &sortedSplitPts, - &outShardIds); - if (!status.isOK()) { - return appendCommandStatus(result, status); - } + try { + Strategy::commandOp( + outDB, finalCmdObj, 0, finalColLong, BSONObj(), &mrCommandResults); + ok = true; + } catch (DBException& e) { + e.addContext(str::stream() << "could not run final reduce on all shards for " + << fullns << ", output " << finalColLong); + throw; } - map<BSONObj, int> chunkSizes; - { - // Take distributed lock to prevent split / migration. - auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock( - finalColLong, - "mr-post-process", - stdx::chrono::milliseconds(-1), // retry indefinitely - stdx::chrono::milliseconds(100)); - - if (!scopedDistLock.isOK()) { - return appendCommandStatus(result, scopedDistLock.getStatus()); + for (const auto& mrResult : mrCommandResults) { + string server; + { + const auto shard = grid.shardRegistry()->getShard(mrResult.shardTargetId); + server = shard->getConnString().toString(); } + singleResult = mrResult.result; - BSONObj finalCmdObj = finalCmd.obj(); - mrCommandResults.clear(); - - try { - Strategy::commandOp(outDB, finalCmdObj, 0, finalColLong, BSONObj(), &mrCommandResults); - ok = true; - } - catch (DBException& e){ - e.addContext(str::stream() << "could not run final reduce on all shards for " - << fullns << ", output " << finalColLong); - throw; + ok = singleResult["ok"].trueValue(); + if (!ok) { + break; } - for (const auto& mrResult : mrCommandResults) { - string server; - { - const auto shard = - grid.shardRegistry()->getShard(mrResult.shardTargetId); - server = shard->getConnString().toString(); - } - singleResult = mrResult.result; - - ok = singleResult["ok"].trueValue(); - if (!ok) { - break; - } - - BSONObj counts = singleResult.getObjectField("counts"); - reduceCount += counts.getIntField("reduce"); - outputCount += counts.getIntField("output"); - postCountsB.append(server, counts); - - // get the size inserted for each chunk - // split cannot be called here since we already have the distributed lock - if (singleResult.hasField("chunkSizes")) { - vector<BSONElement> sizes = singleResult.getField("chunkSizes").Array(); - for (unsigned int i = 0; i < sizes.size(); i += 2) { - BSONObj key = sizes[i].Obj().getOwned(); - const long long size = sizes[i + 1].numberLong(); - - invariant(size < std::numeric_limits<int>::max()); - chunkSizes[key] = static_cast<int>(size); - } + BSONObj counts = singleResult.getObjectField("counts"); + reduceCount += counts.getIntField("reduce"); + outputCount += counts.getIntField("output"); + postCountsB.append(server, counts); + + // get the size inserted for each chunk + // split cannot be called here since we already have the distributed lock + if (singleResult.hasField("chunkSizes")) { + vector<BSONElement> sizes = singleResult.getField("chunkSizes").Array(); + for (unsigned int i = 0; i < sizes.size(); i += 2) { + BSONObj key = sizes[i].Obj().getOwned(); + const long long size = sizes[i + 1].numberLong(); + + invariant(size < std::numeric_limits<int>::max()); + chunkSizes[key] = static_cast<int>(size); } } } + } - // Do the splitting round - ChunkManagerPtr cm = confOut->getChunkManagerIfExists(finalColLong); - for (const auto& chunkSize : chunkSizes) { - BSONObj key = chunkSize.first; - const int size = chunkSize.second; - invariant(size < std::numeric_limits<int>::max()); - - // key reported should be the chunk's minimum - ChunkPtr c = cm->findIntersectingChunk(key); - if (!c) { - warning() << "Mongod reported " << size << " bytes inserted for key " - << key << " but can't find chunk"; - } - else { - c->splitIfShould(size); - } + // Do the splitting round + ChunkManagerPtr cm = confOut->getChunkManagerIfExists(finalColLong); + for (const auto& chunkSize : chunkSizes) { + BSONObj key = chunkSize.first; + const int size = chunkSize.second; + invariant(size < std::numeric_limits<int>::max()); + + // key reported should be the chunk's minimum + ChunkPtr c = cm->findIntersectingChunk(key); + if (!c) { + warning() << "Mongod reported " << size << " bytes inserted for key " << key + << " but can't find chunk"; + } else { + c->splitIfShould(size); } } + } - _cleanUp(servers, dbname, shardResultCollection); - - if (!ok) { - errmsg = str::stream() << "MR post processing failed: " << singleResult.toString(); - return 0; - } + _cleanUp(servers, dbname, shardResultCollection); - // copy some elements from a single result - // annoying that we have to copy all results for inline, but no way around it - if (singleResult.hasField("result")) { - result.append(singleResult.getField("result")); - } - else if (singleResult.hasField("results")) { - result.append(singleResult.getField("results")); - } + if (!ok) { + errmsg = str::stream() << "MR post processing failed: " << singleResult.toString(); + return 0; + } - BSONObjBuilder countsB(32); - // input stat is determined by aggregate MR job - countsB.append("input", aggCounts.getField("input").numberLong()); - countsB.append("emit", aggCounts.getField("emit").numberLong()); + // copy some elements from a single result + // annoying that we have to copy all results for inline, but no way around it + if (singleResult.hasField("result")) { + result.append(singleResult.getField("result")); + } else if (singleResult.hasField("results")) { + result.append(singleResult.getField("results")); + } - // reduce count is sum of all reduces that happened - countsB.append("reduce", aggCounts.getField("reduce").numberLong() + reduceCount); + BSONObjBuilder countsB(32); + // input stat is determined by aggregate MR job + countsB.append("input", aggCounts.getField("input").numberLong()); + countsB.append("emit", aggCounts.getField("emit").numberLong()); - // ouput is determined by post processing on each shard - countsB.append("output", outputCount); - result.append("counts", countsB.done()); + // reduce count is sum of all reduces that happened + countsB.append("reduce", aggCounts.getField("reduce").numberLong() + reduceCount); - timingBuilder.append("postProcessing", t2.millis()); + // ouput is determined by post processing on each shard + countsB.append("output", outputCount); + result.append("counts", countsB.done()); - result.append("timeMillis", t.millis()); - result.append("timing", timingBuilder.done()); - result.append("shardCounts", shardCounts); - result.append("postProcessCounts", postCountsB.done()); + timingBuilder.append("postProcessing", t2.millis()); - return true; - } + result.append("timeMillis", t.millis()); + result.append("timing", timingBuilder.done()); + result.append("shardCounts", shardCounts); + result.append("postProcessCounts", postCountsB.done()); - private: + return true; + } - /** - * Drops the temporary results collections from each shard. - */ - void _cleanUp(const set<string>& servers, string dbName, string shardResultCollection) { - try { - // drop collections with tmp results on each shard - for (const auto& server : servers) { - ScopedDbConnection conn(server); - conn->dropCollection(dbName + "." + shardResultCollection); - conn.done(); - } - } - catch (const DBException& e) { - warning() << "Cannot cleanup shard results" << e.toString(); - } - catch (const std::exception& e) { - severe() << "Cannot cleanup shard results" << causedBy(e); +private: + /** + * Drops the temporary results collections from each shard. + */ + void _cleanUp(const set<string>& servers, string dbName, string shardResultCollection) { + try { + // drop collections with tmp results on each shard + for (const auto& server : servers) { + ScopedDbConnection conn(server); + conn->dropCollection(dbName + "." + shardResultCollection); + conn.done(); } + } catch (const DBException& e) { + warning() << "Cannot cleanup shard results" << e.toString(); + } catch (const std::exception& e) { + severe() << "Cannot cleanup shard results" << causedBy(e); } + } - } clusterMapReduceCmd; +} clusterMapReduceCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_merge_chunks_cmd.cpp b/src/mongo/s/commands/cluster_merge_chunks_cmd.cpp index fba40110919..c6173244fbb 100644 --- a/src/mongo/s/commands/cluster_merge_chunks_cmd.cpp +++ b/src/mongo/s/commands/cluster_merge_chunks_cmd.cpp @@ -45,160 +45,168 @@ namespace mongo { - using std::shared_ptr; - using std::string; - using std::stringstream; - using std::vector; +using std::shared_ptr; +using std::string; +using std::stringstream; +using std::vector; namespace { - /** - * Mongos-side command for merging chunks, passes command to appropriate shard. - */ - class ClusterMergeChunksCommand : public Command { - public: - ClusterMergeChunksCommand() : Command("mergeChunks") {} +/** + * Mongos-side command for merging chunks, passes command to appropriate shard. + */ +class ClusterMergeChunksCommand : public Command { +public: + ClusterMergeChunksCommand() : Command("mergeChunks") {} + + virtual void help(stringstream& h) const { + h << "Merge Chunks command\n" + << "usage: { mergeChunks : <ns>, bounds : [ <min key>, <max key> ] }"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::splitChunk)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); + } + return Status::OK(); + } + + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } + + virtual bool adminOnly() const { + return true; + } + virtual bool slaveOk() const { + return false; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + // Required + static BSONField<string> nsField; + static BSONField<vector<BSONObj>> boundsField; + + // Used to send sharding state + static BSONField<string> shardNameField; + static BSONField<string> configField; + + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + vector<BSONObj> bounds; + if (!FieldParser::extract(cmdObj, boundsField, &bounds, &errmsg)) { + return false; + } + + if (bounds.size() == 0) { + errmsg = "no bounds were specified"; + return false; + } - virtual void help(stringstream& h) const { - h << "Merge Chunks command\n" - << "usage: { mergeChunks : <ns>, bounds : [ <min key>, <max key> ] }"; + if (bounds.size() != 2) { + errmsg = "only a min and max bound may be specified"; + return false; } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), - ActionType::splitChunk)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - return Status::OK(); + BSONObj minKey = bounds[0]; + BSONObj maxKey = bounds[1]; + + if (minKey.isEmpty()) { + errmsg = "no min key specified"; + return false; } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); + if (maxKey.isEmpty()) { + errmsg = "no max key specified"; + return false; } - virtual bool adminOnly() const { return true; } - virtual bool slaveOk() const { return false; } - virtual bool isWriteCommandForConfigServer() const { return false; } - - // Required - static BSONField<string> nsField; - static BSONField<vector<BSONObj> > boundsField; - - // Used to send sharding state - static BSONField<string> shardNameField; - static BSONField<string> configField; - - - bool run(OperationContext* txn, const string& dbname, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - - vector<BSONObj> bounds; - if ( !FieldParser::extract( cmdObj, boundsField, &bounds, &errmsg ) ) { - return false; - } - - if ( bounds.size() == 0 ) { - errmsg = "no bounds were specified"; - return false; - } - - if ( bounds.size() != 2 ) { - errmsg = "only a min and max bound may be specified"; - return false; - } - - BSONObj minKey = bounds[0]; - BSONObj maxKey = bounds[1]; - - if ( minKey.isEmpty() ) { - errmsg = "no min key specified"; - return false; - } - - if ( maxKey.isEmpty() ) { - errmsg = "no max key specified"; - return false; - } - - const NamespaceString nss(parseNs(dbname, cmdObj)); - if (nss.size() == 0) { - return appendCommandStatus(result, Status(ErrorCodes::InvalidNamespace, - "no namespace specified")); - } - - auto status = grid.catalogCache()->getDatabase(nss.db().toString()); - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } - - std::shared_ptr<DBConfig> config = status.getValue(); - if (!config->isSharded(nss.ns())) { - return appendCommandStatus(result, Status(ErrorCodes::NamespaceNotSharded, - "ns [" + nss.ns() + " is not sharded.")); - } - - // This refreshes the chunk metadata if stale. - ChunkManagerPtr manager = config->getChunkManagerIfExists(nss, true); - if (!manager) { - return appendCommandStatus(result, Status(ErrorCodes::NamespaceNotSharded, - "ns [" + nss.ns() + " is not sharded.")); - } - - if (!manager->getShardKeyPattern().isShardKey(minKey) - || !manager->getShardKeyPattern().isShardKey(maxKey)) { - errmsg = stream() << "shard key bounds " << "[" << minKey << "," << maxKey << ")" - << " are not valid for shard key pattern " - << manager->getShardKeyPattern().toBSON(); - return false; - } - - minKey = manager->getShardKeyPattern().normalizeShardKey(minKey); - maxKey = manager->getShardKeyPattern().normalizeShardKey(maxKey); - - ChunkPtr firstChunk = manager->findIntersectingChunk(minKey); - verify(firstChunk); - - BSONObjBuilder remoteCmdObjB; - remoteCmdObjB.append( cmdObj[ ClusterMergeChunksCommand::nsField() ] ); - remoteCmdObjB.append( cmdObj[ ClusterMergeChunksCommand::boundsField() ] ); - remoteCmdObjB.append( ClusterMergeChunksCommand::configField(), - grid.catalogManager()->connectionString().toString() ); - remoteCmdObjB.append( ClusterMergeChunksCommand::shardNameField(), - firstChunk->getShardId() ); - - BSONObj remoteResult; - - // Throws, but handled at level above. Don't want to rewrap to preserve exception - // formatting. - const auto shard = grid.shardRegistry()->getShard(firstChunk->getShardId()); - if (!shard) { - return appendCommandStatus(result, - Status(ErrorCodes::ShardNotFound, - str::stream() << "Can't find shard for chunk: " - << firstChunk->toString())); - } - - ScopedDbConnection conn(shard->getConnString()); - bool ok = conn->runCommand( "admin", remoteCmdObjB.obj(), remoteResult ); - conn.done(); - - result.appendElements( remoteResult ); - return ok; + const NamespaceString nss(parseNs(dbname, cmdObj)); + if (nss.size() == 0) { + return appendCommandStatus( + result, Status(ErrorCodes::InvalidNamespace, "no namespace specified")); } - } clusterMergeChunksCommand; + auto status = grid.catalogCache()->getDatabase(nss.db().toString()); + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); + } + + std::shared_ptr<DBConfig> config = status.getValue(); + if (!config->isSharded(nss.ns())) { + return appendCommandStatus( + result, + Status(ErrorCodes::NamespaceNotSharded, "ns [" + nss.ns() + " is not sharded.")); + } + + // This refreshes the chunk metadata if stale. + ChunkManagerPtr manager = config->getChunkManagerIfExists(nss, true); + if (!manager) { + return appendCommandStatus( + result, + Status(ErrorCodes::NamespaceNotSharded, "ns [" + nss.ns() + " is not sharded.")); + } + + if (!manager->getShardKeyPattern().isShardKey(minKey) || + !manager->getShardKeyPattern().isShardKey(maxKey)) { + errmsg = stream() << "shard key bounds " + << "[" << minKey << "," << maxKey << ")" + << " are not valid for shard key pattern " + << manager->getShardKeyPattern().toBSON(); + return false; + } + + minKey = manager->getShardKeyPattern().normalizeShardKey(minKey); + maxKey = manager->getShardKeyPattern().normalizeShardKey(maxKey); + + ChunkPtr firstChunk = manager->findIntersectingChunk(minKey); + verify(firstChunk); + + BSONObjBuilder remoteCmdObjB; + remoteCmdObjB.append(cmdObj[ClusterMergeChunksCommand::nsField()]); + remoteCmdObjB.append(cmdObj[ClusterMergeChunksCommand::boundsField()]); + remoteCmdObjB.append(ClusterMergeChunksCommand::configField(), + grid.catalogManager()->connectionString().toString()); + remoteCmdObjB.append(ClusterMergeChunksCommand::shardNameField(), firstChunk->getShardId()); + + BSONObj remoteResult; + + // Throws, but handled at level above. Don't want to rewrap to preserve exception + // formatting. + const auto shard = grid.shardRegistry()->getShard(firstChunk->getShardId()); + if (!shard) { + return appendCommandStatus( + result, + Status(ErrorCodes::ShardNotFound, + str::stream() << "Can't find shard for chunk: " << firstChunk->toString())); + } + + ScopedDbConnection conn(shard->getConnString()); + bool ok = conn->runCommand("admin", remoteCmdObjB.obj(), remoteResult); + conn.done(); + + result.appendElements(remoteResult); + return ok; + } + +} clusterMergeChunksCommand; - BSONField<string> ClusterMergeChunksCommand::nsField( "mergeChunks" ); - BSONField<vector<BSONObj> > ClusterMergeChunksCommand::boundsField( "bounds" ); +BSONField<string> ClusterMergeChunksCommand::nsField("mergeChunks"); +BSONField<vector<BSONObj>> ClusterMergeChunksCommand::boundsField("bounds"); - BSONField<string> ClusterMergeChunksCommand::configField( "config" ); - BSONField<string> ClusterMergeChunksCommand::shardNameField( "shardName" ); +BSONField<string> ClusterMergeChunksCommand::configField("config"); +BSONField<string> ClusterMergeChunksCommand::shardNameField("shardName"); -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_move_chunk_cmd.cpp b/src/mongo/s/commands/cluster_move_chunk_cmd.cpp index 602f539ec55..45b9c63cdec 100644 --- a/src/mongo/s/commands/cluster_move_chunk_cmd.cpp +++ b/src/mongo/s/commands/cluster_move_chunk_cmd.cpp @@ -50,230 +50,216 @@ namespace mongo { - using std::shared_ptr; - using std::unique_ptr; - using std::string; +using std::shared_ptr; +using std::unique_ptr; +using std::string; namespace { - class MoveChunkCmd : public Command { - public: - MoveChunkCmd() : Command("moveChunk", false, "movechunk") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; +class MoveChunkCmd : public Command { +public: + MoveChunkCmd() : Command("moveChunk", false, "movechunk") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "Example: move chunk that contains the doc {num : 7} to shard001\n" + << " { movechunk : 'test.foo' , find : { num : 7 } , to : 'shard0001' }\n" + << "Example: move chunk with lower bound 0 and upper bound 10 to shard001\n" + << " { movechunk : 'test.foo' , bounds : [ { num : 0 } , { num : 10 } ] " + << " , to : 'shard001' }\n"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::moveChunk)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); } - virtual bool isWriteCommandForConfigServer() const { - return false; - } - - virtual void help(std::stringstream& help) const { - help << "Example: move chunk that contains the doc {num : 7} to shard001\n" - << " { movechunk : 'test.foo' , find : { num : 7 } , to : 'shard0001' }\n" - << "Example: move chunk with lower bound 0 and upper bound 10 to shard001\n" - << " { movechunk : 'test.foo' , bounds : [ { num : 0 } , { num : 10 } ] " - << " , to : 'shard001' }\n"; - } + return Status::OK(); + } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace( - NamespaceString(parseNs(dbname, - cmdObj))), - ActionType::moveChunk)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - - return Status::OK(); - } - - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); - } + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + ShardConnection::sync(); - ShardConnection::sync(); + Timer t; - Timer t; + const NamespaceString nss(parseNs(dbname, cmdObj)); - const NamespaceString nss(parseNs(dbname, cmdObj)); + std::shared_ptr<DBConfig> config; - std::shared_ptr<DBConfig> config; - - { - if (nss.size() == 0) { - return appendCommandStatus(result, Status(ErrorCodes::InvalidNamespace, - "no namespace specified")); - } - - auto status = grid.catalogCache()->getDatabase(nss.db().toString()); - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } - - config = status.getValue(); + { + if (nss.size() == 0) { + return appendCommandStatus( + result, Status(ErrorCodes::InvalidNamespace, "no namespace specified")); } - if (!config->isSharded(nss.ns())) { - config->reload(); - - if (!config->isSharded(nss.ns())) { - return appendCommandStatus(result, - Status(ErrorCodes::NamespaceNotSharded, - "ns [" + nss.ns() + " is not sharded.")); - } + auto status = grid.catalogCache()->getDatabase(nss.db().toString()); + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); } - string toString = cmdObj["to"].valuestrsafe(); - if (!toString.size()) { - errmsg = "you have to specify where you want to move the chunk"; - return false; - } + config = status.getValue(); + } + + if (!config->isSharded(nss.ns())) { + config->reload(); - const auto to = grid.shardRegistry()->getShard(toString); - if (!to) { - string msg(str::stream() << - "Could not move chunk in '" << nss.ns() << - "' to shard '" << toString << - "' because that shard does not exist"); - log() << msg; + if (!config->isSharded(nss.ns())) { return appendCommandStatus(result, - Status(ErrorCodes::ShardNotFound, msg)); + Status(ErrorCodes::NamespaceNotSharded, + "ns [" + nss.ns() + " is not sharded.")); } + } - // so far, chunk size serves test purposes; it may or may not become a supported parameter - long long maxChunkSizeBytes = cmdObj["maxChunkSizeBytes"].numberLong(); - if (maxChunkSizeBytes == 0) { - maxChunkSizeBytes = Chunk::MaxChunkSize; - } + string toString = cmdObj["to"].valuestrsafe(); + if (!toString.size()) { + errmsg = "you have to specify where you want to move the chunk"; + return false; + } - BSONObj find = cmdObj.getObjectField("find"); - BSONObj bounds = cmdObj.getObjectField("bounds"); + const auto to = grid.shardRegistry()->getShard(toString); + if (!to) { + string msg(str::stream() << "Could not move chunk in '" << nss.ns() << "' to shard '" + << toString << "' because that shard does not exist"); + log() << msg; + return appendCommandStatus(result, Status(ErrorCodes::ShardNotFound, msg)); + } - // check that only one of the two chunk specification methods is used - if (find.isEmpty() == bounds.isEmpty()) { - errmsg = "need to specify either a find query, or both lower and upper bounds."; - return false; - } + // so far, chunk size serves test purposes; it may or may not become a supported parameter + long long maxChunkSizeBytes = cmdObj["maxChunkSizeBytes"].numberLong(); + if (maxChunkSizeBytes == 0) { + maxChunkSizeBytes = Chunk::MaxChunkSize; + } - // This refreshes the chunk metadata if stale. - ChunkManagerPtr info = config->getChunkManager(nss.ns(), true); - ChunkPtr chunk; + BSONObj find = cmdObj.getObjectField("find"); + BSONObj bounds = cmdObj.getObjectField("bounds"); - if (!find.isEmpty()) { + // check that only one of the two chunk specification methods is used + if (find.isEmpty() == bounds.isEmpty()) { + errmsg = "need to specify either a find query, or both lower and upper bounds."; + return false; + } - StatusWith<BSONObj> status = - info->getShardKeyPattern().extractShardKeyFromQuery(find); + // This refreshes the chunk metadata if stale. + ChunkManagerPtr info = config->getChunkManager(nss.ns(), true); + ChunkPtr chunk; - // Bad query - if (!status.isOK()) - return appendCommandStatus(result, status.getStatus()); + if (!find.isEmpty()) { + StatusWith<BSONObj> status = info->getShardKeyPattern().extractShardKeyFromQuery(find); - BSONObj shardKey = status.getValue(); + // Bad query + if (!status.isOK()) + return appendCommandStatus(result, status.getStatus()); - if (shardKey.isEmpty()) { - errmsg = str::stream() << "no shard key found in chunk query " << find; - return false; - } + BSONObj shardKey = status.getValue(); - chunk = info->findIntersectingChunk(shardKey); - verify(chunk.get()); - } - else { - - // Bounds - if (!info->getShardKeyPattern().isShardKey(bounds[0].Obj()) - || !info->getShardKeyPattern().isShardKey(bounds[1].Obj())) { - errmsg = str::stream() << "shard key bounds " << "[" << bounds[0].Obj() << "," - << bounds[1].Obj() << ")" - << " are not valid for shard key pattern " - << info->getShardKeyPattern().toBSON(); - return false; - } - - BSONObj minKey = info->getShardKeyPattern().normalizeShardKey(bounds[0].Obj()); - BSONObj maxKey = info->getShardKeyPattern().normalizeShardKey(bounds[1].Obj()); - - chunk = info->findIntersectingChunk(minKey); - verify(chunk.get()); - - if (chunk->getMin().woCompare(minKey) != 0 - || chunk->getMax().woCompare(maxKey) != 0) { - - errmsg = str::stream() << "no chunk found with the shard key bounds " << "[" - << minKey << "," << maxKey << ")"; - return false; - } + if (shardKey.isEmpty()) { + errmsg = str::stream() << "no shard key found in chunk query " << find; + return false; } - { - const auto from = grid.shardRegistry()->getShard(chunk->getShardId()); - if (from->getId() == to->getId()) { - errmsg = "that chunk is already on that shard"; - return false; - } + chunk = info->findIntersectingChunk(shardKey); + verify(chunk.get()); + } else { + // Bounds + if (!info->getShardKeyPattern().isShardKey(bounds[0].Obj()) || + !info->getShardKeyPattern().isShardKey(bounds[1].Obj())) { + errmsg = str::stream() << "shard key bounds " + << "[" << bounds[0].Obj() << "," << bounds[1].Obj() << ")" + << " are not valid for shard key pattern " + << info->getShardKeyPattern().toBSON(); + return false; } - LOG(0) << "CMD: movechunk: " << cmdObj; + BSONObj minKey = info->getShardKeyPattern().normalizeShardKey(bounds[0].Obj()); + BSONObj maxKey = info->getShardKeyPattern().normalizeShardKey(bounds[1].Obj()); - StatusWith<int> maxTimeMS = LiteParsedQuery::parseMaxTimeMSCommand(cmdObj); + chunk = info->findIntersectingChunk(minKey); + verify(chunk.get()); - if (!maxTimeMS.isOK()) { - errmsg = maxTimeMS.getStatus().reason(); + if (chunk->getMin().woCompare(minKey) != 0 || chunk->getMax().woCompare(maxKey) != 0) { + errmsg = str::stream() << "no chunk found with the shard key bounds " + << "[" << minKey << "," << maxKey << ")"; return false; } + } - unique_ptr<WriteConcernOptions> writeConcern(new WriteConcernOptions()); - - Status status = writeConcern->parseSecondaryThrottle(cmdObj, NULL); - if (!status.isOK()){ - if (status.code() != ErrorCodes::WriteConcernNotDefined) { - errmsg = status.toString(); - return false; - } - - // Let the shard decide what write concern to use. - writeConcern.reset(); + { + const auto from = grid.shardRegistry()->getShard(chunk->getShardId()); + if (from->getId() == to->getId()) { + errmsg = "that chunk is already on that shard"; + return false; } + } - BSONObj res; - if (!chunk->moveAndCommit(to->getId(), - maxChunkSizeBytes, - writeConcern.get(), - cmdObj["_waitForDelete"].trueValue(), - maxTimeMS.getValue(), - res)) { + LOG(0) << "CMD: movechunk: " << cmdObj; - errmsg = "move failed"; - result.append("cause", res); + StatusWith<int> maxTimeMS = LiteParsedQuery::parseMaxTimeMSCommand(cmdObj); + + if (!maxTimeMS.isOK()) { + errmsg = maxTimeMS.getStatus().reason(); + return false; + } - if (!res["code"].eoo()) { - result.append(res["code"]); - } + unique_ptr<WriteConcernOptions> writeConcern(new WriteConcernOptions()); + Status status = writeConcern->parseSecondaryThrottle(cmdObj, NULL); + if (!status.isOK()) { + if (status.code() != ErrorCodes::WriteConcernNotDefined) { + errmsg = status.toString(); return false; } - result.append("millis", t.millis()); + // Let the shard decide what write concern to use. + writeConcern.reset(); + } + + BSONObj res; + if (!chunk->moveAndCommit(to->getId(), + maxChunkSizeBytes, + writeConcern.get(), + cmdObj["_waitForDelete"].trueValue(), + maxTimeMS.getValue(), + res)) { + errmsg = "move failed"; + result.append("cause", res); + + if (!res["code"].eoo()) { + result.append(res["code"]); + } - return true; + return false; } - } moveChunk; + result.append("millis", t.millis()); + + return true; + } + +} moveChunk; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_move_primary_cmd.cpp b/src/mongo/s/commands/cluster_move_primary_cmd.cpp index 227830f410d..f9d401a26b6 100644 --- a/src/mongo/s/commands/cluster_move_primary_cmd.cpp +++ b/src/mongo/s/commands/cluster_move_primary_cmd.cpp @@ -51,242 +51,225 @@ namespace mongo { - using std::shared_ptr; - using std::set; - using std::string; +using std::shared_ptr; +using std::set; +using std::string; namespace { - class MoveDatabasePrimaryCommand : public Command { - public: - MoveDatabasePrimaryCommand() : Command("movePrimary", false, "moveprimary") { } +class MoveDatabasePrimaryCommand : public Command { +public: + MoveDatabasePrimaryCommand() : Command("movePrimary", false, "moveprimary") {} - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual bool adminOnly() const { - return true; - } + virtual bool adminOnly() const { + return true; + } - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << " example: { moveprimary : 'foo' , to : 'localhost:9999' }"; + } - virtual void help(std::stringstream& help) const { - help << " example: { moveprimary : 'foo' , to : 'localhost:9999' }"; + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forDatabaseName(parseNs(dbname, cmdObj)), ActionType::moveChunk)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { + return Status::OK(); + } - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forDatabaseName( - parseNs(dbname, cmdObj)), - ActionType::moveChunk)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return cmdObj.firstElement().str(); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname_unused, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + const string dbname = parseNs("", cmdObj); - return Status::OK(); + if (dbname.empty() || !nsIsDbOnly(dbname)) { + errmsg = "invalid db name specified: " + dbname; + return false; } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return cmdObj.firstElement().str(); + if (dbname == "admin" || dbname == "config" || dbname == "local") { + errmsg = "can't move primary for " + dbname + " database"; + return false; } - virtual bool run(OperationContext* txn, - const std::string& dbname_unused, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + // Flush all cached information. This can't be perfect, but it's better than nothing. + grid.catalogCache()->invalidate(dbname); - const string dbname = parseNs("", cmdObj); + auto status = grid.catalogCache()->getDatabase(dbname); + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); + } - if (dbname.empty() || !nsIsDbOnly(dbname)) { - errmsg = "invalid db name specified: " + dbname; - return false; - } + shared_ptr<DBConfig> config = status.getValue(); - if (dbname == "admin" || dbname == "config" || dbname == "local") { - errmsg = "can't move primary for " + dbname + " database"; - return false; - } + const string to = cmdObj["to"].valuestrsafe(); + if (!to.size()) { + errmsg = "you have to specify where you want to move it"; + return false; + } - // Flush all cached information. This can't be perfect, but it's better than nothing. - grid.catalogCache()->invalidate(dbname); + shared_ptr<Shard> toShard = grid.shardRegistry()->getShard(to); + if (!toShard) { + string msg(str::stream() << "Could not move database '" << dbname << "' to shard '" + << to << "' because the shard does not exist"); + log() << msg; + return appendCommandStatus(result, Status(ErrorCodes::ShardNotFound, msg)); + } - auto status = grid.catalogCache()->getDatabase(dbname); - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } + shared_ptr<Shard> fromShard = grid.shardRegistry()->getShard(config->getPrimaryId()); + invariant(fromShard); - shared_ptr<DBConfig> config = status.getValue(); + if (fromShard->getConnString().sameLogicalEndpoint(toShard->getConnString())) { + errmsg = "it is already the primary"; + return false; + } - const string to = cmdObj["to"].valuestrsafe(); - if (!to.size()) { - errmsg = "you have to specify where you want to move it"; - return false; - } + if (!grid.catalogManager()->isShardHost(toShard->getConnString())) { + errmsg = "that server isn't known to me"; + return false; + } - shared_ptr<Shard> toShard = grid.shardRegistry()->getShard(to); - if (!toShard) { - string msg(str::stream() << "Could not move database '" << dbname - << "' to shard '" << to - << "' because the shard does not exist"); - log() << msg; - return appendCommandStatus(result, - Status(ErrorCodes::ShardNotFound, msg)); - } + log() << "Moving " << dbname << " primary from: " << fromShard->toString() + << " to: " << toShard->toString(); - shared_ptr<Shard> fromShard = - grid.shardRegistry()->getShard(config->getPrimaryId()); - invariant(fromShard); + string whyMessage(str::stream() << "Moving primary shard of " << dbname); + auto scopedDistLock = + grid.catalogManager()->getDistLockManager()->lock(dbname + "-movePrimary", whyMessage); - if (fromShard->getConnString().sameLogicalEndpoint(toShard->getConnString())) { - errmsg = "it is already the primary"; - return false; - } + if (!scopedDistLock.isOK()) { + return appendCommandStatus(result, scopedDistLock.getStatus()); + } - if (!grid.catalogManager()->isShardHost(toShard->getConnString())) { - errmsg = "that server isn't known to me"; - return false; - } + set<string> shardedColls; + config->getAllShardedCollections(shardedColls); - log() << "Moving " << dbname << " primary from: " - << fromShard->toString() << " to: " << toShard->toString(); + // Record start in changelog + BSONObj moveStartDetails = + _buildMoveEntry(dbname, fromShard->toString(), toShard->toString(), shardedColls); - string whyMessage(str::stream() << "Moving primary shard of " << dbname); - auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock( - dbname + "-movePrimary", whyMessage); + grid.catalogManager()->logChange(txn, "movePrimary.start", dbname, moveStartDetails); - if (!scopedDistLock.isOK()) { - return appendCommandStatus(result, scopedDistLock.getStatus()); - } + BSONArrayBuilder barr; + barr.append(shardedColls); - set<string> shardedColls; - config->getAllShardedCollections(shardedColls); - - // Record start in changelog - BSONObj moveStartDetails = _buildMoveEntry(dbname, - fromShard->toString(), - toShard->toString(), - shardedColls); - - grid.catalogManager()->logChange(txn, "movePrimary.start", dbname, moveStartDetails); - - BSONArrayBuilder barr; - barr.append(shardedColls); - - ScopedDbConnection toconn(toShard->getConnString()); - - // TODO ERH - we need a clone command which replays operations from clone start to now - // can just use local.oplog.$main - BSONObj cloneRes; - bool worked = toconn->runCommand( - dbname.c_str(), - BSON("clone" << fromShard->getConnString().toString() - << "collsToIgnore" << barr.arr() - << bypassDocumentValidationCommandOption() << true), - cloneRes); - toconn.done(); - - if (!worked) { - log() << "clone failed" << cloneRes; - errmsg = "clone failed"; - return false; - } + ScopedDbConnection toconn(toShard->getConnString()); - const string oldPrimary = fromShard->getConnString().toString(); + // TODO ERH - we need a clone command which replays operations from clone start to now + // can just use local.oplog.$main + BSONObj cloneRes; + bool worked = toconn->runCommand( + dbname.c_str(), + BSON("clone" << fromShard->getConnString().toString() << "collsToIgnore" << barr.arr() + << bypassDocumentValidationCommandOption() << true), + cloneRes); + toconn.done(); - ScopedDbConnection fromconn(fromShard->getConnString()); + if (!worked) { + log() << "clone failed" << cloneRes; + errmsg = "clone failed"; + return false; + } - config->setPrimary(toShard->getConnString().toString()); + const string oldPrimary = fromShard->getConnString().toString(); - if (shardedColls.empty()){ + ScopedDbConnection fromconn(fromShard->getConnString()); - // TODO: Collections can be created in the meantime, and we should handle in the future. - log() << "movePrimary dropping database on " << oldPrimary - << ", no sharded collections in " << dbname; + config->setPrimary(toShard->getConnString().toString()); - try { - fromconn->dropDatabase(dbname.c_str()); - } - catch (DBException& e){ - e.addContext(str::stream() << "movePrimary could not drop the database " - << dbname << " on " << oldPrimary); - throw; - } + if (shardedColls.empty()) { + // TODO: Collections can be created in the meantime, and we should handle in the future. + log() << "movePrimary dropping database on " << oldPrimary + << ", no sharded collections in " << dbname; + try { + fromconn->dropDatabase(dbname.c_str()); + } catch (DBException& e) { + e.addContext(str::stream() << "movePrimary could not drop the database " << dbname + << " on " << oldPrimary); + throw; } - else if (cloneRes["clonedColls"].type() != Array) { - // Legacy behavior from old mongod with sharded collections, *do not* delete - // database, but inform user they can drop manually (or ignore). - warning() << "movePrimary legacy mongod behavior detected. " - << "User must manually remove unsharded collections in database " - << dbname << " on " << oldPrimary; - } - else { - // We moved some unsharded collections, but not all - BSONObjIterator it(cloneRes["clonedColls"].Obj()); - - while (it.more()){ - BSONElement el = it.next(); - if (el.type() == String){ - try { - log() << "movePrimary dropping cloned collection " << el.String() - << " on " << oldPrimary; - fromconn->dropCollection(el.String()); - } - catch (DBException& e){ - e.addContext(str::stream() << "movePrimary could not drop the cloned collection " - << el.String() << " on " << oldPrimary); - throw; - } + } else if (cloneRes["clonedColls"].type() != Array) { + // Legacy behavior from old mongod with sharded collections, *do not* delete + // database, but inform user they can drop manually (or ignore). + warning() << "movePrimary legacy mongod behavior detected. " + << "User must manually remove unsharded collections in database " << dbname + << " on " << oldPrimary; + + } else { + // We moved some unsharded collections, but not all + BSONObjIterator it(cloneRes["clonedColls"].Obj()); + + while (it.more()) { + BSONElement el = it.next(); + if (el.type() == String) { + try { + log() << "movePrimary dropping cloned collection " << el.String() << " on " + << oldPrimary; + fromconn->dropCollection(el.String()); + } catch (DBException& e) { + e.addContext(str::stream() + << "movePrimary could not drop the cloned collection " + << el.String() << " on " << oldPrimary); + throw; } } } - - fromconn.done(); - - result << "primary" << toShard->toString(); - - // Record finish in changelog - BSONObj moveFinishDetails = _buildMoveEntry(dbname, - oldPrimary, - toShard->toString(), - shardedColls); - - grid.catalogManager()->logChange(txn, "movePrimary", dbname, moveFinishDetails); - return true; } - private: - static BSONObj _buildMoveEntry(const string db, - const string from, - const string to, - set<string> shardedColls) { - - BSONObjBuilder details; - details.append("database", db); - details.append("from", from); - details.append("to", to); - - BSONArrayBuilder collB(details.subarrayStart("shardedCollections")); - { - set<string>::iterator it; - for (it = shardedColls.begin(); it != shardedColls.end(); ++it) { - collB.append(*it); - } + fromconn.done(); + + result << "primary" << toShard->toString(); + + // Record finish in changelog + BSONObj moveFinishDetails = + _buildMoveEntry(dbname, oldPrimary, toShard->toString(), shardedColls); + + grid.catalogManager()->logChange(txn, "movePrimary", dbname, moveFinishDetails); + return true; + } + +private: + static BSONObj _buildMoveEntry(const string db, + const string from, + const string to, + set<string> shardedColls) { + BSONObjBuilder details; + details.append("database", db); + details.append("from", from); + details.append("to", to); + + BSONArrayBuilder collB(details.subarrayStart("shardedCollections")); + { + set<string>::iterator it; + for (it = shardedColls.begin(); it != shardedColls.end(); ++it) { + collB.append(*it); } - collB.done(); - - return details.obj(); } + collB.done(); + + return details.obj(); + } - } movePrimary; +} movePrimary; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_netstat_cmd.cpp b/src/mongo/s/commands/cluster_netstat_cmd.cpp index ca1bf2fd9af..df4c158a8b7 100644 --- a/src/mongo/s/commands/cluster_netstat_cmd.cpp +++ b/src/mongo/s/commands/cluster_netstat_cmd.cpp @@ -35,47 +35,46 @@ namespace mongo { namespace { - class NetStatCmd : public Command { - public: - NetStatCmd() : Command("netstat", false, "netstat") { } +class NetStatCmd : public Command { +public: + NetStatCmd() : Command("netstat", false, "netstat") {} - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual bool adminOnly() const { - return true; - } + virtual bool adminOnly() const { + return true; + } - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual void help(std::stringstream& help) const { - help << " shows status/reachability of servers in the cluster"; - } + virtual void help(std::stringstream& help) const { + help << " shows status/reachability of servers in the cluster"; + } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::netstat); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::netstat); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + result.append("configserver", grid.catalogManager()->connectionString().toString()); + result.append("isdbgrid", 1); + return true; + } - result.append("configserver", grid.catalogManager()->connectionString().toString()); - result.append("isdbgrid", 1); - return true; - } +} netstat; - } netstat; - -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_pipeline_cmd.cpp b/src/mongo/s/commands/cluster_pipeline_cmd.cpp index 60f7512d172..1d1fe933410 100644 --- a/src/mongo/s/commands/cluster_pipeline_cmd.cpp +++ b/src/mongo/s/commands/cluster_pipeline_cmd.cpp @@ -53,372 +53,345 @@ namespace mongo { - using boost::intrusive_ptr; - using std::unique_ptr; - using std::shared_ptr; - using std::string; - using std::vector; +using boost::intrusive_ptr; +using std::unique_ptr; +using std::shared_ptr; +using std::string; +using std::vector; namespace { - /** - * Implements the aggregation (pipeline command for sharding). - */ - class PipelineCommand : public Command { - public: - PipelineCommand() : Command(Pipeline::commandName, false) { } +/** + * Implements the aggregation (pipeline command for sharding). + */ +class PipelineCommand : public Command { +public: + PipelineCommand() : Command(Pipeline::commandName, false) {} - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual bool adminOnly() const { - return false; - } + virtual bool adminOnly() const { + return false; + } - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual void help(std::stringstream& help) const { - help << "Runs the sharded aggregation command"; - } + virtual void help(std::stringstream& help) const { + help << "Runs the sharded aggregation command"; + } - // virtuals from Command - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + // virtuals from Command + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + Pipeline::addRequiredPrivileges(this, dbname, cmdObj, out); + } - Pipeline::addRequiredPrivileges(this, dbname, cmdObj, out); + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + const string fullns = parseNs(dbname, cmdObj); + + auto status = grid.catalogCache()->getDatabase(dbname); + if (!status.isOK()) { + return appendEmptyResultSet(result, status.getStatus(), fullns); } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - const string fullns = parseNs(dbname, cmdObj); + shared_ptr<DBConfig> conf = status.getValue(); - auto status = grid.catalogCache()->getDatabase(dbname); - if (!status.isOK()) { - return appendEmptyResultSet(result, status.getStatus(), fullns); - } - - shared_ptr<DBConfig> conf = status.getValue(); - - // If the system isn't running sharded, or the target collection isn't sharded, pass - // this on to a mongod. - if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { - return aggPassthrough(conf, cmdObj, result, options); - } + // If the system isn't running sharded, or the target collection isn't sharded, pass + // this on to a mongod. + if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { + return aggPassthrough(conf, cmdObj, result, options); + } - intrusive_ptr<ExpressionContext> mergeCtx = - new ExpressionContext(txn, NamespaceString(fullns)); - mergeCtx->inRouter = true; - // explicitly *not* setting mergeCtx->tempDir + intrusive_ptr<ExpressionContext> mergeCtx = + new ExpressionContext(txn, NamespaceString(fullns)); + mergeCtx->inRouter = true; + // explicitly *not* setting mergeCtx->tempDir - // Parse the pipeline specification - intrusive_ptr<Pipeline> pipeline(Pipeline::parseCommand(errmsg, cmdObj, mergeCtx)); - if (!pipeline.get()) { - // There was some parsing error - return false; - } + // Parse the pipeline specification + intrusive_ptr<Pipeline> pipeline(Pipeline::parseCommand(errmsg, cmdObj, mergeCtx)); + if (!pipeline.get()) { + // There was some parsing error + return false; + } - // If the first $match stage is an exact match on the shard key, we only have to send it - // to one shard, so send the command to that shard. - BSONObj firstMatchQuery = pipeline->getInitialQuery(); - ChunkManagerPtr chunkMgr = conf->getChunkManager(fullns); - BSONObj shardKeyMatches = uassertStatusOK( - chunkMgr->getShardKeyPattern().extractShardKeyFromQuery(firstMatchQuery)); - - // Don't need to split pipeline if the first $match is an exact match on shard key, but - // we can't send the entire pipeline to one shard if there is a $out stage, since that - // shard may not be the primary shard for the database. - bool needSplit = shardKeyMatches.isEmpty() || pipeline->hasOutStage(); - - // Split the pipeline into pieces for mongod(s) and this mongos. If needSplit is true, - // 'pipeline' will become the merger side. - intrusive_ptr<Pipeline> shardPipeline(needSplit ? pipeline->splitForSharded() - : pipeline); - - // Create the command for the shards. The 'fromRouter' field means produce output to - // be merged. - MutableDocument commandBuilder(shardPipeline->serialize()); - if (needSplit) { - commandBuilder.setField("fromRouter", Value(true)); - commandBuilder.setField("cursor", Value(DOC("batchSize" << 0))); - } - else { - commandBuilder.setField("cursor", Value(cmdObj["cursor"])); - } + // If the first $match stage is an exact match on the shard key, we only have to send it + // to one shard, so send the command to that shard. + BSONObj firstMatchQuery = pipeline->getInitialQuery(); + ChunkManagerPtr chunkMgr = conf->getChunkManager(fullns); + BSONObj shardKeyMatches = uassertStatusOK( + chunkMgr->getShardKeyPattern().extractShardKeyFromQuery(firstMatchQuery)); + + // Don't need to split pipeline if the first $match is an exact match on shard key, but + // we can't send the entire pipeline to one shard if there is a $out stage, since that + // shard may not be the primary shard for the database. + bool needSplit = shardKeyMatches.isEmpty() || pipeline->hasOutStage(); + + // Split the pipeline into pieces for mongod(s) and this mongos. If needSplit is true, + // 'pipeline' will become the merger side. + intrusive_ptr<Pipeline> shardPipeline(needSplit ? pipeline->splitForSharded() : pipeline); + + // Create the command for the shards. The 'fromRouter' field means produce output to + // be merged. + MutableDocument commandBuilder(shardPipeline->serialize()); + if (needSplit) { + commandBuilder.setField("fromRouter", Value(true)); + commandBuilder.setField("cursor", Value(DOC("batchSize" << 0))); + } else { + commandBuilder.setField("cursor", Value(cmdObj["cursor"])); + } - if (cmdObj.hasField("$queryOptions")) { - commandBuilder.setField("$queryOptions", Value(cmdObj["$queryOptions"])); - } + if (cmdObj.hasField("$queryOptions")) { + commandBuilder.setField("$queryOptions", Value(cmdObj["$queryOptions"])); + } - if (cmdObj.hasField(LiteParsedQuery::cmdOptionMaxTimeMS)) { - commandBuilder.setField(LiteParsedQuery::cmdOptionMaxTimeMS, - Value(cmdObj[LiteParsedQuery::cmdOptionMaxTimeMS])); - } + if (cmdObj.hasField(LiteParsedQuery::cmdOptionMaxTimeMS)) { + commandBuilder.setField(LiteParsedQuery::cmdOptionMaxTimeMS, + Value(cmdObj[LiteParsedQuery::cmdOptionMaxTimeMS])); + } - BSONObj shardedCommand = commandBuilder.freeze().toBson(); - BSONObj shardQuery = shardPipeline->getInitialQuery(); - - // Run the command on the shards - // TODO need to make sure cursors are killed if a retry is needed - vector<Strategy::CommandResult> shardResults; - Strategy::commandOp(dbname, - shardedCommand, - options, - fullns, - shardQuery, - &shardResults); - - if (pipeline->isExplain()) { - // This must be checked before we start modifying result. - uassertAllShardsSupportExplain(shardResults); - - if (needSplit) { - result << "splitPipeline" - << DOC("shardsPart" << shardPipeline->writeExplainOps() - << "mergerPart" << pipeline->writeExplainOps()); - } - else { - result << "splitPipeline" << BSONNULL; - } + BSONObj shardedCommand = commandBuilder.freeze().toBson(); + BSONObj shardQuery = shardPipeline->getInitialQuery(); - BSONObjBuilder shardExplains(result.subobjStart("shards")); - for (size_t i = 0; i < shardResults.size(); i++) { - shardExplains.append(shardResults[i].shardTargetId, - BSON("host" << shardResults[i].target.toString() << - "stages" << shardResults[i].result["stages"])); - } + // Run the command on the shards + // TODO need to make sure cursors are killed if a retry is needed + vector<Strategy::CommandResult> shardResults; + Strategy::commandOp(dbname, shardedCommand, options, fullns, shardQuery, &shardResults); - return true; - } + if (pipeline->isExplain()) { + // This must be checked before we start modifying result. + uassertAllShardsSupportExplain(shardResults); - if (!needSplit) { - invariant(shardResults.size() == 1); - const auto& reply = shardResults[0].result; - storePossibleCursor(shardResults[0].target.toString(), reply); - result.appendElements(reply); - return reply["ok"].trueValue(); + if (needSplit) { + result << "splitPipeline" + << DOC("shardsPart" << shardPipeline->writeExplainOps() << "mergerPart" + << pipeline->writeExplainOps()); + } else { + result << "splitPipeline" << BSONNULL; } - DocumentSourceMergeCursors::CursorIds cursorIds = parseCursors(shardResults, fullns); - pipeline->addInitialSource(DocumentSourceMergeCursors::create(cursorIds, mergeCtx)); - - MutableDocument mergeCmd(pipeline->serialize()); - mergeCmd["cursor"] = Value(cmdObj["cursor"]); - - if (cmdObj.hasField("$queryOptions")) { - mergeCmd["$queryOptions"] = Value(cmdObj["$queryOptions"]); + BSONObjBuilder shardExplains(result.subobjStart("shards")); + for (size_t i = 0; i < shardResults.size(); i++) { + shardExplains.append(shardResults[i].shardTargetId, + BSON("host" << shardResults[i].target.toString() << "stages" + << shardResults[i].result["stages"])); } - if (cmdObj.hasField(LiteParsedQuery::cmdOptionMaxTimeMS)) { - mergeCmd[LiteParsedQuery::cmdOptionMaxTimeMS] - = Value(cmdObj[LiteParsedQuery::cmdOptionMaxTimeMS]); - } + return true; + } - string outputNsOrEmpty; - if (DocumentSourceOut* out = dynamic_cast<DocumentSourceOut*>(pipeline->output())) { - outputNsOrEmpty = out->getOutputNs().ns(); - } + if (!needSplit) { + invariant(shardResults.size() == 1); + const auto& reply = shardResults[0].result; + storePossibleCursor(shardResults[0].target.toString(), reply); + result.appendElements(reply); + return reply["ok"].trueValue(); + } - // Run merging command on primary shard of database. Need to use ShardConnection so - // that the merging mongod is sent the config servers on connection init. - const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); - ShardConnection conn(shard->getConnString(), outputNsOrEmpty); - BSONObj mergedResults = aggRunCommand(conn.get(), - dbname, - mergeCmd.freeze().toBson(), - options); - conn.done(); + DocumentSourceMergeCursors::CursorIds cursorIds = parseCursors(shardResults, fullns); + pipeline->addInitialSource(DocumentSourceMergeCursors::create(cursorIds, mergeCtx)); - // Copy output from merging (primary) shard to the output object from our command. - // Also, propagates errmsg and code if ok == false. - result.appendElements(mergedResults); + MutableDocument mergeCmd(pipeline->serialize()); + mergeCmd["cursor"] = Value(cmdObj["cursor"]); - return mergedResults["ok"].trueValue(); + if (cmdObj.hasField("$queryOptions")) { + mergeCmd["$queryOptions"] = Value(cmdObj["$queryOptions"]); } - private: - DocumentSourceMergeCursors::CursorIds parseCursors( - const vector<Strategy::CommandResult>& shardResults, - const string& fullns); - - void killAllCursors(const vector<Strategy::CommandResult>& shardResults); - void uassertAllShardsSupportExplain(const vector<Strategy::CommandResult>& shardResults); - - // These are temporary hacks because the runCommand method doesn't report the exact - // host the command was run on which is necessary for cursor support. The exact host - // could be different from conn->getServerAddress() for connections that map to - // multiple servers such as for replica sets. These also take care of registering - // returned cursors with mongos's cursorCache. - BSONObj aggRunCommand(DBClientBase* conn, - const string& db, - BSONObj cmd, - int queryOptions); - - bool aggPassthrough(DBConfigPtr conf, - BSONObj cmd, - BSONObjBuilder& result, - int queryOptions); - } clusterPipelineCmd; - - DocumentSourceMergeCursors::CursorIds PipelineCommand::parseCursors( - const vector<Strategy::CommandResult>& shardResults, - const string& fullns) { - try { - DocumentSourceMergeCursors::CursorIds cursors; - - for (size_t i = 0; i < shardResults.size(); i++) { - BSONObj result = shardResults[i].result; - - if (!result["ok"].trueValue()) { - // If the failure of the sharded command can be accounted to a single error, - // throw a UserException with that error code; otherwise, throw with a - // location uassert code. - int errCode = getUniqueCodeFromCommandResults(shardResults); - if (errCode == 0) { - errCode = 17022; - } - - invariant(errCode == result["code"].numberInt() || errCode == 17022); - uasserted(errCode, str::stream() - << "sharded pipeline failed on shard " - << shardResults[i].shardTargetId << ": " - << result.toString()); - } + if (cmdObj.hasField(LiteParsedQuery::cmdOptionMaxTimeMS)) { + mergeCmd[LiteParsedQuery::cmdOptionMaxTimeMS] = + Value(cmdObj[LiteParsedQuery::cmdOptionMaxTimeMS]); + } - BSONObj cursor = result["cursor"].Obj(); + string outputNsOrEmpty; + if (DocumentSourceOut* out = dynamic_cast<DocumentSourceOut*>(pipeline->output())) { + outputNsOrEmpty = out->getOutputNs().ns(); + } - massert(17023, - str::stream() << "shard " << shardResults[i].shardTargetId - << " returned non-empty first batch", - cursor["firstBatch"].Obj().isEmpty()); + // Run merging command on primary shard of database. Need to use ShardConnection so + // that the merging mongod is sent the config servers on connection init. + const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); + ShardConnection conn(shard->getConnString(), outputNsOrEmpty); + BSONObj mergedResults = + aggRunCommand(conn.get(), dbname, mergeCmd.freeze().toBson(), options); + conn.done(); - massert(17024, - str::stream() << "shard " << shardResults[i].shardTargetId - << " returned cursorId 0", - cursor["id"].Long() != 0); + // Copy output from merging (primary) shard to the output object from our command. + // Also, propagates errmsg and code if ok == false. + result.appendElements(mergedResults); - massert(17025, - str::stream() << "shard " << shardResults[i].shardTargetId - << " returned different ns: " << cursor["ns"], - cursor["ns"].String() == fullns); + return mergedResults["ok"].trueValue(); + } - cursors.push_back(std::make_pair(shardResults[i].target, cursor["id"].Long())); - } +private: + DocumentSourceMergeCursors::CursorIds parseCursors( + const vector<Strategy::CommandResult>& shardResults, const string& fullns); - return cursors; - } - catch (...) { - // Need to clean up any cursors we successfully created on the shards - killAllCursors(shardResults); - throw; - } - } + void killAllCursors(const vector<Strategy::CommandResult>& shardResults); + void uassertAllShardsSupportExplain(const vector<Strategy::CommandResult>& shardResults); - void PipelineCommand::uassertAllShardsSupportExplain( - const vector<Strategy::CommandResult>& shardResults) { + // These are temporary hacks because the runCommand method doesn't report the exact + // host the command was run on which is necessary for cursor support. The exact host + // could be different from conn->getServerAddress() for connections that map to + // multiple servers such as for replica sets. These also take care of registering + // returned cursors with mongos's cursorCache. + BSONObj aggRunCommand(DBClientBase* conn, const string& db, BSONObj cmd, int queryOptions); - for (size_t i = 0; i < shardResults.size(); i++) { - uassert(17403, - str::stream() << "Shard " << shardResults[i].target.toString() - << " failed: " << shardResults[i].result, - shardResults[i].result["ok"].trueValue()); - - uassert(17404, - str::stream() << "Shard " << shardResults[i].target.toString() - << " does not support $explain", - shardResults[i].result.hasField("stages")); - } - } + bool aggPassthrough(DBConfigPtr conf, BSONObj cmd, BSONObjBuilder& result, int queryOptions); +} clusterPipelineCmd; - void PipelineCommand::killAllCursors(const vector<Strategy::CommandResult>& shardResults) { - // This function must ignore and log all errors. Callers expect a best-effort attempt at - // cleanup without exceptions. If any cursors aren't cleaned up here, they will be cleaned - // up automatically on the shard after 10 minutes anyway. +DocumentSourceMergeCursors::CursorIds PipelineCommand::parseCursors( + const vector<Strategy::CommandResult>& shardResults, const string& fullns) { + try { + DocumentSourceMergeCursors::CursorIds cursors; for (size_t i = 0; i < shardResults.size(); i++) { - try { - BSONObj result = shardResults[i].result; - if (!result["ok"].trueValue()) { - continue; - } - - const long long cursor = result["cursor"]["id"].Long(); - if (!cursor) { - continue; + BSONObj result = shardResults[i].result; + + if (!result["ok"].trueValue()) { + // If the failure of the sharded command can be accounted to a single error, + // throw a UserException with that error code; otherwise, throw with a + // location uassert code. + int errCode = getUniqueCodeFromCommandResults(shardResults); + if (errCode == 0) { + errCode = 17022; } - ScopedDbConnection conn(shardResults[i].target); - conn->killCursor(cursor); - conn.done(); - } - catch (const DBException& e) { - log() << "Couldn't kill aggregation cursor on shard: " << shardResults[i].target - << " due to DBException: " << e.toString(); - } - catch (const std::exception& e) { - log() << "Couldn't kill aggregation cursor on shard: " << shardResults[i].target - << " due to std::exception: " << e.what(); + invariant(errCode == result["code"].numberInt() || errCode == 17022); + uasserted(errCode, + str::stream() << "sharded pipeline failed on shard " + << shardResults[i].shardTargetId << ": " + << result.toString()); } - catch (...) { - log() << "Couldn't kill aggregation cursor on shard: " << shardResults[i].target - << " due to non-exception"; - } - } - } - BSONObj PipelineCommand::aggRunCommand(DBClientBase* conn, - const string& db, - BSONObj cmd, - int queryOptions) { + BSONObj cursor = result["cursor"].Obj(); - // Temporary hack. See comment on declaration for details. + massert(17023, + str::stream() << "shard " << shardResults[i].shardTargetId + << " returned non-empty first batch", + cursor["firstBatch"].Obj().isEmpty()); - massert(17016, - "should only be running an aggregate command here", - str::equals(cmd.firstElementFieldName(), "aggregate")); + massert(17024, + str::stream() << "shard " << shardResults[i].shardTargetId + << " returned cursorId 0", + cursor["id"].Long() != 0); - auto cursor = conn->query(db + ".$cmd", - cmd, - -1, // nToReturn - 0, // nToSkip - NULL, // fieldsToReturn - queryOptions); - massert(17014, - str::stream() << "aggregate command didn't return results on host: " - << conn->toString(), - cursor && cursor->more()); + massert(17025, + str::stream() << "shard " << shardResults[i].shardTargetId + << " returned different ns: " << cursor["ns"], + cursor["ns"].String() == fullns); - BSONObj result = cursor->nextSafe().getOwned(); - - if (ErrorCodes::SendStaleConfig == getStatusFromCommandResult(result)) { - throw RecvStaleConfigException("command failed because of stale config", result); + cursors.push_back(std::make_pair(shardResults[i].target, cursor["id"].Long())); } - uassertStatusOK(storePossibleCursor(cursor->originalHost(), result)); - return result; + return cursors; + } catch (...) { + // Need to clean up any cursors we successfully created on the shards + killAllCursors(shardResults); + throw; + } +} + +void PipelineCommand::uassertAllShardsSupportExplain( + const vector<Strategy::CommandResult>& shardResults) { + for (size_t i = 0; i < shardResults.size(); i++) { + uassert(17403, + str::stream() << "Shard " << shardResults[i].target.toString() + << " failed: " << shardResults[i].result, + shardResults[i].result["ok"].trueValue()); + + uassert(17404, + str::stream() << "Shard " << shardResults[i].target.toString() + << " does not support $explain", + shardResults[i].result.hasField("stages")); } +} - bool PipelineCommand::aggPassthrough(DBConfigPtr conf, - BSONObj cmd, - BSONObjBuilder& out, - int queryOptions) { +void PipelineCommand::killAllCursors(const vector<Strategy::CommandResult>& shardResults) { + // This function must ignore and log all errors. Callers expect a best-effort attempt at + // cleanup without exceptions. If any cursors aren't cleaned up here, they will be cleaned + // up automatically on the shard after 10 minutes anyway. - // Temporary hack. See comment on declaration for details. - const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); - ShardConnection conn(shard->getConnString(), ""); - BSONObj result = aggRunCommand(conn.get(), conf->name(), cmd, queryOptions); - conn.done(); - out.appendElements(result); - return result["ok"].trueValue(); + for (size_t i = 0; i < shardResults.size(); i++) { + try { + BSONObj result = shardResults[i].result; + if (!result["ok"].trueValue()) { + continue; + } + + const long long cursor = result["cursor"]["id"].Long(); + if (!cursor) { + continue; + } + + ScopedDbConnection conn(shardResults[i].target); + conn->killCursor(cursor); + conn.done(); + } catch (const DBException& e) { + log() << "Couldn't kill aggregation cursor on shard: " << shardResults[i].target + << " due to DBException: " << e.toString(); + } catch (const std::exception& e) { + log() << "Couldn't kill aggregation cursor on shard: " << shardResults[i].target + << " due to std::exception: " << e.what(); + } catch (...) { + log() << "Couldn't kill aggregation cursor on shard: " << shardResults[i].target + << " due to non-exception"; + } + } +} + +BSONObj PipelineCommand::aggRunCommand(DBClientBase* conn, + const string& db, + BSONObj cmd, + int queryOptions) { + // Temporary hack. See comment on declaration for details. + + massert(17016, + "should only be running an aggregate command here", + str::equals(cmd.firstElementFieldName(), "aggregate")); + + auto cursor = conn->query(db + ".$cmd", + cmd, + -1, // nToReturn + 0, // nToSkip + NULL, // fieldsToReturn + queryOptions); + massert( + 17014, + str::stream() << "aggregate command didn't return results on host: " << conn->toString(), + cursor && cursor->more()); + + BSONObj result = cursor->nextSafe().getOwned(); + + if (ErrorCodes::SendStaleConfig == getStatusFromCommandResult(result)) { + throw RecvStaleConfigException("command failed because of stale config", result); } -} // namespace -} // namespace mongo + uassertStatusOK(storePossibleCursor(cursor->originalHost(), result)); + return result; +} + +bool PipelineCommand::aggPassthrough(DBConfigPtr conf, + BSONObj cmd, + BSONObjBuilder& out, + int queryOptions) { + // Temporary hack. See comment on declaration for details. + const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); + ShardConnection conn(shard->getConnString(), ""); + BSONObj result = aggRunCommand(conn.get(), conf->name(), cmd, queryOptions); + conn.done(); + out.appendElements(result); + return result["ok"].trueValue(); +} + +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_plan_cache_cmd.cpp b/src/mongo/s/commands/cluster_plan_cache_cmd.cpp index bb66ce55998..1411f5bc3b1 100644 --- a/src/mongo/s/commands/cluster_plan_cache_cmd.cpp +++ b/src/mongo/s/commands/cluster_plan_cache_cmd.cpp @@ -38,141 +38,137 @@ namespace mongo { - using std::string; - using std::stringstream; - using std::vector; +using std::string; +using std::stringstream; +using std::vector; - /** - * Base class for mongos plan cache commands. - * Cluster plan cache commands don't do much more than - * forwarding the commands to all shards and combining the results. - */ - class ClusterPlanCacheCmd : public Command { +/** + * Base class for mongos plan cache commands. + * Cluster plan cache commands don't do much more than + * forwarding the commands to all shards and combining the results. + */ +class ClusterPlanCacheCmd : public Command { MONGO_DISALLOW_COPYING(ClusterPlanCacheCmd); - public: - virtual ~ClusterPlanCacheCmd() { - } +public: + virtual ~ClusterPlanCacheCmd() {} - bool slaveOk() const { - return false; - } - - bool slaveOverrideOk() const { - return true; - } + bool slaveOk() const { + return false; + } - virtual bool isWriteCommandForConfigServer() const { return false; } + bool slaveOverrideOk() const { + return true; + } - void help(stringstream& ss) const { - ss << _helpText; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - Status checkAuthForCommand( ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj ) { - AuthorizationSession* authzSession = AuthorizationSession::get(client); - ResourcePattern pattern = parseResourcePattern(dbname, cmdObj); + void help(stringstream& ss) const { + ss << _helpText; + } - if (authzSession->isAuthorizedForActionsOnResource(pattern, _actionType)) { - return Status::OK(); - } + Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + AuthorizationSession* authzSession = AuthorizationSession::get(client); + ResourcePattern pattern = parseResourcePattern(dbname, cmdObj); - return Status(ErrorCodes::Unauthorized, "unauthorized"); + if (authzSession->isAuthorizedForActionsOnResource(pattern, _actionType)) { + return Status::OK(); } - // Cluster plan cache command entry point. - bool run(OperationContext* txn, const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result); - - public: - - /** - * Instantiates a command that can be invoked by "name", which will be described by - * "helpText", and will require privilege "actionType" to run. - */ - ClusterPlanCacheCmd( const std::string& name, const std::string& helpText, - ActionType actionType ) : - Command( name ), _helpText( helpText ), _actionType( actionType ) { - } + return Status(ErrorCodes::Unauthorized, "unauthorized"); + } + + // Cluster plan cache command entry point. + bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result); - private: - - std::string _helpText; - ActionType _actionType; - }; - - // - // Cluster plan cache command implementation(s) below - // - - bool ClusterPlanCacheCmd::run(OperationContext* txn, const std::string& dbName, - BSONObj& cmdObj, - int options, - std::string& errMsg, - BSONObjBuilder& result) { - const std::string fullns = parseNs(dbName, cmdObj); - NamespaceString nss(fullns); - - // Dispatch command to all the shards. - // Targeted shard commands are generally data-dependent but plan cache - // commands are tied to query shape (data has no effect on query shape). - vector<Strategy::CommandResult> results; - Strategy::commandOp(dbName, cmdObj, options, nss.ns(), BSONObj(), &results); - - // Set value of first shard result's "ok" field. - bool clusterCmdResult = true; - - for (vector<Strategy::CommandResult>::const_iterator i = results.begin(); - i != results.end(); ++i) { - const Strategy::CommandResult& cmdResult = *i; - - // XXX: In absence of sensible aggregation strategy, - // promote first shard's result to top level. - if (i == results.begin()) { - result.appendElements(cmdResult.result); - clusterCmdResult = cmdResult.result["ok"].trueValue(); - } - - // Append shard result as a sub object. - // Name the field after the shard. - string shardName = cmdResult.shardTargetId; - result.append(shardName, cmdResult.result); +public: + /** + * Instantiates a command that can be invoked by "name", which will be described by + * "helpText", and will require privilege "actionType" to run. + */ + ClusterPlanCacheCmd(const std::string& name, const std::string& helpText, ActionType actionType) + : Command(name), _helpText(helpText), _actionType(actionType) {} + +private: + std::string _helpText; + ActionType _actionType; +}; + +// +// Cluster plan cache command implementation(s) below +// + +bool ClusterPlanCacheCmd::run(OperationContext* txn, + const std::string& dbName, + BSONObj& cmdObj, + int options, + std::string& errMsg, + BSONObjBuilder& result) { + const std::string fullns = parseNs(dbName, cmdObj); + NamespaceString nss(fullns); + + // Dispatch command to all the shards. + // Targeted shard commands are generally data-dependent but plan cache + // commands are tied to query shape (data has no effect on query shape). + vector<Strategy::CommandResult> results; + Strategy::commandOp(dbName, cmdObj, options, nss.ns(), BSONObj(), &results); + + // Set value of first shard result's "ok" field. + bool clusterCmdResult = true; + + for (vector<Strategy::CommandResult>::const_iterator i = results.begin(); i != results.end(); + ++i) { + const Strategy::CommandResult& cmdResult = *i; + + // XXX: In absence of sensible aggregation strategy, + // promote first shard's result to top level. + if (i == results.begin()) { + result.appendElements(cmdResult.result); + clusterCmdResult = cmdResult.result["ok"].trueValue(); } - return clusterCmdResult; + // Append shard result as a sub object. + // Name the field after the shard. + string shardName = cmdResult.shardTargetId; + result.append(shardName, cmdResult.result); } - // - // Register plan cache commands at startup - // + return clusterCmdResult; +} - namespace { +// +// Register plan cache commands at startup +// - MONGO_INITIALIZER(RegisterPlanCacheCommands)(InitializerContext* context) { - // Leaked intentionally: a Command registers itself when constructed. +namespace { - new ClusterPlanCacheCmd( - "planCacheListQueryShapes", - "Displays all query shapes in a collection.", - ActionType::planCacheRead ); +MONGO_INITIALIZER(RegisterPlanCacheCommands)(InitializerContext* context) { + // Leaked intentionally: a Command registers itself when constructed. - new ClusterPlanCacheCmd( - "planCacheClear", - "Drops one or all cached queries in a collection.", - ActionType::planCacheWrite ); + new ClusterPlanCacheCmd("planCacheListQueryShapes", + "Displays all query shapes in a collection.", + ActionType::planCacheRead); - new ClusterPlanCacheCmd( - "planCacheListPlans", - "Displays the cached plans for a query shape.", - ActionType::planCacheRead ); + new ClusterPlanCacheCmd("planCacheClear", + "Drops one or all cached queries in a collection.", + ActionType::planCacheWrite); - return Status::OK(); - } + new ClusterPlanCacheCmd("planCacheListPlans", + "Displays the cached plans for a query shape.", + ActionType::planCacheRead); + + return Status::OK(); +} - } // namespace +} // namespace -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_profile_cmd.cpp b/src/mongo/s/commands/cluster_profile_cmd.cpp index e4084def78b..ca619d80fbe 100644 --- a/src/mongo/s/commands/cluster_profile_cmd.cpp +++ b/src/mongo/s/commands/cluster_profile_cmd.cpp @@ -33,43 +33,41 @@ namespace mongo { namespace { - class ProfileCmd : public Command { - public: - ProfileCmd() : Command("profile", false) { } +class ProfileCmd : public Command { +public: + ProfileCmd() : Command("profile", false) {} - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual bool adminOnly() const { - return false; - } + virtual bool adminOnly() const { + return false; + } - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::enableProfiler); + out->push_back(Privilege(ResourcePattern::forDatabaseName(dbname), actions)); + } - ActionSet actions; - actions.addAction(ActionType::enableProfiler); - out->push_back(Privilege(ResourcePattern::forDatabaseName(dbname), actions)); - } + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + errmsg = "profile currently not supported via mongos"; + return false; + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { +} profileCmd; - errmsg = "profile currently not supported via mongos"; - return false; - } - - } profileCmd; - -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_remove_shard_cmd.cpp b/src/mongo/s/commands/cluster_remove_shard_cmd.cpp index e936e926d3b..14aeaf6bf42 100644 --- a/src/mongo/s/commands/cluster_remove_shard_cmd.cpp +++ b/src/mongo/s/commands/cluster_remove_shard_cmd.cpp @@ -45,90 +45,84 @@ namespace mongo { - using std::string; - using std::vector; +using std::string; +using std::vector; namespace { - class RemoveShardCmd : public Command { - public: - RemoveShardCmd() : Command("removeShard", false, "removeshard") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; - } - - virtual bool isWriteCommandForConfigServer() const { - return false; - } - - virtual void help(std::stringstream& help) const { - help << "remove a shard from the system."; +class RemoveShardCmd : public Command { +public: + RemoveShardCmd() : Command("removeShard", false, "removeshard") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "remove a shard from the system."; + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::removeShard); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + const string target = cmdObj.firstElement().valuestrsafe(); + + const auto s = grid.shardRegistry()->getShard(target); + if (!s) { + string msg(str::stream() << "Could not drop shard '" << target + << "' because it does not exist"); + log() << msg; + return appendCommandStatus(result, Status(ErrorCodes::ShardNotFound, msg)); } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - - ActionSet actions; - actions.addAction(ActionType::removeShard); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + StatusWith<ShardDrainingStatus> removeShardResult = + grid.catalogManager()->removeShard(txn, s->getId()); + if (!removeShardResult.isOK()) { + return appendCommandStatus(result, removeShardResult.getStatus()); } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - const string target = cmdObj.firstElement().valuestrsafe(); - - const auto s = grid.shardRegistry()->getShard(target); - if (!s) { - string msg(str::stream() << - "Could not drop shard '" << target << - "' because it does not exist"); - log() << msg; - return appendCommandStatus(result, - Status(ErrorCodes::ShardNotFound, msg)); - } - - StatusWith<ShardDrainingStatus> removeShardResult = - grid.catalogManager()->removeShard(txn, s->getId()); - if (!removeShardResult.isOK()) { - return appendCommandStatus(result, removeShardResult.getStatus()); - } - - vector<string> databases; - grid.catalogManager()->getDatabasesForShard(s->getId(), &databases); - - // Get BSONObj containing: - // 1) note about moving or dropping databases in a shard - // 2) list of databases (excluding 'local' database) that need to be moved - BSONObj dbInfo; - { - BSONObjBuilder dbInfoBuilder; - dbInfoBuilder.append("note", - "you need to drop or movePrimary these databases"); - BSONArrayBuilder dbs(dbInfoBuilder.subarrayStart("dbsToMove")); - for (vector<string>::const_iterator it = databases.begin(); - it != databases.end(); - it++) { - if (*it != "local") { - dbs.append(*it); - } + vector<string> databases; + grid.catalogManager()->getDatabasesForShard(s->getId(), &databases); + + // Get BSONObj containing: + // 1) note about moving or dropping databases in a shard + // 2) list of databases (excluding 'local' database) that need to be moved + BSONObj dbInfo; + { + BSONObjBuilder dbInfoBuilder; + dbInfoBuilder.append("note", "you need to drop or movePrimary these databases"); + BSONArrayBuilder dbs(dbInfoBuilder.subarrayStart("dbsToMove")); + for (vector<string>::const_iterator it = databases.begin(); it != databases.end(); + it++) { + if (*it != "local") { + dbs.append(*it); } - dbs.doneFast(); - dbInfo = dbInfoBuilder.obj(); } + dbs.doneFast(); + dbInfo = dbInfoBuilder.obj(); + } - // TODO: Standardize/Seperate how we append to the result object - switch (removeShardResult.getValue()) { + // TODO: Standardize/Seperate how we append to the result object + switch (removeShardResult.getValue()) { case ShardDrainingStatus::STARTED: result.append("msg", "draining started successfully"); result.append("state", "started"); @@ -137,10 +131,10 @@ namespace { break; case ShardDrainingStatus::ONGOING: { vector<ChunkType> chunks; - Status status = grid.catalogManager()->getChunks( - Query(BSON(ChunkType::shard(s->getId()))), - 0, // return all - &chunks); + Status status = + grid.catalogManager()->getChunks(Query(BSON(ChunkType::shard(s->getId()))), + 0, // return all + &chunks); if (!status.isOK()) { return appendCommandStatus(result, status); } @@ -161,12 +155,12 @@ namespace { result.append("msg", "removeshard completed successfully"); result.append("state", "completed"); result.append("shard", s->getId()); - } - - return true; } - } removeShardCmd; + return true; + } + +} removeShardCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_repair_database_cmd.cpp b/src/mongo/s/commands/cluster_repair_database_cmd.cpp index 33c849ecde6..5383b5688bf 100644 --- a/src/mongo/s/commands/cluster_repair_database_cmd.cpp +++ b/src/mongo/s/commands/cluster_repair_database_cmd.cpp @@ -33,19 +33,19 @@ namespace mongo { namespace { - class ClusterRepairDatabaseCmd : public RunOnAllShardsCommand { - public: - ClusterRepairDatabaseCmd() : RunOnAllShardsCommand("repairDatabase") {} +class ClusterRepairDatabaseCmd : public RunOnAllShardsCommand { +public: + ClusterRepairDatabaseCmd() : RunOnAllShardsCommand("repairDatabase") {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::repairDatabase); - out->push_back(Privilege(ResourcePattern::forDatabaseName(dbname), actions)); - } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::repairDatabase); + out->push_back(Privilege(ResourcePattern::forDatabaseName(dbname), actions)); + } - } clusterRepairDatabaseCmd; +} clusterRepairDatabaseCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_repl_set_get_status_cmd.cpp b/src/mongo/s/commands/cluster_repl_set_get_status_cmd.cpp index 0164fe3cc15..bd8c0332bb1 100644 --- a/src/mongo/s/commands/cluster_repl_set_get_status_cmd.cpp +++ b/src/mongo/s/commands/cluster_repl_set_get_status_cmd.cpp @@ -36,53 +36,51 @@ namespace mongo { namespace { - class CmdReplSetGetStatus : public Command { - public: - CmdReplSetGetStatus() : Command("replSetGetStatus") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; - } - - virtual bool isWriteCommandForConfigServer() const { - return false; - } - - virtual void help(std::stringstream& help) const { - help << "Not supported through mongos"; +class CmdReplSetGetStatus : public Command { +public: + CmdReplSetGetStatus() : Command("replSetGetStatus") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "Not supported through mongos"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + // Require no auth since this command isn't supported in mongos + return Status::OK(); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + if (cmdObj["forShell"].trueValue()) { + LastError::get(cc()).disable(); + ClusterLastErrorInfo::get(cc()).disableForCommand(); } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { + errmsg = "replSetGetStatus is not supported through mongos"; + result.append("info", "mongos"); - // Require no auth since this command isn't supported in mongos - return Status::OK(); - } - - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - if (cmdObj["forShell"].trueValue()) { - LastError::get(cc()).disable(); - ClusterLastErrorInfo::get(cc()).disableForCommand(); - } - - errmsg = "replSetGetStatus is not supported through mongos"; - result.append("info", "mongos"); - - return false; - } + return false; + } - } cmdReplSetGetStatus; +} cmdReplSetGetStatus; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_reset_error_cmd.cpp b/src/mongo/s/commands/cluster_reset_error_cmd.cpp index efa56f524b3..e82b43db1ec 100644 --- a/src/mongo/s/commands/cluster_reset_error_cmd.cpp +++ b/src/mongo/s/commands/cluster_reset_error_cmd.cpp @@ -40,56 +40,50 @@ namespace mongo { namespace { - class CmdShardingResetError : public Command { - public: - CmdShardingResetError() : Command("resetError", false, "reseterror") { } +class CmdShardingResetError : public Command { +public: + CmdShardingResetError() : Command("resetError", false, "reseterror") {} - virtual bool isWriteCommandForConfigServer() const { - return false; - } - - virtual bool slaveOk() const { - return true; - } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + virtual bool isWriteCommandForConfigServer() const { + return false; + } - // No auth required - } - - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + virtual bool slaveOk() const { + return true; + } - LastError::get(cc()).reset(); + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + // No auth required + } - const std::set<std::string>* shards = - ClusterLastErrorInfo::get(cc()).getPrevShardHosts(); + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + LastError::get(cc()).reset(); - for (std::set<std::string>::const_iterator i = shards->begin(); - i != shards->end(); - i++) { + const std::set<std::string>* shards = ClusterLastErrorInfo::get(cc()).getPrevShardHosts(); - const std::string shardName = *i; + for (std::set<std::string>::const_iterator i = shards->begin(); i != shards->end(); i++) { + const std::string shardName = *i; - ShardConnection conn(ConnectionString(shardName, ConnectionString::SET), ""); + ShardConnection conn(ConnectionString(shardName, ConnectionString::SET), ""); - BSONObj res; + BSONObj res; - // Don't care about result from shards. - conn->runCommand(dbname, cmdObj, res); - conn.done(); - } - - return true; + // Don't care about result from shards. + conn->runCommand(dbname, cmdObj, res); + conn.done(); } - } cmdShardingResetError; + return true; + } + +} cmdShardingResetError; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_shard_collection_cmd.cpp b/src/mongo/s/commands/cluster_shard_collection_cmd.cpp index ccc0edd67f8..6e7c6f2cb55 100644 --- a/src/mongo/s/commands/cluster_shard_collection_cmd.cpp +++ b/src/mongo/s/commands/cluster_shard_collection_cmd.cpp @@ -55,464 +55,436 @@ namespace mongo { - using std::shared_ptr; - using std::list; - using std::set; - using std::string; - using std::vector; +using std::shared_ptr; +using std::list; +using std::set; +using std::string; +using std::vector; namespace { - class ShardCollectionCmd : public Command { - public: - ShardCollectionCmd() : Command("shardCollection", false, "shardcollection") { } - - virtual bool slaveOk() const { - return true; - } - - virtual bool adminOnly() const { - return true; +class ShardCollectionCmd : public Command { +public: + ShardCollectionCmd() : Command("shardCollection", false, "shardcollection") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << "Shard a collection. Requires key. Optional unique." + << " Sharding must already be enabled for the database.\n" + << " { enablesharding : \"<dbname>\" }\n"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::enableSharding)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); } - virtual bool isWriteCommandForConfigServer() const { + return Status::OK(); + } + + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + const string ns = parseNs(dbname, cmdObj); + if (ns.size() == 0) { + errmsg = "no ns"; return false; } - virtual void help(std::stringstream& help) const { - help << "Shard a collection. Requires key. Optional unique." - << " Sharding must already be enabled for the database.\n" - << " { enablesharding : \"<dbname>\" }\n"; + const NamespaceString nsStr(ns); + if (!nsStr.isValid()) { + return appendCommandStatus( + result, + Status(ErrorCodes::InvalidNamespace, "invalid collection namespace [" + ns + "]")); } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace( - NamespaceString(parseNs(dbname, - cmdObj))), - ActionType::enableSharding)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - - return Status::OK(); + auto config = uassertStatusOK(grid.catalogCache()->getDatabase(nsStr.db().toString())); + if (!config->isShardingEnabled()) { + return appendCommandStatus( + result, + Status(ErrorCodes::IllegalOperation, + str::stream() << "sharding not enabled for db " << nsStr.db())); } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); + if (config->isSharded(ns)) { + return appendCommandStatus( + result, + Status(ErrorCodes::IllegalOperation, + str::stream() << "sharding already enabled for collection " << ns)); } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - const string ns = parseNs(dbname, cmdObj); - if (ns.size() == 0) { - errmsg = "no ns"; - return false; - } - - const NamespaceString nsStr(ns); - if (!nsStr.isValid()) { - return appendCommandStatus( - result, - Status(ErrorCodes::InvalidNamespace, - "invalid collection namespace [" + ns + "]")); - } - - auto config = uassertStatusOK(grid.catalogCache()->getDatabase(nsStr.db().toString())); - if (!config->isShardingEnabled()) { - return appendCommandStatus( - result, - Status(ErrorCodes::IllegalOperation, - str::stream() << "sharding not enabled for db " << nsStr.db())); - } - - if (config->isSharded(ns)) { - return appendCommandStatus( - result, - Status(ErrorCodes::IllegalOperation, - str::stream() << "sharding already enabled for collection " << ns)); - } + // NOTE: We *must* take ownership of the key here - otherwise the shared BSONObj + // becomes corrupt as soon as the command ends. + BSONObj proposedKey = cmdObj.getObjectField("key").getOwned(); + if (proposedKey.isEmpty()) { + errmsg = "no shard key"; + return false; + } - // NOTE: We *must* take ownership of the key here - otherwise the shared BSONObj - // becomes corrupt as soon as the command ends. - BSONObj proposedKey = cmdObj.getObjectField("key").getOwned(); - if (proposedKey.isEmpty()) { - errmsg = "no shard key"; - return false; - } + ShardKeyPattern proposedKeyPattern(proposedKey); + if (!proposedKeyPattern.isValid()) { + errmsg = str::stream() << "Unsupported shard key pattern. Pattern must" + << " either be a single hashed field, or a list" + << " of ascending fields."; + return false; + } - ShardKeyPattern proposedKeyPattern(proposedKey); - if (!proposedKeyPattern.isValid()) { - errmsg = str::stream() << "Unsupported shard key pattern. Pattern must" - << " either be a single hashed field, or a list" - << " of ascending fields."; - return false; - } + bool isHashedShardKey = proposedKeyPattern.isHashedPattern(); - bool isHashedShardKey = proposedKeyPattern.isHashedPattern(); + if (isHashedShardKey && cmdObj["unique"].trueValue()) { + dassert(proposedKey.nFields() == 1); - if (isHashedShardKey && cmdObj["unique"].trueValue()) { - dassert(proposedKey.nFields() == 1); + // it's possible to ensure uniqueness on the hashed field by + // declaring an additional (non-hashed) unique index on the field, + // but the hashed shard key itself should not be declared unique + errmsg = "hashed shard keys cannot be declared unique."; + return false; + } - // it's possible to ensure uniqueness on the hashed field by - // declaring an additional (non-hashed) unique index on the field, - // but the hashed shard key itself should not be declared unique - errmsg = "hashed shard keys cannot be declared unique."; - return false; - } + if (ns.find(".system.") != string::npos) { + errmsg = "can't shard system namespaces"; + return false; + } - if (ns.find(".system.") != string::npos) { - errmsg = "can't shard system namespaces"; - return false; + // The rest of the checks require a connection to the primary db + ConnectionString shardConnString; + { + const auto shard = grid.shardRegistry()->getShard(config->getPrimaryId()); + shardConnString = shard->getConnString(); + } + ScopedDbConnection conn(shardConnString); + + // check that collection is not capped + BSONObj res; + { + list<BSONObj> all = conn->getCollectionInfos( + config->name(), BSON("name" << nsToCollectionSubstring(ns))); + if (!all.empty()) { + res = all.front().getOwned(); } + } - // The rest of the checks require a connection to the primary db - ConnectionString shardConnString; - { - const auto shard = grid.shardRegistry()->getShard(config->getPrimaryId()); - shardConnString = shard->getConnString(); - } - ScopedDbConnection conn(shardConnString); - - //check that collection is not capped - BSONObj res; - { - list<BSONObj> all = conn->getCollectionInfos( - config->name(), - BSON("name" << nsToCollectionSubstring(ns))); - if (!all.empty()) { - res = all.front().getOwned(); - } - } + if (res["options"].type() == Object && + res["options"].embeddedObject()["capped"].trueValue()) { + errmsg = "can't shard capped collection"; + conn.done(); + return false; + } - if (res["options"].type() == Object && - res["options"].embeddedObject()["capped"].trueValue()) { - errmsg = "can't shard capped collection"; + // The proposed shard key must be validated against the set of existing indexes. + // In particular, we must ensure the following constraints + // + // 1. All existing unique indexes, except those which start with the _id index, + // must contain the proposed key as a prefix (uniqueness of the _id index is + // ensured by the _id generation process or guaranteed by the user). + // + // 2. If the collection is not empty, there must exist at least one index that + // is "useful" for the proposed key. A "useful" index is defined as follows + // Useful Index: + // i. contains proposedKey as a prefix + // ii. is not a sparse index or partial index + // iii. contains no null values + // iv. is not multikey (maybe lift this restriction later) + // v. if a hashed index, has default seed (lift this restriction later) + // + // 3. If the proposed shard key is specified as unique, there must exist a useful, + // unique index exactly equal to the proposedKey (not just a prefix). + // + // After validating these constraint: + // + // 4. If there is no useful index, and the collection is non-empty, we + // must fail. + // + // 5. If the collection is empty, and it's still possible to create an index + // on the proposed key, we go ahead and do so. + + list<BSONObj> indexes = conn->getIndexSpecs(ns); + + // 1. Verify consistency with existing unique indexes + ShardKeyPattern proposedShardKey(proposedKey); + for (list<BSONObj>::iterator it = indexes.begin(); it != indexes.end(); ++it) { + BSONObj idx = *it; + BSONObj currentKey = idx["key"].embeddedObject(); + bool isUnique = idx["unique"].trueValue(); + + if (isUnique && !proposedShardKey.isUniqueIndexCompatible(currentKey)) { + errmsg = str::stream() << "can't shard collection '" << ns << "' " + << "with unique index on " << currentKey << " " + << "and proposed shard key " << proposedKey << ". " + << "Uniqueness can't be maintained unless " + << "shard key is a prefix"; conn.done(); return false; } + } - // The proposed shard key must be validated against the set of existing indexes. - // In particular, we must ensure the following constraints - // - // 1. All existing unique indexes, except those which start with the _id index, - // must contain the proposed key as a prefix (uniqueness of the _id index is - // ensured by the _id generation process or guaranteed by the user). - // - // 2. If the collection is not empty, there must exist at least one index that - // is "useful" for the proposed key. A "useful" index is defined as follows - // Useful Index: - // i. contains proposedKey as a prefix - // ii. is not a sparse index or partial index - // iii. contains no null values - // iv. is not multikey (maybe lift this restriction later) - // v. if a hashed index, has default seed (lift this restriction later) - // - // 3. If the proposed shard key is specified as unique, there must exist a useful, - // unique index exactly equal to the proposedKey (not just a prefix). - // - // After validating these constraint: - // - // 4. If there is no useful index, and the collection is non-empty, we - // must fail. - // - // 5. If the collection is empty, and it's still possible to create an index - // on the proposed key, we go ahead and do so. - - list<BSONObj> indexes = conn->getIndexSpecs(ns); - - // 1. Verify consistency with existing unique indexes - ShardKeyPattern proposedShardKey(proposedKey); - for (list<BSONObj>::iterator it = indexes.begin(); it != indexes.end(); ++it) { - BSONObj idx = *it; - BSONObj currentKey = idx["key"].embeddedObject(); - bool isUnique = idx["unique"].trueValue(); - - if (isUnique && !proposedShardKey.isUniqueIndexCompatible(currentKey)) { - errmsg = str::stream() << "can't shard collection '" << ns << "' " - << "with unique index on " << currentKey << " " - << "and proposed shard key " << proposedKey << ". " - << "Uniqueness can't be maintained unless " - << "shard key is a prefix"; + // 2. Check for a useful index + bool hasUsefulIndexForKey = false; + + for (list<BSONObj>::iterator it = indexes.begin(); it != indexes.end(); ++it) { + BSONObj idx = *it; + BSONObj currentKey = idx["key"].embeddedObject(); + // Check 2.i. and 2.ii. + if (!idx["sparse"].trueValue() && idx["filter"].eoo() && + proposedKey.isPrefixOf(currentKey)) { + // We can't currently use hashed indexes with a non-default hash seed + // Check v. + // Note that this means that, for sharding, we only support one hashed index + // per field per collection. + if (isHashedShardKey && !idx["seed"].eoo() && + idx["seed"].numberInt() != BSONElementHasher::DEFAULT_HASH_SEED) { + errmsg = str::stream() << "can't shard collection " << ns + << " with hashed shard key " << proposedKey + << " because the hashed index uses a non-default" + << " seed of " << idx["seed"].numberInt(); conn.done(); return false; } + + hasUsefulIndexForKey = true; } + } - // 2. Check for a useful index - bool hasUsefulIndexForKey = false; + // 3. If proposed key is required to be unique, additionally check for exact match. + bool careAboutUnique = cmdObj["unique"].trueValue(); + if (hasUsefulIndexForKey && careAboutUnique) { + BSONObj eqQuery = BSON("ns" << ns << "key" << proposedKey); + BSONObj eqQueryResult; for (list<BSONObj>::iterator it = indexes.begin(); it != indexes.end(); ++it) { BSONObj idx = *it; - BSONObj currentKey = idx["key"].embeddedObject(); - // Check 2.i. and 2.ii. - if (!idx["sparse"].trueValue() && - idx["filter"].eoo() && - proposedKey.isPrefixOf(currentKey)) { - - // We can't currently use hashed indexes with a non-default hash seed - // Check v. - // Note that this means that, for sharding, we only support one hashed index - // per field per collection. - if (isHashedShardKey && - !idx["seed"].eoo() && - idx["seed"].numberInt() != BSONElementHasher::DEFAULT_HASH_SEED) { - - errmsg = str::stream() << "can't shard collection " << ns - << " with hashed shard key " << proposedKey - << " because the hashed index uses a non-default" - << " seed of " << idx["seed"].numberInt(); - conn.done(); - return false; - } - - hasUsefulIndexForKey = true; - } - } - - // 3. If proposed key is required to be unique, additionally check for exact match. - bool careAboutUnique = cmdObj["unique"].trueValue(); - if (hasUsefulIndexForKey && careAboutUnique) { - BSONObj eqQuery = BSON("ns" << ns << "key" << proposedKey); - BSONObj eqQueryResult; - - for (list<BSONObj>::iterator it = indexes.begin(); it != indexes.end(); ++it) { - BSONObj idx = *it; - if (idx["key"].embeddedObject() == proposedKey) { - eqQueryResult = idx; - break; - } - } - - if (eqQueryResult.isEmpty()) { - // If no exact match, index not useful, but still possible to create one later - hasUsefulIndexForKey = false; - } - else { - bool isExplicitlyUnique = eqQueryResult["unique"].trueValue(); - BSONObj currKey = eqQueryResult["key"].embeddedObject(); - bool isCurrentID = str::equals(currKey.firstElementFieldName(), "_id"); - - if (!isExplicitlyUnique && !isCurrentID) { - errmsg = str::stream() << "can't shard collection " << ns << ", " - << proposedKey << " index not unique, " - << "and unique index explicitly specified"; - conn.done(); - return false; - } + if (idx["key"].embeddedObject() == proposedKey) { + eqQueryResult = idx; + break; } } - if (hasUsefulIndexForKey) { - // Check 2.iii and 2.iv. Make sure no null entries in the sharding index - // and that there is a useful, non-multikey index available - BSONObjBuilder checkShardingIndexCmd; - checkShardingIndexCmd.append("checkShardingIndex", ns); - checkShardingIndexCmd.append("keyPattern", proposedKey); - - if (!conn.get()->runCommand("admin", checkShardingIndexCmd.obj(), res)) { - errmsg = res["errmsg"].str(); + if (eqQueryResult.isEmpty()) { + // If no exact match, index not useful, but still possible to create one later + hasUsefulIndexForKey = false; + } else { + bool isExplicitlyUnique = eqQueryResult["unique"].trueValue(); + BSONObj currKey = eqQueryResult["key"].embeddedObject(); + bool isCurrentID = str::equals(currKey.firstElementFieldName(), "_id"); + + if (!isExplicitlyUnique && !isCurrentID) { + errmsg = str::stream() << "can't shard collection " << ns << ", " << proposedKey + << " index not unique, " + << "and unique index explicitly specified"; conn.done(); return false; } } - else if (conn->count(ns) != 0) { - // 4. if no useful index, and collection is non-empty, fail - errmsg = str::stream() << "please create an index that starts with the " - << "shard key before sharding."; - result.append("proposedKey", proposedKey); - result.append("curIndexes", indexes); + } + + if (hasUsefulIndexForKey) { + // Check 2.iii and 2.iv. Make sure no null entries in the sharding index + // and that there is a useful, non-multikey index available + BSONObjBuilder checkShardingIndexCmd; + checkShardingIndexCmd.append("checkShardingIndex", ns); + checkShardingIndexCmd.append("keyPattern", proposedKey); + + if (!conn.get()->runCommand("admin", checkShardingIndexCmd.obj(), res)) { + errmsg = res["errmsg"].str(); conn.done(); return false; } - else { - // 5. If no useful index exists, and collection empty, create one on proposedKey. - // Only need to call ensureIndex on primary shard, since indexes get copied to - // receiving shard whenever a migrate occurs. - Status status = clusterCreateIndex(ns, proposedKey, careAboutUnique, NULL); - if (!status.isOK()) { - errmsg = str::stream() << "ensureIndex failed to create index on " - << "primary shard: " << status.reason(); - conn.done(); - return false; - } + } else if (conn->count(ns) != 0) { + // 4. if no useful index, and collection is non-empty, fail + errmsg = str::stream() << "please create an index that starts with the " + << "shard key before sharding."; + result.append("proposedKey", proposedKey); + result.append("curIndexes", indexes); + conn.done(); + return false; + } else { + // 5. If no useful index exists, and collection empty, create one on proposedKey. + // Only need to call ensureIndex on primary shard, since indexes get copied to + // receiving shard whenever a migrate occurs. + Status status = clusterCreateIndex(ns, proposedKey, careAboutUnique, NULL); + if (!status.isOK()) { + errmsg = str::stream() << "ensureIndex failed to create index on " + << "primary shard: " << status.reason(); + conn.done(); + return false; } + } - bool isEmpty = (conn->count(ns) == 0); - - conn.done(); + bool isEmpty = (conn->count(ns) == 0); - // Pre-splitting: - // For new collections which use hashed shard keys, we can can pre-split the - // range of possible hashes into a large number of chunks, and distribute them - // evenly at creation time. Until we design a better initialization scheme, the - // safest way to pre-split is to - // 1. make one big chunk for each shard - // 2. move them one at a time - // 3. split the big chunks to achieve the desired total number of initial chunks - - vector<ShardId> shardIds; - grid.shardRegistry()->getAllShardIds(&shardIds); - int numShards = shardIds.size(); - - vector<BSONObj> initSplits; // there will be at most numShards-1 of these - vector<BSONObj> allSplits; // all of the initial desired split points - - // only pre-split when using a hashed shard key and collection is still empty - if (isHashedShardKey && isEmpty){ - int numChunks = cmdObj["numInitialChunks"].numberInt(); - if (numChunks <= 0) { - // default number of initial chunks - numChunks = 2 * numShards; - } + conn.done(); - // hashes are signed, 64-bit ints. So we divide the range (-MIN long, +MAX long) - // into intervals of size (2^64/numChunks) and create split points at the - // boundaries. The logic below ensures that initial chunks are all - // symmetric around 0. - long long intervalSize = (std::numeric_limits<long long>::max() / numChunks) * 2; - long long current = 0; + // Pre-splitting: + // For new collections which use hashed shard keys, we can can pre-split the + // range of possible hashes into a large number of chunks, and distribute them + // evenly at creation time. Until we design a better initialization scheme, the + // safest way to pre-split is to + // 1. make one big chunk for each shard + // 2. move them one at a time + // 3. split the big chunks to achieve the desired total number of initial chunks - if (numChunks % 2 == 0){ - allSplits.push_back(BSON(proposedKey.firstElementFieldName() << current)); - current += intervalSize; - } - else { - current += intervalSize / 2; - } + vector<ShardId> shardIds; + grid.shardRegistry()->getAllShardIds(&shardIds); + int numShards = shardIds.size(); - for (int i = 0; i < (numChunks - 1) / 2; i++){ - allSplits.push_back(BSON(proposedKey.firstElementFieldName() << current)); - allSplits.push_back(BSON(proposedKey.firstElementFieldName() << -current)); - current += intervalSize; - } + vector<BSONObj> initSplits; // there will be at most numShards-1 of these + vector<BSONObj> allSplits; // all of the initial desired split points - sort(allSplits.begin(), allSplits.end()); + // only pre-split when using a hashed shard key and collection is still empty + if (isHashedShardKey && isEmpty) { + int numChunks = cmdObj["numInitialChunks"].numberInt(); + if (numChunks <= 0) { + // default number of initial chunks + numChunks = 2 * numShards; + } - // 1. the initial splits define the "big chunks" that we will subdivide later - int lastIndex = -1; - for (int i = 1; i < numShards; i++) { - if (lastIndex < (i*numChunks) / numShards - 1) { - lastIndex = (i*numChunks) / numShards - 1; - initSplits.push_back(allSplits[lastIndex]); - } - } + // hashes are signed, 64-bit ints. So we divide the range (-MIN long, +MAX long) + // into intervals of size (2^64/numChunks) and create split points at the + // boundaries. The logic below ensures that initial chunks are all + // symmetric around 0. + long long intervalSize = (std::numeric_limits<long long>::max() / numChunks) * 2; + long long current = 0; + + if (numChunks % 2 == 0) { + allSplits.push_back(BSON(proposedKey.firstElementFieldName() << current)); + current += intervalSize; + } else { + current += intervalSize / 2; } - LOG(0) << "CMD: shardcollection: " << cmdObj; + for (int i = 0; i < (numChunks - 1) / 2; i++) { + allSplits.push_back(BSON(proposedKey.firstElementFieldName() << current)); + allSplits.push_back(BSON(proposedKey.firstElementFieldName() << -current)); + current += intervalSize; + } - audit::logShardCollection(ClientBasic::getCurrent(), - ns, - proposedKey, - careAboutUnique); + sort(allSplits.begin(), allSplits.end()); - Status status = grid.catalogManager()->shardCollection(ns, - proposedShardKey, - careAboutUnique, - &initSplits); - if (!status.isOK()) { - return appendCommandStatus(result, status); + // 1. the initial splits define the "big chunks" that we will subdivide later + int lastIndex = -1; + for (int i = 1; i < numShards; i++) { + if (lastIndex < (i * numChunks) / numShards - 1) { + lastIndex = (i * numChunks) / numShards - 1; + initSplits.push_back(allSplits[lastIndex]); + } } + } - result << "collectionsharded" << ns; - - // Only initially move chunks when using a hashed shard key - if (isHashedShardKey && isEmpty) { - // Reload the new config info. If we created more than one initial chunk, then - // we need to move them around to balance. - ChunkManagerPtr chunkManager = config->getChunkManager(ns, true); - ChunkMap chunkMap = chunkManager->getChunkMap(); - - // 2. Move and commit each "big chunk" to a different shard. - int i = 0; - for (ChunkMap::const_iterator c = chunkMap.begin(); c != chunkMap.end(); ++c, ++i){ - const ShardId& shardId = shardIds[i % numShards]; - const auto to = grid.shardRegistry()->getShard(shardId); - if (!to) { - continue; - } + LOG(0) << "CMD: shardcollection: " << cmdObj; - ChunkPtr chunk = c->second; + audit::logShardCollection(ClientBasic::getCurrent(), ns, proposedKey, careAboutUnique); - // can't move chunk to shard it's already on - if (to->getId() == chunk->getShardId()) { - continue; - } + Status status = grid.catalogManager()->shardCollection( + ns, proposedShardKey, careAboutUnique, &initSplits); + if (!status.isOK()) { + return appendCommandStatus(result, status); + } - BSONObj moveResult; - WriteConcernOptions noThrottle; - if (!chunk->moveAndCommit(to->getId(), - Chunk::MaxChunkSize, - &noThrottle, - true, - 0, - moveResult)) { - - warning() << "couldn't move chunk " << chunk->toString() - << " to shard " << *to - << " while sharding collection " << ns << "." - << " Reason: " << moveResult; - } + result << "collectionsharded" << ns; + + // Only initially move chunks when using a hashed shard key + if (isHashedShardKey && isEmpty) { + // Reload the new config info. If we created more than one initial chunk, then + // we need to move them around to balance. + ChunkManagerPtr chunkManager = config->getChunkManager(ns, true); + ChunkMap chunkMap = chunkManager->getChunkMap(); + + // 2. Move and commit each "big chunk" to a different shard. + int i = 0; + for (ChunkMap::const_iterator c = chunkMap.begin(); c != chunkMap.end(); ++c, ++i) { + const ShardId& shardId = shardIds[i % numShards]; + const auto to = grid.shardRegistry()->getShard(shardId); + if (!to) { + continue; + } + + ChunkPtr chunk = c->second; + + // can't move chunk to shard it's already on + if (to->getId() == chunk->getShardId()) { + continue; } - if (allSplits.empty()) { - return true; + BSONObj moveResult; + WriteConcernOptions noThrottle; + if (!chunk->moveAndCommit( + to->getId(), Chunk::MaxChunkSize, &noThrottle, true, 0, moveResult)) { + warning() << "couldn't move chunk " << chunk->toString() << " to shard " << *to + << " while sharding collection " << ns << "." + << " Reason: " << moveResult; } + } - // Reload the config info, after all the migrations - chunkManager = config->getChunkManager(ns, true); - - // 3. Subdivide the big chunks by splitting at each of the points in "allSplits" - // that we haven't already split by. - ChunkPtr currentChunk = chunkManager->findIntersectingChunk(allSplits[0]); - - vector<BSONObj> subSplits; - for (unsigned i = 0; i <= allSplits.size(); i++){ - if (i == allSplits.size() || !currentChunk->containsKey(allSplits[i])) { - if (!subSplits.empty()){ - Status status = currentChunk->multiSplit(subSplits, NULL); - if (!status.isOK()){ - warning() << "couldn't split chunk " - << currentChunk->toString() - << " while sharding collection " << ns - << causedBy(status); - } - - subSplits.clear(); - } + if (allSplits.empty()) { + return true; + } - if (i < allSplits.size()) { - currentChunk = chunkManager->findIntersectingChunk(allSplits[i]); + // Reload the config info, after all the migrations + chunkManager = config->getChunkManager(ns, true); + + // 3. Subdivide the big chunks by splitting at each of the points in "allSplits" + // that we haven't already split by. + ChunkPtr currentChunk = chunkManager->findIntersectingChunk(allSplits[0]); + + vector<BSONObj> subSplits; + for (unsigned i = 0; i <= allSplits.size(); i++) { + if (i == allSplits.size() || !currentChunk->containsKey(allSplits[i])) { + if (!subSplits.empty()) { + Status status = currentChunk->multiSplit(subSplits, NULL); + if (!status.isOK()) { + warning() << "couldn't split chunk " << currentChunk->toString() + << " while sharding collection " << ns << causedBy(status); } + + subSplits.clear(); } - else { - BSONObj splitPoint(allSplits[i]); - // Do not split on the boundaries - if (currentChunk->getMin().woCompare(splitPoint) == 0) { - continue; - } + if (i < allSplits.size()) { + currentChunk = chunkManager->findIntersectingChunk(allSplits[i]); + } + } else { + BSONObj splitPoint(allSplits[i]); - subSplits.push_back(splitPoint); + // Do not split on the boundaries + if (currentChunk->getMin().woCompare(splitPoint) == 0) { + continue; } - } - // Proactively refresh the chunk manager. Not really necessary, but this way it's - // immediately up-to-date the next time it's used. - config->getChunkManager(ns, true); + subSplits.push_back(splitPoint); + } } - return true; + // Proactively refresh the chunk manager. Not really necessary, but this way it's + // immediately up-to-date the next time it's used. + config->getChunkManager(ns, true); } - } shardCollectionCmd; + return true; + } + +} shardCollectionCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_shutdown_cmd.cpp b/src/mongo/s/commands/cluster_shutdown_cmd.cpp index 58fe1d47483..72f0fd71e6f 100644 --- a/src/mongo/s/commands/cluster_shutdown_cmd.cpp +++ b/src/mongo/s/commands/cluster_shutdown_cmd.cpp @@ -34,26 +34,25 @@ namespace mongo { namespace { - class ClusterShutdownCmd : public CmdShutdown { - public: - virtual void help(std::stringstream& help) const { - help << "shutdown the database. must be ran against admin db and " - << "either (1) ran from localhost or (2) authenticated."; - } - - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { - - // Never returns - shutdownHelper(); - return true; - } - - } clusterShutdownCmd; - -} // namespace -} // namespace mongo +class ClusterShutdownCmd : public CmdShutdown { +public: + virtual void help(std::stringstream& help) const { + help << "shutdown the database. must be ran against admin db and " + << "either (1) ran from localhost or (2) authenticated."; + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + // Never returns + shutdownHelper(); + return true; + } + +} clusterShutdownCmd; + +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_split_collection_cmd.cpp b/src/mongo/s/commands/cluster_split_collection_cmd.cpp index acd99d32c7b..fd5f462518c 100644 --- a/src/mongo/s/commands/cluster_split_collection_cmd.cpp +++ b/src/mongo/s/commands/cluster_split_collection_cmd.cpp @@ -49,244 +49,231 @@ namespace mongo { - using std::shared_ptr; - using std::string; - using std::vector; +using std::shared_ptr; +using std::string; +using std::vector; namespace { - class SplitCollectionCmd : public Command { - public: - SplitCollectionCmd() : Command("split", false, "split") { } +class SplitCollectionCmd : public Command { +public: + SplitCollectionCmd() : Command("split", false, "split") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(std::stringstream& help) const { + help << " example: - split the shard that contains give key\n" + << " { split : 'alleyinsider.blog.posts' , find : { ts : 1 } }\n" + << " example: - split the shard that contains the key with this as the middle\n" + << " { split : 'alleyinsider.blog.posts' , middle : { ts : 1 } }\n" + << " NOTE: this does not move the chunks, it just creates a logical separation."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::splitChunk)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); + } + return Status::OK(); + } + + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } + + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + ShardConnection::sync(); + + const NamespaceString nss(parseNs(dbname, cmdObj)); + if (nss.size() == 0) { + return appendCommandStatus( + result, Status(ErrorCodes::InvalidNamespace, "no namespace specified")); + } - virtual bool slaveOk() const { - return true; + auto status = grid.catalogCache()->getDatabase(nss.db().toString()); + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); } - virtual bool adminOnly() const { - return true; + std::shared_ptr<DBConfig> config = status.getValue(); + if (!config->isSharded(nss.ns())) { + config->reload(); + + if (!config->isSharded(nss.ns())) { + return appendCommandStatus(result, + Status(ErrorCodes::NamespaceNotSharded, + "ns [" + nss.ns() + " is not sharded.")); + } } - virtual bool isWriteCommandForConfigServer() const { + const BSONField<BSONObj> findField("find", BSONObj()); + const BSONField<BSONArray> boundsField("bounds", BSONArray()); + const BSONField<BSONObj> middleField("middle", BSONObj()); + + BSONObj find; + if (FieldParser::extract(cmdObj, findField, &find, &errmsg) == FieldParser::FIELD_INVALID) { return false; } - virtual void help(std::stringstream& help) const { - help << " example: - split the shard that contains give key\n" - << " { split : 'alleyinsider.blog.posts' , find : { ts : 1 } }\n" - << " example: - split the shard that contains the key with this as the middle\n" - << " { split : 'alleyinsider.blog.posts' , middle : { ts : 1 } }\n" - << " NOTE: this does not move the chunks, it just creates a logical separation."; + BSONArray bounds; + if (FieldParser::extract(cmdObj, boundsField, &bounds, &errmsg) == + FieldParser::FIELD_INVALID) { + return false; } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { + if (!bounds.isEmpty()) { + if (!bounds.hasField("0")) { + errmsg = "lower bound not specified"; + return false; + } - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace( - NamespaceString(parseNs(dbname, - cmdObj))), - ActionType::splitChunk)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); + if (!bounds.hasField("1")) { + errmsg = "upper bound not specified"; + return false; } - return Status::OK(); } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); + if (!find.isEmpty() && !bounds.isEmpty()) { + errmsg = "cannot specify bounds and find at the same time"; + return false; } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { + BSONObj middle; + if (FieldParser::extract(cmdObj, middleField, &middle, &errmsg) == + FieldParser::FIELD_INVALID) { + return false; + } - ShardConnection::sync(); + if (find.isEmpty() && bounds.isEmpty() && middle.isEmpty()) { + errmsg = "need to specify find/bounds or middle"; + return false; + } - const NamespaceString nss(parseNs(dbname, cmdObj)); - if (nss.size() == 0) { - return appendCommandStatus(result, Status(ErrorCodes::InvalidNamespace, - "no namespace specified")); - } + if (!find.isEmpty() && !middle.isEmpty()) { + errmsg = "cannot specify find and middle together"; + return false; + } - auto status = grid.catalogCache()->getDatabase(nss.db().toString()); - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } + if (!bounds.isEmpty() && !middle.isEmpty()) { + errmsg = "cannot specify bounds and middle together"; + return false; + } - std::shared_ptr<DBConfig> config = status.getValue(); - if (!config->isSharded(nss.ns())) { - config->reload(); + // This refreshes the chunk metadata if stale. + ChunkManagerPtr info = config->getChunkManager(nss.ns(), true); + ChunkPtr chunk; - if (!config->isSharded(nss.ns())) { - return appendCommandStatus(result, - Status(ErrorCodes::NamespaceNotSharded, - "ns [" + nss.ns() + " is not sharded.")); - } - } + if (!find.isEmpty()) { + StatusWith<BSONObj> status = info->getShardKeyPattern().extractShardKeyFromQuery(find); - const BSONField<BSONObj> findField("find", BSONObj()); - const BSONField<BSONArray> boundsField("bounds", BSONArray()); - const BSONField<BSONObj> middleField("middle", BSONObj()); + // Bad query + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); + } - BSONObj find; - if (FieldParser::extract(cmdObj, findField, &find, &errmsg) == - FieldParser::FIELD_INVALID) { + BSONObj shardKey = status.getValue(); + if (shardKey.isEmpty()) { + errmsg = stream() << "no shard key found in chunk query " << find; return false; } - BSONArray bounds; - if (FieldParser::extract(cmdObj, boundsField, &bounds, &errmsg) == - FieldParser::FIELD_INVALID) { + chunk = info->findIntersectingChunk(shardKey); + invariant(chunk.get()); + } else if (!bounds.isEmpty()) { + if (!info->getShardKeyPattern().isShardKey(bounds[0].Obj()) || + !info->getShardKeyPattern().isShardKey(bounds[1].Obj())) { + errmsg = stream() << "shard key bounds " + << "[" << bounds[0].Obj() << "," << bounds[1].Obj() << ")" + << " are not valid for shard key pattern " + << info->getShardKeyPattern().toBSON(); return false; } - if (!bounds.isEmpty()) { - if (!bounds.hasField("0")) { - errmsg = "lower bound not specified"; - return false; - } + BSONObj minKey = info->getShardKeyPattern().normalizeShardKey(bounds[0].Obj()); + BSONObj maxKey = info->getShardKeyPattern().normalizeShardKey(bounds[1].Obj()); - if (!bounds.hasField("1")) { - errmsg = "upper bound not specified"; - return false; - } - } + chunk = info->findIntersectingChunk(minKey); + invariant(chunk.get()); - if (!find.isEmpty() && !bounds.isEmpty()) { - errmsg = "cannot specify bounds and find at the same time"; + if (chunk->getMin().woCompare(minKey) != 0 || chunk->getMax().woCompare(maxKey) != 0) { + errmsg = stream() << "no chunk found with the shard key bounds " + << "[" << minKey << "," << maxKey << ")"; return false; } - - BSONObj middle; - if (FieldParser::extract(cmdObj, middleField, &middle, &errmsg) == - FieldParser::FIELD_INVALID) { + } else { + // Middle + if (!info->getShardKeyPattern().isShardKey(middle)) { + errmsg = stream() << "new split key " << middle + << " is not valid for shard key pattern " + << info->getShardKeyPattern().toBSON(); return false; } - if (find.isEmpty() && bounds.isEmpty() && middle.isEmpty()) { - errmsg = "need to specify find/bounds or middle"; - return false; - } + middle = info->getShardKeyPattern().normalizeShardKey(middle); - if (!find.isEmpty() && !middle.isEmpty()) { - errmsg = "cannot specify find and middle together"; - return false; + // Check shard key size when manually provided + Status status = ShardKeyPattern::checkShardKeySize(middle); + if (!status.isOK()) { + return appendCommandStatus(result, status); } - if (!bounds.isEmpty() && !middle.isEmpty()) { - errmsg = "cannot specify bounds and middle together"; + chunk = info->findIntersectingChunk(middle); + invariant(chunk.get()); + + if (chunk->getMin().woCompare(middle) == 0 || chunk->getMax().woCompare(middle) == 0) { + errmsg = stream() << "new split key " << middle + << " is a boundary key of existing chunk " + << "[" << chunk->getMin() << "," << chunk->getMax() << ")"; return false; } + } - // This refreshes the chunk metadata if stale. - ChunkManagerPtr info = config->getChunkManager(nss.ns(), true); - ChunkPtr chunk; - - if (!find.isEmpty()) { - StatusWith<BSONObj> status = - info->getShardKeyPattern().extractShardKeyFromQuery(find); - - // Bad query - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } - - BSONObj shardKey = status.getValue(); - if (shardKey.isEmpty()) { - errmsg = stream() << "no shard key found in chunk query " << find; - return false; - } + invariant(chunk.get()); + log() << "splitting chunk [" << chunk->getMin() << "," << chunk->getMax() << ")" + << " in collection " << nss.ns() << " on shard " << chunk->getShardId(); - chunk = info->findIntersectingChunk(shardKey); - invariant(chunk.get()); - } - else if (!bounds.isEmpty()) { - - if (!info->getShardKeyPattern().isShardKey(bounds[0].Obj()) - || !info->getShardKeyPattern().isShardKey(bounds[1].Obj())) { - errmsg = stream() << "shard key bounds " << "[" << bounds[0].Obj() << "," - << bounds[1].Obj() << ")" - << " are not valid for shard key pattern " - << info->getShardKeyPattern().toBSON(); - return false; - } - - BSONObj minKey = info->getShardKeyPattern().normalizeShardKey(bounds[0].Obj()); - BSONObj maxKey = info->getShardKeyPattern().normalizeShardKey(bounds[1].Obj()); - - chunk = info->findIntersectingChunk(minKey); - invariant(chunk.get()); - - if (chunk->getMin().woCompare(minKey) != 0 - || chunk->getMax().woCompare(maxKey) != 0) { - errmsg = stream() << "no chunk found with the shard key bounds " << "[" - << minKey << "," << maxKey << ")"; - return false; - } - } - else { - // Middle - if (!info->getShardKeyPattern().isShardKey(middle)) { - errmsg = stream() << "new split key " << middle - << " is not valid for shard key pattern " - << info->getShardKeyPattern().toBSON(); - return false; - } - - middle = info->getShardKeyPattern().normalizeShardKey(middle); - - // Check shard key size when manually provided - Status status = ShardKeyPattern::checkShardKeySize(middle); - if (!status.isOK()) { - return appendCommandStatus(result, status); - } - - chunk = info->findIntersectingChunk(middle); - invariant(chunk.get()); - - if (chunk->getMin().woCompare(middle) == 0 - || chunk->getMax().woCompare(middle) == 0) { - errmsg = stream() << "new split key " << middle - << " is a boundary key of existing chunk " << "[" - << chunk->getMin() << "," << chunk->getMax() << ")"; - return false; - } + BSONObj res; + if (middle.isEmpty()) { + Status status = chunk->split(Chunk::atMedian, NULL, NULL); + if (!status.isOK()) { + errmsg = "split failed"; + result.append("cause", status.toString()); + return false; } + } else { + vector<BSONObj> splitPoints; + splitPoints.push_back(middle); - invariant(chunk.get()); - log() << "splitting chunk [" << chunk->getMin() << "," << chunk->getMax() << ")" - << " in collection " << nss.ns() - << " on shard " << chunk->getShardId(); - - BSONObj res; - if (middle.isEmpty()) { - Status status = chunk->split(Chunk::atMedian, NULL, NULL); - if (!status.isOK()) { - errmsg = "split failed"; - result.append("cause", status.toString()); - return false; - } - } - else { - vector<BSONObj> splitPoints; - splitPoints.push_back(middle); - - Status status = chunk->multiSplit(splitPoints, NULL); - if (!status.isOK()) { - errmsg = "split failed"; - result.append("cause", status.toString()); - return false; - } + Status status = chunk->multiSplit(splitPoints, NULL); + if (!status.isOK()) { + errmsg = "split failed"; + result.append("cause", status.toString()); + return false; } - - return true; } - } splitCollectionCmd; + return true; + } + +} splitCollectionCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_user_management_commands.cpp b/src/mongo/s/commands/cluster_user_management_commands.cpp index 2c781ac0e91..d2a6ab667d4 100644 --- a/src/mongo/s/commands/cluster_user_management_commands.cpp +++ b/src/mongo/s/commands/cluster_user_management_commands.cpp @@ -46,746 +46,761 @@ namespace mongo { - using std::string; - using std::stringstream; - using std::vector; +using std::string; +using std::stringstream; +using std::vector; + +class CmdCreateUser : public Command { +public: + CmdCreateUser() : Command("createUser") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Adds a user to the system"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForCreateUserCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + return grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + } + + virtual void redactForLogging(mutablebson::Document* cmdObj) { + auth::redactPasswordData(cmdObj->root()); + } + +} cmdCreateUser; + +class CmdUpdateUser : public Command { +public: + CmdUpdateUser() : Command("updateUser") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Used to update a user, for example to change its password"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForUpdateUserCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + auth::CreateOrUpdateUserArgs args; + Status status = auth::parseCreateOrUpdateUserCommands(cmdObj, this->name, dbname, &args); + if (!status.isOK()) { + return appendCommandStatus(result, status); + } + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserByName(args.userName); + + return ok; + } + + virtual void redactForLogging(mutablebson::Document* cmdObj) { + auth::redactPasswordData(cmdObj->root()); + } + +} cmdUpdateUser; + +class CmdDropUser : public Command { +public: + CmdDropUser() : Command("dropUser") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Drops a single user."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForDropUserCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + UserName userName; + BSONObj unusedWriteConcern; + Status status = + auth::parseAndValidateDropUserCommand(cmdObj, dbname, &userName, &unusedWriteConcern); + if (!status.isOK()) { + return appendCommandStatus(result, status); + } + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserByName(userName); + + return ok; + } + +} cmdDropUser; + +class CmdDropAllUsersFromDatabase : public Command { +public: + CmdDropAllUsersFromDatabase() : Command("dropAllUsersFromDatabase") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Drops all users for a single database."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForDropAllUsersFromDatabaseCommand(client, dbname); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUsersFromDB(dbname); + + return ok; + } + +} cmdDropAllUsersFromDatabase; + +class CmdGrantRolesToUser : public Command { +public: + CmdGrantRolesToUser() : Command("grantRolesToUser") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Grants roles to a user."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForGrantRolesToUserCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + string userNameString; + vector<RoleName> roles; + BSONObj unusedWriteConcern; + Status status = auth::parseRolePossessionManipulationCommands( + cmdObj, this->name, dbname, &userNameString, &roles, &unusedWriteConcern); + if (!status.isOK()) { + return appendCommandStatus(result, status); + } + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserByName(UserName(userNameString, dbname)); + + return ok; + } + +} cmdGrantRolesToUser; + +class CmdRevokeRolesFromUser : public Command { +public: + CmdRevokeRolesFromUser() : Command("revokeRolesFromUser") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Revokes roles from a user."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForRevokeRolesFromUserCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + string userNameString; + vector<RoleName> unusedRoles; + BSONObj unusedWriteConcern; + Status status = auth::parseRolePossessionManipulationCommands( + cmdObj, this->name, dbname, &userNameString, &unusedRoles, &unusedWriteConcern); + if (!status.isOK()) { + return appendCommandStatus(result, status); + } + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserByName(UserName(userNameString, dbname)); + + return ok; + } + +} cmdRevokeRolesFromUser; + +class CmdUsersInfo : public Command { +public: + virtual bool slaveOk() const { + return false; + } + + virtual bool slaveOverrideOk() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + CmdUsersInfo() : Command("usersInfo") {} + + virtual void help(stringstream& ss) const { + ss << "Returns information about users."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForUsersInfoCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + return grid.catalogManager()->runUserManagementReadCommand(dbname, cmdObj, &result); + } + +} cmdUsersInfo; + +class CmdCreateRole : public Command { +public: + CmdCreateRole() : Command("createRole") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Adds a role to the system"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForCreateRoleCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + return grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + } + +} cmdCreateRole; + +class CmdUpdateRole : public Command { +public: + CmdUpdateRole() : Command("updateRole") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Used to update a role"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForUpdateRoleCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserCache(); + + return ok; + } + +} cmdUpdateRole; + +class CmdGrantPrivilegesToRole : public Command { +public: + CmdGrantPrivilegesToRole() : Command("grantPrivilegesToRole") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Grants privileges to a role"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForGrantPrivilegesToRoleCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserCache(); + + return ok; + } + +} cmdGrantPrivilegesToRole; + +class CmdRevokePrivilegesFromRole : public Command { +public: + CmdRevokePrivilegesFromRole() : Command("revokePrivilegesFromRole") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Revokes privileges from a role"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForRevokePrivilegesFromRoleCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserCache(); + + return ok; + } + +} cmdRevokePrivilegesFromRole; + +class CmdGrantRolesToRole : public Command { +public: + CmdGrantRolesToRole() : Command("grantRolesToRole") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Grants roles to another role."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForGrantRolesToRoleCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserCache(); + + return ok; + } + +} cmdGrantRolesToRole; + +class CmdRevokeRolesFromRole : public Command { +public: + CmdRevokeRolesFromRole() : Command("revokeRolesFromRole") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Revokes roles from another role."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForRevokeRolesFromRoleCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserCache(); + + return ok; + } + +} cmdRevokeRolesFromRole; + +class CmdDropRole : public Command { +public: + CmdDropRole() : Command("dropRole") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Drops a single role. Before deleting the role completely it must remove it " + "from any users or roles that reference it. If any errors occur in the middle " + "of that process it's possible to be left in a state where the role has been " + "removed from some user/roles but otherwise still exists."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForDropRoleCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserCache(); + + return ok; + } + +} cmdDropRole; + +class CmdDropAllRolesFromDatabase : public Command { +public: + CmdDropAllRolesFromDatabase() : Command("dropAllRolesFromDatabase") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Drops all roles from the given database. Before deleting the roles completely " + "it must remove them from any users or other roles that reference them. If any " + "errors occur in the middle of that process it's possible to be left in a state " + "where the roles have been removed from some user/roles but otherwise still " + "exist."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForDropAllRolesFromDatabaseCommand(client, dbname); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const bool ok = grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserCache(); + + return ok; + } + +} cmdDropAllRolesFromDatabase; + +class CmdRolesInfo : public Command { +public: + CmdRolesInfo() : Command("rolesInfo") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool slaveOverrideOk() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Returns information about roles."; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForRolesInfoCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + return grid.catalogManager()->runUserManagementReadCommand(dbname, cmdObj, &result); + } + +} cmdRolesInfo; + +class CmdInvalidateUserCache : public Command { +public: + CmdInvalidateUserCache() : Command("invalidateUserCache") {} + + virtual bool slaveOk() const { + return true; + } + + virtual bool adminOnly() const { + return true; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual void help(stringstream& ss) const { + ss << "Invalidates the in-memory cache of user information"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForInvalidateUserCacheCommand(client); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + AuthorizationManager* authzManager = getGlobalAuthorizationManager(); + invariant(authzManager); + authzManager->invalidateUserCache(); + return true; + } + +} cmdInvalidateUserCache; - class CmdCreateUser : public Command { - public: - - CmdCreateUser() : Command("createUser") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Adds a user to the system"; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForCreateUserCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - return grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - } - - virtual void redactForLogging(mutablebson::Document* cmdObj) { - auth::redactPasswordData(cmdObj->root()); - } - - } cmdCreateUser; - - class CmdUpdateUser : public Command { - public: - - CmdUpdateUser() : Command("updateUser") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Used to update a user, for example to change its password"; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForUpdateUserCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - auth::CreateOrUpdateUserArgs args; - Status status = auth::parseCreateOrUpdateUserCommands(cmdObj, - this->name, - dbname, - &args); - if (!status.isOK()) { - return appendCommandStatus(result, status); - } - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserByName(args.userName); - - return ok; - } - - virtual void redactForLogging(mutablebson::Document* cmdObj) { - auth::redactPasswordData(cmdObj->root()); - } - - } cmdUpdateUser; - - class CmdDropUser : public Command { - public: - - CmdDropUser() : Command("dropUser") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Drops a single user."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForDropUserCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - UserName userName; - BSONObj unusedWriteConcern; - Status status = auth::parseAndValidateDropUserCommand(cmdObj, - dbname, - &userName, - &unusedWriteConcern); - if (!status.isOK()) { - return appendCommandStatus(result, status); - } - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserByName(userName); - - return ok; - } - - } cmdDropUser; - - class CmdDropAllUsersFromDatabase : public Command { - public: - - CmdDropAllUsersFromDatabase() : Command("dropAllUsersFromDatabase") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Drops all users for a single database."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForDropAllUsersFromDatabaseCommand(client, dbname); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUsersFromDB(dbname); - - return ok; - } - - } cmdDropAllUsersFromDatabase; - - class CmdGrantRolesToUser: public Command { - public: - - CmdGrantRolesToUser() : Command("grantRolesToUser") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Grants roles to a user."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForGrantRolesToUserCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - string userNameString; - vector<RoleName> roles; - BSONObj unusedWriteConcern; - Status status = auth::parseRolePossessionManipulationCommands(cmdObj, - this->name, - dbname, - &userNameString, - &roles, - &unusedWriteConcern); - if (!status.isOK()) { - return appendCommandStatus(result, status); - } - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserByName(UserName(userNameString, dbname)); - - return ok; - } - - } cmdGrantRolesToUser; - - class CmdRevokeRolesFromUser: public Command { - public: - - CmdRevokeRolesFromUser() : Command("revokeRolesFromUser") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Revokes roles from a user."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForRevokeRolesFromUserCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - string userNameString; - vector<RoleName> unusedRoles; - BSONObj unusedWriteConcern; - Status status = auth::parseRolePossessionManipulationCommands(cmdObj, - this->name, - dbname, - &userNameString, - &unusedRoles, - &unusedWriteConcern); - if (!status.isOK()) { - return appendCommandStatus(result, status); - } - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserByName(UserName(userNameString, dbname)); - - return ok; - } - - } cmdRevokeRolesFromUser; - - class CmdUsersInfo: public Command { - public: - - virtual bool slaveOk() const { return false; } - - virtual bool slaveOverrideOk() const { return true; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - CmdUsersInfo() : Command("usersInfo") {} - - virtual void help(stringstream& ss) const { - ss << "Returns information about users."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForUsersInfoCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - return grid.catalogManager()->runUserManagementReadCommand(dbname, - cmdObj, - &result); - } - - } cmdUsersInfo; - - class CmdCreateRole: public Command { - public: - - CmdCreateRole() : Command("createRole") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Adds a role to the system"; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForCreateRoleCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - return grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - } - - } cmdCreateRole; - - class CmdUpdateRole: public Command { - public: - - CmdUpdateRole() : Command("updateRole") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Used to update a role"; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForUpdateRoleCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserCache(); - - return ok; - } - - } cmdUpdateRole; - - class CmdGrantPrivilegesToRole: public Command { - public: - - CmdGrantPrivilegesToRole() : Command("grantPrivilegesToRole") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Grants privileges to a role"; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForGrantPrivilegesToRoleCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserCache(); - - return ok; - } - - } cmdGrantPrivilegesToRole; - - class CmdRevokePrivilegesFromRole: public Command { - public: - - CmdRevokePrivilegesFromRole() : Command("revokePrivilegesFromRole") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Revokes privileges from a role"; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForRevokePrivilegesFromRoleCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserCache(); - - return ok; - } - - } cmdRevokePrivilegesFromRole; - - class CmdGrantRolesToRole: public Command { - public: - - CmdGrantRolesToRole() : Command("grantRolesToRole") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Grants roles to another role."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForGrantRolesToRoleCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserCache(); - - return ok; - } - - } cmdGrantRolesToRole; - - class CmdRevokeRolesFromRole: public Command { - public: - - CmdRevokeRolesFromRole() : Command("revokeRolesFromRole") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Revokes roles from another role."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForRevokeRolesFromRoleCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserCache(); - - return ok; - } - - } cmdRevokeRolesFromRole; - - class CmdDropRole: public Command { - public: - - CmdDropRole() : Command("dropRole") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Drops a single role. Before deleting the role completely it must remove it " - "from any users or roles that reference it. If any errors occur in the middle " - "of that process it's possible to be left in a state where the role has been " - "removed from some user/roles but otherwise still exists."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForDropRoleCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserCache(); - - return ok; - } - - } cmdDropRole; - - class CmdDropAllRolesFromDatabase: public Command { - public: - - CmdDropAllRolesFromDatabase() : Command("dropAllRolesFromDatabase") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Drops all roles from the given database. Before deleting the roles completely " - "it must remove them from any users or other roles that reference them. If any " - "errors occur in the middle of that process it's possible to be left in a state " - "where the roles have been removed from some user/roles but otherwise still " - "exist."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForDropAllRolesFromDatabaseCommand(client, dbname); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const bool ok = grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserCache(); - - return ok; - } - - } cmdDropAllRolesFromDatabase; - - class CmdRolesInfo: public Command { - public: - - CmdRolesInfo() : Command("rolesInfo") {} - - virtual bool slaveOk() const { return false; } - - virtual bool slaveOverrideOk() const { return true; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Returns information about roles."; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForRolesInfoCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - return grid.catalogManager()->runUserManagementReadCommand(dbname, - cmdObj, - &result); - } - - } cmdRolesInfo; - - class CmdInvalidateUserCache: public Command { - public: - - CmdInvalidateUserCache() : Command("invalidateUserCache") {} - - virtual bool slaveOk() const { return true; } - - virtual bool adminOnly() const { return true; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void help(stringstream& ss) const { - ss << "Invalidates the in-memory cache of user information"; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForInvalidateUserCacheCommand(client); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - AuthorizationManager* authzManager = getGlobalAuthorizationManager(); - invariant(authzManager); - authzManager->invalidateUserCache(); - return true; - } - - } cmdInvalidateUserCache; - - /** - * This command is used only by mongorestore to handle restoring users/roles. We do this so - * that mongorestore doesn't do direct inserts into the admin.system.users and - * admin.system.roles, which would bypass the authzUpdateLock and allow multiple concurrent - * modifications to users/roles. What mongorestore now does instead is it inserts all user/role - * definitions it wants to restore into temporary collections, then this command moves those - * user/role definitions into their proper place in admin.system.users and admin.system.roles. - * It either adds the users/roles to the existing ones or replaces the existing ones, depending - * on whether the "drop" argument is true or false. - */ - class CmdMergeAuthzCollections : public Command { - public: - - CmdMergeAuthzCollections() : Command("_mergeAuthzCollections") {} - - virtual bool slaveOk() const { return false; } - - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual bool adminOnly() const { return true; } - - virtual void help(stringstream& ss) const { - ss << "Internal command used by mongorestore for updating user/role data"; - } - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return auth::checkAuthForMergeAuthzCollectionsCommand(client, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - return grid.catalogManager()->runUserManagementWriteCommand(this->name, - dbname, - cmdObj, - &result); - } - - } cmdMergeAuthzCollections; - -} // namespace mongo +/** + * This command is used only by mongorestore to handle restoring users/roles. We do this so + * that mongorestore doesn't do direct inserts into the admin.system.users and + * admin.system.roles, which would bypass the authzUpdateLock and allow multiple concurrent + * modifications to users/roles. What mongorestore now does instead is it inserts all user/role + * definitions it wants to restore into temporary collections, then this command moves those + * user/role definitions into their proper place in admin.system.users and admin.system.roles. + * It either adds the users/roles to the existing ones or replaces the existing ones, depending + * on whether the "drop" argument is true or false. + */ +class CmdMergeAuthzCollections : public Command { +public: + CmdMergeAuthzCollections() : Command("_mergeAuthzCollections") {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual bool adminOnly() const { + return true; + } + + virtual void help(stringstream& ss) const { + ss << "Internal command used by mongorestore for updating user/role data"; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return auth::checkAuthForMergeAuthzCollectionsCommand(client, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + return grid.catalogManager()->runUserManagementWriteCommand( + this->name, dbname, cmdObj, &result); + } + +} cmdMergeAuthzCollections; + +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_whats_my_uri_cmd.cpp b/src/mongo/s/commands/cluster_whats_my_uri_cmd.cpp index ad4c957e598..6b10eb476e6 100644 --- a/src/mongo/s/commands/cluster_whats_my_uri_cmd.cpp +++ b/src/mongo/s/commands/cluster_whats_my_uri_cmd.cpp @@ -35,41 +35,39 @@ namespace mongo { namespace { - class WhatsMyUriCmd : public Command { - public: - WhatsMyUriCmd() : Command("whatsmyuri") { } +class WhatsMyUriCmd : public Command { +public: + WhatsMyUriCmd() : Command("whatsmyuri") {} - virtual bool slaveOk() const { - return true; - } + virtual bool slaveOk() const { + return true; + } - virtual bool isWriteCommandForConfigServer() const { - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - virtual void help(std::stringstream &help) const { - help << "{whatsmyuri:1}"; - } + virtual void help(std::stringstream& help) const { + help << "{whatsmyuri:1}"; + } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + // No auth required + } - // No auth required - } + virtual bool run(OperationContext* txn, + const std::string& dbname, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& result) { + result << "you" << cc().getRemote().toString(); + return true; + } - virtual bool run(OperationContext* txn, - const std::string& dbname, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& result) { +} whatsMyUriCmd; - result << "you" << cc().getRemote().toString(); - return true; - } - - } whatsMyUriCmd; - -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/cluster_write_cmd.cpp b/src/mongo/s/commands/cluster_write_cmd.cpp index a22dd2c7dc7..e499d708ec3 100644 --- a/src/mongo/s/commands/cluster_write_cmd.cpp +++ b/src/mongo/s/commands/cluster_write_cmd.cpp @@ -27,7 +27,7 @@ */ #include "mongo/platform/basic.h" - + #include "mongo/base/error_codes.h" #include "mongo/db/client.h" #include "mongo/db/client_basic.h" @@ -46,234 +46,209 @@ namespace mongo { - using std::string; - using std::stringstream; - using std::vector; +using std::string; +using std::stringstream; +using std::vector; namespace { - /** - * Base class for mongos write commands. Cluster write commands support batch writes and write - * concern, and return per-item error information. All cluster write commands use the entry - * point ClusterWriteCmd::run(). - * - * Batch execution (targeting and dispatching) is performed by the BatchWriteExec class. - */ - class ClusterWriteCmd : public Command { - public: - virtual ~ClusterWriteCmd() { - - } - - virtual bool slaveOk() const { - return false; - } - - virtual bool isWriteCommandForConfigServer() const { - return false; +/** + * Base class for mongos write commands. Cluster write commands support batch writes and write + * concern, and return per-item error information. All cluster write commands use the entry + * point ClusterWriteCmd::run(). + * + * Batch execution (targeting and dispatching) is performed by the BatchWriteExec class. + */ +class ClusterWriteCmd : public Command { +public: + virtual ~ClusterWriteCmd() {} + + virtual bool slaveOk() const { + return false; + } + + virtual bool isWriteCommandForConfigServer() const { + return false; + } + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + Status status = auth::checkAuthForWriteCommand(AuthorizationSession::get(client), + _writeType, + NamespaceString(parseNs(dbname, cmdObj)), + cmdObj); + + // TODO: Remove this when we standardize GLE reporting from commands + if (!status.isOK()) { + LastError::get(client).setLastError(status.code(), status.reason()); } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { + return status; + } - Status status = auth::checkAuthForWriteCommand(AuthorizationSession::get(client), - _writeType, - NamespaceString(parseNs(dbname, - cmdObj)), - cmdObj); + virtual Status explain(OperationContext* txn, + const std::string& dbname, + const BSONObj& cmdObj, + ExplainCommon::Verbosity verbosity, + BSONObjBuilder* out) const { + BatchedCommandRequest request(_writeType); - // TODO: Remove this when we standardize GLE reporting from commands - if (!status.isOK()) { - LastError::get(client).setLastError(status.code(), status.reason()); - } - - return status; + string errMsg; + if (!request.parseBSON(cmdObj, &errMsg) || !request.isValid(&errMsg)) { + return Status(ErrorCodes::FailedToParse, errMsg); } - virtual Status explain(OperationContext* txn, - const std::string& dbname, - const BSONObj& cmdObj, - ExplainCommon::Verbosity verbosity, - BSONObjBuilder* out) const { + // Fixup the namespace to be a full ns internally + const NamespaceString nss(dbname, request.getNS()); + request.setNS(nss.ns()); - BatchedCommandRequest request(_writeType); + // We can only explain write batches of size 1. + if (request.sizeWriteOps() != 1U) { + return Status(ErrorCodes::InvalidLength, "explained write batches must be of size 1"); + } - string errMsg; - if (!request.parseBSON(cmdObj, &errMsg) || !request.isValid(&errMsg)) { - return Status(ErrorCodes::FailedToParse, errMsg); - } + BSONObjBuilder explainCmdBob; + ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmdBob); - // Fixup the namespace to be a full ns internally - const NamespaceString nss(dbname, request.getNS()); - request.setNS(nss.ns()); + // We will time how long it takes to run the commands on the shards. + Timer timer; - // We can only explain write batches of size 1. - if (request.sizeWriteOps() != 1U) { - return Status(ErrorCodes::InvalidLength, - "explained write batches must be of size 1"); - } + // Target the command to the shards based on the singleton batch item. + BatchItemRef targetingBatchItem(&request, 0); + vector<Strategy::CommandResult> shardResults; + Status status = Strategy::commandOpWrite( + dbname, explainCmdBob.obj(), targetingBatchItem, &shardResults); + if (!status.isOK()) { + return status; + } - BSONObjBuilder explainCmdBob; - ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmdBob); - - // We will time how long it takes to run the commands on the shards. - Timer timer; - - // Target the command to the shards based on the singleton batch item. - BatchItemRef targetingBatchItem(&request, 0); - vector<Strategy::CommandResult> shardResults; - Status status = Strategy::commandOpWrite(dbname, - explainCmdBob.obj(), - targetingBatchItem, - &shardResults); - if (!status.isOK()) { - return status; + return ClusterExplain::buildExplainResult( + shardResults, ClusterExplain::kWriteOnShards, timer.millis(), out); + } + + virtual bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + BatchedCommandRequest request(_writeType); + BatchedCommandResponse response; + + ClusterWriter writer(true, 0); + + LastError* cmdLastError = &LastError::get(cc()); + + { + // Disable the last error object for the duration of the write + LastError::Disabled disableLastError(cmdLastError); + + // TODO: if we do namespace parsing, push this to the type + if (!request.parseBSON(cmdObj, &errmsg) || !request.isValid(&errmsg)) { + // Batch parse failure + response.setOk(false); + response.setErrCode(ErrorCodes::FailedToParse); + response.setErrMessage(errmsg); + } else { + // Fixup the namespace to be a full ns internally + const NamespaceString nss(dbname, request.getNS()); + request.setNSS(nss); + + writer.write(request, &response); } - return ClusterExplain::buildExplainResult(shardResults, - ClusterExplain::kWriteOnShards, - timer.millis(), - out); + dassert(response.isValid(NULL)); } - virtual bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - - BatchedCommandRequest request(_writeType); - BatchedCommandResponse response; - - ClusterWriter writer(true, 0); - - LastError* cmdLastError = &LastError::get(cc()); - - { - // Disable the last error object for the duration of the write - LastError::Disabled disableLastError(cmdLastError); + // Populate the lastError object based on the write response + cmdLastError->reset(); + batchErrorToLastError(request, response, cmdLastError); - // TODO: if we do namespace parsing, push this to the type - if (!request.parseBSON(cmdObj, &errmsg) || !request.isValid(&errmsg)) { - // Batch parse failure - response.setOk(false); - response.setErrCode(ErrorCodes::FailedToParse); - response.setErrMessage(errmsg); - } - else { - // Fixup the namespace to be a full ns internally - const NamespaceString nss(dbname, request.getNS()); - request.setNSS(nss); + size_t numAttempts; - writer.write(request, &response); - } - - dassert(response.isValid(NULL)); - } - - // Populate the lastError object based on the write response - cmdLastError->reset(); - batchErrorToLastError(request, response, cmdLastError); - - size_t numAttempts; - - if (!response.getOk()) { - numAttempts = 0; - } - else if (request.getOrdered() && response.isErrDetailsSet()) { - // Add one failed attempt - numAttempts = response.getErrDetailsAt(0)->getIndex() + 1; - } - else { - numAttempts = request.sizeWriteOps(); - } + if (!response.getOk()) { + numAttempts = 0; + } else if (request.getOrdered() && response.isErrDetailsSet()) { + // Add one failed attempt + numAttempts = response.getErrDetailsAt(0)->getIndex() + 1; + } else { + numAttempts = request.sizeWriteOps(); + } - // TODO: increase opcounters by more than one - if (_writeType == BatchedCommandRequest::BatchType_Insert) { - for (size_t i = 0; i < numAttempts; ++i) { - globalOpCounters.gotInsert(); - } + // TODO: increase opcounters by more than one + if (_writeType == BatchedCommandRequest::BatchType_Insert) { + for (size_t i = 0; i < numAttempts; ++i) { + globalOpCounters.gotInsert(); } - else if (_writeType == BatchedCommandRequest::BatchType_Update) { - for (size_t i = 0; i < numAttempts; ++i) { - globalOpCounters.gotUpdate(); - } + } else if (_writeType == BatchedCommandRequest::BatchType_Update) { + for (size_t i = 0; i < numAttempts; ++i) { + globalOpCounters.gotUpdate(); } - else if (_writeType == BatchedCommandRequest::BatchType_Delete) { - for (size_t i = 0; i < numAttempts; ++i) { - globalOpCounters.gotDelete(); - } - } - - // Save the last opTimes written on each shard for this client, to allow GLE to work - if (haveClient() && writer.getStats().hasShardStats()) { - ClusterLastErrorInfo::get(cc()).addHostOpTimes( - writer.getStats().getShardStats().getWriteOpTimes()); + } else if (_writeType == BatchedCommandRequest::BatchType_Delete) { + for (size_t i = 0; i < numAttempts; ++i) { + globalOpCounters.gotDelete(); } - - // TODO - // There's a pending issue about how to report response here. If we use - // the command infra-structure, we should reuse the 'errmsg' field. But - // we have already filed that message inside the BatchCommandResponse. - // return response.getOk(); - result.appendElements(response.toBSON()); - return true; } - protected: - /** - * Instantiates a command that can be invoked by "name", which will be capable of issuing - * write batches of type "writeType", and will require privilege "action" to run. - */ - ClusterWriteCmd(StringData name, BatchedCommandRequest::BatchType writeType) - : Command(name), - _writeType(writeType) { - + // Save the last opTimes written on each shard for this client, to allow GLE to work + if (haveClient() && writer.getStats().hasShardStats()) { + ClusterLastErrorInfo::get(cc()) + .addHostOpTimes(writer.getStats().getShardStats().getWriteOpTimes()); } - private: - // Type of batch (e.g. insert, update). - const BatchedCommandRequest::BatchType _writeType; - }; + // TODO + // There's a pending issue about how to report response here. If we use + // the command infra-structure, we should reuse the 'errmsg' field. But + // we have already filed that message inside the BatchCommandResponse. + // return response.getOk(); + result.appendElements(response.toBSON()); + return true; + } +protected: + /** + * Instantiates a command that can be invoked by "name", which will be capable of issuing + * write batches of type "writeType", and will require privilege "action" to run. + */ + ClusterWriteCmd(StringData name, BatchedCommandRequest::BatchType writeType) + : Command(name), _writeType(writeType) {} - class ClusterCmdInsert : public ClusterWriteCmd { - public: - ClusterCmdInsert() : ClusterWriteCmd("insert", BatchedCommandRequest::BatchType_Insert) { +private: + // Type of batch (e.g. insert, update). + const BatchedCommandRequest::BatchType _writeType; +}; - } - void help(stringstream& help) const { - help << "insert documents"; - } +class ClusterCmdInsert : public ClusterWriteCmd { +public: + ClusterCmdInsert() : ClusterWriteCmd("insert", BatchedCommandRequest::BatchType_Insert) {} - } clusterInsertCmd; + void help(stringstream& help) const { + help << "insert documents"; + } - class ClusterCmdUpdate : public ClusterWriteCmd { - public: - ClusterCmdUpdate() : ClusterWriteCmd("update", BatchedCommandRequest::BatchType_Update) { +} clusterInsertCmd; - } +class ClusterCmdUpdate : public ClusterWriteCmd { +public: + ClusterCmdUpdate() : ClusterWriteCmd("update", BatchedCommandRequest::BatchType_Update) {} - void help( stringstream& help ) const { - help << "update documents"; - } + void help(stringstream& help) const { + help << "update documents"; + } - } clusterUpdateCmd; +} clusterUpdateCmd; - class ClusterCmdDelete : public ClusterWriteCmd { - public: - ClusterCmdDelete() : ClusterWriteCmd("delete", BatchedCommandRequest::BatchType_Delete) { +class ClusterCmdDelete : public ClusterWriteCmd { +public: + ClusterCmdDelete() : ClusterWriteCmd("delete", BatchedCommandRequest::BatchType_Delete) {} - } - - void help(stringstream& help) const { - help << "delete documents"; - } + void help(stringstream& help) const { + help << "delete documents"; + } - } clusterDeleteCmd; +} clusterDeleteCmd; -} // namespace -} // namespace mongo +} // namespace +} // namespace mongo diff --git a/src/mongo/s/commands/commands_public.cpp b/src/mongo/s/commands/commands_public.cpp index 6f0d57bfdd8..395f84c8c7e 100644 --- a/src/mongo/s/commands/commands_public.cpp +++ b/src/mongo/s/commands/commands_public.cpp @@ -64,1394 +64,1369 @@ namespace mongo { - using boost::intrusive_ptr; - using std::unique_ptr; - using std::shared_ptr; - using std::list; - using std::make_pair; - using std::map; - using std::multimap; - using std::set; - using std::string; - using std::stringstream; - using std::vector; - - namespace dbgrid_pub_cmds { - - class PublicGridCommand : public Command { - public: - PublicGridCommand( const char* n, const char* oldname=NULL ) : Command( n, false, oldname ) { - } - virtual bool slaveOk() const { - return true; - } - virtual bool adminOnly() const { - return false; - } - - // Override if passthrough should also send query options - // Safer as off by default, can slowly enable as we add more tests - virtual bool passOptions() const { return false; } - - // all grid commands are designed not to lock - virtual bool isWriteCommandForConfigServer() const { return false; } - - protected: - - bool passthrough( DBConfigPtr conf, const BSONObj& cmdObj , BSONObjBuilder& result ) { - return _passthrough(conf->name(), conf, cmdObj, 0, result); - } - - bool adminPassthrough( DBConfigPtr conf, const BSONObj& cmdObj , BSONObjBuilder& result ) { - return _passthrough("admin", conf, cmdObj, 0, result); - } - - bool passthrough( DBConfigPtr conf, const BSONObj& cmdObj , int options, BSONObjBuilder& result ) { - return _passthrough(conf->name(), conf, cmdObj, options, result); - } - - private: - bool _passthrough(const string& db, - DBConfigPtr conf, - const BSONObj& cmdObj, - int options, BSONObjBuilder& result) { - const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); - ShardConnection conn(shard->getConnString(), ""); - - BSONObj res; - bool ok = conn->runCommand(db, cmdObj, res, passOptions() ? options : 0); - conn.done(); - - result.appendElements(res); - return ok; +using boost::intrusive_ptr; +using std::unique_ptr; +using std::shared_ptr; +using std::list; +using std::make_pair; +using std::map; +using std::multimap; +using std::set; +using std::string; +using std::stringstream; +using std::vector; + +namespace dbgrid_pub_cmds { + +class PublicGridCommand : public Command { +public: + PublicGridCommand(const char* n, const char* oldname = NULL) : Command(n, false, oldname) {} + virtual bool slaveOk() const { + return true; + } + virtual bool adminOnly() const { + return false; + } + + // Override if passthrough should also send query options + // Safer as off by default, can slowly enable as we add more tests + virtual bool passOptions() const { + return false; + } + + // all grid commands are designed not to lock + virtual bool isWriteCommandForConfigServer() const { + return false; + } + +protected: + bool passthrough(DBConfigPtr conf, const BSONObj& cmdObj, BSONObjBuilder& result) { + return _passthrough(conf->name(), conf, cmdObj, 0, result); + } + + bool adminPassthrough(DBConfigPtr conf, const BSONObj& cmdObj, BSONObjBuilder& result) { + return _passthrough("admin", conf, cmdObj, 0, result); + } + + bool passthrough(DBConfigPtr conf, const BSONObj& cmdObj, int options, BSONObjBuilder& result) { + return _passthrough(conf->name(), conf, cmdObj, options, result); + } + +private: + bool _passthrough(const string& db, + DBConfigPtr conf, + const BSONObj& cmdObj, + int options, + BSONObjBuilder& result) { + const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); + ShardConnection conn(shard->getConnString(), ""); + + BSONObj res; + bool ok = conn->runCommand(db, cmdObj, res, passOptions() ? options : 0); + conn.done(); + + result.appendElements(res); + return ok; + } +}; + +class AllShardsCollectionCommand : public RunOnAllShardsCommand { +public: + AllShardsCollectionCommand(const char* n, + const char* oldname = NULL, + bool useShardConn = false, + bool implicitCreateDb = false) + : RunOnAllShardsCommand(n, oldname, useShardConn, implicitCreateDb) {} + + virtual void getShardIds(const string& dbName, BSONObj& cmdObj, vector<ShardId>& shardIds) { + const string fullns = dbName + '.' + cmdObj.firstElement().valuestrsafe(); + + auto status = grid.catalogCache()->getDatabase(dbName); + uassertStatusOK(status.getStatus()); + + shared_ptr<DBConfig> conf = status.getValue(); + + if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { + shardIds.push_back(conf->getShardId(fullns)); + } else { + grid.shardRegistry()->getAllShardIds(&shardIds); + } + } +}; + + +class NotAllowedOnShardedCollectionCmd : public PublicGridCommand { +public: + NotAllowedOnShardedCollectionCmd(const char* n) : PublicGridCommand(n) {} + + virtual bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const string fullns = parseNs(dbName, cmdObj); + + auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); + if (!conf->isSharded(fullns)) { + return passthrough(conf, cmdObj, options, result); + } + + return appendCommandStatus( + result, + Status(ErrorCodes::IllegalOperation, + str::stream() << "can't do command: " << name << " on sharded collection")); + } +}; + +// ---- + +class DropIndexesCmd : public AllShardsCollectionCommand { +public: + DropIndexesCmd() : AllShardsCollectionCommand("dropIndexes", "deleteIndexes") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::dropIndex); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } +} dropIndexesCmd; + +class CreateIndexesCmd : public AllShardsCollectionCommand { +public: + CreateIndexesCmd() + : AllShardsCollectionCommand("createIndexes", + NULL, /* oldName */ + true /* use ShardConnection */, + true /* implicit create db */) { + // createIndexes command should use ShardConnection so the getLastError would + // be able to properly enforce the write concern (via the saveGLEStats callback). + } + + /** + * the createIndexes command doesn't require the 'ns' field to be populated + * so we make sure its here as its needed for the system.indexes insert + */ + BSONObj fixSpec(const NamespaceString& ns, const BSONObj& original) const { + if (original["ns"].type() == String) + return original; + BSONObjBuilder bb; + bb.appendElements(original); + bb.append("ns", ns.toString()); + return bb.obj(); + } + + /** + * @return equivalent of gle + */ + BSONObj createIndexLegacy(const string& server, + const NamespaceString& nss, + const BSONObj& spec) const { + try { + ScopedDbConnection conn(server); + conn->insert(nss.getSystemIndexesCollection(), spec); + BSONObj gle = conn->getLastErrorDetailed(nss.db().toString()); + conn.done(); + return gle; + } catch (DBException& e) { + BSONObjBuilder b; + b.append("errmsg", e.toString()); + b.append("code", e.getCode()); + return b.obj(); + } + } + + virtual BSONObj specialErrorHandler(const string& server, + const string& dbName, + const BSONObj& cmdObj, + const BSONObj& originalResult) const { + string errmsg = originalResult["errmsg"]; + if (errmsg.find("no such cmd") == string::npos) { + // cannot use codes as 2.4 didn't have a code for this + return originalResult; + } + + // we need to down convert + + NamespaceString nss(dbName, cmdObj["createIndexes"].String()); + + if (cmdObj["indexes"].type() != Array) + return originalResult; + + BSONObjBuilder newResult; + newResult.append("note", "downgraded"); + newResult.append("sentTo", server); + + BSONArrayBuilder individualResults; + + bool ok = true; + + BSONObjIterator indexIterator(cmdObj["indexes"].Obj()); + while (indexIterator.more()) { + BSONObj spec = indexIterator.next().Obj(); + spec = fixSpec(nss, spec); + + BSONObj gle = createIndexLegacy(server, nss, spec); + + individualResults.append(BSON("spec" << spec << "gle" << gle)); + + BSONElement e = gle["errmsg"]; + if (e.type() == String && e.String().size() > 0) { + ok = false; + newResult.appendAs(e, "errmsg"); + break; } - }; - - class AllShardsCollectionCommand : public RunOnAllShardsCommand { - public: - AllShardsCollectionCommand(const char* n, - const char* oldname = NULL, - bool useShardConn = false, - bool implicitCreateDb = false) - : RunOnAllShardsCommand(n, oldname, useShardConn, implicitCreateDb) { + e = gle["err"]; + if (e.type() == String && e.String().size() > 0) { + ok = false; + newResult.appendAs(e, "errmsg"); + break; } - - virtual void getShardIds(const string& dbName, - BSONObj& cmdObj, - vector<ShardId>& shardIds) { - const string fullns = dbName + '.' + cmdObj.firstElement().valuestrsafe(); - - auto status = grid.catalogCache()->getDatabase(dbName); - uassertStatusOK(status.getStatus()); - - shared_ptr<DBConfig> conf = status.getValue(); - - if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { - shardIds.push_back(conf->getShardId(fullns)); + } + + newResult.append("eachIndex", individualResults.arr()); + + newResult.append("ok", ok ? 1 : 0); + return newResult.obj(); + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::createIndex); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + +} createIndexesCmd; + +class ReIndexCmd : public AllShardsCollectionCommand { +public: + ReIndexCmd() : AllShardsCollectionCommand("reIndex") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::reIndex); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } +} reIndexCmd; + +class CollectionModCmd : public AllShardsCollectionCommand { +public: + CollectionModCmd() : AllShardsCollectionCommand("collMod") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::collMod); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } +} collectionModCmd; + + +class ValidateCmd : public AllShardsCollectionCommand { +public: + ValidateCmd() : AllShardsCollectionCommand("validate") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::validate); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + virtual void aggregateResults(const vector<ShardAndReply>& results, BSONObjBuilder& output) { + for (vector<ShardAndReply>::const_iterator it(results.begin()), end(results.end()); + it != end; + it++) { + const BSONObj& result = std::get<1>(*it); + const BSONElement valid = result["valid"]; + if (!valid.eoo()) { + if (!valid.trueValue()) { + output.appendBool("valid", false); + return; } - else { - grid.shardRegistry()->getAllShardIds(&shardIds); + } else { + // Support pre-1.9.0 output with everything in a big string + const char* s = result["result"].valuestrsafe(); + if (strstr(s, "exception") || strstr(s, "corrupt")) { + output.appendBool("valid", false); + return; } } - }; - - - class NotAllowedOnShardedCollectionCmd : public PublicGridCommand { - public: - NotAllowedOnShardedCollectionCmd( const char * n ) : PublicGridCommand( n ) {} - - virtual bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - - const string fullns = parseNs(dbName, cmdObj); - - auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); - if (!conf->isSharded(fullns)) { - return passthrough( conf , cmdObj , options, result ); - } - - return appendCommandStatus(result, - Status(ErrorCodes::IllegalOperation, - str::stream() << "can't do command: " << name - << " on sharded collection")); - } - - }; - - // ---- - - class DropIndexesCmd : public AllShardsCollectionCommand { - public: - DropIndexesCmd() : AllShardsCollectionCommand("dropIndexes", "deleteIndexes") {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::dropIndex); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } - } dropIndexesCmd; - - class CreateIndexesCmd : public AllShardsCollectionCommand { - public: - CreateIndexesCmd(): - AllShardsCollectionCommand("createIndexes", - NULL, /* oldName */ - true /* use ShardConnection */, - true /* implicit create db */) { - // createIndexes command should use ShardConnection so the getLastError would - // be able to properly enforce the write concern (via the saveGLEStats callback). + } + + output.appendBool("valid", true); + } +} validateCmd; + +class CreateCmd : public PublicGridCommand { +public: + CreateCmd() : PublicGridCommand("create") {} + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + AuthorizationSession* authzSession = AuthorizationSession::get(client); + if (cmdObj["capped"].trueValue()) { + if (!authzSession->isAuthorizedForActionsOnResource( + parseResourcePattern(dbname, cmdObj), ActionType::convertToCapped)) { + return Status(ErrorCodes::Unauthorized, "unauthorized"); } - - /** - * the createIndexes command doesn't require the 'ns' field to be populated - * so we make sure its here as its needed for the system.indexes insert - */ - BSONObj fixSpec( const NamespaceString& ns, const BSONObj& original ) const { - if ( original["ns"].type() == String ) - return original; - BSONObjBuilder bb; - bb.appendElements( original ); - bb.append( "ns", ns.toString() ); - return bb.obj(); + } + + // ActionType::createCollection or ActionType::insert are both acceptable + if (authzSession->isAuthorizedForActionsOnResource(parseResourcePattern(dbname, cmdObj), + ActionType::createCollection) || + authzSession->isAuthorizedForActionsOnResource(parseResourcePattern(dbname, cmdObj), + ActionType::insert)) { + return Status::OK(); + } + + return Status(ErrorCodes::Unauthorized, "unauthorized"); + } + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + auto status = grid.implicitCreateDb(dbName); + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); + } + + shared_ptr<DBConfig> conf = status.getValue(); + + return passthrough(conf, cmdObj, result); + } + +} createCmd; + +class DropCmd : public PublicGridCommand { +public: + DropCmd() : PublicGridCommand("drop") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::dropCollection); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + auto status = grid.catalogCache()->getDatabase(dbName); + if (!status.isOK()) { + if (status == ErrorCodes::DatabaseNotFound) { + return true; } - /** - * @return equivalent of gle - */ - BSONObj createIndexLegacy( const string& server, - const NamespaceString& nss, - const BSONObj& spec ) const { - try { - ScopedDbConnection conn( server ); - conn->insert( nss.getSystemIndexesCollection(), spec ); - BSONObj gle = conn->getLastErrorDetailed( nss.db().toString() ); - conn.done(); - return gle; - } - catch ( DBException& e ) { - BSONObjBuilder b; - b.append( "errmsg", e.toString() ); - b.append( "code", e.getCode() ); - return b.obj(); + return appendCommandStatus(result, status.getStatus()); + } + + shared_ptr<DBConfig> conf = status.getValue(); + + const string fullns = dbName + "." + cmdObj.firstElement().valuestrsafe(); + log() << "DROP: " << fullns; + + if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { + log() << "\tdrop going to do passthrough"; + return passthrough(conf, cmdObj, result); + } + + // + // TODO: There will be problems if we simultaneously shard and drop a collection + // + + ChunkManagerPtr cm; + ShardPtr primary; + conf->getChunkManagerOrPrimary(fullns, cm, primary); + + if (!cm) { + log() << "\tdrop going to do passthrough after re-check"; + return passthrough(conf, cmdObj, result); + } + + uassertStatusOK(grid.catalogManager()->dropCollection(fullns)); + + if (!conf->removeSharding(fullns)) { + warning() << "collection " << fullns + << " was reloaded as unsharded before drop completed" + << " during single drop"; + } + + return 1; + } +} dropCmd; + +class RenameCollectionCmd : public PublicGridCommand { +public: + RenameCollectionCmd() : PublicGridCommand("renameCollection") {} + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return rename_collection::checkAuthForRenameCollectionCommand(client, dbname, cmdObj); + } + virtual bool adminOnly() const { + return true; + } + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + const string fullnsFrom = cmdObj.firstElement().valuestrsafe(); + const string dbNameFrom = nsToDatabase(fullnsFrom); + auto confFrom = uassertStatusOK(grid.catalogCache()->getDatabase(dbNameFrom)); + + const string fullnsTo = cmdObj["to"].valuestrsafe(); + const string dbNameTo = nsToDatabase(fullnsTo); + auto confTo = uassertStatusOK(grid.catalogCache()->getDatabase(dbNameTo)); + + uassert(13138, "You can't rename a sharded collection", !confFrom->isSharded(fullnsFrom)); + uassert(13139, "You can't rename to a sharded collection", !confTo->isSharded(fullnsTo)); + + const ShardId& shardTo = confTo->getShardId(fullnsTo); + const ShardId& shardFrom = confFrom->getShardId(fullnsFrom); + + uassert(13137, + "Source and destination collections must be on same shard", + shardFrom == shardTo); + + return adminPassthrough(confFrom, cmdObj, result); + } +} renameCollectionCmd; + +class CopyDBCmd : public PublicGridCommand { +public: + CopyDBCmd() : PublicGridCommand("copydb") {} + virtual bool adminOnly() const { + return true; + } + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return copydb::checkAuthForCopydbCommand(client, dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + const string todb = cmdObj.getStringField("todb"); + uassert(ErrorCodes::EmptyFieldName, "missing todb argument", !todb.empty()); + uassert(ErrorCodes::InvalidNamespace, "invalid todb argument", nsIsDbOnly(todb)); + + auto confTo = uassertStatusOK(grid.implicitCreateDb(todb)); + uassert(ErrorCodes::IllegalOperation, + "cannot copy to a sharded database", + !confTo->isShardingEnabled()); + + const string fromhost = cmdObj.getStringField("fromhost"); + if (!fromhost.empty()) { + return adminPassthrough(confTo, cmdObj, result); + } else { + const string fromdb = cmdObj.getStringField("fromdb"); + uassert(13399, "need a fromdb argument", !fromdb.empty()); + + shared_ptr<DBConfig> confFrom = + uassertStatusOK(grid.catalogCache()->getDatabase(fromdb)); + + uassert(13400, "don't know where source DB is", confFrom); + uassert(13401, "cant copy from sharded DB", !confFrom->isShardingEnabled()); + + BSONObjBuilder b; + BSONForEach(e, cmdObj) { + if (strcmp(e.fieldName(), "fromhost") != 0) { + b.append(e); } } - virtual BSONObj specialErrorHandler( const string& server, - const string& dbName, - const BSONObj& cmdObj, - const BSONObj& originalResult ) const { - string errmsg = originalResult["errmsg"]; - if ( errmsg.find( "no such cmd" ) == string::npos ) { - // cannot use codes as 2.4 didn't have a code for this - return originalResult; - } - - // we need to down convert - - NamespaceString nss( dbName, cmdObj["createIndexes"].String() ); - - if ( cmdObj["indexes"].type() != Array ) - return originalResult; - - BSONObjBuilder newResult; - newResult.append( "note", "downgraded" ); - newResult.append( "sentTo", server ); - - BSONArrayBuilder individualResults; - - bool ok = true; - - BSONObjIterator indexIterator( cmdObj["indexes"].Obj() ); - while ( indexIterator.more() ) { - BSONObj spec = indexIterator.next().Obj(); - spec = fixSpec( nss, spec ); - - BSONObj gle = createIndexLegacy( server, nss, spec ); - - individualResults.append( BSON( "spec" << spec << - "gle" << gle ) ); - - BSONElement e = gle["errmsg"]; - if ( e.type() == String && e.String().size() > 0 ) { - ok = false; - newResult.appendAs( e, "errmsg" ); - break; - } + { + const auto& shard = grid.shardRegistry()->getShard(confFrom->getPrimaryId()); + b.append("fromhost", shard->getConnString().toString()); + } + BSONObj fixed = b.obj(); + + return adminPassthrough(confTo, fixed, result); + } + } + +} clusterCopyDBCmd; + +class CollectionStats : public PublicGridCommand { +public: + CollectionStats() : PublicGridCommand("collStats", "collstats") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::collStats); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + const string fullns = parseNs(dbName, cmdObj); + + auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); + if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { + result.appendBool("sharded", false); + result.append("primary", conf->getPrimaryId()); + + return passthrough(conf, cmdObj, result); + } + + result.appendBool("sharded", true); + + ChunkManagerPtr cm = conf->getChunkManager(fullns); + massert(12594, "how could chunk manager be null!", cm); + + BSONObjBuilder shardStats; + map<string, long long> counts; + map<string, long long> indexSizes; + /* + long long count=0; + long long size=0; + long long storageSize=0; + */ + int nindexes = 0; + bool warnedAboutIndexes = false; + + set<ShardId> shardIds; + cm->getAllShardIds(&shardIds); + + for (const ShardId& shardId : shardIds) { + const auto shard = grid.shardRegistry()->getShard(shardId); + if (!shard) { + continue; + } - e = gle["err"]; - if ( e.type() == String && e.String().size() > 0 ) { - ok = false; - newResult.appendAs( e, "errmsg" ); - break; + BSONObj res; + { + ScopedDbConnection conn(shard->getConnString()); + if (!conn->runCommand(dbName, cmdObj, res)) { + if (!res["code"].eoo()) { + result.append(res["code"]); } - + errmsg = "failed on shard: " + res.toString(); + return false; } - - newResult.append( "eachIndex", individualResults.arr() ); - - newResult.append( "ok", ok ? 1 : 0 ); - return newResult.obj(); - } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::createIndex); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + conn.done(); } - } createIndexesCmd; - - class ReIndexCmd : public AllShardsCollectionCommand { - public: - ReIndexCmd() : AllShardsCollectionCommand("reIndex") {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::reIndex); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } - } reIndexCmd; - - class CollectionModCmd : public AllShardsCollectionCommand { - public: - CollectionModCmd() : AllShardsCollectionCommand("collMod") {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::collMod); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } - } collectionModCmd; - - - class ValidateCmd : public AllShardsCollectionCommand { - public: - ValidateCmd() : AllShardsCollectionCommand("validate") {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::validate); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } - virtual void aggregateResults(const vector<ShardAndReply>& results, - BSONObjBuilder& output) { - - for (vector<ShardAndReply>::const_iterator it(results.begin()), end(results.end()); - it!=end; it++) { - const BSONObj& result = std::get<1>(*it); - const BSONElement valid = result["valid"]; - if (!valid.eoo()){ - if (!valid.trueValue()) { - output.appendBool("valid", false); - return; - } + BSONObjIterator j(res); + while (j.more()) { + BSONElement e = j.next(); + + if (str::equals(e.fieldName(), "ns") || str::equals(e.fieldName(), "ok") || + str::equals(e.fieldName(), "avgObjSize") || + str::equals(e.fieldName(), "lastExtentSize") || + str::equals(e.fieldName(), "paddingFactor")) { + continue; + } else if (str::equals(e.fieldName(), "count") || + str::equals(e.fieldName(), "size") || + str::equals(e.fieldName(), "storageSize") || + str::equals(e.fieldName(), "numExtents") || + str::equals(e.fieldName(), "totalIndexSize")) { + counts[e.fieldName()] += e.numberLong(); + } else if (str::equals(e.fieldName(), "indexSizes")) { + BSONObjIterator k(e.Obj()); + while (k.more()) { + BSONElement temp = k.next(); + indexSizes[temp.fieldName()] += temp.numberLong(); } - else { - // Support pre-1.9.0 output with everything in a big string - const char* s = result["result"].valuestrsafe(); - if (strstr(s, "exception") || strstr(s, "corrupt")){ - output.appendBool("valid", false); - return; + } + // no longer used since 2.2 + else if (str::equals(e.fieldName(), "flags")) { + if (!result.hasField(e.fieldName())) + result.append(e); + } + // flags broken out in 2.4+ + else if (str::equals(e.fieldName(), "systemFlags")) { + if (!result.hasField(e.fieldName())) + result.append(e); + } else if (str::equals(e.fieldName(), "userFlags")) { + if (!result.hasField(e.fieldName())) + result.append(e); + } else if (str::equals(e.fieldName(), "capped")) { + if (!result.hasField(e.fieldName())) + result.append(e); + } else if (str::equals(e.fieldName(), "paddingFactorNote")) { + if (!result.hasField(e.fieldName())) + result.append(e); + } else if (str::equals(e.fieldName(), "indexDetails")) { + // skip this field in the rollup + } else if (str::equals(e.fieldName(), "wiredTiger")) { + // skip this field in the rollup + } else if (str::equals(e.fieldName(), "nindexes")) { + int myIndexes = e.numberInt(); + + if (nindexes == 0) { + nindexes = myIndexes; + } else if (nindexes == myIndexes) { + // no-op + } else { + // hopefully this means we're building an index + + if (myIndexes > nindexes) + nindexes = myIndexes; + + if (!warnedAboutIndexes) { + result.append("warning", + "indexes don't all match - ok if ensureIndex is running"); + warnedAboutIndexes = true; } } + } else { + warning() << "mongos collstats doesn't know about: " << e.fieldName(); } - - output.appendBool("valid", true); } - } validateCmd; - - class CreateCmd : public PublicGridCommand { - public: - CreateCmd() : PublicGridCommand( "create" ) {} - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - AuthorizationSession* authzSession = AuthorizationSession::get(client); - if (cmdObj["capped"].trueValue()) { - if (!authzSession->isAuthorizedForActionsOnResource( - parseResourcePattern(dbname, cmdObj), ActionType::convertToCapped)) { - return Status(ErrorCodes::Unauthorized, "unauthorized"); - } - } - - // ActionType::createCollection or ActionType::insert are both acceptable - if (authzSession->isAuthorizedForActionsOnResource( - parseResourcePattern(dbname, cmdObj), ActionType::createCollection) || - authzSession->isAuthorizedForActionsOnResource( - parseResourcePattern(dbname, cmdObj), ActionType::insert)) { - return Status::OK(); - } - - return Status(ErrorCodes::Unauthorized, "unauthorized"); + shardStats.append(shardId, res); + } + + result.append("ns", fullns); + + for (map<string, long long>::iterator i = counts.begin(); i != counts.end(); ++i) + result.appendNumber(i->first, i->second); + + { + BSONObjBuilder ib(result.subobjStart("indexSizes")); + for (map<string, long long>::iterator i = indexSizes.begin(); i != indexSizes.end(); + ++i) + ib.appendNumber(i->first, i->second); + ib.done(); + } + + if (counts["count"] > 0) + result.append("avgObjSize", (double)counts["size"] / (double)counts["count"]); + else + result.append("avgObjSize", 0.0); + + result.append("nindexes", nindexes); + + result.append("nchunks", cm->numChunks()); + result.append("shards", shardStats.obj()); + + return true; + } +} collectionStatsCmd; + +class DataSizeCmd : public PublicGridCommand { +public: + DataSizeCmd() : PublicGridCommand("dataSize", "datasize") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::find); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + const string fullns = parseNs(dbName, cmdObj); + + auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); + if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { + return passthrough(conf, cmdObj, result); + } + + ChunkManagerPtr cm = conf->getChunkManager(fullns); + massert(13407, "how could chunk manager be null!", cm); + + BSONObj min = cmdObj.getObjectField("min"); + BSONObj max = cmdObj.getObjectField("max"); + BSONObj keyPattern = cmdObj.getObjectField("keyPattern"); + + uassert(13408, + "keyPattern must equal shard key", + cm->getShardKeyPattern().toBSON() == keyPattern); + uassert(13405, + str::stream() << "min value " << min << " does not have shard key", + cm->getShardKeyPattern().isShardKey(min)); + uassert(13406, + str::stream() << "max value " << max << " does not have shard key", + cm->getShardKeyPattern().isShardKey(max)); + + min = cm->getShardKeyPattern().normalizeShardKey(min); + max = cm->getShardKeyPattern().normalizeShardKey(max); + + // yes these are doubles... + double size = 0; + double numObjects = 0; + int millis = 0; + + set<ShardId> shardIds; + cm->getShardIdsForRange(shardIds, min, max); + for (const ShardId& shardId : shardIds) { + const auto shard = grid.shardRegistry()->getShard(shardId); + if (!shard) { + continue; } - bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - auto status = grid.implicitCreateDb(dbName); - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } - - shared_ptr<DBConfig> conf = status.getValue(); + ScopedDbConnection conn(shard->getConnString()); + BSONObj res; + bool ok = conn->runCommand(conf->name(), cmdObj, res); + conn.done(); - return passthrough(conf, cmdObj, result); - } - - } createCmd; - - class DropCmd : public PublicGridCommand { - public: - DropCmd() : PublicGridCommand( "drop" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::dropCollection); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + if (!ok) { + result.appendElements(res); + return false; } - bool run(OperationContext* txn, + size += res["size"].number(); + numObjects += res["numObjects"].number(); + millis += res["millis"].numberInt(); + } + + result.append("size", size); + result.append("numObjects", numObjects); + result.append("millis", millis); + return true; + } + +} DataSizeCmd; + +class ConvertToCappedCmd : public NotAllowedOnShardedCollectionCmd { +public: + ConvertToCappedCmd() : NotAllowedOnShardedCollectionCmd("convertToCapped") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::convertToCapped); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + +} convertToCappedCmd; + + +class GroupCmd : public NotAllowedOnShardedCollectionCmd { +public: + GroupCmd() : NotAllowedOnShardedCollectionCmd("group") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::find); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + + virtual bool passOptions() const { + return true; + } + + virtual std::string parseNs(const std::string& dbName, const BSONObj& cmdObj) const { + return dbName + "." + cmdObj.firstElement().embeddedObjectUserCheck()["ns"].valuestrsafe(); + } + + Status explain(OperationContext* txn, + const std::string& dbname, + const BSONObj& cmdObj, + ExplainCommon::Verbosity verbosity, + BSONObjBuilder* out) const { + const string fullns = parseNs(dbname, cmdObj); + + BSONObjBuilder explainCmdBob; + ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmdBob); + + // We will time how long it takes to run the commands on the shards. + Timer timer; + + Strategy::CommandResult singleResult; + Status commandStat = + Strategy::commandOpUnsharded(dbname, explainCmdBob.obj(), 0, fullns, &singleResult); + if (!commandStat.isOK()) { + return commandStat; + } + + long long millisElapsed = timer.millis(); + + vector<Strategy::CommandResult> shardResults; + shardResults.push_back(singleResult); + + return ClusterExplain::buildExplainResult( + shardResults, ClusterExplain::kSingleShard, millisElapsed, out); + } + +} groupCmd; + +class SplitVectorCmd : public NotAllowedOnShardedCollectionCmd { +public: + SplitVectorCmd() : NotAllowedOnShardedCollectionCmd("splitVector") {} + virtual bool passOptions() const { + return true; + } + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::splitVector)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); + } + return Status::OK(); + } + virtual bool run(OperationContext* txn, const string& dbName, BSONObj& cmdObj, int options, string& errmsg, BSONObjBuilder& result) { - - auto status = grid.catalogCache()->getDatabase(dbName); - if (!status.isOK()) { - if (status == ErrorCodes::DatabaseNotFound) { - return true; - } - - return appendCommandStatus(result, status.getStatus()); - } - - shared_ptr<DBConfig> conf = status.getValue(); - - const string fullns = dbName + "." + cmdObj.firstElement().valuestrsafe(); - log() << "DROP: " << fullns; - - if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { - log() << "\tdrop going to do passthrough"; - return passthrough( conf , cmdObj , result ); - } - - // - // TODO: There will be problems if we simultaneously shard and drop a collection - // - - ChunkManagerPtr cm; - ShardPtr primary; - conf->getChunkManagerOrPrimary( fullns, cm, primary ); - - if( ! cm ) { - log() << "\tdrop going to do passthrough after re-check"; - return passthrough( conf , cmdObj , result ); - } - - uassertStatusOK(grid.catalogManager()->dropCollection(fullns)); - - if( ! conf->removeSharding( fullns ) ){ - warning() << "collection " << fullns - << " was reloaded as unsharded before drop completed" - << " during single drop"; - } - - return 1; - } - } dropCmd; - - class RenameCollectionCmd : public PublicGridCommand { - public: - RenameCollectionCmd() : PublicGridCommand( "renameCollection" ) {} - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return rename_collection::checkAuthForRenameCollectionCommand(client, - dbname, - cmdObj); + string x = parseNs(dbName, cmdObj); + if (!str::startsWith(x, dbName)) { + errmsg = str::stream() << "doing a splitVector across dbs isn't supported via mongos"; + return false; + } + return NotAllowedOnShardedCollectionCmd::run(txn, dbName, cmdObj, options, errmsg, result); + } + virtual std::string parseNs(const string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } + +} splitVectorCmd; + + +class DistinctCmd : public PublicGridCommand { +public: + DistinctCmd() : PublicGridCommand("distinct") {} + virtual void help(stringstream& help) const { + help << "{ distinct : 'collection name' , key : 'a.b' , query : {} }"; + } + virtual bool passOptions() const { + return true; + } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::find); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const string fullns = parseNs(dbName, cmdObj); + + auto status = grid.catalogCache()->getDatabase(dbName); + if (!status.isOK()) { + return appendEmptyResultSet(result, status.getStatus(), fullns); + } + + shared_ptr<DBConfig> conf = status.getValue(); + if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { + return passthrough(conf, cmdObj, options, result); + } + + ChunkManagerPtr cm = conf->getChunkManager(fullns); + massert(10420, "how could chunk manager be null!", cm); + + BSONObj query = getQuery(cmdObj); + set<ShardId> shardIds; + cm->getShardIdsForQuery(shardIds, query); + + set<BSONObj, BSONObjCmp> all; + int size = 32; + + for (const ShardId& shardId : shardIds) { + const auto shard = grid.shardRegistry()->getShard(shardId); + if (!shard) { + continue; } - virtual bool adminOnly() const { - return true; - } - bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - const string fullnsFrom = cmdObj.firstElement().valuestrsafe(); - const string dbNameFrom = nsToDatabase(fullnsFrom); - auto confFrom = uassertStatusOK(grid.catalogCache()->getDatabase(dbNameFrom)); - - const string fullnsTo = cmdObj["to"].valuestrsafe(); - const string dbNameTo = nsToDatabase(fullnsTo); - auto confTo = uassertStatusOK(grid.catalogCache()->getDatabase(dbNameTo)); - - uassert(13138, "You can't rename a sharded collection", !confFrom->isSharded(fullnsFrom)); - uassert(13139, "You can't rename to a sharded collection", !confTo->isSharded(fullnsTo)); - - const ShardId& shardTo = confTo->getShardId(fullnsTo); - const ShardId& shardFrom = confFrom->getShardId(fullnsFrom); - uassert(13137, - "Source and destination collections must be on same shard", - shardFrom == shardTo); + ShardConnection conn(shard->getConnString(), fullns); + BSONObj res; + bool ok = conn->runCommand(conf->name(), cmdObj, res, options); + conn.done(); - return adminPassthrough( confFrom , cmdObj , result ); + if (!ok) { + result.appendElements(res); + return false; } - } renameCollectionCmd; - class CopyDBCmd : public PublicGridCommand { - public: - CopyDBCmd() : PublicGridCommand( "copydb" ) {} - virtual bool adminOnly() const { - return true; + BSONObjIterator it(res["values"].embeddedObject()); + while (it.more()) { + BSONElement nxt = it.next(); + BSONObjBuilder temp(32); + temp.appendAs(nxt, ""); + all.insert(temp.obj()); } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return copydb::checkAuthForCopydbCommand(client, dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - - const string todb = cmdObj.getStringField("todb"); - uassert(ErrorCodes::EmptyFieldName, "missing todb argument", !todb.empty()); - uassert(ErrorCodes::InvalidNamespace, "invalid todb argument", nsIsDbOnly(todb)); - - auto confTo = uassertStatusOK(grid.implicitCreateDb(todb)); - uassert(ErrorCodes::IllegalOperation, - "cannot copy to a sharded database", - !confTo->isShardingEnabled()); - - const string fromhost = cmdObj.getStringField("fromhost"); - if (!fromhost.empty()) { - return adminPassthrough( confTo , cmdObj , result ); - } - else { - const string fromdb = cmdObj.getStringField("fromdb"); - uassert(13399, "need a fromdb argument", !fromdb.empty()); - - shared_ptr<DBConfig> confFrom = - uassertStatusOK(grid.catalogCache()->getDatabase(fromdb)); - - uassert(13400, "don't know where source DB is", confFrom); - uassert(13401, "cant copy from sharded DB", !confFrom->isShardingEnabled()); - - BSONObjBuilder b; - BSONForEach(e, cmdObj) { - if (strcmp(e.fieldName(), "fromhost") != 0) { - b.append(e); - } - } - - { - const auto& shard = - grid.shardRegistry()->getShard(confFrom->getPrimaryId()); - b.append("fromhost", shard->getConnString().toString()); - } - BSONObj fixed = b.obj(); - - return adminPassthrough( confTo , fixed , result ); + } + + BSONObjBuilder b(size); + int n = 0; + for (set<BSONObj, BSONObjCmp>::iterator i = all.begin(); i != all.end(); i++) { + b.appendAs(i->firstElement(), b.numStr(n++)); + } + + result.appendArray("values", b.obj()); + return true; + } +} disinctCmd; + +class FileMD5Cmd : public PublicGridCommand { +public: + FileMD5Cmd() : PublicGridCommand("filemd5") {} + virtual void help(stringstream& help) const { + help << " example: { filemd5 : ObjectId(aaaaaaa) , root : \"fs\" }"; + } + + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + std::string collectionName = cmdObj.getStringField("root"); + if (collectionName.empty()) + collectionName = "fs"; + collectionName += ".chunks"; + return NamespaceString(dbname, collectionName).ns(); + } + + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), ActionType::find)); + } + + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + const string fullns = parseNs(dbName, cmdObj); + + auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); + if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { + return passthrough(conf, cmdObj, result); + } + + ChunkManagerPtr cm = conf->getChunkManager(fullns); + massert(13091, "how could chunk manager be null!", cm); + if (cm->getShardKeyPattern().toBSON() == BSON("files_id" << 1)) { + BSONObj finder = BSON("files_id" << cmdObj.firstElement()); + + vector<Strategy::CommandResult> results; + Strategy::commandOp(dbName, cmdObj, 0, fullns, finder, &results); + verify(results.size() == 1); // querying on shard key so should only talk to one shard + BSONObj res = results.begin()->result; + + result.appendElements(res); + return res["ok"].trueValue(); + } else if (cm->getShardKeyPattern().toBSON() == BSON("files_id" << 1 << "n" << 1)) { + int n = 0; + BSONObj lastResult; + + while (true) { + // Theory of operation: Starting with n=0, send filemd5 command to shard + // with that chunk (gridfs chunk not sharding chunk). That shard will then + // compute a partial md5 state (passed in the "md5state" field) for all + // contiguous chunks that it has. When it runs out or hits a discontinuity + // (eg [1,2,7]) it returns what it has done so far. This is repeated as + // long as we keep getting more chunks. The end condition is when we go to + // look for chunk n and it doesn't exist. This means that the file's last + // chunk is n-1, so we return the computed md5 results. + BSONObjBuilder bb; + bb.appendElements(cmdObj); + bb.appendBool("partialOk", true); + bb.append("startAt", n); + if (!lastResult.isEmpty()) { + bb.append(lastResult["md5state"]); } + BSONObj shardCmd = bb.obj(); - } - - } clusterCopyDBCmd; - - class CollectionStats : public PublicGridCommand { - public: - CollectionStats() : PublicGridCommand("collStats", "collstats") { } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::collStats); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } + BSONObj finder = BSON("files_id" << cmdObj.firstElement() << "n" << n); - bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - const string fullns = parseNs(dbName, cmdObj); - - auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); - if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { - result.appendBool("sharded", false); - result.append("primary", conf->getPrimaryId()); - - return passthrough( conf , cmdObj , result); + vector<Strategy::CommandResult> results; + try { + Strategy::commandOp(dbName, shardCmd, 0, fullns, finder, &results); + } catch (DBException& e) { + // This is handled below and logged + Strategy::CommandResult errResult; + errResult.shardTargetId = ""; + errResult.result = BSON("errmsg" << e.what() << "ok" << 0); + results.push_back(errResult); } - result.appendBool("sharded", true); - - ChunkManagerPtr cm = conf->getChunkManager( fullns ); - massert( 12594 , "how could chunk manager be null!" , cm ); - - BSONObjBuilder shardStats; - map<string,long long> counts; - map<string,long long> indexSizes; - /* - long long count=0; - long long size=0; - long long storageSize=0; - */ - int nindexes=0; - bool warnedAboutIndexes = false; - - set<ShardId> shardIds; - cm->getAllShardIds(&shardIds); - - for (const ShardId& shardId : shardIds) { - const auto shard = grid.shardRegistry()->getShard(shardId); - if (!shard) { - continue; - } - - BSONObj res; - { - ScopedDbConnection conn(shard->getConnString()); - if ( ! conn->runCommand( dbName , cmdObj , res ) ) { - if ( !res["code"].eoo() ) { - result.append( res["code"] ); - } - errmsg = "failed on shard: " + res.toString(); - return false; - } - conn.done(); - } - - BSONObjIterator j( res ); - while ( j.more() ) { - BSONElement e = j.next(); - - if ( str::equals( e.fieldName() , "ns" ) || - str::equals( e.fieldName() , "ok" ) || - str::equals( e.fieldName() , "avgObjSize" ) || - str::equals( e.fieldName() , "lastExtentSize" ) || - str::equals( e.fieldName() , "paddingFactor" ) ) { - continue; - } - else if ( str::equals( e.fieldName() , "count" ) || - str::equals( e.fieldName() , "size" ) || - str::equals( e.fieldName() , "storageSize" ) || - str::equals( e.fieldName() , "numExtents" ) || - str::equals( e.fieldName() , "totalIndexSize" ) ) { - counts[e.fieldName()] += e.numberLong(); - } - else if ( str::equals( e.fieldName() , "indexSizes" ) ) { - BSONObjIterator k( e.Obj() ); - while ( k.more() ) { - BSONElement temp = k.next(); - indexSizes[temp.fieldName()] += temp.numberLong(); - } - } - // no longer used since 2.2 - else if ( str::equals( e.fieldName() , "flags" ) ) { - if ( ! result.hasField( e.fieldName() ) ) - result.append( e ); - } - // flags broken out in 2.4+ - else if ( str::equals( e.fieldName() , "systemFlags" ) ) { - if ( ! result.hasField( e.fieldName() ) ) - result.append( e ); - } - else if ( str::equals( e.fieldName() , "userFlags" ) ) { - if ( ! result.hasField( e.fieldName() ) ) - result.append( e ); - } - else if ( str::equals( e.fieldName() , "capped" ) ) { - if ( ! result.hasField( e.fieldName() ) ) - result.append( e ); - } - else if ( str::equals( e.fieldName() , "paddingFactorNote" ) ) { - if ( ! result.hasField( e.fieldName() ) ) - result.append( e ); - } - else if ( str::equals( e.fieldName() , "indexDetails" ) ) { - //skip this field in the rollup - } - else if ( str::equals( e.fieldName() , "wiredTiger" ) ) { - //skip this field in the rollup - } - else if ( str::equals( e.fieldName() , "nindexes" ) ) { - int myIndexes = e.numberInt(); - - if ( nindexes == 0 ) { - nindexes = myIndexes; - } - else if ( nindexes == myIndexes ) { - // no-op - } - else { - // hopefully this means we're building an index - - if ( myIndexes > nindexes ) - nindexes = myIndexes; - - if ( ! warnedAboutIndexes ) { - result.append( "warning" , "indexes don't all match - ok if ensureIndex is running" ); - warnedAboutIndexes = true; - } - } - } - else { - warning() << "mongos collstats doesn't know about: " << e.fieldName(); - } - + verify(results.size() == + 1); // querying on shard key so should only talk to one shard + BSONObj res = results.begin()->result; + bool ok = res["ok"].trueValue(); + + if (!ok) { + // Add extra info to make debugging easier + result.append("failedAt", n); + result.append("sentCommand", shardCmd); + BSONForEach(e, res) { + if (!str::equals(e.fieldName(), "errmsg")) + result.append(e); } - shardStats.append(shardId, res); - } - result.append("ns", fullns); - - for ( map<string,long long>::iterator i=counts.begin(); i!=counts.end(); ++i ) - result.appendNumber( i->first , i->second ); - - { - BSONObjBuilder ib( result.subobjStart( "indexSizes" ) ); - for ( map<string,long long>::iterator i=indexSizes.begin(); i!=indexSizes.end(); ++i ) - ib.appendNumber( i->first , i->second ); - ib.done(); - } + log() << "Sharded filemd5 failed: " << result.asTempObj(); - if ( counts["count"] > 0 ) - result.append("avgObjSize", (double)counts["size"] / (double)counts["count"] ); - else - result.append( "avgObjSize", 0.0 ); - - result.append("nindexes", nindexes); - - result.append("nchunks", cm->numChunks()); - result.append("shards", shardStats.obj()); - - return true; - } - } collectionStatsCmd; - - class DataSizeCmd : public PublicGridCommand { - public: - DataSizeCmd() : PublicGridCommand("dataSize", "datasize") { } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::find); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } - bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - const string fullns = parseNs(dbName, cmdObj); - - auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); - if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { - return passthrough( conf , cmdObj , result); + errmsg = + string("sharded filemd5 failed because: ") + res["errmsg"].valuestrsafe(); + return false; } - ChunkManagerPtr cm = conf->getChunkManager( fullns ); - massert( 13407 , "how could chunk manager be null!" , cm ); - - BSONObj min = cmdObj.getObjectField( "min" ); - BSONObj max = cmdObj.getObjectField( "max" ); - BSONObj keyPattern = cmdObj.getObjectField( "keyPattern" ); - - uassert( 13408, "keyPattern must equal shard key", - cm->getShardKeyPattern().toBSON() == keyPattern ); - uassert( 13405, str::stream() << "min value " << min << " does not have shard key", - cm->getShardKeyPattern().isShardKey(min) ); - uassert( 13406, str::stream() << "max value " << max << " does not have shard key", - cm->getShardKeyPattern().isShardKey(max) ); - - min = cm->getShardKeyPattern().normalizeShardKey(min); - max = cm->getShardKeyPattern().normalizeShardKey(max); - - // yes these are doubles... - double size = 0; - double numObjects = 0; - int millis = 0; - - set<ShardId> shardIds; - cm->getShardIdsForRange(shardIds, min, max); - for (const ShardId& shardId : shardIds) { - const auto shard = grid.shardRegistry()->getShard(shardId); - if (!shard) { - continue; - } + uassert(16246, + "Shard " + conf->name() + + " is too old to support GridFS sharded by {files_id:1, n:1}", + res.hasField("md5state")); - ScopedDbConnection conn(shard->getConnString()); - BSONObj res; - bool ok = conn->runCommand( conf->name() , cmdObj , res ); - conn.done(); - - if ( ! ok ) { - result.appendElements( res ); - return false; - } + lastResult = res; + int nNext = res["numChunks"].numberInt(); - size += res["size"].number(); - numObjects += res["numObjects"].number(); - millis += res["millis"].numberInt(); + if (n == nNext) { + // no new data means we've reached the end of the file + result.appendElements(res); + return true; } - result.append( "size", size ); - result.append( "numObjects" , numObjects ); - result.append( "millis" , millis ); - return true; - } - - } DataSizeCmd; - - class ConvertToCappedCmd : public NotAllowedOnShardedCollectionCmd { - public: - ConvertToCappedCmd() : NotAllowedOnShardedCollectionCmd("convertToCapped") {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::convertToCapped); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + verify(nNext > n); + n = nNext; } - } convertToCappedCmd; - - - class GroupCmd : public NotAllowedOnShardedCollectionCmd { - public: - GroupCmd() : NotAllowedOnShardedCollectionCmd("group") {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::find); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + verify(0); + } + + // We could support arbitrary shard keys by sending commands to all shards but I don't think we should + errmsg = + "GridFS fs.chunks collection must be sharded on either {files_id:1} or {files_id:1, " + "n:1}"; + return false; + } +} fileMD5Cmd; + +class Geo2dFindNearCmd : public PublicGridCommand { +public: + Geo2dFindNearCmd() : PublicGridCommand("geoNear") {} + void help(stringstream& h) const { + h << "http://dochub.mongodb.org/core/geo#GeospatialIndexing-geoNearCommand"; + } + virtual bool passOptions() const { + return true; + } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::find); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + const string fullns = parseNs(dbName, cmdObj); + + auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); + if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { + return passthrough(conf, cmdObj, options, result); + } + + ChunkManagerPtr cm = conf->getChunkManager(fullns); + massert(13500, "how could chunk manager be null!", cm); + + BSONObj query = getQuery(cmdObj); + set<ShardId> shardIds; + cm->getShardIdsForQuery(shardIds, query); + + // We support both "num" and "limit" options to control limit + int limit = 100; + const char* limitName = cmdObj["num"].isNumber() ? "num" : "limit"; + if (cmdObj[limitName].isNumber()) + limit = cmdObj[limitName].numberInt(); + + list<shared_ptr<Future::CommandResult>> futures; + BSONArrayBuilder shardArray; + for (const ShardId& shardId : shardIds) { + const auto shard = grid.shardRegistry()->getShard(shardId); + if (!shard) { + continue; } - virtual bool passOptions() const { return true; } - - virtual std::string parseNs(const std::string& dbName, const BSONObj& cmdObj) const { - return dbName + "." + cmdObj.firstElement() - .embeddedObjectUserCheck()["ns"] - .valuestrsafe(); - } - - Status explain(OperationContext* txn, - const std::string& dbname, - const BSONObj& cmdObj, - ExplainCommon::Verbosity verbosity, - BSONObjBuilder* out) const { - const string fullns = parseNs(dbname, cmdObj); - - BSONObjBuilder explainCmdBob; - ClusterExplain::wrapAsExplain(cmdObj, verbosity, &explainCmdBob); - - // We will time how long it takes to run the commands on the shards. - Timer timer; - - Strategy::CommandResult singleResult; - Status commandStat = Strategy::commandOpUnsharded(dbname, - explainCmdBob.obj(), - 0, - fullns, - &singleResult); - if (!commandStat.isOK()) { - return commandStat; + futures.push_back( + Future::spawnCommand(shard->getConnString().toString(), dbName, cmdObj, options)); + shardArray.append(shardId); + } + + multimap<double, BSONObj> results; // TODO: maybe use merge-sort instead + string nearStr; + double time = 0; + double btreelocs = 0; + double nscanned = 0; + double objectsLoaded = 0; + for (list<shared_ptr<Future::CommandResult>>::iterator i = futures.begin(); + i != futures.end(); + i++) { + shared_ptr<Future::CommandResult> res = *i; + if (!res->join()) { + errmsg = res->result()["errmsg"].String(); + if (res->result().hasField("code")) { + result.append(res->result()["code"]); } - - long long millisElapsed = timer.millis(); - - vector<Strategy::CommandResult> shardResults; - shardResults.push_back(singleResult); - - return ClusterExplain::buildExplainResult(shardResults, - ClusterExplain::kSingleShard, - millisElapsed, - out); + return false; } - } groupCmd; - - class SplitVectorCmd : public NotAllowedOnShardedCollectionCmd { - public: - SplitVectorCmd() : NotAllowedOnShardedCollectionCmd("splitVector") {} - virtual bool passOptions() const { return true; } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, - cmdObj))), - ActionType::splitVector)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - return Status::OK(); + if (res->result().hasField("near")) { + nearStr = res->result()["near"].String(); } - virtual bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - string x = parseNs(dbName, cmdObj); - if ( ! str::startsWith( x , dbName ) ) { - errmsg = str::stream() << "doing a splitVector across dbs isn't supported via mongos"; - return false; - } - return NotAllowedOnShardedCollectionCmd::run(txn, - dbName, - cmdObj, - options, - errmsg, - result); + time += res->result()["stats"]["time"].Number(); + if (!res->result()["stats"]["btreelocs"].eoo()) { + btreelocs += res->result()["stats"]["btreelocs"].Number(); } - virtual std::string parseNs(const string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); + nscanned += res->result()["stats"]["nscanned"].Number(); + if (!res->result()["stats"]["objectsLoaded"].eoo()) { + objectsLoaded += res->result()["stats"]["objectsLoaded"].Number(); } - } splitVectorCmd; - - - class DistinctCmd : public PublicGridCommand { - public: - DistinctCmd() : PublicGridCommand("distinct") {} - virtual void help( stringstream &help ) const { - help << "{ distinct : 'collection name' , key : 'a.b' , query : {} }"; - } - virtual bool passOptions() const { return true; } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::find); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + BSONForEach(obj, res->result()["results"].embeddedObject()) { + results.insert(make_pair(obj["dis"].Number(), obj.embeddedObject().getOwned())); } - bool run(OperationContext* txn, - const string& dbName , - BSONObj& cmdObj, - int options, - string& errmsg, - BSONObjBuilder& result) { - const string fullns = parseNs(dbName, cmdObj); - - auto status = grid.catalogCache()->getDatabase(dbName); - if (!status.isOK()) { - return appendEmptyResultSet(result, status.getStatus(), fullns); - } - - shared_ptr<DBConfig> conf = status.getValue(); - if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { - return passthrough(conf, cmdObj, options, result); - } - - ChunkManagerPtr cm = conf->getChunkManager( fullns ); - massert( 10420 , "how could chunk manager be null!" , cm ); - - BSONObj query = getQuery(cmdObj); - set<ShardId> shardIds; - cm->getShardIdsForQuery(shardIds, query); - - set<BSONObj,BSONObjCmp> all; - int size = 32; - - for (const ShardId& shardId : shardIds) { - const auto shard = grid.shardRegistry()->getShard(shardId); - if (!shard) { - continue; - } + // TODO: maybe shrink results if size() > limit + } - ShardConnection conn(shard->getConnString(), fullns); - BSONObj res; - bool ok = conn->runCommand( conf->name() , cmdObj , res, options ); - conn.done(); + result.append("ns", fullns); + result.append("near", nearStr); - if ( ! ok ) { - result.appendElements( res ); - return false; - } - - BSONObjIterator it( res["values"].embeddedObject() ); - while ( it.more() ) { - BSONElement nxt = it.next(); - BSONObjBuilder temp(32); - temp.appendAs( nxt , "" ); - all.insert( temp.obj() ); - } - - } - - BSONObjBuilder b( size ); - int n=0; - for ( set<BSONObj,BSONObjCmp>::iterator i = all.begin() ; i != all.end(); i++ ) { - b.appendAs( i->firstElement() , b.numStr( n++ ) ); - } - - result.appendArray( "values" , b.obj() ); - return true; - } - } disinctCmd; + int outCount = 0; + double totalDistance = 0; + double maxDistance = 0; + { + BSONArrayBuilder sub(result.subarrayStart("results")); + for (multimap<double, BSONObj>::const_iterator it(results.begin()), end(results.end()); + it != end && outCount < limit; + ++it, ++outCount) { + totalDistance += it->first; + maxDistance = it->first; // guaranteed to be highest so far - class FileMD5Cmd : public PublicGridCommand { - public: - FileMD5Cmd() : PublicGridCommand("filemd5") {} - virtual void help( stringstream &help ) const { - help << " example: { filemd5 : ObjectId(aaaaaaa) , root : \"fs\" }"; + sub.append(it->second); } - - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - std::string collectionName = cmdObj.getStringField("root"); - if (collectionName.empty()) - collectionName = "fs"; - collectionName += ".chunks"; - return NamespaceString(dbname, collectionName).ns(); - } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), ActionType::find)); - } - - bool run(OperationContext* txn, + sub.done(); + } + + { + BSONObjBuilder sub(result.subobjStart("stats")); + sub.append("time", time); + sub.append("btreelocs", btreelocs); + sub.append("nscanned", nscanned); + sub.append("objectsLoaded", objectsLoaded); + sub.append("avgDistance", (outCount == 0) ? 0 : (totalDistance / outCount)); + sub.append("maxDistance", maxDistance); + sub.append("shards", shardArray.arr()); + sub.done(); + } + + return true; + } +} geo2dFindNearCmd; + +class ApplyOpsCmd : public PublicGridCommand { +public: + ApplyOpsCmd() : PublicGridCommand("applyOps") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + // applyOps can do pretty much anything, so require all privileges. + RoleGraph::generateUniversalPrivileges(out); + } + virtual bool run(OperationContext* txn, const string& dbName, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result) { - const string fullns = parseNs(dbName, cmdObj); - - auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); - if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { - return passthrough( conf , cmdObj , result ); - } - - ChunkManagerPtr cm = conf->getChunkManager( fullns ); - massert( 13091 , "how could chunk manager be null!" , cm ); - if(cm->getShardKeyPattern().toBSON() == BSON("files_id" << 1)) { - BSONObj finder = BSON("files_id" << cmdObj.firstElement()); - - vector<Strategy::CommandResult> results; - Strategy::commandOp(dbName, cmdObj, 0, fullns, finder, &results); - verify(results.size() == 1); // querying on shard key so should only talk to one shard - BSONObj res = results.begin()->result; - - result.appendElements(res); - return res["ok"].trueValue(); - } - else if (cm->getShardKeyPattern().toBSON() == BSON("files_id" << 1 << "n" << 1)) { - int n = 0; - BSONObj lastResult; - - while (true) { - // Theory of operation: Starting with n=0, send filemd5 command to shard - // with that chunk (gridfs chunk not sharding chunk). That shard will then - // compute a partial md5 state (passed in the "md5state" field) for all - // contiguous chunks that it has. When it runs out or hits a discontinuity - // (eg [1,2,7]) it returns what it has done so far. This is repeated as - // long as we keep getting more chunks. The end condition is when we go to - // look for chunk n and it doesn't exist. This means that the file's last - // chunk is n-1, so we return the computed md5 results. - BSONObjBuilder bb; - bb.appendElements(cmdObj); - bb.appendBool("partialOk", true); - bb.append("startAt", n); - if (!lastResult.isEmpty()){ - bb.append(lastResult["md5state"]); - } - BSONObj shardCmd = bb.obj(); - - BSONObj finder = BSON("files_id" << cmdObj.firstElement() << "n" << n); - - vector<Strategy::CommandResult> results; - try { - Strategy::commandOp(dbName, shardCmd, 0, fullns, finder, &results); - } - catch( DBException& e ){ - //This is handled below and logged - Strategy::CommandResult errResult; - errResult.shardTargetId = ""; - errResult.result = BSON("errmsg" << e.what() << "ok" << 0 ); - results.push_back( errResult ); - } - - verify(results.size() == 1); // querying on shard key so should only talk to one shard - BSONObj res = results.begin()->result; - bool ok = res["ok"].trueValue(); - - if (!ok) { - // Add extra info to make debugging easier - result.append("failedAt", n); - result.append("sentCommand", shardCmd); - BSONForEach(e, res){ - if (!str::equals(e.fieldName(), "errmsg")) - result.append(e); - } - - log() << "Sharded filemd5 failed: " << result.asTempObj(); - - errmsg = string("sharded filemd5 failed because: ") + res["errmsg"].valuestrsafe(); - return false; - } - - uassert(16246, "Shard " + conf->name() + " is too old to support GridFS sharded by {files_id:1, n:1}", - res.hasField("md5state")); - - lastResult = res; - int nNext = res["numChunks"].numberInt(); - - if (n == nNext){ - // no new data means we've reached the end of the file - result.appendElements(res); - return true; - } - - verify(nNext > n); - n = nNext; - } - - verify(0); - } - - // We could support arbitrary shard keys by sending commands to all shards but I don't think we should - errmsg = "GridFS fs.chunks collection must be sharded on either {files_id:1} or {files_id:1, n:1}"; - return false; - } - } fileMD5Cmd; - - class Geo2dFindNearCmd : public PublicGridCommand { - public: - Geo2dFindNearCmd() : PublicGridCommand( "geoNear" ) {} - void help(stringstream& h) const { h << "http://dochub.mongodb.org/core/geo#GeospatialIndexing-geoNearCommand"; } - virtual bool passOptions() const { return true; } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::find); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } - - bool run(OperationContext* txn, + errmsg = "applyOps not allowed through mongos"; + return false; + } +} applyOpsCmd; + + +class CompactCmd : public PublicGridCommand { +public: + CompactCmd() : PublicGridCommand("compact") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::compact); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + virtual bool run(OperationContext* txn, const string& dbName, BSONObj& cmdObj, - int options, + int, string& errmsg, BSONObjBuilder& result) { - const string fullns = parseNs(dbName, cmdObj); - - auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); - if (!conf->isShardingEnabled() || !conf->isSharded(fullns)) { - return passthrough( conf , cmdObj , options, result ); - } - - ChunkManagerPtr cm = conf->getChunkManager( fullns ); - massert( 13500 , "how could chunk manager be null!" , cm ); - - BSONObj query = getQuery(cmdObj); - set<ShardId> shardIds; - cm->getShardIdsForQuery(shardIds, query); - - // We support both "num" and "limit" options to control limit - int limit = 100; - const char* limitName = cmdObj["num"].isNumber() ? "num" : "limit"; - if (cmdObj[limitName].isNumber()) - limit = cmdObj[limitName].numberInt(); - - list< shared_ptr<Future::CommandResult> > futures; - BSONArrayBuilder shardArray; - for (const ShardId& shardId : shardIds) { - const auto shard = grid.shardRegistry()->getShard(shardId); - if (!shard) { - continue; - } - - futures.push_back(Future::spawnCommand(shard->getConnString().toString(), - dbName, - cmdObj, - options)); - shardArray.append(shardId); - } - - multimap<double, BSONObj> results; // TODO: maybe use merge-sort instead - string nearStr; - double time = 0; - double btreelocs = 0; - double nscanned = 0; - double objectsLoaded = 0; - for ( list< shared_ptr<Future::CommandResult> >::iterator i=futures.begin(); i!=futures.end(); i++ ) { - shared_ptr<Future::CommandResult> res = *i; - if ( ! res->join() ) { - errmsg = res->result()["errmsg"].String(); - if (res->result().hasField("code")) { - result.append(res->result()["code"]); - } - return false; - } - - if (res->result().hasField("near")) { - nearStr = res->result()["near"].String(); - } - time += res->result()["stats"]["time"].Number(); - if (!res->result()["stats"]["btreelocs"].eoo()) { - btreelocs += res->result()["stats"]["btreelocs"].Number(); - } - nscanned += res->result()["stats"]["nscanned"].Number(); - if (!res->result()["stats"]["objectsLoaded"].eoo()) { - objectsLoaded += res->result()["stats"]["objectsLoaded"].Number(); - } - - BSONForEach(obj, res->result()["results"].embeddedObject()) { - results.insert(make_pair(obj["dis"].Number(), obj.embeddedObject().getOwned())); - } - - // TODO: maybe shrink results if size() > limit - } - - result.append("ns" , fullns); - result.append("near", nearStr); - - int outCount = 0; - double totalDistance = 0; - double maxDistance = 0; - { - BSONArrayBuilder sub (result.subarrayStart("results")); - for (multimap<double, BSONObj>::const_iterator it(results.begin()), end(results.end()); it!= end && outCount < limit; ++it, ++outCount) { - totalDistance += it->first; - maxDistance = it->first; // guaranteed to be highest so far - - sub.append(it->second); - } - sub.done(); - } - - { - BSONObjBuilder sub (result.subobjStart("stats")); - sub.append("time", time); - sub.append("btreelocs", btreelocs); - sub.append("nscanned", nscanned); - sub.append("objectsLoaded", objectsLoaded); - sub.append("avgDistance", (outCount == 0) ? 0: (totalDistance / outCount)); - sub.append("maxDistance", maxDistance); - sub.append("shards", shardArray.arr()); - sub.done(); - } - - return true; - } - } geo2dFindNearCmd; - - class ApplyOpsCmd : public PublicGridCommand { - public: - ApplyOpsCmd() : PublicGridCommand( "applyOps" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - // applyOps can do pretty much anything, so require all privileges. - RoleGraph::generateUniversalPrivileges(out); - } - virtual bool run(OperationContext* txn, const string& dbName , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result) { - errmsg = "applyOps not allowed through mongos"; - return false; - } - } applyOpsCmd; - - - class CompactCmd : public PublicGridCommand { - public: - CompactCmd() : PublicGridCommand( "compact" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::compact); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } - virtual bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - errmsg = "compact not allowed through mongos"; - return false; - } - } compactCmd; - - class EvalCmd : public PublicGridCommand { - public: - EvalCmd() : PublicGridCommand( "eval", "$eval" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - // $eval can do pretty much anything, so require all privileges. - RoleGraph::generateUniversalPrivileges(out); - } - virtual bool run(OperationContext* txn, - const string& dbName, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - - RARELY { - warning() << "the eval command is deprecated" << startupWarningsLog; - } - - // $eval isn't allowed to access sharded collections, but we need to leave the - // shard to detect that. - auto status = grid.catalogCache()->getDatabase(dbName); - if (!status.isOK()) { - return appendCommandStatus(result, status.getStatus()); - } - - shared_ptr<DBConfig> conf = status.getValue(); - return passthrough( conf , cmdObj , result ); - } - } evalCmd; - - class CmdListCollections : public PublicGridCommand { - public: - CmdListCollections() : PublicGridCommand( "listCollections" ) {} - - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - AuthorizationSession* authzSession = AuthorizationSession::get(client); - - // Check for the listCollections ActionType on the database - // or find on system.namespaces for pre 3.0 systems. - if (authzSession->isAuthorizedForActionsOnResource( - ResourcePattern::forDatabaseName(dbname), - ActionType::listCollections) || - authzSession->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace( - NamespaceString(dbname, "system.namespaces")), - ActionType::find)) { - return Status::OK(); - } - - return Status(ErrorCodes::Unauthorized, - str::stream() << "Not authorized to create users on db: " << - dbname); - } - - bool run(OperationContext* txn, + errmsg = "compact not allowed through mongos"; + return false; + } +} compactCmd; + +class EvalCmd : public PublicGridCommand { +public: + EvalCmd() : PublicGridCommand("eval", "$eval") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + // $eval can do pretty much anything, so require all privileges. + RoleGraph::generateUniversalPrivileges(out); + } + virtual bool run(OperationContext* txn, const string& dbName, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result) { - - auto status = grid.catalogCache()->getDatabase(dbName); - if (!status.isOK()) { - return appendEmptyResultSet(result, - status.getStatus(), - dbName + ".$cmd.listCollections"); - } - - shared_ptr<DBConfig> conf = status.getValue(); - bool retval = passthrough( conf, cmdObj, result ); - - const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); - Status storeCursorStatus = storePossibleCursor(shard->getConnString().toString(), - result.asTempObj()); - if (!storeCursorStatus.isOK()) { - return appendCommandStatus(result, storeCursorStatus); - } - - return retval; - } - } cmdListCollections; - - class CmdListIndexes : public PublicGridCommand { - public: - CmdListIndexes() : PublicGridCommand( "listIndexes" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - string ns = parseNs( dbname, cmdObj ); - ActionSet actions; - actions.addAction(ActionType::listIndexes); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); - } - - bool run(OperationContext* txn, - const string& dbName, + RARELY { + warning() << "the eval command is deprecated" << startupWarningsLog; + } + + // $eval isn't allowed to access sharded collections, but we need to leave the + // shard to detect that. + auto status = grid.catalogCache()->getDatabase(dbName); + if (!status.isOK()) { + return appendCommandStatus(result, status.getStatus()); + } + + shared_ptr<DBConfig> conf = status.getValue(); + return passthrough(conf, cmdObj, result); + } +} evalCmd; + +class CmdListCollections : public PublicGridCommand { +public: + CmdListCollections() : PublicGridCommand("listCollections") {} + + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + AuthorizationSession* authzSession = AuthorizationSession::get(client); + + // Check for the listCollections ActionType on the database + // or find on system.namespaces for pre 3.0 systems. + if (authzSession->isAuthorizedForActionsOnResource(ResourcePattern::forDatabaseName(dbname), + ActionType::listCollections) || + authzSession->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(dbname, "system.namespaces")), + ActionType::find)) { + return Status::OK(); + } + + return Status(ErrorCodes::Unauthorized, + str::stream() << "Not authorized to create users on db: " << dbname); + } + + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + auto status = grid.catalogCache()->getDatabase(dbName); + if (!status.isOK()) { + return appendEmptyResultSet( + result, status.getStatus(), dbName + ".$cmd.listCollections"); + } + + shared_ptr<DBConfig> conf = status.getValue(); + bool retval = passthrough(conf, cmdObj, result); + + const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); + Status storeCursorStatus = + storePossibleCursor(shard->getConnString().toString(), result.asTempObj()); + if (!storeCursorStatus.isOK()) { + return appendCommandStatus(result, storeCursorStatus); + } + + return retval; + } +} cmdListCollections; + +class CmdListIndexes : public PublicGridCommand { +public: + CmdListIndexes() : PublicGridCommand("listIndexes") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + string ns = parseNs(dbname, cmdObj); + ActionSet actions; + actions.addAction(ActionType::listIndexes); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + + bool run(OperationContext* txn, + const string& dbName, + BSONObj& cmdObj, + int options, + string& errmsg, + BSONObjBuilder& result) { + auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); + bool retval = passthrough(conf, cmdObj, result); + + const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); + Status storeCursorStatus = + storePossibleCursor(shard->getConnString().toString(), result.asTempObj()); + if (!storeCursorStatus.isOK()) { + return appendCommandStatus(result, storeCursorStatus); + } + + return retval; + } + +} cmdListIndexes; + +class AvailableQueryOptions : public Command { +public: + AvailableQueryOptions() : Command("availableQueryOptions", false, "availablequeryoptions") {} + + virtual bool slaveOk() const { + return true; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + return Status::OK(); + } + + + virtual bool run(OperationContext* txn, + const string& dbname, BSONObj& cmdObj, - int options, + int, string& errmsg, BSONObjBuilder& result) { + result << "options" << QueryOption_AllSupportedForSharding; + return true; + } +} availableQueryOptionsCmd; - auto conf = uassertStatusOK(grid.catalogCache()->getDatabase(dbName)); - bool retval = passthrough( conf, cmdObj, result ); - - const auto shard = grid.shardRegistry()->getShard(conf->getPrimaryId()); - Status storeCursorStatus = storePossibleCursor(shard->getConnString().toString(), - result.asTempObj()); - if (!storeCursorStatus.isOK()) { - return appendCommandStatus(result, storeCursorStatus); - } - - return retval; - } - - } cmdListIndexes; - - class AvailableQueryOptions : public Command { - public: - AvailableQueryOptions(): Command("availableQueryOptions", - false , - "availablequeryoptions") { - } - - virtual bool slaveOk() const { return true; } - virtual bool isWriteCommandForConfigServer() const { return false; } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - return Status::OK(); - } - - - virtual bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - result << "options" << QueryOption_AllSupportedForSharding; - return true; - } - } availableQueryOptionsCmd; - - } // namespace pub_grid_cmds - -} // namespace mongo +} // namespace pub_grid_cmds +} // namespace mongo diff --git a/src/mongo/s/commands/run_on_all_shards_cmd.cpp b/src/mongo/s/commands/run_on_all_shards_cmd.cpp index 68e9629227a..7c206b6a978 100644 --- a/src/mongo/s/commands/run_on_all_shards_cmd.cpp +++ b/src/mongo/s/commands/run_on_all_shards_cmd.cpp @@ -44,141 +44,128 @@ namespace mongo { - RunOnAllShardsCommand::RunOnAllShardsCommand(const char* name, - const char* oldName, - bool useShardConn, - bool implicitCreateDb) - : Command(name, false, oldName), - _useShardConn(useShardConn), - _implicitCreateDb(implicitCreateDb) { +RunOnAllShardsCommand::RunOnAllShardsCommand(const char* name, + const char* oldName, + bool useShardConn, + bool implicitCreateDb) + : Command(name, false, oldName), + _useShardConn(useShardConn), + _implicitCreateDb(implicitCreateDb) {} + +void RunOnAllShardsCommand::aggregateResults(const std::vector<ShardAndReply>& results, + BSONObjBuilder& output) {} + +BSONObj RunOnAllShardsCommand::specialErrorHandler(const std::string& server, + const std::string& db, + const BSONObj& cmdObj, + const BSONObj& originalResult) const { + return originalResult; +} - } +void RunOnAllShardsCommand::getShardIds(const std::string& db, + BSONObj& cmdObj, + std::vector<ShardId>& shardIds) { + grid.shardRegistry()->getAllShardIds(&shardIds); +} - void RunOnAllShardsCommand::aggregateResults(const std::vector<ShardAndReply>& results, - BSONObjBuilder& output) - {} +bool RunOnAllShardsCommand::run(OperationContext* txn, + const std::string& dbName, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& output) { + LOG(1) << "RunOnAllShardsCommand db: " << dbName << " cmd:" << cmdObj; - BSONObj RunOnAllShardsCommand::specialErrorHandler(const std::string& server, - const std::string& db, - const BSONObj& cmdObj, - const BSONObj& originalResult) const { - return originalResult; + if (_implicitCreateDb) { + uassertStatusOK(grid.implicitCreateDb(dbName)); } - void RunOnAllShardsCommand::getShardIds(const std::string& db, - BSONObj& cmdObj, - std::vector<ShardId>& shardIds) { - grid.shardRegistry()->getAllShardIds(&shardIds); - } + std::vector<ShardId> shardIds; + getShardIds(dbName, cmdObj, shardIds); - bool RunOnAllShardsCommand::run(OperationContext* txn, - const std::string& dbName, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& output) { - - LOG(1) << "RunOnAllShardsCommand db: " << dbName << " cmd:" << cmdObj; - - if (_implicitCreateDb) { - uassertStatusOK(grid.implicitCreateDb(dbName)); + std::list<std::shared_ptr<Future::CommandResult>> futures; + for (const ShardId& shardId : shardIds) { + const auto shard = grid.shardRegistry()->getShard(shardId); + if (!shard) { + continue; } - std::vector<ShardId> shardIds; - getShardIds(dbName, cmdObj, shardIds); - - std::list<std::shared_ptr<Future::CommandResult>> futures; - for (const ShardId& shardId : shardIds) { - const auto shard = grid.shardRegistry()->getShard(shardId); - if (!shard) { - continue; - } + futures.push_back(Future::spawnCommand( + shard->getConnString().toString(), dbName, cmdObj, 0, NULL, _useShardConn)); + } - futures.push_back(Future::spawnCommand(shard->getConnString().toString(), - dbName, - cmdObj, - 0, - NULL, - _useShardConn)); + std::vector<ShardAndReply> results; + BSONObjBuilder subobj(output.subobjStart("raw")); + BSONObjBuilder errors; + int commonErrCode = -1; + + std::list<std::shared_ptr<Future::CommandResult>>::iterator futuresit; + std::vector<ShardId>::const_iterator shardIdsIt; + // We iterate over the set of shard ids and their corresponding futures in parallel. + // TODO: replace with zip iterator if we ever decide to use one from Boost or elsewhere + for (futuresit = futures.begin(), shardIdsIt = shardIds.cbegin(); + futuresit != futures.end() && shardIdsIt != shardIds.end(); + ++futuresit, ++shardIdsIt) { + std::shared_ptr<Future::CommandResult> res = *futuresit; + + if (res->join()) { + // success :) + BSONObj result = res->result(); + results.emplace_back(*shardIdsIt, result); + subobj.append(res->getServer(), result); + continue; } - std::vector<ShardAndReply> results; - BSONObjBuilder subobj (output.subobjStart("raw")); - BSONObjBuilder errors; - int commonErrCode = -1; - - std::list< std::shared_ptr<Future::CommandResult> >::iterator futuresit; - std::vector<ShardId>::const_iterator shardIdsIt; - // We iterate over the set of shard ids and their corresponding futures in parallel. - // TODO: replace with zip iterator if we ever decide to use one from Boost or elsewhere - for (futuresit = futures.begin(), shardIdsIt = shardIds.cbegin(); - futuresit != futures.end() && shardIdsIt != shardIds.end(); - ++futuresit, ++shardIdsIt) { - - std::shared_ptr<Future::CommandResult> res = *futuresit; - - if ( res->join() ) { - // success :) - BSONObj result = res->result(); - results.emplace_back(*shardIdsIt, result ); - subobj.append( res->getServer(), result ); - continue; - } + BSONObj result = res->result(); - BSONObj result = res->result(); - - if ( result["errmsg"].type() || - result["code"].numberInt() != 0 ) { - result = specialErrorHandler( res->getServer(), dbName, cmdObj, result ); - - BSONElement errmsg = result["errmsg"]; - if ( errmsg.eoo() || errmsg.String().empty() ) { - // it was fixed! - results.emplace_back(*shardIdsIt, result ); - subobj.append( res->getServer(), result ); - continue; - } - } + if (result["errmsg"].type() || result["code"].numberInt() != 0) { + result = specialErrorHandler(res->getServer(), dbName, cmdObj, result); - // Handle "errmsg". - if( ! result["errmsg"].eoo() ){ - errors.appendAs(result["errmsg"], res->getServer()); - } - else { - // Can happen if message is empty, for some reason - errors.append( res->getServer(), str::stream() << - "result without error message returned : " << result ); + BSONElement errmsg = result["errmsg"]; + if (errmsg.eoo() || errmsg.String().empty()) { + // it was fixed! + results.emplace_back(*shardIdsIt, result); + subobj.append(res->getServer(), result); + continue; } + } - // Handle "code". - int errCode = result["code"].numberInt(); - if ( commonErrCode == -1 ) { - commonErrCode = errCode; - } - else if ( commonErrCode != errCode ) { - commonErrCode = 0; - } - results.emplace_back(*shardIdsIt, result ); - subobj.append( res->getServer(), result ); + // Handle "errmsg". + if (!result["errmsg"].eoo()) { + errors.appendAs(result["errmsg"], res->getServer()); + } else { + // Can happen if message is empty, for some reason + errors.append(res->getServer(), + str::stream() << "result without error message returned : " << result); } - subobj.done(); + // Handle "code". + int errCode = result["code"].numberInt(); + if (commonErrCode == -1) { + commonErrCode = errCode; + } else if (commonErrCode != errCode) { + commonErrCode = 0; + } + results.emplace_back(*shardIdsIt, result); + subobj.append(res->getServer(), result); + } - BSONObj errobj = errors.done(); - if (! errobj.isEmpty()) { - errmsg = errobj.toString(false, true); + subobj.done(); - // If every error has a code, and the code for all errors is the same, then add - // a top-level field "code" with this value to the output object. - if ( commonErrCode > 0 ) { - output.append( "code", commonErrCode ); - } + BSONObj errobj = errors.done(); + if (!errobj.isEmpty()) { + errmsg = errobj.toString(false, true); - return false; + // If every error has a code, and the code for all errors is the same, then add + // a top-level field "code" with this value to the output object. + if (commonErrCode > 0) { + output.append("code", commonErrCode); } - aggregateResults(results, output); - return true; + return false; } + aggregateResults(results, output); + return true; +} } diff --git a/src/mongo/s/commands/run_on_all_shards_cmd.h b/src/mongo/s/commands/run_on_all_shards_cmd.h index bc0b6f22084..149887864a7 100644 --- a/src/mongo/s/commands/run_on_all_shards_cmd.h +++ b/src/mongo/s/commands/run_on_all_shards_cmd.h @@ -36,62 +36,68 @@ namespace mongo { - class BSONObj; - class BSONObjBuilder; - class OperationContext; +class BSONObj; +class BSONObjBuilder; +class OperationContext; - /** - * Logic for commands that simply map out to all shards then fold the results into - * a single response. - * - * All shards are contacted in parallel. - * - * When extending, don't override run() - but rather aggregateResults(). If you need - * to implement some kind of fall back logic for multiversion clusters, - * override specialErrorHandler(). - */ - class RunOnAllShardsCommand : public Command { - public: - RunOnAllShardsCommand(const char* name, - const char* oldName=NULL, - bool useShardConn=false, - bool implicitCreateDb=false); +/** + * Logic for commands that simply map out to all shards then fold the results into + * a single response. + * + * All shards are contacted in parallel. + * + * When extending, don't override run() - but rather aggregateResults(). If you need + * to implement some kind of fall back logic for multiversion clusters, + * override specialErrorHandler(). + */ +class RunOnAllShardsCommand : public Command { +public: + RunOnAllShardsCommand(const char* name, + const char* oldName = NULL, + bool useShardConn = false, + bool implicitCreateDb = false); - bool slaveOk() const override { return true; } - bool adminOnly() const override { return false; } - bool isWriteCommandForConfigServer() const override { return false; } + bool slaveOk() const override { + return true; + } + bool adminOnly() const override { + return false; + } + bool isWriteCommandForConfigServer() const override { + return false; + } - // The StringData contains the shard ident. - // This can be used to create an instance of Shard - using ShardAndReply = std::tuple<StringData, BSONObj>; + // The StringData contains the shard ident. + // This can be used to create an instance of Shard + using ShardAndReply = std::tuple<StringData, BSONObj>; - virtual void aggregateResults(const std::vector<ShardAndReply>& results, - BSONObjBuilder& output); + virtual void aggregateResults(const std::vector<ShardAndReply>& results, + BSONObjBuilder& output); - // The default implementation is the identity function. - virtual BSONObj specialErrorHandler(const std::string& server, - const std::string& db, - const BSONObj& cmdObj, - const BSONObj& originalResult) const; + // The default implementation is the identity function. + virtual BSONObj specialErrorHandler(const std::string& server, + const std::string& db, + const BSONObj& cmdObj, + const BSONObj& originalResult) const; - // The default implementation uses all shards. - virtual void getShardIds(const std::string& db, - BSONObj& cmdObj, - std::vector<ShardId>& shardIds); + // The default implementation uses all shards. + virtual void getShardIds(const std::string& db, + BSONObj& cmdObj, + std::vector<ShardId>& shardIds); - bool run(OperationContext* txn, - const std::string& db, - BSONObj& cmdObj, - int options, - std::string& errmsg, - BSONObjBuilder& output) final; + bool run(OperationContext* txn, + const std::string& db, + BSONObj& cmdObj, + int options, + std::string& errmsg, + BSONObjBuilder& output) final; - private: - // Use ShardConnection as opposed to ScopedDbConnection - const bool _useShardConn; +private: + // Use ShardConnection as opposed to ScopedDbConnection + const bool _useShardConn; - // Whether the requested database should be created implicitly - const bool _implicitCreateDb; - }; + // Whether the requested database should be created implicitly + const bool _implicitCreateDb; +}; } // namespace mongo diff --git a/src/mongo/s/config.cpp b/src/mongo/s/config.cpp index 0a1904c9552..7481f949a2a 100644 --- a/src/mongo/s/config.cpp +++ b/src/mongo/s/config.cpp @@ -56,751 +56,712 @@ namespace mongo { - using std::endl; - using std::set; - using std::string; - using std::unique_ptr; - using std::vector; - - CollectionInfo::CollectionInfo(const CollectionType& coll) { - _dropped = coll.getDropped(); - - shard(new ChunkManager(coll)); - _dirty = false; +using std::endl; +using std::set; +using std::string; +using std::unique_ptr; +using std::vector; + +CollectionInfo::CollectionInfo(const CollectionType& coll) { + _dropped = coll.getDropped(); + + shard(new ChunkManager(coll)); + _dirty = false; +} + +CollectionInfo::~CollectionInfo() {} + +void CollectionInfo::resetCM(ChunkManager* cm) { + invariant(cm); + invariant(_cm); + + _cm.reset(cm); +} + +void CollectionInfo::shard(ChunkManager* manager) { + // Do this *first* so we're invisible to everyone else + manager->loadExistingRanges(nullptr); + + // + // Collections with no chunks are unsharded, no matter what the collections entry says + // This helps prevent errors when dropping in a different process + // + + if (manager->numChunks() != 0) { + useChunkManager(ChunkManagerPtr(manager)); + } else { + warning() << "no chunks found for collection " << manager->getns() + << ", assuming unsharded"; + unshard(); } - - CollectionInfo::~CollectionInfo() { - +} + +void CollectionInfo::unshard() { + _cm.reset(); + _dropped = true; + _dirty = true; + _key = BSONObj(); +} + +void CollectionInfo::useChunkManager(ChunkManagerPtr manager) { + _cm = manager; + _key = manager->getShardKeyPattern().toBSON().getOwned(); + _unique = manager->isUnique(); + _dirty = true; + _dropped = false; +} + +void CollectionInfo::save(const string& ns) { + CollectionType coll; + coll.setNs(NamespaceString{ns}); + + if (_cm) { + invariant(!_dropped); + coll.setEpoch(_cm->getVersion().epoch()); + // TODO(schwerin): The following isn't really a date, but is stored as one in-memory and + // in config.collections, as a historical oddity. + coll.setUpdatedAt(Date_t::fromMillisSinceEpoch(_cm->getVersion().toLong())); + coll.setKeyPattern(_cm->getShardKeyPattern().toBSON()); + coll.setUnique(_cm->isUnique()); + } else { + invariant(_dropped); + coll.setDropped(true); + coll.setEpoch(ChunkVersion::DROPPED().epoch()); + coll.setUpdatedAt(Date_t::now()); } - void CollectionInfo::resetCM(ChunkManager* cm) { - invariant(cm); - invariant(_cm); - - _cm.reset(cm); + uassertStatusOK(grid.catalogManager()->updateCollection(ns, coll)); + _dirty = false; +} + +DBConfig::DBConfig(std::string name, const DatabaseType& dbt) : _name(name) { + invariant(_name == dbt.getName()); + _primaryId = dbt.getPrimary(); + _shardingEnabled = dbt.getSharded(); +} + +bool DBConfig::isSharded(const string& ns) { + if (!_shardingEnabled) + return false; + stdx::lock_guard<stdx::mutex> lk(_lock); + return _isSharded(ns); +} + +bool DBConfig::_isSharded(const string& ns) { + if (!_shardingEnabled) { + return false; } - - void CollectionInfo::shard(ChunkManager* manager) { - // Do this *first* so we're invisible to everyone else - manager->loadExistingRanges(nullptr); - - // - // Collections with no chunks are unsharded, no matter what the collections entry says - // This helps prevent errors when dropping in a different process - // - if (manager->numChunks() != 0) { - useChunkManager(ChunkManagerPtr(manager)); - } - else{ - warning() << "no chunks found for collection " << manager->getns() - << ", assuming unsharded"; - unshard(); - } + CollectionInfoMap::iterator i = _collections.find(ns); + if (i == _collections.end()) { + return false; } - void CollectionInfo::unshard() { - _cm.reset(); - _dropped = true; - _dirty = true; - _key = BSONObj(); - } + return i->second.isSharded(); +} - void CollectionInfo::useChunkManager(ChunkManagerPtr manager) { - _cm = manager; - _key = manager->getShardKeyPattern().toBSON().getOwned(); - _unique = manager->isUnique(); - _dirty = true; - _dropped = false; - } +const ShardId& DBConfig::getShardId(const string& ns) { + uassert(28679, "ns can't be sharded", !isSharded(ns)); - void CollectionInfo::save(const string& ns) { - CollectionType coll; - coll.setNs(NamespaceString{ns}); + uassert(10178, "no primary!", grid.shardRegistry()->getShard(_primaryId)); + return _primaryId; +} - if (_cm) { - invariant(!_dropped); - coll.setEpoch(_cm->getVersion().epoch()); - // TODO(schwerin): The following isn't really a date, but is stored as one in-memory and - // in config.collections, as a historical oddity. - coll.setUpdatedAt(Date_t::fromMillisSinceEpoch(_cm->getVersion().toLong())); - coll.setKeyPattern(_cm->getShardKeyPattern().toBSON()); - coll.setUnique(_cm->isUnique()); - } - else { - invariant(_dropped); - coll.setDropped(true); - coll.setEpoch(ChunkVersion::DROPPED().epoch()); - coll.setUpdatedAt(Date_t::now()); - } +void DBConfig::enableSharding(bool save) { + if (_shardingEnabled) + return; - uassertStatusOK(grid.catalogManager()->updateCollection(ns, coll)); - _dirty = false; - } + verify(_name != "config"); - DBConfig::DBConfig(std::string name, const DatabaseType& dbt) - : _name(name) { + stdx::lock_guard<stdx::mutex> lk(_lock); + _shardingEnabled = true; + if (save) + _save(); +} - invariant(_name == dbt.getName()); - _primaryId = dbt.getPrimary(); - _shardingEnabled = dbt.getSharded(); +bool DBConfig::removeSharding(const string& ns) { + if (!_shardingEnabled) { + warning() << "could not remove sharding for collection " << ns + << ", sharding not enabled for db" << endl; + return false; } - bool DBConfig::isSharded( const string& ns ) { - if ( ! _shardingEnabled ) - return false; - stdx::lock_guard<stdx::mutex> lk( _lock ); - return _isSharded( ns ); - } + stdx::lock_guard<stdx::mutex> lk(_lock); - bool DBConfig::_isSharded( const string& ns ) { - if (!_shardingEnabled) { - return false; - } + CollectionInfoMap::iterator i = _collections.find(ns); - CollectionInfoMap::iterator i = _collections.find( ns ); - if (i == _collections.end()) { - return false; - } + if (i == _collections.end()) + return false; - return i->second.isSharded(); + CollectionInfo& ci = _collections[ns]; + if (!ci.isSharded()) { + warning() << "could not remove sharding for collection " << ns + << ", no sharding information found" << endl; + return false; } - const ShardId& DBConfig::getShardId(const string& ns) { - uassert(28679, "ns can't be sharded", !isSharded(ns)); + ci.unshard(); + _save(false, true); + return true; +} - uassert(10178, - "no primary!", - grid.shardRegistry()->getShard(_primaryId)); - return _primaryId; - } +// Handles weird logic related to getting *either* a chunk manager *or* the collection primary shard +void DBConfig::getChunkManagerOrPrimary(const string& ns, + std::shared_ptr<ChunkManager>& manager, + std::shared_ptr<Shard>& primary) { + // The logic here is basically that at any time, our collection can become sharded or unsharded + // via a command. If we're not sharded, we want to send data to the primary, if sharded, we want + // to send data to the correct chunks, and we can't check both w/o the lock. - void DBConfig::enableSharding( bool save ) { - if ( _shardingEnabled ) - return; - - verify( _name != "config" ); + manager.reset(); + primary.reset(); - stdx::lock_guard<stdx::mutex> lk( _lock ); - _shardingEnabled = true; - if( save ) _save(); - } + { + stdx::lock_guard<stdx::mutex> lk(_lock); - bool DBConfig::removeSharding( const string& ns ) { - if ( ! _shardingEnabled ) { + CollectionInfoMap::iterator i = _collections.find(ns); - warning() << "could not remove sharding for collection " << ns - << ", sharding not enabled for db" << endl; - return false; + // No namespace + if (i == _collections.end()) { + // If we don't know about this namespace, it's unsharded by default + primary = grid.shardRegistry()->getShard(_primaryId); + } else { + CollectionInfo& cInfo = i->second; + + // TODO: we need to be careful about handling shardingEnabled, b/c in some places we seem to use and + // some we don't. If we use this function in combination with just getChunkManager() on a slightly + // borked config db, we'll get lots of staleconfig retries + if (_shardingEnabled && cInfo.isSharded()) { + manager = cInfo.getCM(); + } else { + primary = grid.shardRegistry()->getShard(_primaryId); + } } + } - stdx::lock_guard<stdx::mutex> lk( _lock ); + verify(manager || primary); + verify(!manager || !primary); +} - CollectionInfoMap::iterator i = _collections.find( ns ); - if ( i == _collections.end() ) - return false; +ChunkManagerPtr DBConfig::getChunkManagerIfExists(const string& ns, + bool shouldReload, + bool forceReload) { + // Don't report exceptions here as errors in GetLastError + LastError::Disabled ignoreForGLE(&LastError::get(cc())); - CollectionInfo& ci = _collections[ns]; - if ( ! ci.isSharded() ){ - - warning() << "could not remove sharding for collection " << ns - << ", no sharding information found" << endl; - return false; - } - - ci.unshard(); - _save( false, true ); - return true; + try { + return getChunkManager(ns, shouldReload, forceReload); + } catch (AssertionException& e) { + warning() << "chunk manager not found for " << ns << causedBy(e); + return ChunkManagerPtr(); } +} - // Handles weird logic related to getting *either* a chunk manager *or* the collection primary shard - void DBConfig::getChunkManagerOrPrimary(const string& ns, - std::shared_ptr<ChunkManager>& manager, - std::shared_ptr<Shard>& primary) { - - // The logic here is basically that at any time, our collection can become sharded or unsharded - // via a command. If we're not sharded, we want to send data to the primary, if sharded, we want - // to send data to the correct chunks, and we can't check both w/o the lock. +std::shared_ptr<ChunkManager> DBConfig::getChunkManager(const string& ns, + bool shouldReload, + bool forceReload) { + BSONObj key; + ChunkVersion oldVersion; + ChunkManagerPtr oldManager; - manager.reset(); - primary.reset(); - - { - stdx::lock_guard<stdx::mutex> lk( _lock ); - - CollectionInfoMap::iterator i = _collections.find( ns ); + { + stdx::lock_guard<stdx::mutex> lk(_lock); - // No namespace - if( i == _collections.end() ){ - // If we don't know about this namespace, it's unsharded by default - primary = grid.shardRegistry()->getShard(_primaryId); - } - else { - CollectionInfo& cInfo = i->second; - - // TODO: we need to be careful about handling shardingEnabled, b/c in some places we seem to use and - // some we don't. If we use this function in combination with just getChunkManager() on a slightly - // borked config db, we'll get lots of staleconfig retries - if( _shardingEnabled && cInfo.isSharded() ){ - manager = cInfo.getCM(); - } - else{ - primary = grid.shardRegistry()->getShard(_primaryId); - } - } + bool earlyReload = !_collections[ns].isSharded() && (shouldReload || forceReload); + if (earlyReload) { + // This is to catch cases where there this is a new sharded collection + _reload(); } - verify( manager || primary ); - verify( ! manager || ! primary ); + CollectionInfo& ci = _collections[ns]; + uassert(10181, str::stream() << "not sharded:" << ns, ci.isSharded()); - } + invariant(!ci.key().isEmpty()); + if (!(shouldReload || forceReload) || earlyReload) { + return ci.getCM(); + } - ChunkManagerPtr DBConfig::getChunkManagerIfExists(const string& ns, - bool shouldReload, - bool forceReload) { + key = ci.key().copy(); - // Don't report exceptions here as errors in GetLastError - LastError::Disabled ignoreForGLE(&LastError::get(cc())); - - try{ - return getChunkManager(ns, shouldReload, forceReload); - } - catch (AssertionException& e) { - warning() << "chunk manager not found for " << ns << causedBy(e); - return ChunkManagerPtr(); + if (ci.getCM()) { + oldManager = ci.getCM(); + oldVersion = ci.getCM()->getVersion(); } } - std::shared_ptr<ChunkManager> DBConfig::getChunkManager(const string& ns, - bool shouldReload, - bool forceReload) { - BSONObj key; - ChunkVersion oldVersion; - ChunkManagerPtr oldManager; - - { - stdx::lock_guard<stdx::mutex> lk(_lock); - - bool earlyReload = !_collections[ns].isSharded() && (shouldReload || forceReload); - if (earlyReload) { - // This is to catch cases where there this is a new sharded collection - _reload(); - } - - CollectionInfo& ci = _collections[ns]; - uassert(10181, str::stream() << "not sharded:" << ns, ci.isSharded()); - - invariant(!ci.key().isEmpty()); - - if (!(shouldReload || forceReload) || earlyReload) { + invariant(!key.isEmpty()); + + // TODO: We need to keep this first one-chunk check in until we have a more efficient way of + // creating/reusing a chunk manager, as doing so requires copying the full set of chunks currently + vector<ChunkType> newestChunk; + if (oldVersion.isSet() && !forceReload) { + uassertStatusOK(grid.catalogManager()->getChunks( + Query(BSON(ChunkType::ns(ns))).sort(ChunkType::DEPRECATED_lastmod(), -1), + 1, + &newestChunk)); + + if (!newestChunk.empty()) { + invariant(newestChunk.size() == 1); + ChunkVersion v = newestChunk[0].getVersion(); + if (v.equals(oldVersion)) { + stdx::lock_guard<stdx::mutex> lk(_lock); + const CollectionInfo& ci = _collections[ns]; + uassert(15885, + str::stream() << "not sharded after reloading from chunks : " << ns, + ci.isSharded()); return ci.getCM(); } - - key = ci.key().copy(); - - if (ci.getCM()){ - oldManager = ci.getCM(); - oldVersion = ci.getCM()->getVersion(); - } } - - invariant(!key.isEmpty()); - - // TODO: We need to keep this first one-chunk check in until we have a more efficient way of - // creating/reusing a chunk manager, as doing so requires copying the full set of chunks currently - vector<ChunkType> newestChunk; - if ( oldVersion.isSet() && ! forceReload ) { - uassertStatusOK(grid.catalogManager()->getChunks( - Query(BSON(ChunkType::ns(ns))) - .sort(ChunkType::DEPRECATED_lastmod(), -1), - 1, - &newestChunk)); - - if (!newestChunk.empty()) { - invariant(newestChunk.size() == 1); - ChunkVersion v = newestChunk[0].getVersion(); - if (v.equals(oldVersion)) { - stdx::lock_guard<stdx::mutex> lk( _lock ); - const CollectionInfo& ci = _collections[ns]; - uassert(15885, - str::stream() << "not sharded after reloading from chunks : " - << ns, ci.isSharded()); - return ci.getCM(); - } - } - } - else if (!oldVersion.isSet()) { - warning() << "version 0 found when " << (forceReload ? "reloading" : "checking") - << " chunk manager; collection '" << ns << "' initially detected as sharded"; - } + } else if (!oldVersion.isSet()) { + warning() << "version 0 found when " << (forceReload ? "reloading" : "checking") + << " chunk manager; collection '" << ns << "' initially detected as sharded"; + } - // we are not locked now, and want to load a new ChunkManager - - unique_ptr<ChunkManager> tempChunkManager; + // we are not locked now, and want to load a new ChunkManager - { - stdx::lock_guard<stdx::mutex> lll ( _hitConfigServerLock ); + unique_ptr<ChunkManager> tempChunkManager; - if (!newestChunk.empty() && !forceReload) { - // If we have a target we're going for see if we've hit already - stdx::lock_guard<stdx::mutex> lk( _lock ); + { + stdx::lock_guard<stdx::mutex> lll(_hitConfigServerLock); - CollectionInfo& ci = _collections[ns]; + if (!newestChunk.empty() && !forceReload) { + // If we have a target we're going for see if we've hit already + stdx::lock_guard<stdx::mutex> lk(_lock); - if ( ci.isSharded() && ci.getCM() ) { - ChunkVersion currentVersion = newestChunk[0].getVersion(); + CollectionInfo& ci = _collections[ns]; - // Only reload if the version we found is newer than our own in the same epoch - if (currentVersion <= ci.getCM()->getVersion() && - ci.getCM()->getVersion().hasEqualEpoch(currentVersion)) { + if (ci.isSharded() && ci.getCM()) { + ChunkVersion currentVersion = newestChunk[0].getVersion(); - return ci.getCM(); - } + // Only reload if the version we found is newer than our own in the same epoch + if (currentVersion <= ci.getCM()->getVersion() && + ci.getCM()->getVersion().hasEqualEpoch(currentVersion)) { + return ci.getCM(); } } - - tempChunkManager.reset(new ChunkManager(oldManager->getns(), - oldManager->getShardKeyPattern(), - oldManager->isUnique())); - tempChunkManager->loadExistingRanges(oldManager.get()); - - if (tempChunkManager->numChunks() == 0) { - // Maybe we're not sharded any more, so do a full reload - reload(); - - return getChunkManager(ns, false); - } } - stdx::lock_guard<stdx::mutex> lk( _lock ); + tempChunkManager.reset(new ChunkManager( + oldManager->getns(), oldManager->getShardKeyPattern(), oldManager->isUnique())); + tempChunkManager->loadExistingRanges(oldManager.get()); - CollectionInfo& ci = _collections[ns]; - uassert(14822, (string)"state changed in the middle: " + ns, ci.isSharded()); - - // Reset if our versions aren't the same - bool shouldReset = !tempChunkManager->getVersion().equals(ci.getCM()->getVersion()); - - // Also reset if we're forced to do so - if (!shouldReset && forceReload) { - shouldReset = true; - warning() << "chunk manager reload forced for collection '" << ns - << "', config version is " << tempChunkManager->getVersion(); - } + if (tempChunkManager->numChunks() == 0) { + // Maybe we're not sharded any more, so do a full reload + reload(); - // - // LEGACY BEHAVIOR - // - // It's possible to get into a state when dropping collections when our new version is - // less than our prev version. Behave identically to legacy mongos, for now, and warn to - // draw attention to the problem. - // - // TODO: Assert in next version, to allow smooth upgrades - // - - if (shouldReset && tempChunkManager->getVersion() < ci.getCM()->getVersion()) { - shouldReset = false; - - warning() << "not resetting chunk manager for collection '" << ns - << "', config version is " << tempChunkManager->getVersion() << " and " - << "old version is " << ci.getCM()->getVersion(); + return getChunkManager(ns, false); } + } - // end legacy behavior + stdx::lock_guard<stdx::mutex> lk(_lock); - if (shouldReset) { - ci.resetCM(tempChunkManager.release()); - } + CollectionInfo& ci = _collections[ns]; + uassert(14822, (string) "state changed in the middle: " + ns, ci.isSharded()); - uassert(15883, str::stream() << "not sharded after chunk manager reset : " - << ns, ci.isSharded()); + // Reset if our versions aren't the same + bool shouldReset = !tempChunkManager->getVersion().equals(ci.getCM()->getVersion()); - return ci.getCM(); + // Also reset if we're forced to do so + if (!shouldReset && forceReload) { + shouldReset = true; + warning() << "chunk manager reload forced for collection '" << ns << "', config version is " + << tempChunkManager->getVersion(); } - void DBConfig::setPrimary(const std::string& s) { - const auto shard = grid.shardRegistry()->getShard(s); - - stdx::lock_guard<stdx::mutex> lk( _lock ); - _primaryId = shard->getId(); - _save(); + // + // LEGACY BEHAVIOR + // + // It's possible to get into a state when dropping collections when our new version is + // less than our prev version. Behave identically to legacy mongos, for now, and warn to + // draw attention to the problem. + // + // TODO: Assert in next version, to allow smooth upgrades + // + + if (shouldReset && tempChunkManager->getVersion() < ci.getCM()->getVersion()) { + shouldReset = false; + + warning() << "not resetting chunk manager for collection '" << ns << "', config version is " + << tempChunkManager->getVersion() << " and " + << "old version is " << ci.getCM()->getVersion(); } - bool DBConfig::load() { - stdx::lock_guard<stdx::mutex> lk( _lock ); - return _load(); + // end legacy behavior + + if (shouldReset) { + ci.resetCM(tempChunkManager.release()); } - bool DBConfig::_load() { - StatusWith<DatabaseType> status = grid.catalogManager()->getDatabase(_name); - if (status == ErrorCodes::DatabaseNotFound) { - return false; - } + uassert( + 15883, str::stream() << "not sharded after chunk manager reset : " << ns, ci.isSharded()); - // All other errors are connectivity, etc so throw an exception. - uassertStatusOK(status.getStatus()); + return ci.getCM(); +} - DatabaseType dbt = status.getValue(); - invariant(_name == dbt.getName()); - _primaryId = dbt.getPrimary(); - _shardingEnabled = dbt.getSharded(); +void DBConfig::setPrimary(const std::string& s) { + const auto shard = grid.shardRegistry()->getShard(s); - // Load all collections - vector<CollectionType> collections; - uassertStatusOK(grid.catalogManager()->getCollections(&_name, &collections)); + stdx::lock_guard<stdx::mutex> lk(_lock); + _primaryId = shard->getId(); + _save(); +} - int numCollsErased = 0; - int numCollsSharded = 0; +bool DBConfig::load() { + stdx::lock_guard<stdx::mutex> lk(_lock); + return _load(); +} - for (const auto& coll : collections) { - if (coll.getDropped()) { - _collections.erase(coll.getNs()); - numCollsErased++; - } - else { - _collections[coll.getNs()] = CollectionInfo(coll); - numCollsSharded++; - } - } - - LOG(2) << "found " << numCollsSharded << " collections left and " - << numCollsErased << " collections dropped for database " << _name; - - return true; +bool DBConfig::_load() { + StatusWith<DatabaseType> status = grid.catalogManager()->getDatabase(_name); + if (status == ErrorCodes::DatabaseNotFound) { + return false; } - void DBConfig::_save(bool db, bool coll) { - if (db) { - DatabaseType dbt; - dbt.setName(_name); - dbt.setPrimary(_primaryId); - dbt.setSharded(_shardingEnabled); + // All other errors are connectivity, etc so throw an exception. + uassertStatusOK(status.getStatus()); - uassertStatusOK(grid.catalogManager()->updateDatabase(_name, dbt)); - } + DatabaseType dbt = status.getValue(); + invariant(_name == dbt.getName()); + _primaryId = dbt.getPrimary(); + _shardingEnabled = dbt.getSharded(); - if (coll) { - for (CollectionInfoMap::iterator i = _collections.begin(); - i != _collections.end(); - ++i) { + // Load all collections + vector<CollectionType> collections; + uassertStatusOK(grid.catalogManager()->getCollections(&_name, &collections)); - if (!i->second.isDirty()) { - continue; - } + int numCollsErased = 0; + int numCollsSharded = 0; - i->second.save(i->first); - } + for (const auto& coll : collections) { + if (coll.getDropped()) { + _collections.erase(coll.getNs()); + numCollsErased++; + } else { + _collections[coll.getNs()] = CollectionInfo(coll); + numCollsSharded++; } } - bool DBConfig::reload() { - bool successful = false; + LOG(2) << "found " << numCollsSharded << " collections left and " << numCollsErased + << " collections dropped for database " << _name; - { - stdx::lock_guard<stdx::mutex> lk( _lock ); - successful = _reload(); - } + return true; +} - // If we aren't successful loading the database entry, we don't want to keep the stale - // object around which has invalid data. - if (!successful) { - grid.catalogCache()->invalidate(_name); - } +void DBConfig::_save(bool db, bool coll) { + if (db) { + DatabaseType dbt; + dbt.setName(_name); + dbt.setPrimary(_primaryId); + dbt.setSharded(_shardingEnabled); - return successful; + uassertStatusOK(grid.catalogManager()->updateDatabase(_name, dbt)); } - bool DBConfig::_reload() { - // TODO: i don't think is 100% correct - return _load(); + if (coll) { + for (CollectionInfoMap::iterator i = _collections.begin(); i != _collections.end(); ++i) { + if (!i->second.isDirty()) { + continue; + } + + i->second.save(i->first); + } } +} - bool DBConfig::dropDatabase(string& errmsg) { - /** - * 1) update config server - * 2) drop and reset sharded collections - * 3) drop and reset primary - * 4) drop everywhere to clean up loose ends - */ +bool DBConfig::reload() { + bool successful = false; - log() << "DBConfig::dropDatabase: " << _name << endl; - grid.catalogManager()->logChange(NULL, "dropDatabase.start", _name, BSONObj()); + { + stdx::lock_guard<stdx::mutex> lk(_lock); + successful = _reload(); + } - // 1 + // If we aren't successful loading the database entry, we don't want to keep the stale + // object around which has invalid data. + if (!successful) { grid.catalogCache()->invalidate(_name); + } - Status result = grid.catalogManager()->remove(DatabaseType::ConfigNS, - BSON(DatabaseType::name(_name)), - 0, - NULL); - if (!result.isOK()) { - errmsg = result.reason(); - log() << "could not drop '" << _name << "': " << errmsg << endl; - return false; - } + return successful; +} + +bool DBConfig::_reload() { + // TODO: i don't think is 100% correct + return _load(); +} + +bool DBConfig::dropDatabase(string& errmsg) { + /** + * 1) update config server + * 2) drop and reset sharded collections + * 3) drop and reset primary + * 4) drop everywhere to clean up loose ends + */ + + log() << "DBConfig::dropDatabase: " << _name << endl; + grid.catalogManager()->logChange(NULL, "dropDatabase.start", _name, BSONObj()); + + // 1 + grid.catalogCache()->invalidate(_name); + + Status result = grid.catalogManager()->remove( + DatabaseType::ConfigNS, BSON(DatabaseType::name(_name)), 0, NULL); + if (!result.isOK()) { + errmsg = result.reason(); + log() << "could not drop '" << _name << "': " << errmsg << endl; + return false; + } - LOG(1) << "\t removed entry from config server for: " << _name << endl; + LOG(1) << "\t removed entry from config server for: " << _name << endl; - set<ShardId> shardIds; + set<ShardId> shardIds; - // 2 - while ( true ) { - int num = 0; - if (!_dropShardedCollections(num, shardIds, errmsg)) { - return 0; - } + // 2 + while (true) { + int num = 0; + if (!_dropShardedCollections(num, shardIds, errmsg)) { + return 0; + } - log() << " DBConfig::dropDatabase: " << _name - << " dropped sharded collections: " << num; + log() << " DBConfig::dropDatabase: " << _name << " dropped sharded collections: " << num; - if (num == 0) { - break; - } + if (num == 0) { + break; } + } - // 3 - { - const auto shard = grid.shardRegistry()->getShard(_primaryId); - ScopedDbConnection conn(shard->getConnString(), 30.0); - BSONObj res; - if ( ! conn->dropDatabase( _name , &res ) ) { - errmsg = res.toString(); - return 0; - } - conn.done(); + // 3 + { + const auto shard = grid.shardRegistry()->getShard(_primaryId); + ScopedDbConnection conn(shard->getConnString(), 30.0); + BSONObj res; + if (!conn->dropDatabase(_name, &res)) { + errmsg = res.toString(); + return 0; } + conn.done(); + } - // 4 - for (const ShardId& shardId : shardIds) { - const auto shard = grid.shardRegistry()->getShard(shardId); - if (!shard) { - continue; - } - - ScopedDbConnection conn(shard->getConnString(), 30.0); - BSONObj res; - if ( ! conn->dropDatabase( _name , &res ) ) { - errmsg = res.toString(); - return 0; - } - conn.done(); + // 4 + for (const ShardId& shardId : shardIds) { + const auto shard = grid.shardRegistry()->getShard(shardId); + if (!shard) { + continue; } - LOG(1) << "\t dropped primary db for: " << _name << endl; - - grid.catalogManager()->logChange(NULL, "dropDatabase", _name, BSONObj()); - - return true; + ScopedDbConnection conn(shard->getConnString(), 30.0); + BSONObj res; + if (!conn->dropDatabase(_name, &res)) { + errmsg = res.toString(); + return 0; + } + conn.done(); } - bool DBConfig::_dropShardedCollections(int& num, set<ShardId>& shardIds, string& errmsg) { - num = 0; - set<string> seen; - while ( true ) { - CollectionInfoMap::iterator i = _collections.begin(); - for ( ; i != _collections.end(); ++i ) { - // log() << "coll : " << i->first << " and " << i->second.isSharded() << endl; - if ( i->second.isSharded() ) - break; - } - - if ( i == _collections.end() ) - break; - - if ( seen.count( i->first ) ) { - errmsg = "seen a collection twice!"; - return false; - } + LOG(1) << "\t dropped primary db for: " << _name << endl; - seen.insert( i->first ); - LOG(1) << "\t dropping sharded collection: " << i->first << endl; + grid.catalogManager()->logChange(NULL, "dropDatabase", _name, BSONObj()); - i->second.getCM()->getAllShardIds(&shardIds); + return true; +} - uassertStatusOK(grid.catalogManager()->dropCollection(i->first)); +bool DBConfig::_dropShardedCollections(int& num, set<ShardId>& shardIds, string& errmsg) { + num = 0; + set<string> seen; + while (true) { + CollectionInfoMap::iterator i = _collections.begin(); + for (; i != _collections.end(); ++i) { + // log() << "coll : " << i->first << " and " << i->second.isSharded() << endl; + if (i->second.isSharded()) + break; + } - // We should warn, but it's not a fatal error if someone else reloaded the db/coll as - // unsharded in the meantime - if( ! removeSharding( i->first ) ){ - warning() << "collection " << i->first - << " was reloaded as unsharded before drop completed" - << " during drop of all collections" << endl; - } + if (i == _collections.end()) + break; - num++; - uassert( 10184 , "_dropShardedCollections too many collections - bailing" , num < 100000 ); - LOG(2) << "\t\t dropped " << num << " so far" << endl; + if (seen.count(i->first)) { + errmsg = "seen a collection twice!"; + return false; } - return true; - } + seen.insert(i->first); + LOG(1) << "\t dropping sharded collection: " << i->first << endl; - void DBConfig::getAllShardIds(set<ShardId>* shardIds) { - dassert(shardIds); + i->second.getCM()->getAllShardIds(&shardIds); - stdx::lock_guard<stdx::mutex> lk(_lock); - shardIds->insert(getPrimaryId()); - for (CollectionInfoMap::const_iterator it(_collections.begin()), end(_collections.end()); - it != end; - ++it) { - if (it->second.isSharded()) { - it->second.getCM()->getAllShardIds(shardIds); - } // TODO: handle collections on non-primary shard - } - } + uassertStatusOK(grid.catalogManager()->dropCollection(i->first)); - void DBConfig::getAllShardedCollections( set<string>& namespaces ) { - stdx::lock_guard<stdx::mutex> lk(_lock); - - for( CollectionInfoMap::const_iterator i = _collections.begin(); i != _collections.end(); i++ ) { - log() << "Coll : " << i->first << " sharded? " << i->second.isSharded() << endl; - if( i->second.isSharded() ) namespaces.insert( i->first ); + // We should warn, but it's not a fatal error if someone else reloaded the db/coll as + // unsharded in the meantime + if (!removeSharding(i->first)) { + warning() << "collection " << i->first + << " was reloaded as unsharded before drop completed" + << " during drop of all collections" << endl; } + num++; + uassert(10184, "_dropShardedCollections too many collections - bailing", num < 100000); + LOG(2) << "\t\t dropped " << num << " so far" << endl; } - /* --- ConfigServer ---- */ + return true; +} - void ConfigServer::reloadSettings() { - auto chunkSizeResult = grid.catalogManager()->getGlobalSettings(SettingsType::ChunkSizeDocKey); - if (chunkSizeResult.isOK()) { - const int csize = chunkSizeResult.getValue().getChunkSizeMB(); - LOG(1) << "Found MaxChunkSize: " << csize; +void DBConfig::getAllShardIds(set<ShardId>* shardIds) { + dassert(shardIds); - if (!Chunk::setMaxChunkSizeSizeMB(csize)) { - warning() << "invalid chunksize: " << csize; - } - } - else if (chunkSizeResult.getStatus() == ErrorCodes::NoMatchingDocument) { - const int chunkSize = Chunk::MaxChunkSize / (1024 * 1024); - Status result = - grid.catalogManager()->insert(SettingsType::ConfigNS, - BSON(SettingsType::key(SettingsType::ChunkSizeDocKey) - << SettingsType::chunkSizeMB(chunkSize)), - NULL); - if (!result.isOK()) { - warning() << "couldn't set chunkSize on config db" << causedBy(result); - } - } - else { - warning() << "couldn't load settings on config db: " << chunkSizeResult.getStatus(); - } + stdx::lock_guard<stdx::mutex> lk(_lock); + shardIds->insert(getPrimaryId()); + for (CollectionInfoMap::const_iterator it(_collections.begin()), end(_collections.end()); + it != end; + ++it) { + if (it->second.isSharded()) { + it->second.getCM()->getAllShardIds(shardIds); + } // TODO: handle collections on non-primary shard + } +} - // indexes - Status result = clusterCreateIndex( ChunkType::ConfigNS, - BSON( ChunkType::ns() << 1 << ChunkType::min() << 1 ), - true, // unique - NULL ); +void DBConfig::getAllShardedCollections(set<string>& namespaces) { + stdx::lock_guard<stdx::mutex> lk(_lock); + for (CollectionInfoMap::const_iterator i = _collections.begin(); i != _collections.end(); i++) { + log() << "Coll : " << i->first << " sharded? " << i->second.isSharded() << endl; + if (i->second.isSharded()) + namespaces.insert(i->first); + } +} + +/* --- ConfigServer ---- */ + +void ConfigServer::reloadSettings() { + auto chunkSizeResult = grid.catalogManager()->getGlobalSettings(SettingsType::ChunkSizeDocKey); + if (chunkSizeResult.isOK()) { + const int csize = chunkSizeResult.getValue().getChunkSizeMB(); + LOG(1) << "Found MaxChunkSize: " << csize; + + if (!Chunk::setMaxChunkSizeSizeMB(csize)) { + warning() << "invalid chunksize: " << csize; + } + } else if (chunkSizeResult.getStatus() == ErrorCodes::NoMatchingDocument) { + const int chunkSize = Chunk::MaxChunkSize / (1024 * 1024); + Status result = + grid.catalogManager()->insert(SettingsType::ConfigNS, + BSON(SettingsType::key(SettingsType::ChunkSizeDocKey) + << SettingsType::chunkSizeMB(chunkSize)), + NULL); if (!result.isOK()) { - warning() << "couldn't create ns_1_min_1 index on config db" << causedBy(result); + warning() << "couldn't set chunkSize on config db" << causedBy(result); } + } else { + warning() << "couldn't load settings on config db: " << chunkSizeResult.getStatus(); + } - result = clusterCreateIndex( ChunkType::ConfigNS, - BSON( ChunkType::ns() << 1 << - ChunkType::shard() << 1 << - ChunkType::min() << 1 ), - true, // unique - NULL ); + // indexes + Status result = clusterCreateIndex(ChunkType::ConfigNS, + BSON(ChunkType::ns() << 1 << ChunkType::min() << 1), + true, // unique + NULL); - if (!result.isOK()) { - warning() << "couldn't create ns_1_shard_1_min_1 index on config db" - << causedBy(result); - } + if (!result.isOK()) { + warning() << "couldn't create ns_1_min_1 index on config db" << causedBy(result); + } - result = clusterCreateIndex( ChunkType::ConfigNS, - BSON( ChunkType::ns() << 1 << - ChunkType::DEPRECATED_lastmod() << 1 ), - true, // unique - NULL ); + result = clusterCreateIndex( + ChunkType::ConfigNS, + BSON(ChunkType::ns() << 1 << ChunkType::shard() << 1 << ChunkType::min() << 1), + true, // unique + NULL); - if (!result.isOK()) { - warning() << "couldn't create ns_1_lastmod_1 index on config db" << causedBy(result); - } + if (!result.isOK()) { + warning() << "couldn't create ns_1_shard_1_min_1 index on config db" << causedBy(result); + } - result = clusterCreateIndex( ShardType::ConfigNS, - BSON( ShardType::host() << 1 ), - true, // unique - NULL ); + result = clusterCreateIndex(ChunkType::ConfigNS, + BSON(ChunkType::ns() << 1 << ChunkType::DEPRECATED_lastmod() << 1), + true, // unique + NULL); - if (!result.isOK()) { - warning() << "couldn't create host_1 index on config db" << causedBy(result); - } + if (!result.isOK()) { + warning() << "couldn't create ns_1_lastmod_1 index on config db" << causedBy(result); + } - result = clusterCreateIndex(LocksType::ConfigNS, - BSON(LocksType::lockID() << 1), - false, // unique - NULL); + result = clusterCreateIndex(ShardType::ConfigNS, + BSON(ShardType::host() << 1), + true, // unique + NULL); - if (!result.isOK()) { - warning() << "couldn't create lock id index on config db" << causedBy(result); - } + if (!result.isOK()) { + warning() << "couldn't create host_1 index on config db" << causedBy(result); + } - result = clusterCreateIndex( LocksType::ConfigNS, - BSON( LocksType::state() << 1 << - LocksType::process() << 1 ), - false, // unique - NULL ); + result = clusterCreateIndex(LocksType::ConfigNS, + BSON(LocksType::lockID() << 1), + false, // unique + NULL); - if (!result.isOK()) { - warning() << "couldn't create state and process id index on config db" - << causedBy(result); - } + if (!result.isOK()) { + warning() << "couldn't create lock id index on config db" << causedBy(result); + } - result = clusterCreateIndex( LockpingsType::ConfigNS, - BSON( LockpingsType::ping() << 1 ), - false, // unique - NULL ); + result = clusterCreateIndex(LocksType::ConfigNS, + BSON(LocksType::state() << 1 << LocksType::process() << 1), + false, // unique + NULL); - if (!result.isOK()) { - warning() << "couldn't create lockping ping time index on config db" - << causedBy(result); - } + if (!result.isOK()) { + warning() << "couldn't create state and process id index on config db" << causedBy(result); + } - result = clusterCreateIndex(TagsType::ConfigNS, - BSON(TagsType::ns() << 1 << TagsType::min() << 1), - true, // unique - NULL); + result = clusterCreateIndex(LockpingsType::ConfigNS, + BSON(LockpingsType::ping() << 1), + false, // unique + NULL); - if (!result.isOK()) { - warning() << "could not create index ns_1_min_1: " << causedBy(result); - } + if (!result.isOK()) { + warning() << "couldn't create lockping ping time index on config db" << causedBy(result); } - void ConfigServer::replicaSetChange(const string& setName, const string& newConnectionString) { - // This is run in it's own thread. Exceptions escaping would result in a call to terminate. - Client::initThread("replSetChange"); + result = clusterCreateIndex(TagsType::ConfigNS, + BSON(TagsType::ns() << 1 << TagsType::min() << 1), + true, // unique + NULL); - try { - ShardPtr s = Shard::lookupRSName(setName); - if (!s) { - LOG(1) << "shard not found for set: " << newConnectionString; - return; - } + if (!result.isOK()) { + warning() << "could not create index ns_1_min_1: " << causedBy(result); + } +} - Status result = grid.catalogManager()->update( - ShardType::ConfigNS, - BSON(ShardType::name(s->getId())), - BSON("$set" << BSON(ShardType::host( - newConnectionString))), - false, // upsert - false, // multi - NULL); - if (!result.isOK()) { - error() << "RSChangeWatcher: could not update config db for set: " - << setName - << " to: " << newConnectionString - << ": " << result.reason(); - } - } - catch (const std::exception& e) { - log() << "caught exception while updating config servers: " << e.what(); +void ConfigServer::replicaSetChange(const string& setName, const string& newConnectionString) { + // This is run in it's own thread. Exceptions escaping would result in a call to terminate. + Client::initThread("replSetChange"); + + try { + ShardPtr s = Shard::lookupRSName(setName); + if (!s) { + LOG(1) << "shard not found for set: " << newConnectionString; + return; } - catch (...) { - log() << "caught unknown exception while updating config servers"; + + Status result = grid.catalogManager()->update( + ShardType::ConfigNS, + BSON(ShardType::name(s->getId())), + BSON("$set" << BSON(ShardType::host(newConnectionString))), + false, // upsert + false, // multi + NULL); + if (!result.isOK()) { + error() << "RSChangeWatcher: could not update config db for set: " << setName + << " to: " << newConnectionString << ": " << result.reason(); } + } catch (const std::exception& e) { + log() << "caught exception while updating config servers: " << e.what(); + } catch (...) { + log() << "caught unknown exception while updating config servers"; } +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/config.h b/src/mongo/s/config.h index e2216b67786..9622e83f3ac 100644 --- a/src/mongo/s/config.h +++ b/src/mongo/s/config.h @@ -36,149 +36,167 @@ namespace mongo { - class ChunkManager; - class CollectionType; - class DatabaseType; - class DBConfig; +class ChunkManager; +class CollectionType; +class DatabaseType; +class DBConfig; - typedef std::shared_ptr<DBConfig> DBConfigPtr; +typedef std::shared_ptr<DBConfig> DBConfigPtr; - struct CollectionInfo { - CollectionInfo() { - _dirty = false; - _dropped = false; - } +struct CollectionInfo { + CollectionInfo() { + _dirty = false; + _dropped = false; + } - CollectionInfo(const CollectionType& in); - ~CollectionInfo(); + CollectionInfo(const CollectionType& in); + ~CollectionInfo(); - bool isSharded() const { - return _cm.get(); - } + bool isSharded() const { + return _cm.get(); + } - std::shared_ptr<ChunkManager> getCM() const { - return _cm; - } + std::shared_ptr<ChunkManager> getCM() const { + return _cm; + } - void resetCM(ChunkManager * cm); + void resetCM(ChunkManager* cm); - void shard(ChunkManager* cm); - void unshard(); + void shard(ChunkManager* cm); + void unshard(); - bool isDirty() const { return _dirty; } - bool wasDropped() const { return _dropped; } + bool isDirty() const { + return _dirty; + } + bool wasDropped() const { + return _dropped; + } - void save(const std::string& ns); + void save(const std::string& ns); - void useChunkManager(std::shared_ptr<ChunkManager> manager); + void useChunkManager(std::shared_ptr<ChunkManager> manager); - bool unique() const { return _unique; } - BSONObj key() const { return _key; } + bool unique() const { + return _unique; + } + BSONObj key() const { + return _key; + } - private: - BSONObj _key; - bool _unique; - std::shared_ptr<ChunkManager> _cm; - bool _dirty; - bool _dropped; - }; +private: + BSONObj _key; + bool _unique; + std::shared_ptr<ChunkManager> _cm; + bool _dirty; + bool _dropped; +}; + +/** + * top level configuration for a database + */ +class DBConfig { +public: + DBConfig(std::string name, const DatabaseType& dbt); /** - * top level configuration for a database + * The name of the database which this entry caches. */ - class DBConfig { - public: - DBConfig(std::string name, const DatabaseType& dbt); - - /** - * The name of the database which this entry caches. - */ - const std::string& name() const { return _name; } + const std::string& name() const { + return _name; + } - /** - * Whether sharding is enabled for this database. - */ - bool isShardingEnabled() const { return _shardingEnabled; } + /** + * Whether sharding is enabled for this database. + */ + bool isShardingEnabled() const { + return _shardingEnabled; + } - const ShardId& getPrimaryId() const { return _primaryId; } + const ShardId& getPrimaryId() const { + return _primaryId; + } - void enableSharding( bool save = true ); + void enableSharding(bool save = true); - /** - @return true if there was sharding info to remove - */ - bool removeSharding( const std::string& ns ); + /** + @return true if there was sharding info to remove + */ + bool removeSharding(const std::string& ns); - /** - * @return whether or not the 'ns' collection is partitioned - */ - bool isSharded( const std::string& ns ); + /** + * @return whether or not the 'ns' collection is partitioned + */ + bool isSharded(const std::string& ns); - // Atomically returns *either* the chunk manager *or* the primary shard for the collection, - // neither if the collection doesn't exist. - void getChunkManagerOrPrimary(const std::string& ns, - std::shared_ptr<ChunkManager>& manager, - std::shared_ptr<Shard>& primary); + // Atomically returns *either* the chunk manager *or* the primary shard for the collection, + // neither if the collection doesn't exist. + void getChunkManagerOrPrimary(const std::string& ns, + std::shared_ptr<ChunkManager>& manager, + std::shared_ptr<Shard>& primary); - std::shared_ptr<ChunkManager> getChunkManager(const std::string& ns, bool reload = false, bool forceReload = false); - std::shared_ptr<ChunkManager> getChunkManagerIfExists(const std::string& ns, bool reload = false, bool forceReload = false); + std::shared_ptr<ChunkManager> getChunkManager(const std::string& ns, + bool reload = false, + bool forceReload = false); + std::shared_ptr<ChunkManager> getChunkManagerIfExists(const std::string& ns, + bool reload = false, + bool forceReload = false); - /** - * Returns shard id for primary shard for the database for which this DBConfig represents. - */ - const ShardId& getShardId(const std::string& ns); + /** + * Returns shard id for primary shard for the database for which this DBConfig represents. + */ + const ShardId& getShardId(const std::string& ns); - void setPrimary( const std::string& s ); + void setPrimary(const std::string& s); - bool load(); - bool reload(); + bool load(); + bool reload(); - bool dropDatabase( std::string& errmsg ); + bool dropDatabase(std::string& errmsg); - void getAllShardIds(std::set<ShardId>* shardIds); - void getAllShardedCollections(std::set<std::string>& namespaces); + void getAllShardIds(std::set<ShardId>* shardIds); + void getAllShardedCollections(std::set<std::string>& namespaces); - protected: - typedef std::map<std::string, CollectionInfo> CollectionInfoMap; +protected: + typedef std::map<std::string, CollectionInfo> CollectionInfoMap; - /** - lockless - */ - bool _isSharded( const std::string& ns ); + /** + lockless + */ + bool _isSharded(const std::string& ns); - bool _dropShardedCollections(int& num, std::set<ShardId>& shardIds, std::string& errmsg); + bool _dropShardedCollections(int& num, std::set<ShardId>& shardIds, std::string& errmsg); - bool _load(); - bool _reload(); - void _save( bool db = true, bool coll = true ); + bool _load(); + bool _reload(); + void _save(bool db = true, bool coll = true); - // Name of the database which this entry caches - const std::string _name; + // Name of the database which this entry caches + const std::string _name; - // Primary shard id - ShardId _primaryId; + // Primary shard id + ShardId _primaryId; - // Whether sharding has been enabled for this database - bool _shardingEnabled; + // Whether sharding has been enabled for this database + bool _shardingEnabled; - // Set of collections and lock to protect access - stdx::mutex _lock; - CollectionInfoMap _collections; + // Set of collections and lock to protect access + stdx::mutex _lock; + CollectionInfoMap _collections; - // Ensures that only one thread at a time loads collection configuration data from - // the config server - stdx::mutex _hitConfigServerLock; - }; + // Ensures that only one thread at a time loads collection configuration data from + // the config server + stdx::mutex _hitConfigServerLock; +}; - class ConfigServer { - public: - static void reloadSettings(); +class ConfigServer { +public: + static void reloadSettings(); - static void replicaSetChange(const std::string& setName, - const std::string& newConnectionString); - }; + static void replicaSetChange(const std::string& setName, + const std::string& newConnectionString); +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/cursors.cpp b/src/mongo/s/cursors.cpp index 31f62c1e2b6..6b4b4e6f2c6 100644 --- a/src/mongo/s/cursors.cpp +++ b/src/mongo/s/cursors.cpp @@ -54,466 +54,461 @@ namespace mongo { - using std::unique_ptr; - using std::endl; - using std::string; - using std::stringstream; - - const int ShardedClientCursor::INIT_REPLY_BUFFER_SIZE = 32768; - - // Note: There is no counter for shardedEver from cursorInfo since it is deprecated - static Counter64 cursorStatsMultiTarget; - static Counter64 cursorStatsSingleTarget; - - // Simple class to report the sum total open cursors = sharded + refs - class CursorStatsSum { - public: - operator long long() const { - return get(); - } - long long get() const { - return cursorStatsMultiTarget.get() + cursorStatsSingleTarget.get(); - } - }; - - static CursorStatsSum cursorStatsTotalOpen; +using std::unique_ptr; +using std::endl; +using std::string; +using std::stringstream; + +const int ShardedClientCursor::INIT_REPLY_BUFFER_SIZE = 32768; + +// Note: There is no counter for shardedEver from cursorInfo since it is deprecated +static Counter64 cursorStatsMultiTarget; +static Counter64 cursorStatsSingleTarget; + +// Simple class to report the sum total open cursors = sharded + refs +class CursorStatsSum { +public: + operator long long() const { + return get(); + } + long long get() const { + return cursorStatsMultiTarget.get() + cursorStatsSingleTarget.get(); + } +}; - static ServerStatusMetricField<Counter64> dCursorStatsMultiTarget( "cursor.open.multiTarget", - &cursorStatsMultiTarget); - static ServerStatusMetricField<Counter64> dCursorStatsSingleTarget( "cursor.open.singleTarget", - &cursorStatsSingleTarget); - static ServerStatusMetricField<CursorStatsSum> dCursorStatsTotalOpen( "cursor.open.total", - &cursorStatsTotalOpen); +static CursorStatsSum cursorStatsTotalOpen; +static ServerStatusMetricField<Counter64> dCursorStatsMultiTarget("cursor.open.multiTarget", + &cursorStatsMultiTarget); +static ServerStatusMetricField<Counter64> dCursorStatsSingleTarget("cursor.open.singleTarget", + &cursorStatsSingleTarget); +static ServerStatusMetricField<CursorStatsSum> dCursorStatsTotalOpen("cursor.open.total", + &cursorStatsTotalOpen); - // -------- ShardedCursor ----------- - ShardedClientCursor::ShardedClientCursor( QueryMessage& q, - ParallelSortClusteredCursor * cursor ) { - verify( cursor ); - _cursor = cursor; +// -------- ShardedCursor ----------- - _skip = q.ntoskip; - _ntoreturn = q.ntoreturn; +ShardedClientCursor::ShardedClientCursor(QueryMessage& q, ParallelSortClusteredCursor* cursor) { + verify(cursor); + _cursor = cursor; - _totalSent = 0; - _done = false; + _skip = q.ntoskip; + _ntoreturn = q.ntoreturn; - _id = 0; + _totalSent = 0; + _done = false; - if ( q.queryOptions & QueryOption_NoCursorTimeout ) { - _lastAccessMillis = 0; - } - else - _lastAccessMillis = Listener::getElapsedTimeMillis(); + _id = 0; - cursorStatsMultiTarget.increment(); - } + if (q.queryOptions & QueryOption_NoCursorTimeout) { + _lastAccessMillis = 0; + } else + _lastAccessMillis = Listener::getElapsedTimeMillis(); - ShardedClientCursor::~ShardedClientCursor() { - verify( _cursor ); - delete _cursor; - _cursor = 0; - cursorStatsMultiTarget.decrement(); - } + cursorStatsMultiTarget.increment(); +} - long long ShardedClientCursor::getId() { - if ( _id <= 0 ) { - _id = cursorCache.genId(); - verify( _id >= 0 ); - } - return _id; - } +ShardedClientCursor::~ShardedClientCursor() { + verify(_cursor); + delete _cursor; + _cursor = 0; + cursorStatsMultiTarget.decrement(); +} - int ShardedClientCursor::getTotalSent() const { - return _totalSent; +long long ShardedClientCursor::getId() { + if (_id <= 0) { + _id = cursorCache.genId(); + verify(_id >= 0); } + return _id; +} - void ShardedClientCursor::accessed() { - if ( _lastAccessMillis > 0 ) - _lastAccessMillis = Listener::getElapsedTimeMillis(); - } +int ShardedClientCursor::getTotalSent() const { + return _totalSent; +} - long long ShardedClientCursor::idleTime( long long now ) { - if ( _lastAccessMillis == 0 ) - return 0; - return now - _lastAccessMillis; - } +void ShardedClientCursor::accessed() { + if (_lastAccessMillis > 0) + _lastAccessMillis = Listener::getElapsedTimeMillis(); +} - bool ShardedClientCursor::sendNextBatch(int ntoreturn, BufBuilder& buffer, int& docCount) { - uassert( 10191 , "cursor already done" , ! _done ); - - int maxSize = 1024 * 1024; - if ( _totalSent > 0 ) - maxSize *= 3; - - docCount = 0; - - // If ntoreturn is negative, it means that we should send up to -ntoreturn results - // back to the client, and that we should only send a *single batch*. An ntoreturn of - // 1 is also a special case which means "return up to 1 result in a single batch" (so - // that +1 actually has the same meaning of -1). For all other values of ntoreturn, we - // may have to return multiple batches. - const bool sendMoreBatches = ntoreturn == 0 || ntoreturn > 1; - ntoreturn = abs( ntoreturn ); - - bool cursorHasMore = true; - while ( ( cursorHasMore = _cursor->more() ) ) { - BSONObj o = _cursor->next(); - - buffer.appendBuf( (void*)o.objdata() , o.objsize() ); - docCount++; - // Ensure that the next batch will never wind up requesting more docs from the shard - // than are remaining to satisfy the initial ntoreturn. - if (ntoreturn != 0) { - _cursor->setBatchSize(ntoreturn - docCount); - } +long long ShardedClientCursor::idleTime(long long now) { + if (_lastAccessMillis == 0) + return 0; + return now - _lastAccessMillis; +} - if ( buffer.len() > maxSize ) { - break; - } +bool ShardedClientCursor::sendNextBatch(int ntoreturn, BufBuilder& buffer, int& docCount) { + uassert(10191, "cursor already done", !_done); + + int maxSize = 1024 * 1024; + if (_totalSent > 0) + maxSize *= 3; + + docCount = 0; + + // If ntoreturn is negative, it means that we should send up to -ntoreturn results + // back to the client, and that we should only send a *single batch*. An ntoreturn of + // 1 is also a special case which means "return up to 1 result in a single batch" (so + // that +1 actually has the same meaning of -1). For all other values of ntoreturn, we + // may have to return multiple batches. + const bool sendMoreBatches = ntoreturn == 0 || ntoreturn > 1; + ntoreturn = abs(ntoreturn); + + bool cursorHasMore = true; + while ((cursorHasMore = _cursor->more())) { + BSONObj o = _cursor->next(); + + buffer.appendBuf((void*)o.objdata(), o.objsize()); + docCount++; + // Ensure that the next batch will never wind up requesting more docs from the shard + // than are remaining to satisfy the initial ntoreturn. + if (ntoreturn != 0) { + _cursor->setBatchSize(ntoreturn - docCount); + } - if ( docCount == ntoreturn ) { - // soft limit aka batch size - break; - } + if (buffer.len() > maxSize) { + break; + } - if ( ntoreturn == 0 && _totalSent == 0 && docCount >= 100 ) { - // first batch should be max 100 unless batch size specified - break; - } + if (docCount == ntoreturn) { + // soft limit aka batch size + break; } - // We need to request another batch if the following two conditions hold: - // - // 1. ntoreturn is positive and not equal to 1 (see the comment above). This condition - // is stored in 'sendMoreBatches'. - // - // 2. The last call to _cursor->more() was true (i.e. we never explicitly got a false - // value from _cursor->more()). This condition is stored in 'cursorHasMore'. If the server - // hits EOF while executing a query or a getmore, it will pass a cursorId of 0 in the - // query response to indicate that there are no more results. In this case, _cursor->more() - // will be explicitly false, and we know for sure that we do not have to send more batches. - // - // On the other hand, if _cursor->more() is true there may or may not be more results. - // Suppose that the mongod generates enough results to fill this batch. In this case it - // does not know whether not there are more, because doing so would require requesting an - // extra result and seeing whether we get EOF. The mongod sends a valid cursorId to - // indicate that there may be more. We do the same here: we indicate that there may be - // more results to retrieve by setting 'hasMoreBatches' to true. - bool hasMoreBatches = sendMoreBatches && cursorHasMore; - - LOG(5) << "\t hasMoreBatches: " << hasMoreBatches - << " sendMoreBatches: " << sendMoreBatches - << " cursorHasMore: " << cursorHasMore - << " ntoreturn: " << ntoreturn - << " num: " << docCount - << " id:" << getId() - << " totalSent: " << _totalSent << endl; - - _totalSent += docCount; - _done = ! hasMoreBatches; - - return hasMoreBatches; + if (ntoreturn == 0 && _totalSent == 0 && docCount >= 100) { + // first batch should be max 100 unless batch size specified + break; + } } - // ---- CursorCache ----- + // We need to request another batch if the following two conditions hold: + // + // 1. ntoreturn is positive and not equal to 1 (see the comment above). This condition + // is stored in 'sendMoreBatches'. + // + // 2. The last call to _cursor->more() was true (i.e. we never explicitly got a false + // value from _cursor->more()). This condition is stored in 'cursorHasMore'. If the server + // hits EOF while executing a query or a getmore, it will pass a cursorId of 0 in the + // query response to indicate that there are no more results. In this case, _cursor->more() + // will be explicitly false, and we know for sure that we do not have to send more batches. + // + // On the other hand, if _cursor->more() is true there may or may not be more results. + // Suppose that the mongod generates enough results to fill this batch. In this case it + // does not know whether not there are more, because doing so would require requesting an + // extra result and seeing whether we get EOF. The mongod sends a valid cursorId to + // indicate that there may be more. We do the same here: we indicate that there may be + // more results to retrieve by setting 'hasMoreBatches' to true. + bool hasMoreBatches = sendMoreBatches && cursorHasMore; + + LOG(5) << "\t hasMoreBatches: " << hasMoreBatches << " sendMoreBatches: " << sendMoreBatches + << " cursorHasMore: " << cursorHasMore << " ntoreturn: " << ntoreturn + << " num: " << docCount << " id:" << getId() << " totalSent: " << _totalSent << endl; + + _totalSent += docCount; + _done = !hasMoreBatches; + + return hasMoreBatches; +} - long long CursorCache::TIMEOUT = 10 * 60 * 1000 /* 10 minutes */; - ExportedServerParameter<long long> cursorCacheTimeoutConfig(ServerParameterSet::getGlobal(), - "cursorTimeoutMillis", - &CursorCache::TIMEOUT, - true, true); +// ---- CursorCache ----- - unsigned getCCRandomSeed() { - unique_ptr<SecureRandom> sr( SecureRandom::create() ); - return sr->nextInt64(); - } +long long CursorCache::TIMEOUT = 10 * 60 * 1000 /* 10 minutes */; +ExportedServerParameter<long long> cursorCacheTimeoutConfig( + ServerParameterSet::getGlobal(), "cursorTimeoutMillis", &CursorCache::TIMEOUT, true, true); - CursorCache::CursorCache() - : _random( getCCRandomSeed() ), - _shardedTotal(0) { - } +unsigned getCCRandomSeed() { + unique_ptr<SecureRandom> sr(SecureRandom::create()); + return sr->nextInt64(); +} - CursorCache::~CursorCache() { - // TODO: delete old cursors? - bool print = shouldLog(logger::LogSeverity::Debug(1)); - if ( _cursors.size() || _refs.size() ) - print = true; - verify(_refs.size() == _refsNS.size()); - - if ( print ) - log() << " CursorCache at shutdown - " - << " sharded: " << _cursors.size() - << " passthrough: " << _refs.size() - << endl; - } +CursorCache::CursorCache() : _random(getCCRandomSeed()), _shardedTotal(0) {} - ShardedClientCursorPtr CursorCache::get( long long id ) const { - LOG(_myLogLevel) << "CursorCache::get id: " << id << endl; - stdx::lock_guard<stdx::mutex> lk( _mutex ); - MapSharded::const_iterator i = _cursors.find( id ); - if ( i == _cursors.end() ) { - return ShardedClientCursorPtr(); - } - i->second->accessed(); - return i->second; - } +CursorCache::~CursorCache() { + // TODO: delete old cursors? + bool print = shouldLog(logger::LogSeverity::Debug(1)); + if (_cursors.size() || _refs.size()) + print = true; + verify(_refs.size() == _refsNS.size()); - int CursorCache::getMaxTimeMS( long long id ) const { - verify( id ); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - MapShardedInt::const_iterator i = _cursorsMaxTimeMS.find( id ); - return ( i != _cursorsMaxTimeMS.end() ) ? i->second : 0; - } + if (print) + log() << " CursorCache at shutdown - " + << " sharded: " << _cursors.size() << " passthrough: " << _refs.size() << endl; +} - void CursorCache::store( ShardedClientCursorPtr cursor, int maxTimeMS ) { - LOG(_myLogLevel) << "CursorCache::store cursor " << " id: " << cursor->getId() - << (maxTimeMS != kMaxTimeCursorNoTimeLimit ? str::stream() << "maxTimeMS: " << maxTimeMS - : string("")) - << endl; - verify( cursor->getId() ); - verify( maxTimeMS == kMaxTimeCursorTimeLimitExpired - || maxTimeMS == kMaxTimeCursorNoTimeLimit - || maxTimeMS > 0 ); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - _cursorsMaxTimeMS[cursor->getId()] = maxTimeMS; - _cursors[cursor->getId()] = cursor; - _shardedTotal++; +ShardedClientCursorPtr CursorCache::get(long long id) const { + LOG(_myLogLevel) << "CursorCache::get id: " << id << endl; + stdx::lock_guard<stdx::mutex> lk(_mutex); + MapSharded::const_iterator i = _cursors.find(id); + if (i == _cursors.end()) { + return ShardedClientCursorPtr(); } + i->second->accessed(); + return i->second; +} - void CursorCache::updateMaxTimeMS( long long id, int maxTimeMS ) { - verify( id ); - verify( maxTimeMS == kMaxTimeCursorTimeLimitExpired - || maxTimeMS == kMaxTimeCursorNoTimeLimit - || maxTimeMS > 0 ); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - _cursorsMaxTimeMS[id] = maxTimeMS; - } +int CursorCache::getMaxTimeMS(long long id) const { + verify(id); + stdx::lock_guard<stdx::mutex> lk(_mutex); + MapShardedInt::const_iterator i = _cursorsMaxTimeMS.find(id); + return (i != _cursorsMaxTimeMS.end()) ? i->second : 0; +} - void CursorCache::remove( long long id ) { - verify( id ); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - _cursorsMaxTimeMS.erase( id ); - _cursors.erase( id ); - } +void CursorCache::store(ShardedClientCursorPtr cursor, int maxTimeMS) { + LOG(_myLogLevel) << "CursorCache::store cursor " + << " id: " << cursor->getId() + << (maxTimeMS != kMaxTimeCursorNoTimeLimit + ? str::stream() << "maxTimeMS: " << maxTimeMS + : string("")) << endl; + verify(cursor->getId()); + verify(maxTimeMS == kMaxTimeCursorTimeLimitExpired || maxTimeMS == kMaxTimeCursorNoTimeLimit || + maxTimeMS > 0); + stdx::lock_guard<stdx::mutex> lk(_mutex); + _cursorsMaxTimeMS[cursor->getId()] = maxTimeMS; + _cursors[cursor->getId()] = cursor; + _shardedTotal++; +} - void CursorCache::removeRef( long long id ) { - verify( id ); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - _refs.erase( id ); - _refsNS.erase( id ); - cursorStatsSingleTarget.decrement(); - } +void CursorCache::updateMaxTimeMS(long long id, int maxTimeMS) { + verify(id); + verify(maxTimeMS == kMaxTimeCursorTimeLimitExpired || maxTimeMS == kMaxTimeCursorNoTimeLimit || + maxTimeMS > 0); + stdx::lock_guard<stdx::mutex> lk(_mutex); + _cursorsMaxTimeMS[id] = maxTimeMS; +} - void CursorCache::storeRef(const std::string& server, long long id, const std::string& ns) { - LOG(_myLogLevel) << "CursorCache::storeRef server: " << server << " id: " << id << endl; - verify( id ); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - _refs[id] = server; - _refsNS[id] = ns; - cursorStatsSingleTarget.increment(); - } +void CursorCache::remove(long long id) { + verify(id); + stdx::lock_guard<stdx::mutex> lk(_mutex); + _cursorsMaxTimeMS.erase(id); + _cursors.erase(id); +} - string CursorCache::getRef( long long id ) const { - verify( id ); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - MapNormal::const_iterator i = _refs.find( id ); +void CursorCache::removeRef(long long id) { + verify(id); + stdx::lock_guard<stdx::mutex> lk(_mutex); + _refs.erase(id); + _refsNS.erase(id); + cursorStatsSingleTarget.decrement(); +} - LOG(_myLogLevel) << "CursorCache::getRef id: " << id << " out: " << ( i == _refs.end() ? " NONE " : i->second ) << endl; +void CursorCache::storeRef(const std::string& server, long long id, const std::string& ns) { + LOG(_myLogLevel) << "CursorCache::storeRef server: " << server << " id: " << id << endl; + verify(id); + stdx::lock_guard<stdx::mutex> lk(_mutex); + _refs[id] = server; + _refsNS[id] = ns; + cursorStatsSingleTarget.increment(); +} - if ( i == _refs.end() ) - return ""; - return i->second; - } +string CursorCache::getRef(long long id) const { + verify(id); + stdx::lock_guard<stdx::mutex> lk(_mutex); + MapNormal::const_iterator i = _refs.find(id); - std::string CursorCache::getRefNS(long long id) const { - verify(id); - stdx::lock_guard<stdx::mutex> lk(_mutex); - MapNormal::const_iterator i = _refsNS.find(id); + LOG(_myLogLevel) << "CursorCache::getRef id: " << id + << " out: " << (i == _refs.end() ? " NONE " : i->second) << endl; - LOG(_myLogLevel) << "CursorCache::getRefNs id: " << id - << " out: " << ( i == _refsNS.end() ? " NONE " : i->second ) << std::endl; + if (i == _refs.end()) + return ""; + return i->second; +} - if ( i == _refsNS.end() ) - return ""; - return i->second; - } +std::string CursorCache::getRefNS(long long id) const { + verify(id); + stdx::lock_guard<stdx::mutex> lk(_mutex); + MapNormal::const_iterator i = _refsNS.find(id); + LOG(_myLogLevel) << "CursorCache::getRefNs id: " << id + << " out: " << (i == _refsNS.end() ? " NONE " : i->second) << std::endl; - long long CursorCache::genId() { - while ( true ) { - stdx::lock_guard<stdx::mutex> lk( _mutex ); + if (i == _refsNS.end()) + return ""; + return i->second; +} - long long x = Listener::getElapsedTimeMillis() << 32; - x |= _random.nextInt32(); - if ( x == 0 ) - continue; +long long CursorCache::genId() { + while (true) { + stdx::lock_guard<stdx::mutex> lk(_mutex); - if ( x < 0 ) - x *= -1; + long long x = Listener::getElapsedTimeMillis() << 32; + x |= _random.nextInt32(); - MapSharded::iterator i = _cursors.find( x ); - if ( i != _cursors.end() ) - continue; + if (x == 0) + continue; - MapNormal::iterator j = _refs.find( x ); - if ( j != _refs.end() ) - continue; + if (x < 0) + x *= -1; - return x; - } + MapSharded::iterator i = _cursors.find(x); + if (i != _cursors.end()) + continue; + + MapNormal::iterator j = _refs.find(x); + if (j != _refs.end()) + continue; + + return x; } +} - void CursorCache::gotKillCursors(Message& m ) { - LastError::get(cc()).disable(); - DbMessage dbmessage(m); - int n = dbmessage.pullInt(); +void CursorCache::gotKillCursors(Message& m) { + LastError::get(cc()).disable(); + DbMessage dbmessage(m); + int n = dbmessage.pullInt(); - if ( n > 2000 ) { - ( n < 30000 ? warning() : error() ) << "receivedKillCursors, n=" << n << endl; - } + if (n > 2000) { + (n < 30000 ? warning() : error()) << "receivedKillCursors, n=" << n << endl; + } - uassert( 13286 , "sent 0 cursors to kill" , n >= 1 ); - uassert( 13287 , "too many cursors to kill" , n < 30000 ); - massert( 18632 , str::stream() << "bad kill cursors size: " << m.dataSize(), - m.dataSize() == 8 + ( 8 * n ) ); + uassert(13286, "sent 0 cursors to kill", n >= 1); + uassert(13287, "too many cursors to kill", n < 30000); + massert(18632, + str::stream() << "bad kill cursors size: " << m.dataSize(), + m.dataSize() == 8 + (8 * n)); - ConstDataCursor cursors(dbmessage.getArray(n)); + ConstDataCursor cursors(dbmessage.getArray(n)); - ClientBasic* client = ClientBasic::getCurrent(); - AuthorizationSession* authSession = AuthorizationSession::get(client); - for ( int i=0; i<n; i++ ) { - long long id = cursors.readAndAdvance<LittleEndian<int64_t>>(); - LOG(_myLogLevel) << "CursorCache::gotKillCursors id: " << id << endl; + ClientBasic* client = ClientBasic::getCurrent(); + AuthorizationSession* authSession = AuthorizationSession::get(client); + for (int i = 0; i < n; i++) { + long long id = cursors.readAndAdvance<LittleEndian<int64_t>>(); + LOG(_myLogLevel) << "CursorCache::gotKillCursors id: " << id << endl; - if ( ! id ) { - warning() << " got cursor id of 0 to kill" << endl; - continue; - } + if (!id) { + warning() << " got cursor id of 0 to kill" << endl; + continue; + } - string server; - { - stdx::lock_guard<stdx::mutex> lk( _mutex ); - - MapSharded::iterator i = _cursors.find( id ); - if ( i != _cursors.end() ) { - Status authorizationStatus = authSession->checkAuthForKillCursors( - NamespaceString(i->second->getNS()), id); - audit::logKillCursorsAuthzCheck( - client, - NamespaceString(i->second->getNS()), - id, - authorizationStatus.isOK() ? ErrorCodes::OK : ErrorCodes::Unauthorized); - if (authorizationStatus.isOK()) { - _cursorsMaxTimeMS.erase( i->second->getId() ); - _cursors.erase( i ); - } - continue; - } + string server; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); - MapNormal::iterator refsIt = _refs.find(id); - MapNormal::iterator refsNSIt = _refsNS.find(id); - if (refsIt == _refs.end()) { - warning() << "can't find cursor: " << id << endl; - continue; - } - verify(refsNSIt != _refsNS.end()); - Status authorizationStatus = authSession->checkAuthForKillCursors( - NamespaceString(refsNSIt->second), id); + MapSharded::iterator i = _cursors.find(id); + if (i != _cursors.end()) { + Status authorizationStatus = + authSession->checkAuthForKillCursors(NamespaceString(i->second->getNS()), id); audit::logKillCursorsAuthzCheck( - client, - NamespaceString(refsNSIt->second), - id, - authorizationStatus.isOK() ? ErrorCodes::OK : ErrorCodes::Unauthorized); - if (!authorizationStatus.isOK()) { - continue; + client, + NamespaceString(i->second->getNS()), + id, + authorizationStatus.isOK() ? ErrorCodes::OK : ErrorCodes::Unauthorized); + if (authorizationStatus.isOK()) { + _cursorsMaxTimeMS.erase(i->second->getId()); + _cursors.erase(i); } - server = refsIt->second; - _refs.erase(refsIt); - _refsNS.erase(refsNSIt); - cursorStatsSingleTarget.decrement(); + continue; } - LOG(_myLogLevel) << "CursorCache::found gotKillCursors id: " << id << " server: " << server << endl; - - verify( server.size() ); - ScopedDbConnection conn(server); - conn->killCursor( id ); - conn.done(); + MapNormal::iterator refsIt = _refs.find(id); + MapNormal::iterator refsNSIt = _refsNS.find(id); + if (refsIt == _refs.end()) { + warning() << "can't find cursor: " << id << endl; + continue; + } + verify(refsNSIt != _refsNS.end()); + Status authorizationStatus = + authSession->checkAuthForKillCursors(NamespaceString(refsNSIt->second), id); + audit::logKillCursorsAuthzCheck(client, + NamespaceString(refsNSIt->second), + id, + authorizationStatus.isOK() ? ErrorCodes::OK + : ErrorCodes::Unauthorized); + if (!authorizationStatus.isOK()) { + continue; + } + server = refsIt->second; + _refs.erase(refsIt); + _refsNS.erase(refsNSIt); + cursorStatsSingleTarget.decrement(); } - } - void CursorCache::appendInfo( BSONObjBuilder& result ) const { - stdx::lock_guard<stdx::mutex> lk( _mutex ); - result.append( "sharded", static_cast<int>(cursorStatsMultiTarget.get())); - result.appendNumber( "shardedEver" , _shardedTotal ); - result.append( "refs", static_cast<int>(cursorStatsSingleTarget.get())); - result.append( "totalOpen", static_cast<int>(cursorStatsTotalOpen.get())); + LOG(_myLogLevel) << "CursorCache::found gotKillCursors id: " << id << " server: " << server + << endl; + + verify(server.size()); + ScopedDbConnection conn(server); + conn->killCursor(id); + conn.done(); } +} - void CursorCache::doTimeouts() { - long long now = Listener::getElapsedTimeMillis(); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - for ( MapSharded::iterator i=_cursors.begin(); i!=_cursors.end(); ++i ) { - // Note: cursors with no timeout will always have an idleTime of 0 - long long idleFor = i->second->idleTime( now ); - if ( idleFor < TIMEOUT ) { - continue; - } - log() << "killing old cursor " << i->second->getId() << " idle for: " << idleFor << "ms" << endl; // TODO: make LOG(1) - _cursorsMaxTimeMS.erase( i->second->getId() ); - _cursors.erase( i ); - i = _cursors.begin(); // possible 2nd entry will get skipped, will get on next pass - if ( i == _cursors.end() ) - break; +void CursorCache::appendInfo(BSONObjBuilder& result) const { + stdx::lock_guard<stdx::mutex> lk(_mutex); + result.append("sharded", static_cast<int>(cursorStatsMultiTarget.get())); + result.appendNumber("shardedEver", _shardedTotal); + result.append("refs", static_cast<int>(cursorStatsSingleTarget.get())); + result.append("totalOpen", static_cast<int>(cursorStatsTotalOpen.get())); +} + +void CursorCache::doTimeouts() { + long long now = Listener::getElapsedTimeMillis(); + stdx::lock_guard<stdx::mutex> lk(_mutex); + for (MapSharded::iterator i = _cursors.begin(); i != _cursors.end(); ++i) { + // Note: cursors with no timeout will always have an idleTime of 0 + long long idleFor = i->second->idleTime(now); + if (idleFor < TIMEOUT) { + continue; } + log() << "killing old cursor " << i->second->getId() << " idle for: " << idleFor << "ms" + << endl; // TODO: make LOG(1) + _cursorsMaxTimeMS.erase(i->second->getId()); + _cursors.erase(i); + i = _cursors.begin(); // possible 2nd entry will get skipped, will get on next pass + if (i == _cursors.end()) + break; } +} - CursorCache cursorCache; - - const int CursorCache::_myLogLevel = 3; +CursorCache cursorCache; - class CursorTimeoutTask : public task::Task { - public: - virtual string name() const { return "cursorTimeout"; } - virtual void doWork() { - cursorCache.doTimeouts(); - } - }; +const int CursorCache::_myLogLevel = 3; - void CursorCache::startTimeoutThread() { - task::repeat( new CursorTimeoutTask , 4000 ); +class CursorTimeoutTask : public task::Task { +public: + virtual string name() const { + return "cursorTimeout"; + } + virtual void doWork() { + cursorCache.doTimeouts(); } +}; - class CmdCursorInfo : public Command { - public: - CmdCursorInfo() : Command( "cursorInfo" ) {} - virtual bool slaveOk() const { return true; } - virtual void help( stringstream& help ) const { - help << " example: { cursorInfo : 1 }"; - } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::cursorInfo); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - virtual bool isWriteCommandForConfigServer() const { return false; } - bool run(OperationContext* txn, - const string&, - BSONObj& jsobj, - int, - string& errmsg, - BSONObjBuilder& result) { - cursorCache.appendInfo( result ); - if ( jsobj["setTimeout"].isNumber() ) - CursorCache::TIMEOUT = jsobj["setTimeout"].numberLong(); - return true; - } - } cmdCursorInfo; +void CursorCache::startTimeoutThread() { + task::repeat(new CursorTimeoutTask, 4000); +} +class CmdCursorInfo : public Command { +public: + CmdCursorInfo() : Command("cursorInfo") {} + virtual bool slaveOk() const { + return true; + } + virtual void help(stringstream& help) const { + help << " example: { cursorInfo : 1 }"; + } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::cursorInfo); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + bool run(OperationContext* txn, + const string&, + BSONObj& jsobj, + int, + string& errmsg, + BSONObjBuilder& result) { + cursorCache.appendInfo(result); + if (jsobj["setTimeout"].isNumber()) + CursorCache::TIMEOUT = jsobj["setTimeout"].numberLong(); + return true; + } +} cmdCursorInfo; } diff --git a/src/mongo/s/cursors.h b/src/mongo/s/cursors.h index 2e0d805402e..4ae1de00771 100644 --- a/src/mongo/s/cursors.h +++ b/src/mongo/s/cursors.h @@ -38,119 +38,120 @@ namespace mongo { - class QueryMessage; +class QueryMessage; - class ShardedClientCursor { - MONGO_DISALLOW_COPYING(ShardedClientCursor); - public: - ShardedClientCursor( QueryMessage& q , ParallelSortClusteredCursor * cursor ); - ~ShardedClientCursor(); +class ShardedClientCursor { + MONGO_DISALLOW_COPYING(ShardedClientCursor); - long long getId(); +public: + ShardedClientCursor(QueryMessage& q, ParallelSortClusteredCursor* cursor); + ~ShardedClientCursor(); - /** - * @return the cumulative number of documents seen by this cursor. - */ - int getTotalSent() const; + long long getId(); - /** - * Sends queries to the shards and gather the result for this batch. - * - * @param r The request object from the client - * @param ntoreturn Number of documents to return - * @param buffer The buffer to use to store the results. - * @param docCount This will contain the number of documents gathered for this batch after - * a successful call. - * - * @return true if this is not the final batch. - */ - bool sendNextBatch(int ntoreturn, BufBuilder& buffer, int& docCount); + /** + * @return the cumulative number of documents seen by this cursor. + */ + int getTotalSent() const; - void accessed(); - /** @return idle time in ms */ - long long idleTime( long long now ); + /** + * Sends queries to the shards and gather the result for this batch. + * + * @param r The request object from the client + * @param ntoreturn Number of documents to return + * @param buffer The buffer to use to store the results. + * @param docCount This will contain the number of documents gathered for this batch after + * a successful call. + * + * @return true if this is not the final batch. + */ + bool sendNextBatch(int ntoreturn, BufBuilder& buffer, int& docCount); - std::string getNS() { return _cursor->getNS(); } + void accessed(); + /** @return idle time in ms */ + long long idleTime(long long now); - // The default initial buffer size for sending responses. - static const int INIT_REPLY_BUFFER_SIZE; + std::string getNS() { + return _cursor->getNS(); + } - protected: + // The default initial buffer size for sending responses. + static const int INIT_REPLY_BUFFER_SIZE; - ParallelSortClusteredCursor * _cursor; +protected: + ParallelSortClusteredCursor* _cursor; - int _skip; - int _ntoreturn; + int _skip; + int _ntoreturn; - int _totalSent; - bool _done; + int _totalSent; + bool _done; - long long _id; - long long _lastAccessMillis; // 0 means no timeout + long long _id; + long long _lastAccessMillis; // 0 means no timeout +}; - }; +typedef std::shared_ptr<ShardedClientCursor> ShardedClientCursorPtr; - typedef std::shared_ptr<ShardedClientCursor> ShardedClientCursorPtr; +class CursorCache { +public: + static long long TIMEOUT; - class CursorCache { - public: + typedef std::map<long long, ShardedClientCursorPtr> MapSharded; + typedef std::map<long long, int> MapShardedInt; + typedef std::map<long long, std::string> MapNormal; - static long long TIMEOUT; + CursorCache(); + ~CursorCache(); - typedef std::map<long long,ShardedClientCursorPtr> MapSharded; - typedef std::map<long long,int> MapShardedInt; - typedef std::map<long long,std::string> MapNormal; + ShardedClientCursorPtr get(long long id) const; + int getMaxTimeMS(long long id) const; + void store(ShardedClientCursorPtr cursor, int maxTimeMS); + void updateMaxTimeMS(long long id, int maxTimeMS); + void remove(long long id); - CursorCache(); - ~CursorCache(); + void storeRef(const std::string& server, long long id, const std::string& ns); + void removeRef(long long id); - ShardedClientCursorPtr get( long long id ) const; - int getMaxTimeMS( long long id ) const; - void store( ShardedClientCursorPtr cursor, int maxTimeMS ); - void updateMaxTimeMS( long long id, int maxTimeMS ); - void remove( long long id ); + /** @return the server for id or "" */ + std::string getRef(long long id) const; + /** @return the ns for id or "" */ + std::string getRefNS(long long id) const; - void storeRef(const std::string& server, long long id, const std::string& ns); - void removeRef( long long id ); + void gotKillCursors(Message& m); - /** @return the server for id or "" */ - std::string getRef( long long id ) const ; - /** @return the ns for id or "" */ - std::string getRefNS(long long id) const ; - - void gotKillCursors(Message& m ); + void appendInfo(BSONObjBuilder& result) const; - void appendInfo( BSONObjBuilder& result ) const ; + long long genId(); - long long genId(); + void doTimeouts(); + void startTimeoutThread(); - void doTimeouts(); - void startTimeoutThread(); - private: - mutable stdx::mutex _mutex; +private: + mutable stdx::mutex _mutex; - PseudoRandom _random; + PseudoRandom _random; - // Maps sharded cursor ID to ShardedClientCursorPtr. - MapSharded _cursors; + // Maps sharded cursor ID to ShardedClientCursorPtr. + MapSharded _cursors; - // Maps sharded cursor ID to remaining max time. Value can be any of: - // - the constant "kMaxTimeCursorNoTimeLimit", or - // - the constant "kMaxTimeCursorTimeLimitExpired", or - // - a positive integer representing milliseconds of remaining time - MapShardedInt _cursorsMaxTimeMS; + // Maps sharded cursor ID to remaining max time. Value can be any of: + // - the constant "kMaxTimeCursorNoTimeLimit", or + // - the constant "kMaxTimeCursorTimeLimitExpired", or + // - a positive integer representing milliseconds of remaining time + MapShardedInt _cursorsMaxTimeMS; - // Maps passthrough cursor ID to shard name. - MapNormal _refs; + // Maps passthrough cursor ID to shard name. + MapNormal _refs; - // Maps passthrough cursor ID to namespace. - MapNormal _refsNS; - - long long _shardedTotal; + // Maps passthrough cursor ID to namespace. + MapNormal _refsNS; - static const int _myLogLevel; - }; + long long _shardedTotal; - extern CursorCache cursorCache; + static const int _myLogLevel; +}; + +extern CursorCache cursorCache; } diff --git a/src/mongo/s/d_merge.cpp b/src/mongo/s/d_merge.cpp index 897b3dfd32f..d14c036b0e1 100644 --- a/src/mongo/s/d_merge.cpp +++ b/src/mongo/s/d_merge.cpp @@ -47,344 +47,322 @@ namespace mongo { - using std::endl; - using std::string; - using mongoutils::str::stream; - - static Status runApplyOpsCmd(const std::vector<ChunkType>&, - const ChunkVersion&, - const ChunkVersion&); - - static BSONObj buildMergeLogEntry(const std::vector<ChunkType>&, - const ChunkVersion&, - const ChunkVersion&); - - static bool isEmptyChunk( const ChunkType& ); - - bool mergeChunks( OperationContext* txn, - const NamespaceString& nss, - const BSONObj& minKey, - const BSONObj& maxKey, - const OID& epoch, - string* errMsg ) { - - // - // Get sharding state up-to-date - // - - ConnectionString configLoc = ConnectionString::parse( shardingState.getConfigServer(), - *errMsg ); - if ( !configLoc.isValid() ){ - warning() << *errMsg << endl; - return false; - } - - // - // Get the distributed lock - // - - string whyMessage = stream() << "merging chunks in " << nss.ns() - << " from " << minKey << " to " << maxKey; - auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock( - nss.ns(), whyMessage); - - if (!scopedDistLock.isOK()) { - *errMsg = stream() << "could not acquire collection lock for " << nss.ns() - << " to merge chunks in [" << minKey << "," << maxKey << ")" - << causedBy(scopedDistLock.getStatus()); - - warning() << *errMsg << endl; - return false; - } - - // - // We now have the collection lock, refresh metadata to latest version and sanity check - // +using std::endl; +using std::string; +using mongoutils::str::stream; + +static Status runApplyOpsCmd(const std::vector<ChunkType>&, + const ChunkVersion&, + const ChunkVersion&); + +static BSONObj buildMergeLogEntry(const std::vector<ChunkType>&, + const ChunkVersion&, + const ChunkVersion&); + +static bool isEmptyChunk(const ChunkType&); + +bool mergeChunks(OperationContext* txn, + const NamespaceString& nss, + const BSONObj& minKey, + const BSONObj& maxKey, + const OID& epoch, + string* errMsg) { + // + // Get sharding state up-to-date + // - ChunkVersion shardVersion; - Status status = shardingState.refreshMetadataNow(txn, nss.ns(), &shardVersion); + ConnectionString configLoc = ConnectionString::parse(shardingState.getConfigServer(), *errMsg); + if (!configLoc.isValid()) { + warning() << *errMsg << endl; + return false; + } - if ( !status.isOK() ) { + // + // Get the distributed lock + // - *errMsg = str::stream() << "could not merge chunks, failed to refresh metadata for " - << nss.ns() << causedBy( status.reason() ); + string whyMessage = stream() << "merging chunks in " << nss.ns() << " from " << minKey << " to " + << maxKey; + auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock(nss.ns(), whyMessage); - warning() << *errMsg << endl; - return false; - } + if (!scopedDistLock.isOK()) { + *errMsg = stream() << "could not acquire collection lock for " << nss.ns() + << " to merge chunks in [" << minKey << "," << maxKey << ")" + << causedBy(scopedDistLock.getStatus()); - if ( epoch.isSet() && shardVersion.epoch() != epoch ) { + warning() << *errMsg << endl; + return false; + } - *errMsg = stream() << "could not merge chunks, collection " << nss.ns() - << " has changed" << " since merge was sent" << "(sent epoch : " - << epoch.toString() - << ", current epoch : " << shardVersion.epoch().toString() << ")"; + // + // We now have the collection lock, refresh metadata to latest version and sanity check + // - warning() << *errMsg << endl; - return false; - } + ChunkVersion shardVersion; + Status status = shardingState.refreshMetadataNow(txn, nss.ns(), &shardVersion); - CollectionMetadataPtr metadata = shardingState.getCollectionMetadata( nss.ns() ); + if (!status.isOK()) { + *errMsg = str::stream() << "could not merge chunks, failed to refresh metadata for " + << nss.ns() << causedBy(status.reason()); - if ( !metadata || metadata->getKeyPattern().isEmpty() ) { + warning() << *errMsg << endl; + return false; + } - *errMsg = stream() << "could not merge chunks, collection " << nss.ns() - << " is not sharded"; + if (epoch.isSet() && shardVersion.epoch() != epoch) { + *errMsg = stream() << "could not merge chunks, collection " << nss.ns() << " has changed" + << " since merge was sent" + << "(sent epoch : " << epoch.toString() + << ", current epoch : " << shardVersion.epoch().toString() << ")"; - warning() << *errMsg << endl; - return false; - } + warning() << *errMsg << endl; + return false; + } - dassert( metadata->getShardVersion().equals( shardVersion ) ); + CollectionMetadataPtr metadata = shardingState.getCollectionMetadata(nss.ns()); - if ( !metadata->isValidKey( minKey ) || !metadata->isValidKey( maxKey ) ) { + if (!metadata || metadata->getKeyPattern().isEmpty()) { + *errMsg = stream() << "could not merge chunks, collection " << nss.ns() + << " is not sharded"; - *errMsg = stream() << "could not merge chunks, the range " - << rangeToString( minKey, maxKey ) << " is not valid" - << " for collection " << nss.ns() << " with key pattern " - << metadata->getKeyPattern(); + warning() << *errMsg << endl; + return false; + } - warning() << *errMsg << endl; - return false; - } + dassert(metadata->getShardVersion().equals(shardVersion)); - // - // Get merged chunk information - // + if (!metadata->isValidKey(minKey) || !metadata->isValidKey(maxKey)) { + *errMsg = stream() << "could not merge chunks, the range " << rangeToString(minKey, maxKey) + << " is not valid" + << " for collection " << nss.ns() << " with key pattern " + << metadata->getKeyPattern(); - ChunkVersion mergeVersion = metadata->getCollVersion(); - mergeVersion.incMinor(); + warning() << *errMsg << endl; + return false; + } - std::vector<ChunkType> chunksToMerge; + // + // Get merged chunk information + // - ChunkType itChunk; - itChunk.setMin( minKey ); - itChunk.setMax( minKey ); - itChunk.setNS( nss.ns() ); - itChunk.setShard( shardingState.getShardName() ); + ChunkVersion mergeVersion = metadata->getCollVersion(); + mergeVersion.incMinor(); - while (itChunk.getMax().woCompare(maxKey) < 0 && - metadata->getNextChunk(itChunk.getMax(), &itChunk)) { - chunksToMerge.push_back(itChunk); - } + std::vector<ChunkType> chunksToMerge; - if ( chunksToMerge.empty() ) { + ChunkType itChunk; + itChunk.setMin(minKey); + itChunk.setMax(minKey); + itChunk.setNS(nss.ns()); + itChunk.setShard(shardingState.getShardName()); - *errMsg = stream() << "could not merge chunks, collection " << nss.ns() - << " range starting at " << minKey - << " and ending at " << maxKey - << " does not belong to shard " << shardingState.getShardName(); + while (itChunk.getMax().woCompare(maxKey) < 0 && + metadata->getNextChunk(itChunk.getMax(), &itChunk)) { + chunksToMerge.push_back(itChunk); + } - warning() << *errMsg << endl; - return false; - } + if (chunksToMerge.empty()) { + *errMsg = stream() << "could not merge chunks, collection " << nss.ns() + << " range starting at " << minKey << " and ending at " << maxKey + << " does not belong to shard " << shardingState.getShardName(); - // - // Validate the range starts and ends at chunks and has no holes, error if not valid - // + warning() << *errMsg << endl; + return false; + } - BSONObj firstDocMin = chunksToMerge.front().getMin(); - BSONObj firstDocMax = chunksToMerge.front().getMax(); - // minKey is inclusive - bool minKeyInRange = rangeContains( firstDocMin, firstDocMax, minKey ); + // + // Validate the range starts and ends at chunks and has no holes, error if not valid + // - if ( !minKeyInRange ) { + BSONObj firstDocMin = chunksToMerge.front().getMin(); + BSONObj firstDocMax = chunksToMerge.front().getMax(); + // minKey is inclusive + bool minKeyInRange = rangeContains(firstDocMin, firstDocMax, minKey); - *errMsg = stream() << "could not merge chunks, collection " << nss.ns() - << " range starting at " << minKey - << " does not belong to shard " << shardingState.getShardName(); + if (!minKeyInRange) { + *errMsg = stream() << "could not merge chunks, collection " << nss.ns() + << " range starting at " << minKey << " does not belong to shard " + << shardingState.getShardName(); - warning() << *errMsg << endl; - return false; - } + warning() << *errMsg << endl; + return false; + } - BSONObj lastDocMin = chunksToMerge.back().getMin(); - BSONObj lastDocMax = chunksToMerge.back().getMax(); - // maxKey is exclusive - bool maxKeyInRange = lastDocMin.woCompare( maxKey ) < 0 && - lastDocMax.woCompare( maxKey ) >= 0; + BSONObj lastDocMin = chunksToMerge.back().getMin(); + BSONObj lastDocMax = chunksToMerge.back().getMax(); + // maxKey is exclusive + bool maxKeyInRange = lastDocMin.woCompare(maxKey) < 0 && lastDocMax.woCompare(maxKey) >= 0; - if ( !maxKeyInRange ) { - *errMsg = stream() << "could not merge chunks, collection " << nss.ns() - << " range ending at " << maxKey - << " does not belong to shard " << shardingState.getShardName(); + if (!maxKeyInRange) { + *errMsg = stream() << "could not merge chunks, collection " << nss.ns() + << " range ending at " << maxKey << " does not belong to shard " + << shardingState.getShardName(); - warning() << *errMsg << endl; - return false; - } + warning() << *errMsg << endl; + return false; + } - bool validRangeStartKey = firstDocMin.woCompare( minKey ) == 0; - bool validRangeEndKey = lastDocMax.woCompare( maxKey ) == 0; + bool validRangeStartKey = firstDocMin.woCompare(minKey) == 0; + bool validRangeEndKey = lastDocMax.woCompare(maxKey) == 0; - if ( !validRangeStartKey || !validRangeEndKey ) { + if (!validRangeStartKey || !validRangeEndKey) { + *errMsg = stream() << "could not merge chunks, collection " << nss.ns() + << " does not contain a chunk " + << (!validRangeStartKey ? "starting at " + minKey.toString() : "") + << (!validRangeStartKey && !validRangeEndKey ? " or " : "") + << (!validRangeEndKey ? "ending at " + maxKey.toString() : ""); - *errMsg = stream() << "could not merge chunks, collection " << nss.ns() - << " does not contain a chunk " - << ( !validRangeStartKey ? "starting at " + minKey.toString() : "" ) - << ( !validRangeStartKey && !validRangeEndKey ? " or " : "" ) - << ( !validRangeEndKey ? "ending at " + maxKey.toString() : "" ); + warning() << *errMsg << endl; + return false; + } - warning() << *errMsg << endl; - return false; - } + if (chunksToMerge.size() == 1) { + *errMsg = stream() << "could not merge chunks, collection " << nss.ns() + << " already contains chunk for " << rangeToString(minKey, maxKey); - if ( chunksToMerge.size() == 1 ) { + warning() << *errMsg << endl; + return false; + } - *errMsg = stream() << "could not merge chunks, collection " << nss.ns() - << " already contains chunk for " << rangeToString( minKey, maxKey ); + // Look for hole in range + for (size_t i = 1; i < chunksToMerge.size(); ++i) { + if (chunksToMerge[i - 1].getMax().woCompare(chunksToMerge[i].getMin()) != 0) { + *errMsg = + stream() << "could not merge chunks, collection " << nss.ns() + << " has a hole in the range " << rangeToString(minKey, maxKey) << " at " + << rangeToString(chunksToMerge[i - 1].getMax(), chunksToMerge[i].getMin()); warning() << *errMsg << endl; return false; } + } - // Look for hole in range - for (size_t i = 1; i < chunksToMerge.size(); ++i) { - if (chunksToMerge[i-1].getMax().woCompare(chunksToMerge[i].getMin()) != 0) { - *errMsg = stream() << "could not merge chunks, collection " << nss.ns() - << " has a hole in the range " << rangeToString(minKey, maxKey) - << " at " << rangeToString(chunksToMerge[i-1].getMax(), - chunksToMerge[i].getMin()); - - warning() << *errMsg << endl; - return false; - } - } - - // - // Run apply ops command - // - Status applyOpsStatus = runApplyOpsCmd(chunksToMerge, shardVersion, mergeVersion); - if (!applyOpsStatus.isOK()) { - warning() << applyOpsStatus; - return false; - } - - // - // Install merged chunk metadata - // - - { - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock writeLk(txn->lockState(), nss.db(), MODE_IX); - Lock::CollectionLock collLock(txn->lockState(), nss.ns(), MODE_X); - shardingState.mergeChunks(txn, nss.ns(), minKey, maxKey, mergeVersion); - } - - // - // Log change - // - - BSONObj mergeLogEntry = buildMergeLogEntry( chunksToMerge, - shardVersion, - mergeVersion ); + // + // Run apply ops command + // + Status applyOpsStatus = runApplyOpsCmd(chunksToMerge, shardVersion, mergeVersion); + if (!applyOpsStatus.isOK()) { + warning() << applyOpsStatus; + return false; + } - grid.catalogManager()->logChange(txn, "merge", nss.ns(), mergeLogEntry); + // + // Install merged chunk metadata + // - return true; + { + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock writeLk(txn->lockState(), nss.db(), MODE_IX); + Lock::CollectionLock collLock(txn->lockState(), nss.ns(), MODE_X); + shardingState.mergeChunks(txn, nss.ns(), minKey, maxKey, mergeVersion); } // - // Utilities for building BSONObjs for applyOps and change logging + // Log change // - BSONObj buildMergeLogEntry(const std::vector<ChunkType>& chunksToMerge, - const ChunkVersion& currShardVersion, - const ChunkVersion& newMergedVersion) { + BSONObj mergeLogEntry = buildMergeLogEntry(chunksToMerge, shardVersion, mergeVersion); - BSONObjBuilder logDetailB; + grid.catalogManager()->logChange(txn, "merge", nss.ns(), mergeLogEntry); - BSONArrayBuilder mergedB( logDetailB.subarrayStart( "merged" ) ); + return true; +} - for (const ChunkType& chunkToMerge : chunksToMerge) { - mergedB.append(chunkToMerge.toBSON()); - } +// +// Utilities for building BSONObjs for applyOps and change logging +// - mergedB.done(); +BSONObj buildMergeLogEntry(const std::vector<ChunkType>& chunksToMerge, + const ChunkVersion& currShardVersion, + const ChunkVersion& newMergedVersion) { + BSONObjBuilder logDetailB; - currShardVersion.addToBSON( logDetailB, "prevShardVersion" ); - newMergedVersion.addToBSON( logDetailB, "mergedVersion" ); + BSONArrayBuilder mergedB(logDetailB.subarrayStart("merged")); - return logDetailB.obj(); + for (const ChunkType& chunkToMerge : chunksToMerge) { + mergedB.append(chunkToMerge.toBSON()); } + mergedB.done(); - BSONObj buildOpMergeChunk( const ChunkType& mergedChunk ) { + currShardVersion.addToBSON(logDetailB, "prevShardVersion"); + newMergedVersion.addToBSON(logDetailB, "mergedVersion"); - BSONObjBuilder opB; + return logDetailB.obj(); +} - // Op basics - opB.append( "op" , "u" ); - opB.appendBool( "b" , false ); // no upserting - opB.append( "ns" , ChunkType::ConfigNS ); - // New object - opB.append( "o", mergedChunk.toBSON() ); +BSONObj buildOpMergeChunk(const ChunkType& mergedChunk) { + BSONObjBuilder opB; - // Query object - opB.append( "o2", - BSON( ChunkType::name( mergedChunk.getName() ) ) ); + // Op basics + opB.append("op", "u"); + opB.appendBool("b", false); // no upserting + opB.append("ns", ChunkType::ConfigNS); - return opB.obj(); - } + // New object + opB.append("o", mergedChunk.toBSON()); - BSONObj buildOpRemoveChunk( const ChunkType& chunkToRemove ) { + // Query object + opB.append("o2", BSON(ChunkType::name(mergedChunk.getName()))); - BSONObjBuilder opB; - - // Op basics - opB.append( "op", "d" ); // delete - opB.append( "ns", ChunkType::ConfigNS ); + return opB.obj(); +} - opB.append( "o", BSON( ChunkType::name( chunkToRemove.getName() ) ) ); +BSONObj buildOpRemoveChunk(const ChunkType& chunkToRemove) { + BSONObjBuilder opB; - return opB.obj(); - } + // Op basics + opB.append("op", "d"); // delete + opB.append("ns", ChunkType::ConfigNS); - BSONArray buildOpPrecond(const string& ns, - const string& shardName, - const ChunkVersion& shardVersion) { - BSONArrayBuilder preCond; - BSONObjBuilder condB; - condB.append( "ns", ChunkType::ConfigNS ); - condB.append( "q", BSON( "query" << BSON( ChunkType::ns( ns ) ) - << "orderby" << BSON( ChunkType::DEPRECATED_lastmod() << -1 ) ) ); - { - BSONObjBuilder resB( condB.subobjStart( "res" ) ); - shardVersion.addToBSON( resB, ChunkType::DEPRECATED_lastmod() ); - resB.done(); - } - preCond.append(condB.obj()); - return preCond.arr(); - } + opB.append("o", BSON(ChunkType::name(chunkToRemove.getName()))); - Status runApplyOpsCmd(const std::vector<ChunkType>& chunksToMerge, - const ChunkVersion& currShardVersion, - const ChunkVersion& newMergedVersion) { + return opB.obj(); +} - BSONArrayBuilder updatesB; +BSONArray buildOpPrecond(const string& ns, + const string& shardName, + const ChunkVersion& shardVersion) { + BSONArrayBuilder preCond; + BSONObjBuilder condB; + condB.append("ns", ChunkType::ConfigNS); + condB.append("q", + BSON("query" << BSON(ChunkType::ns(ns)) << "orderby" + << BSON(ChunkType::DEPRECATED_lastmod() << -1))); + { + BSONObjBuilder resB(condB.subobjStart("res")); + shardVersion.addToBSON(resB, ChunkType::DEPRECATED_lastmod()); + resB.done(); + } + preCond.append(condB.obj()); + return preCond.arr(); +} - // The chunk we'll be "expanding" is the first chunk - const ChunkType& firstChunk = chunksToMerge.front(); +Status runApplyOpsCmd(const std::vector<ChunkType>& chunksToMerge, + const ChunkVersion& currShardVersion, + const ChunkVersion& newMergedVersion) { + BSONArrayBuilder updatesB; - // Fill in details not tracked by metadata - ChunkType mergedChunk(firstChunk); - mergedChunk.setName(Chunk::genID(firstChunk.getNS(), firstChunk.getMin())); - mergedChunk.setMax(chunksToMerge.back().getMax()); - mergedChunk.setVersion(newMergedVersion); + // The chunk we'll be "expanding" is the first chunk + const ChunkType& firstChunk = chunksToMerge.front(); - updatesB.append(buildOpMergeChunk(mergedChunk)); + // Fill in details not tracked by metadata + ChunkType mergedChunk(firstChunk); + mergedChunk.setName(Chunk::genID(firstChunk.getNS(), firstChunk.getMin())); + mergedChunk.setMax(chunksToMerge.back().getMax()); + mergedChunk.setVersion(newMergedVersion); - // Don't remove chunk we're expanding - for (size_t i = 1; i < chunksToMerge.size(); ++i) { - ChunkType chunkToMerge(chunksToMerge[i]); - chunkToMerge.setName(Chunk::genID(chunkToMerge.getNS(), chunkToMerge.getMin())); - updatesB.append(buildOpRemoveChunk(chunkToMerge)); - } + updatesB.append(buildOpMergeChunk(mergedChunk)); - BSONArray preCond = buildOpPrecond(firstChunk.getNS(), - firstChunk.getShard(), - currShardVersion); - return grid.catalogManager()->applyChunkOpsDeprecated(updatesB.arr(), preCond); + // Don't remove chunk we're expanding + for (size_t i = 1; i < chunksToMerge.size(); ++i) { + ChunkType chunkToMerge(chunksToMerge[i]); + chunkToMerge.setName(Chunk::genID(chunkToMerge.getNS(), chunkToMerge.getMin())); + updatesB.append(buildOpRemoveChunk(chunkToMerge)); } + BSONArray preCond = buildOpPrecond(firstChunk.getNS(), firstChunk.getShard(), currShardVersion); + return grid.catalogManager()->applyChunkOpsDeprecated(updatesB.arr(), preCond); +} } diff --git a/src/mongo/s/d_merge.h b/src/mongo/s/d_merge.h index 85d39fa7f11..79fa37a2dbb 100644 --- a/src/mongo/s/d_merge.h +++ b/src/mongo/s/d_merge.h @@ -33,30 +33,30 @@ namespace mongo { - class OperationContext; +class OperationContext; - /** - * Merges a chunks in the specified [minKey, maxKey) range of the specified namespace. - * Updates the local and remote metadata by expanding the bounds of the first chunk in the - * range, dropping the others, and incrementing the minor version of the shard to the next - * higher version. Returns true on success. - * - * Fails with errMsg if the 'epoch' was set and has changed (indicating the range is no longer - * valid), or if the range does not exactly start and stop on chunks owned by this shard, or - * if the chunks in this range are not contiguous. - * - * WARNING: On network failure, it is possible that the chunks in our local metadata may not - * match the remote metadata, however the key ranges protected will be the same. All metadata - * operations are responsible for updating the metadata before performing any actions. - * - * Locking note: - * + Takes a distributed lock over the namespace - * + Cannot be called with any other locks held - */ - bool mergeChunks( OperationContext* txn, - const NamespaceString& nss, - const BSONObj& minKey, - const BSONObj& maxKey, - const OID& epoch, - std::string* errMsg ); +/** + * Merges a chunks in the specified [minKey, maxKey) range of the specified namespace. + * Updates the local and remote metadata by expanding the bounds of the first chunk in the + * range, dropping the others, and incrementing the minor version of the shard to the next + * higher version. Returns true on success. + * + * Fails with errMsg if the 'epoch' was set and has changed (indicating the range is no longer + * valid), or if the range does not exactly start and stop on chunks owned by this shard, or + * if the chunks in this range are not contiguous. + * + * WARNING: On network failure, it is possible that the chunks in our local metadata may not + * match the remote metadata, however the key ranges protected will be the same. All metadata + * operations are responsible for updating the metadata before performing any actions. + * + * Locking note: + * + Takes a distributed lock over the namespace + * + Cannot be called with any other locks held + */ +bool mergeChunks(OperationContext* txn, + const NamespaceString& nss, + const BSONObj& minKey, + const BSONObj& maxKey, + const OID& epoch, + std::string* errMsg); } diff --git a/src/mongo/s/d_migrate.cpp b/src/mongo/s/d_migrate.cpp index 8ed12b4d082..a0893be2524 100644 --- a/src/mongo/s/d_migrate.cpp +++ b/src/mongo/s/d_migrate.cpp @@ -91,653 +91,645 @@ #include "mongo/util/startup_test.h" // Pause while a fail point is enabled. -#define MONGO_FP_PAUSE_WHILE(symbol) while (MONGO_FAIL_POINT(symbol)) { sleepmillis(100); } +#define MONGO_FP_PAUSE_WHILE(symbol) \ + while (MONGO_FAIL_POINT(symbol)) { \ + sleepmillis(100); \ + } namespace mongo { - using namespace std::chrono; - using std::list; - using std::set; - using std::string; - using std::vector; +using namespace std::chrono; +using std::list; +using std::set; +using std::string; +using std::vector; namespace { - const int kDefaultWTimeoutMs = 60 * 1000; - const WriteConcernOptions DefaultWriteConcern(2, WriteConcernOptions::NONE, kDefaultWTimeoutMs); - - /** - * Returns the default write concern for migration cleanup (at donor shard) and - * cloning documents (at recipient shard). - */ - WriteConcernOptions getDefaultWriteConcern() { - repl::ReplicationCoordinator* replCoordinator = repl::getGlobalReplicationCoordinator(); - if (replCoordinator->getReplicationMode() == - mongo::repl::ReplicationCoordinator::modeReplSet) { +const int kDefaultWTimeoutMs = 60 * 1000; +const WriteConcernOptions DefaultWriteConcern(2, WriteConcernOptions::NONE, kDefaultWTimeoutMs); - Status status = - replCoordinator->checkIfWriteConcernCanBeSatisfied(DefaultWriteConcern); - if (status.isOK()) { - return DefaultWriteConcern; - } +/** + * Returns the default write concern for migration cleanup (at donor shard) and + * cloning documents (at recipient shard). + */ +WriteConcernOptions getDefaultWriteConcern() { + repl::ReplicationCoordinator* replCoordinator = repl::getGlobalReplicationCoordinator(); + if (replCoordinator->getReplicationMode() == mongo::repl::ReplicationCoordinator::modeReplSet) { + Status status = replCoordinator->checkIfWriteConcernCanBeSatisfied(DefaultWriteConcern); + if (status.isOK()) { + return DefaultWriteConcern; } - - return WriteConcernOptions(1, WriteConcernOptions::NONE, 0); } -} // namespace - - MONGO_FP_DECLARE(failMigrationCommit); - MONGO_FP_DECLARE(failMigrationConfigWritePrepare); - MONGO_FP_DECLARE(failMigrationApplyOps); - - Tee* migrateLog = RamLog::get("migrate"); + return WriteConcernOptions(1, WriteConcernOptions::NONE, 0); +} - class MoveTimingHelper { - public: - MoveTimingHelper(OperationContext* txn, - const string& where, - const string& ns, - BSONObj min, - BSONObj max , - int total, - string* cmdErrmsg, - string toShard, - string fromShard) - : _txn(txn), - _where(where), - _ns(ns), - _to(toShard), - _from(fromShard), - _next(0), - _total(total), - _cmdErrmsg(cmdErrmsg) { - - _b.append( "min" , min ); - _b.append( "max" , max ); - } - - ~MoveTimingHelper() { - // even if logChange doesn't throw, bson does - // sigh - try { - if ( !_to.empty() ){ - _b.append( "to", _to ); - } - if ( !_from.empty() ){ - _b.append( "from", _from ); - } - if ( _next != _total ) { - _b.append( "note" , "aborted" ); - } - else { - _b.append( "note" , "success" ); - } - if ( !_cmdErrmsg->empty() ) { - _b.append( "errmsg" , *_cmdErrmsg ); - } +} // namespace + +MONGO_FP_DECLARE(failMigrationCommit); +MONGO_FP_DECLARE(failMigrationConfigWritePrepare); +MONGO_FP_DECLARE(failMigrationApplyOps); + +Tee* migrateLog = RamLog::get("migrate"); + +class MoveTimingHelper { +public: + MoveTimingHelper(OperationContext* txn, + const string& where, + const string& ns, + BSONObj min, + BSONObj max, + int total, + string* cmdErrmsg, + string toShard, + string fromShard) + : _txn(txn), + _where(where), + _ns(ns), + _to(toShard), + _from(fromShard), + _next(0), + _total(total), + _cmdErrmsg(cmdErrmsg) { + _b.append("min", min); + _b.append("max", max); + } - grid.catalogManager()->logChange(_txn, - (string)"moveChunk." + _where, - _ns, - _b.obj()); + ~MoveTimingHelper() { + // even if logChange doesn't throw, bson does + // sigh + try { + if (!_to.empty()) { + _b.append("to", _to); } - catch ( const std::exception& e ) { - warning() << "couldn't record timing for moveChunk '" << _where << "': " << e.what() << migrateLog; + if (!_from.empty()) { + _b.append("from", _from); + } + if (_next != _total) { + _b.append("note", "aborted"); + } else { + _b.append("note", "success"); + } + if (!_cmdErrmsg->empty()) { + _b.append("errmsg", *_cmdErrmsg); } - } - void done(int step) { - invariant(step == ++_next); - invariant(step <= _total); + grid.catalogManager()->logChange(_txn, (string) "moveChunk." + _where, _ns, _b.obj()); + } catch (const std::exception& e) { + warning() << "couldn't record timing for moveChunk '" << _where << "': " << e.what() + << migrateLog; + } + } - const string s = str::stream() << "step " << step << " of " << _total; + void done(int step) { + invariant(step == ++_next); + invariant(step <= _total); - CurOp * op = CurOp::get(_txn); - { - std::lock_guard<Client> lk(*_txn->getClient()); - op->setMessage_inlock(s.c_str()); - } + const string s = str::stream() << "step " << step << " of " << _total; - _b.appendNumber(s, _t.millis()); - _t.reset(); + CurOp* op = CurOp::get(_txn); + { + std::lock_guard<Client> lk(*_txn->getClient()); + op->setMessage_inlock(s.c_str()); } - private: - OperationContext* const _txn; - Timer _t; + _b.appendNumber(s, _t.millis()); + _t.reset(); + } - string _where; - string _ns; - string _to; - string _from; +private: + OperationContext* const _txn; + Timer _t; - int _next; - int _total; // expected # of steps + string _where; + string _ns; + string _to; + string _from; - const string* _cmdErrmsg; + int _next; + int _total; // expected # of steps - BSONObjBuilder _b; - }; + const string* _cmdErrmsg; - class ChunkCommandHelper : public Command { - public: - ChunkCommandHelper( const char * name ) - : Command( name ) { - } + BSONObjBuilder _b; +}; - virtual void help( std::stringstream& help ) const { - help << "internal - should not be called directly"; - } - virtual bool slaveOk() const { return false; } - virtual bool adminOnly() const { return true; } - virtual bool isWriteCommandForConfigServer() const { return false; } +class ChunkCommandHelper : public Command { +public: + ChunkCommandHelper(const char* name) : Command(name) {} - }; - - bool isInRange( const BSONObj& obj , - const BSONObj& min , - const BSONObj& max , - const BSONObj& shardKeyPattern ) { - ShardKeyPattern shardKey( shardKeyPattern ); - BSONObj k = shardKey.extractShardKeyFromDoc( obj ); - return k.woCompare( min ) >= 0 && k.woCompare( max ) < 0; + virtual void help(std::stringstream& help) const { + help << "internal - should not be called directly"; } + virtual bool slaveOk() const { + return false; + } + virtual bool adminOnly() const { + return true; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } +}; + +bool isInRange(const BSONObj& obj, + const BSONObj& min, + const BSONObj& max, + const BSONObj& shardKeyPattern) { + ShardKeyPattern shardKey(shardKeyPattern); + BSONObj k = shardKey.extractShardKeyFromDoc(obj); + return k.woCompare(min) >= 0 && k.woCompare(max) < 0; +} - class MigrateFromStatus { - public: - MigrateFromStatus(): - _inCriticalSection(false), - _memoryUsed(0), - _active(false) { - } - - /** - * @return false if cannot start. One of the reason for not being able to - * start is there is already an existing migration in progress. - */ - bool start(OperationContext* txn, - const std::string& ns, - const BSONObj& min, - const BSONObj& max, - const BSONObj& shardKeyPattern) { - verify(!min.isEmpty()); - verify(!max.isEmpty()); - verify(!ns.empty()); +class MigrateFromStatus { +public: + MigrateFromStatus() : _inCriticalSection(false), _memoryUsed(0), _active(false) {} - // Get global shared to synchronize with logOp. Also see comments in the class - // members declaration for more details. - Lock::GlobalRead globalShared(txn->lockState()); - std::lock_guard<std::mutex> lk(_mutex); + /** + * @return false if cannot start. One of the reason for not being able to + * start is there is already an existing migration in progress. + */ + bool start(OperationContext* txn, + const std::string& ns, + const BSONObj& min, + const BSONObj& max, + const BSONObj& shardKeyPattern) { + verify(!min.isEmpty()); + verify(!max.isEmpty()); + verify(!ns.empty()); + + // Get global shared to synchronize with logOp. Also see comments in the class + // members declaration for more details. + Lock::GlobalRead globalShared(txn->lockState()); + std::lock_guard<std::mutex> lk(_mutex); + + if (_active) { + return false; + } - if (_active) { - return false; - } + _ns = ns; + _min = min; + _max = max; + _shardKeyPattern = shardKeyPattern; - _ns = ns; - _min = min; - _max = max; - _shardKeyPattern = shardKeyPattern; + verify(_deleted.size() == 0); + verify(_reload.size() == 0); + verify(_memoryUsed == 0); - verify(_deleted.size() == 0); - verify(_reload.size() == 0); - verify(_memoryUsed == 0); + _active = true; - _active = true; + std::lock_guard<std::mutex> tLock(_cloneLocsMutex); + verify(_cloneLocs.size() == 0); - std::lock_guard<std::mutex> tLock(_cloneLocsMutex); - verify(_cloneLocs.size() == 0); + return true; + } - return true; - } + void done(OperationContext* txn) { + log() << "MigrateFromStatus::done About to acquire global lock to exit critical " + "section"; - void done(OperationContext* txn) { - log() << "MigrateFromStatus::done About to acquire global lock to exit critical " - "section"; + // Get global shared to synchronize with logOp. Also see comments in the class + // members declaration for more details. + Lock::GlobalRead globalShared(txn->lockState()); + std::lock_guard<std::mutex> lk(_mutex); - // Get global shared to synchronize with logOp. Also see comments in the class - // members declaration for more details. - Lock::GlobalRead globalShared(txn->lockState()); - std::lock_guard<std::mutex> lk(_mutex); + _active = false; + _deleteNotifyExec.reset(NULL); + _inCriticalSection = false; + _inCriticalSectionCV.notify_all(); - _active = false; - _deleteNotifyExec.reset( NULL ); - _inCriticalSection = false; - _inCriticalSectionCV.notify_all(); + _deleted.clear(); + _reload.clear(); + _memoryUsed = 0; - _deleted.clear(); - _reload.clear(); - _memoryUsed = 0; + std::lock_guard<std::mutex> cloneLock(_cloneLocsMutex); + _cloneLocs.clear(); + } - std::lock_guard<std::mutex> cloneLock(_cloneLocsMutex); - _cloneLocs.clear(); + void logOp(OperationContext* txn, + const char* opstr, + const char* ns, + const BSONObj& obj, + BSONObj* patt, + bool notInActiveChunk) { + ensureShardVersionOKOrThrow(txn->getClient(), ns); + + const char op = opstr[0]; + + if (notInActiveChunk) { + // Ignore writes that came from the migration process like cleanup so they + // won't be transferred to the recipient shard. Also ignore ops from + // _migrateClone and _transferMods since it is impossible to move a chunk + // to self. + return; } - void logOp(OperationContext* txn, - const char* opstr, - const char* ns, - const BSONObj& obj, - BSONObj* patt, - bool notInActiveChunk) { - ensureShardVersionOKOrThrow(txn->getClient(), ns); + dassert(txn->lockState()->isWriteLocked()); // Must have Global IX. - const char op = opstr[0]; + if (!_active) + return; - if (notInActiveChunk) { - // Ignore writes that came from the migration process like cleanup so they - // won't be transferred to the recipient shard. Also ignore ops from - // _migrateClone and _transferMods since it is impossible to move a chunk - // to self. - return; - } + if (_ns != ns) + return; - dassert(txn->lockState()->isWriteLocked()); // Must have Global IX. + // no need to log if this is not an insertion, an update, or an actual deletion + // note: opstr 'db' isn't a deletion but a mention that a database exists + // (for replication machinery mostly). + if (op == 'n' || op == 'c' || (op == 'd' && opstr[1] == 'b')) + return; - if (!_active) - return; + BSONElement ide; + if (patt) + ide = patt->getField("_id"); + else + ide = obj["_id"]; - if (_ns != ns) - return; + if (ide.eoo()) { + warning() << "logOpForSharding got mod with no _id, ignoring obj: " << obj + << migrateLog; + return; + } - // no need to log if this is not an insertion, an update, or an actual deletion - // note: opstr 'db' isn't a deletion but a mention that a database exists - // (for replication machinery mostly). - if (op == 'n' || op == 'c' || (op == 'd' && opstr[1] == 'b')) - return; + if (op == 'i' && (!isInRange(obj, _min, _max, _shardKeyPattern))) { + return; + } - BSONElement ide; - if (patt) - ide = patt->getField("_id"); - else - ide = obj["_id"]; + BSONObj idObj(ide.wrap()); - if (ide.eoo()) { - warning() << "logOpForSharding got mod with no _id, ignoring obj: " - << obj << migrateLog; + if (op == 'u') { + BSONObj fullDoc; + OldClientContext ctx(txn, _ns, false); + if (!Helpers::findById(txn, ctx.db(), _ns.c_str(), idObj, fullDoc)) { + warning() << "logOpForSharding couldn't find: " << idObj + << " even though should have" << migrateLog; + dassert(false); // TODO: Abort the migration. return; } - if (op == 'i' && (!isInRange(obj, _min, _max, _shardKeyPattern))) { + if (!isInRange(fullDoc, _min, _max, _shardKeyPattern)) { return; } + } - BSONObj idObj(ide.wrap()); + // Note: can't check if delete is in active chunk since the document is gone! - if (op == 'u') { - BSONObj fullDoc; - OldClientContext ctx(txn, _ns, false); - if (!Helpers::findById(txn, ctx.db(), _ns.c_str(), idObj, fullDoc)) { - warning() << "logOpForSharding couldn't find: " << idObj - << " even though should have" << migrateLog; - dassert(false); // TODO: Abort the migration. - return; - } + txn->recoveryUnit()->registerChange(new LogOpForShardingHandler(this, idObj, op)); + } - if (!isInRange(fullDoc, _min, _max, _shardKeyPattern)) { - return; + /** + * Insert items from docIdList to a new array with the given fieldName in the given + * builder. If explode is true, the inserted object will be the full version of the + * document. Note that the whenever an item from the docList is inserted to the array, + * it will also be removed from docList. + * + * Should be holding the collection lock for ns if explode is true. + */ + void xfer(OperationContext* txn, + const string& ns, + Database* db, + list<BSONObj>* docIdList, + BSONObjBuilder& builder, + const char* fieldName, + long long& size, + bool explode) { + const long long maxSize = 1024 * 1024; + + if (docIdList->size() == 0 || size > maxSize) + return; + + BSONArrayBuilder arr(builder.subarrayStart(fieldName)); + + list<BSONObj>::iterator docIdIter = docIdList->begin(); + while (docIdIter != docIdList->end() && size < maxSize) { + BSONObj idDoc = *docIdIter; + if (explode) { + BSONObj fullDoc; + if (Helpers::findById(txn, db, ns.c_str(), idDoc, fullDoc)) { + arr.append(fullDoc); + size += fullDoc.objsize(); } + } else { + arr.append(idDoc); + size += idDoc.objsize(); } - // Note: can't check if delete is in active chunk since the document is gone! - - txn->recoveryUnit()->registerChange(new LogOpForShardingHandler(this, idObj, op)); + docIdIter = docIdList->erase(docIdIter); } - /** - * Insert items from docIdList to a new array with the given fieldName in the given - * builder. If explode is true, the inserted object will be the full version of the - * document. Note that the whenever an item from the docList is inserted to the array, - * it will also be removed from docList. - * - * Should be holding the collection lock for ns if explode is true. - */ - void xfer(OperationContext* txn, - const string& ns, - Database* db, - list<BSONObj> *docIdList, - BSONObjBuilder& builder, - const char* fieldName, - long long& size, - bool explode) { - const long long maxSize = 1024 * 1024; - - if (docIdList->size() == 0 || size > maxSize) - return; + arr.done(); + } - BSONArrayBuilder arr(builder.subarrayStart(fieldName)); + /** + * called from the dest of a migrate + * transfers mods from src to dest + */ + bool transferMods(OperationContext* txn, string& errmsg, BSONObjBuilder& b) { + long long size = 0; - list<BSONObj>::iterator docIdIter = docIdList->begin(); - while (docIdIter != docIdList->end() && size < maxSize) { - BSONObj idDoc = *docIdIter; - if (explode) { - BSONObj fullDoc; - if (Helpers::findById(txn, db, ns.c_str(), idDoc, fullDoc)) { - arr.append( fullDoc ); - size += fullDoc.objsize(); - } - } - else { - arr.append(idDoc); - size += idDoc.objsize(); - } + { + AutoGetCollectionForRead ctx(txn, getNS()); - docIdIter = docIdList->erase(docIdIter); + std::lock_guard<std::mutex> sl(_mutex); + if (!_active) { + errmsg = "no active migration!"; + return false; } - arr.done(); + // TODO: fix SERVER-16540 race + xfer(txn, _ns, ctx.getDb(), &_deleted, b, "deleted", size, false); + xfer(txn, _ns, ctx.getDb(), &_reload, b, "reload", size, true); } - /** - * called from the dest of a migrate - * transfers mods from src to dest - */ - bool transferMods(OperationContext* txn, string& errmsg, BSONObjBuilder& b) { - long long size = 0; - - { - AutoGetCollectionForRead ctx(txn, getNS()); - - std::lock_guard<std::mutex> sl(_mutex); - if (!_active) { - errmsg = "no active migration!"; - return false; - } + b.append("size", size); - // TODO: fix SERVER-16540 race - xfer(txn, _ns, ctx.getDb(), &_deleted, b, "deleted", size, false); - xfer(txn, _ns, ctx.getDb(), &_reload, b, "reload", size, true); - } - - b.append( "size" , size ); + return true; + } - return true; + /** + * Get the disklocs that belong to the chunk migrated and sort them in _cloneLocs + * (to avoid seeking disk later). + * + * @param maxChunkSize number of bytes beyond which a chunk's base data (no indices) + * is considered too large to move. + * @param errmsg filled with textual description of error if this call return false. + * @return false if approximate chunk size is too big to move or true otherwise. + */ + bool storeCurrentLocs(OperationContext* txn, + long long maxChunkSize, + string& errmsg, + BSONObjBuilder& result) { + AutoGetCollectionForRead ctx(txn, getNS()); + Collection* collection = ctx.getCollection(); + if (!collection) { + errmsg = "ns not found, should be impossible"; + return false; } - /** - * Get the disklocs that belong to the chunk migrated and sort them in _cloneLocs - * (to avoid seeking disk later). - * - * @param maxChunkSize number of bytes beyond which a chunk's base data (no indices) - * is considered too large to move. - * @param errmsg filled with textual description of error if this call return false. - * @return false if approximate chunk size is too big to move or true otherwise. - */ - bool storeCurrentLocs(OperationContext* txn, - long long maxChunkSize, - string& errmsg, - BSONObjBuilder& result ) { - AutoGetCollectionForRead ctx(txn, getNS()); - Collection* collection = ctx.getCollection(); - if ( !collection ) { - errmsg = "ns not found, should be impossible"; - return false; - } - - // Allow multiKey based on the invariant that shard keys must be single-valued. - // Therefore, any multi-key index prefixed by shard key cannot be multikey over - // the shard key fields. - IndexDescriptor *idx = - collection->getIndexCatalog()->findShardKeyPrefixedIndex(txn, - _shardKeyPattern , - false); // requireSingleKey + // Allow multiKey based on the invariant that shard keys must be single-valued. + // Therefore, any multi-key index prefixed by shard key cannot be multikey over + // the shard key fields. + IndexDescriptor* idx = + collection->getIndexCatalog()->findShardKeyPrefixedIndex(txn, + _shardKeyPattern, + false); // requireSingleKey + + if (idx == NULL) { + errmsg = str::stream() << "can't find index with prefix " << _shardKeyPattern + << " in storeCurrentLocs for " << _ns; + return false; + } - if (idx == NULL) { - errmsg = str::stream() << "can't find index with prefix " << _shardKeyPattern - << " in storeCurrentLocs for " << _ns; - return false; - } + // Assume both min and max non-empty, append MinKey's to make them fit chosen index + BSONObj min; + BSONObj max; + KeyPattern kp(idx->keyPattern()); - // Assume both min and max non-empty, append MinKey's to make them fit chosen index - BSONObj min; - BSONObj max; - KeyPattern kp(idx->keyPattern()); + { + // It's alright not to lock _mutex all the way through based on the assumption + // that this is only called by the main thread that drives the migration and + // only it can start and stop the current migration. + std::lock_guard<std::mutex> sl(_mutex); - { - // It's alright not to lock _mutex all the way through based on the assumption - // that this is only called by the main thread that drives the migration and - // only it can start and stop the current migration. - std::lock_guard<std::mutex> sl(_mutex); + invariant(_deleteNotifyExec.get() == NULL); + WorkingSet* ws = new WorkingSet(); + DeleteNotificationStage* dns = new DeleteNotificationStage(); + PlanExecutor* deleteNotifyExec; + // Takes ownership of 'ws' and 'dns'. + Status execStatus = PlanExecutor::make( + txn, ws, dns, collection, PlanExecutor::YIELD_MANUAL, &deleteNotifyExec); + invariant(execStatus.isOK()); + deleteNotifyExec->registerExec(); + _deleteNotifyExec.reset(deleteNotifyExec); + + min = Helpers::toKeyFormat(kp.extendRangeBound(_min, false)); + max = Helpers::toKeyFormat(kp.extendRangeBound(_max, false)); + } - invariant( _deleteNotifyExec.get() == NULL ); - WorkingSet* ws = new WorkingSet(); - DeleteNotificationStage* dns = new DeleteNotificationStage(); - PlanExecutor* deleteNotifyExec; - // Takes ownership of 'ws' and 'dns'. - Status execStatus = PlanExecutor::make(txn, - ws, - dns, - collection, - PlanExecutor::YIELD_MANUAL, - &deleteNotifyExec); - invariant(execStatus.isOK()); - deleteNotifyExec->registerExec(); - _deleteNotifyExec.reset(deleteNotifyExec); - - min = Helpers::toKeyFormat(kp.extendRangeBound(_min, false)); - max = Helpers::toKeyFormat(kp.extendRangeBound(_max, false)); - } - - std::unique_ptr<PlanExecutor> exec( - InternalPlanner::indexScan(txn, collection, idx, min, max, false)); - // We can afford to yield here because any change to the base data that we might - // miss is already being queued and will migrate in the 'transferMods' stage. - exec->setYieldPolicy(PlanExecutor::YIELD_AUTO); - - // use the average object size to estimate how many objects a full chunk would carry - // do that while traversing the chunk's range using the sharding index, below - // there's a fair amount of slack before we determine a chunk is too large because object sizes will vary - unsigned long long maxRecsWhenFull; - long long avgRecSize; - const long long totalRecs = collection->numRecords(txn); - if ( totalRecs > 0 ) { - avgRecSize = collection->dataSize(txn) / totalRecs; - maxRecsWhenFull = maxChunkSize / avgRecSize; - maxRecsWhenFull = std::min((unsigned long long)(Chunk::MaxObjectPerChunk + 1) , 130 * maxRecsWhenFull / 100 /* slack */ ); - } - else { - avgRecSize = 0; - maxRecsWhenFull = Chunk::MaxObjectPerChunk + 1; - } - - // do a full traversal of the chunk and don't stop even if we think it is a large chunk - // we want the number of records to better report, in that case - bool isLargeChunk = false; - unsigned long long recCount = 0;; - RecordId dl; - while (PlanExecutor::ADVANCED == exec->getNext(NULL, &dl)) { - if ( ! isLargeChunk ) { - std::lock_guard<std::mutex> lk(_cloneLocsMutex); - _cloneLocs.insert( dl ); - } + std::unique_ptr<PlanExecutor> exec( + InternalPlanner::indexScan(txn, collection, idx, min, max, false)); + // We can afford to yield here because any change to the base data that we might + // miss is already being queued and will migrate in the 'transferMods' stage. + exec->setYieldPolicy(PlanExecutor::YIELD_AUTO); + + // use the average object size to estimate how many objects a full chunk would carry + // do that while traversing the chunk's range using the sharding index, below + // there's a fair amount of slack before we determine a chunk is too large because object sizes will vary + unsigned long long maxRecsWhenFull; + long long avgRecSize; + const long long totalRecs = collection->numRecords(txn); + if (totalRecs > 0) { + avgRecSize = collection->dataSize(txn) / totalRecs; + maxRecsWhenFull = maxChunkSize / avgRecSize; + maxRecsWhenFull = std::min((unsigned long long)(Chunk::MaxObjectPerChunk + 1), + 130 * maxRecsWhenFull / 100 /* slack */); + } else { + avgRecSize = 0; + maxRecsWhenFull = Chunk::MaxObjectPerChunk + 1; + } - if ( ++recCount > maxRecsWhenFull ) { - isLargeChunk = true; - // continue on despite knowing that it will fail, - // just to get the correct value for recCount. - } + // do a full traversal of the chunk and don't stop even if we think it is a large chunk + // we want the number of records to better report, in that case + bool isLargeChunk = false; + unsigned long long recCount = 0; + ; + RecordId dl; + while (PlanExecutor::ADVANCED == exec->getNext(NULL, &dl)) { + if (!isLargeChunk) { + std::lock_guard<std::mutex> lk(_cloneLocsMutex); + _cloneLocs.insert(dl); } - exec.reset(); - if ( isLargeChunk ) { - std::lock_guard<std::mutex> sl(_mutex); - warning() << "cannot move chunk: the maximum number of documents for a chunk is " - << maxRecsWhenFull << " , the maximum chunk size is " << maxChunkSize - << " , average document size is " << avgRecSize - << ". Found " << recCount << " documents in chunk " - << " ns: " << _ns << " " - << _min << " -> " << _max << migrateLog; - - result.appendBool( "chunkTooBig" , true ); - result.appendNumber( "estimatedChunkSize" , (long long)(recCount * avgRecSize) ); - errmsg = "chunk too big to move"; - return false; + if (++recCount > maxRecsWhenFull) { + isLargeChunk = true; + // continue on despite knowing that it will fail, + // just to get the correct value for recCount. } + } + exec.reset(); - log() << "moveChunk number of documents: " << cloneLocsRemaining() << migrateLog; - - txn->recoveryUnit()->abandonSnapshot(); - return true; + if (isLargeChunk) { + std::lock_guard<std::mutex> sl(_mutex); + warning() << "cannot move chunk: the maximum number of documents for a chunk is " + << maxRecsWhenFull << " , the maximum chunk size is " << maxChunkSize + << " , average document size is " << avgRecSize << ". Found " << recCount + << " documents in chunk " + << " ns: " << _ns << " " << _min << " -> " << _max << migrateLog; + + result.appendBool("chunkTooBig", true); + result.appendNumber("estimatedChunkSize", (long long)(recCount * avgRecSize)); + errmsg = "chunk too big to move"; + return false; } - bool clone(OperationContext* txn, string& errmsg , BSONObjBuilder& result ) { - ElapsedTracker tracker(internalQueryExecYieldIterations, - internalQueryExecYieldPeriodMS); + log() << "moveChunk number of documents: " << cloneLocsRemaining() << migrateLog; - int allocSize = 0; - { - AutoGetCollectionForRead ctx(txn, getNS()); + txn->recoveryUnit()->abandonSnapshot(); + return true; + } - std::lock_guard<std::mutex> sl(_mutex); - if (!_active) { - errmsg = "not active"; - return false; - } + bool clone(OperationContext* txn, string& errmsg, BSONObjBuilder& result) { + ElapsedTracker tracker(internalQueryExecYieldIterations, internalQueryExecYieldPeriodMS); - Collection* collection = ctx.getCollection(); - if (!collection) { - errmsg = str::stream() << "collection " << _ns << " does not exist"; - return false; - } + int allocSize = 0; + { + AutoGetCollectionForRead ctx(txn, getNS()); - allocSize = - std::min(BSONObjMaxUserSize, - static_cast<int>((12 + collection->averageObjectSize(txn)) * - cloneLocsRemaining())); + std::lock_guard<std::mutex> sl(_mutex); + if (!_active) { + errmsg = "not active"; + return false; } - bool isBufferFilled = false; - BSONArrayBuilder clonedDocsArrayBuilder(allocSize); - while (!isBufferFilled) { - AutoGetCollectionForRead ctx(txn, getNS()); + Collection* collection = ctx.getCollection(); + if (!collection) { + errmsg = str::stream() << "collection " << _ns << " does not exist"; + return false; + } - std::lock_guard<std::mutex> sl(_mutex); - if (!_active) { - errmsg = "not active"; - return false; - } + allocSize = std::min( + BSONObjMaxUserSize, + static_cast<int>((12 + collection->averageObjectSize(txn)) * cloneLocsRemaining())); + } - // TODO: fix SERVER-16540 race + bool isBufferFilled = false; + BSONArrayBuilder clonedDocsArrayBuilder(allocSize); + while (!isBufferFilled) { + AutoGetCollectionForRead ctx(txn, getNS()); - Collection* collection = ctx.getCollection(); + std::lock_guard<std::mutex> sl(_mutex); + if (!_active) { + errmsg = "not active"; + return false; + } - if (!collection) { - errmsg = str::stream() << "collection " << _ns << " does not exist"; - return false; - } + // TODO: fix SERVER-16540 race - std::lock_guard<std::mutex> lk(_cloneLocsMutex); - set<RecordId>::iterator cloneLocsIter = _cloneLocs.begin(); - for ( ; cloneLocsIter != _cloneLocs.end(); ++cloneLocsIter) { - if (tracker.intervalHasElapsed()) // should I yield? - break; + Collection* collection = ctx.getCollection(); - RecordId dl = *cloneLocsIter; - Snapshotted<BSONObj> doc; - if (!collection->findDoc(txn, dl, &doc)) { - // doc was deleted - continue; - } + if (!collection) { + errmsg = str::stream() << "collection " << _ns << " does not exist"; + return false; + } - // Use the builder size instead of accumulating 'doc's size so that we take - // into consideration the overhead of BSONArray indices, and *always* - // append one doc. - if (clonedDocsArrayBuilder.arrSize() != 0 && - (clonedDocsArrayBuilder.len() + doc.value().objsize() + 1024) - > BSONObjMaxUserSize) { - isBufferFilled = true; // break out of outer while loop - break; - } + std::lock_guard<std::mutex> lk(_cloneLocsMutex); + set<RecordId>::iterator cloneLocsIter = _cloneLocs.begin(); + for (; cloneLocsIter != _cloneLocs.end(); ++cloneLocsIter) { + if (tracker.intervalHasElapsed()) // should I yield? + break; - clonedDocsArrayBuilder.append(doc.value()); + RecordId dl = *cloneLocsIter; + Snapshotted<BSONObj> doc; + if (!collection->findDoc(txn, dl, &doc)) { + // doc was deleted + continue; } - _cloneLocs.erase(_cloneLocs.begin(), cloneLocsIter); - - // Note: must be holding _cloneLocsMutex, don't move this inside while condition! - if (_cloneLocs.empty()) { + // Use the builder size instead of accumulating 'doc's size so that we take + // into consideration the overhead of BSONArray indices, and *always* + // append one doc. + if (clonedDocsArrayBuilder.arrSize() != 0 && + (clonedDocsArrayBuilder.len() + doc.value().objsize() + 1024) > + BSONObjMaxUserSize) { + isBufferFilled = true; // break out of outer while loop break; } + + clonedDocsArrayBuilder.append(doc.value()); } - result.appendArray("objects", clonedDocsArrayBuilder.arr()); - return true; + _cloneLocs.erase(_cloneLocs.begin(), cloneLocsIter); + + // Note: must be holding _cloneLocsMutex, don't move this inside while condition! + if (_cloneLocs.empty()) { + break; + } } - void aboutToDelete( const RecordId& dl ) { - // Even though above we call findDoc to check for existance - // that check only works for non-mmapv1 engines, and this is needed - // for mmapv1. + result.appendArray("objects", clonedDocsArrayBuilder.arr()); + return true; + } - std::lock_guard<std::mutex> lk(_cloneLocsMutex); - _cloneLocs.erase( dl ); - } + void aboutToDelete(const RecordId& dl) { + // Even though above we call findDoc to check for existance + // that check only works for non-mmapv1 engines, and this is needed + // for mmapv1. - std::size_t cloneLocsRemaining() { - std::lock_guard<std::mutex> lk(_cloneLocsMutex); - return _cloneLocs.size(); - } + std::lock_guard<std::mutex> lk(_cloneLocsMutex); + _cloneLocs.erase(dl); + } - long long mbUsed() const { - std::lock_guard<std::mutex> lk(_mutex); - return _memoryUsed / ( 1024 * 1024 ); - } + std::size_t cloneLocsRemaining() { + std::lock_guard<std::mutex> lk(_cloneLocsMutex); + return _cloneLocs.size(); + } - bool getInCriticalSection() const { - std::lock_guard<std::mutex> lk(_mutex); - return _inCriticalSection; - } + long long mbUsed() const { + std::lock_guard<std::mutex> lk(_mutex); + return _memoryUsed / (1024 * 1024); + } - void setInCriticalSection( bool b ) { - std::lock_guard<std::mutex> lk(_mutex); - _inCriticalSection = b; - _inCriticalSectionCV.notify_all(); - } + bool getInCriticalSection() const { + std::lock_guard<std::mutex> lk(_mutex); + return _inCriticalSection; + } - std::string getNS() const { - std::lock_guard<std::mutex> sl(_mutex); - return _ns; - } + void setInCriticalSection(bool b) { + std::lock_guard<std::mutex> lk(_mutex); + _inCriticalSection = b; + _inCriticalSectionCV.notify_all(); + } - /** - * @return true if we are NOT in the critical section - */ - bool waitTillNotInCriticalSection( int maxSecondsToWait ) { - const auto deadline = system_clock::now() + seconds(maxSecondsToWait); - std::unique_lock<std::mutex> lk(_mutex); - while (_inCriticalSection) { - if (std::cv_status::timeout == _inCriticalSectionCV.wait_until(lk, deadline)) - return false; - } + std::string getNS() const { + std::lock_guard<std::mutex> sl(_mutex); + return _ns; + } - return true; + /** + * @return true if we are NOT in the critical section + */ + bool waitTillNotInCriticalSection(int maxSecondsToWait) { + const auto deadline = system_clock::now() + seconds(maxSecondsToWait); + std::unique_lock<std::mutex> lk(_mutex); + while (_inCriticalSection) { + if (std::cv_status::timeout == _inCriticalSectionCV.wait_until(lk, deadline)) + return false; } - bool isActive() const { return _getActive(); } + return true; + } - private: - bool _getActive() const { std::lock_guard<std::mutex> lk(_mutex); return _active; } - void _setActive( bool b ) { std::lock_guard<std::mutex> lk(_mutex); _active = b; } + bool isActive() const { + return _getActive(); + } + +private: + bool _getActive() const { + std::lock_guard<std::mutex> lk(_mutex); + return _active; + } + void _setActive(bool b) { + std::lock_guard<std::mutex> lk(_mutex); + _active = b; + } + /** + * Used to commit work for LogOpForSharding. Used to keep track of changes in documents + * that are part of a chunk being migrated. + */ + class LogOpForShardingHandler : public RecoveryUnit::Change { + public: /** - * Used to commit work for LogOpForSharding. Used to keep track of changes in documents - * that are part of a chunk being migrated. + * Invariant: idObj should belong to a document that is part of the active chunk + * being migrated. */ - class LogOpForShardingHandler : public RecoveryUnit::Change { - public: - /** - * Invariant: idObj should belong to a document that is part of the active chunk - * being migrated. - */ - LogOpForShardingHandler(MigrateFromStatus* migrateFromStatus, - const BSONObj& idObj, - const char op): - _migrateFromStatus(migrateFromStatus), - _idObj(idObj.getOwned()), - _op(op) { - } - - virtual void commit() { - switch (_op) { + LogOpForShardingHandler(MigrateFromStatus* migrateFromStatus, + const BSONObj& idObj, + const char op) + : _migrateFromStatus(migrateFromStatus), _idObj(idObj.getOwned()), _op(op) {} + + virtual void commit() { + switch (_op) { case 'd': { std::lock_guard<std::mutex> sl(_migrateFromStatus->_mutex); _migrateFromStatus->_deleted.push_back(_idObj); @@ -746,8 +738,7 @@ namespace { } case 'i': - case 'u': - { + case 'u': { std::lock_guard<std::mutex> sl(_migrateFromStatus->_mutex); _migrateFromStatus->_reload.push_back(_idObj); _migrateFromStatus->_memoryUsed += _idObj.firstElement().size() + 5; @@ -756,2184 +747,2146 @@ namespace { default: invariant(false); - - } } + } - virtual void rollback() { } + virtual void rollback() {} - private: - MigrateFromStatus* _migrateFromStatus; - const BSONObj _idObj; - const char _op; - }; + private: + MigrateFromStatus* _migrateFromStatus; + const BSONObj _idObj; + const char _op; + }; - /** - * Used to receive invalidation notifications. - * - * XXX: move to the exec/ directory. - */ - class DeleteNotificationStage : public PlanStage { - public: - virtual void invalidate(OperationContext* txn, - const RecordId& dl, - InvalidationType type); + /** + * Used to receive invalidation notifications. + * + * XXX: move to the exec/ directory. + */ + class DeleteNotificationStage : public PlanStage { + public: + virtual void invalidate(OperationContext* txn, const RecordId& dl, InvalidationType type); - virtual StageState work(WorkingSetID* out) { - invariant( false ); - } - virtual bool isEOF() { - invariant( false ); - return false; - } - virtual void kill() { - } - virtual void saveState() { - invariant( false ); - } - virtual void restoreState(OperationContext* opCtx) { - invariant( false ); - } - virtual PlanStageStats* getStats() { - invariant( false ); - return NULL; - } - virtual CommonStats* getCommonStats() const { - invariant( false ); - return NULL; - } - virtual SpecificStats* getSpecificStats() const { - invariant( false ); - return NULL; - } - virtual std::vector<PlanStage*> getChildren() const { - vector<PlanStage*> empty; - return empty; - } - virtual StageType stageType() const { - return STAGE_NOTIFY_DELETE; - } - }; + virtual StageState work(WorkingSetID* out) { + invariant(false); + } + virtual bool isEOF() { + invariant(false); + return false; + } + virtual void kill() {} + virtual void saveState() { + invariant(false); + } + virtual void restoreState(OperationContext* opCtx) { + invariant(false); + } + virtual PlanStageStats* getStats() { + invariant(false); + return NULL; + } + virtual CommonStats* getCommonStats() const { + invariant(false); + return NULL; + } + virtual SpecificStats* getSpecificStats() const { + invariant(false); + return NULL; + } + virtual std::vector<PlanStage*> getChildren() const { + vector<PlanStage*> empty; + return empty; + } + virtual StageType stageType() const { + return STAGE_NOTIFY_DELETE; + } + }; - // - // All member variables are labeled with one of the following codes indicating the - // synchronization rules for accessing them. - // - // (M) Must hold _mutex for access. - // (MG) For reads, _mutex *OR* Global IX Lock must be held. - // For writes, the _mutex *AND* (Global Shared or Exclusive Lock) must be held. - // (C) Must hold _cloneLocsMutex for access. - // - // Locking order: - // - // Global Lock -> _mutex -> _cloneLocsMutex + // + // All member variables are labeled with one of the following codes indicating the + // synchronization rules for accessing them. + // + // (M) Must hold _mutex for access. + // (MG) For reads, _mutex *OR* Global IX Lock must be held. + // For writes, the _mutex *AND* (Global Shared or Exclusive Lock) must be held. + // (C) Must hold _cloneLocsMutex for access. + // + // Locking order: + // + // Global Lock -> _mutex -> _cloneLocsMutex - mutable std::mutex _mutex; + mutable std::mutex _mutex; - std::condition_variable _inCriticalSectionCV; // (M) + std::condition_variable _inCriticalSectionCV; // (M) - // Is migration currently in critical section. This can be used to block new writes. - bool _inCriticalSection; // (M) + // Is migration currently in critical section. This can be used to block new writes. + bool _inCriticalSection; // (M) - std::unique_ptr<PlanExecutor> _deleteNotifyExec; // (M) + std::unique_ptr<PlanExecutor> _deleteNotifyExec; // (M) - // List of _id of documents that were modified that must be re-cloned. - list<BSONObj> _reload; // (M) + // List of _id of documents that were modified that must be re-cloned. + list<BSONObj> _reload; // (M) - // List of _id of documents that were deleted during clone that should be deleted later. - list<BSONObj> _deleted; // (M) + // List of _id of documents that were deleted during clone that should be deleted later. + list<BSONObj> _deleted; // (M) - // bytes in _reload + _deleted - long long _memoryUsed; // (M) + // bytes in _reload + _deleted + long long _memoryUsed; // (M) - // If a migration is currently active. - bool _active; // (MG) + // If a migration is currently active. + bool _active; // (MG) - string _ns; // (MG) - BSONObj _min; // (MG) - BSONObj _max; // (MG) - BSONObj _shardKeyPattern; // (MG) + string _ns; // (MG) + BSONObj _min; // (MG) + BSONObj _max; // (MG) + BSONObj _shardKeyPattern; // (MG) - mutable std::mutex _cloneLocsMutex; + mutable std::mutex _cloneLocsMutex; - // List of record id that needs to be transferred from here to the other side. - set<RecordId> _cloneLocs; // (C) + // List of record id that needs to be transferred from here to the other side. + set<RecordId> _cloneLocs; // (C) - } migrateFromStatus; +} migrateFromStatus; - void MigrateFromStatus::DeleteNotificationStage::invalidate(OperationContext *txn, - const RecordId& dl, - InvalidationType type) { - if ( type == INVALIDATION_DELETION ) { - migrateFromStatus.aboutToDelete( dl ); - } +void MigrateFromStatus::DeleteNotificationStage::invalidate(OperationContext* txn, + const RecordId& dl, + InvalidationType type) { + if (type == INVALIDATION_DELETION) { + migrateFromStatus.aboutToDelete(dl); } +} - struct MigrateStatusHolder { - MigrateStatusHolder( OperationContext* txn, - const std::string& ns , - const BSONObj& min , - const BSONObj& max , - const BSONObj& shardKeyPattern ) - : _txn(txn) { - _isAnotherMigrationActive = - !migrateFromStatus.start(txn, ns, min, max, shardKeyPattern); - } - ~MigrateStatusHolder() { - if (!_isAnotherMigrationActive) { - migrateFromStatus.done(_txn); - } +struct MigrateStatusHolder { + MigrateStatusHolder(OperationContext* txn, + const std::string& ns, + const BSONObj& min, + const BSONObj& max, + const BSONObj& shardKeyPattern) + : _txn(txn) { + _isAnotherMigrationActive = !migrateFromStatus.start(txn, ns, min, max, shardKeyPattern); + } + ~MigrateStatusHolder() { + if (!_isAnotherMigrationActive) { + migrateFromStatus.done(_txn); } + } - bool isAnotherMigrationActive() const { - return _isAnotherMigrationActive; - } + bool isAnotherMigrationActive() const { + return _isAnotherMigrationActive; + } - private: - OperationContext* _txn; - bool _isAnotherMigrationActive; - }; +private: + OperationContext* _txn; + bool _isAnotherMigrationActive; +}; + +void logOpForSharding(OperationContext* txn, + const char* opstr, + const char* ns, + const BSONObj& obj, + BSONObj* patt, + bool notInActiveChunk) { + migrateFromStatus.logOp(txn, opstr, ns, obj, patt, notInActiveChunk); +} - void logOpForSharding(OperationContext* txn, - const char * opstr, - const char * ns, - const BSONObj& obj, - BSONObj * patt, - bool notInActiveChunk) { - migrateFromStatus.logOp(txn, opstr, ns, obj, patt, notInActiveChunk); +class TransferModsCommand : public ChunkCommandHelper { +public: + void help(std::stringstream& h) const { + h << "internal"; } + TransferModsCommand() : ChunkCommandHelper("_transferMods") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::internal); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + return migrateFromStatus.transferMods(txn, errmsg, result); + } +} transferModsCommand; - class TransferModsCommand : public ChunkCommandHelper { - public: - void help(std::stringstream& h) const { h << "internal"; } - TransferModsCommand() : ChunkCommandHelper( "_transferMods" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::internal); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - return migrateFromStatus.transferMods(txn, errmsg, result); - } - } transferModsCommand; - - - class InitialCloneCommand : public ChunkCommandHelper { - public: - void help(std::stringstream& h) const { h << "internal"; } - InitialCloneCommand() : ChunkCommandHelper( "_migrateClone" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::internal); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - return migrateFromStatus.clone(txn, errmsg, result); - } - } initialCloneCommand; - - // Tests can pause / resume moveChunk's progress at each step by enabling / disabling each fail point. - MONGO_FP_DECLARE(moveChunkHangAtStep1); - MONGO_FP_DECLARE(moveChunkHangAtStep2); - MONGO_FP_DECLARE(moveChunkHangAtStep3); - MONGO_FP_DECLARE(moveChunkHangAtStep4); - MONGO_FP_DECLARE(moveChunkHangAtStep5); - MONGO_FP_DECLARE(moveChunkHangAtStep6); - /** - * this is the main entry for moveChunk - * called to initial a move - * usually by a mongos - * this is called on the "from" side - * - * Format: - * { - * moveChunk: "namespace", - * from: "hostAndPort", - * fromShard: "shardName", - * to: "hostAndPort", - * toShard: "shardName", - * min: {}, - * max: {}, - * maxChunkBytes: numeric, - * configdb: "hostAndPort", - * - * // optional - * secondaryThrottle: bool, //defaults to true. - * writeConcern: {} // applies to individual writes. - * } - */ - class MoveChunkCommand : public Command { - public: - MoveChunkCommand() : Command( "moveChunk" ) {} - virtual void help( std::stringstream& help ) const { - help << "should not be calling this directly"; - } - - virtual bool slaveOk() const { return false; } - virtual bool adminOnly() const { return true; } - virtual bool isWriteCommandForConfigServer() const { return false; } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), - ActionType::moveChunk)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - return Status::OK(); - } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); - } - - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - // 1. Parse options - // 2. Make sure my view is complete and lock the distributed lock to ensure shard - // metadata stability. - // 3. Migration - // Retrieve all RecordIds, which need to be migrated in order to do as little seeking - // as possible during transfer. Retrieval of the RecordIds happens under a collection - // lock, but then the collection lock is dropped. This opens up an opportunity for - // repair or compact to invalidate these RecordIds, because these commands do not - // synchronized with migration. Note that data modifications are not a problem, - // because we are registered for change notifications. - // - // 4. pause till migrate caught up - // 5. LOCK - // a) update my config, essentially locking - // b) finish migrate - // c) update config server - // d) logChange to config server - // 6. wait for all current cursors to expire - // 7. remove data locally - - // ------------------------------- - - // 1. - string ns = parseNs(dbname, cmdObj); - - // The shard addresses, redundant, but allows for validation - string toShardHost = cmdObj["to"].str(); - string fromShardHost = cmdObj["from"].str(); - - // The shard names - string toShardName = cmdObj["toShard"].str(); - string fromShardName = cmdObj["fromShard"].str(); - - // Process secondary throttle settings and assign defaults if necessary. - BSONObj secThrottleObj; - WriteConcernOptions writeConcern; - Status status = writeConcern.parseSecondaryThrottle(cmdObj, &secThrottleObj); - - if (!status.isOK()){ - if (status.code() != ErrorCodes::WriteConcernNotDefined) { - warning() << status.toString(); - return appendCommandStatus(result, status); - } +class InitialCloneCommand : public ChunkCommandHelper { +public: + void help(std::stringstream& h) const { + h << "internal"; + } + InitialCloneCommand() : ChunkCommandHelper("_migrateClone") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::internal); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + return migrateFromStatus.clone(txn, errmsg, result); + } +} initialCloneCommand; - writeConcern = getDefaultWriteConcern(); - } - else { - repl::ReplicationCoordinator* replCoordinator = - repl::getGlobalReplicationCoordinator(); +// Tests can pause / resume moveChunk's progress at each step by enabling / disabling each fail point. +MONGO_FP_DECLARE(moveChunkHangAtStep1); +MONGO_FP_DECLARE(moveChunkHangAtStep2); +MONGO_FP_DECLARE(moveChunkHangAtStep3); +MONGO_FP_DECLARE(moveChunkHangAtStep4); +MONGO_FP_DECLARE(moveChunkHangAtStep5); +MONGO_FP_DECLARE(moveChunkHangAtStep6); - if (replCoordinator->getReplicationMode() == - repl::ReplicationCoordinator::modeMasterSlave && - writeConcern.shouldWaitForOtherNodes()) { - warning() << "moveChunk cannot check if secondary throttle setting " - << writeConcern.toBSON() - << " can be enforced in a master slave configuration"; - } +/** + * this is the main entry for moveChunk + * called to initial a move + * usually by a mongos + * this is called on the "from" side + * + * Format: + * { + * moveChunk: "namespace", + * from: "hostAndPort", + * fromShard: "shardName", + * to: "hostAndPort", + * toShard: "shardName", + * min: {}, + * max: {}, + * maxChunkBytes: numeric, + * configdb: "hostAndPort", + * + * // optional + * secondaryThrottle: bool, //defaults to true. + * writeConcern: {} // applies to individual writes. + * } + */ +class MoveChunkCommand : public Command { +public: + MoveChunkCommand() : Command("moveChunk") {} + virtual void help(std::stringstream& help) const { + help << "should not be calling this directly"; + } - Status status = replCoordinator->checkIfWriteConcernCanBeSatisfied(writeConcern); - if (!status.isOK() && status != ErrorCodes::NoReplicationEnabled) { - warning() << status.toString(); - return appendCommandStatus(result, status); - } - } + virtual bool slaveOk() const { + return false; + } + virtual bool adminOnly() const { + return true; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::moveChunk)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); + } + return Status::OK(); + } + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } - if (writeConcern.shouldWaitForOtherNodes() && - writeConcern.wTimeout == WriteConcernOptions::kNoTimeout) { - // Don't allow no timeout. - writeConcern.wTimeout = kDefaultWTimeoutMs; - } + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + // 1. Parse options + // 2. Make sure my view is complete and lock the distributed lock to ensure shard + // metadata stability. + // 3. Migration + // Retrieve all RecordIds, which need to be migrated in order to do as little seeking + // as possible during transfer. Retrieval of the RecordIds happens under a collection + // lock, but then the collection lock is dropped. This opens up an opportunity for + // repair or compact to invalidate these RecordIds, because these commands do not + // synchronized with migration. Note that data modifications are not a problem, + // because we are registered for change notifications. + // + // 4. pause till migrate caught up + // 5. LOCK + // a) update my config, essentially locking + // b) finish migrate + // c) update config server + // d) logChange to config server + // 6. wait for all current cursors to expire + // 7. remove data locally - // Do inline deletion - bool waitForDelete = cmdObj["waitForDelete"].trueValue(); - if (waitForDelete) { - log() << "moveChunk waiting for full cleanup after move"; - } + // ------------------------------- - BSONObj min = cmdObj["min"].Obj(); - BSONObj max = cmdObj["max"].Obj(); - BSONElement maxSizeElem = cmdObj["maxChunkSizeBytes"]; + // 1. + string ns = parseNs(dbname, cmdObj); - if ( ns.empty() ) { - errmsg = "need to specify namespace in command"; - return false; - } + // The shard addresses, redundant, but allows for validation + string toShardHost = cmdObj["to"].str(); + string fromShardHost = cmdObj["from"].str(); - if ( toShardName.empty() ) { - errmsg = "need to specify shard to move chunk to"; - return false; - } - if ( fromShardName.empty() ) { - errmsg = "need to specify shard to move chunk from"; - return false; - } + // The shard names + string toShardName = cmdObj["toShard"].str(); + string fromShardName = cmdObj["fromShard"].str(); - if ( min.isEmpty() ) { - errmsg = "need to specify a min"; - return false; - } + // Process secondary throttle settings and assign defaults if necessary. + BSONObj secThrottleObj; + WriteConcernOptions writeConcern; + Status status = writeConcern.parseSecondaryThrottle(cmdObj, &secThrottleObj); - if ( max.isEmpty() ) { - errmsg = "need to specify a max"; - return false; + if (!status.isOK()) { + if (status.code() != ErrorCodes::WriteConcernNotDefined) { + warning() << status.toString(); + return appendCommandStatus(result, status); } - if ( maxSizeElem.eoo() || ! maxSizeElem.isNumber() ) { - errmsg = "need to specify maxChunkSizeBytes"; - return false; - } - const long long maxChunkSize = maxSizeElem.numberLong(); // in bytes + writeConcern = getDefaultWriteConcern(); + } else { + repl::ReplicationCoordinator* replCoordinator = repl::getGlobalReplicationCoordinator(); - // This could be the first call that enables sharding - make sure we initialize the - // sharding state for this shard. - if ( ! shardingState.enabled() ) { - if ( cmdObj["configdb"].type() != String ) { - errmsg = "sharding not enabled"; - warning() << errmsg; - return false; - } - string configdb = cmdObj["configdb"].String(); - ShardingState::initialize(configdb); + if (replCoordinator->getReplicationMode() == + repl::ReplicationCoordinator::modeMasterSlave && + writeConcern.shouldWaitForOtherNodes()) { + warning() << "moveChunk cannot check if secondary throttle setting " + << writeConcern.toBSON() + << " can be enforced in a master slave configuration"; } - // Initialize our current shard name in the shard state if needed - shardingState.gotShardName(fromShardName); - - // Make sure we're as up-to-date as possible with shard information - // This catches the case where we had to previously changed a shard's host by - // removing/adding a shard with the same name - Shard::reloadShardInfo(); - - ConnectionString configLoc = ConnectionString::parse(shardingState.getConfigServer(), - errmsg); - if (!configLoc.isValid()) { - warning() << errmsg; - return false; + Status status = replCoordinator->checkIfWriteConcernCanBeSatisfied(writeConcern); + if (!status.isOK() && status != ErrorCodes::NoReplicationEnabled) { + warning() << status.toString(); + return appendCommandStatus(result, status); } + } - MoveTimingHelper timing(txn, "from" , ns , min , max , 6 /* steps */ , &errmsg, - toShardName, fromShardName ); + if (writeConcern.shouldWaitForOtherNodes() && + writeConcern.wTimeout == WriteConcernOptions::kNoTimeout) { + // Don't allow no timeout. + writeConcern.wTimeout = kDefaultWTimeoutMs; + } - log() << "received moveChunk request: " << cmdObj << migrateLog; + // Do inline deletion + bool waitForDelete = cmdObj["waitForDelete"].trueValue(); + if (waitForDelete) { + log() << "moveChunk waiting for full cleanup after move"; + } - timing.done(1); - MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep1); + BSONObj min = cmdObj["min"].Obj(); + BSONObj max = cmdObj["max"].Obj(); + BSONElement maxSizeElem = cmdObj["maxChunkSizeBytes"]; - // 2. + if (ns.empty()) { + errmsg = "need to specify namespace in command"; + return false; + } - if ( migrateFromStatus.isActive() ) { - errmsg = "migration already in progress"; - warning() << errmsg; - return false; - } + if (toShardName.empty()) { + errmsg = "need to specify shard to move chunk to"; + return false; + } + if (fromShardName.empty()) { + errmsg = "need to specify shard to move chunk from"; + return false; + } - // - // Get the distributed lock - // + if (min.isEmpty()) { + errmsg = "need to specify a min"; + return false; + } - string whyMessage(str::stream() << "migrating chunk [" << minKey << ", " << maxKey - << ") in " << ns); - auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock( - ns, whyMessage); + if (max.isEmpty()) { + errmsg = "need to specify a max"; + return false; + } - if (!scopedDistLock.isOK()) { - errmsg = stream() << "could not acquire collection lock for " << ns - << " to migrate chunk [" << minKey << "," << maxKey << ")" - << causedBy(scopedDistLock.getStatus()); + if (maxSizeElem.eoo() || !maxSizeElem.isNumber()) { + errmsg = "need to specify maxChunkSizeBytes"; + return false; + } + const long long maxChunkSize = maxSizeElem.numberLong(); // in bytes + // This could be the first call that enables sharding - make sure we initialize the + // sharding state for this shard. + if (!shardingState.enabled()) { + if (cmdObj["configdb"].type() != String) { + errmsg = "sharding not enabled"; warning() << errmsg; return false; } + string configdb = cmdObj["configdb"].String(); + ShardingState::initialize(configdb); + } - BSONObj chunkInfo = BSON("min" << min << "max" << max << - "from" << fromShardName << "to" << toShardName); + // Initialize our current shard name in the shard state if needed + shardingState.gotShardName(fromShardName); - grid.catalogManager()->logChange(txn, "moveChunk.start", ns, chunkInfo); + // Make sure we're as up-to-date as possible with shard information + // This catches the case where we had to previously changed a shard's host by + // removing/adding a shard with the same name + Shard::reloadShardInfo(); - // Always refresh our metadata remotely - ChunkVersion origShardVersion; - Status refreshStatus = shardingState.refreshMetadataNow(txn, ns, &origShardVersion); + ConnectionString configLoc = + ConnectionString::parse(shardingState.getConfigServer(), errmsg); + if (!configLoc.isValid()) { + warning() << errmsg; + return false; + } - if (!refreshStatus.isOK()) { - errmsg = str::stream() << "moveChunk cannot start migrate of chunk " - << "[" << minKey << "," << maxKey << ")" - << causedBy(refreshStatus.reason()); + MoveTimingHelper timing( + txn, "from", ns, min, max, 6 /* steps */, &errmsg, toShardName, fromShardName); - warning() << errmsg; - return false; - } - - if (origShardVersion.majorVersion() == 0) { - // It makes no sense to migrate if our version is zero and we have no chunks - errmsg = str::stream() << "moveChunk cannot start migrate of chunk " - << "[" << minKey << "," << maxKey << ")" - << " with zero shard version"; + log() << "received moveChunk request: " << cmdObj << migrateLog; - warning() << errmsg; - return false; - } + timing.done(1); + MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep1); - // From mongos >= v3.0. - BSONElement epochElem(cmdObj["epoch"]); - if (epochElem.type() == jstOID) { - OID cmdEpoch = epochElem.OID(); + // 2. - if (cmdEpoch != origShardVersion.epoch()) { - errmsg = str::stream() << "moveChunk cannot move chunk " - << "[" << minKey << "," - << maxKey << "), " - << "collection may have been dropped. " - << "current epoch: " << origShardVersion.epoch() - << ", cmd epoch: " << cmdEpoch; - warning() << errmsg; - return false; - } - } + if (migrateFromStatus.isActive()) { + errmsg = "migration already in progress"; + warning() << errmsg; + return false; + } - // Get collection metadata - const CollectionMetadataPtr origCollMetadata(shardingState.getCollectionMetadata(ns)); + // + // Get the distributed lock + // - // With nonzero shard version, we must have metadata - invariant(NULL != origCollMetadata); + string whyMessage(str::stream() << "migrating chunk [" << minKey << ", " << maxKey + << ") in " << ns); + auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock(ns, whyMessage); - ChunkVersion origCollVersion = origCollMetadata->getCollVersion(); - BSONObj shardKeyPattern = origCollMetadata->getKeyPattern(); + if (!scopedDistLock.isOK()) { + errmsg = stream() << "could not acquire collection lock for " << ns + << " to migrate chunk [" << minKey << "," << maxKey << ")" + << causedBy(scopedDistLock.getStatus()); - // With nonzero shard version, we must have a coll version >= our shard version - invariant(origCollVersion >= origShardVersion); + warning() << errmsg; + return false; + } - // With nonzero shard version, we must have a shard key - invariant(!shardKeyPattern.isEmpty()); + BSONObj chunkInfo = + BSON("min" << min << "max" << max << "from" << fromShardName << "to" << toShardName); - ChunkType origChunk; - if (!origCollMetadata->getNextChunk(min, &origChunk) - || origChunk.getMin().woCompare(min) - || origChunk.getMax().woCompare(max)) { + grid.catalogManager()->logChange(txn, "moveChunk.start", ns, chunkInfo); - // Our boundaries are different from those passed in - errmsg = str::stream() << "moveChunk cannot find chunk " - << "[" << minKey << "," << maxKey << ")" - << " to migrate, the chunk boundaries may be stale"; + // Always refresh our metadata remotely + ChunkVersion origShardVersion; + Status refreshStatus = shardingState.refreshMetadataNow(txn, ns, &origShardVersion); - warning() << errmsg; - return false; - } + if (!refreshStatus.isOK()) { + errmsg = str::stream() << "moveChunk cannot start migrate of chunk " + << "[" << minKey << "," << maxKey << ")" + << causedBy(refreshStatus.reason()); - log() << "moveChunk request accepted at version " << origShardVersion; + warning() << errmsg; + return false; + } - timing.done(2); - MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep2); + if (origShardVersion.majorVersion() == 0) { + // It makes no sense to migrate if our version is zero and we have no chunks + errmsg = str::stream() << "moveChunk cannot start migrate of chunk " + << "[" << minKey << "," << maxKey << ")" + << " with zero shard version"; - // 3. - MigrateStatusHolder statusHolder(txn, ns, min, max, shardKeyPattern); + warning() << errmsg; + return false; + } - if (statusHolder.isAnotherMigrationActive()) { - errmsg = "moveChunk is already in progress from this shard"; + // From mongos >= v3.0. + BSONElement epochElem(cmdObj["epoch"]); + if (epochElem.type() == jstOID) { + OID cmdEpoch = epochElem.OID(); + + if (cmdEpoch != origShardVersion.epoch()) { + errmsg = str::stream() << "moveChunk cannot move chunk " + << "[" << minKey << "," << maxKey << "), " + << "collection may have been dropped. " + << "current epoch: " << origShardVersion.epoch() + << ", cmd epoch: " << cmdEpoch; warning() << errmsg; return false; } + } - ConnectionString fromShardCS; - ConnectionString toShardCS; + // Get collection metadata + const CollectionMetadataPtr origCollMetadata(shardingState.getCollectionMetadata(ns)); - // Resolve the shard connection strings. - { - std::shared_ptr<Shard> fromShard = - grid.shardRegistry()->getShard(fromShardName); - uassert(28674, - str::stream() << "Source shard " << fromShardName - << " is missing. This indicates metadata corruption.", - fromShard); + // With nonzero shard version, we must have metadata + invariant(NULL != origCollMetadata); - fromShardCS = fromShard->getConnString(); + ChunkVersion origCollVersion = origCollMetadata->getCollVersion(); + BSONObj shardKeyPattern = origCollMetadata->getKeyPattern(); - std::shared_ptr<Shard> toShard = grid.shardRegistry()->getShard(toShardName); - uassert(28675, - str::stream() << "Destination shard " << toShardName - << " is missing. This indicates metadata corruption.", - toShard); + // With nonzero shard version, we must have a coll version >= our shard version + invariant(origCollVersion >= origShardVersion); - toShardCS = toShard->getConnString(); - } + // With nonzero shard version, we must have a shard key + invariant(!shardKeyPattern.isEmpty()); - { - // See comment at the top of the function for more information on what - // synchronization is used here. - if (!migrateFromStatus.storeCurrentLocs(txn, maxChunkSize, errmsg, result)) { - warning() << errmsg; - return false; - } + ChunkType origChunk; + if (!origCollMetadata->getNextChunk(min, &origChunk) || origChunk.getMin().woCompare(min) || + origChunk.getMax().woCompare(max)) { + // Our boundaries are different from those passed in + errmsg = str::stream() << "moveChunk cannot find chunk " + << "[" << minKey << "," << maxKey << ")" + << " to migrate, the chunk boundaries may be stale"; - ScopedDbConnection connTo(toShardCS); - BSONObj res; - bool ok; - - const bool isSecondaryThrottle(writeConcern.shouldWaitForOtherNodes()); - - BSONObjBuilder recvChunkStartBuilder; - recvChunkStartBuilder.append("_recvChunkStart", ns); - recvChunkStartBuilder.append("from", fromShardCS.toString()); - recvChunkStartBuilder.append("fromShardName", fromShardName); - recvChunkStartBuilder.append("toShardName", toShardName); - recvChunkStartBuilder.append("min", min); - recvChunkStartBuilder.append("max", max); - recvChunkStartBuilder.append("shardKeyPattern", shardKeyPattern); - recvChunkStartBuilder.append("configServer", shardingState.getConfigServer()); - recvChunkStartBuilder.append("secondaryThrottle", isSecondaryThrottle); - - // Follow the same convention in moveChunk. - if (isSecondaryThrottle && !secThrottleObj.isEmpty()) { - recvChunkStartBuilder.append("writeConcern", secThrottleObj); - } + warning() << errmsg; + return false; + } - try{ - ok = connTo->runCommand("admin", recvChunkStartBuilder.done(), res); - } - catch( DBException& e ){ - errmsg = str::stream() << "moveChunk could not contact to: shard " - << toShardName << " to start transfer" << causedBy( e ); - warning() << errmsg; - return false; - } + log() << "moveChunk request accepted at version " << origShardVersion; - connTo.done(); + timing.done(2); + MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep2); - if ( ! ok ) { - errmsg = "moveChunk failed to engage TO-shard in the data transfer: "; - verify( res["errmsg"].type() ); - errmsg += res["errmsg"].String(); - result.append( "cause" , res ); - warning() << errmsg; - return false; - } + // 3. + MigrateStatusHolder statusHolder(txn, ns, min, max, shardKeyPattern); - } - timing.done( 3 ); - MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep3); + if (statusHolder.isAnotherMigrationActive()) { + errmsg = "moveChunk is already in progress from this shard"; + warning() << errmsg; + return false; + } - // 4. + ConnectionString fromShardCS; + ConnectionString toShardCS; - // Track last result from TO shard for sanity check - BSONObj res; - for ( int i=0; i<86400; i++ ) { // don't want a single chunk move to take more than a day - invariant(!txn->lockState()->isLocked()); + // Resolve the shard connection strings. + { + std::shared_ptr<Shard> fromShard = grid.shardRegistry()->getShard(fromShardName); + uassert(28674, + str::stream() << "Source shard " << fromShardName + << " is missing. This indicates metadata corruption.", + fromShard); - // Exponential sleep backoff, up to 1024ms. Don't sleep much on the first few - // iterations, since we want empty chunk migrations to be fast. - sleepmillis(1 << std::min(i, 10)); + fromShardCS = fromShard->getConnString(); - ScopedDbConnection conn(toShardCS); - bool ok; - res = BSONObj(); - try { - ok = conn->runCommand( "admin" , BSON( "_recvChunkStatus" << 1 ) , res ); - res = res.getOwned(); - } - catch( DBException& e ){ - errmsg = str::stream() << "moveChunk could not contact to: shard " << toShardName << " to monitor transfer" << causedBy( e ); - warning() << errmsg; - return false; - } - - conn.done(); - - if ( res["ns"].str() != ns || - res["from"].str() != fromShardCS.toString() || - !res["min"].isABSONObj() || - res["min"].Obj().woCompare(min) != 0 || - !res["max"].isABSONObj() || - res["max"].Obj().woCompare(max) != 0 ) { - // This can happen when the destination aborted the migration and - // received another recvChunk before this thread sees the transition - // to the abort state. This is currently possible only if multiple migrations - // are happening at once. This is an unfortunate consequence of the shards not - // being able to keep track of multiple incoming and outgoing migrations. - errmsg = str::stream() << "Destination shard aborted migration, " - "now running a new one: " << res; - warning() << errmsg; - return false; - } + std::shared_ptr<Shard> toShard = grid.shardRegistry()->getShard(toShardName); + uassert(28675, + str::stream() << "Destination shard " << toShardName + << " is missing. This indicates metadata corruption.", + toShard); - LOG(0) << "moveChunk data transfer progress: " << res << " my mem used: " << migrateFromStatus.mbUsed() << migrateLog; + toShardCS = toShard->getConnString(); + } - if ( ! ok || res["state"].String() == "fail" ) { - warning() << "moveChunk error transferring data caused migration abort: " << res << migrateLog; - errmsg = "data transfer error"; - result.append( "cause" , res ); - return false; - } + { + // See comment at the top of the function for more information on what + // synchronization is used here. + if (!migrateFromStatus.storeCurrentLocs(txn, maxChunkSize, errmsg, result)) { + warning() << errmsg; + return false; + } - if ( res["state"].String() == "steady" ) - break; + ScopedDbConnection connTo(toShardCS); + BSONObj res; + bool ok; - if ( migrateFromStatus.mbUsed() > (500 * 1024 * 1024) ) { - // This is too much memory for us to use for this so we're going to abort - // the migrate - ScopedDbConnection conn(toShardCS); + const bool isSecondaryThrottle(writeConcern.shouldWaitForOtherNodes()); - BSONObj res; - if (!conn->runCommand("admin", BSON("_recvChunkAbort" << 1), res)) { - warning() << "Error encountered while trying to abort migration on " - << "destination shard" << toShardCS; - } + BSONObjBuilder recvChunkStartBuilder; + recvChunkStartBuilder.append("_recvChunkStart", ns); + recvChunkStartBuilder.append("from", fromShardCS.toString()); + recvChunkStartBuilder.append("fromShardName", fromShardName); + recvChunkStartBuilder.append("toShardName", toShardName); + recvChunkStartBuilder.append("min", min); + recvChunkStartBuilder.append("max", max); + recvChunkStartBuilder.append("shardKeyPattern", shardKeyPattern); + recvChunkStartBuilder.append("configServer", shardingState.getConfigServer()); + recvChunkStartBuilder.append("secondaryThrottle", isSecondaryThrottle); - res = res.getOwned(); - conn.done(); - error() << "aborting migrate because too much memory used res: " << res << migrateLog; - errmsg = "aborting migrate because too much memory used"; - result.appendBool( "split" , true ); - return false; - } + // Follow the same convention in moveChunk. + if (isSecondaryThrottle && !secThrottleObj.isEmpty()) { + recvChunkStartBuilder.append("writeConcern", secThrottleObj); + } - txn->checkForInterrupt(); + try { + ok = connTo->runCommand("admin", recvChunkStartBuilder.done(), res); + } catch (DBException& e) { + errmsg = str::stream() << "moveChunk could not contact to: shard " << toShardName + << " to start transfer" << causedBy(e); + warning() << errmsg; + return false; } - timing.done(4); - MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep4); + connTo.done(); - // 5. + if (!ok) { + errmsg = "moveChunk failed to engage TO-shard in the data transfer: "; + verify(res["errmsg"].type()); + errmsg += res["errmsg"].String(); + result.append("cause", res); + warning() << errmsg; + return false; + } + } + timing.done(3); + MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep3); - // Before we get into the critical section of the migration, let's double check - // that the docs have been cloned, the config servers are reachable, - // and the lock is in place. - log() << "About to check if it is safe to enter critical section"; + // 4. - // Ensure all cloned docs have actually been transferred - std::size_t locsRemaining = migrateFromStatus.cloneLocsRemaining(); - if ( locsRemaining != 0 ) { + // Track last result from TO shard for sanity check + BSONObj res; + for (int i = 0; i < 86400; i++) { // don't want a single chunk move to take more than a day + invariant(!txn->lockState()->isLocked()); - errmsg = - str::stream() << "moveChunk cannot enter critical section before all data is" - << " cloned, " << locsRemaining << " locs were not transferred" - << " but to-shard reported " << res; + // Exponential sleep backoff, up to 1024ms. Don't sleep much on the first few + // iterations, since we want empty chunk migrations to be fast. + sleepmillis(1 << std::min(i, 10)); - // Should never happen, but safe to abort before critical section - error() << errmsg << migrateLog; - dassert( false ); + ScopedDbConnection conn(toShardCS); + bool ok; + res = BSONObj(); + try { + ok = conn->runCommand("admin", BSON("_recvChunkStatus" << 1), res); + res = res.getOwned(); + } catch (DBException& e) { + errmsg = str::stream() << "moveChunk could not contact to: shard " << toShardName + << " to monitor transfer" << causedBy(e); + warning() << errmsg; return false; } - // Ensure distributed lock still held - Status lockStatus = scopedDistLock.getValue().checkStatus(); - if (!lockStatus.isOK()) { - errmsg = str::stream() << "not entering migrate critical section because " - << lockStatus.toString(); + conn.done(); + + if (res["ns"].str() != ns || res["from"].str() != fromShardCS.toString() || + !res["min"].isABSONObj() || res["min"].Obj().woCompare(min) != 0 || + !res["max"].isABSONObj() || res["max"].Obj().woCompare(max) != 0) { + // This can happen when the destination aborted the migration and + // received another recvChunk before this thread sees the transition + // to the abort state. This is currently possible only if multiple migrations + // are happening at once. This is an unfortunate consequence of the shards not + // being able to keep track of multiple incoming and outgoing migrations. + errmsg = str::stream() << "Destination shard aborted migration, " + "now running a new one: " << res; warning() << errmsg; return false; } - log() << "About to enter migrate critical section"; + LOG(0) << "moveChunk data transfer progress: " << res + << " my mem used: " << migrateFromStatus.mbUsed() << migrateLog; - { - // 5.a - // we're under the collection lock here, so no other migrate can change maxVersion - // or CollectionMetadata state - migrateFromStatus.setInCriticalSection( true ); - ChunkVersion myVersion = origCollVersion; - myVersion.incMajor(); + if (!ok || res["state"].String() == "fail") { + warning() << "moveChunk error transferring data caused migration abort: " << res + << migrateLog; + errmsg = "data transfer error"; + result.append("cause", res); + return false; + } - { - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock lk(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); - Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); - verify( myVersion > shardingState.getVersion( ns ) ); + if (res["state"].String() == "steady") + break; - // bump the metadata's version up and "forget" about the chunk being moved - // this is not the commit point but in practice the state in this shard won't - // until the commit it done - shardingState.donateChunk(txn, ns, min, max, myVersion); + if (migrateFromStatus.mbUsed() > (500 * 1024 * 1024)) { + // This is too much memory for us to use for this so we're going to abort + // the migrate + ScopedDbConnection conn(toShardCS); + + BSONObj res; + if (!conn->runCommand("admin", BSON("_recvChunkAbort" << 1), res)) { + warning() << "Error encountered while trying to abort migration on " + << "destination shard" << toShardCS; } - log() << "moveChunk setting version to: " << myVersion << migrateLog; + res = res.getOwned(); + conn.done(); + error() << "aborting migrate because too much memory used res: " << res + << migrateLog; + errmsg = "aborting migrate because too much memory used"; + result.appendBool("split", true); + return false; + } - // 5.b - // we're under the collection lock here, too, so we can undo the chunk donation because no other state change - // could be ongoing + txn->checkForInterrupt(); + } - BSONObj res; - bool ok; + timing.done(4); + MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep4); - try { - ScopedDbConnection connTo(toShardCS, 35.0); - ok = connTo->runCommand( "admin", BSON( "_recvChunkCommit" << 1 ), res ); - connTo.done(); - } - catch ( DBException& e ) { - errmsg = str::stream() << "moveChunk could not contact to: shard " - << toShardCS.toString() - << " to commit transfer" << causedBy(e); - warning() << errmsg; - ok = false; - } + // 5. - if ( !ok || MONGO_FAIL_POINT(failMigrationCommit) ) { - log() << "moveChunk migrate commit not accepted by TO-shard: " << res - << " resetting shard version to: " << origShardVersion << migrateLog; - { - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); - Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); + // Before we get into the critical section of the migration, let's double check + // that the docs have been cloned, the config servers are reachable, + // and the lock is in place. + log() << "About to check if it is safe to enter critical section"; - log() << "moveChunk collection lock acquired to reset shard version from " - "failed migration" - ; + // Ensure all cloned docs have actually been transferred + std::size_t locsRemaining = migrateFromStatus.cloneLocsRemaining(); + if (locsRemaining != 0) { + errmsg = str::stream() << "moveChunk cannot enter critical section before all data is" + << " cloned, " << locsRemaining << " locs were not transferred" + << " but to-shard reported " << res; - // revert the chunk manager back to the state before "forgetting" about the - // chunk - shardingState.undoDonateChunk(txn, ns, origCollMetadata); - } - log() << "Shard version successfully reset to clean up failed migration" - ; + // Should never happen, but safe to abort before critical section + error() << errmsg << migrateLog; + dassert(false); + return false; + } - errmsg = "_recvChunkCommit failed!"; - result.append( "cause", res ); - return false; - } + // Ensure distributed lock still held + Status lockStatus = scopedDistLock.getValue().checkStatus(); + if (!lockStatus.isOK()) { + errmsg = str::stream() << "not entering migrate critical section because " + << lockStatus.toString(); + warning() << errmsg; + return false; + } - log() << "moveChunk migrate commit accepted by TO-shard: " << res << migrateLog; + log() << "About to enter migrate critical section"; - // 5.c + { + // 5.a + // we're under the collection lock here, so no other migrate can change maxVersion + // or CollectionMetadata state + migrateFromStatus.setInCriticalSection(true); + ChunkVersion myVersion = origCollVersion; + myVersion.incMajor(); - // version at which the next highest lastmod will be set - // if the chunk being moved is the last in the shard, nextVersion is that chunk's lastmod - // otherwise the highest version is from the chunk being bumped on the FROM-shard - ChunkVersion nextVersion; + { + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock lk(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); + Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); + verify(myVersion > shardingState.getVersion(ns)); - // we want to go only once to the configDB but perhaps change two chunks, the one being migrated and another - // local one (so to bump version for the entire shard) - // we use the 'applyOps' mechanism to group the two updates and make them safer - // TODO pull config update code to a module + // bump the metadata's version up and "forget" about the chunk being moved + // this is not the commit point but in practice the state in this shard won't + // until the commit it done + shardingState.donateChunk(txn, ns, min, max, myVersion); + } - BSONArrayBuilder updates; - { - // update for the chunk being moved - BSONObjBuilder op; - op.append( "op" , "u" ); - op.appendBool( "b" , false /* no upserting */ ); - op.append( "ns" , ChunkType::ConfigNS ); - - BSONObjBuilder n( op.subobjStart( "o" ) ); - n.append(ChunkType::name(), Chunk::genID(ns, min)); - myVersion.addToBSON(n, ChunkType::DEPRECATED_lastmod()); - n.append(ChunkType::ns(), ns); - n.append(ChunkType::min(), min); - n.append(ChunkType::max(), max); - n.append(ChunkType::shard(), toShardName); - n.done(); - - BSONObjBuilder q( op.subobjStart( "o2" ) ); - q.append(ChunkType::name(), Chunk::genID(ns, min)); - q.done(); - - updates.append( op.obj() ); - } + log() << "moveChunk setting version to: " << myVersion << migrateLog; - nextVersion = myVersion; + // 5.b + // we're under the collection lock here, too, so we can undo the chunk donation because no other state change + // could be ongoing - // if we have chunks left on the FROM shard, update the version of one of them as - // well. we can figure that out by grabbing the metadata installed on 5.a + BSONObj res; + bool ok; - const CollectionMetadataPtr bumpedCollMetadata( shardingState.getCollectionMetadata( ns ) ); - if( bumpedCollMetadata->getNumChunks() > 0 ) { + try { + ScopedDbConnection connTo(toShardCS, 35.0); + ok = connTo->runCommand("admin", BSON("_recvChunkCommit" << 1), res); + connTo.done(); + } catch (DBException& e) { + errmsg = str::stream() << "moveChunk could not contact to: shard " + << toShardCS.toString() << " to commit transfer" + << causedBy(e); + warning() << errmsg; + ok = false; + } - // get another chunk on that shard - ChunkType bumpChunk; - bool chunkRes = - bumpedCollMetadata->getNextChunk(bumpedCollMetadata->getMinKey(), - &bumpChunk); - BSONObj bumpMin = bumpChunk.getMin(); - BSONObj bumpMax = bumpChunk.getMax(); + if (!ok || MONGO_FAIL_POINT(failMigrationCommit)) { + log() << "moveChunk migrate commit not accepted by TO-shard: " << res + << " resetting shard version to: " << origShardVersion << migrateLog; + { + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); + Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); - (void)chunkRes; // for compile warning on non-debug - dassert(chunkRes); - dassert( bumpMin.woCompare( min ) != 0 ); + log() << "moveChunk collection lock acquired to reset shard version from " + "failed migration"; - BSONObjBuilder op; - op.append( "op" , "u" ); - op.appendBool( "b" , false ); - op.append( "ns" , ChunkType::ConfigNS ); + // revert the chunk manager back to the state before "forgetting" about the + // chunk + shardingState.undoDonateChunk(txn, ns, origCollMetadata); + } + log() << "Shard version successfully reset to clean up failed migration"; - nextVersion.incMinor(); // same as used on donateChunk - BSONObjBuilder n( op.subobjStart( "o" ) ); - n.append(ChunkType::name(), Chunk::genID(ns, bumpMin)); - nextVersion.addToBSON(n, ChunkType::DEPRECATED_lastmod()); - n.append(ChunkType::ns(), ns); - n.append(ChunkType::min(), bumpMin); - n.append(ChunkType::max(), bumpMax); - n.append(ChunkType::shard(), fromShardName); - n.done(); + errmsg = "_recvChunkCommit failed!"; + result.append("cause", res); + return false; + } - BSONObjBuilder q( op.subobjStart( "o2" ) ); - q.append(ChunkType::name(), Chunk::genID(ns, bumpMin)); - q.done(); + log() << "moveChunk migrate commit accepted by TO-shard: " << res << migrateLog; - updates.append( op.obj() ); + // 5.c - log() << "moveChunk updating self version to: " << nextVersion << " through " - << bumpMin << " -> " << bumpMax << " for collection '" << ns << "'" << migrateLog; + // version at which the next highest lastmod will be set + // if the chunk being moved is the last in the shard, nextVersion is that chunk's lastmod + // otherwise the highest version is from the chunk being bumped on the FROM-shard + ChunkVersion nextVersion; - } - else { + // we want to go only once to the configDB but perhaps change two chunks, the one being migrated and another + // local one (so to bump version for the entire shard) + // we use the 'applyOps' mechanism to group the two updates and make them safer + // TODO pull config update code to a module - log() << "moveChunk moved last chunk out for collection '" << ns << "'" << migrateLog; - } + BSONArrayBuilder updates; + { + // update for the chunk being moved + BSONObjBuilder op; + op.append("op", "u"); + op.appendBool("b", false /* no upserting */); + op.append("ns", ChunkType::ConfigNS); + + BSONObjBuilder n(op.subobjStart("o")); + n.append(ChunkType::name(), Chunk::genID(ns, min)); + myVersion.addToBSON(n, ChunkType::DEPRECATED_lastmod()); + n.append(ChunkType::ns(), ns); + n.append(ChunkType::min(), min); + n.append(ChunkType::max(), max); + n.append(ChunkType::shard(), toShardName); + n.done(); + + BSONObjBuilder q(op.subobjStart("o2")); + q.append(ChunkType::name(), Chunk::genID(ns, min)); + q.done(); + + updates.append(op.obj()); + } + + nextVersion = myVersion; + + // if we have chunks left on the FROM shard, update the version of one of them as + // well. we can figure that out by grabbing the metadata installed on 5.a + + const CollectionMetadataPtr bumpedCollMetadata(shardingState.getCollectionMetadata(ns)); + if (bumpedCollMetadata->getNumChunks() > 0) { + // get another chunk on that shard + ChunkType bumpChunk; + bool chunkRes = + bumpedCollMetadata->getNextChunk(bumpedCollMetadata->getMinKey(), &bumpChunk); + BSONObj bumpMin = bumpChunk.getMin(); + BSONObj bumpMax = bumpChunk.getMax(); + + (void)chunkRes; // for compile warning on non-debug + dassert(chunkRes); + dassert(bumpMin.woCompare(min) != 0); + + BSONObjBuilder op; + op.append("op", "u"); + op.appendBool("b", false); + op.append("ns", ChunkType::ConfigNS); + + nextVersion.incMinor(); // same as used on donateChunk + BSONObjBuilder n(op.subobjStart("o")); + n.append(ChunkType::name(), Chunk::genID(ns, bumpMin)); + nextVersion.addToBSON(n, ChunkType::DEPRECATED_lastmod()); + n.append(ChunkType::ns(), ns); + n.append(ChunkType::min(), bumpMin); + n.append(ChunkType::max(), bumpMax); + n.append(ChunkType::shard(), fromShardName); + n.done(); + + BSONObjBuilder q(op.subobjStart("o2")); + q.append(ChunkType::name(), Chunk::genID(ns, bumpMin)); + q.done(); + + updates.append(op.obj()); + + log() << "moveChunk updating self version to: " << nextVersion << " through " + << bumpMin << " -> " << bumpMax << " for collection '" << ns << "'" + << migrateLog; + + } else { + log() << "moveChunk moved last chunk out for collection '" << ns << "'" + << migrateLog; + } - BSONArrayBuilder preCond; + BSONArrayBuilder preCond; + { + BSONObjBuilder b; + b.append("ns", ChunkType::ConfigNS); + b.append("q", + BSON("query" << BSON(ChunkType::ns(ns)) << "orderby" + << BSON(ChunkType::DEPRECATED_lastmod() << -1))); { - BSONObjBuilder b; - b.append("ns", ChunkType::ConfigNS); - b.append("q", BSON("query" << BSON(ChunkType::ns(ns)) << - "orderby" << BSON(ChunkType::DEPRECATED_lastmod() << -1))); - { - BSONObjBuilder bb( b.subobjStart( "res" ) ); - // TODO: For backwards compatibility, we can't yet require an epoch here - bb.appendTimestamp(ChunkType::DEPRECATED_lastmod(), origCollVersion.toLong()); - bb.done(); - } - preCond.append( b.obj() ); + BSONObjBuilder bb(b.subobjStart("res")); + // TODO: For backwards compatibility, we can't yet require an epoch here + bb.appendTimestamp(ChunkType::DEPRECATED_lastmod(), origCollVersion.toLong()); + bb.done(); } + preCond.append(b.obj()); + } - Status applyOpsStatus{Status::OK()}; - try { - - // For testing migration failures - if ( MONGO_FAIL_POINT(failMigrationConfigWritePrepare) ) { - throw DBException( "mock migration failure before config write", - PrepareConfigsFailedCode ); - } + Status applyOpsStatus{Status::OK()}; + try { + // For testing migration failures + if (MONGO_FAIL_POINT(failMigrationConfigWritePrepare)) { + throw DBException("mock migration failure before config write", + PrepareConfigsFailedCode); + } - applyOpsStatus = grid.catalogManager()->applyChunkOpsDeprecated(updates.arr(), - preCond.arr()); + applyOpsStatus = + grid.catalogManager()->applyChunkOpsDeprecated(updates.arr(), preCond.arr()); - if (MONGO_FAIL_POINT(failMigrationApplyOps)) { - throw SocketException(SocketException::RECV_ERROR, - shardingState.getConfigServer()); - } - } - catch (const DBException& ex) { - warning() << ex << migrateLog; - applyOpsStatus = ex.toStatus(); + if (MONGO_FAIL_POINT(failMigrationApplyOps)) { + throw SocketException(SocketException::RECV_ERROR, + shardingState.getConfigServer()); } + } catch (const DBException& ex) { + warning() << ex << migrateLog; + applyOpsStatus = ex.toStatus(); + } - if (applyOpsStatus == ErrorCodes::PrepareConfigsFailedCode) { - - // In the process of issuing the migrate commit, the SyncClusterConnection - // checks that the config servers are reachable. If they are not, we are - // sure that the applyOps command was not sent to any of the configs, so we - // can safely back out of the migration here, by resetting the shard - // version that we bumped up to in the donateChunk() call above. + if (applyOpsStatus == ErrorCodes::PrepareConfigsFailedCode) { + // In the process of issuing the migrate commit, the SyncClusterConnection + // checks that the config servers are reachable. If they are not, we are + // sure that the applyOps command was not sent to any of the configs, so we + // can safely back out of the migration here, by resetting the shard + // version that we bumped up to in the donateChunk() call above. - log() << "About to acquire moveChunk coll lock to reset shard version from " - << "failed migration"; + log() << "About to acquire moveChunk coll lock to reset shard version from " + << "failed migration"; - { - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); - Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); + { + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); + Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); - // Revert the metadata back to the state before "forgetting" - // about the chunk. - shardingState.undoDonateChunk(txn, ns, origCollMetadata); - } + // Revert the metadata back to the state before "forgetting" + // about the chunk. + shardingState.undoDonateChunk(txn, ns, origCollMetadata); + } - log() << "Shard version successfully reset to clean up failed migration"; + log() << "Shard version successfully reset to clean up failed migration"; - errmsg = "Failed to send migrate commit to configs because " + errmsg; - return false; + errmsg = "Failed to send migrate commit to configs because " + errmsg; + return false; - } - else if (!applyOpsStatus.isOK()) { - - // this could be a blip in the connectivity - // wait out a few seconds and check if the commit request made it - // - // if the commit made it to the config, we'll see the chunk in the new shard and there's no action - // if the commit did not make it, currently the only way to fix this state is to bounce the mongod so - // that the old state (before migrating) be brought in - - warning() << "moveChunk commit outcome ongoing" << migrateLog; - sleepsecs( 10 ); - - // look for the chunk in this shard whose version got bumped - // we assume that if that mod made it to the config, the applyOps was successful - try { - std::vector<ChunkType> newestChunk; - Status status = grid.catalogManager()->getChunks( - Query(BSON(ChunkType::ns(ns))) - .sort(ChunkType::DEPRECATED_lastmod(), -1), - 1, - &newestChunk); - uassertStatusOK(status); - - ChunkVersion checkVersion; - if (!newestChunk.empty()) { - invariant(newestChunk.size() == 1); - checkVersion = newestChunk[0].getVersion(); - } + } else if (!applyOpsStatus.isOK()) { + // this could be a blip in the connectivity + // wait out a few seconds and check if the commit request made it + // + // if the commit made it to the config, we'll see the chunk in the new shard and there's no action + // if the commit did not make it, currently the only way to fix this state is to bounce the mongod so + // that the old state (before migrating) be brought in - if (checkVersion.equals(nextVersion)) { - log() << "moveChunk commit confirmed" << migrateLog; - errmsg.clear(); + warning() << "moveChunk commit outcome ongoing" << migrateLog; + sleepsecs(10); - } - else { - error() << "moveChunk commit failed: version is at " - << checkVersion << " instead of " << nextVersion << migrateLog; - error() << "TERMINATING" << migrateLog; - dbexit( EXIT_SHARDING_ERROR ); - } + // look for the chunk in this shard whose version got bumped + // we assume that if that mod made it to the config, the applyOps was successful + try { + std::vector<ChunkType> newestChunk; + Status status = grid.catalogManager()->getChunks( + Query(BSON(ChunkType::ns(ns))).sort(ChunkType::DEPRECATED_lastmod(), -1), + 1, + &newestChunk); + uassertStatusOK(status); + + ChunkVersion checkVersion; + if (!newestChunk.empty()) { + invariant(newestChunk.size() == 1); + checkVersion = newestChunk[0].getVersion(); } - catch ( ... ) { - error() << "moveChunk failed to get confirmation of commit" << migrateLog; + + if (checkVersion.equals(nextVersion)) { + log() << "moveChunk commit confirmed" << migrateLog; + errmsg.clear(); + + } else { + error() << "moveChunk commit failed: version is at " << checkVersion + << " instead of " << nextVersion << migrateLog; error() << "TERMINATING" << migrateLog; - dbexit( EXIT_SHARDING_ERROR ); + dbexit(EXIT_SHARDING_ERROR); } + } catch (...) { + error() << "moveChunk failed to get confirmation of commit" << migrateLog; + error() << "TERMINATING" << migrateLog; + dbexit(EXIT_SHARDING_ERROR); } + } - migrateFromStatus.setInCriticalSection( false ); - - // 5.d - BSONObjBuilder commitInfo; - commitInfo.appendElements( chunkInfo ); - if (res["counts"].type() == Object) { - commitInfo.appendElements(res["counts"].Obj()); - } + migrateFromStatus.setInCriticalSection(false); - grid.catalogManager()->logChange(txn, "moveChunk.commit", ns, commitInfo.obj()); + // 5.d + BSONObjBuilder commitInfo; + commitInfo.appendElements(chunkInfo); + if (res["counts"].type() == Object) { + commitInfo.appendElements(res["counts"].Obj()); } - migrateFromStatus.done(txn); - timing.done(5); - MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep5); - - // 6. - // NOTE: It is important that the distributed collection lock be held for this step. - RangeDeleter* deleter = getDeleter(); - RangeDeleterOptions deleterOptions(KeyRange(ns, - min.getOwned(), - max.getOwned(), - shardKeyPattern)); - deleterOptions.writeConcern = writeConcern; - deleterOptions.waitForOpenCursors = true; - deleterOptions.fromMigrate = true; - deleterOptions.onlyRemoveOrphanedDocs = true; - deleterOptions.removeSaverReason = "post-cleanup"; - - if (waitForDelete) { - log() << "doing delete inline for cleanup of chunk data" << migrateLog; - - string errMsg; - // This is an immediate delete, and as a consequence, there could be more - // deletes happening simultaneously than there are deleter worker threads. - if (!deleter->deleteNow(txn, - deleterOptions, - &errMsg)) { - log() << "Error occured while performing cleanup: " << errMsg; - } - } - else { - log() << "forking for cleanup of chunk data" << migrateLog; + grid.catalogManager()->logChange(txn, "moveChunk.commit", ns, commitInfo.obj()); + } - string errMsg; - if (!deleter->queueDelete(txn, - deleterOptions, - NULL, // Don't want to be notified. - &errMsg)) { - log() << "could not queue migration cleanup: " << errMsg; - } + migrateFromStatus.done(txn); + timing.done(5); + MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep5); + + // 6. + // NOTE: It is important that the distributed collection lock be held for this step. + RangeDeleter* deleter = getDeleter(); + RangeDeleterOptions deleterOptions( + KeyRange(ns, min.getOwned(), max.getOwned(), shardKeyPattern)); + deleterOptions.writeConcern = writeConcern; + deleterOptions.waitForOpenCursors = true; + deleterOptions.fromMigrate = true; + deleterOptions.onlyRemoveOrphanedDocs = true; + deleterOptions.removeSaverReason = "post-cleanup"; + + if (waitForDelete) { + log() << "doing delete inline for cleanup of chunk data" << migrateLog; + + string errMsg; + // This is an immediate delete, and as a consequence, there could be more + // deletes happening simultaneously than there are deleter worker threads. + if (!deleter->deleteNow(txn, deleterOptions, &errMsg)) { + log() << "Error occured while performing cleanup: " << errMsg; + } + } else { + log() << "forking for cleanup of chunk data" << migrateLog; + + string errMsg; + if (!deleter->queueDelete(txn, + deleterOptions, + NULL, // Don't want to be notified. + &errMsg)) { + log() << "could not queue migration cleanup: " << errMsg; } - timing.done(6); - MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep6); + } + timing.done(6); + MONGO_FP_PAUSE_WHILE(moveChunkHangAtStep6); - return true; + return true; + } - } +} moveChunkCmd; + +bool ShardingState::inCriticalMigrateSection() { + return migrateFromStatus.getInCriticalSection(); +} - } moveChunkCmd; +bool ShardingState::waitTillNotInCriticalSection(int maxSecondsToWait) { + return migrateFromStatus.waitTillNotInCriticalSection(maxSecondsToWait); +} - bool ShardingState::inCriticalMigrateSection() { - return migrateFromStatus.getInCriticalSection(); +/* ----- + below this are the "to" side commands + + command to initiate + worker thread + does initial clone + pulls initial change set + keeps pulling + keeps state + command to get state + commend to "commit" +*/ + +// Enabling / disabling these fail points pauses / resumes MigrateStatus::_go(), the thread +// that receives a chunk migration from the donor. +MONGO_FP_DECLARE(migrateThreadHangAtStep1); +MONGO_FP_DECLARE(migrateThreadHangAtStep2); +MONGO_FP_DECLARE(migrateThreadHangAtStep3); +MONGO_FP_DECLARE(migrateThreadHangAtStep4); +MONGO_FP_DECLARE(migrateThreadHangAtStep5); + +class MigrateStatus { +public: + enum State { READY, CLONE, CATCHUP, STEADY, COMMIT_START, DONE, FAIL, ABORT }; + + MigrateStatus() + : _active(false), + _numCloned(0), + _clonedBytes(0), + _numCatchup(0), + _numSteady(0), + _state(READY) {} + + void setState(State newState) { + std::lock_guard<std::mutex> sl(_mutex); + _state = newState; } - bool ShardingState::waitTillNotInCriticalSection( int maxSecondsToWait ) { - return migrateFromStatus.waitTillNotInCriticalSection( maxSecondsToWait ); + State getState() const { + std::lock_guard<std::mutex> sl(_mutex); + return _state; } - /* ----- - below this are the "to" side commands + /** + * Returns OK if preparation was successful. + */ + Status prepare(const std::string& ns, + const std::string& fromShard, + const BSONObj& min, + const BSONObj& max, + const BSONObj& shardKeyPattern) { + std::lock_guard<std::mutex> lk(_mutex); + + if (_active) { + return Status(ErrorCodes::ConflictingOperationInProgress, + str::stream() << "Active migration already in progress " + << "ns: " << _ns << ", from: " << _from << ", min: " << _min + << ", max: " << _max); + } - command to initiate - worker thread - does initial clone - pulls initial change set - keeps pulling - keeps state - command to get state - commend to "commit" - */ + _state = READY; + _errmsg = ""; - // Enabling / disabling these fail points pauses / resumes MigrateStatus::_go(), the thread - // that receives a chunk migration from the donor. - MONGO_FP_DECLARE(migrateThreadHangAtStep1); - MONGO_FP_DECLARE(migrateThreadHangAtStep2); - MONGO_FP_DECLARE(migrateThreadHangAtStep3); - MONGO_FP_DECLARE(migrateThreadHangAtStep4); - MONGO_FP_DECLARE(migrateThreadHangAtStep5); + _ns = ns; + _from = fromShard; + _min = min; + _max = max; + _shardKeyPattern = shardKeyPattern; - class MigrateStatus { - public: - enum State { - READY, - CLONE, - CATCHUP, - STEADY, - COMMIT_START, - DONE, - FAIL, - ABORT - }; - - MigrateStatus(): - _active(false), - _numCloned(0), - _clonedBytes(0), - _numCatchup(0), - _numSteady(0), - _state(READY) { - } - - void setState(State newState) { - std::lock_guard<std::mutex> sl(_mutex); - _state = newState; - } + _numCloned = 0; + _clonedBytes = 0; + _numCatchup = 0; + _numSteady = 0; - State getState() const { - std::lock_guard<std::mutex> sl(_mutex); - return _state; - } + _active = true; - /** - * Returns OK if preparation was successful. - */ - Status prepare(const std::string& ns, - const std::string& fromShard, - const BSONObj& min, - const BSONObj& max, - const BSONObj& shardKeyPattern) { - std::lock_guard<std::mutex> lk(_mutex); - - if (_active) { - return Status(ErrorCodes::ConflictingOperationInProgress, - str::stream() << "Active migration already in progress " - << "ns: " << _ns - << ", from: " << _from - << ", min: " << _min - << ", max: " << _max); - } - - _state = READY; - _errmsg = ""; - - _ns = ns; - _from = fromShard; - _min = min; - _max = max; - _shardKeyPattern = shardKeyPattern; - - _numCloned = 0; - _clonedBytes = 0; - _numCatchup = 0; - _numSteady = 0; - - _active = true; - - return Status::OK(); - } - - void go(OperationContext* txn, - const std::string& ns, - BSONObj min, - BSONObj max, - BSONObj shardKeyPattern, - const std::string& fromShard, - const OID& epoch, - const WriteConcernOptions& writeConcern) { - try { - _go(txn, ns, min, max, shardKeyPattern, fromShard, epoch, writeConcern); - } - catch ( std::exception& e ) { - { - std::lock_guard<std::mutex> sl(_mutex); - _state = FAIL; - _errmsg = e.what(); - } + return Status::OK(); + } - error() << "migrate failed: " << e.what() << migrateLog; + void go(OperationContext* txn, + const std::string& ns, + BSONObj min, + BSONObj max, + BSONObj shardKeyPattern, + const std::string& fromShard, + const OID& epoch, + const WriteConcernOptions& writeConcern) { + try { + _go(txn, ns, min, max, shardKeyPattern, fromShard, epoch, writeConcern); + } catch (std::exception& e) { + { + std::lock_guard<std::mutex> sl(_mutex); + _state = FAIL; + _errmsg = e.what(); } - catch ( ... ) { - { - std::lock_guard<std::mutex> sl(_mutex); - _state = FAIL; - _errmsg = "UNKNOWN ERROR"; - } - error() << "migrate failed with unknown exception" << migrateLog; + error() << "migrate failed: " << e.what() << migrateLog; + } catch (...) { + { + std::lock_guard<std::mutex> sl(_mutex); + _state = FAIL; + _errmsg = "UNKNOWN ERROR"; } - if ( getState() != DONE ) { - // Unprotect the range if needed/possible on unsuccessful TO migration - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); - Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); + error() << "migrate failed with unknown exception" << migrateLog; + } - string errMsg; - if (!shardingState.forgetPending(txn, ns, min, max, epoch, &errMsg)) { - warning() << errMsg; - } - } + if (getState() != DONE) { + // Unprotect the range if needed/possible on unsuccessful TO migration + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); + Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); - setActive( false ); + string errMsg; + if (!shardingState.forgetPending(txn, ns, min, max, epoch, &errMsg)) { + warning() << errMsg; + } } - void _go(OperationContext* txn, - const std::string& ns, - BSONObj min, - BSONObj max, - BSONObj shardKeyPattern, - const std::string& fromShard, - const OID& epoch, - const WriteConcernOptions& writeConcern) { - verify( getActive() ); - verify( getState() == READY ); - verify( ! min.isEmpty() ); - verify( ! max.isEmpty() ); + setActive(false); + } - DisableDocumentValidation validationDisabler(txn); + void _go(OperationContext* txn, + const std::string& ns, + BSONObj min, + BSONObj max, + BSONObj shardKeyPattern, + const std::string& fromShard, + const OID& epoch, + const WriteConcernOptions& writeConcern) { + verify(getActive()); + verify(getState() == READY); + verify(!min.isEmpty()); + verify(!max.isEmpty()); + + DisableDocumentValidation validationDisabler(txn); + + log() << "starting receiving-end of migration of chunk " << min << " -> " << max + << " for collection " << ns << " from " << fromShard << " at epoch " + << epoch.toString(); + + string errmsg; + MoveTimingHelper timing(txn, "to", ns, min, max, 5 /* steps */, &errmsg, "", ""); + + ScopedDbConnection conn(fromShard); + conn->getLastError(); // just test connection + + NamespaceString nss(ns); + { + // 0. copy system.namespaces entry if collection doesn't already exist + OldClientWriteContext ctx(txn, ns); + if (!repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nss)) { + errmsg = str::stream() << "Not primary during migration: " << ns + << ": checking if collection exists"; + warning() << errmsg; + setState(FAIL); + return; + } - log() << "starting receiving-end of migration of chunk " << min << " -> " << max << - " for collection " << ns << " from " << fromShard - << " at epoch " << epoch.toString(); + // Only copy if ns doesn't already exist + Database* db = ctx.db(); + Collection* collection = db->getCollection(ns); - string errmsg; - MoveTimingHelper timing(txn, "to", ns, min, max, 5 /* steps */, &errmsg, "", ""); + if (!collection) { + list<BSONObj> infos = conn->getCollectionInfos( + nsToDatabase(ns), BSON("name" << nsToCollectionSubstring(ns))); - ScopedDbConnection conn(fromShard); - conn->getLastError(); // just test connection + BSONObj options; + if (infos.size() > 0) { + BSONObj entry = infos.front(); + if (entry["options"].isABSONObj()) { + options = entry["options"].Obj(); + } + } - NamespaceString nss(ns); - { - // 0. copy system.namespaces entry if collection doesn't already exist - OldClientWriteContext ctx(txn, ns ); - if (!repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nss)) { - errmsg = str::stream() << "Not primary during migration: " << ns - << ": checking if collection exists"; - warning() << errmsg; - setState(FAIL); - return; + WriteUnitOfWork wuow(txn); + Status status = userCreateNS(txn, db, ns, options, false); + if (!status.isOK()) { + warning() << "failed to create collection [" << ns << "] " + << " with options " << options << ": " << status; } + wuow.commit(); + } + } - // Only copy if ns doesn't already exist - Database* db = ctx.db(); - Collection* collection = db->getCollection( ns ); + { + // 1. copy indexes - if ( !collection ) { - list<BSONObj> infos = - conn->getCollectionInfos(nsToDatabase(ns), - BSON("name" << nsToCollectionSubstring(ns))); + vector<BSONObj> indexSpecs; + { + const std::list<BSONObj> indexes = conn->getIndexSpecs(ns); + indexSpecs.insert(indexSpecs.begin(), indexes.begin(), indexes.end()); + } - BSONObj options; - if (infos.size() > 0) { - BSONObj entry = infos.front(); - if (entry["options"].isABSONObj()) { - options = entry["options"].Obj(); - } - } + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock lk(txn->lockState(), nsToDatabaseSubstring(ns), MODE_X); + OldClientContext ctx(txn, ns); + if (!repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nss)) { + errmsg = str::stream() << "Not primary during migration: " << ns; + warning() << errmsg; + setState(FAIL); + return; + } - WriteUnitOfWork wuow(txn); - Status status = userCreateNS(txn, db, ns, options, false); - if ( !status.isOK() ) { - warning() << "failed to create collection [" << ns << "] " - << " with options " << options << ": " << status; - } - wuow.commit(); - } + Database* db = ctx.db(); + Collection* collection = db->getCollection(ns); + if (!collection) { + errmsg = str::stream() << "collection dropped during migration: " << ns; + warning() << errmsg; + setState(FAIL); + return; } - { - // 1. copy indexes + MultiIndexBlock indexer(txn, collection); - vector<BSONObj> indexSpecs; - { - const std::list<BSONObj> indexes = conn->getIndexSpecs(ns); - indexSpecs.insert(indexSpecs.begin(), indexes.begin(), indexes.end()); - } + indexer.removeExistingIndexes(&indexSpecs); - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock lk(txn->lockState(), nsToDatabaseSubstring(ns), MODE_X); - OldClientContext ctx(txn, ns); - if (!repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nss)) { - errmsg = str::stream() << "Not primary during migration: " << ns; + if (!indexSpecs.empty()) { + // Only copy indexes if the collection does not have any documents. + if (collection->numRecords(txn) > 0) { + errmsg = str::stream() << "aborting migration, shard is missing " + << indexSpecs.size() << " indexes and " + << "collection is not empty. Non-trivial " + << "index creation should be scheduled manually"; warning() << errmsg; setState(FAIL); return; } - Database* db = ctx.db(); - Collection* collection = db->getCollection( ns ); - if ( !collection ) { - errmsg = str::stream() << "collection dropped during migration: " << ns; + Status status = indexer.init(indexSpecs); + if (!status.isOK()) { + errmsg = str::stream() << "failed to create index before migrating data. " + << " error: " << status.toString(); warning() << errmsg; setState(FAIL); return; } - MultiIndexBlock indexer(txn, collection); + status = indexer.insertAllDocumentsInCollection(); + if (!status.isOK()) { + errmsg = str::stream() << "failed to create index before migrating data. " + << " error: " << status.toString(); + warning() << errmsg; + setState(FAIL); + return; + } - indexer.removeExistingIndexes(&indexSpecs); + WriteUnitOfWork wunit(txn); + indexer.commit(); - if (!indexSpecs.empty()) { - // Only copy indexes if the collection does not have any documents. - if (collection->numRecords(txn) > 0) { - errmsg = str::stream() << "aborting migration, shard is missing " - << indexSpecs.size() << " indexes and " - << "collection is not empty. Non-trivial " - << "index creation should be scheduled manually"; - warning() << errmsg; - setState(FAIL); - return; - } + for (size_t i = 0; i < indexSpecs.size(); i++) { + // make sure to create index on secondaries as well + getGlobalServiceContext()->getOpObserver()->onCreateIndex( + txn, db->getSystemIndexesName(), indexSpecs[i], true /* fromMigrate */); + } - Status status = indexer.init(indexSpecs); - if ( !status.isOK() ) { - errmsg = str::stream() << "failed to create index before migrating data. " - << " error: " << status.toString(); - warning() << errmsg; - setState(FAIL); - return; - } + wunit.commit(); + } - status = indexer.insertAllDocumentsInCollection(); - if ( !status.isOK() ) { - errmsg = str::stream() << "failed to create index before migrating data. " - << " error: " << status.toString(); - warning() << errmsg; - setState(FAIL); - return; - } + timing.done(1); + MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep1); + } - WriteUnitOfWork wunit(txn); - indexer.commit(); + { + // 2. delete any data already in range + RangeDeleterOptions deleterOptions( + KeyRange(ns, min.getOwned(), max.getOwned(), shardKeyPattern)); + deleterOptions.writeConcern = writeConcern; + // No need to wait since all existing cursors will filter out this range when + // returning the results. + deleterOptions.waitForOpenCursors = false; + deleterOptions.fromMigrate = true; + deleterOptions.onlyRemoveOrphanedDocs = true; + deleterOptions.removeSaverReason = "preCleanup"; - for (size_t i = 0; i < indexSpecs.size(); i++) { - // make sure to create index on secondaries as well - getGlobalServiceContext()->getOpObserver()->onCreateIndex( - txn, - db->getSystemIndexesName(), - indexSpecs[i], - true /* fromMigrate */); - } + string errMsg; - wunit.commit(); - } - - timing.done(1); - MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep1); + if (!getDeleter()->deleteNow(txn, deleterOptions, &errMsg)) { + warning() << "Failed to queue delete for migrate abort: " << errMsg; + setState(FAIL); + return; } { - // 2. delete any data already in range - RangeDeleterOptions deleterOptions(KeyRange(ns, - min.getOwned(), - max.getOwned(), - shardKeyPattern)); - deleterOptions.writeConcern = writeConcern; - // No need to wait since all existing cursors will filter out this range when - // returning the results. - deleterOptions.waitForOpenCursors = false; - deleterOptions.fromMigrate = true; - deleterOptions.onlyRemoveOrphanedDocs = true; - deleterOptions.removeSaverReason = "preCleanup"; - - string errMsg; - - if (!getDeleter()->deleteNow(txn, deleterOptions, &errMsg)) { - warning() << "Failed to queue delete for migrate abort: " << errMsg; + // Protect the range by noting that we're now starting a migration to it + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); + Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); + + if (!shardingState.notePending(txn, ns, min, max, epoch, &errmsg)) { + warning() << errmsg; setState(FAIL); return; } + } - { - // Protect the range by noting that we're now starting a migration to it - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); - Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); + timing.done(2); + MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep2); + } - if (!shardingState.notePending(txn, ns, min, max, epoch, &errmsg)) { - warning() << errmsg; - setState(FAIL); - return; - } - } + State currentState = getState(); + if (currentState == FAIL || currentState == ABORT) { + string errMsg; + RangeDeleterOptions deleterOptions( + KeyRange(ns, min.getOwned(), max.getOwned(), shardKeyPattern)); + deleterOptions.writeConcern = writeConcern; + // No need to wait since all existing cursors will filter out this range when + // returning the results. + deleterOptions.waitForOpenCursors = false; + deleterOptions.fromMigrate = true; + deleterOptions.onlyRemoveOrphanedDocs = true; - timing.done(2); - MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep2); - } - - State currentState = getState(); - if (currentState == FAIL || currentState == ABORT) { - string errMsg; - RangeDeleterOptions deleterOptions(KeyRange(ns, - min.getOwned(), - max.getOwned(), - shardKeyPattern)); - deleterOptions.writeConcern = writeConcern; - // No need to wait since all existing cursors will filter out this range when - // returning the results. - deleterOptions.waitForOpenCursors = false; - deleterOptions.fromMigrate = true; - deleterOptions.onlyRemoveOrphanedDocs = true; - - if (!getDeleter()->queueDelete(txn, deleterOptions, NULL /* notifier */, &errMsg)) { - warning() << "Failed to queue delete for migrate abort: " << errMsg; - } + if (!getDeleter()->queueDelete(txn, deleterOptions, NULL /* notifier */, &errMsg)) { + warning() << "Failed to queue delete for migrate abort: " << errMsg; } + } - { - // 3. initial bulk clone - setState(CLONE); - - while ( true ) { - BSONObj res; - if ( ! conn->runCommand( "admin" , BSON( "_migrateClone" << 1 ) , res ) ) { // gets array of objects to copy, in disk order - setState(FAIL); - errmsg = "_migrateClone failed: "; - errmsg += res.toString(); + { + // 3. initial bulk clone + setState(CLONE); + + while (true) { + BSONObj res; + if (!conn->runCommand("admin", + BSON("_migrateClone" << 1), + res)) { // gets array of objects to copy, in disk order + setState(FAIL); + errmsg = "_migrateClone failed: "; + errmsg += res.toString(); + error() << errmsg << migrateLog; + conn.done(); + return; + } + + BSONObj arr = res["objects"].Obj(); + int thisTime = 0; + + BSONObjIterator i(arr); + while (i.more()) { + txn->checkForInterrupt(); + + if (getState() == ABORT) { + errmsg = str::stream() << "Migration abort requested while " + << "copying documents"; error() << errmsg << migrateLog; - conn.done(); return; } - BSONObj arr = res["objects"].Obj(); - int thisTime = 0; - - BSONObjIterator i( arr ); - while( i.more() ) { - txn->checkForInterrupt(); - - if ( getState() == ABORT ) { - errmsg = str::stream() << "Migration abort requested while " - << "copying documents"; - error() << errmsg << migrateLog; - return; + BSONObj docToClone = i.next().Obj(); + { + OldClientWriteContext cx(txn, ns); + + BSONObj localDoc; + if (willOverrideLocalId(txn, + ns, + min, + max, + shardKeyPattern, + cx.db(), + docToClone, + &localDoc)) { + string errMsg = str::stream() << "cannot migrate chunk, local document " + << localDoc << " has same _id as cloned " + << "remote document " << docToClone; + + warning() << errMsg; + + // Exception will abort migration cleanly + uasserted(16976, errMsg); } - BSONObj docToClone = i.next().Obj(); - { - OldClientWriteContext cx(txn, ns ); - - BSONObj localDoc; - if (willOverrideLocalId(txn, - ns, - min, - max, - shardKeyPattern, - cx.db(), - docToClone, - &localDoc)) { - string errMsg = - str::stream() << "cannot migrate chunk, local document " - << localDoc - << " has same _id as cloned " - << "remote document " << docToClone; - - warning() << errMsg; - - // Exception will abort migration cleanly - uasserted( 16976, errMsg ); - } - - Helpers::upsert( txn, ns, docToClone, true ); - } - thisTime++; + Helpers::upsert(txn, ns, docToClone, true); + } + thisTime++; - { - std::lock_guard<std::mutex> statsLock(_mutex); - _numCloned++; - _clonedBytes += docToClone.objsize(); - } + { + std::lock_guard<std::mutex> statsLock(_mutex); + _numCloned++; + _clonedBytes += docToClone.objsize(); + } - if (writeConcern.shouldWaitForOtherNodes()) { - repl::ReplicationCoordinator::StatusAndDuration replStatus = - repl::getGlobalReplicationCoordinator()->awaitReplication( - txn, - repl::ReplClientInfo::forClient( - txn->getClient()).getLastOp(), - writeConcern); - if (replStatus.status.code() == ErrorCodes::WriteConcernFailed) { - warning() << "secondaryThrottle on, but doc insert timed out; " - "continuing"; - } - else { - massertStatusOK(replStatus.status); - } + if (writeConcern.shouldWaitForOtherNodes()) { + repl::ReplicationCoordinator::StatusAndDuration replStatus = + repl::getGlobalReplicationCoordinator()->awaitReplication( + txn, + repl::ReplClientInfo::forClient(txn->getClient()).getLastOp(), + writeConcern); + if (replStatus.status.code() == ErrorCodes::WriteConcernFailed) { + warning() << "secondaryThrottle on, but doc insert timed out; " + "continuing"; + } else { + massertStatusOK(replStatus.status); } } - - if ( thisTime == 0 ) - break; } - timing.done(3); - MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep3); + if (thisTime == 0) + break; } - // If running on a replicated system, we'll need to flush the docs we cloned to the - // secondaries - repl::OpTime lastOpApplied = - repl::ReplClientInfo::forClient(txn->getClient()).getLastOp(); + timing.done(3); + MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep3); + } - { - // 4. do bulk of mods - setState(CATCHUP); - while ( true ) { - BSONObj res; - if ( ! conn->runCommand( "admin" , BSON( "_transferMods" << 1 ) , res ) ) { - setState(FAIL); - errmsg = "_transferMods failed: "; - errmsg += res.toString(); - error() << "_transferMods failed: " << res << migrateLog; - conn.done(); - return; - } - if ( res["size"].number() == 0 ) - break; + // If running on a replicated system, we'll need to flush the docs we cloned to the + // secondaries + repl::OpTime lastOpApplied = repl::ReplClientInfo::forClient(txn->getClient()).getLastOp(); - apply(txn, ns, min, max, shardKeyPattern, res, &lastOpApplied); + { + // 4. do bulk of mods + setState(CATCHUP); + while (true) { + BSONObj res; + if (!conn->runCommand("admin", BSON("_transferMods" << 1), res)) { + setState(FAIL); + errmsg = "_transferMods failed: "; + errmsg += res.toString(); + error() << "_transferMods failed: " << res << migrateLog; + conn.done(); + return; + } + if (res["size"].number() == 0) + break; - const int maxIterations = 3600*50; - int i; - for ( i=0;i<maxIterations; i++) { - txn->checkForInterrupt(); + apply(txn, ns, min, max, shardKeyPattern, res, &lastOpApplied); - if ( getState() == ABORT ) { - errmsg = str::stream() << "Migration abort requested while waiting " - << "for replication at catch up stage"; - error() << errmsg << migrateLog; + const int maxIterations = 3600 * 50; + int i; + for (i = 0; i < maxIterations; i++) { + txn->checkForInterrupt(); - return; - } + if (getState() == ABORT) { + errmsg = str::stream() << "Migration abort requested while waiting " + << "for replication at catch up stage"; + error() << errmsg << migrateLog; - if (opReplicatedEnough(txn, lastOpApplied, writeConcern)) - break; + return; + } - if ( i > 100 ) { - warning() << "secondaries having hard time keeping up with migrate" << migrateLog; - } + if (opReplicatedEnough(txn, lastOpApplied, writeConcern)) + break; - sleepmillis( 20 ); + if (i > 100) { + warning() << "secondaries having hard time keeping up with migrate" + << migrateLog; } - if ( i == maxIterations ) { - errmsg = "secondary can't keep up with migrate"; - error() << errmsg << migrateLog; - conn.done(); - setState(FAIL); - return; - } + sleepmillis(20); } - timing.done(4); - MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep4); + if (i == maxIterations) { + errmsg = "secondary can't keep up with migrate"; + error() << errmsg << migrateLog; + conn.done(); + setState(FAIL); + return; + } } - { - // pause to wait for replication - // this will prevent us from going into critical section until we're ready - Timer t; - while ( t.minutes() < 600 ) { - txn->checkForInterrupt(); + timing.done(4); + MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep4); + } - if (getState() == ABORT) { - errmsg = "Migration abort requested while waiting for replication"; - error() << errmsg << migrateLog; - return; - } + { + // pause to wait for replication + // this will prevent us from going into critical section until we're ready + Timer t; + while (t.minutes() < 600) { + txn->checkForInterrupt(); - log() << "Waiting for replication to catch up before entering critical section" - ; - if (flushPendingWrites(txn, ns, min, max, lastOpApplied, writeConcern)) - break; - sleepsecs(1); + if (getState() == ABORT) { + errmsg = "Migration abort requested while waiting for replication"; + error() << errmsg << migrateLog; + return; } - if (t.minutes() >= 600) { - setState(FAIL); - errmsg = "Cannot go to critical section because secondaries cannot keep up"; - error() << errmsg << migrateLog; - return; - } + log() << "Waiting for replication to catch up before entering critical section"; + if (flushPendingWrites(txn, ns, min, max, lastOpApplied, writeConcern)) + break; + sleepsecs(1); } - { - // 5. wait for commit - - setState(STEADY); - bool transferAfterCommit = false; - while ( getState() == STEADY || getState() == COMMIT_START ) { - txn->checkForInterrupt(); + if (t.minutes() >= 600) { + setState(FAIL); + errmsg = "Cannot go to critical section because secondaries cannot keep up"; + error() << errmsg << migrateLog; + return; + } + } - // Make sure we do at least one transfer after recv'ing the commit message - // If we aren't sure that at least one transfer happens *after* our state - // changes to COMMIT_START, there could be mods still on the FROM shard that - // got logged *after* our _transferMods but *before* the critical section. - if ( getState() == COMMIT_START ) transferAfterCommit = true; - - BSONObj res; - if ( ! conn->runCommand( "admin" , BSON( "_transferMods" << 1 ) , res ) ) { - log() << "_transferMods failed in STEADY state: " << res << migrateLog; - errmsg = res.toString(); - setState(FAIL); - conn.done(); - return; - } + { + // 5. wait for commit - if (res["size"].number() > 0 && - apply(txn, ns, min, max, shardKeyPattern, res, &lastOpApplied)) { - continue; - } + setState(STEADY); + bool transferAfterCommit = false; + while (getState() == STEADY || getState() == COMMIT_START) { + txn->checkForInterrupt(); - if ( getState() == ABORT ) { - return; - } + // Make sure we do at least one transfer after recv'ing the commit message + // If we aren't sure that at least one transfer happens *after* our state + // changes to COMMIT_START, there could be mods still on the FROM shard that + // got logged *after* our _transferMods but *before* the critical section. + if (getState() == COMMIT_START) + transferAfterCommit = true; - // We know we're finished when: - // 1) The from side has told us that it has locked writes (COMMIT_START) - // 2) We've checked at least one more time for un-transmitted mods - if ( getState() == COMMIT_START && transferAfterCommit == true ) { - if (flushPendingWrites(txn, ns, min, max, lastOpApplied, writeConcern)) - break; - } + BSONObj res; + if (!conn->runCommand("admin", BSON("_transferMods" << 1), res)) { + log() << "_transferMods failed in STEADY state: " << res << migrateLog; + errmsg = res.toString(); + setState(FAIL); + conn.done(); + return; + } - // Only sleep if we aren't committing - if ( getState() == STEADY ) sleepmillis( 10 ); + if (res["size"].number() > 0 && + apply(txn, ns, min, max, shardKeyPattern, res, &lastOpApplied)) { + continue; } - if ( getState() == FAIL ) { - errmsg = "timed out waiting for commit"; + if (getState() == ABORT) { return; } - timing.done(5); - MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep5); + // We know we're finished when: + // 1) The from side has told us that it has locked writes (COMMIT_START) + // 2) We've checked at least one more time for un-transmitted mods + if (getState() == COMMIT_START && transferAfterCommit == true) { + if (flushPendingWrites(txn, ns, min, max, lastOpApplied, writeConcern)) + break; + } + + // Only sleep if we aren't committing + if (getState() == STEADY) + sleepmillis(10); } - setState(DONE); - conn.done(); + if (getState() == FAIL) { + errmsg = "timed out waiting for commit"; + return; + } + + timing.done(5); + MONGO_FP_PAUSE_WHILE(migrateThreadHangAtStep5); } - void status(BSONObjBuilder& b) { - std::lock_guard<std::mutex> sl(_mutex); + setState(DONE); + conn.done(); + } - b.appendBool("active", _active); + void status(BSONObjBuilder& b) { + std::lock_guard<std::mutex> sl(_mutex); - b.append("ns", _ns); - b.append("from", _from); - b.append("min", _min); - b.append("max", _max); - b.append("shardKeyPattern", _shardKeyPattern); + b.appendBool("active", _active); - b.append("state", stateToString(_state)); - if (_state == FAIL) { - b.append("errmsg", _errmsg); - } + b.append("ns", _ns); + b.append("from", _from); + b.append("min", _min); + b.append("max", _max); + b.append("shardKeyPattern", _shardKeyPattern); - BSONObjBuilder bb(b.subobjStart("counts")); - bb.append("cloned", _numCloned); - bb.append("clonedBytes", _clonedBytes); - bb.append("catchup", _numCatchup); - bb.append("steady", _numSteady); - bb.done(); + b.append("state", stateToString(_state)); + if (_state == FAIL) { + b.append("errmsg", _errmsg); } - bool apply(OperationContext* txn, - const string& ns, - BSONObj min, - BSONObj max, - BSONObj shardKeyPattern, - const BSONObj& xfer, - repl::OpTime* lastOpApplied) { - - repl::OpTime dummy; - if ( lastOpApplied == NULL ) { - lastOpApplied = &dummy; - } + BSONObjBuilder bb(b.subobjStart("counts")); + bb.append("cloned", _numCloned); + bb.append("clonedBytes", _clonedBytes); + bb.append("catchup", _numCatchup); + bb.append("steady", _numSteady); + bb.done(); + } - bool didAnything = false; + bool apply(OperationContext* txn, + const string& ns, + BSONObj min, + BSONObj max, + BSONObj shardKeyPattern, + const BSONObj& xfer, + repl::OpTime* lastOpApplied) { + repl::OpTime dummy; + if (lastOpApplied == NULL) { + lastOpApplied = &dummy; + } - if ( xfer["deleted"].isABSONObj() ) { - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock dlk(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); - Helpers::RemoveSaver rs( "moveChunk" , ns , "removedDuring" ); + bool didAnything = false; - BSONObjIterator i( xfer["deleted"].Obj() ); - while ( i.more() ) { - Lock::CollectionLock clk(txn->lockState(), ns, MODE_X); - OldClientContext ctx(txn, ns); + if (xfer["deleted"].isABSONObj()) { + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock dlk(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); + Helpers::RemoveSaver rs("moveChunk", ns, "removedDuring"); - BSONObj id = i.next().Obj(); + BSONObjIterator i(xfer["deleted"].Obj()); + while (i.more()) { + Lock::CollectionLock clk(txn->lockState(), ns, MODE_X); + OldClientContext ctx(txn, ns); - // do not apply deletes if they do not belong to the chunk being migrated - BSONObj fullObj; - if (Helpers::findById(txn, ctx.db(), ns.c_str(), id, fullObj)) { - if (!isInRange(fullObj , min , max , shardKeyPattern)) { - log() << "not applying out of range deletion: " << fullObj << migrateLog; + BSONObj id = i.next().Obj(); - continue; - } - } + // do not apply deletes if they do not belong to the chunk being migrated + BSONObj fullObj; + if (Helpers::findById(txn, ctx.db(), ns.c_str(), id, fullObj)) { + if (!isInRange(fullObj, min, max, shardKeyPattern)) { + log() << "not applying out of range deletion: " << fullObj << migrateLog; - if (serverGlobalParams.moveParanoia) { - rs.goingToDelete(fullObj); + continue; } + } - deleteObjects(txn, - ctx.db(), - ns, - id, - PlanExecutor::YIELD_MANUAL, - true /* justOne */, - false /* god */, - true /* fromMigrate */); - - *lastOpApplied = repl::ReplClientInfo::forClient(txn->getClient()).getLastOp(); - didAnything = true; + if (serverGlobalParams.moveParanoia) { + rs.goingToDelete(fullObj); } + + deleteObjects(txn, + ctx.db(), + ns, + id, + PlanExecutor::YIELD_MANUAL, + true /* justOne */, + false /* god */, + true /* fromMigrate */); + + *lastOpApplied = repl::ReplClientInfo::forClient(txn->getClient()).getLastOp(); + didAnything = true; } + } - if ( xfer["reload"].isABSONObj() ) { - BSONObjIterator i( xfer["reload"].Obj() ); - while ( i.more() ) { - OldClientWriteContext cx(txn, ns); - - BSONObj updatedDoc = i.next().Obj(); - - BSONObj localDoc; - if (willOverrideLocalId(txn, - ns, - min, - max, - shardKeyPattern, - cx.db(), - updatedDoc, - &localDoc)) { - string errMsg = - str::stream() << "cannot migrate chunk, local document " - << localDoc - << " has same _id as reloaded remote document " - << updatedDoc; - - warning() << errMsg; - - // Exception will abort migration cleanly - uasserted( 16977, errMsg ); - } + if (xfer["reload"].isABSONObj()) { + BSONObjIterator i(xfer["reload"].Obj()); + while (i.more()) { + OldClientWriteContext cx(txn, ns); - // We are in write lock here, so sure we aren't killing - Helpers::upsert( txn, ns , updatedDoc , true ); + BSONObj updatedDoc = i.next().Obj(); - *lastOpApplied = repl::ReplClientInfo::forClient(txn->getClient()).getLastOp(); - didAnything = true; - } - } + BSONObj localDoc; + if (willOverrideLocalId( + txn, ns, min, max, shardKeyPattern, cx.db(), updatedDoc, &localDoc)) { + string errMsg = str::stream() + << "cannot migrate chunk, local document " << localDoc + << " has same _id as reloaded remote document " << updatedDoc; - return didAnything; - } + warning() << errMsg; - /** - * Checks if an upsert of a remote document will override a local document with the same _id - * but in a different range on this shard. - * Must be in WriteContext to avoid races and DBHelper errors. - * TODO: Could optimize this check out if sharding on _id. - */ - bool willOverrideLocalId(OperationContext* txn, - const string& ns, - BSONObj min, - BSONObj max, - BSONObj shardKeyPattern, - Database* db, - BSONObj remoteDoc, - BSONObj* localDoc) { + // Exception will abort migration cleanly + uasserted(16977, errMsg); + } + + // We are in write lock here, so sure we aren't killing + Helpers::upsert(txn, ns, updatedDoc, true); - *localDoc = BSONObj(); - if ( Helpers::findById( txn, db, ns.c_str(), remoteDoc, *localDoc ) ) { - return !isInRange( *localDoc , min , max , shardKeyPattern ); + *lastOpApplied = repl::ReplClientInfo::forClient(txn->getClient()).getLastOp(); + didAnything = true; } + } - return false; + return didAnything; + } + + /** + * Checks if an upsert of a remote document will override a local document with the same _id + * but in a different range on this shard. + * Must be in WriteContext to avoid races and DBHelper errors. + * TODO: Could optimize this check out if sharding on _id. + */ + bool willOverrideLocalId(OperationContext* txn, + const string& ns, + BSONObj min, + BSONObj max, + BSONObj shardKeyPattern, + Database* db, + BSONObj remoteDoc, + BSONObj* localDoc) { + *localDoc = BSONObj(); + if (Helpers::findById(txn, db, ns.c_str(), remoteDoc, *localDoc)) { + return !isInRange(*localDoc, min, max, shardKeyPattern); } - /** - * Returns true if the majority of the nodes and the nodes corresponding to the given - * writeConcern (if not empty) have applied till the specified lastOp. - */ - bool opReplicatedEnough(OperationContext* txn, - const repl::OpTime& lastOpApplied, - const WriteConcernOptions& writeConcern) { - WriteConcernOptions majorityWriteConcern; - majorityWriteConcern.wTimeout = -1; - majorityWriteConcern.wMode = WriteConcernOptions::kMajority; - Status majorityStatus = repl::getGlobalReplicationCoordinator()->awaitReplication( - txn, lastOpApplied, majorityWriteConcern).status; - - if (!writeConcern.shouldWaitForOtherNodes()) { - return majorityStatus.isOK(); - } - - // Also enforce the user specified write concern after "majority" so it covers - // the union of the 2 write concerns. - - WriteConcernOptions userWriteConcern(writeConcern); - userWriteConcern.wTimeout = -1; - Status userStatus = repl::getGlobalReplicationCoordinator()->awaitReplication( - txn, lastOpApplied, userWriteConcern).status; - - return majorityStatus.isOK() && userStatus.isOK(); - } - - bool flushPendingWrites(OperationContext* txn, - const std::string& ns, - BSONObj min, - BSONObj max, - const repl::OpTime& lastOpApplied, - const WriteConcernOptions& writeConcern) { - if (!opReplicatedEnough(txn, lastOpApplied, writeConcern)) { - repl::OpTime op(lastOpApplied); - OCCASIONALLY warning() << "migrate commit waiting for a majority of slaves for '" - << ns << "' " << min << " -> " << max - << " waiting for: " << op - << migrateLog; - return false; - } + return false; + } - log() << "migrate commit succeeded flushing to secondaries for '" << ns << "' " << min << " -> " << max << migrateLog; + /** + * Returns true if the majority of the nodes and the nodes corresponding to the given + * writeConcern (if not empty) have applied till the specified lastOp. + */ + bool opReplicatedEnough(OperationContext* txn, + const repl::OpTime& lastOpApplied, + const WriteConcernOptions& writeConcern) { + WriteConcernOptions majorityWriteConcern; + majorityWriteConcern.wTimeout = -1; + majorityWriteConcern.wMode = WriteConcernOptions::kMajority; + Status majorityStatus = repl::getGlobalReplicationCoordinator() + ->awaitReplication(txn, lastOpApplied, majorityWriteConcern) + .status; + + if (!writeConcern.shouldWaitForOtherNodes()) { + return majorityStatus.isOK(); + } - { - // Get global lock to wait for write to be commited to journal. - ScopedTransaction transaction(txn, MODE_S); - Lock::GlobalRead lk(txn->lockState()); + // Also enforce the user specified write concern after "majority" so it covers + // the union of the 2 write concerns. - // if durability is on, force a write to journal - if (getDur().commitNow(txn)) { - log() << "migrate commit flushed to journal for '" << ns << "' " << min << " -> " << max << migrateLog; - } - } + WriteConcernOptions userWriteConcern(writeConcern); + userWriteConcern.wTimeout = -1; + Status userStatus = repl::getGlobalReplicationCoordinator() + ->awaitReplication(txn, lastOpApplied, userWriteConcern) + .status; - return true; - } + return majorityStatus.isOK() && userStatus.isOK(); + } - static string stateToString(State state) { - switch (state) { - case READY: return "ready"; - case CLONE: return "clone"; - case CATCHUP: return "catchup"; - case STEADY: return "steady"; - case COMMIT_START: return "commitStart"; - case DONE: return "done"; - case FAIL: return "fail"; - case ABORT: return "abort"; - } - verify(0); - return ""; + bool flushPendingWrites(OperationContext* txn, + const std::string& ns, + BSONObj min, + BSONObj max, + const repl::OpTime& lastOpApplied, + const WriteConcernOptions& writeConcern) { + if (!opReplicatedEnough(txn, lastOpApplied, writeConcern)) { + repl::OpTime op(lastOpApplied); + OCCASIONALLY warning() << "migrate commit waiting for a majority of slaves for '" << ns + << "' " << min << " -> " << max << " waiting for: " << op + << migrateLog; + return false; } - bool startCommit() { - std::unique_lock<std::mutex> lock(_mutex); + log() << "migrate commit succeeded flushing to secondaries for '" << ns << "' " << min + << " -> " << max << migrateLog; - if (_state != STEADY) { - return false; - } + { + // Get global lock to wait for write to be commited to journal. + ScopedTransaction transaction(txn, MODE_S); + Lock::GlobalRead lk(txn->lockState()); - const auto deadline = system_clock::now() + seconds(30); - _state = COMMIT_START; - while (_active) { - if (std::cv_status::timeout == isActiveCV.wait_until(lock, deadline)) { - _state = FAIL; - log() << "startCommit never finished!" << migrateLog; - return false; - } + // if durability is on, force a write to journal + if (getDur().commitNow(txn)) { + log() << "migrate commit flushed to journal for '" << ns << "' " << min << " -> " + << max << migrateLog; } + } - if (_state == DONE) { - return true; - } + return true; + } - log() << "startCommit failed, final data failed to transfer" << migrateLog; + static string stateToString(State state) { + switch (state) { + case READY: + return "ready"; + case CLONE: + return "clone"; + case CATCHUP: + return "catchup"; + case STEADY: + return "steady"; + case COMMIT_START: + return "commitStart"; + case DONE: + return "done"; + case FAIL: + return "fail"; + case ABORT: + return "abort"; + } + verify(0); + return ""; + } + + bool startCommit() { + std::unique_lock<std::mutex> lock(_mutex); + + if (_state != STEADY) { return false; } - void abort() { - std::lock_guard<std::mutex> sl(_mutex); - _state = ABORT; - _errmsg = "aborted"; + const auto deadline = system_clock::now() + seconds(30); + _state = COMMIT_START; + while (_active) { + if (std::cv_status::timeout == isActiveCV.wait_until(lock, deadline)) { + _state = FAIL; + log() << "startCommit never finished!" << migrateLog; + return false; + } } - bool getActive() const { std::lock_guard<std::mutex> lk(_mutex); return _active; } - void setActive( bool b ) { - std::lock_guard<std::mutex> lk(_mutex); - _active = b; - isActiveCV.notify_all(); + if (_state == DONE) { + return true; } - // Guards all fields. - mutable std::mutex _mutex; - bool _active; - std::condition_variable isActiveCV; + log() << "startCommit failed, final data failed to transfer" << migrateLog; + return false; + } + + void abort() { + std::lock_guard<std::mutex> sl(_mutex); + _state = ABORT; + _errmsg = "aborted"; + } + + bool getActive() const { + std::lock_guard<std::mutex> lk(_mutex); + return _active; + } + void setActive(bool b) { + std::lock_guard<std::mutex> lk(_mutex); + _active = b; + isActiveCV.notify_all(); + } - std::string _ns; - std::string _from; - BSONObj _min; - BSONObj _max; - BSONObj _shardKeyPattern; + // Guards all fields. + mutable std::mutex _mutex; + bool _active; + std::condition_variable isActiveCV; - long long _numCloned; - long long _clonedBytes; - long long _numCatchup; - long long _numSteady; + std::string _ns; + std::string _from; + BSONObj _min; + BSONObj _max; + BSONObj _shardKeyPattern; - State _state; - std::string _errmsg; + long long _numCloned; + long long _clonedBytes; + long long _numCatchup; + long long _numSteady; - } migrateStatus; + State _state; + std::string _errmsg; - void migrateThread(std::string ns, - BSONObj min, - BSONObj max, - BSONObj shardKeyPattern, - std::string fromShard, - OID epoch, - WriteConcernOptions writeConcern) { - Client::initThread( "migrateThread" ); - OperationContextImpl txn; - if (getGlobalAuthorizationManager()->isAuthEnabled()) { - ShardedConnectionInfo::addHook(); - AuthorizationSession::get(txn.getClient())->grantInternalAuthorization(); - } +} migrateStatus; - migrateStatus.go(&txn, ns, min, max, shardKeyPattern, fromShard, epoch, writeConcern); +void migrateThread(std::string ns, + BSONObj min, + BSONObj max, + BSONObj shardKeyPattern, + std::string fromShard, + OID epoch, + WriteConcernOptions writeConcern) { + Client::initThread("migrateThread"); + OperationContextImpl txn; + if (getGlobalAuthorizationManager()->isAuthEnabled()) { + ShardedConnectionInfo::addHook(); + AuthorizationSession::get(txn.getClient())->grantInternalAuthorization(); } - /** - * Command for initiating the recipient side of the migration to start copying data - * from the donor shard. - * - * { - * _recvChunkStart: "namespace", - * congfigServer: "hostAndPort", - * from: "hostAndPort", - * fromShardName: "shardName", - * toShardName: "shardName", - * min: {}, - * max: {}, - * shardKeyPattern: {}, - * - * // optional - * secondaryThrottle: bool, // defaults to true - * writeConcern: {} // applies to individual writes. - * } - */ - class RecvChunkStartCommand : public ChunkCommandHelper { - public: - void help(std::stringstream& h) const { h << "internal"; } - RecvChunkStartCommand() : ChunkCommandHelper( "_recvChunkStart" ) {} - - virtual bool isWriteCommandForConfigServer() const { return false; } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::internal); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - - // Active state of TO-side migrations (MigrateStatus) is serialized by distributed - // collection lock. - if ( migrateStatus.getActive() ) { - errmsg = "migrate already in progress"; - return false; - } + migrateStatus.go(&txn, ns, min, max, shardKeyPattern, fromShard, epoch, writeConcern); +} - // Pending deletes (for migrations) are serialized by the distributed collection lock, - // we are sure we registered a delete for a range *before* we can migrate-in a - // subrange. - const size_t numDeletes = getDeleter()->getTotalDeletes(); - if (numDeletes > 0) { +/** + * Command for initiating the recipient side of the migration to start copying data + * from the donor shard. + * + * { + * _recvChunkStart: "namespace", + * congfigServer: "hostAndPort", + * from: "hostAndPort", + * fromShardName: "shardName", + * toShardName: "shardName", + * min: {}, + * max: {}, + * shardKeyPattern: {}, + * + * // optional + * secondaryThrottle: bool, // defaults to true + * writeConcern: {} // applies to individual writes. + * } + */ +class RecvChunkStartCommand : public ChunkCommandHelper { +public: + void help(std::stringstream& h) const { + h << "internal"; + } + RecvChunkStartCommand() : ChunkCommandHelper("_recvChunkStart") {} - errmsg = str::stream() << "can't accept new chunks because " - << " there are still " << numDeletes - << " deletes from previous migration"; + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::internal); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + // Active state of TO-side migrations (MigrateStatus) is serialized by distributed + // collection lock. + if (migrateStatus.getActive()) { + errmsg = "migrate already in progress"; + return false; + } + + // Pending deletes (for migrations) are serialized by the distributed collection lock, + // we are sure we registered a delete for a range *before* we can migrate-in a + // subrange. + const size_t numDeletes = getDeleter()->getTotalDeletes(); + if (numDeletes > 0) { + errmsg = str::stream() << "can't accept new chunks because " + << " there are still " << numDeletes + << " deletes from previous migration"; + + warning() << errmsg; + return false; + } + + if (!shardingState.enabled()) { + if (!cmdObj["configServer"].eoo()) { + dassert(cmdObj["configServer"].type() == String); + ShardingState::initialize(cmdObj["configServer"].String()); + } else { + errmsg = str::stream() + << "cannot start recv'ing chunk, " + << "sharding is not enabled and no config server was provided"; warning() << errmsg; return false; } + } - if (!shardingState.enabled()) { - if (!cmdObj["configServer"].eoo()) { - dassert(cmdObj["configServer"].type() == String); - ShardingState::initialize(cmdObj["configServer"].String()); - } - else { - - errmsg = str::stream() - << "cannot start recv'ing chunk, " - << "sharding is not enabled and no config server was provided"; + if (!cmdObj["toShardName"].eoo()) { + dassert(cmdObj["toShardName"].type() == String); + shardingState.gotShardName(cmdObj["toShardName"].String()); + } - warning() << errmsg; - return false; - } - } + string ns = cmdObj.firstElement().String(); + BSONObj min = cmdObj["min"].Obj().getOwned(); + BSONObj max = cmdObj["max"].Obj().getOwned(); - if ( !cmdObj["toShardName"].eoo() ) { - dassert( cmdObj["toShardName"].type() == String ); - shardingState.gotShardName( cmdObj["toShardName"].String() ); - } + // Refresh our collection manager from the config server, we need a collection manager + // to start registering pending chunks. + // We force the remote refresh here to make the behavior consistent and predictable, + // generally we'd refresh anyway, and to be paranoid. + ChunkVersion currentVersion; + Status status = shardingState.refreshMetadataNow(txn, ns, ¤tVersion); - string ns = cmdObj.firstElement().String(); - BSONObj min = cmdObj["min"].Obj().getOwned(); - BSONObj max = cmdObj["max"].Obj().getOwned(); + if (!status.isOK()) { + errmsg = str::stream() << "cannot start recv'ing chunk " + << "[" << min << "," << max << ")" << causedBy(status.reason()); - // Refresh our collection manager from the config server, we need a collection manager - // to start registering pending chunks. - // We force the remote refresh here to make the behavior consistent and predictable, - // generally we'd refresh anyway, and to be paranoid. - ChunkVersion currentVersion; - Status status = shardingState.refreshMetadataNow(txn, ns, ¤tVersion ); + warning() << errmsg; + return false; + } - if ( !status.isOK() ) { - errmsg = str::stream() << "cannot start recv'ing chunk " - << "[" << min << "," << max << ")" - << causedBy( status.reason() ); + // Process secondary throttle settings and assign defaults if necessary. + WriteConcernOptions writeConcern; + status = writeConcern.parseSecondaryThrottle(cmdObj, NULL); - warning() << errmsg; - return false; + if (!status.isOK()) { + if (status.code() != ErrorCodes::WriteConcernNotDefined) { + warning() << status.toString(); + return appendCommandStatus(result, status); } - // Process secondary throttle settings and assign defaults if necessary. - WriteConcernOptions writeConcern; - status = writeConcern.parseSecondaryThrottle(cmdObj, NULL); + writeConcern = getDefaultWriteConcern(); + } else { + repl::ReplicationCoordinator* replCoordinator = repl::getGlobalReplicationCoordinator(); - if (!status.isOK()){ - if (status.code() != ErrorCodes::WriteConcernNotDefined) { - warning() << status.toString(); - return appendCommandStatus(result, status); - } + if (replCoordinator->getReplicationMode() == + repl::ReplicationCoordinator::modeMasterSlave && + writeConcern.shouldWaitForOtherNodes()) { + warning() << "recvChunk cannot check if secondary throttle setting " + << writeConcern.toBSON() + << " can be enforced in a master slave configuration"; + } - writeConcern = getDefaultWriteConcern(); + Status status = replCoordinator->checkIfWriteConcernCanBeSatisfied(writeConcern); + if (!status.isOK() && status != ErrorCodes::NoReplicationEnabled) { + warning() << status.toString(); + return appendCommandStatus(result, status); } - else { - repl::ReplicationCoordinator* replCoordinator = - repl::getGlobalReplicationCoordinator(); + } - if (replCoordinator->getReplicationMode() == - repl::ReplicationCoordinator::modeMasterSlave && - writeConcern.shouldWaitForOtherNodes()) { - warning() << "recvChunk cannot check if secondary throttle setting " - << writeConcern.toBSON() - << " can be enforced in a master slave configuration"; - } + if (writeConcern.shouldWaitForOtherNodes() && + writeConcern.wTimeout == WriteConcernOptions::kNoTimeout) { + // Don't allow no timeout. + writeConcern.wTimeout = kDefaultWTimeoutMs; + } - Status status = replCoordinator->checkIfWriteConcernCanBeSatisfied(writeConcern); - if (!status.isOK() && status != ErrorCodes::NoReplicationEnabled) { - warning() << status.toString(); - return appendCommandStatus(result, status); - } - } + BSONObj shardKeyPattern; + if (cmdObj.hasField("shardKeyPattern")) { + shardKeyPattern = cmdObj["shardKeyPattern"].Obj().getOwned(); + } else { + // shardKeyPattern may not be provided if another shard is from pre 2.2 + // In that case, assume the shard key pattern is the same as the range + // specifiers provided. + BSONObj keya = Helpers::inferKeyPattern(min); + BSONObj keyb = Helpers::inferKeyPattern(max); + verify(keya == keyb); + + warning() + << "No shard key pattern provided by source shard for migration." + " This is likely because the source shard is running a version prior to 2.2." + " Falling back to assuming the shard key matches the pattern of the min and max" + " chunk range specifiers. Inferred shard key: " << keya; + + shardKeyPattern = keya.getOwned(); + } - if (writeConcern.shouldWaitForOtherNodes() && - writeConcern.wTimeout == WriteConcernOptions::kNoTimeout) { - // Don't allow no timeout. - writeConcern.wTimeout = kDefaultWTimeoutMs; - } + const string fromShard(cmdObj["from"].String()); - BSONObj shardKeyPattern; - if (cmdObj.hasField("shardKeyPattern")) { - shardKeyPattern = cmdObj["shardKeyPattern"].Obj().getOwned(); - } else { - // shardKeyPattern may not be provided if another shard is from pre 2.2 - // In that case, assume the shard key pattern is the same as the range - // specifiers provided. - BSONObj keya = Helpers::inferKeyPattern(min); - BSONObj keyb = Helpers::inferKeyPattern(max); - verify( keya == keyb ); + // Set the TO-side migration to active + Status prepareStatus = migrateStatus.prepare(ns, fromShard, min, max, shardKeyPattern); - warning() << "No shard key pattern provided by source shard for migration." - " This is likely because the source shard is running a version prior to 2.2." - " Falling back to assuming the shard key matches the pattern of the min and max" - " chunk range specifiers. Inferred shard key: " << keya; + if (!prepareStatus.isOK()) { + return appendCommandStatus(result, prepareStatus); + } - shardKeyPattern = keya.getOwned(); - } + std::thread m(migrateThread, + ns, + min, + max, + shardKeyPattern, + fromShard, + currentVersion.epoch(), + writeConcern); + + m.detach(); + result.appendBool("started", true); + return true; + } - const string fromShard(cmdObj["from"].String()); +} recvChunkStartCmd; - // Set the TO-side migration to active - Status prepareStatus = migrateStatus.prepare(ns, fromShard, min, max, shardKeyPattern); +class RecvChunkStatusCommand : public ChunkCommandHelper { +public: + void help(std::stringstream& h) const { + h << "internal"; + } + RecvChunkStatusCommand() : ChunkCommandHelper("_recvChunkStatus") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::internal); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + migrateStatus.status(result); + return 1; + } - if (!prepareStatus.isOK()) { - return appendCommandStatus(result, prepareStatus); - } +} recvChunkStatusCommand; - std::thread m(migrateThread, - ns, - min, - max, - shardKeyPattern, - fromShard, - currentVersion.epoch(), - writeConcern); +class RecvChunkCommitCommand : public ChunkCommandHelper { +public: + void help(std::stringstream& h) const { + h << "internal"; + } + RecvChunkCommitCommand() : ChunkCommandHelper("_recvChunkCommit") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::internal); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + bool ok = migrateStatus.startCommit(); + migrateStatus.status(result); + return ok; + } - m.detach(); - result.appendBool( "started" , true ); - return true; - } +} recvChunkCommitCommand; - } recvChunkStartCmd; +class RecvChunkAbortCommand : public ChunkCommandHelper { +public: + void help(std::stringstream& h) const { + h << "internal"; + } + RecvChunkAbortCommand() : ChunkCommandHelper("_recvChunkAbort") {} + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::internal); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + migrateStatus.abort(); + migrateStatus.status(result); + return true; + } - class RecvChunkStatusCommand : public ChunkCommandHelper { - public: - void help(std::stringstream& h) const { h << "internal"; } - RecvChunkStatusCommand() : ChunkCommandHelper( "_recvChunkStatus" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::internal); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - migrateStatus.status( result ); - return 1; - } - - } recvChunkStatusCommand; - - class RecvChunkCommitCommand : public ChunkCommandHelper { - public: - void help(std::stringstream& h) const { h << "internal"; } - RecvChunkCommitCommand() : ChunkCommandHelper( "_recvChunkCommit" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::internal); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - bool ok = migrateStatus.startCommit(); - migrateStatus.status( result ); - return ok; - } - - } recvChunkCommitCommand; - - class RecvChunkAbortCommand : public ChunkCommandHelper { - public: - void help(std::stringstream& h) const { h << "internal"; } - RecvChunkAbortCommand() : ChunkCommandHelper( "_recvChunkAbort" ) {} - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::internal); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - migrateStatus.abort(); - migrateStatus.status( result ); - return true; - } +} recvChunkAboortCommand; - } recvChunkAboortCommand; +class IsInRangeTest : public StartupTest { +public: + void run() { + BSONObj min = BSON("x" << 1); + BSONObj max = BSON("x" << 5); + BSONObj skey = BSON("x" << 1); - class IsInRangeTest : public StartupTest { - public: - void run() { - BSONObj min = BSON( "x" << 1 ); - BSONObj max = BSON( "x" << 5 ); - BSONObj skey = BSON( "x" << 1 ); - - verify( ! isInRange( BSON( "x" << 0 ) , min , max , skey ) ); - verify( isInRange( BSON( "x" << 1 ) , min , max , skey ) ); - verify( isInRange( BSON( "x" << 3 ) , min , max , skey ) ); - verify( isInRange( BSON( "x" << 4 ) , min , max , skey ) ); - verify( ! isInRange( BSON( "x" << 5 ) , min , max , skey ) ); - verify( ! isInRange( BSON( "x" << 6 ) , min , max , skey ) ); - - BSONObj obj = BSON( "n" << 3 ); - BSONObj min2 = BSON( "x" << BSONElementHasher::hash64( obj.firstElement() , 0 ) - 2 ); - BSONObj max2 = BSON( "x" << BSONElementHasher::hash64( obj.firstElement() , 0 ) + 2 ); - BSONObj hashedKey = BSON( "x" << "hashed" ); - - verify( isInRange( BSON( "x" << 3 ) , min2 , max2 , hashedKey ) ); - verify( ! isInRange( BSON( "x" << 3 ) , min , max , hashedKey ) ); - verify( ! isInRange( BSON( "x" << 4 ) , min2 , max2 , hashedKey ) ); - - LOG(1) << "isInRangeTest passed" << migrateLog; - } - } isInRangeTest; + verify(!isInRange(BSON("x" << 0), min, max, skey)); + verify(isInRange(BSON("x" << 1), min, max, skey)); + verify(isInRange(BSON("x" << 3), min, max, skey)); + verify(isInRange(BSON("x" << 4), min, max, skey)); + verify(!isInRange(BSON("x" << 5), min, max, skey)); + verify(!isInRange(BSON("x" << 6), min, max, skey)); + + BSONObj obj = BSON("n" << 3); + BSONObj min2 = BSON("x" << BSONElementHasher::hash64(obj.firstElement(), 0) - 2); + BSONObj max2 = BSON("x" << BSONElementHasher::hash64(obj.firstElement(), 0) + 2); + BSONObj hashedKey = BSON("x" + << "hashed"); + + verify(isInRange(BSON("x" << 3), min2, max2, hashedKey)); + verify(!isInRange(BSON("x" << 3), min, max, hashedKey)); + verify(!isInRange(BSON("x" << 4), min2, max2, hashedKey)); + + LOG(1) << "isInRangeTest passed" << migrateLog; + } +} isInRangeTest; } diff --git a/src/mongo/s/d_split.cpp b/src/mongo/s/d_split.cpp index 98d62c7029a..cbf81006723 100644 --- a/src/mongo/s/d_split.cpp +++ b/src/mongo/s/d_split.cpp @@ -67,866 +67,875 @@ namespace mongo { - using std::unique_ptr; - using std::endl; - using std::ostringstream; - using std::set; - using std::string; - using std::stringstream; - using std::vector; - - class CmdMedianKey : public Command { - public: - CmdMedianKey() : Command( "medianKey" ) {} - virtual bool slaveOk() const { return true; } - virtual bool isWriteCommandForConfigServer() const { return false; } - virtual void help( stringstream &help ) const { - help << "Deprecated internal command. Use splitVector command instead. \n"; - } - // No auth required as this command no longer does anything. - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) {} - bool run(OperationContext* txn, - const string& dbname, - BSONObj& jsobj, - int, - string& errmsg, - BSONObjBuilder& result) { - errmsg = "medianKey command no longer supported. Calling this indicates mismatch between mongo versions."; +using std::unique_ptr; +using std::endl; +using std::ostringstream; +using std::set; +using std::string; +using std::stringstream; +using std::vector; + +class CmdMedianKey : public Command { +public: + CmdMedianKey() : Command("medianKey") {} + virtual bool slaveOk() const { + return true; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual void help(stringstream& help) const { + help << "Deprecated internal command. Use splitVector command instead. \n"; + } + // No auth required as this command no longer does anything. + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) {} + bool run(OperationContext* txn, + const string& dbname, + BSONObj& jsobj, + int, + string& errmsg, + BSONObjBuilder& result) { + errmsg = + "medianKey command no longer supported. Calling this indicates mismatch between mongo " + "versions."; + return false; + } +} cmdMedianKey; + +class CheckShardingIndex : public Command { +public: + CheckShardingIndex() : Command("checkShardingIndex", false) {} + virtual bool slaveOk() const { + return false; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual void help(stringstream& help) const { + help << "Internal command.\n"; + } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::find); + out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + } + + std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } + + bool run(OperationContext* txn, + const string& dbname, + BSONObj& jsobj, + int, + string& errmsg, + BSONObjBuilder& result) { + std::string ns = parseNs(dbname, jsobj); + BSONObj keyPattern = jsobj.getObjectField("keyPattern"); + + if (keyPattern.isEmpty()) { + errmsg = "no key pattern found in checkShardingindex"; return false; } - } cmdMedianKey; - class CheckShardingIndex : public Command { - public: - CheckShardingIndex() : Command( "checkShardingIndex" , false ) {} - virtual bool slaveOk() const { return false; } - virtual bool isWriteCommandForConfigServer() const { return false; } - virtual void help( stringstream &help ) const { - help << "Internal command.\n"; + if (keyPattern.nFields() == 1 && str::equals("_id", keyPattern.firstElementFieldName())) { + result.appendBool("idskip", true); + return true; + } + + BSONObj min = jsobj.getObjectField("min"); + BSONObj max = jsobj.getObjectField("max"); + if (min.isEmpty() != max.isEmpty()) { + errmsg = "either provide both min and max or leave both empty"; + return false; } - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::find); - out->push_back(Privilege(parseResourcePattern(dbname, cmdObj), actions)); + + AutoGetCollectionForRead ctx(txn, ns); + Collection* collection = ctx.getCollection(); + if (!collection) { + errmsg = "ns not found"; + return false; } - std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); + IndexDescriptor* idx = + collection->getIndexCatalog()->findShardKeyPrefixedIndex(txn, + keyPattern, + true); // requireSingleKey + if (idx == NULL) { + errmsg = "couldn't find valid index for shard key"; + return false; + } + // extend min to get (min, MinKey, MinKey, ....) + KeyPattern kp(idx->keyPattern()); + min = Helpers::toKeyFormat(kp.extendRangeBound(min, false)); + if (max.isEmpty()) { + // if max not specified, make it (MaxKey, Maxkey, MaxKey...) + max = Helpers::toKeyFormat(kp.extendRangeBound(max, true)); + } else { + // otherwise make it (max,MinKey,MinKey...) so that bound is non-inclusive + max = Helpers::toKeyFormat(kp.extendRangeBound(max, false)); } - bool run(OperationContext* txn, - const string& dbname, - BSONObj& jsobj, - int, - string& errmsg, - BSONObjBuilder& result) { + unique_ptr<PlanExecutor> exec(InternalPlanner::indexScan( + txn, collection, idx, min, max, false, InternalPlanner::FORWARD)); + exec->setYieldPolicy(PlanExecutor::YIELD_AUTO); + + // Find the 'missingField' value used to represent a missing document field in a key of + // this index. + // NOTE A local copy of 'missingField' is made because indices may be + // invalidated during a db lock yield. + BSONObj missingFieldObj = IndexLegacy::getMissingField(txn, collection, idx->infoObj()); + BSONElement missingField = missingFieldObj.firstElement(); + + // for now, the only check is that all shard keys are filled + // a 'missingField' valued index key is ok if the field is present in the document, + // TODO if $exist for nulls were picking the index, it could be used instead efficiently + int keyPatternLength = keyPattern.nFields(); + + RecordId loc; + BSONObj currKey; + while (PlanExecutor::ADVANCED == exec->getNext(&currKey, &loc)) { + // check that current key contains non missing elements for all fields in keyPattern + BSONObjIterator i(currKey); + for (int k = 0; k < keyPatternLength; k++) { + if (!i.more()) { + errmsg = str::stream() << "index key " << currKey << " too short for pattern " + << keyPattern; + return false; + } + BSONElement currKeyElt = i.next(); - std::string ns = parseNs(dbname, jsobj); - BSONObj keyPattern = jsobj.getObjectField( "keyPattern" ); + if (!currKeyElt.eoo() && !currKeyElt.valuesEqual(missingField)) + continue; - if ( keyPattern.isEmpty() ) { - errmsg = "no key pattern found in checkShardingindex"; - return false; - } + // This is a fetch, but it's OK. The underlying code won't throw a page fault + // exception. + BSONObj obj = collection->docFor(txn, loc).value(); + BSONObjIterator j(keyPattern); + BSONElement real; + for (int x = 0; x <= k; x++) + real = j.next(); - if ( keyPattern.nFields() == 1 && str::equals( "_id" , keyPattern.firstElementFieldName() ) ) { - result.appendBool( "idskip" , true ); - return true; - } + real = obj.getFieldDotted(real.fieldName()); + + if (real.type()) + continue; - BSONObj min = jsobj.getObjectField( "min" ); - BSONObj max = jsobj.getObjectField( "max" ); - if ( min.isEmpty() != max.isEmpty() ) { - errmsg = "either provide both min and max or leave both empty"; + ostringstream os; + os << "found missing value in key " << currKey << " for doc: " + << (obj.hasField("_id") ? obj.toString() : obj["_id"].toString()); + log() << "checkShardingIndex for '" << ns << "' failed: " << os.str() << endl; + + errmsg = os.str(); return false; } + } + + return true; + } +} cmdCheckShardingIndex; + +BSONObj prettyKey(const BSONObj& keyPattern, const BSONObj& key) { + return key.replaceFieldNames(keyPattern).clientReadable(); +} + +class SplitVector : public Command { +public: + SplitVector() : Command("splitVector", false) {} + virtual bool slaveOk() const { + return false; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual void help(stringstream& help) const { + help << "Internal command.\n" + "examples:\n" + " { splitVector : \"blog.post\" , keyPattern:{x:1} , min:{x:10} , max:{x:20}, " + "maxChunkSize:200 }\n" + " maxChunkSize unit in MBs\n" + " May optionally specify 'maxSplitPoints' and 'maxChunkObjects' to avoid " + "traversing the whole chunk\n" + " \n" + " { splitVector : \"blog.post\" , keyPattern:{x:1} , min:{x:10} , max:{x:20}, " + "force: true }\n" + " 'force' will produce one split point even if data is small; defaults to false\n" + "NOTE: This command may take a while to run"; + } + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::splitVector)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); + } + return Status::OK(); + } + virtual std::string parseNs(const string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } + bool run(OperationContext* txn, + const string& dbname, + BSONObj& jsobj, + int, + string& errmsg, + BSONObjBuilder& result) { + // + // 1.a We'll parse the parameters in two steps. First, make sure the we can use the split index to get + // a good approximation of the size of the chunk -- without needing to access the actual data. + // + + const std::string ns = parseNs(dbname, jsobj); + BSONObj keyPattern = jsobj.getObjectField("keyPattern"); + + if (keyPattern.isEmpty()) { + errmsg = "no key pattern found in splitVector"; + return false; + } + + // If min and max are not provided use the "minKey" and "maxKey" for the sharding key pattern. + BSONObj min = jsobj.getObjectField("min"); + BSONObj max = jsobj.getObjectField("max"); + if (min.isEmpty() != max.isEmpty()) { + errmsg = "either provide both min and max or leave both empty"; + return false; + } + + long long maxSplitPoints = 0; + BSONElement maxSplitPointsElem = jsobj["maxSplitPoints"]; + if (maxSplitPointsElem.isNumber()) { + maxSplitPoints = maxSplitPointsElem.numberLong(); + } + + long long maxChunkObjects = Chunk::MaxObjectPerChunk; + BSONElement MaxChunkObjectsElem = jsobj["maxChunkObjects"]; + if (MaxChunkObjectsElem.isNumber()) { + maxChunkObjects = MaxChunkObjectsElem.numberLong(); + } + + vector<BSONObj> splitKeys; + { + // Get the size estimate for this namespace AutoGetCollectionForRead ctx(txn, ns); Collection* collection = ctx.getCollection(); - if ( !collection ) { + if (!collection) { errmsg = "ns not found"; return false; } - IndexDescriptor *idx = - collection->getIndexCatalog() - ->findShardKeyPrefixedIndex( txn, - keyPattern, - true ); // requireSingleKey - if ( idx == NULL ) { - errmsg = "couldn't find valid index for shard key"; + // Allow multiKey based on the invariant that shard keys must be single-valued. + // Therefore, any multi-key index prefixed by shard key cannot be multikey over + // the shard key fields. + IndexDescriptor* idx = + collection->getIndexCatalog()->findShardKeyPrefixedIndex(txn, keyPattern, false); + if (idx == NULL) { + errmsg = (string) "couldn't find index over splitting key " + + keyPattern.clientReadable().toString(); return false; } // extend min to get (min, MinKey, MinKey, ....) - KeyPattern kp( idx->keyPattern() ); - min = Helpers::toKeyFormat( kp.extendRangeBound( min, false ) ); - if ( max.isEmpty() ) { + KeyPattern kp(idx->keyPattern()); + min = Helpers::toKeyFormat(kp.extendRangeBound(min, false)); + if (max.isEmpty()) { // if max not specified, make it (MaxKey, Maxkey, MaxKey...) - max = Helpers::toKeyFormat( kp.extendRangeBound( max, true ) ); + max = Helpers::toKeyFormat(kp.extendRangeBound(max, true)); } else { // otherwise make it (max,MinKey,MinKey...) so that bound is non-inclusive - max = Helpers::toKeyFormat( kp.extendRangeBound( max, false ) ); + max = Helpers::toKeyFormat(kp.extendRangeBound(max, false)); } - unique_ptr<PlanExecutor> exec(InternalPlanner::indexScan(txn, collection, idx, - min, max, false, - InternalPlanner::FORWARD)); - exec->setYieldPolicy(PlanExecutor::YIELD_AUTO); + const long long recCount = collection->numRecords(txn); + const long long dataSize = collection->dataSize(txn); - // Find the 'missingField' value used to represent a missing document field in a key of - // this index. - // NOTE A local copy of 'missingField' is made because indices may be - // invalidated during a db lock yield. - BSONObj missingFieldObj = IndexLegacy::getMissingField(txn, collection, idx->infoObj()); - BSONElement missingField = missingFieldObj.firstElement(); - - // for now, the only check is that all shard keys are filled - // a 'missingField' valued index key is ok if the field is present in the document, - // TODO if $exist for nulls were picking the index, it could be used instead efficiently - int keyPatternLength = keyPattern.nFields(); - - RecordId loc; - BSONObj currKey; - while (PlanExecutor::ADVANCED == exec->getNext(&currKey, &loc)) { - //check that current key contains non missing elements for all fields in keyPattern - BSONObjIterator i( currKey ); - for( int k = 0; k < keyPatternLength ; k++ ) { - if( ! i.more() ) { - errmsg = str::stream() << "index key " << currKey - << " too short for pattern " << keyPattern; - return false; + // + // 1.b Now that we have the size estimate, go over the remaining parameters and apply any maximum size + // restrictions specified there. + // + + // 'force'-ing a split is equivalent to having maxChunkSize be the size of the current chunk, i.e., the + // logic below will split that chunk in half + long long maxChunkSize = 0; + bool forceMedianSplit = false; + { + BSONElement maxSizeElem = jsobj["maxChunkSize"]; + BSONElement forceElem = jsobj["force"]; + + if (forceElem.trueValue()) { + forceMedianSplit = true; + // This chunk size is effectively ignored if force is true + maxChunkSize = dataSize; + + } else if (maxSizeElem.isNumber()) { + maxChunkSize = maxSizeElem.numberLong() * 1 << 20; + + } else { + maxSizeElem = jsobj["maxChunkSizeBytes"]; + if (maxSizeElem.isNumber()) { + maxChunkSize = maxSizeElem.numberLong(); } - BSONElement currKeyElt = i.next(); - - if ( !currKeyElt.eoo() && !currKeyElt.valuesEqual( missingField ) ) - continue; - - // This is a fetch, but it's OK. The underlying code won't throw a page fault - // exception. - BSONObj obj = collection->docFor(txn, loc).value(); - BSONObjIterator j( keyPattern ); - BSONElement real; - for ( int x=0; x <= k; x++ ) - real = j.next(); - - real = obj.getFieldDotted( real.fieldName() ); - - if ( real.type() ) - continue; - - ostringstream os; - os << "found missing value in key " << currKey << " for doc: " - << ( obj.hasField( "_id" ) ? obj.toString() : obj["_id"].toString() ); - log() << "checkShardingIndex for '" << ns << "' failed: " << os.str() << endl; - - errmsg = os.str(); + } + + // We need a maximum size for the chunk, unless we're not actually capable of finding any + // split points. + if (maxChunkSize <= 0 && recCount != 0) { + errmsg = + "need to specify the desired max chunk size (maxChunkSize or " + "maxChunkSizeBytes)"; return false; } } - return true; - } - } cmdCheckShardingIndex; - BSONObj prettyKey(const BSONObj& keyPattern, const BSONObj& key) { - return key.replaceFieldNames(keyPattern).clientReadable(); - } + // If there's not enough data for more than one chunk, no point continuing. + if (dataSize < maxChunkSize || recCount == 0) { + vector<BSONObj> emptyVector; + result.append("splitKeys", emptyVector); + return true; + } + + log() << "request split points lookup for chunk " << ns << " " << min << " -->> " << max + << endl; - class SplitVector : public Command { - public: - SplitVector() : Command( "splitVector" , false ) {} - virtual bool slaveOk() const { return false; } - virtual bool isWriteCommandForConfigServer() const { return false; } - virtual void help( stringstream &help ) const { - help << - "Internal command.\n" - "examples:\n" - " { splitVector : \"blog.post\" , keyPattern:{x:1} , min:{x:10} , max:{x:20}, maxChunkSize:200 }\n" - " maxChunkSize unit in MBs\n" - " May optionally specify 'maxSplitPoints' and 'maxChunkObjects' to avoid traversing the whole chunk\n" - " \n" - " { splitVector : \"blog.post\" , keyPattern:{x:1} , min:{x:10} , max:{x:20}, force: true }\n" - " 'force' will produce one split point even if data is small; defaults to false\n" - "NOTE: This command may take a while to run"; - } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), - ActionType::splitVector)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); + // We'll use the average object size and number of object to find approximately how many keys + // each chunk should have. We'll split at half the maxChunkSize or maxChunkObjects, if + // provided. + const long long avgRecSize = dataSize / recCount; + long long keyCount = maxChunkSize / (2 * avgRecSize); + if (maxChunkObjects && (maxChunkObjects < keyCount)) { + log() << "limiting split vector to " << maxChunkObjects << " (from " << keyCount + << ") objects " << endl; + keyCount = maxChunkObjects; } - return Status::OK(); - } - virtual std::string parseNs(const string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); - } - bool run(OperationContext* txn, - const string& dbname, - BSONObj& jsobj, - int, - string& errmsg, - BSONObjBuilder& result) { // - // 1.a We'll parse the parameters in two steps. First, make sure the we can use the split index to get - // a good approximation of the size of the chunk -- without needing to access the actual data. + // 2. Traverse the index and add the keyCount-th key to the result vector. If that key + // appeared in the vector before, we omit it. The invariant here is that all the + // instances of a given key value live in the same chunk. // - const std::string ns = parseNs(dbname, jsobj); - BSONObj keyPattern = jsobj.getObjectField( "keyPattern" ); + Timer timer; + long long currCount = 0; + long long numChunks = 0; - if ( keyPattern.isEmpty() ) { - errmsg = "no key pattern found in splitVector"; - return false; - } + unique_ptr<PlanExecutor> exec(InternalPlanner::indexScan( + txn, collection, idx, min, max, false, InternalPlanner::FORWARD)); - // If min and max are not provided use the "minKey" and "maxKey" for the sharding key pattern. - BSONObj min = jsobj.getObjectField( "min" ); - BSONObj max = jsobj.getObjectField( "max" ); - if ( min.isEmpty() != max.isEmpty() ){ - errmsg = "either provide both min and max or leave both empty"; + BSONObj currKey; + PlanExecutor::ExecState state = exec->getNext(&currKey, NULL); + if (PlanExecutor::ADVANCED != state) { + errmsg = "can't open a cursor for splitting (desired range is possibly empty)"; return false; } - long long maxSplitPoints = 0; - BSONElement maxSplitPointsElem = jsobj[ "maxSplitPoints" ]; - if ( maxSplitPointsElem.isNumber() ) { - maxSplitPoints = maxSplitPointsElem.numberLong(); - } - - long long maxChunkObjects = Chunk::MaxObjectPerChunk; - BSONElement MaxChunkObjectsElem = jsobj[ "maxChunkObjects" ]; - if ( MaxChunkObjectsElem.isNumber() ) { - maxChunkObjects = MaxChunkObjectsElem.numberLong(); - } + // Use every 'keyCount'-th key as a split point. We add the initial key as a sentinel, to be removed + // at the end. If a key appears more times than entries allowed on a chunk, we issue a warning and + // split on the following key. + set<BSONObj> tooFrequentKeys; + splitKeys.push_back( + prettyKey(idx->keyPattern(), currKey.getOwned()).extractFields(keyPattern)); - vector<BSONObj> splitKeys; + exec->setYieldPolicy(PlanExecutor::YIELD_AUTO); + while (1) { + while (PlanExecutor::ADVANCED == state) { + currCount++; + + if (currCount > keyCount && !forceMedianSplit) { + currKey = prettyKey(idx->keyPattern(), currKey.getOwned()) + .extractFields(keyPattern); + // Do not use this split key if it is the same used in the previous split point. + if (currKey.woCompare(splitKeys.back()) == 0) { + tooFrequentKeys.insert(currKey.getOwned()); + } else { + splitKeys.push_back(currKey.getOwned()); + currCount = 0; + numChunks++; + LOG(4) << "picked a split key: " << currKey << endl; + } + } - { - // Get the size estimate for this namespace - AutoGetCollectionForRead ctx(txn, ns); - Collection* collection = ctx.getCollection(); - if ( !collection ) { - errmsg = "ns not found"; - return false; - } + // Stop if we have enough split points. + if (maxSplitPoints && (numChunks >= maxSplitPoints)) { + log() << "max number of requested split points reached (" << numChunks + << ") before the end of chunk " << ns << " " << min << " -->> " << max + << endl; + break; + } - // Allow multiKey based on the invariant that shard keys must be single-valued. - // Therefore, any multi-key index prefixed by shard key cannot be multikey over - // the shard key fields. - IndexDescriptor *idx = - collection->getIndexCatalog()->findShardKeyPrefixedIndex( txn, - keyPattern, - false ); - if ( idx == NULL ) { - errmsg = (string)"couldn't find index over splitting key " + - keyPattern.clientReadable().toString(); - return false; - } - // extend min to get (min, MinKey, MinKey, ....) - KeyPattern kp( idx->keyPattern() ); - min = Helpers::toKeyFormat( kp.extendRangeBound ( min, false ) ); - if ( max.isEmpty() ) { - // if max not specified, make it (MaxKey, Maxkey, MaxKey...) - max = Helpers::toKeyFormat( kp.extendRangeBound( max, true ) ); - } else { - // otherwise make it (max,MinKey,MinKey...) so that bound is non-inclusive - max = Helpers::toKeyFormat( kp.extendRangeBound( max, false ) ); + state = exec->getNext(&currKey, NULL); } - const long long recCount = collection->numRecords(txn); - const long long dataSize = collection->dataSize(txn); + if (!forceMedianSplit) + break; // - // 1.b Now that we have the size estimate, go over the remaining parameters and apply any maximum size - // restrictions specified there. + // If we're forcing a split at the halfway point, then the first pass was just + // to count the keys, and we still need a second pass. // - - // 'force'-ing a split is equivalent to having maxChunkSize be the size of the current chunk, i.e., the - // logic below will split that chunk in half - long long maxChunkSize = 0; - bool forceMedianSplit = false; - { - BSONElement maxSizeElem = jsobj[ "maxChunkSize" ]; - BSONElement forceElem = jsobj[ "force" ]; - - if ( forceElem.trueValue() ) { - forceMedianSplit = true; - // This chunk size is effectively ignored if force is true - maxChunkSize = dataSize; - - } - else if ( maxSizeElem.isNumber() ) { - maxChunkSize = maxSizeElem.numberLong() * 1<<20; - - } - else { - maxSizeElem = jsobj["maxChunkSizeBytes"]; - if ( maxSizeElem.isNumber() ) { - maxChunkSize = maxSizeElem.numberLong(); - } - } - // We need a maximum size for the chunk, unless we're not actually capable of finding any - // split points. - if ( maxChunkSize <= 0 && recCount != 0 ) { - errmsg = "need to specify the desired max chunk size (maxChunkSize or maxChunkSizeBytes)"; - return false; - } - } - - - // If there's not enough data for more than one chunk, no point continuing. - if ( dataSize < maxChunkSize || recCount == 0 ) { - vector<BSONObj> emptyVector; - result.append( "splitKeys" , emptyVector ); - return true; - } - - log() << "request split points lookup for chunk " << ns << " " << min << " -->> " << max << endl; - - // We'll use the average object size and number of object to find approximately how many keys - // each chunk should have. We'll split at half the maxChunkSize or maxChunkObjects, if - // provided. - const long long avgRecSize = dataSize / recCount; - long long keyCount = maxChunkSize / (2 * avgRecSize); - if ( maxChunkObjects && ( maxChunkObjects < keyCount ) ) { - log() << "limiting split vector to " << maxChunkObjects << " (from " << keyCount << ") objects " << endl; - keyCount = maxChunkObjects; - } - - // - // 2. Traverse the index and add the keyCount-th key to the result vector. If that key - // appeared in the vector before, we omit it. The invariant here is that all the - // instances of a given key value live in the same chunk. - // - - Timer timer; - long long currCount = 0; - long long numChunks = 0; - - unique_ptr<PlanExecutor> exec( - InternalPlanner::indexScan(txn, collection, idx, min, max, - false, InternalPlanner::FORWARD)); - - BSONObj currKey; - PlanExecutor::ExecState state = exec->getNext(&currKey, NULL); - if (PlanExecutor::ADVANCED != state) { - errmsg = "can't open a cursor for splitting (desired range is possibly empty)"; - return false; - } - - // Use every 'keyCount'-th key as a split point. We add the initial key as a sentinel, to be removed - // at the end. If a key appears more times than entries allowed on a chunk, we issue a warning and - // split on the following key. - set<BSONObj> tooFrequentKeys; - splitKeys.push_back(prettyKey(idx->keyPattern(), currKey.getOwned()).extractFields( keyPattern ) ); + forceMedianSplit = false; + keyCount = currCount / 2; + currCount = 0; + log() << "splitVector doing another cycle because of force, keyCount now: " + << keyCount << endl; + + exec.reset(InternalPlanner::indexScan( + txn, collection, idx, min, max, false, InternalPlanner::FORWARD)); exec->setYieldPolicy(PlanExecutor::YIELD_AUTO); - while ( 1 ) { - while (PlanExecutor::ADVANCED == state) { - currCount++; - - if ( currCount > keyCount && !forceMedianSplit ) { - currKey = prettyKey(idx->keyPattern(), currKey.getOwned()).extractFields(keyPattern); - // Do not use this split key if it is the same used in the previous split point. - if ( currKey.woCompare( splitKeys.back() ) == 0 ) { - tooFrequentKeys.insert( currKey.getOwned() ); - } - else { - splitKeys.push_back( currKey.getOwned() ); - currCount = 0; - numChunks++; - LOG(4) << "picked a split key: " << currKey << endl; - } - } + state = exec->getNext(&currKey, NULL); + } - // Stop if we have enough split points. - if ( maxSplitPoints && ( numChunks >= maxSplitPoints ) ) { - log() << "max number of requested split points reached (" << numChunks - << ") before the end of chunk " << ns << " " << min << " -->> " << max - << endl; - break; - } + // + // 3. Format the result and issue any warnings about the data we gathered while traversing the + // index + // - state = exec->getNext(&currKey, NULL); - } - - if ( ! forceMedianSplit ) - break; - - // - // If we're forcing a split at the halfway point, then the first pass was just - // to count the keys, and we still need a second pass. - // + // Warn for keys that are more numerous than maxChunkSize allows. + for (set<BSONObj>::const_iterator it = tooFrequentKeys.begin(); + it != tooFrequentKeys.end(); + ++it) { + warning() << "possible low cardinality key detected in " << ns << " - key is " + << prettyKey(idx->keyPattern(), *it) << endl; + } - forceMedianSplit = false; - keyCount = currCount / 2; - currCount = 0; - log() << "splitVector doing another cycle because of force, keyCount now: " << keyCount << endl; + // Remove the sentinel at the beginning before returning + splitKeys.erase(splitKeys.begin()); - exec.reset(InternalPlanner::indexScan(txn, collection, idx, min, max, - false, InternalPlanner::FORWARD)); + if (timer.millis() > serverGlobalParams.slowMS) { + warning() << "Finding the split vector for " << ns << " over " << keyPattern + << " keyCount: " << keyCount << " numSplits: " << splitKeys.size() + << " lookedAt: " << currCount << " took " << timer.millis() << "ms" + << endl; + } - exec->setYieldPolicy(PlanExecutor::YIELD_AUTO); - state = exec->getNext(&currKey, NULL); - } + // Warning: we are sending back an array of keys but are currently limited to + // 4MB work of 'result' size. This should be okay for now. - // - // 3. Format the result and issue any warnings about the data we gathered while traversing the - // index - // - - // Warn for keys that are more numerous than maxChunkSize allows. - for ( set<BSONObj>::const_iterator it = tooFrequentKeys.begin(); it != tooFrequentKeys.end(); ++it ) { - warning() << "possible low cardinality key detected in " << ns - << " - key is " << prettyKey(idx->keyPattern(), *it ) << endl; - } - - // Remove the sentinel at the beginning before returning - splitKeys.erase( splitKeys.begin() ); - - if (timer.millis() > serverGlobalParams.slowMS) { - warning() << "Finding the split vector for " << ns << " over "<< keyPattern - << " keyCount: " << keyCount << " numSplits: " << splitKeys.size() - << " lookedAt: " << currCount << " took " << timer.millis() << "ms" - << endl; - } - - // Warning: we are sending back an array of keys but are currently limited to - // 4MB work of 'result' size. This should be okay for now. + result.append("timeMillis", timer.millis()); + } - result.append( "timeMillis", timer.millis() ); - } + result.append("splitKeys", splitKeys); + return true; + } +} cmdSplitVector; + +class SplitChunkCommand : public Command { +public: + SplitChunkCommand() : Command("splitChunk") {} + virtual void help(stringstream& help) const { + help << "internal command usage only\n" + "example:\n" + " { splitChunk:\"db.foo\" , keyPattern: {a:1} , min : {a:100} , max: {a:200} { " + "splitKeys : [ {a:150} , ... ]}"; + } - result.append( "splitKeys" , splitKeys ); - return true; + virtual bool slaveOk() const { + return false; + } + virtual bool adminOnly() const { + return true; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::splitChunk)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); + } + return Status::OK(); + } + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + // + // 1. check whether parameters passed to splitChunk are sound + // + + const string ns = parseNs(dbname, cmdObj); + if (ns.empty()) { + errmsg = "need to specify namespace in command"; + return false; + } + const BSONObj keyPattern = cmdObj["keyPattern"].Obj(); + if (keyPattern.isEmpty()) { + errmsg = "need to specify the key pattern the collection is sharded over"; + return false; } - } cmdSplitVector; - - class SplitChunkCommand : public Command { - public: - SplitChunkCommand() : Command( "splitChunk" ) {} - virtual void help( stringstream& help ) const { - help << - "internal command usage only\n" - "example:\n" - " { splitChunk:\"db.foo\" , keyPattern: {a:1} , min : {a:100} , max: {a:200} { splitKeys : [ {a:150} , ... ]}"; - } - - virtual bool slaveOk() const { return false; } - virtual bool adminOnly() const { return true; } - virtual bool isWriteCommandForConfigServer() const { return false; } - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), - ActionType::splitChunk)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - return Status::OK(); + + const BSONObj min = cmdObj["min"].Obj(); + if (min.isEmpty()) { + errmsg = "need to specify the min key for the chunk"; + return false; } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); + + const BSONObj max = cmdObj["max"].Obj(); + if (max.isEmpty()) { + errmsg = "need to specify the max key for the chunk"; + return false; } - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - // - // 1. check whether parameters passed to splitChunk are sound - // + const string shardName = cmdObj["from"].str(); + if (shardName.empty()) { + errmsg = "need specify server to split chunk at"; + return false; + } - const string ns = parseNs(dbname, cmdObj); - if ( ns.empty() ) { - errmsg = "need to specify namespace in command"; - return false; - } + const BSONObj splitKeysElem = cmdObj["splitKeys"].Obj(); + if (splitKeysElem.isEmpty()) { + errmsg = "need to provide the split points to chunk over"; + return false; + } + vector<BSONObj> splitKeys; + BSONObjIterator it(splitKeysElem); + while (it.more()) { + splitKeys.push_back(it.next().Obj().getOwned()); + } - const BSONObj keyPattern = cmdObj["keyPattern"].Obj(); - if ( keyPattern.isEmpty() ) { - errmsg = "need to specify the key pattern the collection is sharded over"; - return false; - } + // + // Get sharding state up-to-date + // - const BSONObj min = cmdObj["min"].Obj(); - if ( min.isEmpty() ) { - errmsg = "need to specify the min key for the chunk"; + // This could be the first call that enables sharding - make sure we initialize the + // sharding state for this shard. + if (!shardingState.enabled()) { + if (cmdObj["configdb"].type() != String) { + errmsg = "sharding not enabled"; + warning() << errmsg << endl; return false; } + string configdb = cmdObj["configdb"].String(); + ShardingState::initialize(configdb); + } - const BSONObj max = cmdObj["max"].Obj(); - if ( max.isEmpty() ) { - errmsg = "need to specify the max key for the chunk"; - return false; - } + // Initialize our current shard name in the shard state if needed + shardingState.gotShardName(shardName); - const string shardName = cmdObj["from"].str(); - if ( shardName.empty() ) { - errmsg = "need specify server to split chunk at"; - return false; - } + ConnectionString configLoc = + ConnectionString::parse(shardingState.getConfigServer(), errmsg); + if (!configLoc.isValid()) { + warning() << errmsg; + return false; + } - const BSONObj splitKeysElem = cmdObj["splitKeys"].Obj(); - if ( splitKeysElem.isEmpty() ) { - errmsg = "need to provide the split points to chunk over"; - return false; - } - vector<BSONObj> splitKeys; - BSONObjIterator it( splitKeysElem ); - while ( it.more() ) { - splitKeys.push_back( it.next().Obj().getOwned() ); - } + log() << "received splitChunk request: " << cmdObj; - // - // Get sharding state up-to-date - // + // + // 2. lock the collection's metadata and get highest version for the current shard + // - // This could be the first call that enables sharding - make sure we initialize the - // sharding state for this shard. - if ( ! shardingState.enabled() ) { - if ( cmdObj["configdb"].type() != String ) { - errmsg = "sharding not enabled"; - warning() << errmsg << endl; - return false; - } - string configdb = cmdObj["configdb"].String(); - ShardingState::initialize(configdb); - } + string whyMessage(str::stream() << "splitting chunk [" << minKey << ", " << maxKey + << ") in " << ns); + auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock(ns, whyMessage); + + if (!scopedDistLock.isOK()) { + errmsg = str::stream() << "could not acquire collection lock for " << ns + << " to split chunk [" << minKey << "," << maxKey << ")" + << causedBy(scopedDistLock.getStatus()); + warning() << errmsg; + return false; + } - // Initialize our current shard name in the shard state if needed - shardingState.gotShardName(shardName); + // Always check our version remotely + ChunkVersion shardVersion; + Status refreshStatus = shardingState.refreshMetadataNow(txn, ns, &shardVersion); - ConnectionString configLoc = ConnectionString::parse(shardingState.getConfigServer(), - errmsg); - if (!configLoc.isValid()) { - warning() << errmsg; - return false; - } + if (!refreshStatus.isOK()) { + errmsg = str::stream() << "splitChunk cannot split chunk " + << "[" << minKey << "," << maxKey << ")" + << causedBy(refreshStatus.reason()); - log() << "received splitChunk request: " << cmdObj; + warning() << errmsg; + return false; + } - // - // 2. lock the collection's metadata and get highest version for the current shard - // + if (shardVersion.majorVersion() == 0) { + // It makes no sense to split if our version is zero and we have no chunks + errmsg = str::stream() << "splitChunk cannot split chunk " + << "[" << minKey << "," << maxKey << ")" + << " with zero shard version"; - string whyMessage(str::stream() << "splitting chunk [" << minKey - << ", " << maxKey - << ") in " << ns); - auto scopedDistLock = grid.catalogManager()->getDistLockManager()->lock( - ns, whyMessage); + warning() << errmsg; + return false; + } - if (!scopedDistLock.isOK()) { - errmsg = str::stream() << "could not acquire collection lock for " << ns - << " to split chunk [" << minKey << "," << maxKey << ")" - << causedBy(scopedDistLock.getStatus()); + // From mongos >= v3.0. + BSONElement epochElem(cmdObj["epoch"]); + if (epochElem.type() == jstOID) { + OID cmdEpoch = epochElem.OID(); + + if (cmdEpoch != shardVersion.epoch()) { + errmsg = str::stream() << "splitChunk cannot split chunk " + << "[" << minKey << "," << maxKey << "), " + << "collection may have been dropped. " + << "current epoch: " << shardVersion.epoch() + << ", cmd epoch: " << cmdEpoch; warning() << errmsg; return false; } + } - // Always check our version remotely - ChunkVersion shardVersion; - Status refreshStatus = shardingState.refreshMetadataNow(txn, ns, &shardVersion); + // Get collection metadata + const CollectionMetadataPtr collMetadata(shardingState.getCollectionMetadata(ns)); + // With nonzero shard version, we must have metadata + invariant(NULL != collMetadata); - if (!refreshStatus.isOK()) { + ChunkVersion collVersion = collMetadata->getCollVersion(); + // With nonzero shard version, we must have a coll version >= our shard version + invariant(collVersion >= shardVersion); - errmsg = str::stream() << "splitChunk cannot split chunk " << "[" << minKey << "," - << maxKey << ")" << causedBy(refreshStatus.reason()); + ChunkType origChunk; + if (!collMetadata->getNextChunk(min, &origChunk) || origChunk.getMin().woCompare(min) || + origChunk.getMax().woCompare(max)) { + // Our boundaries are different from those passed in + errmsg = str::stream() << "splitChunk cannot find chunk " + << "[" << minKey << "," << maxKey << ")" + << " to split, the chunk boundaries may be stale"; - warning() << errmsg; - return false; - } + warning() << errmsg; + return false; + } - if (shardVersion.majorVersion() == 0) { + log() << "splitChunk accepted at version " << shardVersion; - // It makes no sense to split if our version is zero and we have no chunks - errmsg = str::stream() << "splitChunk cannot split chunk " << "[" << minKey << "," - << maxKey << ")" << " with zero shard version"; + // + // 3. create the batch of updates to metadata ( the new chunks ) to be applied via 'applyOps' command + // - warning() << errmsg; - return false; - } + BSONObjBuilder logDetail; + appendShortVersion(logDetail.subobjStart("before"), origChunk); + LOG(1) << "before split on " << origChunk << endl; + OwnedPointerVector<ChunkType> newChunks; - // From mongos >= v3.0. - BSONElement epochElem(cmdObj["epoch"]); - if (epochElem.type() == jstOID) { - OID cmdEpoch = epochElem.OID(); - - if (cmdEpoch != shardVersion.epoch()) { - errmsg = str::stream() << "splitChunk cannot split chunk " - << "[" << minKey << "," - << maxKey << "), " - << "collection may have been dropped. " - << "current epoch: " << shardVersion.epoch() - << ", cmd epoch: " << cmdEpoch; - warning() << errmsg; - return false; - } - } + ChunkVersion nextChunkVersion = collVersion; + BSONObj startKey = min; + splitKeys.push_back(max); // makes it easier to have 'max' in the next loop. remove later. - // Get collection metadata - const CollectionMetadataPtr collMetadata(shardingState.getCollectionMetadata(ns)); - // With nonzero shard version, we must have metadata - invariant(NULL != collMetadata); + BSONArrayBuilder updates; - ChunkVersion collVersion = collMetadata->getCollVersion(); - // With nonzero shard version, we must have a coll version >= our shard version - invariant(collVersion >= shardVersion); + for (vector<BSONObj>::const_iterator it = splitKeys.begin(); it != splitKeys.end(); ++it) { + BSONObj endKey = *it; - ChunkType origChunk; - if (!collMetadata->getNextChunk(min, &origChunk) - || origChunk.getMin().woCompare(min) || origChunk.getMax().woCompare(max)) { + if (endKey.woCompare(startKey) == 0) { + errmsg = str::stream() << "split on the lower bound of chunk " + << "[" << min << ", " << max << ")" + << " is not allowed"; - // Our boundaries are different from those passed in - errmsg = str::stream() << "splitChunk cannot find chunk " - << "[" << minKey << "," << maxKey << ")" - << " to split, the chunk boundaries may be stale"; + warning() << errmsg << endl; + return false; + } + // Make sure splits don't create too-big shard keys + Status status = ShardKeyPattern::checkShardKeySize(endKey); + if (!status.isOK()) { + errmsg = status.reason(); warning() << errmsg; return false; } - log() << "splitChunk accepted at version " << shardVersion; + // splits only update the 'minor' portion of version + nextChunkVersion.incMinor(); - // - // 3. create the batch of updates to metadata ( the new chunks ) to be applied via 'applyOps' command - // + // build an update operation against the chunks collection of the config database with + // upsert true + BSONObjBuilder op; + op.append("op", "u"); + op.appendBool("b", true); + op.append("ns", ChunkType::ConfigNS); - BSONObjBuilder logDetail; - appendShortVersion(logDetail.subobjStart("before"), origChunk); - LOG(1) << "before split on " << origChunk << endl; - OwnedPointerVector<ChunkType> newChunks; + // add the modified (new) chunk information as the update object + BSONObjBuilder n(op.subobjStart("o")); + n.append(ChunkType::name(), Chunk::genID(ns, startKey)); + nextChunkVersion.addToBSON(n, ChunkType::DEPRECATED_lastmod()); + n.append(ChunkType::ns(), ns); + n.append(ChunkType::min(), startKey); + n.append(ChunkType::max(), endKey); + n.append(ChunkType::shard(), shardName); + n.done(); - ChunkVersion nextChunkVersion = collVersion; - BSONObj startKey = min; - splitKeys.push_back( max ); // makes it easier to have 'max' in the next loop. remove later. + // add the chunk's _id as the query part of the update statement + BSONObjBuilder q(op.subobjStart("o2")); + q.append(ChunkType::name(), Chunk::genID(ns, startKey)); + q.done(); - BSONArrayBuilder updates; + updates.append(op.obj()); - for ( vector<BSONObj>::const_iterator it = splitKeys.begin(); it != splitKeys.end(); ++it ) { - BSONObj endKey = *it; + // remember this chunk info for logging later + unique_ptr<ChunkType> chunk(new ChunkType()); + chunk->setMin(startKey); + chunk->setMax(endKey); + chunk->setVersion(nextChunkVersion); - if ( endKey.woCompare( startKey ) == 0) { - errmsg = str::stream() << "split on the lower bound of chunk " - << "[" << min << ", " << max << ")" - << " is not allowed"; + newChunks.push_back(chunk.release()); - warning() << errmsg << endl; - return false; - } + startKey = endKey; + } - // Make sure splits don't create too-big shard keys - Status status = ShardKeyPattern::checkShardKeySize(endKey); - if (!status.isOK()) { - errmsg = status.reason(); - warning() << errmsg; - return false; - } + splitKeys.pop_back(); // 'max' was used as sentinel - // splits only update the 'minor' portion of version - nextChunkVersion.incMinor(); - - // build an update operation against the chunks collection of the config database with - // upsert true - BSONObjBuilder op; - op.append( "op" , "u" ); - op.appendBool( "b" , true ); - op.append( "ns" , ChunkType::ConfigNS ); - - // add the modified (new) chunk information as the update object - BSONObjBuilder n( op.subobjStart( "o" ) ); - n.append(ChunkType::name(), Chunk::genID(ns, startKey)); - nextChunkVersion.addToBSON(n, ChunkType::DEPRECATED_lastmod()); - n.append(ChunkType::ns(), ns); - n.append(ChunkType::min(), startKey); - n.append(ChunkType::max(), endKey); - n.append(ChunkType::shard(), shardName); - n.done(); - - // add the chunk's _id as the query part of the update statement - BSONObjBuilder q( op.subobjStart( "o2" ) ); - q.append(ChunkType::name(), Chunk::genID(ns, startKey)); - q.done(); - - updates.append( op.obj() ); - - // remember this chunk info for logging later - unique_ptr<ChunkType> chunk(new ChunkType()); - chunk->setMin(startKey); - chunk->setMax(endKey); - chunk->setVersion(nextChunkVersion); - - newChunks.push_back(chunk.release()); - - startKey = endKey; + BSONArrayBuilder preCond; + { + BSONObjBuilder b; + b.append("ns", ChunkType::ConfigNS); + b.append("q", + BSON("query" << BSON(ChunkType::ns(ns)) << "orderby" + << BSON(ChunkType::DEPRECATED_lastmod() << -1))); + { + BSONObjBuilder bb(b.subobjStart("res")); + // TODO: For backwards compatibility, we can't yet require an epoch here + bb.appendTimestamp(ChunkType::DEPRECATED_lastmod(), collVersion.toLong()); + bb.done(); } + preCond.append(b.obj()); + } - splitKeys.pop_back(); // 'max' was used as sentinel + // + // 4. apply the batch of updates to remote and local metadata + // + Status applyOpsStatus = + grid.catalogManager()->applyChunkOpsDeprecated(updates.arr(), preCond.arr()); + if (!applyOpsStatus.isOK()) { + return appendCommandStatus(result, applyOpsStatus); + } - BSONArrayBuilder preCond; - { - BSONObjBuilder b; - b.append("ns", ChunkType::ConfigNS); - b.append("q", BSON("query" << BSON(ChunkType::ns(ns)) << - "orderby" << BSON(ChunkType::DEPRECATED_lastmod() << -1))); - { - BSONObjBuilder bb( b.subobjStart( "res" ) ); - // TODO: For backwards compatibility, we can't yet require an epoch here - bb.appendTimestamp(ChunkType::DEPRECATED_lastmod(), collVersion.toLong()); - bb.done(); - } - preCond.append( b.obj() ); - } + // + // Install chunk metadata with knowledge about newly split chunks in this shard's state + // - // - // 4. apply the batch of updates to remote and local metadata - // - Status applyOpsStatus = grid.catalogManager()->applyChunkOpsDeprecated(updates.arr(), - preCond.arr()); - if (!applyOpsStatus.isOK()) { - return appendCommandStatus(result, applyOpsStatus); - } + { + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock writeLk(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); + Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); - // - // Install chunk metadata with knowledge about newly split chunks in this shard's state - // - - { - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock writeLk(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); - Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); + // NOTE: The newShardVersion resulting from this split is higher than any + // other chunk version, so it's also implicitly the newCollVersion + ChunkVersion newShardVersion = collVersion; - // NOTE: The newShardVersion resulting from this split is higher than any - // other chunk version, so it's also implicitly the newCollVersion - ChunkVersion newShardVersion = collVersion; + // Increment the minor version once, shardingState.splitChunk increments once + // per split point (resulting in the correct final shard/collection version) + // TODO: Revisit this interface, it's a bit clunky + newShardVersion.incMinor(); - // Increment the minor version once, shardingState.splitChunk increments once - // per split point (resulting in the correct final shard/collection version) - // TODO: Revisit this interface, it's a bit clunky - newShardVersion.incMinor(); + shardingState.splitChunk(txn, ns, min, max, splitKeys, newShardVersion); + } - shardingState.splitChunk(txn, ns, min, max, splitKeys, newShardVersion); - } + // + // 5. logChanges + // - // - // 5. logChanges - // + // single splits are logged different than multisplits + if (newChunks.size() == 2) { + appendShortVersion(logDetail.subobjStart("left"), *newChunks[0]); + appendShortVersion(logDetail.subobjStart("right"), *newChunks[1]); - // single splits are logged different than multisplits - if ( newChunks.size() == 2 ) { - appendShortVersion(logDetail.subobjStart("left"), *newChunks[0]); - appendShortVersion(logDetail.subobjStart("right"), *newChunks[1]); + grid.catalogManager()->logChange(txn, "split", ns, logDetail.obj()); + } else { + BSONObj beforeDetailObj = logDetail.obj(); + BSONObj firstDetailObj = beforeDetailObj.getOwned(); + const int newChunksSize = newChunks.size(); - grid.catalogManager()->logChange(txn, "split", ns, logDetail.obj()); - } - else { - BSONObj beforeDetailObj = logDetail.obj(); - BSONObj firstDetailObj = beforeDetailObj.getOwned(); - const int newChunksSize = newChunks.size(); - - for ( int i=0; i < newChunksSize; i++ ) { - BSONObjBuilder chunkDetail; - chunkDetail.appendElements( beforeDetailObj ); - chunkDetail.append( "number", i+1 ); - chunkDetail.append( "of" , newChunksSize ); - appendShortVersion(chunkDetail.subobjStart("chunk"), *newChunks[i]); - - grid.catalogManager()->logChange(txn, "multi-split", ns, chunkDetail.obj()); - } + for (int i = 0; i < newChunksSize; i++) { + BSONObjBuilder chunkDetail; + chunkDetail.appendElements(beforeDetailObj); + chunkDetail.append("number", i + 1); + chunkDetail.append("of", newChunksSize); + appendShortVersion(chunkDetail.subobjStart("chunk"), *newChunks[i]); + + grid.catalogManager()->logChange(txn, "multi-split", ns, chunkDetail.obj()); } + } - dassert(newChunks.size() > 1); + dassert(newChunks.size() > 1); - { - // Select chunk to move out for "top chunk optimization". - KeyPattern shardKeyPattern(collMetadata->getKeyPattern()); - - AutoGetCollectionForRead ctx(txn, ns); - Collection* collection = ctx.getCollection(); - if (!collection) { - warning() << "will not perform top-chunk checking since " << ns - << " does not exist after splitting"; - return true; - } + { + // Select chunk to move out for "top chunk optimization". + KeyPattern shardKeyPattern(collMetadata->getKeyPattern()); - // Allow multiKey based on the invariant that shard keys must be - // single-valued. Therefore, any multi-key index prefixed by shard - // key cannot be multikey over the shard key fields. - IndexDescriptor *idx = - collection->getIndexCatalog()->findShardKeyPrefixedIndex(txn, - keyPattern, - false); - - if (idx == NULL) { - return true; - } + AutoGetCollectionForRead ctx(txn, ns); + Collection* collection = ctx.getCollection(); + if (!collection) { + warning() << "will not perform top-chunk checking since " << ns + << " does not exist after splitting"; + return true; + } - const ChunkType* backChunk = newChunks.vector().back(); - const ChunkType* frontChunk = newChunks.vector().front(); + // Allow multiKey based on the invariant that shard keys must be + // single-valued. Therefore, any multi-key index prefixed by shard + // key cannot be multikey over the shard key fields. + IndexDescriptor* idx = + collection->getIndexCatalog()->findShardKeyPrefixedIndex(txn, keyPattern, false); - if (shardKeyPattern.globalMax().woCompare(backChunk->getMax()) == 0 && - checkIfSingleDoc(txn, collection, idx, backChunk)) { - result.append("shouldMigrate", - BSON("min" << backChunk->getMin() - << "max" << backChunk->getMax())); - } - else if (shardKeyPattern.globalMin().woCompare(frontChunk->getMin()) == 0 && - checkIfSingleDoc(txn, collection, idx, frontChunk)) { - result.append("shouldMigrate", - BSON("min" << frontChunk->getMin() - << "max" << frontChunk->getMax())); - } + if (idx == NULL) { + return true; } - return true; + const ChunkType* backChunk = newChunks.vector().back(); + const ChunkType* frontChunk = newChunks.vector().front(); + + if (shardKeyPattern.globalMax().woCompare(backChunk->getMax()) == 0 && + checkIfSingleDoc(txn, collection, idx, backChunk)) { + result.append("shouldMigrate", + BSON("min" << backChunk->getMin() << "max" << backChunk->getMax())); + } else if (shardKeyPattern.globalMin().woCompare(frontChunk->getMin()) == 0 && + checkIfSingleDoc(txn, collection, idx, frontChunk)) { + result.append("shouldMigrate", + BSON("min" << frontChunk->getMin() << "max" << frontChunk->getMax())); + } } - private: + return true; + } - /** - * Append min, max and version information from chunk to the buffer. - */ - static void appendShortVersion(BufBuilder& b, const ChunkType& chunk) { - BSONObjBuilder bb(b); - bb.append(ChunkType::min(), chunk.getMin()); - bb.append(ChunkType::max(), chunk.getMax()); - if (chunk.isVersionSet()) - chunk.getVersion().addToBSON(bb, ChunkType::DEPRECATED_lastmod()); - bb.done(); - } +private: + /** + * Append min, max and version information from chunk to the buffer. + */ + static void appendShortVersion(BufBuilder& b, const ChunkType& chunk) { + BSONObjBuilder bb(b); + bb.append(ChunkType::min(), chunk.getMin()); + bb.append(ChunkType::max(), chunk.getMax()); + if (chunk.isVersionSet()) + chunk.getVersion().addToBSON(bb, ChunkType::DEPRECATED_lastmod()); + bb.done(); + } - static bool checkIfSingleDoc(OperationContext* txn, - Collection* collection, - const IndexDescriptor* idx, - const ChunkType* chunk) { - KeyPattern kp(idx->keyPattern()); - BSONObj newmin = Helpers::toKeyFormat(kp.extendRangeBound(chunk->getMin(), false)); - BSONObj newmax = Helpers::toKeyFormat(kp.extendRangeBound(chunk->getMax(), true)); + static bool checkIfSingleDoc(OperationContext* txn, + Collection* collection, + const IndexDescriptor* idx, + const ChunkType* chunk) { + KeyPattern kp(idx->keyPattern()); + BSONObj newmin = Helpers::toKeyFormat(kp.extendRangeBound(chunk->getMin(), false)); + BSONObj newmax = Helpers::toKeyFormat(kp.extendRangeBound(chunk->getMax(), true)); - unique_ptr<PlanExecutor> exec( - InternalPlanner::indexScan(txn, collection, idx, newmin, newmax, false)); + unique_ptr<PlanExecutor> exec( + InternalPlanner::indexScan(txn, collection, idx, newmin, newmax, false)); - // check if exactly one document found - if (PlanExecutor::ADVANCED == exec->getNext(NULL, NULL)) { - if (PlanExecutor::IS_EOF == exec->getNext(NULL, NULL)) { - return true; - } + // check if exactly one document found + if (PlanExecutor::ADVANCED == exec->getNext(NULL, NULL)) { + if (PlanExecutor::IS_EOF == exec->getNext(NULL, NULL)) { + return true; } - - return false; } - } cmdSplitChunk; + return false; + } + +} cmdSplitChunk; } // namespace mongo diff --git a/src/mongo/s/d_state.cpp b/src/mongo/s/d_state.cpp index e7341e65093..3c7739bc804 100644 --- a/src/mongo/s/d_state.cpp +++ b/src/mongo/s/d_state.cpp @@ -72,1390 +72,1343 @@ namespace mongo { - using boost::optional; - using std::endl; - using std::string; - using std::stringstream; - using std::vector; +using boost::optional; +using std::endl; +using std::string; +using std::stringstream; +using std::vector; + +namespace { +const auto clientSCI = Client::declareDecoration<optional<ShardedConnectionInfo>>(); +} // namespace + +bool isMongos() { + return false; +} - namespace { - const auto clientSCI = Client::declareDecoration<optional<ShardedConnectionInfo>>(); - } // namespace - bool isMongos() { - return false; - } +// -----ShardingState START ---- +ShardingState::ShardingState() + : _enabled(false), + _configServerTickets(3 /* max number of concurrent config server refresh threads */) {} - // -----ShardingState START ---- +bool ShardingState::enabled() { + stdx::lock_guard<stdx::mutex> lk(_mutex); + return _enabled; +} - ShardingState::ShardingState() - : _enabled(false), - _configServerTickets( 3 /* max number of concurrent config server refresh threads */ ) { - } +string ShardingState::getConfigServer() { + stdx::lock_guard<stdx::mutex> lk(_mutex); + invariant(_enabled); - bool ShardingState::enabled() { - stdx::lock_guard<stdx::mutex> lk(_mutex); - return _enabled; - } + return grid.catalogManager()->connectionString().toString(); +} - string ShardingState::getConfigServer() { - stdx::lock_guard<stdx::mutex> lk(_mutex); - invariant(_enabled); +void ShardingState::initialize(const string& server) { + uassert(18509, + "Unable to obtain host name during sharding initialization.", + !getHostName().empty()); - return grid.catalogManager()->connectionString().toString(); - } + shardingState._initialize(server); +} - void ShardingState::initialize(const string& server) { - uassert(18509, - "Unable to obtain host name during sharding initialization.", - !getHostName().empty()); +// TODO: Consolidate and eliminate these various ways of setting / validating shard names +bool ShardingState::setShardName(const string& name) { + return setShardNameAndHost(name, ""); +} - shardingState._initialize(server); - } +std::string ShardingState::getShardName() { + stdx::lock_guard<stdx::mutex> lk(_mutex); + return _shardName; +} - // TODO: Consolidate and eliminate these various ways of setting / validating shard names - bool ShardingState::setShardName( const string& name ) { - return setShardNameAndHost( name, "" ); - } +bool ShardingState::setShardNameAndHost(const string& name, const string& host) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + if (_shardName.size() == 0) { + // TODO SERVER-2299 remotely verify the name is sound w.r.t IPs + _shardName = name; - std::string ShardingState::getShardName() { - stdx::lock_guard<stdx::mutex> lk(_mutex); - return _shardName; - } + string clientAddr = cc().clientAddress(true); - bool ShardingState::setShardNameAndHost( const string& name, const string& host ) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - if ( _shardName.size() == 0 ) { - // TODO SERVER-2299 remotely verify the name is sound w.r.t IPs - _shardName = name; + log() << "remote client " << clientAddr << " initialized this host " + << (host.empty() ? string("") : string("(") + host + ") ") << "as shard " << name; - string clientAddr = cc().clientAddress(true); + return true; + } - log() << "remote client " << clientAddr << " initialized this host " - << ( host.empty() ? string( "" ) : string( "(" ) + host + ") " ) - << "as shard " << name; + if (_shardName == name) + return true; - return true; - } + string clientAddr = cc().clientAddress(true); - if ( _shardName == name ) - return true; + warning() << "remote client " << clientAddr << " tried to initialize this host " + << (host.empty() ? string("") : string("(") + host + ") ") << "as shard " << name + << ", but shard name was previously initialized as " << _shardName; - string clientAddr = cc().clientAddress(true); + return false; +} - warning() << "remote client " << clientAddr << " tried to initialize this host " - << ( host.empty() ? string( "" ) : string( "(" ) + host + ") " ) - << "as shard " << name - << ", but shard name was previously initialized as " << _shardName; +void ShardingState::gotShardName(const string& name) { + gotShardNameAndHost(name, ""); +} - return false; - } +void ShardingState::gotShardNameAndHost(const string& name, const string& host) { + if (setShardNameAndHost(name, host)) + return; - void ShardingState::gotShardName( const string& name ) { - gotShardNameAndHost( name, "" ); - } + string clientAddr = cc().clientAddress(true); + stringstream ss; - void ShardingState::gotShardNameAndHost( const string& name, const string& host ) { - if ( setShardNameAndHost( name, host ) ) - return; + // Same error as above, to match for reporting + ss << "remote client " << clientAddr << " tried to initialize this host " + << (host.empty() ? string("") : string("(") + host + ") ") << "as shard " << name + << ", but shard name was previously initialized as " << _shardName; - string clientAddr = cc().clientAddress(true); - stringstream ss; + msgasserted(13298, ss.str()); +} - // Same error as above, to match for reporting - ss << "remote client " << clientAddr << " tried to initialize this host " - << ( host.empty() ? string( "" ) : string( "(" ) + host + ") " ) - << "as shard " << name - << ", but shard name was previously initialized as " << _shardName; +void ShardingState::clearCollectionMetadata() { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _collMetadata.clear(); +} - msgasserted( 13298 , ss.str() ); - } +// TODO we shouldn't need three ways for checking the version. Fix this. +bool ShardingState::hasVersion(const string& ns) { + stdx::lock_guard<stdx::mutex> lk(_mutex); - void ShardingState::clearCollectionMetadata() { - stdx::lock_guard<stdx::mutex> lk(_mutex); - _collMetadata.clear(); - } + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + return it != _collMetadata.end(); +} - // TODO we shouldn't need three ways for checking the version. Fix this. - bool ShardingState::hasVersion( const string& ns ) { - stdx::lock_guard<stdx::mutex> lk(_mutex); +bool ShardingState::hasVersion(const string& ns, ChunkVersion& version) { + stdx::lock_guard<stdx::mutex> lk(_mutex); - CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); - return it != _collMetadata.end(); - } + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + if (it == _collMetadata.end()) + return false; - bool ShardingState::hasVersion( const string& ns , ChunkVersion& version ) { - stdx::lock_guard<stdx::mutex> lk(_mutex); + CollectionMetadataPtr p = it->second; + version = p->getShardVersion(); + return true; +} - CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); - if ( it == _collMetadata.end() ) - return false; +ChunkVersion ShardingState::getVersion(const string& ns) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + if (it != _collMetadata.end()) { CollectionMetadataPtr p = it->second; - version = p->getShardVersion(); - return true; + return p->getShardVersion(); + } else { + return ChunkVersion(0, 0, OID()); } +} - ChunkVersion ShardingState::getVersion(const string& ns) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - - CollectionMetadataMap::const_iterator it = _collMetadata.find( ns ); - if ( it != _collMetadata.end() ) { - CollectionMetadataPtr p = it->second; - return p->getShardVersion(); - } - else { - return ChunkVersion( 0, 0, OID() ); - } - } +void ShardingState::donateChunk(OperationContext* txn, + const string& ns, + const BSONObj& min, + const BSONObj& max, + ChunkVersion version) { + invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); + stdx::lock_guard<stdx::mutex> lk(_mutex); + + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + verify(it != _collMetadata.end()); + CollectionMetadataPtr p = it->second; + + // empty shards should have version 0 + version = (p->getNumChunks() > 1) ? version : ChunkVersion(0, 0, p->getCollVersion().epoch()); + + ChunkType chunk; + chunk.setMin(min); + chunk.setMax(max); + string errMsg; + + CollectionMetadataPtr cloned(p->cloneMigrate(chunk, version, &errMsg)); + // uassert to match old behavior, TODO: report errors w/o throwing + uassert(16855, errMsg, NULL != cloned.get()); + + // TODO: a bit dangerous to have two different zero-version states - no-metadata and + // no-version + _collMetadata[ns] = cloned; +} - void ShardingState::donateChunk(OperationContext* txn, +void ShardingState::undoDonateChunk(OperationContext* txn, const string& ns, - const BSONObj& min, - const BSONObj& max, - ChunkVersion version) { + CollectionMetadataPtr prevMetadata) { + invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); + stdx::lock_guard<stdx::mutex> lk(_mutex); - invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); - stdx::lock_guard<stdx::mutex> lk( _mutex ); - - CollectionMetadataMap::const_iterator it = _collMetadata.find( ns ); - verify( it != _collMetadata.end() ) ; - CollectionMetadataPtr p = it->second; + log() << "ShardingState::undoDonateChunk acquired _mutex" << endl; - // empty shards should have version 0 - version = - ( p->getNumChunks() > 1 ) ? - version : ChunkVersion( 0, 0, p->getCollVersion().epoch() ); - - ChunkType chunk; - chunk.setMin( min ); - chunk.setMax( max ); - string errMsg; + CollectionMetadataMap::iterator it = _collMetadata.find(ns); + verify(it != _collMetadata.end()); + it->second = prevMetadata; +} - CollectionMetadataPtr cloned( p->cloneMigrate( chunk, version, &errMsg ) ); - // uassert to match old behavior, TODO: report errors w/o throwing - uassert( 16855, errMsg, NULL != cloned.get() ); +bool ShardingState::notePending(OperationContext* txn, + const string& ns, + const BSONObj& min, + const BSONObj& max, + const OID& epoch, + string* errMsg) { + invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); + stdx::lock_guard<stdx::mutex> lk(_mutex); + + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + if (it == _collMetadata.end()) { + *errMsg = str::stream() << "could not note chunk " + << "[" << min << "," << max << ")" + << " as pending because the local metadata for " << ns + << " has changed"; - // TODO: a bit dangerous to have two different zero-version states - no-metadata and - // no-version - _collMetadata[ns] = cloned; + return false; } - void ShardingState::undoDonateChunk(OperationContext* txn, - const string& ns, - CollectionMetadataPtr prevMetadata) { - - invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); - stdx::lock_guard<stdx::mutex> lk( _mutex ); + CollectionMetadataPtr metadata = it->second; - log() << "ShardingState::undoDonateChunk acquired _mutex" << endl; + // This can currently happen because drops aren't synchronized with in-migrations + // The idea for checking this here is that in the future we shouldn't have this problem + if (metadata->getCollVersion().epoch() != epoch) { + *errMsg = str::stream() << "could not note chunk " + << "[" << min << "," << max << ")" + << " as pending because the epoch for " << ns + << " has changed from " << epoch << " to " + << metadata->getCollVersion().epoch(); - CollectionMetadataMap::iterator it = _collMetadata.find( ns ); - verify( it != _collMetadata.end() ); - it->second = prevMetadata; + return false; } - bool ShardingState::notePending(OperationContext* txn, - const string& ns, - const BSONObj& min, - const BSONObj& max, - const OID& epoch, - string* errMsg ) { + ChunkType chunk; + chunk.setMin(min); + chunk.setMax(max); - invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); - stdx::lock_guard<stdx::mutex> lk( _mutex ); + CollectionMetadataPtr cloned(metadata->clonePlusPending(chunk, errMsg)); + if (!cloned) + return false; - CollectionMetadataMap::const_iterator it = _collMetadata.find( ns ); - if ( it == _collMetadata.end() ) { + _collMetadata[ns] = cloned; + return true; +} - *errMsg = str::stream() << "could not note chunk " << "[" << min << "," << max << ")" - << " as pending because the local metadata for " << ns - << " has changed"; +bool ShardingState::forgetPending(OperationContext* txn, + const string& ns, + const BSONObj& min, + const BSONObj& max, + const OID& epoch, + string* errMsg) { + invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); + stdx::lock_guard<stdx::mutex> lk(_mutex); + + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + if (it == _collMetadata.end()) { + *errMsg = str::stream() << "no need to forget pending chunk " + << "[" << min << "," << max << ")" + << " because the local metadata for " << ns << " has changed"; - return false; - } + return false; + } - CollectionMetadataPtr metadata = it->second; + CollectionMetadataPtr metadata = it->second; - // This can currently happen because drops aren't synchronized with in-migrations - // The idea for checking this here is that in the future we shouldn't have this problem - if ( metadata->getCollVersion().epoch() != epoch ) { + // This can currently happen because drops aren't synchronized with in-migrations + // The idea for checking this here is that in the future we shouldn't have this problem + if (metadata->getCollVersion().epoch() != epoch) { + *errMsg = str::stream() << "no need to forget pending chunk " + << "[" << min << "," << max << ")" + << " because the epoch for " << ns << " has changed from " << epoch + << " to " << metadata->getCollVersion().epoch(); - *errMsg = str::stream() << "could not note chunk " << "[" << min << "," << max << ")" - << " as pending because the epoch for " << ns - << " has changed from " - << epoch << " to " << metadata->getCollVersion().epoch(); + return false; + } - return false; - } + ChunkType chunk; + chunk.setMin(min); + chunk.setMax(max); - ChunkType chunk; - chunk.setMin( min ); - chunk.setMax( max ); + CollectionMetadataPtr cloned(metadata->cloneMinusPending(chunk, errMsg)); + if (!cloned) + return false; - CollectionMetadataPtr cloned( metadata->clonePlusPending( chunk, errMsg ) ); - if ( !cloned ) return false; + _collMetadata[ns] = cloned; + return true; +} - _collMetadata[ns] = cloned; - return true; - } +void ShardingState::splitChunk(OperationContext* txn, + const string& ns, + const BSONObj& min, + const BSONObj& max, + const vector<BSONObj>& splitKeys, + ChunkVersion version) { + invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); + stdx::lock_guard<stdx::mutex> lk(_mutex); + + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + verify(it != _collMetadata.end()); + + ChunkType chunk; + chunk.setMin(min); + chunk.setMax(max); + string errMsg; + + CollectionMetadataPtr cloned(it->second->cloneSplit(chunk, splitKeys, version, &errMsg)); + // uassert to match old behavior, TODO: report errors w/o throwing + uassert(16857, errMsg, NULL != cloned.get()); + + _collMetadata[ns] = cloned; +} - bool ShardingState::forgetPending(OperationContext* txn, - const string& ns, - const BSONObj& min, - const BSONObj& max, - const OID& epoch, - string* errMsg ) { +void ShardingState::mergeChunks(OperationContext* txn, + const string& ns, + const BSONObj& minKey, + const BSONObj& maxKey, + ChunkVersion mergedVersion) { + invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); + stdx::lock_guard<stdx::mutex> lk(_mutex); - invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); - stdx::lock_guard<stdx::mutex> lk( _mutex ); + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + verify(it != _collMetadata.end()); - CollectionMetadataMap::const_iterator it = _collMetadata.find( ns ); - if ( it == _collMetadata.end() ) { + string errMsg; - *errMsg = str::stream() << "no need to forget pending chunk " - << "[" << min << "," << max << ")" - << " because the local metadata for " << ns << " has changed"; + CollectionMetadataPtr cloned(it->second->cloneMerge(minKey, maxKey, mergedVersion, &errMsg)); + // uassert to match old behavior, TODO: report errors w/o throwing + uassert(17004, errMsg, NULL != cloned.get()); - return false; - } + _collMetadata[ns] = cloned; +} - CollectionMetadataPtr metadata = it->second; +void ShardingState::resetMetadata(const string& ns) { + stdx::lock_guard<stdx::mutex> lk(_mutex); - // This can currently happen because drops aren't synchronized with in-migrations - // The idea for checking this here is that in the future we shouldn't have this problem - if ( metadata->getCollVersion().epoch() != epoch ) { + warning() << "resetting metadata for " << ns << ", this should only be used in testing" << endl; - *errMsg = str::stream() << "no need to forget pending chunk " - << "[" << min << "," << max << ")" - << " because the epoch for " << ns << " has changed from " - << epoch << " to " << metadata->getCollVersion().epoch(); + _collMetadata.erase(ns); +} - return false; - } +Status ShardingState::refreshMetadataIfNeeded(OperationContext* txn, + const string& ns, + const ChunkVersion& reqShardVersion, + ChunkVersion* latestShardVersion) { + // The _configServerTickets serializes this process such that only a small number of threads + // can try to refresh at the same time. - ChunkType chunk; - chunk.setMin( min ); - chunk.setMax( max ); + LOG(2) << "metadata refresh requested for " << ns << " at shard version " << reqShardVersion + << endl; - CollectionMetadataPtr cloned( metadata->cloneMinusPending( chunk, errMsg ) ); - if ( !cloned ) return false; + // + // Queuing of refresh requests starts here when remote reload is needed. This may take time. + // TODO: Explicitly expose the queuing discipline. + // - _collMetadata[ns] = cloned; - return true; - } + _configServerTickets.waitForTicket(); + TicketHolderReleaser needTicketFrom(&_configServerTickets); - void ShardingState::splitChunk(OperationContext* txn, - const string& ns, - const BSONObj& min, - const BSONObj& max, - const vector<BSONObj>& splitKeys, - ChunkVersion version ) { + // + // Fast path - check if the requested version is at a higher version than the current + // metadata version or a different epoch before verifying against config server. + // - invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); - stdx::lock_guard<stdx::mutex> lk( _mutex ); + CollectionMetadataPtr storedMetadata; + { + stdx::lock_guard<stdx::mutex> lk(_mutex); + CollectionMetadataMap::iterator it = _collMetadata.find(ns); + if (it != _collMetadata.end()) + storedMetadata = it->second; + } + ChunkVersion storedShardVersion; + if (storedMetadata) + storedShardVersion = storedMetadata->getShardVersion(); + *latestShardVersion = storedShardVersion; + + if (storedShardVersion >= reqShardVersion && + storedShardVersion.epoch() == reqShardVersion.epoch()) { + // Don't need to remotely reload if we're in the same epoch with a >= version + return Status::OK(); + } + + // + // Slow path - remotely reload + // + // Cases: + // A) Initial config load and/or secondary take-over. + // B) Migration TO this shard finished, notified by mongos. + // C) Dropping a collection, notified (currently) by mongos. + // D) Stale client wants to reload metadata with a different *epoch*, so we aren't sure. + + if (storedShardVersion.epoch() != reqShardVersion.epoch()) { + // Need to remotely reload if our epochs aren't the same, to verify + LOG(1) << "metadata change requested for " << ns << ", from shard version " + << storedShardVersion << " to " << reqShardVersion + << ", need to verify with config server" << endl; + } else { + // Need to remotely reload since our epochs aren't the same but our version is greater + LOG(1) << "metadata version update requested for " << ns << ", from shard version " + << storedShardVersion << " to " << reqShardVersion + << ", need to verify with config server" << endl; + } - CollectionMetadataMap::const_iterator it = _collMetadata.find( ns ); - verify( it != _collMetadata.end() ) ; + return doRefreshMetadata(txn, ns, reqShardVersion, true, latestShardVersion); +} - ChunkType chunk; - chunk.setMin( min ); - chunk.setMax( max ); - string errMsg; +Status ShardingState::refreshMetadataNow(OperationContext* txn, + const string& ns, + ChunkVersion* latestShardVersion) { + return doRefreshMetadata(txn, ns, ChunkVersion(0, 0, OID()), false, latestShardVersion); +} - CollectionMetadataPtr cloned( it->second->cloneSplit( chunk, splitKeys, version, &errMsg ) ); - // uassert to match old behavior, TODO: report errors w/o throwing - uassert( 16857, errMsg, NULL != cloned.get() ); +void ShardingState::_initialize(const string& server) { + // Ensure only one caller at a time initializes + stdx::lock_guard<stdx::mutex> lk(_mutex); - _collMetadata[ns] = cloned; + if (_enabled) { + // TODO: Do we need to throw exception if the config servers have changed from what we + // already have in place? How do we test for that? + return; } - void ShardingState::mergeChunks(OperationContext* txn, - const string& ns, - const BSONObj& minKey, - const BSONObj& maxKey, - ChunkVersion mergedVersion ) { + ShardedConnectionInfo::addHook(); - invariant(txn->lockState()->isCollectionLockedForMode(ns, MODE_X)); - stdx::lock_guard<stdx::mutex> lk( _mutex ); + std::string errmsg; + ConnectionString configServerCS = ConnectionString::parse(server, errmsg); + uassert(28633, + str::stream() << "Invalid config server connection string: " << errmsg, + configServerCS.isValid()); - CollectionMetadataMap::const_iterator it = _collMetadata.find( ns ); - verify( it != _collMetadata.end() ); + auto catalogManager = stdx::make_unique<CatalogManagerLegacy>(); + uassertStatusOK(catalogManager->init(configServerCS)); - string errMsg; + auto shardRegistry = + stdx::make_unique<ShardRegistry>(stdx::make_unique<RemoteCommandTargeterFactoryImpl>(), + stdx::make_unique<RemoteCommandRunnerImpl>(0), + std::unique_ptr<executor::TaskExecutor>{nullptr}, + catalogManager.get()); - CollectionMetadataPtr cloned( it->second->cloneMerge( minKey, - maxKey, - mergedVersion, - &errMsg ) ); - // uassert to match old behavior, TODO: report errors w/o throwing - uassert( 17004, errMsg, NULL != cloned.get() ); + grid.init(std::move(catalogManager), std::move(shardRegistry)); - _collMetadata[ns] = cloned; - } + _enabled = true; +} - void ShardingState::resetMetadata( const string& ns ) { - stdx::lock_guard<stdx::mutex> lk( _mutex ); +Status ShardingState::doRefreshMetadata(OperationContext* txn, + const string& ns, + const ChunkVersion& reqShardVersion, + bool useRequestedVersion, + ChunkVersion* latestShardVersion) { + // The idea here is that we're going to reload the metadata from the config server, but + // we need to do so outside any locks. When we get our result back, if the current metadata + // has changed, we may not be able to install the new metadata. - warning() << "resetting metadata for " << ns << ", this should only be used in testing" - << endl; + // + // Get the initial metadata + // No DBLock is needed since the metadata is expected to change during reload. + // - _collMetadata.erase( ns ); - } + CollectionMetadataPtr beforeMetadata; - Status ShardingState::refreshMetadataIfNeeded( OperationContext* txn, - const string& ns, - const ChunkVersion& reqShardVersion, - ChunkVersion* latestShardVersion ) { - // The _configServerTickets serializes this process such that only a small number of threads - // can try to refresh at the same time. - - LOG( 2 ) << "metadata refresh requested for " << ns << " at shard version " - << reqShardVersion << endl; - - // - // Queuing of refresh requests starts here when remote reload is needed. This may take time. - // TODO: Explicitly expose the queuing discipline. - // - - _configServerTickets.waitForTicket(); - TicketHolderReleaser needTicketFrom( &_configServerTickets ); + stdx::lock_guard<stdx::mutex> lk(_mutex); - // - // Fast path - check if the requested version is at a higher version than the current - // metadata version or a different epoch before verifying against config server. - // + // We can't reload if sharding is not enabled - i.e. without a config server location + if (!_enabled) { + string errMsg = str::stream() << "cannot refresh metadata for " << ns + << " before sharding has been enabled"; - CollectionMetadataPtr storedMetadata; - { - stdx::lock_guard<stdx::mutex> lk( _mutex ); - CollectionMetadataMap::iterator it = _collMetadata.find( ns ); - if ( it != _collMetadata.end() ) storedMetadata = it->second; + warning() << errMsg; + return Status(ErrorCodes::NotYetInitialized, errMsg); } - ChunkVersion storedShardVersion; - if ( storedMetadata ) storedShardVersion = storedMetadata->getShardVersion(); - *latestShardVersion = storedShardVersion; - if ( storedShardVersion >= reqShardVersion && - storedShardVersion.epoch() == reqShardVersion.epoch() ) { + // We also can't reload if a shard name has not yet been set. + if (_shardName.empty()) { + string errMsg = str::stream() << "cannot refresh metadata for " << ns + << " before shard name has been set"; - // Don't need to remotely reload if we're in the same epoch with a >= version - return Status::OK(); + warning() << errMsg; + return Status(ErrorCodes::NotYetInitialized, errMsg); } - // - // Slow path - remotely reload - // - // Cases: - // A) Initial config load and/or secondary take-over. - // B) Migration TO this shard finished, notified by mongos. - // C) Dropping a collection, notified (currently) by mongos. - // D) Stale client wants to reload metadata with a different *epoch*, so we aren't sure. - - if ( storedShardVersion.epoch() != reqShardVersion.epoch() ) { - // Need to remotely reload if our epochs aren't the same, to verify - LOG( 1 ) << "metadata change requested for " << ns << ", from shard version " - << storedShardVersion << " to " << reqShardVersion - << ", need to verify with config server" << endl; - } - else { - // Need to remotely reload since our epochs aren't the same but our version is greater - LOG( 1 ) << "metadata version update requested for " << ns - << ", from shard version " << storedShardVersion << " to " << reqShardVersion - << ", need to verify with config server" << endl; + CollectionMetadataMap::iterator it = _collMetadata.find(ns); + if (it != _collMetadata.end()) { + beforeMetadata = it->second; } - - return doRefreshMetadata(txn, ns, reqShardVersion, true, latestShardVersion); } - Status ShardingState::refreshMetadataNow(OperationContext* txn, - const string& ns, - ChunkVersion* latestShardVersion) { - return doRefreshMetadata(txn, ns, ChunkVersion(0, 0, OID()), false, latestShardVersion); + ChunkVersion beforeShardVersion; + ChunkVersion beforeCollVersion; + if (beforeMetadata) { + beforeShardVersion = beforeMetadata->getShardVersion(); + beforeCollVersion = beforeMetadata->getCollVersion(); } - void ShardingState::_initialize(const string& server) { - // Ensure only one caller at a time initializes - stdx::lock_guard<stdx::mutex> lk(_mutex); + *latestShardVersion = beforeShardVersion; - if (_enabled) { - // TODO: Do we need to throw exception if the config servers have changed from what we - // already have in place? How do we test for that? - return; - } + // + // Determine whether we need to diff or fully reload + // - ShardedConnectionInfo::addHook(); + bool fullReload = false; + if (!beforeMetadata) { + // We don't have any metadata to reload from + fullReload = true; + } else if (useRequestedVersion && reqShardVersion.epoch() != beforeShardVersion.epoch()) { + // It's not useful to use the metadata as a base because we think the epoch will differ + fullReload = true; + } + + // + // Load the metadata from the remote server, start construction + // + + LOG(0) << "remotely refreshing metadata for " << ns + << (useRequestedVersion + ? string(" with requested shard version ") + reqShardVersion.toString() + : "") + << (fullReload ? ", current shard version is " : " based on current shard version ") + << beforeShardVersion << ", current metadata version is " << beforeCollVersion << endl; + + string errMsg; + + MetadataLoader mdLoader; + CollectionMetadata* remoteMetadataRaw = new CollectionMetadata(); + CollectionMetadataPtr remoteMetadata(remoteMetadataRaw); + + Timer refreshTimer; + Status status = mdLoader.makeCollectionMetadata(grid.catalogManager(), + ns, + getShardName(), + fullReload ? NULL : beforeMetadata.get(), + remoteMetadataRaw); + long long refreshMillis = refreshTimer.millis(); + + if (status.code() == ErrorCodes::NamespaceNotFound) { + remoteMetadata.reset(); + remoteMetadataRaw = NULL; + } else if (!status.isOK()) { + warning() << "could not remotely refresh metadata for " << ns << causedBy(status.reason()) + << endl; - std::string errmsg; - ConnectionString configServerCS = ConnectionString::parse(server, errmsg); - uassert(28633, - str::stream() << "Invalid config server connection string: " << errmsg, - configServerCS.isValid()); + return status; + } - auto catalogManager = stdx::make_unique<CatalogManagerLegacy>(); - uassertStatusOK(catalogManager->init(configServerCS)); + ChunkVersion remoteShardVersion; + ChunkVersion remoteCollVersion; + if (remoteMetadata) { + remoteShardVersion = remoteMetadata->getShardVersion(); + remoteCollVersion = remoteMetadata->getCollVersion(); + } - auto shardRegistry = stdx::make_unique<ShardRegistry>( - stdx::make_unique<RemoteCommandTargeterFactoryImpl>(), - stdx::make_unique<RemoteCommandRunnerImpl>(0), - std::unique_ptr<executor::TaskExecutor>{nullptr}, - catalogManager.get()); + // + // Get ready to install loaded metadata if needed + // - grid.init(std::move(catalogManager), std::move(shardRegistry)); + CollectionMetadataPtr afterMetadata; + ChunkVersion afterShardVersion; + ChunkVersion afterCollVersion; + ChunkVersion::VersionChoice choice; - _enabled = true; - } + // If we choose to install the new metadata, this describes the kind of install + enum InstallType { + InstallType_New, + InstallType_Update, + InstallType_Replace, + InstallType_Drop, + InstallType_None + } installType = InstallType_None; // compiler complains otherwise - Status ShardingState::doRefreshMetadata( OperationContext* txn, - const string& ns, - const ChunkVersion& reqShardVersion, - bool useRequestedVersion, - ChunkVersion* latestShardVersion ) { - // The idea here is that we're going to reload the metadata from the config server, but - // we need to do so outside any locks. When we get our result back, if the current metadata - // has changed, we may not be able to install the new metadata. + // Exclusive collection lock needed since we're now potentially changing the metadata, + // and don't want reads/writes to be ongoing. + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); + Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); // - // Get the initial metadata - // No DBLock is needed since the metadata is expected to change during reload. + // Get the metadata now that the load has completed // - CollectionMetadataPtr beforeMetadata; - - { - stdx::lock_guard<stdx::mutex> lk( _mutex ); - - // We can't reload if sharding is not enabled - i.e. without a config server location - if (!_enabled) { - string errMsg = str::stream() << "cannot refresh metadata for " << ns - << " before sharding has been enabled"; - - warning() << errMsg; - return Status(ErrorCodes::NotYetInitialized, errMsg); - } - - // We also can't reload if a shard name has not yet been set. - if (_shardName.empty()) { - - string errMsg = str::stream() << "cannot refresh metadata for " << ns - << " before shard name has been set"; - - warning() << errMsg; - return Status(ErrorCodes::NotYetInitialized, errMsg); - } + stdx::lock_guard<stdx::mutex> lk(_mutex); - CollectionMetadataMap::iterator it = _collMetadata.find(ns); - if (it != _collMetadata.end()) { - beforeMetadata = it->second; - } - } + // Don't reload if our config server has changed or sharding is no longer enabled + if (!_enabled) { + string errMsg = str::stream() << "could not refresh metadata for " << ns + << ", sharding is no longer enabled"; - ChunkVersion beforeShardVersion; - ChunkVersion beforeCollVersion; - if ( beforeMetadata ) { - beforeShardVersion = beforeMetadata->getShardVersion(); - beforeCollVersion = beforeMetadata->getCollVersion(); + warning() << errMsg; + return Status(ErrorCodes::NotYetInitialized, errMsg); } - *latestShardVersion = beforeShardVersion; - - // - // Determine whether we need to diff or fully reload - // + CollectionMetadataMap::iterator it = _collMetadata.find(ns); + if (it != _collMetadata.end()) + afterMetadata = it->second; - bool fullReload = false; - if ( !beforeMetadata ) { - // We don't have any metadata to reload from - fullReload = true; - } - else if ( useRequestedVersion && reqShardVersion.epoch() != beforeShardVersion.epoch() ) { - // It's not useful to use the metadata as a base because we think the epoch will differ - fullReload = true; + if (afterMetadata) { + afterShardVersion = afterMetadata->getShardVersion(); + afterCollVersion = afterMetadata->getCollVersion(); } + *latestShardVersion = afterShardVersion; + // - // Load the metadata from the remote server, start construction + // Resolve newer pending chunks with the remote metadata, finish construction // - LOG( 0 ) << "remotely refreshing metadata for " << ns - << ( useRequestedVersion ? - string( " with requested shard version " ) + reqShardVersion.toString() : "" ) - << ( fullReload ? - ", current shard version is " : " based on current shard version " ) - << beforeShardVersion - << ", current metadata version is " << beforeCollVersion << endl; - - string errMsg; - - MetadataLoader mdLoader; - CollectionMetadata* remoteMetadataRaw = new CollectionMetadata(); - CollectionMetadataPtr remoteMetadata( remoteMetadataRaw ); - - Timer refreshTimer; - Status status = - mdLoader.makeCollectionMetadata(grid.catalogManager(), - ns, - getShardName(), - fullReload ? NULL : beforeMetadata.get(), - remoteMetadataRaw); - long long refreshMillis = refreshTimer.millis(); - - if ( status.code() == ErrorCodes::NamespaceNotFound ) { - remoteMetadata.reset(); - remoteMetadataRaw = NULL; - } - else if ( !status.isOK() ) { + status = mdLoader.promotePendingChunks(afterMetadata.get(), remoteMetadataRaw); - warning() << "could not remotely refresh metadata for " << ns - << causedBy( status.reason() ) << endl; + if (!status.isOK()) { + warning() << "remote metadata for " << ns + << " is inconsistent with current pending chunks" << causedBy(status.reason()) + << endl; return status; } - ChunkVersion remoteShardVersion; - ChunkVersion remoteCollVersion; - if ( remoteMetadata ) { - remoteShardVersion = remoteMetadata->getShardVersion(); - remoteCollVersion = remoteMetadata->getCollVersion(); - } - // - // Get ready to install loaded metadata if needed + // Compare the 'before', 'after', and 'remote' versions/epochs and choose newest + // Zero-epochs (sentinel value for "dropped" collections), are tested by + // !epoch.isSet(). // - CollectionMetadataPtr afterMetadata; - ChunkVersion afterShardVersion; - ChunkVersion afterCollVersion; - ChunkVersion::VersionChoice choice; - - // If we choose to install the new metadata, this describes the kind of install - enum InstallType { - InstallType_New, InstallType_Update, InstallType_Replace, InstallType_Drop, - InstallType_None - } installType = InstallType_None; // compiler complains otherwise - - { - // Exclusive collection lock needed since we're now potentially changing the metadata, - // and don't want reads/writes to be ongoing. - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock dbLock(txn->lockState(), nsToDatabaseSubstring(ns), MODE_IX); - Lock::CollectionLock collLock(txn->lockState(), ns, MODE_X); - - // - // Get the metadata now that the load has completed - // - - stdx::lock_guard<stdx::mutex> lk( _mutex ); - - // Don't reload if our config server has changed or sharding is no longer enabled - if (!_enabled) { - string errMsg = str::stream() << "could not refresh metadata for " << ns - << ", sharding is no longer enabled"; - - warning() << errMsg; - return Status(ErrorCodes::NotYetInitialized, errMsg); + choice = ChunkVersion::chooseNewestVersion( + beforeCollVersion, afterCollVersion, remoteCollVersion); + + if (choice == ChunkVersion::VersionChoice_Remote) { + dassert(!remoteCollVersion.epoch().isSet() || remoteShardVersion >= beforeShardVersion); + + if (!afterCollVersion.epoch().isSet()) { + // First metadata load + installType = InstallType_New; + dassert(it == _collMetadata.end()); + _collMetadata.insert(make_pair(ns, remoteMetadata)); + } else if (remoteCollVersion.epoch().isSet() && + remoteCollVersion.epoch() == afterCollVersion.epoch()) { + // Update to existing metadata + installType = InstallType_Update; + + // Invariant: If CollMetadata was not found, version should be have been 0. + dassert(it != _collMetadata.end()); + it->second = remoteMetadata; + } else if (remoteCollVersion.epoch().isSet()) { + // New epoch detected, replacing metadata + installType = InstallType_Replace; + + // Invariant: If CollMetadata was not found, version should be have been 0. + dassert(it != _collMetadata.end()); + it->second = remoteMetadata; + } else { + dassert(!remoteCollVersion.epoch().isSet()); + + // Drop detected + installType = InstallType_Drop; + _collMetadata.erase(it); } - CollectionMetadataMap::iterator it = _collMetadata.find( ns ); - if ( it != _collMetadata.end() ) afterMetadata = it->second; + *latestShardVersion = remoteShardVersion; + } + } + // End _mutex + // End DBWrite + + // + // Do messaging based on what happened above + // + string localShardVersionMsg = beforeShardVersion.epoch() == afterShardVersion.epoch() + ? afterShardVersion.toString() + : beforeShardVersion.toString() + " / " + afterShardVersion.toString(); + + if (choice == ChunkVersion::VersionChoice_Unknown) { + string errMsg = str::stream() + << "need to retry loading metadata for " << ns + << ", collection may have been dropped or recreated during load" + << " (loaded shard version : " << remoteShardVersion.toString() + << ", stored shard versions : " << localShardVersionMsg << ", took " << refreshMillis + << "ms)"; + + warning() << errMsg; + return Status(ErrorCodes::RemoteChangeDetected, errMsg); + } - if ( afterMetadata ) { - afterShardVersion = afterMetadata->getShardVersion(); - afterCollVersion = afterMetadata->getCollVersion(); - } + if (choice == ChunkVersion::VersionChoice_Local) { + LOG(0) << "metadata of collection " << ns + << " already up to date (shard version : " << afterShardVersion.toString() + << ", took " << refreshMillis << "ms)" << endl; + return Status::OK(); + } - *latestShardVersion = afterShardVersion; + dassert(choice == ChunkVersion::VersionChoice_Remote); - // - // Resolve newer pending chunks with the remote metadata, finish construction - // + switch (installType) { + case InstallType_New: + LOG(0) << "collection " << ns << " was previously unsharded" + << ", new metadata loaded with shard version " << remoteShardVersion << endl; + break; + case InstallType_Update: + LOG(0) << "updating metadata for " << ns << " from shard version " + << localShardVersionMsg << " to shard version " << remoteShardVersion << endl; + break; + case InstallType_Replace: + LOG(0) << "replacing metadata for " << ns << " at shard version " + << localShardVersionMsg << " with a new epoch (shard version " + << remoteShardVersion << ")" << endl; + break; + case InstallType_Drop: + LOG(0) << "dropping metadata for " << ns << " at shard version " << localShardVersionMsg + << ", took " << refreshMillis << "ms" << endl; + break; + default: + verify(false); + break; + } - status = mdLoader.promotePendingChunks( afterMetadata.get(), remoteMetadataRaw ); + if (installType != InstallType_Drop) { + LOG(0) << "collection version was loaded at version " << remoteCollVersion << ", took " + << refreshMillis << "ms" << endl; + } - if ( !status.isOK() ) { + return Status::OK(); +} - warning() << "remote metadata for " << ns - << " is inconsistent with current pending chunks" - << causedBy( status.reason() ) << endl; +void ShardingState::appendInfo(BSONObjBuilder& builder) { + stdx::lock_guard<stdx::mutex> lk(_mutex); - return status; - } + builder.appendBool("enabled", _enabled); + if (!_enabled) { + return; + } - // - // Compare the 'before', 'after', and 'remote' versions/epochs and choose newest - // Zero-epochs (sentinel value for "dropped" collections), are tested by - // !epoch.isSet(). - // + builder.append("configServer", grid.catalogManager()->connectionString().toString()); + builder.append("shardName", _shardName); - choice = ChunkVersion::chooseNewestVersion( beforeCollVersion, - afterCollVersion, - remoteCollVersion ); + BSONObjBuilder versionB(builder.subobjStart("versions")); + for (CollectionMetadataMap::const_iterator it = _collMetadata.begin(); + it != _collMetadata.end(); + ++it) { + CollectionMetadataPtr metadata = it->second; + versionB.appendTimestamp(it->first, metadata->getShardVersion().toLong()); + } - if ( choice == ChunkVersion::VersionChoice_Remote ) { - dassert(!remoteCollVersion.epoch().isSet() || - remoteShardVersion >= beforeShardVersion); + versionB.done(); +} - if ( !afterCollVersion.epoch().isSet() ) { +bool ShardingState::needCollectionMetadata(Client* client, const string& ns) const { + if (!_enabled) + return false; - // First metadata load - installType = InstallType_New; - dassert( it == _collMetadata.end() ); - _collMetadata.insert( make_pair( ns, remoteMetadata ) ); - } - else if ( remoteCollVersion.epoch().isSet() && - remoteCollVersion.epoch() == afterCollVersion.epoch() ) { + if (!ShardedConnectionInfo::get(client, false)) + return false; - // Update to existing metadata - installType = InstallType_Update; + return true; +} - // Invariant: If CollMetadata was not found, version should be have been 0. - dassert( it != _collMetadata.end() ); - it->second = remoteMetadata; - } - else if ( remoteCollVersion.epoch().isSet() ) { +CollectionMetadataPtr ShardingState::getCollectionMetadata(const string& ns) { + stdx::lock_guard<stdx::mutex> lk(_mutex); - // New epoch detected, replacing metadata - installType = InstallType_Replace; + CollectionMetadataMap::const_iterator it = _collMetadata.find(ns); + if (it == _collMetadata.end()) { + return CollectionMetadataPtr(); + } else { + return it->second; + } +} - // Invariant: If CollMetadata was not found, version should be have been 0. - dassert( it != _collMetadata.end() ); - it->second = remoteMetadata; - } - else { - dassert( !remoteCollVersion.epoch().isSet() ); +ShardingState shardingState; - // Drop detected - installType = InstallType_Drop; - _collMetadata.erase( it ); - } +// -----ShardingState END ---- - *latestShardVersion = remoteShardVersion; - } - } - // End _mutex - // End DBWrite +// -----ShardedConnectionInfo START ---- - // - // Do messaging based on what happened above - // - string localShardVersionMsg = - beforeShardVersion.epoch() == afterShardVersion.epoch() ? - afterShardVersion.toString() : - beforeShardVersion.toString() + " / " + afterShardVersion.toString(); +ShardedConnectionInfo::ShardedConnectionInfo() { + _forceVersionOk = false; +} - if ( choice == ChunkVersion::VersionChoice_Unknown ) { +ShardedConnectionInfo* ShardedConnectionInfo::get(Client* client, bool create) { + auto& current = clientSCI(client); - string errMsg = str::stream() - << "need to retry loading metadata for " << ns - << ", collection may have been dropped or recreated during load" - << " (loaded shard version : " << remoteShardVersion.toString() - << ", stored shard versions : " << localShardVersionMsg - << ", took " << refreshMillis << "ms)"; + if (!current && create) { + LOG(1) << "entering shard mode for connection" << endl; + current = boost::in_place(); + } - warning() << errMsg; - return Status( ErrorCodes::RemoteChangeDetected, errMsg ); - } + return current ? ¤t.value() : nullptr; +} - if ( choice == ChunkVersion::VersionChoice_Local ) { +void ShardedConnectionInfo::reset(Client* client) { + clientSCI(client) = boost::none; +} - LOG( 0 ) << "metadata of collection " << ns << " already up to date (shard version : " - << afterShardVersion.toString() << ", took " << refreshMillis << "ms)" - << endl; - return Status::OK(); - } +const ChunkVersion ShardedConnectionInfo::getVersion(const string& ns) const { + NSVersionMap::const_iterator it = _versions.find(ns); + if (it != _versions.end()) { + return it->second; + } else { + return ChunkVersion(0, 0, OID()); + } +} - dassert( choice == ChunkVersion::VersionChoice_Remote ); +void ShardedConnectionInfo::setVersion(const string& ns, const ChunkVersion& version) { + _versions[ns] = version; +} - switch( installType ) { - case InstallType_New: - LOG( 0 ) << "collection " << ns << " was previously unsharded" - << ", new metadata loaded with shard version " << remoteShardVersion - << endl; - break; - case InstallType_Update: - LOG( 0 ) << "updating metadata for " << ns << " from shard version " - << localShardVersionMsg << " to shard version " << remoteShardVersion - << endl; - break; - case InstallType_Replace: - LOG( 0 ) << "replacing metadata for " << ns << " at shard version " - << localShardVersionMsg << " with a new epoch (shard version " - << remoteShardVersion << ")" << endl; - break; - case InstallType_Drop: - LOG( 0 ) << "dropping metadata for " << ns << " at shard version " - << localShardVersionMsg << ", took " << refreshMillis << "ms" << endl; - break; - default: - verify( false ); - break; - } +void ShardedConnectionInfo::addHook() { + static stdx::mutex lock; + static bool done = false; - if ( installType != InstallType_Drop ) { - LOG( 0 ) << "collection version was loaded at version " << remoteCollVersion - << ", took " << refreshMillis << "ms" << endl; - } + stdx::lock_guard<stdx::mutex> lk(lock); + if (!done) { + log() << "first cluster operation detected, adding sharding hook to enable versioning " + "and authentication to remote servers"; - return Status::OK(); + globalConnPool.addHook(new ShardingConnectionHook(false)); + shardConnectionPool.addHook(new ShardingConnectionHook(true)); + done = true; } +} - void ShardingState::appendInfo(BSONObjBuilder& builder) { - stdx::lock_guard<stdx::mutex> lk(_mutex); +class MongodShardCommand : public Command { +public: + MongodShardCommand(const char* n) : Command(n) {} + virtual bool slaveOk() const { + return false; + } + virtual bool adminOnly() const { + return true; + } +}; - builder.appendBool("enabled", _enabled); - if (!_enabled) { - return; - } - builder.append("configServer", grid.catalogManager()->connectionString().toString()); - builder.append("shardName", _shardName); +bool haveLocalShardingInfo(Client* client, const string& ns) { + if (!shardingState.enabled()) + return false; - BSONObjBuilder versionB(builder.subobjStart("versions")); - for (CollectionMetadataMap::const_iterator it = _collMetadata.begin(); - it != _collMetadata.end(); - ++it) { + if (!shardingState.hasVersion(ns)) + return false; - CollectionMetadataPtr metadata = it->second; - versionB.appendTimestamp(it->first, metadata->getShardVersion().toLong()); - } + return ShardedConnectionInfo::get(client, false) != NULL; +} - versionB.done(); - } +class UnsetShardingCommand : public MongodShardCommand { +public: + UnsetShardingCommand() : MongodShardCommand("unsetSharding") {} - bool ShardingState::needCollectionMetadata( Client* client, const string& ns ) const { - if ( ! _enabled ) - return false; + virtual void help(stringstream& help) const { + help << "internal"; + } - if ( ! ShardedConnectionInfo::get( client, false ) ) - return false; + virtual bool isWriteCommandForConfigServer() const { + return false; + } + virtual bool slaveOk() const { return true; } - CollectionMetadataPtr ShardingState::getCollectionMetadata( const string& ns ) { - stdx::lock_guard<stdx::mutex> lk( _mutex ); - - CollectionMetadataMap::const_iterator it = _collMetadata.find( ns ); - if ( it == _collMetadata.end() ) { - return CollectionMetadataPtr(); - } - else { - return it->second; - } + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::internal); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); } - ShardingState shardingState; - - // -----ShardingState END ---- - - // -----ShardedConnectionInfo START ---- - - ShardedConnectionInfo::ShardedConnectionInfo() { - _forceVersionOk = false; + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + ShardedConnectionInfo::reset(txn->getClient()); + return true; } - ShardedConnectionInfo* ShardedConnectionInfo::get( Client* client, bool create ) { - auto& current = clientSCI(client); +} unsetShardingCommand; - if (!current && create) { - LOG(1) << "entering shard mode for connection" << endl; - current = boost::in_place(); - } +class SetShardVersion : public MongodShardCommand { +public: + SetShardVersion() : MongodShardCommand("setShardVersion") {} - return current ? ¤t.value() : nullptr; + virtual void help(stringstream& help) const { + help << "internal"; } - void ShardedConnectionInfo::reset(Client* client) { - clientSCI(client) = boost::none; + virtual bool slaveOk() const { + return true; } - - const ChunkVersion ShardedConnectionInfo::getVersion( const string& ns ) const { - NSVersionMap::const_iterator it = _versions.find( ns ); - if ( it != _versions.end() ) { - return it->second; - } - else { - return ChunkVersion( 0, 0, OID() ); - } + virtual bool isWriteCommandForConfigServer() const { + return false; } - void ShardedConnectionInfo::setVersion( const string& ns , const ChunkVersion& version ) { - _versions[ns] = version; + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::internal); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); } - void ShardedConnectionInfo::addHook() { - static stdx::mutex lock; - static bool done = false; + bool checkConfigOrInit(OperationContext* txn, + const string& configdb, + bool authoritative, + string& errmsg, + BSONObjBuilder& result, + bool locked = false) const { + if (configdb.size() == 0) { + errmsg = "no configdb"; + return false; + } - stdx::lock_guard<stdx::mutex> lk(lock); - if (!done) { - log() << "first cluster operation detected, adding sharding hook to enable versioning " - "and authentication to remote servers"; + if (shardingState.enabled()) { + if (configdb == shardingState.getConfigServer()) + return true; - globalConnPool.addHook(new ShardingConnectionHook(false)); - shardConnectionPool.addHook(new ShardingConnectionHook(true)); - done = true; - } - } + result.append("configdb", + BSON("stored" << shardingState.getConfigServer() << "given" << configdb)); - class MongodShardCommand : public Command { - public: - MongodShardCommand( const char * n ) : Command( n ) { - } - virtual bool slaveOk() const { + errmsg = str::stream() << "mongos specified a different config database string : " + << "stored : " << shardingState.getConfigServer() + << " vs given : " << configdb; return false; } - virtual bool adminOnly() const { - return true; - } - }; - - bool haveLocalShardingInfo( Client* client, const string& ns ) { - if ( ! shardingState.enabled() ) + if (!authoritative) { + result.appendBool("need_authoritative", true); + errmsg = "first setShardVersion"; return false; + } - if ( ! shardingState.hasVersion( ns ) ) - return false; + if (locked) { + ShardingState::initialize(configdb); + return true; + } - return ShardedConnectionInfo::get(client, false) != NULL; + ScopedTransaction transaction(txn, MODE_X); + Lock::GlobalWrite lk(txn->lockState()); + return checkConfigOrInit(txn, configdb, authoritative, errmsg, result, true); } - class UnsetShardingCommand : public MongodShardCommand { - public: - UnsetShardingCommand() : MongodShardCommand("unsetSharding") {} - - virtual void help( stringstream& help ) const { - help << "internal"; + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + // Compatibility error for < v3.0 mongoses still active in the cluster + // TODO: Remove post-3.0 + if (!cmdObj["serverID"].eoo()) { + // This mongos is too old to talk to us + string errMsg = stream() << "v3.0 mongod is incompatible with v2.6 mongos, " + << "a v2.6 mongos may be running in the v3.0 cluster at " + << txn->getClient()->clientAddress(false); + error() << errMsg; + return appendCommandStatus(result, Status(ErrorCodes::ProtocolError, errMsg)); } - virtual bool isWriteCommandForConfigServer() const { return false; } + // Steps + // 1. check basic config + // 2. extract params from command + // 3. fast check + // 4. slow check (LOCKS) - virtual bool slaveOk() const { return true; } + // step 1 - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::internal); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + Client* client = txn->getClient(); + LastError::get(client).disable(); + ShardedConnectionInfo* info = ShardedConnectionInfo::get(client, true); + + bool authoritative = cmdObj.getBoolField("authoritative"); + + // check config server is ok or enable sharding + if (!checkConfigOrInit( + txn, cmdObj["configdb"].valuestrsafe(), authoritative, errmsg, result)) { + return false; } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - ShardedConnectionInfo::reset(txn->getClient()); - return true; + // check shard name is correct + if (cmdObj["shard"].type() == String) { + // The shard host is also sent when using setShardVersion, report this host if there + // is an error. + shardingState.gotShardNameAndHost(cmdObj["shard"].String(), cmdObj["shardHost"].str()); } - } unsetShardingCommand; + // Handle initial shard connection + if (cmdObj["version"].eoo() && cmdObj["init"].trueValue()) { + result.append("initialized", true); - class SetShardVersion : public MongodShardCommand { - public: - SetShardVersion() : MongodShardCommand("setShardVersion") {} + // Send back wire version to let mongos know what protocol we can speak + result.append("minWireVersion", minWireVersion); + result.append("maxWireVersion", maxWireVersion); - virtual void help( stringstream& help ) const { - help << "internal"; + return true; } - virtual bool slaveOk() const { return true; } - virtual bool isWriteCommandForConfigServer() const { return false; } - - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::internal); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + string ns = cmdObj["setShardVersion"].valuestrsafe(); + if (ns.size() == 0) { + errmsg = "need to specify namespace"; + return false; } - bool checkConfigOrInit(OperationContext* txn, - const string& configdb, - bool authoritative, - string& errmsg, - BSONObjBuilder& result, - bool locked = false ) const { - if ( configdb.size() == 0 ) { - errmsg = "no configdb"; - return false; - } - - if ( shardingState.enabled() ) { - if ( configdb == shardingState.getConfigServer() ) - return true; - - result.append( "configdb" , BSON( "stored" << shardingState.getConfigServer() << - "given" << configdb ) ); - - errmsg = str::stream() << "mongos specified a different config database string : " - << "stored : " << shardingState.getConfigServer() - << " vs given : " << configdb; - return false; - } - - if ( ! authoritative ) { - result.appendBool( "need_authoritative" , true ); - errmsg = "first setShardVersion"; - return false; - } - - if ( locked ) { - ShardingState::initialize(configdb); - return true; - } - ScopedTransaction transaction(txn, MODE_X); - Lock::GlobalWrite lk(txn->lockState()); - return checkConfigOrInit(txn, configdb, authoritative, errmsg, result, true); + // we can run on a slave up to here + if (!repl::getGlobalReplicationCoordinator()->canAcceptWritesForDatabase( + nsToDatabase(ns))) { + result.append("errmsg", "not master"); + result.append("note", "from post init in setShardVersion"); + return false; } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - - // Compatibility error for < v3.0 mongoses still active in the cluster - // TODO: Remove post-3.0 - if (!cmdObj["serverID"].eoo()) { - - // This mongos is too old to talk to us - string errMsg = stream() << "v3.0 mongod is incompatible with v2.6 mongos, " - << "a v2.6 mongos may be running in the v3.0 cluster at " - << txn->getClient()->clientAddress(false); - error() << errMsg; - return appendCommandStatus(result, Status(ErrorCodes::ProtocolError, errMsg)); - } + // step 2 + if (!ChunkVersion::canParseBSON(cmdObj, "version")) { + errmsg = "need to specify version"; + return false; + } - // Steps - // 1. check basic config - // 2. extract params from command - // 3. fast check - // 4. slow check (LOCKS) - - // step 1 - - Client* client = txn->getClient(); - LastError::get(client).disable(); - ShardedConnectionInfo* info = ShardedConnectionInfo::get( client, true ); - - bool authoritative = cmdObj.getBoolField( "authoritative" ); - - // check config server is ok or enable sharding - if (!checkConfigOrInit( - txn, cmdObj["configdb"].valuestrsafe(), authoritative, errmsg, result)) { - return false; - } + const ChunkVersion version = ChunkVersion::fromBSON(cmdObj, "version"); - // check shard name is correct - if ( cmdObj["shard"].type() == String ) { - // The shard host is also sent when using setShardVersion, report this host if there - // is an error. - shardingState.gotShardNameAndHost( cmdObj["shard"].String(), - cmdObj["shardHost"].str() ); - } - - // Handle initial shard connection - if( cmdObj["version"].eoo() && cmdObj["init"].trueValue() ){ + // step 3 - result.append( "initialized", true ); + const ChunkVersion oldVersion = info->getVersion(ns); + const ChunkVersion globalVersion = shardingState.getVersion(ns); - // Send back wire version to let mongos know what protocol we can speak - result.append( "minWireVersion", minWireVersion ); - result.append( "maxWireVersion", maxWireVersion ); + oldVersion.addToBSON(result, "oldVersion"); - return true; + if (version.isWriteCompatibleWith(globalVersion)) { + // mongos and mongod agree! + if (!oldVersion.isWriteCompatibleWith(version)) { + if (oldVersion < globalVersion && oldVersion.hasEqualEpoch(globalVersion)) { + info->setVersion(ns, version); + } else if (authoritative) { + // this means there was a drop and our version is reset + info->setVersion(ns, version); + } else { + result.append("ns", ns); + result.appendBool("need_authoritative", true); + errmsg = "verifying drop on '" + ns + "'"; + return false; + } } - string ns = cmdObj["setShardVersion"].valuestrsafe(); - if ( ns.size() == 0 ) { - errmsg = "need to specify namespace"; - return false; - } + return true; + } + // step 4 + // Cases below all either return OR fall-through to remote metadata reload. + const bool isDropRequested = !version.isSet() && globalVersion.isSet(); - // we can run on a slave up to here - if (!repl::getGlobalReplicationCoordinator()->canAcceptWritesForDatabase( - nsToDatabase(ns))) { - result.append( "errmsg" , "not master" ); - result.append( "note" , "from post init in setShardVersion" ); + if (isDropRequested) { + if (!authoritative) { + result.appendBool("need_authoritative", true); + result.append("ns", ns); + globalVersion.addToBSON(result, "globalVersion"); + errmsg = "dropping needs to be authoritative"; return false; } - // step 2 - if( ! ChunkVersion::canParseBSON( cmdObj, "version" ) ){ - errmsg = "need to specify version"; + // Fall through to metadata reload below + } else { + // Not Dropping + + // TODO: Refactor all of this + if (version < oldVersion && version.hasEqualEpoch(oldVersion)) { + errmsg = str::stream() << "this connection already had a newer version " + << "of collection '" << ns << "'"; + result.append("ns", ns); + version.addToBSON(result, "newVersion"); + globalVersion.addToBSON(result, "globalVersion"); return false; } - const ChunkVersion version = ChunkVersion::fromBSON( cmdObj, "version" ); - - // step 3 - - const ChunkVersion oldVersion = info->getVersion(ns); - const ChunkVersion globalVersion = shardingState.getVersion(ns); - - oldVersion.addToBSON( result, "oldVersion" ); - - if ( version.isWriteCompatibleWith( globalVersion )) { - // mongos and mongod agree! - if ( !oldVersion.isWriteCompatibleWith( version )) { - if ( oldVersion < globalVersion && - oldVersion.hasEqualEpoch( globalVersion )) { - info->setVersion( ns, version ); - } - else if ( authoritative ) { - // this means there was a drop and our version is reset - info->setVersion( ns, version ); - } - else { - result.append( "ns", ns ); - result.appendBool( "need_authoritative", true ); - errmsg = "verifying drop on '" + ns + "'"; - return false; - } - } - - return true; - } - - // step 4 - // Cases below all either return OR fall-through to remote metadata reload. - const bool isDropRequested = !version.isSet() && globalVersion.isSet(); - - if (isDropRequested) { - if ( ! authoritative ) { - result.appendBool( "need_authoritative" , true ); - result.append( "ns" , ns ); - globalVersion.addToBSON( result, "globalVersion" ); - errmsg = "dropping needs to be authoritative"; - return false; + // TODO: Refactor all of this + if (version < globalVersion && version.hasEqualEpoch(globalVersion)) { + while (shardingState.inCriticalMigrateSection()) { + log() << "waiting till out of critical section" << endl; + shardingState.waitTillNotInCriticalSection(10); } - - // Fall through to metadata reload below + errmsg = str::stream() << "shard global version for collection is higher " + << "than trying to set to '" << ns << "'"; + result.append("ns", ns); + version.addToBSON(result, "version"); + globalVersion.addToBSON(result, "globalVersion"); + result.appendBool("reloadConfig", true); + return false; } - else { - // Not Dropping - - // TODO: Refactor all of this - if ( version < oldVersion && version.hasEqualEpoch( oldVersion ) ) { - errmsg = str::stream() << "this connection already had a newer version " - << "of collection '" << ns << "'"; - result.append( "ns" , ns ); - version.addToBSON( result, "newVersion" ); - globalVersion.addToBSON( result, "globalVersion" ); - return false; - } - - // TODO: Refactor all of this - if ( version < globalVersion && version.hasEqualEpoch( globalVersion ) ) { - while ( shardingState.inCriticalMigrateSection() ) { - log() << "waiting till out of critical section" << endl; - shardingState.waitTillNotInCriticalSection( 10 ); - } - errmsg = str::stream() << "shard global version for collection is higher " - << "than trying to set to '" << ns << "'"; - result.append( "ns" , ns ); - version.addToBSON( result, "version" ); - globalVersion.addToBSON( result, "globalVersion" ); - result.appendBool( "reloadConfig" , true ); - return false; - } - if ( ! globalVersion.isSet() && ! authoritative ) { - // Needed b/c when the last chunk is moved off a shard, - // the version gets reset to zero, which should require a reload. - while ( shardingState.inCriticalMigrateSection() ) { - log() << "waiting till out of critical section" << endl; - shardingState.waitTillNotInCriticalSection( 10 ); - } - - // need authoritative for first look - result.append( "ns" , ns ); - result.appendBool( "need_authoritative" , true ); - errmsg = "first time for collection '" + ns + "'"; - return false; + if (!globalVersion.isSet() && !authoritative) { + // Needed b/c when the last chunk is moved off a shard, + // the version gets reset to zero, which should require a reload. + while (shardingState.inCriticalMigrateSection()) { + log() << "waiting till out of critical section" << endl; + shardingState.waitTillNotInCriticalSection(10); } - // Fall through to metadata reload below + // need authoritative for first look + result.append("ns", ns); + result.appendBool("need_authoritative", true); + errmsg = "first time for collection '" + ns + "'"; + return false; } - ChunkVersion currVersion; - Status status = shardingState.refreshMetadataIfNeeded(txn, ns, version, &currVersion); + // Fall through to metadata reload below + } - if (!status.isOK()) { + ChunkVersion currVersion; + Status status = shardingState.refreshMetadataIfNeeded(txn, ns, version, &currVersion); - // The reload itself was interrupted or confused here + if (!status.isOK()) { + // The reload itself was interrupted or confused here - errmsg = str::stream() << "could not refresh metadata for " << ns - << " with requested shard version " << version.toString() - << ", stored shard version is " << currVersion.toString() - << causedBy( status.reason() ); + errmsg = str::stream() << "could not refresh metadata for " << ns + << " with requested shard version " << version.toString() + << ", stored shard version is " << currVersion.toString() + << causedBy(status.reason()); - warning() << errmsg << endl; + warning() << errmsg << endl; - result.append( "ns" , ns ); - version.addToBSON( result, "version" ); - currVersion.addToBSON( result, "globalVersion" ); - result.appendBool( "reloadConfig", true ); + result.append("ns", ns); + version.addToBSON(result, "version"); + currVersion.addToBSON(result, "globalVersion"); + result.appendBool("reloadConfig", true); - return false; + return false; + } else if (!version.isWriteCompatibleWith(currVersion)) { + // We reloaded a version that doesn't match the version mongos was trying to + // set. + + errmsg = str::stream() << "requested shard version differs from" + << " config shard version for " << ns + << ", requested version is " << version.toString() + << " but found version " << currVersion.toString(); + + OCCASIONALLY warning() << errmsg << endl; + + // WARNING: the exact fields below are important for compatibility with mongos + // version reload. + + result.append("ns", ns); + currVersion.addToBSON(result, "globalVersion"); + + // If this was a reset of a collection or the last chunk moved out, inform mongos to + // do a full reload. + if (currVersion.epoch() != version.epoch() || !currVersion.isSet()) { + result.appendBool("reloadConfig", true); + // Zero-version also needed to trigger full mongos reload, sadly + // TODO: Make this saner, and less impactful (full reload on last chunk is bad) + ChunkVersion(0, 0, OID()).addToBSON(result, "version"); + // For debugging + version.addToBSON(result, "origVersion"); + } else { + version.addToBSON(result, "version"); } - else if ( !version.isWriteCompatibleWith( currVersion ) ) { - - // We reloaded a version that doesn't match the version mongos was trying to - // set. - errmsg = str::stream() << "requested shard version differs from" - << " config shard version for " << ns - << ", requested version is " << version.toString() - << " but found version " << currVersion.toString(); + return false; + } - OCCASIONALLY warning() << errmsg << endl; + info->setVersion(ns, version); + return true; + } - // WARNING: the exact fields below are important for compatibility with mongos - // version reload. +} setShardVersionCmd; - result.append( "ns" , ns ); - currVersion.addToBSON( result, "globalVersion" ); +class GetShardVersion : public MongodShardCommand { +public: + GetShardVersion() : MongodShardCommand("getShardVersion") {} - // If this was a reset of a collection or the last chunk moved out, inform mongos to - // do a full reload. - if (currVersion.epoch() != version.epoch() || !currVersion.isSet() ) { - result.appendBool( "reloadConfig", true ); - // Zero-version also needed to trigger full mongos reload, sadly - // TODO: Make this saner, and less impactful (full reload on last chunk is bad) - ChunkVersion( 0, 0, OID() ).addToBSON( result, "version" ); - // For debugging - version.addToBSON( result, "origVersion" ); - } - else { - version.addToBSON( result, "version" ); - } + virtual void help(stringstream& help) const { + help << " example: { getShardVersion : 'alleyinsider.foo' } "; + } - return false; - } + virtual bool isWriteCommandForConfigServer() const { + return false; + } - info->setVersion( ns , version ); - return true; + virtual Status checkAuthForCommand(ClientBasic* client, + const std::string& dbname, + const BSONObj& cmdObj) { + if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( + ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), + ActionType::getShardVersion)) { + return Status(ErrorCodes::Unauthorized, "Unauthorized"); } + return Status::OK(); + } + virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { + return parseNsFullyQualified(dbname, cmdObj); + } - } setShardVersionCmd; - - class GetShardVersion : public MongodShardCommand { - public: - GetShardVersion() : MongodShardCommand("getShardVersion") {} + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + const string ns = cmdObj["getShardVersion"].valuestrsafe(); + if (ns.size() == 0) { + errmsg = "need to specify full namespace"; + return false; + } - virtual void help( stringstream& help ) const { - help << " example: { getShardVersion : 'alleyinsider.foo' } "; + if (shardingState.enabled()) { + result.append("configServer", shardingState.getConfigServer()); + } else { + result.append("configServer", ""); } - virtual bool isWriteCommandForConfigServer() const { return false; } + result.appendTimestamp("global", shardingState.getVersion(ns).toLong()); - virtual Status checkAuthForCommand(ClientBasic* client, - const std::string& dbname, - const BSONObj& cmdObj) { - if (!AuthorizationSession::get(client)->isAuthorizedForActionsOnResource( - ResourcePattern::forExactNamespace(NamespaceString(parseNs(dbname, cmdObj))), - ActionType::getShardVersion)) { - return Status(ErrorCodes::Unauthorized, "Unauthorized"); - } - return Status::OK(); - } - virtual std::string parseNs(const std::string& dbname, const BSONObj& cmdObj) const { - return parseNsFullyQualified(dbname, cmdObj); + ShardedConnectionInfo* const info = ShardedConnectionInfo::get(txn->getClient(), false); + result.appendBool("inShardedMode", info != NULL); + if (info) { + result.appendTimestamp("mine", info->getVersion(ns).toLong()); + } else { + result.appendTimestamp("mine", 0); } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - const string ns = cmdObj["getShardVersion"].valuestrsafe(); - if (ns.size() == 0) { - errmsg = "need to specify full namespace"; - return false; - } - - if (shardingState.enabled()) { - result.append("configServer", shardingState.getConfigServer()); - } - else { - result.append("configServer", ""); + if (cmdObj["fullMetadata"].trueValue()) { + CollectionMetadataPtr metadata = shardingState.getCollectionMetadata(ns); + if (metadata) { + result.append("metadata", metadata->toBSON()); + } else { + result.append("metadata", BSONObj()); } + } - result.appendTimestamp("global", shardingState.getVersion(ns).toLong()); + return true; + } - ShardedConnectionInfo* const info = ShardedConnectionInfo::get(txn->getClient(), false); - result.appendBool("inShardedMode", info != NULL); - if (info) { - result.appendTimestamp("mine", info->getVersion(ns).toLong()); - } - else { - result.appendTimestamp("mine", 0); - } +} getShardVersion; - if (cmdObj["fullMetadata"].trueValue()) { - CollectionMetadataPtr metadata = shardingState.getCollectionMetadata(ns); - if (metadata) { - result.append("metadata", metadata->toBSON()); - } - else { - result.append("metadata", BSONObj()); - } - } +class ShardingStateCmd : public MongodShardCommand { +public: + ShardingStateCmd() : MongodShardCommand("shardingState") {} - return true; - } + virtual bool isWriteCommandForConfigServer() const { + return true; + } - } getShardVersion; + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) { + ActionSet actions; + actions.addAction(ActionType::shardingState); + out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); + } - class ShardingStateCmd : public MongodShardCommand { - public: - ShardingStateCmd() : MongodShardCommand( "shardingState" ) {} + bool run(OperationContext* txn, + const string& dbname, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + ScopedTransaction transaction(txn, MODE_IX); + Lock::DBLock dbXLock(txn->lockState(), dbname, MODE_X); + OldClientContext ctx(txn, dbname); + + shardingState.appendInfo(result); + return true; + } - virtual bool isWriteCommandForConfigServer() const { return true; } +} shardingStateCmd; - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) { - ActionSet actions; - actions.addAction(ActionType::shardingState); - out->push_back(Privilege(ResourcePattern::forClusterResource(), actions)); - } +/** + * @ return true if not in sharded mode + or if version for this client is ok + */ +static bool shardVersionOk(Client* client, + const string& ns, + string& errmsg, + ChunkVersion& received, + ChunkVersion& wanted) { + if (!shardingState.enabled()) + return true; - bool run(OperationContext* txn, - const string& dbname, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - ScopedTransaction transaction(txn, MODE_IX); - Lock::DBLock dbXLock(txn->lockState(), dbname, MODE_X); - OldClientContext ctx(txn, dbname); - - shardingState.appendInfo( result ); - return true; - } + if (!repl::getGlobalReplicationCoordinator()->canAcceptWritesForDatabase(nsToDatabase(ns))) { + // right now connections to secondaries aren't versioned at all + return true; + } - } shardingStateCmd; + ShardedConnectionInfo* info = ShardedConnectionInfo::get(client, false); - /** - * @ return true if not in sharded mode - or if version for this client is ok - */ - static bool shardVersionOk(Client* client, - const string& ns, - string& errmsg, - ChunkVersion& received, - ChunkVersion& wanted) { + if (!info) { + // this means the client has nothing sharded + // so this allows direct connections to do whatever they want + // which i think is the correct behavior + return true; + } - if ( ! shardingState.enabled() ) - return true; + if (info->inForceVersionOkMode()) { + return true; + } - if (!repl::getGlobalReplicationCoordinator()->canAcceptWritesForDatabase( - nsToDatabase(ns))) { - // right now connections to secondaries aren't versioned at all - return true; - } + // TODO : all collections at some point, be sharded or not, will have a version + // (and a CollectionMetadata) + received = info->getVersion(ns); - ShardedConnectionInfo* info = ShardedConnectionInfo::get( client, false ); + if (ChunkVersion::isIgnoredVersion(received)) { + return true; + } - if ( ! info ) { - // this means the client has nothing sharded - // so this allows direct connections to do whatever they want - // which i think is the correct behavior - return true; - } + wanted = shardingState.getVersion(ns); - if ( info->inForceVersionOkMode() ) { - return true; - } + if (received.isWriteCompatibleWith(wanted)) + return true; - // TODO : all collections at some point, be sharded or not, will have a version - // (and a CollectionMetadata) - received = info->getVersion( ns ); + // + // Figure out exactly why not compatible, send appropriate error message + // The versions themselves are returned in the error, so not needed in messages here + // - if (ChunkVersion::isIgnoredVersion(received)) { - return true; - } + // Check epoch first, to send more meaningful message, since other parameters probably + // won't match either + if (!wanted.hasEqualEpoch(received)) { + errmsg = str::stream() << "version epoch mismatch detected for " << ns << ", " + << "the collection may have been dropped and recreated"; + return false; + } - wanted = shardingState.getVersion( ns ); + if (!wanted.isSet() && received.isSet()) { + errmsg = str::stream() << "this shard no longer contains chunks for " << ns << ", " + << "the collection may have been dropped"; + return false; + } - if( received.isWriteCompatibleWith( wanted ) ) return true; + if (wanted.isSet() && !received.isSet()) { + errmsg = str::stream() << "this shard contains versioned chunks for " << ns << ", " + << "but no version set in request"; + return false; + } + if (wanted.majorVersion() != received.majorVersion()) { // - // Figure out exactly why not compatible, send appropriate error message - // The versions themselves are returned in the error, so not needed in messages here + // Could be > or < - wanted is > if this is the source of a migration, + // wanted < if this is the target of a migration // - // Check epoch first, to send more meaningful message, since other parameters probably - // won't match either - if( ! wanted.hasEqualEpoch( received ) ){ - errmsg = str::stream() << "version epoch mismatch detected for " << ns << ", " - << "the collection may have been dropped and recreated"; - return false; - } - - if( ! wanted.isSet() && received.isSet() ){ - errmsg = str::stream() << "this shard no longer contains chunks for " << ns << ", " - << "the collection may have been dropped"; - return false; - } - - if( wanted.isSet() && ! received.isSet() ){ - errmsg = str::stream() << "this shard contains versioned chunks for " << ns << ", " - << "but no version set in request"; - return false; - } - - if( wanted.majorVersion() != received.majorVersion() ){ - - // - // Could be > or < - wanted is > if this is the source of a migration, - // wanted < if this is the target of a migration - // - - errmsg = str::stream() << "version mismatch detected for " << ns << ", " - << "stored major version " << wanted.majorVersion() - << " does not match received " << received.majorVersion(); - return false; - } - - // Those are all the reasons the versions can mismatch - verify( false ); - + errmsg = str::stream() << "version mismatch detected for " << ns << ", " + << "stored major version " << wanted.majorVersion() + << " does not match received " << received.majorVersion(); return false; - } - void ensureShardVersionOKOrThrow(Client* client, const std::string& ns) { - string errmsg; - ChunkVersion received; - ChunkVersion wanted; - if (!shardVersionOk(client, ns, errmsg, received, wanted)) { - StringBuilder sb; - sb << "[" << ns << "] shard version not ok: " << errmsg; - throw SendStaleConfigException(ns, sb.str(), received, wanted); - } - } + // Those are all the reasons the versions can mismatch + verify(false); - void usingAShardConnection( const string& addr ) { - } + return false; +} - void saveGLEStats(const BSONObj& result, StringData hostString) { - // Declared in cluster_last_error_info.h. - // - // This can be called in mongod, which is unfortunate. To fix this, - // we can redesign how connection pooling works on mongod for sharded operations. +void ensureShardVersionOKOrThrow(Client* client, const std::string& ns) { + string errmsg; + ChunkVersion received; + ChunkVersion wanted; + if (!shardVersionOk(client, ns, errmsg, received, wanted)) { + StringBuilder sb; + sb << "[" << ns << "] shard version not ok: " << errmsg; + throw SendStaleConfigException(ns, sb.str(), received, wanted); } +} + +void usingAShardConnection(const string& addr) {} +void saveGLEStats(const BSONObj& result, StringData hostString) { + // Declared in cluster_last_error_info.h. + // + // This can be called in mongod, which is unfortunate. To fix this, + // we can redesign how connection pooling works on mongod for sharded operations. +} } diff --git a/src/mongo/s/d_state.h b/src/mongo/s/d_state.h index 6bd43ce6865..d0e1d502ae4 100644 --- a/src/mongo/s/d_state.h +++ b/src/mongo/s/d_state.h @@ -38,342 +38,343 @@ namespace mongo { - class Database; - class RecordId; - class OperationContext; - - // -------------- - // --- global state --- - // -------------- - - class ShardingState { - public: - ShardingState(); - - bool enabled(); - std::string getConfigServer(); - - // Initialize sharding state and begin authenticating outgoing connections and handling - // shard versions. If this is not run before sharded operations occur auth will not work - // and versions will not be tracked. - static void initialize(const std::string& server); - - void gotShardName( const std::string& name ); - bool setShardName( const std::string& name ); // Same as above, does not throw - std::string getShardName(); - - // Helpers for SetShardVersion which report the host name sent to this shard when the shard - // name does not match. Do not use in other places. - // TODO: Remove once SSV is deprecated - void gotShardNameAndHost( const std::string& name, const std::string& host ); - bool setShardNameAndHost( const std::string& name, const std::string& host ); - - /** - * Clears the collection metadata cache after step down. - */ - void clearCollectionMetadata(); - - // versioning support - - bool hasVersion( const std::string& ns ); - bool hasVersion( const std::string& ns , ChunkVersion& version ); - ChunkVersion getVersion(const std::string& ns); - - /** - * If the metadata for 'ns' at this shard is at or above the requested version, - * 'reqShardVersion', returns OK and fills in 'latestShardVersion' with the latest shard - * version. The latter is always greater or equal than 'reqShardVersion' if in the same - * epoch. - * - * Otherwise, falls back to refreshMetadataNow. - * - * This call blocks if there are more than N threads - * currently refreshing metadata. (N is the number of - * tickets in ShardingState::_configServerTickets, - * currently 3.) - * - * Locking Note: - * + Must NOT be called with the write lock because this call may go into the network, - * and deadlocks may occur with shard-as-a-config. Therefore, nothing here guarantees - * that 'latestShardVersion' is indeed the current one on return. - */ - Status refreshMetadataIfNeeded( OperationContext* txn, - const std::string& ns, - const ChunkVersion& reqShardVersion, - ChunkVersion* latestShardVersion ); - - /** - * Refreshes collection metadata by asking the config server for the latest information. - * Starts a new config server request. - * - * Locking Notes: - * + Must NOT be called with the write lock because this call may go into the network, - * and deadlocks may occur with shard-as-a-config. Therefore, nothing here guarantees - * that 'latestShardVersion' is indeed the current one on return. - * - * + Because this call must not be issued with the DBLock held, by the time the config - * server sent us back the collection metadata information, someone else may have - * updated the previously stored collection metadata. There are cases when one can't - * tell which of updated or loaded metadata are the freshest. There are also cases where - * the data coming from configs do not correspond to a consistent snapshot. - * In these cases, return RemoteChangeDetected. (This usually means this call needs to - * be issued again, at caller discretion) - * - * @return OK if remote metadata successfully loaded (may or may not have been installed) - * @return RemoteChangeDetected if something changed while reloading and we may retry - * @return !OK if something else went wrong during reload - * @return latestShardVersion the version that is now stored for this collection - */ - Status refreshMetadataNow(OperationContext* txn, - const std::string& ns, - ChunkVersion* latestShardVersion); - - void appendInfo( BSONObjBuilder& b ); - - // querying support - - bool needCollectionMetadata( Client* client, const std::string& ns ) const; - CollectionMetadataPtr getCollectionMetadata( const std::string& ns ); - - // chunk migrate and split support - - /** - * Creates and installs a new chunk metadata for a given collection by "forgetting" about - * one of its chunks. The new metadata uses the provided version, which has to be higher - * than the current metadata's shard version. - * - * One exception: if the forgotten chunk is the last one in this shard for the collection, - * version has to be 0. - * - * If it runs successfully, clients need to grab the new version to access the collection. - * - * LOCKING NOTE: - * Only safe to do inside the - * - * @param ns the collection - * @param min max the chunk to eliminate from the current metadata - * @param version at which the new metadata should be at - */ - void donateChunk(OperationContext* txn, - const std::string& ns, - const BSONObj& min, - const BSONObj& max, - ChunkVersion version); - - /** - * Creates and installs new chunk metadata for a given collection by reclaiming a previously - * donated chunk. The previous metadata's shard version has to be provided. - * - * If it runs successfully, clients that became stale by the previous donateChunk will be - * able to access the collection again. - * - * Note: If a migration has aborted but not yet unregistered a pending chunk, replacing the - * metadata may leave the chunk as pending - this is not dangerous and should be rare, but - * will require a stepdown to fully recover. - * - * @param ns the collection - * @param prevMetadata the previous metadata before we donated a chunk - */ - void undoDonateChunk(OperationContext* txn, - const std::string& ns, - CollectionMetadataPtr prevMetadata); - - /** - * Remembers a chunk range between 'min' and 'max' as a range which will have data migrated - * into it. This data can then be protected against cleanup of orphaned data. - * - * Overlapping pending ranges will be removed, so it is only safe to use this when you know - * your metadata view is definitive, such as at the start of a migration. - * - * @return false with errMsg if the range is owned by this shard - */ - bool notePending(OperationContext* txn, - const std::string& ns, - const BSONObj& min, - const BSONObj& max, - const OID& epoch, - std::string* errMsg ); - - /** - * Stops tracking a chunk range between 'min' and 'max' that previously was having data - * migrated into it. This data is no longer protected against cleanup of orphaned data. - * - * To avoid removing pending ranges of other operations, ensure that this is only used when - * a migration is still active. - * TODO: Because migrations may currently be active when a collection drops, an epoch is - * necessary to ensure the pending metadata change is still applicable. - * - * @return false with errMsg if the range is owned by the shard or the epoch of the metadata - * has changed - */ - bool forgetPending(OperationContext* txn, - const std::string& ns, - const BSONObj& min, - const BSONObj& max, - const OID& epoch, - std::string* errMsg ); - - /** - * Creates and installs a new chunk metadata for a given collection by splitting one of its - * chunks in two or more. The version for the first split chunk should be provided. The - * subsequent chunks' version would be the latter with the minor portion incremented. - * - * The effect on clients will depend on the version used. If the major portion is the same - * as the current shards, clients shouldn't perceive the split. - * - * @param ns the collection - * @param min max the chunk that should be split - * @param splitKeys point in which to split - * @param version at which the new metadata should be at - */ - void splitChunk(OperationContext* txn, - const std::string& ns, - const BSONObj& min, - const BSONObj& max, - const std::vector<BSONObj>& splitKeys, - ChunkVersion version ); - - /** - * Creates and installs a new chunk metadata for a given collection by merging a range of - * chunks ['minKey', 'maxKey') into a single chunk with version 'mergedVersion'. - * The current metadata must overlap the range completely and minKey and maxKey must not - * divide an existing chunk. - * - * The merged chunk version must have a greater version than the current shard version, - * and if it has a greater major version clients will need to reload metadata. - * - * @param ns the collection - * @param minKey maxKey the range which should be merged - * @param newShardVersion the shard version the newly merged chunk should have - */ - void mergeChunks(OperationContext* txn, - const std::string& ns, - const BSONObj& minKey, - const BSONObj& maxKey, - ChunkVersion mergedVersion ); - - bool inCriticalMigrateSection(); - - /** - * @return true if we are NOT in the critical section - */ - bool waitTillNotInCriticalSection( int maxSecondsToWait ); - - /** - * TESTING ONLY - * Uninstalls the metadata for a given collection. - */ - void resetMetadata( const std::string& ns ); - - private: - - void _initialize(const std::string& server); - - /** - * Refreshes collection metadata by asking the config server for the latest information. - * May or may not be based on a requested version. - */ - Status doRefreshMetadata( OperationContext* txn, - const std::string& ns, - const ChunkVersion& reqShardVersion, - bool useRequestedVersion, - ChunkVersion* latestShardVersion ); - - // protects state below - stdx::mutex _mutex; - - // Whether ::initialize has been called - bool _enabled; - - // Sets the shard name for this host (comes through setShardVersion) - std::string _shardName; - - // protects accessing the config server - // Using a ticket holder so we can have multiple redundant tries at any given time - mutable TicketHolder _configServerTickets; - - // Map from a namespace into the metadata we need for each collection on this shard - typedef std::map<std::string,CollectionMetadataPtr> CollectionMetadataMap; - CollectionMetadataMap _collMetadata; - }; - - extern ShardingState shardingState; +class Database; +class RecordId; +class OperationContext; + +// -------------- +// --- global state --- +// -------------- + +class ShardingState { +public: + ShardingState(); + + bool enabled(); + std::string getConfigServer(); + + // Initialize sharding state and begin authenticating outgoing connections and handling + // shard versions. If this is not run before sharded operations occur auth will not work + // and versions will not be tracked. + static void initialize(const std::string& server); + + void gotShardName(const std::string& name); + bool setShardName(const std::string& name); // Same as above, does not throw + std::string getShardName(); + + // Helpers for SetShardVersion which report the host name sent to this shard when the shard + // name does not match. Do not use in other places. + // TODO: Remove once SSV is deprecated + void gotShardNameAndHost(const std::string& name, const std::string& host); + bool setShardNameAndHost(const std::string& name, const std::string& host); /** - * one per connection from mongos - * holds version state for each namespace + * Clears the collection metadata cache after step down. */ - class ShardedConnectionInfo { - public: + void clearCollectionMetadata(); - ShardedConnectionInfo(); + // versioning support - const ChunkVersion getVersion( const std::string& ns ) const; - void setVersion( const std::string& ns , const ChunkVersion& version ); + bool hasVersion(const std::string& ns); + bool hasVersion(const std::string& ns, ChunkVersion& version); + ChunkVersion getVersion(const std::string& ns); - static ShardedConnectionInfo* get( Client* client, bool create ); - static void reset( Client* client ); - static void addHook(); + /** + * If the metadata for 'ns' at this shard is at or above the requested version, + * 'reqShardVersion', returns OK and fills in 'latestShardVersion' with the latest shard + * version. The latter is always greater or equal than 'reqShardVersion' if in the same + * epoch. + * + * Otherwise, falls back to refreshMetadataNow. + * + * This call blocks if there are more than N threads + * currently refreshing metadata. (N is the number of + * tickets in ShardingState::_configServerTickets, + * currently 3.) + * + * Locking Note: + * + Must NOT be called with the write lock because this call may go into the network, + * and deadlocks may occur with shard-as-a-config. Therefore, nothing here guarantees + * that 'latestShardVersion' is indeed the current one on return. + */ + Status refreshMetadataIfNeeded(OperationContext* txn, + const std::string& ns, + const ChunkVersion& reqShardVersion, + ChunkVersion* latestShardVersion); + + /** + * Refreshes collection metadata by asking the config server for the latest information. + * Starts a new config server request. + * + * Locking Notes: + * + Must NOT be called with the write lock because this call may go into the network, + * and deadlocks may occur with shard-as-a-config. Therefore, nothing here guarantees + * that 'latestShardVersion' is indeed the current one on return. + * + * + Because this call must not be issued with the DBLock held, by the time the config + * server sent us back the collection metadata information, someone else may have + * updated the previously stored collection metadata. There are cases when one can't + * tell which of updated or loaded metadata are the freshest. There are also cases where + * the data coming from configs do not correspond to a consistent snapshot. + * In these cases, return RemoteChangeDetected. (This usually means this call needs to + * be issued again, at caller discretion) + * + * @return OK if remote metadata successfully loaded (may or may not have been installed) + * @return RemoteChangeDetected if something changed while reloading and we may retry + * @return !OK if something else went wrong during reload + * @return latestShardVersion the version that is now stored for this collection + */ + Status refreshMetadataNow(OperationContext* txn, + const std::string& ns, + ChunkVersion* latestShardVersion); - bool inForceVersionOkMode() const { - return _forceVersionOk; - } + void appendInfo(BSONObjBuilder& b); - void enterForceVersionOkMode() { _forceVersionOk = true; } - void leaveForceVersionOkMode() { _forceVersionOk = false; } + // querying support - private: + bool needCollectionMetadata(Client* client, const std::string& ns) const; + CollectionMetadataPtr getCollectionMetadata(const std::string& ns); - bool _forceVersionOk; // if this is true, then chunk version #s aren't check, and all ops are allowed + // chunk migrate and split support - typedef std::map<std::string,ChunkVersion> NSVersionMap; - NSVersionMap _versions; - }; + /** + * Creates and installs a new chunk metadata for a given collection by "forgetting" about + * one of its chunks. The new metadata uses the provided version, which has to be higher + * than the current metadata's shard version. + * + * One exception: if the forgotten chunk is the last one in this shard for the collection, + * version has to be 0. + * + * If it runs successfully, clients need to grab the new version to access the collection. + * + * LOCKING NOTE: + * Only safe to do inside the + * + * @param ns the collection + * @param min max the chunk to eliminate from the current metadata + * @param version at which the new metadata should be at + */ + void donateChunk(OperationContext* txn, + const std::string& ns, + const BSONObj& min, + const BSONObj& max, + ChunkVersion version); - struct ShardForceVersionOkModeBlock { - ShardForceVersionOkModeBlock(Client* client) { - info = ShardedConnectionInfo::get( client, false ); - if ( info ) - info->enterForceVersionOkMode(); - } - ~ShardForceVersionOkModeBlock() { - if ( info ) - info->leaveForceVersionOkMode(); - } + /** + * Creates and installs new chunk metadata for a given collection by reclaiming a previously + * donated chunk. The previous metadata's shard version has to be provided. + * + * If it runs successfully, clients that became stale by the previous donateChunk will be + * able to access the collection again. + * + * Note: If a migration has aborted but not yet unregistered a pending chunk, replacing the + * metadata may leave the chunk as pending - this is not dangerous and should be rare, but + * will require a stepdown to fully recover. + * + * @param ns the collection + * @param prevMetadata the previous metadata before we donated a chunk + */ + void undoDonateChunk(OperationContext* txn, + const std::string& ns, + CollectionMetadataPtr prevMetadata); - ShardedConnectionInfo * info; - }; + /** + * Remembers a chunk range between 'min' and 'max' as a range which will have data migrated + * into it. This data can then be protected against cleanup of orphaned data. + * + * Overlapping pending ranges will be removed, so it is only safe to use this when you know + * your metadata view is definitive, such as at the start of a migration. + * + * @return false with errMsg if the range is owned by this shard + */ + bool notePending(OperationContext* txn, + const std::string& ns, + const BSONObj& min, + const BSONObj& max, + const OID& epoch, + std::string* errMsg); - // ----------------- - // --- core --- - // ----------------- + /** + * Stops tracking a chunk range between 'min' and 'max' that previously was having data + * migrated into it. This data is no longer protected against cleanup of orphaned data. + * + * To avoid removing pending ranges of other operations, ensure that this is only used when + * a migration is still active. + * TODO: Because migrations may currently be active when a collection drops, an epoch is + * necessary to ensure the pending metadata change is still applicable. + * + * @return false with errMsg if the range is owned by the shard or the epoch of the metadata + * has changed + */ + bool forgetPending(OperationContext* txn, + const std::string& ns, + const BSONObj& min, + const BSONObj& max, + const OID& epoch, + std::string* errMsg); /** - * @return true if we have any shard info for the ns + * Creates and installs a new chunk metadata for a given collection by splitting one of its + * chunks in two or more. The version for the first split chunk should be provided. The + * subsequent chunks' version would be the latter with the minor portion incremented. + * + * The effect on clients will depend on the version used. If the major portion is the same + * as the current shards, clients shouldn't perceive the split. + * + * @param ns the collection + * @param min max the chunk that should be split + * @param splitKeys point in which to split + * @param version at which the new metadata should be at */ - bool haveLocalShardingInfo( Client* client, const std::string& ns ); + void splitChunk(OperationContext* txn, + const std::string& ns, + const BSONObj& min, + const BSONObj& max, + const std::vector<BSONObj>& splitKeys, + ChunkVersion version); /** - * Validates whether the shard chunk version for the specified collection is up to date and if - * not, throws SendStaleConfigException. + * Creates and installs a new chunk metadata for a given collection by merging a range of + * chunks ['minKey', 'maxKey') into a single chunk with version 'mergedVersion'. + * The current metadata must overlap the range completely and minKey and maxKey must not + * divide an existing chunk. * - * It is important (but not enforced) that method be called with the collection locked in at - * least IS mode in order to ensure that the shard version won't change. + * The merged chunk version must have a greater version than the current shard version, + * and if it has a greater major version clients will need to reload metadata. * - * @param ns Complete collection namespace to be cheched. + * @param ns the collection + * @param minKey maxKey the range which should be merged + * @param newShardVersion the shard version the newly merged chunk should have + */ + void mergeChunks(OperationContext* txn, + const std::string& ns, + const BSONObj& minKey, + const BSONObj& maxKey, + ChunkVersion mergedVersion); + + bool inCriticalMigrateSection(); + + /** + * @return true if we are NOT in the critical section + */ + bool waitTillNotInCriticalSection(int maxSecondsToWait); + + /** + * TESTING ONLY + * Uninstalls the metadata for a given collection. */ - void ensureShardVersionOKOrThrow(Client* client, const std::string& ns); + void resetMetadata(const std::string& ns); + +private: + void _initialize(const std::string& server); /** - * If a migration for the chunk in 'ns' where 'obj' lives is occurring, save this log entry - * if it's relevant. The entries saved here are later transferred to the receiving side of - * the migration. A relevant entry is an insertion, a deletion, or an update. + * Refreshes collection metadata by asking the config server for the latest information. + * May or may not be based on a requested version. */ - void logOpForSharding( OperationContext* txn, - const char * opstr, - const char * ns, - const BSONObj& obj, - BSONObj * patt, - bool forMigrateCleanup ); + Status doRefreshMetadata(OperationContext* txn, + const std::string& ns, + const ChunkVersion& reqShardVersion, + bool useRequestedVersion, + ChunkVersion* latestShardVersion); + + // protects state below + stdx::mutex _mutex; + + // Whether ::initialize has been called + bool _enabled; + + // Sets the shard name for this host (comes through setShardVersion) + std::string _shardName; + + // protects accessing the config server + // Using a ticket holder so we can have multiple redundant tries at any given time + mutable TicketHolder _configServerTickets; + + // Map from a namespace into the metadata we need for each collection on this shard + typedef std::map<std::string, CollectionMetadataPtr> CollectionMetadataMap; + CollectionMetadataMap _collMetadata; +}; + +extern ShardingState shardingState; +/** + * one per connection from mongos + * holds version state for each namespace + */ +class ShardedConnectionInfo { +public: + ShardedConnectionInfo(); + + const ChunkVersion getVersion(const std::string& ns) const; + void setVersion(const std::string& ns, const ChunkVersion& version); + + static ShardedConnectionInfo* get(Client* client, bool create); + static void reset(Client* client); + static void addHook(); + + bool inForceVersionOkMode() const { + return _forceVersionOk; + } + + void enterForceVersionOkMode() { + _forceVersionOk = true; + } + void leaveForceVersionOkMode() { + _forceVersionOk = false; + } + +private: + bool + _forceVersionOk; // if this is true, then chunk version #s aren't check, and all ops are allowed + + typedef std::map<std::string, ChunkVersion> NSVersionMap; + NSVersionMap _versions; +}; + +struct ShardForceVersionOkModeBlock { + ShardForceVersionOkModeBlock(Client* client) { + info = ShardedConnectionInfo::get(client, false); + if (info) + info->enterForceVersionOkMode(); + } + ~ShardForceVersionOkModeBlock() { + if (info) + info->leaveForceVersionOkMode(); + } + + ShardedConnectionInfo* info; +}; + +// ----------------- +// --- core --- +// ----------------- + +/** + * @return true if we have any shard info for the ns + */ +bool haveLocalShardingInfo(Client* client, const std::string& ns); + +/** + * Validates whether the shard chunk version for the specified collection is up to date and if + * not, throws SendStaleConfigException. + * + * It is important (but not enforced) that method be called with the collection locked in at + * least IS mode in order to ensure that the shard version won't change. + * + * @param ns Complete collection namespace to be cheched. + */ +void ensureShardVersionOKOrThrow(Client* client, const std::string& ns); + +/** + * If a migration for the chunk in 'ns' where 'obj' lives is occurring, save this log entry + * if it's relevant. The entries saved here are later transferred to the receiving side of + * the migration. A relevant entry is an insertion, a deletion, or an update. + */ +void logOpForSharding(OperationContext* txn, + const char* opstr, + const char* ns, + const BSONObj& obj, + BSONObj* patt, + bool forMigrateCleanup); } diff --git a/src/mongo/s/dbclient_shard_resolver.cpp b/src/mongo/s/dbclient_shard_resolver.cpp index 5caef187a73..92694a51243 100644 --- a/src/mongo/s/dbclient_shard_resolver.cpp +++ b/src/mongo/s/dbclient_shard_resolver.cpp @@ -37,67 +37,63 @@ namespace mongo { - using std::string; +using std::string; + +Status DBClientShardResolver::chooseWriteHost(const string& shardName, + ConnectionString* shardHost) const { + // Internally uses our shard cache, does no reload + std::shared_ptr<Shard> shard = grid.shardRegistry()->getShard(shardName); + if (!shard) { + return Status(ErrorCodes::ShardNotFound, + str::stream() << "unknown shard name " << shardName); + } + + return findMaster(shard->getConnString(), shardHost); +} + +Status DBClientShardResolver::findMaster(const ConnectionString& connString, + ConnectionString* resolvedHost) { + if (connString.type() == ConnectionString::MASTER) { + *resolvedHost = connString; + return Status::OK(); + } - Status DBClientShardResolver::chooseWriteHost(const string& shardName, - ConnectionString* shardHost) const { + dassert(connString.type() == ConnectionString::SET); - // Internally uses our shard cache, does no reload - std::shared_ptr<Shard> shard = grid.shardRegistry()->getShard(shardName); - if (!shard) { - return Status(ErrorCodes::ShardNotFound, - str::stream() << "unknown shard name " << shardName); - } + // + // If we need to, then get the particular node we're targeting in the replica set + // + + // Don't create the monitor unless we need to - fast path + ReplicaSetMonitorPtr replMonitor = ReplicaSetMonitor::get(connString.getSetName()); + + if (!replMonitor) { + // Slow path + std::set<HostAndPort> seedServers(connString.getServers().begin(), + connString.getServers().end()); + ReplicaSetMonitor::createIfNeeded(connString.getSetName(), seedServers); + + replMonitor = ReplicaSetMonitor::get(connString.getSetName()); + } - return findMaster(shard->getConnString(), shardHost); + if (!replMonitor) { + return Status(ErrorCodes::ReplicaSetNotFound, + str::stream() << "unknown replica set " << connString.getSetName()); } - Status DBClientShardResolver::findMaster(const ConnectionString& connString, - ConnectionString* resolvedHost) { - - if (connString.type() == ConnectionString::MASTER) { - *resolvedHost = connString; - return Status::OK(); - } - - dassert(connString.type() == ConnectionString::SET); - - // - // If we need to, then get the particular node we're targeting in the replica set - // - - // Don't create the monitor unless we need to - fast path - ReplicaSetMonitorPtr replMonitor = ReplicaSetMonitor::get(connString.getSetName()); - - if (!replMonitor) { - // Slow path - std::set<HostAndPort> seedServers(connString.getServers().begin(), - connString.getServers().end()); - ReplicaSetMonitor::createIfNeeded(connString.getSetName(), seedServers); - - replMonitor = ReplicaSetMonitor::get(connString.getSetName()); - } - - if (!replMonitor) { - return Status(ErrorCodes::ReplicaSetNotFound, - str::stream() << "unknown replica set " << connString.getSetName()); - } - - try { - // This can throw when we don't find a master! - HostAndPort masterHostAndPort = replMonitor->getMasterOrUassert(); - *resolvedHost = fassertStatusOK(28687, - ConnectionString::parse(masterHostAndPort.toString())); - return Status::OK(); - } - catch ( const DBException& ) { - return Status( ErrorCodes::HostNotFound, - string("could not contact primary for replica set ") - + replMonitor->getName() ); - } - - MONGO_UNREACHABLE; + try { + // This can throw when we don't find a master! + HostAndPort masterHostAndPort = replMonitor->getMasterOrUassert(); + *resolvedHost = + fassertStatusOK(28687, ConnectionString::parse(masterHostAndPort.toString())); + return Status::OK(); + } catch (const DBException&) { + return Status(ErrorCodes::HostNotFound, + string("could not contact primary for replica set ") + + replMonitor->getName()); } -} // namespace mongo + MONGO_UNREACHABLE; +} +} // namespace mongo diff --git a/src/mongo/s/dbclient_shard_resolver.h b/src/mongo/s/dbclient_shard_resolver.h index fb7beb26ee5..076dd078562 100644 --- a/src/mongo/s/dbclient_shard_resolver.h +++ b/src/mongo/s/dbclient_shard_resolver.h @@ -32,42 +32,37 @@ namespace mongo { +/** + * ShardResolver based on the Shard and ReplicaSetMonitor caches. + * + * TODO: Currently it's possible for the shard resolver to be stale after we target and remove + * a shard. We need to figure out how to refresh. + */ +class DBClientShardResolver : public ShardResolver { +public: + DBClientShardResolver() {} + + virtual ~DBClientShardResolver() {} + /** - * ShardResolver based on the Shard and ReplicaSetMonitor caches. + * Returns the current host ConnectionString for a write to a shard. + * + * Note: Does *not* trigger a refresh of either the shard or replica set monitor caches, + * though refreshes may happen unexpectedly between calls. * - * TODO: Currently it's possible for the shard resolver to be stale after we target and remove - * a shard. We need to figure out how to refresh. + * Returns ShardNotFound if the shard name is unknown + * Returns ReplicaSetNotFound if the replica set is not being tracked + * Returns !OK with message if the shard host could not be found for other reasons. */ - class DBClientShardResolver : public ShardResolver { - public: + Status chooseWriteHost(const std::string& shardName, ConnectionString* shardHost) const; - DBClientShardResolver() { - } - - virtual ~DBClientShardResolver() { - } - - /** - * Returns the current host ConnectionString for a write to a shard. - * - * Note: Does *not* trigger a refresh of either the shard or replica set monitor caches, - * though refreshes may happen unexpectedly between calls. - * - * Returns ShardNotFound if the shard name is unknown - * Returns ReplicaSetNotFound if the replica set is not being tracked - * Returns !OK with message if the shard host could not be found for other reasons. - */ - Status chooseWriteHost( const std::string& shardName, ConnectionString* shardHost ) const; - - /** - * Resolves a replica set connection string to a master or returns an error. - * - * Returns HostNotFound if the master is not reachable - * Returns ReplicaSetNotFound if the replica set is not being tracked - */ - static Status findMaster(const ConnectionString& connString, - ConnectionString* resolvedHost); - - }; + /** + * Resolves a replica set connection string to a master or returns an error. + * + * Returns HostNotFound if the master is not reachable + * Returns ReplicaSetNotFound if the replica set is not being tracked + */ + static Status findMaster(const ConnectionString& connString, ConnectionString* resolvedHost); +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/distlock_test.cpp b/src/mongo/s/distlock_test.cpp index 9b1dd3f8ab3..d0fea0b3448 100644 --- a/src/mongo/s/distlock_test.cpp +++ b/src/mongo/s/distlock_test.cpp @@ -73,375 +73,372 @@ // TODO: Make a method in BSONObj if useful, don't modify for now -#define string_field(obj, name, def) ( obj.hasField(name) ? obj[name].String() : def ) -#define number_field(obj, name, def) ( obj.hasField(name) ? obj[name].Number() : def ) +#define string_field(obj, name, def) (obj.hasField(name) ? obj[name].String() : def) +#define number_field(obj, name, def) (obj.hasField(name) ? obj[name].Number() : def) namespace mongo { - using std::shared_ptr; - using std::endl; - using std::string; - using std::stringstream; - using std::vector; - - /** - * Stress test distributed lock by running multiple threads to contend with a single lock. - * Also has an option to make some thread terminate while holding the lock and have some - * other thread take over it after takeoverMS has elapsed. Note that this test does not check - * whether the lock was eventually overtaken and this is only valid if the LockPinger frequency - * is faster than takeoverMS. - * - * { - * _testDistLockWithSkew: 1, - * - * lockName: <string for distributed lock>, - * host: <connection string for config server>, - * seed: <numeric seed for random generator>, - * numThreads: <num of threads to spawn and grab the lock>, - * - * takeoverMS: <duration of missed ping in milliSeconds before a lock can be overtaken>, - * wait: <time in milliseconds before stopping the test threads>, - * skewHosts: <Array<Numeric>, numbers to be used when calling _skewClockCommand - * against each config server>, - * threadWait: <upper bound wait in milliSeconds while holding a lock>, - * - * hangThreads: <integer n, where 1 out of n threads will abort after acquiring lock>, - * threadSleep: <upper bound sleep duration in mSecs between each round of lock operation>, - * skewRange: <maximum skew variance in milliSeconds for a thread's clock, delta will never - * be greater than skewRange/2> - * } - */ - class TestDistLockWithSkew: public Command { - public: - - static const int logLvl = 1; +using std::shared_ptr; +using std::endl; +using std::string; +using std::stringstream; +using std::vector; + +/** + * Stress test distributed lock by running multiple threads to contend with a single lock. + * Also has an option to make some thread terminate while holding the lock and have some + * other thread take over it after takeoverMS has elapsed. Note that this test does not check + * whether the lock was eventually overtaken and this is only valid if the LockPinger frequency + * is faster than takeoverMS. + * + * { + * _testDistLockWithSkew: 1, + * + * lockName: <string for distributed lock>, + * host: <connection string for config server>, + * seed: <numeric seed for random generator>, + * numThreads: <num of threads to spawn and grab the lock>, + * + * takeoverMS: <duration of missed ping in milliSeconds before a lock can be overtaken>, + * wait: <time in milliseconds before stopping the test threads>, + * skewHosts: <Array<Numeric>, numbers to be used when calling _skewClockCommand + * against each config server>, + * threadWait: <upper bound wait in milliSeconds while holding a lock>, + * + * hangThreads: <integer n, where 1 out of n threads will abort after acquiring lock>, + * threadSleep: <upper bound sleep duration in mSecs between each round of lock operation>, + * skewRange: <maximum skew variance in milliSeconds for a thread's clock, delta will never + * be greater than skewRange/2> + * } + */ +class TestDistLockWithSkew : public Command { +public: + static const int logLvl = 1; - TestDistLockWithSkew() : - Command("_testDistLockWithSkew") { - } - virtual void help(stringstream& help) const { - help << "should not be calling this directly" << endl; - } + TestDistLockWithSkew() : Command("_testDistLockWithSkew") {} + virtual void help(stringstream& help) const { + help << "should not be calling this directly" << endl; + } - virtual bool slaveOk() const { - return false; - } - virtual bool adminOnly() const { - return true; + virtual bool slaveOk() const { + return false; + } + virtual bool adminOnly() const { + return true; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + // No auth needed because it only works when enabled via command line. + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) {} + + void runThread(ConnectionString& hostConn, + unsigned threadId, + unsigned seed, + BSONObj& cmdObj, + BSONObjBuilder& result) { + stringstream ss; + ss << "thread-" << threadId; + setThreadName(ss.str().c_str()); + + // Lock name + string lockName = string_field(cmdObj, "lockName", this->name + "_lock"); + + // Range of clock skew in diff threads + int skewRange = (int)number_field(cmdObj, "skewRange", 1); + + // How long to wait with the lock + int threadWait = (int)number_field(cmdObj, "threadWait", 30); + if (threadWait <= 0) + threadWait = 1; + + // Max amount of time (ms) a thread waits before checking the lock again + int threadSleep = (int)number_field(cmdObj, "threadSleep", 30); + if (threadSleep <= 0) + threadSleep = 1; + + // How long until the lock is forced in ms, only compared locally + unsigned long long takeoverMS = (unsigned long long)number_field(cmdObj, "takeoverMS", 0); + + // Whether or not we should hang some threads + int hangThreads = (int)number_field(cmdObj, "hangThreads", 0); + + + boost::mt19937 gen((boost::mt19937::result_type)seed); + + boost::variate_generator<boost::mt19937&, boost::uniform_int<>> randomSkew( + gen, boost::uniform_int<>(0, skewRange)); + boost::variate_generator<boost::mt19937&, boost::uniform_int<>> randomWait( + gen, boost::uniform_int<>(1, threadWait)); + boost::variate_generator<boost::mt19937&, boost::uniform_int<>> randomSleep( + gen, boost::uniform_int<>(1, threadSleep)); + boost::variate_generator<boost::mt19937&, boost::uniform_int<>> randomNewLock( + gen, boost::uniform_int<>(0, 3)); + + + int skew = 0; + if (!lock.get()) { + // Pick a skew, but the first two threads skew the whole range + if (threadId == 0) + skew = -skewRange / 2; + else if (threadId == 1) + skew = skewRange / 2; + else + skew = randomSkew() - (skewRange / 2); + + // Skew this thread + jsTimeVirtualThreadSkew(skew); + + log() << "Initializing lock with skew of " << skew << " for thread " << threadId + << endl; + + lock.reset(new DistributedLock(hostConn, lockName, takeoverMS, true)); + + log() << "Skewed time " << jsTime() << " for thread " << threadId << endl + << " max wait (with lock: " << threadWait << ", after lock: " << threadSleep + << ")" << endl + << " takeover in " << takeoverMS << "(ms remote)" << endl; } - virtual bool isWriteCommandForConfigServer() const { return false; } - // No auth needed because it only works when enabled via command line. - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) {} - - void runThread(ConnectionString& hostConn, unsigned threadId, unsigned seed, - BSONObj& cmdObj, BSONObjBuilder& result) { - - stringstream ss; - ss << "thread-" << threadId; - setThreadName(ss.str().c_str()); - - // Lock name - string lockName = string_field(cmdObj, "lockName", this->name + "_lock"); - - // Range of clock skew in diff threads - int skewRange = (int) number_field(cmdObj, "skewRange", 1); - - // How long to wait with the lock - int threadWait = (int) number_field(cmdObj, "threadWait", 30); - if(threadWait <= 0) threadWait = 1; - - // Max amount of time (ms) a thread waits before checking the lock again - int threadSleep = (int) number_field(cmdObj, "threadSleep", 30); - if(threadSleep <= 0) threadSleep = 1; - - // How long until the lock is forced in ms, only compared locally - unsigned long long takeoverMS = (unsigned long long) number_field(cmdObj, "takeoverMS", 0); - // Whether or not we should hang some threads - int hangThreads = (int) number_field(cmdObj, "hangThreads", 0); + DistributedLock* myLock = lock.get(); + bool errors = false; + BSONObj lockObj; + while (keepGoing.loadRelaxed()) { + Status pingStatus = _pinger.startPing( + *myLock, stdx::chrono::milliseconds(takeoverMS / LOCK_SKEW_FACTOR)); - boost::mt19937 gen((boost::mt19937::result_type) seed); - - boost::variate_generator<boost::mt19937&, boost::uniform_int<> > randomSkew(gen, boost::uniform_int<>(0, skewRange)); - boost::variate_generator<boost::mt19937&, boost::uniform_int<> > randomWait(gen, boost::uniform_int<>(1, threadWait)); - boost::variate_generator<boost::mt19937&, boost::uniform_int<> > randomSleep(gen, boost::uniform_int<>(1, threadSleep)); - boost::variate_generator<boost::mt19937&, boost::uniform_int<> > randomNewLock(gen, boost::uniform_int<>(0, 3)); - - - int skew = 0; - if (!lock.get()) { - - // Pick a skew, but the first two threads skew the whole range - if(threadId == 0) - skew = -skewRange / 2; - else if(threadId == 1) - skew = skewRange / 2; - else skew = randomSkew() - (skewRange / 2); - - // Skew this thread - jsTimeVirtualThreadSkew( skew ); - - log() << "Initializing lock with skew of " << skew << " for thread " << threadId << endl; - - lock.reset(new DistributedLock(hostConn, lockName, takeoverMS, true )); - - log() << "Skewed time " << jsTime() << " for thread " << threadId << endl - << " max wait (with lock: " << threadWait << ", after lock: " << threadSleep << ")" << endl - << " takeover in " << takeoverMS << "(ms remote)" << endl; - + if (!pingStatus.isOK()) { + log() << "**** Not good for pinging: " << pingStatus; + break; } - DistributedLock* myLock = lock.get(); - - bool errors = false; - BSONObj lockObj; - while (keepGoing.loadRelaxed()) { - Status pingStatus = _pinger.startPing(*myLock, - stdx::chrono::milliseconds(takeoverMS / LOCK_SKEW_FACTOR)); + try { + if (myLock->lock_try("Testing distributed lock with skew.", &lockObj)) { + log() << "**** Locked for thread " << threadId << " with ts " << lockObj["ts"] + << endl; + + if (count.loadRelaxed() % 3 == 1 && + myLock->lock_try("Testing lock non-re-entry.")) { + errors = true; + log() << "**** !Invalid lock re-entry" << endl; + break; + } - if (!pingStatus.isOK()) { - log() << "**** Not good for pinging: " << pingStatus; - break; - } + int before = count.addAndFetch(1); + int sleep = randomWait(); + sleepmillis(sleep); + int after = count.loadRelaxed(); - try { - - if (myLock->lock_try("Testing distributed lock with skew.", &lockObj)) { - - log() << "**** Locked for thread " << threadId << " with ts " << lockObj["ts"] << endl; - - if (count.loadRelaxed() % 3 == 1 && - myLock->lock_try( "Testing lock non-re-entry.")) { - errors = true; - log() << "**** !Invalid lock re-entry" << endl; - break; - } - - int before = count.addAndFetch(1); - int sleep = randomWait(); - sleepmillis(sleep); - int after = count.loadRelaxed(); - - if(after != before) { - errors = true; - log() << "**** !Bad increment while sleeping with lock for: " << sleep << "ms" << endl; - break; - } - - // Unlock only half the time... - if(hangThreads == 0 || threadId % hangThreads != 0) { - log() << "**** Unlocking for thread " << threadId << " with ts " << lockObj["ts"] << endl; - myLock->unlock(lockObj["ts"].OID()); - } - else { - log() << "**** Not unlocking for thread " << threadId << endl; - _pinger.stopPing(myLock->getRemoteConnection(), myLock->getProcessId()); - // We're simulating a crashed process... - break; - } + if (after != before) { + errors = true; + log() << "**** !Bad increment while sleeping with lock for: " << sleep + << "ms" << endl; + break; } - } - catch( const DBException& ex ) { - log() << "*** !Could not try distributed lock." << causedBy( ex ) << endl; - break; - } - - // Create a new lock 1/3 of the time - if( randomNewLock() > 1 ){ - lock.reset(new DistributedLock( hostConn, lockName, takeoverMS, true )); - myLock = lock.get(); + // Unlock only half the time... + if (hangThreads == 0 || threadId % hangThreads != 0) { + log() << "**** Unlocking for thread " << threadId << " with ts " + << lockObj["ts"] << endl; + myLock->unlock(lockObj["ts"].OID()); + } else { + log() << "**** Not unlocking for thread " << threadId << endl; + _pinger.stopPing(myLock->getRemoteConnection(), myLock->getProcessId()); + // We're simulating a crashed process... + break; + } } - sleepmillis(randomSleep()); + } catch (const DBException& ex) { + log() << "*** !Could not try distributed lock." << causedBy(ex) << endl; + break; } - result << "errors" << errors - << "skew" << skew - << "takeover" << (long long) takeoverMS - << "localTimeout" << (takeoverMS > 0); - - } + // Create a new lock 1/3 of the time + if (randomNewLock() > 1) { + lock.reset(new DistributedLock(hostConn, lockName, takeoverMS, true)); + myLock = lock.get(); + } - void test(ConnectionString& hostConn, string& lockName, unsigned seed) { - return; + sleepmillis(randomSleep()); } - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { - - Timer t; - - ConnectionString hostConn(cmdObj["host"].String(), - ConnectionString::SYNC); - - unsigned seed = (unsigned) number_field(cmdObj, "seed", 0); - int numThreads = (int) number_field(cmdObj, "numThreads", 4); - int wait = (int) number_field(cmdObj, "wait", 10000); - - log() << "Starting " << this->name << " with -" << endl - << " seed: " << seed << endl - << " numThreads: " << numThreads << endl - << " total wait: " << wait << endl << endl; - - // Skew host clocks if needed - try { - skewClocks( hostConn, cmdObj ); - } - catch( DBException e ) { - errmsg = str::stream() << "Clocks could not be skewed." << causedBy( e ); - return false; - } - - count.store(0); - keepGoing.store(true); - - vector<shared_ptr<stdx::thread> > threads; - vector<shared_ptr<BSONObjBuilder> > results; - for (int i = 0; i < numThreads; i++) { - results.push_back(shared_ptr<BSONObjBuilder> (new BSONObjBuilder())); - threads.push_back(shared_ptr<stdx::thread> (new stdx::thread( - stdx::bind(&TestDistLockWithSkew::runThread, this, - hostConn, (unsigned) i, seed + i, boost::ref(cmdObj), - boost::ref(*(results[i].get())))))); - } + result << "errors" << errors << "skew" << skew << "takeover" << (long long)takeoverMS + << "localTimeout" << (takeoverMS > 0); + } - sleepsecs(wait / 1000); - keepGoing.store(false); + void test(ConnectionString& hostConn, string& lockName, unsigned seed) { + return; + } - bool errors = false; - for (unsigned i = 0; i < threads.size(); i++) { - threads[i]->join(); - errors = errors || results[i].get()->obj()["errors"].Bool(); - } + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + Timer t; + + ConnectionString hostConn(cmdObj["host"].String(), ConnectionString::SYNC); + + unsigned seed = (unsigned)number_field(cmdObj, "seed", 0); + int numThreads = (int)number_field(cmdObj, "numThreads", 4); + int wait = (int)number_field(cmdObj, "wait", 10000); + + log() << "Starting " << this->name << " with -" << endl + << " seed: " << seed << endl + << " numThreads: " << numThreads << endl + << " total wait: " << wait << endl + << endl; + + // Skew host clocks if needed + try { + skewClocks(hostConn, cmdObj); + } catch (DBException e) { + errmsg = str::stream() << "Clocks could not be skewed." << causedBy(e); + return false; + } - result.append("count", count.loadRelaxed()); - result.append("errors", errors); - result.append("timeMS", t.millis()); + count.store(0); + keepGoing.store(true); + + vector<shared_ptr<stdx::thread>> threads; + vector<shared_ptr<BSONObjBuilder>> results; + for (int i = 0; i < numThreads; i++) { + results.push_back(shared_ptr<BSONObjBuilder>(new BSONObjBuilder())); + threads.push_back(shared_ptr<stdx::thread>( + new stdx::thread(stdx::bind(&TestDistLockWithSkew::runThread, + this, + hostConn, + (unsigned)i, + seed + i, + boost::ref(cmdObj), + boost::ref(*(results[i].get())))))); + } - return !errors; + sleepsecs(wait / 1000); + keepGoing.store(false); + bool errors = false; + for (unsigned i = 0; i < threads.size(); i++) { + threads[i]->join(); + errors = errors || results[i].get()->obj()["errors"].Bool(); } - /** - * Skews the clocks of a remote cluster by a particular amount, specified by - * the "skewHosts" element in a BSONObj. - */ - static void skewClocks( ConnectionString& cluster, BSONObj& cmdObj ) { - - vector<long long> skew; - if(cmdObj.hasField("skewHosts")) { - bsonArrToNumVector<long long>(cmdObj["skewHosts"], skew); - } - else { - LOG( logLvl ) << "No host clocks to skew." << endl; - return; - } + result.append("count", count.loadRelaxed()); + result.append("errors", errors); + result.append("timeMS", t.millis()); - LOG( logLvl ) << "Skewing clocks of hosts " << cluster << endl; + return !errors; + } - unsigned s = 0; - for(vector<long long>::iterator i = skew.begin(); i != skew.end(); ++i,s++) { + /** + * Skews the clocks of a remote cluster by a particular amount, specified by + * the "skewHosts" element in a BSONObj. + */ + static void skewClocks(ConnectionString& cluster, BSONObj& cmdObj) { + vector<long long> skew; + if (cmdObj.hasField("skewHosts")) { + bsonArrToNumVector<long long>(cmdObj["skewHosts"], skew); + } else { + LOG(logLvl) << "No host clocks to skew." << endl; + return; + } - ConnectionString server( cluster.getServers()[s] ); - ScopedDbConnection conn(server.toString()); + LOG(logLvl) << "Skewing clocks of hosts " << cluster << endl; - BSONObj result; - try { - bool success = conn->runCommand( string("admin"), - BSON( "_skewClockCommand" << 1 - << "skew" << *i ), - result ); + unsigned s = 0; + for (vector<long long>::iterator i = skew.begin(); i != skew.end(); ++i, s++) { + ConnectionString server(cluster.getServers()[s]); + ScopedDbConnection conn(server.toString()); - uassert(13678, str::stream() << "Could not communicate with server " << server.toString() << " in cluster " << cluster.toString() << " to change skew by " << *i, success ); + BSONObj result; + try { + bool success = conn->runCommand( + string("admin"), BSON("_skewClockCommand" << 1 << "skew" << *i), result); - LOG( logLvl + 1 ) << " Skewed host " << server << " clock by " << *i << endl; - } - catch(...) { - conn.done(); - throw; - } + uassert(13678, + str::stream() << "Could not communicate with server " << server.toString() + << " in cluster " << cluster.toString() + << " to change skew by " << *i, + success); + LOG(logLvl + 1) << " Skewed host " << server << " clock by " << *i << endl; + } catch (...) { conn.done(); - + throw; } + conn.done(); } - - // variables for test - boost::thread_specific_ptr<DistributedLock> lock; - AtomicUInt32 count; - AtomicWord<bool> keepGoing; - - private: - LegacyDistLockPinger _pinger; - }; - MONGO_INITIALIZER(RegisterDistLockWithSkewCmd)(InitializerContext* context) { - if (Command::testCommandsEnabled) { - // Leaked intentionally: a Command registers itself when constructed. - new TestDistLockWithSkew(); - } - return Status::OK(); } - /** - * Utility command to virtually skew the clock of a mongo server a particular amount. - * This skews the clock globally, per-thread skew is also possible. - */ - class SkewClockCommand: public Command { - public: - SkewClockCommand() : - Command("_skewClockCommand") { - } - virtual void help(stringstream& help) const { - help << "should not be calling this directly" << endl; - } - - virtual bool slaveOk() const { - return false; - } - virtual bool adminOnly() const { - return true; - } - virtual bool isWriteCommandForConfigServer() const { return false; } - // No auth needed because it only works when enabled via command line. - virtual void addRequiredPrivileges(const std::string& dbname, - const BSONObj& cmdObj, - std::vector<Privilege>* out) {} - - bool run(OperationContext* txn, - const string&, - BSONObj& cmdObj, - int, - string& errmsg, - BSONObjBuilder& result) { + // variables for test + boost::thread_specific_ptr<DistributedLock> lock; + AtomicUInt32 count; + AtomicWord<bool> keepGoing; + +private: + LegacyDistLockPinger _pinger; +}; +MONGO_INITIALIZER(RegisterDistLockWithSkewCmd)(InitializerContext* context) { + if (Command::testCommandsEnabled) { + // Leaked intentionally: a Command registers itself when constructed. + new TestDistLockWithSkew(); + } + return Status::OK(); +} - long long skew = (long long) number_field(cmdObj, "skew", 0); +/** + * Utility command to virtually skew the clock of a mongo server a particular amount. + * This skews the clock globally, per-thread skew is also possible. + */ +class SkewClockCommand : public Command { +public: + SkewClockCommand() : Command("_skewClockCommand") {} + virtual void help(stringstream& help) const { + help << "should not be calling this directly" << endl; + } - log() << "Adjusting jsTime() clock skew to " << skew << endl; + virtual bool slaveOk() const { + return false; + } + virtual bool adminOnly() const { + return true; + } + virtual bool isWriteCommandForConfigServer() const { + return false; + } + // No auth needed because it only works when enabled via command line. + virtual void addRequiredPrivileges(const std::string& dbname, + const BSONObj& cmdObj, + std::vector<Privilege>* out) {} - jsTimeVirtualSkew( skew ); + bool run(OperationContext* txn, + const string&, + BSONObj& cmdObj, + int, + string& errmsg, + BSONObjBuilder& result) { + long long skew = (long long)number_field(cmdObj, "skew", 0); - log() << "JSTime adjusted, now is " << jsTime() << endl; + log() << "Adjusting jsTime() clock skew to " << skew << endl; - return true; + jsTimeVirtualSkew(skew); - } + log() << "JSTime adjusted, now is " << jsTime() << endl; - }; - MONGO_INITIALIZER(RegisterSkewClockCmd)(InitializerContext* context) { - if (Command::testCommandsEnabled) { - // Leaked intentionally: a Command registers itself when constructed. - new SkewClockCommand(); - } - return Status::OK(); + return true; } +}; +MONGO_INITIALIZER(RegisterSkewClockCmd)(InitializerContext* context) { + if (Command::testCommandsEnabled) { + // Leaked intentionally: a Command registers itself when constructed. + new SkewClockCommand(); + } + return Status::OK(); +} } - diff --git a/src/mongo/s/grid.cpp b/src/mongo/s/grid.cpp index d687a7f8129..e36e1efac54 100644 --- a/src/mongo/s/grid.cpp +++ b/src/mongo/s/grid.cpp @@ -41,87 +41,83 @@ namespace mongo { - Grid::Grid() : _allowLocalShard(true) { +Grid::Grid() : _allowLocalShard(true) {} - } - - void Grid::init(std::unique_ptr<CatalogManager> catalogManager, - std::unique_ptr<ShardRegistry> shardRegistry) { +void Grid::init(std::unique_ptr<CatalogManager> catalogManager, + std::unique_ptr<ShardRegistry> shardRegistry) { + invariant(!_catalogManager); + invariant(!_catalogCache); + invariant(!_shardRegistry); - invariant(!_catalogManager); - invariant(!_catalogCache); - invariant(!_shardRegistry); + _catalogManager = std::move(catalogManager); + _catalogCache = stdx::make_unique<CatalogCache>(_catalogManager.get()); + _shardRegistry = std::move(shardRegistry); +} - _catalogManager = std::move(catalogManager); - _catalogCache = stdx::make_unique<CatalogCache>(_catalogManager.get()); - _shardRegistry = std::move(shardRegistry); +StatusWith<std::shared_ptr<DBConfig>> Grid::implicitCreateDb(const std::string& dbName) { + auto status = catalogCache()->getDatabase(dbName); + if (status.isOK()) { + return status; } - StatusWith<std::shared_ptr<DBConfig>> Grid::implicitCreateDb(const std::string& dbName) { - auto status = catalogCache()->getDatabase(dbName); - if (status.isOK()) { - return status; + if (status == ErrorCodes::DatabaseNotFound) { + auto statusCreateDb = catalogManager()->createDatabase(dbName); + if (statusCreateDb.isOK() || statusCreateDb == ErrorCodes::NamespaceExists) { + return catalogCache()->getDatabase(dbName); } - if (status == ErrorCodes::DatabaseNotFound) { - auto statusCreateDb = catalogManager()->createDatabase(dbName); - if (statusCreateDb.isOK() || statusCreateDb == ErrorCodes::NamespaceExists) { - return catalogCache()->getDatabase(dbName); - } + return statusCreateDb; + } - return statusCreateDb; - } + return status; +} - return status; - } +bool Grid::allowLocalHost() const { + return _allowLocalShard; +} - bool Grid::allowLocalHost() const { - return _allowLocalShard; +void Grid::setAllowLocalHost(bool allow) { + _allowLocalShard = allow; +} + +/* + * Returns whether balancing is enabled, with optional namespace "ns" parameter for balancing on a particular + * collection. + */ +bool Grid::shouldBalance(const SettingsType& balancerSettings) const { + if (balancerSettings.isBalancerStoppedSet() && balancerSettings.getBalancerStopped()) { + return false; } - void Grid::setAllowLocalHost( bool allow ) { - _allowLocalShard = allow; + if (balancerSettings.isBalancerActiveWindowSet()) { + boost::posix_time::ptime now = boost::posix_time::second_clock::local_time(); + return balancerSettings.inBalancingWindow(now); } - /* - * Returns whether balancing is enabled, with optional namespace "ns" parameter for balancing on a particular - * collection. - */ - bool Grid::shouldBalance(const SettingsType& balancerSettings) const { - if (balancerSettings.isBalancerStoppedSet() && balancerSettings.getBalancerStopped()) { - return false; - } + return true; +} - if (balancerSettings.isBalancerActiveWindowSet()) { - boost::posix_time::ptime now = boost::posix_time::second_clock::local_time(); - return balancerSettings.inBalancingWindow(now); - } +bool Grid::getConfigShouldBalance() const { + auto balSettingsResult = grid.catalogManager()->getGlobalSettings(SettingsType::BalancerDocKey); + if (!balSettingsResult.isOK()) { + warning() << balSettingsResult.getStatus(); + return false; + } + SettingsType balSettings = balSettingsResult.getValue(); + if (!balSettings.isKeySet()) { + // Balancer settings doc does not exist. Default to yes. return true; } - bool Grid::getConfigShouldBalance() const { - auto balSettingsResult = - grid.catalogManager()->getGlobalSettings(SettingsType::BalancerDocKey); - if (!balSettingsResult.isOK()) { - warning() << balSettingsResult.getStatus(); - return false; - } - SettingsType balSettings = balSettingsResult.getValue(); - - if (!balSettings.isKeySet()) { - // Balancer settings doc does not exist. Default to yes. - return true; - } - - return shouldBalance(balSettings); - } + return shouldBalance(balSettings); +} - void Grid::clearForUnitTests() { - _catalogManager.reset(); - _catalogCache.reset(); - _shardRegistry.reset(); - } +void Grid::clearForUnitTests() { + _catalogManager.reset(); + _catalogCache.reset(); + _shardRegistry.reset(); +} - Grid grid; +Grid grid; } diff --git a/src/mongo/s/grid.h b/src/mongo/s/grid.h index d15768ae21e..7262e71a629 100644 --- a/src/mongo/s/grid.h +++ b/src/mongo/s/grid.h @@ -35,82 +35,89 @@ namespace mongo { - class BSONObj; - class CatalogCache; - class CatalogManager; - class DBConfig; - class SettingsType; - class ShardRegistry; - template<typename T> class StatusWith; +class BSONObj; +class CatalogCache; +class CatalogManager; +class DBConfig; +class SettingsType; +class ShardRegistry; +template <typename T> +class StatusWith; +/** + * Holds the global sharding context. Single instance exists for a running server. Exists on + * both MongoD and MongoS. + */ +class Grid { +public: + Grid(); + /** - * Holds the global sharding context. Single instance exists for a running server. Exists on - * both MongoD and MongoS. + * Called at startup time so the global sharding services (catalog manager, shard registry) + * can be set. This method must be called once and once only for the lifetime of the + * service. + * + * NOTE: Unit-tests are allowed to call it more than once, provided they reset the object's + * state using clearForUnitTests. */ - class Grid { - public: - Grid(); - - /** - * Called at startup time so the global sharding services (catalog manager, shard registry) - * can be set. This method must be called once and once only for the lifetime of the - * service. - * - * NOTE: Unit-tests are allowed to call it more than once, provided they reset the object's - * state using clearForUnitTests. - */ - void init(std::unique_ptr<CatalogManager> catalogManager, - std::unique_ptr<ShardRegistry> shardRegistry); - - /** - * Implicitly creates the specified database as non-sharded. - */ - StatusWith<std::shared_ptr<DBConfig>> implicitCreateDb(const std::string& dbName); - - /** - * @return true if shards and config servers are allowed to use 'localhost' in address - */ - bool allowLocalHost() const; - - /** - * @param whether to allow shards and config servers to use 'localhost' in address - */ - void setAllowLocalHost( bool allow ); - - /** - * Returns true if the balancer should be running. Caller is responsible - * for making sure settings has the balancer key. - */ - bool shouldBalance(const SettingsType& balancerSettings) const; - - /** - * Returns true if the config server settings indicate that the balancer should be active. - */ - bool getConfigShouldBalance() const; - - CatalogManager* catalogManager() const { return _catalogManager.get(); } - CatalogCache* catalogCache() const { return _catalogCache.get(); } - ShardRegistry* shardRegistry() const { return _shardRegistry.get(); } - - /** - * Clears the grid object so that it can be reused between test executions. This will not - * be necessary if grid is hanging off the global ServiceContext and each test gets its - * own service context. - * - * NOTE: Do not use this outside of unit-tests. - */ - void clearForUnitTests(); - - private: - std::unique_ptr<CatalogManager> _catalogManager; - std::unique_ptr<CatalogCache> _catalogCache; - std::unique_ptr<ShardRegistry> _shardRegistry; - - // can 'localhost' be used in shard addresses? - bool _allowLocalShard; - }; - - extern Grid grid; - -} // namespace mongo + void init(std::unique_ptr<CatalogManager> catalogManager, + std::unique_ptr<ShardRegistry> shardRegistry); + + /** + * Implicitly creates the specified database as non-sharded. + */ + StatusWith<std::shared_ptr<DBConfig>> implicitCreateDb(const std::string& dbName); + + /** + * @return true if shards and config servers are allowed to use 'localhost' in address + */ + bool allowLocalHost() const; + + /** + * @param whether to allow shards and config servers to use 'localhost' in address + */ + void setAllowLocalHost(bool allow); + + /** + * Returns true if the balancer should be running. Caller is responsible + * for making sure settings has the balancer key. + */ + bool shouldBalance(const SettingsType& balancerSettings) const; + + /** + * Returns true if the config server settings indicate that the balancer should be active. + */ + bool getConfigShouldBalance() const; + + CatalogManager* catalogManager() const { + return _catalogManager.get(); + } + CatalogCache* catalogCache() const { + return _catalogCache.get(); + } + ShardRegistry* shardRegistry() const { + return _shardRegistry.get(); + } + + /** + * Clears the grid object so that it can be reused between test executions. This will not + * be necessary if grid is hanging off the global ServiceContext and each test gets its + * own service context. + * + * NOTE: Do not use this outside of unit-tests. + */ + void clearForUnitTests(); + +private: + std::unique_ptr<CatalogManager> _catalogManager; + std::unique_ptr<CatalogCache> _catalogCache; + std::unique_ptr<ShardRegistry> _shardRegistry; + + // can 'localhost' be used in shard addresses? + bool _allowLocalShard; +}; + +extern Grid grid; + +} // namespace mongo diff --git a/src/mongo/s/metadata_loader.cpp b/src/mongo/s/metadata_loader.cpp index 74ca1597a7d..1a0758f5734 100644 --- a/src/mongo/s/metadata_loader.cpp +++ b/src/mongo/s/metadata_loader.cpp @@ -44,265 +44,242 @@ namespace mongo { - using std::unique_ptr; - using std::endl; - using std::make_pair; - using std::map; - using std::pair; - using std::string; - - /** - * This is an adapter so we can use config diffs - mongos and mongod do them slightly - * differently. - * - * The mongod adapter here tracks only a single shard, and stores ranges by (min, max). - */ - class SCMConfigDiffTracker : public ConfigDiffTracker<BSONObj, string> { - public: - SCMConfigDiffTracker(const string& currShard) : _currShard( currShard ) { } - - virtual bool isTracked(const ChunkType& chunk) const { - return chunk.getShard() == _currShard; - } - - virtual pair<BSONObj, BSONObj> rangeFor(const ChunkType& chunk) const { - return make_pair(chunk.getMin(), chunk.getMax()); - } - - virtual string shardFor( const string& name ) const { - return name; - } - - virtual string nameFrom( const string& shard ) const { - return shard; - } +using std::unique_ptr; +using std::endl; +using std::make_pair; +using std::map; +using std::pair; +using std::string; - string _currShard; - }; - - // - // MetadataLoader implementation - // +/** + * This is an adapter so we can use config diffs - mongos and mongod do them slightly + * differently. + * + * The mongod adapter here tracks only a single shard, and stores ranges by (min, max). + */ +class SCMConfigDiffTracker : public ConfigDiffTracker<BSONObj, string> { +public: + SCMConfigDiffTracker(const string& currShard) : _currShard(currShard) {} - MetadataLoader::MetadataLoader() { } + virtual bool isTracked(const ChunkType& chunk) const { + return chunk.getShard() == _currShard; + } - MetadataLoader::~MetadataLoader() { } + virtual pair<BSONObj, BSONObj> rangeFor(const ChunkType& chunk) const { + return make_pair(chunk.getMin(), chunk.getMax()); + } - Status MetadataLoader::makeCollectionMetadata(CatalogManager* catalogManager, - const string& ns, - const string& shard, - const CollectionMetadata* oldMetadata, - CollectionMetadata* metadata) const - { - Status status = _initCollection(catalogManager, ns, shard, metadata); - if (!status.isOK() || metadata->getKeyPattern().isEmpty()) { - return status; - } + virtual string shardFor(const string& name) const { + return name; + } - return initChunks(catalogManager, ns, shard, oldMetadata, metadata ); + virtual string nameFrom(const string& shard) const { + return shard; } - Status MetadataLoader::_initCollection(CatalogManager* catalogManager, - const string& ns, - const string& shard, - CollectionMetadata* metadata) const { + string _currShard; +}; - auto coll = catalogManager->getCollection(ns); - if (!coll.isOK()) { - return coll.getStatus(); - } +// +// MetadataLoader implementation +// - CollectionType collInfo = coll.getValue(); - if (collInfo.getDropped()) { - return Status(ErrorCodes::NamespaceNotFound, - str::stream() << "could not load metadata, collection " - << ns << " was dropped"); - } +MetadataLoader::MetadataLoader() {} - metadata->_keyPattern = collInfo.getKeyPattern().toBSON(); - metadata->fillKeyPatternFields(); - metadata->_shardVersion = ChunkVersion(0, 0, collInfo.getEpoch()); - metadata->_collVersion = ChunkVersion(0, 0, collInfo.getEpoch()); +MetadataLoader::~MetadataLoader() {} - return Status::OK(); +Status MetadataLoader::makeCollectionMetadata(CatalogManager* catalogManager, + const string& ns, + const string& shard, + const CollectionMetadata* oldMetadata, + CollectionMetadata* metadata) const { + Status status = _initCollection(catalogManager, ns, shard, metadata); + if (!status.isOK() || metadata->getKeyPattern().isEmpty()) { + return status; } - Status MetadataLoader::initChunks(CatalogManager* catalogManager, - const string& ns, - const string& shard, - const CollectionMetadata* oldMetadata, - CollectionMetadata* metadata) const - { - map<string, ChunkVersion> versionMap; + return initChunks(catalogManager, ns, shard, oldMetadata, metadata); +} - // Preserve the epoch - versionMap[shard] = metadata->_shardVersion; - OID epoch = metadata->getCollVersion().epoch(); - bool fullReload = true; - - // Check to see if we should use the old version or not. - if ( oldMetadata ) { - - // If our epochs are compatible, it's useful to use the old metadata for diffs - if ( oldMetadata->getCollVersion().hasEqualEpoch( epoch ) ) { - - fullReload = false; - invariant( oldMetadata->isValid() ); - - versionMap[shard] = oldMetadata->_shardVersion; - metadata->_collVersion = oldMetadata->_collVersion; +Status MetadataLoader::_initCollection(CatalogManager* catalogManager, + const string& ns, + const string& shard, + CollectionMetadata* metadata) const { + auto coll = catalogManager->getCollection(ns); + if (!coll.isOK()) { + return coll.getStatus(); + } - // TODO: This could be made more efficient if copying not required, but - // not as frequently reloaded as in mongos. - metadata->_chunksMap = oldMetadata->_chunksMap; + CollectionType collInfo = coll.getValue(); + if (collInfo.getDropped()) { + return Status(ErrorCodes::NamespaceNotFound, + str::stream() << "could not load metadata, collection " << ns + << " was dropped"); + } - LOG( 2 ) << "loading new chunks for collection " << ns - << " using old metadata w/ version " << oldMetadata->getShardVersion() - << " and " << metadata->_chunksMap.size() << " chunks" << endl; - } - else { - warning() << "reloading collection metadata for " << ns << " with new epoch " - << epoch.toString() << ", the current epoch is " - << oldMetadata->getCollVersion().epoch().toString() << endl; - } + metadata->_keyPattern = collInfo.getKeyPattern().toBSON(); + metadata->fillKeyPatternFields(); + metadata->_shardVersion = ChunkVersion(0, 0, collInfo.getEpoch()); + metadata->_collVersion = ChunkVersion(0, 0, collInfo.getEpoch()); + + return Status::OK(); +} + +Status MetadataLoader::initChunks(CatalogManager* catalogManager, + const string& ns, + const string& shard, + const CollectionMetadata* oldMetadata, + CollectionMetadata* metadata) const { + map<string, ChunkVersion> versionMap; + + // Preserve the epoch + versionMap[shard] = metadata->_shardVersion; + OID epoch = metadata->getCollVersion().epoch(); + bool fullReload = true; + + // Check to see if we should use the old version or not. + if (oldMetadata) { + // If our epochs are compatible, it's useful to use the old metadata for diffs + if (oldMetadata->getCollVersion().hasEqualEpoch(epoch)) { + fullReload = false; + invariant(oldMetadata->isValid()); + + versionMap[shard] = oldMetadata->_shardVersion; + metadata->_collVersion = oldMetadata->_collVersion; + + // TODO: This could be made more efficient if copying not required, but + // not as frequently reloaded as in mongos. + metadata->_chunksMap = oldMetadata->_chunksMap; + + LOG(2) << "loading new chunks for collection " << ns + << " using old metadata w/ version " << oldMetadata->getShardVersion() << " and " + << metadata->_chunksMap.size() << " chunks" << endl; + } else { + warning() << "reloading collection metadata for " << ns << " with new epoch " + << epoch.toString() << ", the current epoch is " + << oldMetadata->getCollVersion().epoch().toString() << endl; } + } - // Exposes the new metadata's range map and version to the "differ," who - // would ultimately be responsible of filling them up. - SCMConfigDiffTracker differ( shard ); - differ.attach( ns, metadata->_chunksMap, metadata->_collVersion, versionMap ); - - try { - std::vector<ChunkType> chunks; - Status status = catalogManager->getChunks(differ.configDiffQuery(), 0, &chunks); - if (!status.isOK()) { - if (status == ErrorCodes::HostUnreachable) { - // Make our metadata invalid - metadata->_collVersion = ChunkVersion( 0, 0, OID() ); - metadata->_chunksMap.clear(); - } - return status; - } - - // - // The diff tracker should always find at least one chunk (the highest chunk we saw - // last time). If not, something has changed on the config server (potentially between - // when we read the collection data and when we read the chunks data). - // - int diffsApplied = differ.calculateConfigDiff(chunks); - if ( diffsApplied > 0 ) { - // Chunks found, return ok - LOG(2) << "loaded " << diffsApplied << " chunks into new metadata for " << ns - << " with version " << metadata->_collVersion; - - metadata->_shardVersion = versionMap[shard]; - metadata->fillRanges(); - - invariant( metadata->isValid() ); - return Status::OK(); - } - else if ( diffsApplied == 0 ) { - - // No chunks found, the collection is dropping or we're confused - // If this is a full reload, assume it is a drop for backwards compatibility - // TODO: drop the config.collections entry *before* the chunks and eliminate this - // ambiguity - - string errMsg = - str::stream() << "no chunks found when reloading " << ns - << ", previous version was " - << metadata->_collVersion.toString() - << ( fullReload ? ", this is a drop" : "" ); - - warning() << errMsg << endl; - - metadata->_collVersion = ChunkVersion( 0, 0, OID() ); - metadata->_chunksMap.clear(); + // Exposes the new metadata's range map and version to the "differ," who + // would ultimately be responsible of filling them up. + SCMConfigDiffTracker differ(shard); + differ.attach(ns, metadata->_chunksMap, metadata->_collVersion, versionMap); - return fullReload ? Status( ErrorCodes::NamespaceNotFound, errMsg ) : - Status( ErrorCodes::RemoteChangeDetected, errMsg ); - } - else { - // Invalid chunks found, our epoch may have changed because we dropped/recreated - // the collection. - string errMsg = str::stream() << "invalid chunks found when reloading " << ns - << ", previous version was " - << metadata->_collVersion.toString() - << ", this should be rare"; - warning() << errMsg; - - metadata->_collVersion = ChunkVersion( 0, 0, OID() ); + try { + std::vector<ChunkType> chunks; + Status status = catalogManager->getChunks(differ.configDiffQuery(), 0, &chunks); + if (!status.isOK()) { + if (status == ErrorCodes::HostUnreachable) { + // Make our metadata invalid + metadata->_collVersion = ChunkVersion(0, 0, OID()); metadata->_chunksMap.clear(); - - return Status(ErrorCodes::RemoteChangeDetected, errMsg); } + return status; } - catch (const DBException& e) { - // We deliberately do not return connPtr to the pool, since it was involved with the - // error here. - return Status(ErrorCodes::HostUnreachable, - str::stream() << "problem querying chunks metadata" << causedBy(e)); + + // + // The diff tracker should always find at least one chunk (the highest chunk we saw + // last time). If not, something has changed on the config server (potentially between + // when we read the collection data and when we read the chunks data). + // + int diffsApplied = differ.calculateConfigDiff(chunks); + if (diffsApplied > 0) { + // Chunks found, return ok + LOG(2) << "loaded " << diffsApplied << " chunks into new metadata for " << ns + << " with version " << metadata->_collVersion; + + metadata->_shardVersion = versionMap[shard]; + metadata->fillRanges(); + + invariant(metadata->isValid()); + return Status::OK(); + } else if (diffsApplied == 0) { + // No chunks found, the collection is dropping or we're confused + // If this is a full reload, assume it is a drop for backwards compatibility + // TODO: drop the config.collections entry *before* the chunks and eliminate this + // ambiguity + + string errMsg = str::stream() + << "no chunks found when reloading " << ns << ", previous version was " + << metadata->_collVersion.toString() << (fullReload ? ", this is a drop" : ""); + + warning() << errMsg << endl; + + metadata->_collVersion = ChunkVersion(0, 0, OID()); + metadata->_chunksMap.clear(); + + return fullReload ? Status(ErrorCodes::NamespaceNotFound, errMsg) + : Status(ErrorCodes::RemoteChangeDetected, errMsg); + } else { + // Invalid chunks found, our epoch may have changed because we dropped/recreated + // the collection. + string errMsg = str::stream() + << "invalid chunks found when reloading " << ns << ", previous version was " + << metadata->_collVersion.toString() << ", this should be rare"; + warning() << errMsg; + + metadata->_collVersion = ChunkVersion(0, 0, OID()); + metadata->_chunksMap.clear(); + + return Status(ErrorCodes::RemoteChangeDetected, errMsg); } + } catch (const DBException& e) { + // We deliberately do not return connPtr to the pool, since it was involved with the + // error here. + return Status(ErrorCodes::HostUnreachable, + str::stream() << "problem querying chunks metadata" << causedBy(e)); } +} + +Status MetadataLoader::promotePendingChunks(const CollectionMetadata* afterMetadata, + CollectionMetadata* remoteMetadata) const { + // Ensure pending chunks are applicable + bool notApplicable = (NULL == afterMetadata || NULL == remoteMetadata) || + (afterMetadata->getShardVersion() > remoteMetadata->getShardVersion()) || + (afterMetadata->getShardVersion().epoch() != remoteMetadata->getShardVersion().epoch()); + if (notApplicable) + return Status::OK(); - Status MetadataLoader::promotePendingChunks( const CollectionMetadata* afterMetadata, - CollectionMetadata* remoteMetadata ) const { - - // Ensure pending chunks are applicable - bool notApplicable = - ( NULL == afterMetadata || NULL == remoteMetadata ) || - ( afterMetadata->getShardVersion() > remoteMetadata->getShardVersion() ) || - ( afterMetadata->getShardVersion().epoch() != - remoteMetadata->getShardVersion().epoch() ); - if ( notApplicable ) return Status::OK(); - - // The chunks from remoteMetadata are the latest version, and the pending chunks - // from afterMetadata are the latest version. If no trickery is afoot, pending chunks - // should match exactly zero or one loaded chunk. + // The chunks from remoteMetadata are the latest version, and the pending chunks + // from afterMetadata are the latest version. If no trickery is afoot, pending chunks + // should match exactly zero or one loaded chunk. - remoteMetadata->_pendingMap = afterMetadata->_pendingMap; + remoteMetadata->_pendingMap = afterMetadata->_pendingMap; - // Resolve our pending chunks against the chunks we've loaded - for ( RangeMap::iterator it = remoteMetadata->_pendingMap.begin(); - it != remoteMetadata->_pendingMap.end(); ) { + // Resolve our pending chunks against the chunks we've loaded + for (RangeMap::iterator it = remoteMetadata->_pendingMap.begin(); + it != remoteMetadata->_pendingMap.end();) { + if (!rangeMapOverlaps(remoteMetadata->_chunksMap, it->first, it->second)) { + ++it; + continue; + } - if ( !rangeMapOverlaps( remoteMetadata->_chunksMap, it->first, it->second ) ) { - ++it; - continue; - } + // Our pending range overlaps at least one chunk - // Our pending range overlaps at least one chunk + if (rangeMapContains(remoteMetadata->_chunksMap, it->first, it->second)) { + // Chunk was promoted from pending, successful migration + LOG(2) << "verified chunk " << rangeToString(it->first, it->second) + << " was migrated earlier to this shard" << endl; - if ( rangeMapContains( remoteMetadata->_chunksMap, it->first, it->second ) ) { + remoteMetadata->_pendingMap.erase(it++); + } else { + // Something strange happened, maybe manual editing of config? + RangeVector overlap; + getRangeMapOverlap(remoteMetadata->_chunksMap, it->first, it->second, &overlap); - // Chunk was promoted from pending, successful migration - LOG( 2 ) << "verified chunk " << rangeToString( it->first, it->second ) - << " was migrated earlier to this shard" << endl; + string errMsg = str::stream() + << "the remote metadata changed unexpectedly, pending range " + << rangeToString(it->first, it->second) + << " does not exactly overlap loaded chunks " << overlapToString(overlap); - remoteMetadata->_pendingMap.erase( it++ ); - } - else { - - // Something strange happened, maybe manual editing of config? - RangeVector overlap; - getRangeMapOverlap( remoteMetadata->_chunksMap, - it->first, - it->second, - &overlap ); - - string errMsg = str::stream() - << "the remote metadata changed unexpectedly, pending range " - << rangeToString( it->first, it->second ) - << " does not exactly overlap loaded chunks " - << overlapToString( overlap ); - - return Status( ErrorCodes::RemoteChangeDetected, errMsg ); - } + return Status(ErrorCodes::RemoteChangeDetected, errMsg); } - - return Status::OK(); } + return Status::OK(); +} + -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/metadata_loader.h b/src/mongo/s/metadata_loader.h index 7d6e3d77798..15ca227926e 100644 --- a/src/mongo/s/metadata_loader.h +++ b/src/mongo/s/metadata_loader.h @@ -36,126 +36,124 @@ namespace mongo { - class CatalogManager; - class CollectionMetadata; - class CollectionType; - class DBClientCursor; +class CatalogManager; +class CollectionMetadata; +class CollectionType; +class DBClientCursor; + +/** + * The MetadataLoader is responsible for interfacing with the config servers and previous + * metadata to build new instances of CollectionMetadata. MetadataLoader is the "builder" + * class for metadata. + * + * CollectionMetadata has both persisted and volatile state (for now) - the persisted + * config server chunk state and the volatile pending state which is only tracked locally + * while a server is the primary. This requires a two-step loading process - the persisted + * chunk state *cannot* be loaded in a DBLock lock while the pending chunk state *must* be. + * + * Example usage: + * beforeMetadata = <get latest local metadata>; + * remoteMetadata = makeCollectionMetadata( beforeMetadata, remoteMetadata ); + * DBLock lock(txn, dbname, MODE_X); + * afterMetadata = <get latest local metadata>; + * promotePendingChunks( afterMetadata, remoteMetadata ); + * + * The loader will go out of its way to try to fetch the smaller amount possible of data + * from the config server without sacrificing the freshness and accuracy of the metadata it + * builds. (See ConfigDiffTracker class.) + * + * The class is not thread safe. + */ +class MetadataLoader { +public: + explicit MetadataLoader(); + + ~MetadataLoader(); /** - * The MetadataLoader is responsible for interfacing with the config servers and previous - * metadata to build new instances of CollectionMetadata. MetadataLoader is the "builder" - * class for metadata. + * Fills a new metadata instance representing the chunkset of the collection 'ns' + * (or its entirety, if not sharded) that lives on 'shard' with data from the config server. + * Optionally, uses an 'oldMetadata' for the same 'ns'/'shard'; the contents of + * 'oldMetadata' can help reducing the amount of data read from the config servers. * - * CollectionMetadata has both persisted and volatile state (for now) - the persisted - * config server chunk state and the volatile pending state which is only tracked locally - * while a server is the primary. This requires a two-step loading process - the persisted - * chunk state *cannot* be loaded in a DBLock lock while the pending chunk state *must* be. - * - * Example usage: - * beforeMetadata = <get latest local metadata>; - * remoteMetadata = makeCollectionMetadata( beforeMetadata, remoteMetadata ); - * DBLock lock(txn, dbname, MODE_X); - * afterMetadata = <get latest local metadata>; - * promotePendingChunks( afterMetadata, remoteMetadata ); + * Locking note: + * + Must not be called in a DBLock, since this loads over the network * - * The loader will go out of its way to try to fetch the smaller amount possible of data - * from the config server without sacrificing the freshness and accuracy of the metadata it - * builds. (See ConfigDiffTracker class.) + * OK on success. * - * The class is not thread safe. + * Failure return values: + * Abnormal: + * @return FailedToParse if there was an error parsing the remote config data + * Normal: + * @return NamespaceNotFound if the collection no longer exists + * @return HostUnreachable if there was an error contacting the config servers + * @return RemoteChangeDetected if the data loaded was modified by another operation */ - class MetadataLoader { - public: - - explicit MetadataLoader(); - - ~MetadataLoader(); + Status makeCollectionMetadata(CatalogManager* catalogManager, + const std::string& ns, + const std::string& shard, + const CollectionMetadata* oldMetadata, + CollectionMetadata* metadata) const; - /** - * Fills a new metadata instance representing the chunkset of the collection 'ns' - * (or its entirety, if not sharded) that lives on 'shard' with data from the config server. - * Optionally, uses an 'oldMetadata' for the same 'ns'/'shard'; the contents of - * 'oldMetadata' can help reducing the amount of data read from the config servers. - * - * Locking note: - * + Must not be called in a DBLock, since this loads over the network - * - * OK on success. - * - * Failure return values: - * Abnormal: - * @return FailedToParse if there was an error parsing the remote config data - * Normal: - * @return NamespaceNotFound if the collection no longer exists - * @return HostUnreachable if there was an error contacting the config servers - * @return RemoteChangeDetected if the data loaded was modified by another operation - */ - Status makeCollectionMetadata(CatalogManager* catalogManager, - const std::string& ns, - const std::string& shard, - const CollectionMetadata* oldMetadata, - CollectionMetadata* metadata) const; - - /** - * Replaces the pending chunks of the remote metadata with the more up-to-date pending - * chunks of the 'after' metadata (metadata from after the remote load), and removes pending - * chunks which are now regular chunks. - * - * Pending chunks should always correspond to one or zero chunks in the remoteMetadata - * if the epochs are the same and the remote version is the same or higher, otherwise they - * are not applicable. - * - * Locking note: - * + Must be called in a DBLock, to ensure validity of afterMetadata - * - * Returns OK if pending chunks correctly follow the rule above or are not applicable - * Returns RemoteChangeDetected if pending chunks do not follow the rule above, indicating - * either the config server or us has changed unexpectedly. - * This should only occur with manual editing of the config - * server. - * - * TODO: This is a bit ugly but necessary for now. If/when pending chunk info is stored on - * the config server, this should go away. - */ - Status promotePendingChunks( const CollectionMetadata* afterMetadata, - CollectionMetadata* remoteMetadata ) const; - - private: + /** + * Replaces the pending chunks of the remote metadata with the more up-to-date pending + * chunks of the 'after' metadata (metadata from after the remote load), and removes pending + * chunks which are now regular chunks. + * + * Pending chunks should always correspond to one or zero chunks in the remoteMetadata + * if the epochs are the same and the remote version is the same or higher, otherwise they + * are not applicable. + * + * Locking note: + * + Must be called in a DBLock, to ensure validity of afterMetadata + * + * Returns OK if pending chunks correctly follow the rule above or are not applicable + * Returns RemoteChangeDetected if pending chunks do not follow the rule above, indicating + * either the config server or us has changed unexpectedly. + * This should only occur with manual editing of the config + * server. + * + * TODO: This is a bit ugly but necessary for now. If/when pending chunk info is stored on + * the config server, this should go away. + */ + Status promotePendingChunks(const CollectionMetadata* afterMetadata, + CollectionMetadata* remoteMetadata) const; - /** - * Returns OK and fills in the internal state of 'metadata' with general collection - * information, not including chunks. - * - * If information about the collection can be accessed or is invalid, returns: - * @return NamespaceNotFound if the collection no longer exists - * @return FailedToParse if there was an error parsing the remote config data - * @return HostUnreachable if there was an error contacting the config servers - * @return RemoteChangeDetected if the collection doc loaded is unexpectedly different - * - */ - Status _initCollection(CatalogManager* catalogManager, - const std::string& ns, - const std::string& shard, - CollectionMetadata* metadata) const; +private: + /** + * Returns OK and fills in the internal state of 'metadata' with general collection + * information, not including chunks. + * + * If information about the collection can be accessed or is invalid, returns: + * @return NamespaceNotFound if the collection no longer exists + * @return FailedToParse if there was an error parsing the remote config data + * @return HostUnreachable if there was an error contacting the config servers + * @return RemoteChangeDetected if the collection doc loaded is unexpectedly different + * + */ + Status _initCollection(CatalogManager* catalogManager, + const std::string& ns, + const std::string& shard, + CollectionMetadata* metadata) const; - /** - * Returns OK and fills in the chunk state of 'metadata' to portray the chunks of the - * collection 'ns' that sit in 'shard'. If provided, uses the contents of 'oldMetadata' - * as a base (see description in initCollection above). - * - * If information about the chunks can be accessed or is invalid, returns: - * @return HostUnreachable if there was an error contacting the config servers - * @return RemoteChangeDetected if the chunks loaded are unexpectedly different - * - * For backwards compatibility, - * @return NamespaceNotFound if there are no chunks loaded and an epoch change is detected - * TODO: @return FailedToParse - */ - Status initChunks(CatalogManager* catalogManager, - const std::string& ns, - const std::string& shard, - const CollectionMetadata* oldMetadata, - CollectionMetadata* metadata ) const; - }; + /** + * Returns OK and fills in the chunk state of 'metadata' to portray the chunks of the + * collection 'ns' that sit in 'shard'. If provided, uses the contents of 'oldMetadata' + * as a base (see description in initCollection above). + * + * If information about the chunks can be accessed or is invalid, returns: + * @return HostUnreachable if there was an error contacting the config servers + * @return RemoteChangeDetected if the chunks loaded are unexpectedly different + * + * For backwards compatibility, + * @return NamespaceNotFound if there are no chunks loaded and an epoch change is detected + * TODO: @return FailedToParse + */ + Status initChunks(CatalogManager* catalogManager, + const std::string& ns, + const std::string& shard, + const CollectionMetadata* oldMetadata, + CollectionMetadata* metadata) const; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/metadata_loader_test.cpp b/src/mongo/s/metadata_loader_test.cpp index ad98acf1e94..bf0fc14b3c8 100644 --- a/src/mongo/s/metadata_loader_test.cpp +++ b/src/mongo/s/metadata_loader_test.cpp @@ -47,673 +47,666 @@ namespace { - using namespace mongo; +using namespace mongo; - using std::unique_ptr; - using std::string; - using std::vector; +using std::unique_ptr; +using std::string; +using std::vector; - const std::string CONFIG_HOST_PORT = "$dummy_config:27017"; +const std::string CONFIG_HOST_PORT = "$dummy_config:27017"; - // TODO: Test config server down - // TODO: Test read of chunks with new epoch - // TODO: Test that you can properly load config using format with deprecated fields? +// TODO: Test config server down +// TODO: Test read of chunks with new epoch +// TODO: Test that you can properly load config using format with deprecated fields? - class MetadataLoaderFixture : public mongo::unittest::Test { - public: - void setUp() { - ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); - ASSERT(configLoc.isValid()); - ASSERT_OK(_catalogManager.init(configLoc)); - } - - protected: - CatalogManager* catalogManager() { return &_catalogManager; } - - private: - CatalogManagerLegacy _catalogManager; - }; - - - TEST_F(MetadataLoaderFixture, DroppedColl) { - MockRemoteDBServer dummyConfig( CONFIG_HOST_PORT ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( &dummyConfig ); +class MetadataLoaderFixture : public mongo::unittest::Test { +public: + void setUp() { + ConnectionString configLoc = ConnectionString(HostAndPort(CONFIG_HOST_PORT)); + ASSERT(configLoc.isValid()); + ASSERT_OK(_catalogManager.init(configLoc)); + } - CollectionType collInfo; - collInfo.setNs(NamespaceString{"test.foo"}); - collInfo.setKeyPattern(BSON("a" << 1)); - collInfo.setUpdatedAt( Date_t() ); - collInfo.setEpoch( OID() ); - collInfo.setDropped( true ); - ASSERT_OK(collInfo.validate()); +protected: + CatalogManager* catalogManager() { + return &_catalogManager; + } - dummyConfig.insert( CollectionType::ConfigNS, collInfo.toBSON() ); +private: + CatalogManagerLegacy _catalogManager; +}; - MetadataLoader loader; - string errmsg; - CollectionMetadata metadata; - Status status = loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); +TEST_F(MetadataLoaderFixture, DroppedColl) { + MockRemoteDBServer dummyConfig(CONFIG_HOST_PORT); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(&dummyConfig); - ASSERT_EQUALS( status.code(), ErrorCodes::NamespaceNotFound ); + CollectionType collInfo; + collInfo.setNs(NamespaceString{"test.foo"}); + collInfo.setKeyPattern(BSON("a" << 1)); + collInfo.setUpdatedAt(Date_t()); + collInfo.setEpoch(OID()); + collInfo.setDropped(true); + ASSERT_OK(collInfo.validate()); - MockConnRegistry::get()->clear(); - ScopedDbConnection::clearPool(); - } + dummyConfig.insert(CollectionType::ConfigNS, collInfo.toBSON()); - TEST_F(MetadataLoaderFixture, EmptyColl) { - MockRemoteDBServer dummyConfig( CONFIG_HOST_PORT ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( &dummyConfig ); + MetadataLoader loader; - MetadataLoader loader; + string errmsg; + CollectionMetadata metadata; + Status status = loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); - string errmsg; - CollectionMetadata metadata; - Status status = loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); + ASSERT_EQUALS(status.code(), ErrorCodes::NamespaceNotFound); - ASSERT_EQUALS( status.code(), ErrorCodes::NamespaceNotFound ); + MockConnRegistry::get()->clear(); + ScopedDbConnection::clearPool(); +} - MockConnRegistry::get()->clear(); - ScopedDbConnection::clearPool(); - } +TEST_F(MetadataLoaderFixture, EmptyColl) { + MockRemoteDBServer dummyConfig(CONFIG_HOST_PORT); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(&dummyConfig); - TEST_F(MetadataLoaderFixture, BadColl) { - MockRemoteDBServer dummyConfig( CONFIG_HOST_PORT ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( &dummyConfig ); + MetadataLoader loader; - dummyConfig.insert(CollectionType::ConfigNS, BSON(CollectionType::fullNs("test.foo"))); + string errmsg; + CollectionMetadata metadata; + Status status = loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); - MetadataLoader loader; + ASSERT_EQUALS(status.code(), ErrorCodes::NamespaceNotFound); - string errmsg; - CollectionMetadata metadata; - Status status = loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); + MockConnRegistry::get()->clear(); + ScopedDbConnection::clearPool(); +} - ASSERT_EQUALS(status.code(), ErrorCodes::NoSuchKey); +TEST_F(MetadataLoaderFixture, BadColl) { + MockRemoteDBServer dummyConfig(CONFIG_HOST_PORT); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(&dummyConfig); - MockConnRegistry::get()->clear(); - ScopedDbConnection::clearPool(); - } + dummyConfig.insert(CollectionType::ConfigNS, BSON(CollectionType::fullNs("test.foo"))); - TEST_F(MetadataLoaderFixture, BadChunk) { - MockRemoteDBServer dummyConfig( CONFIG_HOST_PORT ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( &dummyConfig ); + MetadataLoader loader; - CollectionType collInfo; - collInfo.setNs(NamespaceString{"test.foo"}); - collInfo.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); - collInfo.setKeyPattern( BSON("a" << 1) ); - collInfo.setEpoch( OID::gen() ); - ASSERT_OK(collInfo.validate()); + string errmsg; + CollectionMetadata metadata; + Status status = loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); - dummyConfig.insert( CollectionType::ConfigNS, collInfo.toBSON() ); + ASSERT_EQUALS(status.code(), ErrorCodes::NoSuchKey); - ChunkType chunkInfo; - chunkInfo.setNS(NamespaceString{"test.foo"}); - chunkInfo.setVersion( ChunkVersion( 1, 0, collInfo.getEpoch() ) ); - ASSERT(!chunkInfo.validate().isOK()); + MockConnRegistry::get()->clear(); + ScopedDbConnection::clearPool(); +} - dummyConfig.insert( ChunkType::ConfigNS, chunkInfo.toBSON() ); +TEST_F(MetadataLoaderFixture, BadChunk) { + MockRemoteDBServer dummyConfig(CONFIG_HOST_PORT); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(&dummyConfig); - MetadataLoader loader; + CollectionType collInfo; + collInfo.setNs(NamespaceString{"test.foo"}); + collInfo.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + collInfo.setKeyPattern(BSON("a" << 1)); + collInfo.setEpoch(OID::gen()); + ASSERT_OK(collInfo.validate()); - string errmsg; - CollectionMetadata metadata; - Status status = loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); + dummyConfig.insert(CollectionType::ConfigNS, collInfo.toBSON()); - ASSERT_EQUALS(status.code(), ErrorCodes::FailedToParse); + ChunkType chunkInfo; + chunkInfo.setNS(NamespaceString{"test.foo"}); + chunkInfo.setVersion(ChunkVersion(1, 0, collInfo.getEpoch())); + ASSERT(!chunkInfo.validate().isOK()); - MockConnRegistry::get()->clear(); - ScopedDbConnection::clearPool(); - } + dummyConfig.insert(ChunkType::ConfigNS, chunkInfo.toBSON()); - class NoChunkFixture : public MetadataLoaderFixture { - protected: - void setUp() { - MetadataLoaderFixture::setUp(); + MetadataLoader loader; - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); + string errmsg; + CollectionMetadata metadata; + Status status = loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); - OID epoch = OID::gen(); + ASSERT_EQUALS(status.code(), ErrorCodes::FailedToParse); - CollectionType collType; - collType.setNs(NamespaceString{"test.foo"}); - collType.setKeyPattern(BSON("a" << 1)); - collType.setUnique(false); - collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); - collType.setEpoch(epoch); - _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); - } + MockConnRegistry::get()->clear(); + ScopedDbConnection::clearPool(); +} - void tearDown() { - MockConnRegistry::get()->clear(); - ScopedDbConnection::clearPool(); - } +class NoChunkFixture : public MetadataLoaderFixture { +protected: + void setUp() { + MetadataLoaderFixture::setUp(); - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - }; + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); - TEST_F(NoChunkFixture, NoChunksIsDropped) { - MetadataLoader loader; + OID epoch = OID::gen(); - CollectionMetadata metadata; - Status status = loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); - - // This is interpreted as a dropped ns, since we drop the chunks first - ASSERT_EQUALS( status.code(), ErrorCodes::NamespaceNotFound ); + CollectionType collType; + collType.setNs(NamespaceString{"test.foo"}); + collType.setKeyPattern(BSON("a" << 1)); + collType.setUnique(false); + collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + collType.setEpoch(epoch); + _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); } - class NoChunkHereFixture : public MetadataLoaderFixture { - protected: - void setUp() { - MetadataLoaderFixture::setUp(); + void tearDown() { + MockConnRegistry::get()->clear(); + ScopedDbConnection::clearPool(); + } - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; +}; - OID epoch = OID::gen(); +TEST_F(NoChunkFixture, NoChunksIsDropped) { + MetadataLoader loader; - CollectionType collType; - collType.setNs(NamespaceString{"test.foo"}); - collType.setKeyPattern( BSON("a" << 1) ); - collType.setUnique( false ); - collType.setUpdatedAt( Date_t::fromMillisSinceEpoch(1) ); - collType.setEpoch( epoch ); - ASSERT_OK(collType.validate()); - - _dummyConfig->insert( CollectionType::ConfigNS, collType.toBSON() ); - - // Need a chunk on another shard, otherwise the chunks are invalid in general and we - // can't load metadata - ChunkType chunkType; - chunkType.setNS( "test.foo" ); - chunkType.setShard( "shard0001" ); - chunkType.setMin( BSON( "a" << MINKEY ) ); - chunkType.setMax( BSON( "a" << MAXKEY ) ); - chunkType.setVersion( ChunkVersion( 1, 0, epoch ) ); - chunkType.setName( OID::gen().toString() ); - ASSERT(chunkType.validate().isOK()); - - _dummyConfig->insert( ChunkType::ConfigNS, chunkType.toBSON() ); - } + CollectionMetadata metadata; + Status status = loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); - MockRemoteDBServer* getDummyConfig() { - return _dummyConfig.get(); - } + // This is interpreted as a dropped ns, since we drop the chunks first + ASSERT_EQUALS(status.code(), ErrorCodes::NamespaceNotFound); +} - void tearDown() { - MockConnRegistry::get()->clear(); - ScopedDbConnection::clearPool(); - } +class NoChunkHereFixture : public MetadataLoaderFixture { +protected: + void setUp() { + MetadataLoaderFixture::setUp(); - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - }; + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); - TEST_F(NoChunkHereFixture, CheckNumChunk) { - MetadataLoader loader; + OID epoch = OID::gen(); - CollectionMetadata metadata; - Status status = loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); - - ASSERT( status.isOK() ); - ASSERT_EQUALS( 0U, metadata.getNumChunks() ); - ASSERT_EQUALS( 1, metadata.getCollVersion().majorVersion() ); - ASSERT_EQUALS( 0, metadata.getShardVersion().majorVersion() ); - ASSERT_NOT_EQUALS( OID(), metadata.getCollVersion().epoch() ); - ASSERT_NOT_EQUALS( OID(), metadata.getShardVersion().epoch() ); + CollectionType collType; + collType.setNs(NamespaceString{"test.foo"}); + collType.setKeyPattern(BSON("a" << 1)); + collType.setUnique(false); + collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + collType.setEpoch(epoch); + ASSERT_OK(collType.validate()); + + _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); + + // Need a chunk on another shard, otherwise the chunks are invalid in general and we + // can't load metadata + ChunkType chunkType; + chunkType.setNS("test.foo"); + chunkType.setShard("shard0001"); + chunkType.setMin(BSON("a" << MINKEY)); + chunkType.setMax(BSON("a" << MAXKEY)); + chunkType.setVersion(ChunkVersion(1, 0, epoch)); + chunkType.setName(OID::gen().toString()); + ASSERT(chunkType.validate().isOK()); + + _dummyConfig->insert(ChunkType::ConfigNS, chunkType.toBSON()); } - class ConfigServerFixture : public MetadataLoaderFixture { - protected: - void setUp() { - MetadataLoaderFixture::setUp(); - - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); - - OID epoch = OID::gen(); - _maxCollVersion = ChunkVersion( 1, 0, epoch ); - - CollectionType collType; - collType.setNs(NamespaceString{"test.foo"}); - collType.setKeyPattern(BSON("a" << 1)); - collType.setUnique(false); - collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); - collType.setEpoch(epoch); - _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); + MockRemoteDBServer* getDummyConfig() { + return _dummyConfig.get(); + } - BSONObj fooSingle = BSON( - ChunkType::name("test.foo-a_MinKey") << - ChunkType::ns("test.foo") << - ChunkType::min(BSON("a" << MINKEY)) << - ChunkType::max(BSON("a" << MAXKEY)) << - ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch( - _maxCollVersion.toLong())) << - ChunkType::DEPRECATED_epoch(epoch) << - ChunkType::shard("shard0000")); - _dummyConfig->insert( ChunkType::ConfigNS, fooSingle ); - } + void tearDown() { + MockConnRegistry::get()->clear(); + ScopedDbConnection::clearPool(); + } - void tearDown() { - MockConnRegistry::get()->clear(); - } +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; +}; + +TEST_F(NoChunkHereFixture, CheckNumChunk) { + MetadataLoader loader; + + CollectionMetadata metadata; + Status status = loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); + + ASSERT(status.isOK()); + ASSERT_EQUALS(0U, metadata.getNumChunks()); + ASSERT_EQUALS(1, metadata.getCollVersion().majorVersion()); + ASSERT_EQUALS(0, metadata.getShardVersion().majorVersion()); + ASSERT_NOT_EQUALS(OID(), metadata.getCollVersion().epoch()); + ASSERT_NOT_EQUALS(OID(), metadata.getShardVersion().epoch()); +} - ChunkVersion getMaxCollVersion() const { - return _maxCollVersion; - } +class ConfigServerFixture : public MetadataLoaderFixture { +protected: + void setUp() { + MetadataLoaderFixture::setUp(); - ChunkVersion getMaxShardVersion() const { - return _maxCollVersion; - } + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); - MockRemoteDBServer* getConfigServer() const { - return _dummyConfig.get(); - } + OID epoch = OID::gen(); + _maxCollVersion = ChunkVersion(1, 0, epoch); + + CollectionType collType; + collType.setNs(NamespaceString{"test.foo"}); + collType.setKeyPattern(BSON("a" << 1)); + collType.setUnique(false); + collType.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + collType.setEpoch(epoch); + _dummyConfig->insert(CollectionType::ConfigNS, collType.toBSON()); + + BSONObj fooSingle = BSON( + ChunkType::name("test.foo-a_MinKey") + << ChunkType::ns("test.foo") << ChunkType::min(BSON("a" << MINKEY)) + << ChunkType::max(BSON("a" << MAXKEY)) + << ChunkType::DEPRECATED_lastmod(Date_t::fromMillisSinceEpoch(_maxCollVersion.toLong())) + << ChunkType::DEPRECATED_epoch(epoch) << ChunkType::shard("shard0000")); + _dummyConfig->insert(ChunkType::ConfigNS, fooSingle); + } - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - ChunkVersion _maxCollVersion; - }; + void tearDown() { + MockConnRegistry::get()->clear(); + } - TEST_F(ConfigServerFixture, SingleChunkCheckNumChunk) { - MetadataLoader loader; - CollectionMetadata metadata; - Status status = loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); - ASSERT( status.isOK() ); - ASSERT_EQUALS( 1U, metadata.getNumChunks() ); + ChunkVersion getMaxCollVersion() const { + return _maxCollVersion; } - TEST_F(ConfigServerFixture, SingleChunkGetNext) { - MetadataLoader loader; - CollectionMetadata metadata; - loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); - ChunkType chunkInfo; - ASSERT_TRUE(metadata.getNextChunk(metadata.getMinKey(), &chunkInfo)); + ChunkVersion getMaxShardVersion() const { + return _maxCollVersion; } - TEST_F(ConfigServerFixture, SingleChunkGetShardKey) { - MetadataLoader loader; - CollectionMetadata metadata; - loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); - ASSERT_TRUE( metadata.getKeyPattern().equal(BSON("a" << 1)) ); + MockRemoteDBServer* getConfigServer() const { + return _dummyConfig.get(); } - TEST_F(ConfigServerFixture, SingleChunkGetMaxCollVersion) { - MetadataLoader loader; - CollectionMetadata metadata; - loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; + ChunkVersion _maxCollVersion; +}; + +TEST_F(ConfigServerFixture, SingleChunkCheckNumChunk) { + MetadataLoader loader; + CollectionMetadata metadata; + Status status = loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); + ASSERT(status.isOK()); + ASSERT_EQUALS(1U, metadata.getNumChunks()); +} - ASSERT_TRUE( getMaxCollVersion().equals( metadata.getCollVersion() ) ); - } +TEST_F(ConfigServerFixture, SingleChunkGetNext) { + MetadataLoader loader; + CollectionMetadata metadata; + loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); + ChunkType chunkInfo; + ASSERT_TRUE(metadata.getNextChunk(metadata.getMinKey(), &chunkInfo)); +} - TEST_F(ConfigServerFixture, SingleChunkGetMaxShardVersion) { - MetadataLoader loader; - CollectionMetadata metadata; - loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); +TEST_F(ConfigServerFixture, SingleChunkGetShardKey) { + MetadataLoader loader; + CollectionMetadata metadata; + loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); + ASSERT_TRUE(metadata.getKeyPattern().equal(BSON("a" << 1))); +} - ASSERT_TRUE( getMaxShardVersion().equals( metadata.getShardVersion() ) ); - } +TEST_F(ConfigServerFixture, SingleChunkGetMaxCollVersion) { + MetadataLoader loader; + CollectionMetadata metadata; + loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); - TEST_F(ConfigServerFixture, NoChunks) { - getConfigServer()->remove( ChunkType::ConfigNS, BSONObj() ); + ASSERT_TRUE(getMaxCollVersion().equals(metadata.getCollVersion())); +} - MetadataLoader loader; - CollectionMetadata metadata; - Status status = loader.makeCollectionMetadata(catalogManager(), - "test.foo", - "shard0000", - NULL, /* no old metadata */ - &metadata); - - // NSNotFound because we're reloading with no old metadata - ASSERT_EQUALS( status.code(), ErrorCodes::NamespaceNotFound ); - } +TEST_F(ConfigServerFixture, SingleChunkGetMaxShardVersion) { + MetadataLoader loader; + CollectionMetadata metadata; + loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); - class MultipleMetadataFixture : public MetadataLoaderFixture { - protected: - void setUp() { - MetadataLoaderFixture::setUp(); + ASSERT_TRUE(getMaxShardVersion().equals(metadata.getShardVersion())); +} - _dummyConfig.reset( new MockRemoteDBServer( CONFIG_HOST_PORT ) ); - mongo::ConnectionString::setConnectionHook( MockConnRegistry::get()->getConnStrHook() ); - MockConnRegistry::get()->addServer( _dummyConfig.get() ); +TEST_F(ConfigServerFixture, NoChunks) { + getConfigServer()->remove(ChunkType::ConfigNS, BSONObj()); - ConnectionString confServerStr((HostAndPort(CONFIG_HOST_PORT))); - ConnectionString configLoc( confServerStr ); - _loader.reset(new MetadataLoader); - } + MetadataLoader loader; + CollectionMetadata metadata; + Status status = loader.makeCollectionMetadata(catalogManager(), + "test.foo", + "shard0000", + NULL, /* no old metadata */ + &metadata); - MetadataLoader& loader() { - return *_loader; - } + // NSNotFound because we're reloading with no old metadata + ASSERT_EQUALS(status.code(), ErrorCodes::NamespaceNotFound); +} - void getMetadataFor( const OwnedPointerVector<ChunkType>& chunks, - CollectionMetadata* metadata ) { +class MultipleMetadataFixture : public MetadataLoaderFixture { +protected: + void setUp() { + MetadataLoaderFixture::setUp(); - // Infer namespace, shard, epoch, keypattern from first chunk - const ChunkType* firstChunk = *( chunks.vector().begin() ); + _dummyConfig.reset(new MockRemoteDBServer(CONFIG_HOST_PORT)); + mongo::ConnectionString::setConnectionHook(MockConnRegistry::get()->getConnStrHook()); + MockConnRegistry::get()->addServer(_dummyConfig.get()); - const string ns = firstChunk->getNS(); - const string shardName = firstChunk->getShard(); - const OID epoch = firstChunk->getVersion().epoch(); + ConnectionString confServerStr((HostAndPort(CONFIG_HOST_PORT))); + ConnectionString configLoc(confServerStr); + _loader.reset(new MetadataLoader); + } - BSONObjBuilder keyPatternB; - BSONObjIterator keyPatternIt( firstChunk->getMin() ); - while ( keyPatternIt.more() ) - keyPatternB.append( keyPatternIt.next().fieldName(), 1 ); - BSONObj keyPattern = keyPatternB.obj(); + MetadataLoader& loader() { + return *_loader; + } - _dummyConfig->remove( CollectionType::ConfigNS, BSONObj() ); - _dummyConfig->remove( ChunkType::ConfigNS, BSONObj() ); + void getMetadataFor(const OwnedPointerVector<ChunkType>& chunks, CollectionMetadata* metadata) { + // Infer namespace, shard, epoch, keypattern from first chunk + const ChunkType* firstChunk = *(chunks.vector().begin()); + + const string ns = firstChunk->getNS(); + const string shardName = firstChunk->getShard(); + const OID epoch = firstChunk->getVersion().epoch(); - CollectionType coll; - coll.setNs(NamespaceString{ns}); - coll.setKeyPattern( BSON( "a" << 1 ) ); - coll.setUpdatedAt( Date_t::fromMillisSinceEpoch(1) ); - coll.setEpoch( epoch ); - ASSERT_OK(coll.validate()); + BSONObjBuilder keyPatternB; + BSONObjIterator keyPatternIt(firstChunk->getMin()); + while (keyPatternIt.more()) + keyPatternB.append(keyPatternIt.next().fieldName(), 1); + BSONObj keyPattern = keyPatternB.obj(); - _dummyConfig->insert( CollectionType::ConfigNS, coll.toBSON() ); + _dummyConfig->remove(CollectionType::ConfigNS, BSONObj()); + _dummyConfig->remove(ChunkType::ConfigNS, BSONObj()); - ChunkVersion version( 1, 0, epoch ); - for (const auto chunkVal : chunks.vector()) { - ChunkType chunk(*chunkVal); + CollectionType coll; + coll.setNs(NamespaceString{ns}); + coll.setKeyPattern(BSON("a" << 1)); + coll.setUpdatedAt(Date_t::fromMillisSinceEpoch(1)); + coll.setEpoch(epoch); + ASSERT_OK(coll.validate()); - chunk.setName(OID::gen().toString()); - if (!chunk.isVersionSet()) { - chunk.setVersion(version); - version.incMajor(); - } + _dummyConfig->insert(CollectionType::ConfigNS, coll.toBSON()); - ASSERT(chunk.validate().isOK()); + ChunkVersion version(1, 0, epoch); + for (const auto chunkVal : chunks.vector()) { + ChunkType chunk(*chunkVal); - _dummyConfig->insert(ChunkType::ConfigNS, chunk.toBSON()); + chunk.setName(OID::gen().toString()); + if (!chunk.isVersionSet()) { + chunk.setVersion(version); + version.incMajor(); } - Status status = loader().makeCollectionMetadata(catalogManager(), - ns, - shardName, - NULL, - metadata); - ASSERT(status.isOK()); - } + ASSERT(chunk.validate().isOK()); - void tearDown() { - MockConnRegistry::get()->clear(); + _dummyConfig->insert(ChunkType::ConfigNS, chunk.toBSON()); } - private: - unique_ptr<MockRemoteDBServer> _dummyConfig; - unique_ptr<MetadataLoader> _loader; - }; + Status status = + loader().makeCollectionMetadata(catalogManager(), ns, shardName, NULL, metadata); + ASSERT(status.isOK()); + } - TEST_F(MultipleMetadataFixture, PromotePendingNA) { - unique_ptr<ChunkType> chunk( new ChunkType() ); - chunk->setNS("foo.bar"); - chunk->setShard("shard0000"); - chunk->setMin( BSON( "x" << MINKEY ) ); - chunk->setMax( BSON( "x" << 0 ) ); - chunk->setVersion( ChunkVersion( 1, 0, OID::gen() ) ); + void tearDown() { + MockConnRegistry::get()->clear(); + } - OwnedPointerVector<ChunkType> chunks; - chunks.mutableVector().push_back( chunk.release() ); +private: + unique_ptr<MockRemoteDBServer> _dummyConfig; + unique_ptr<MetadataLoader> _loader; +}; - CollectionMetadata afterMetadata; - getMetadataFor( chunks, &afterMetadata ); +TEST_F(MultipleMetadataFixture, PromotePendingNA) { + unique_ptr<ChunkType> chunk(new ChunkType()); + chunk->setNS("foo.bar"); + chunk->setShard("shard0000"); + chunk->setMin(BSON("x" << MINKEY)); + chunk->setMax(BSON("x" << 0)); + chunk->setVersion(ChunkVersion(1, 0, OID::gen())); - // Metadata of different epoch - ( *chunks.vector().begin() )->setVersion( ChunkVersion( 1, 0, OID::gen() ) ); + OwnedPointerVector<ChunkType> chunks; + chunks.mutableVector().push_back(chunk.release()); - CollectionMetadata remoteMetadata; - getMetadataFor( chunks, &remoteMetadata ); + CollectionMetadata afterMetadata; + getMetadataFor(chunks, &afterMetadata); - Status status = loader().promotePendingChunks( &afterMetadata, &remoteMetadata ); - ASSERT( status.isOK() ); + // Metadata of different epoch + (*chunks.vector().begin())->setVersion(ChunkVersion(1, 0, OID::gen())); - string errMsg; - ChunkType pending; - pending.setMin( BSON( "x" << 0 ) ); - pending.setMax( BSON( "x" << 10 ) ); + CollectionMetadata remoteMetadata; + getMetadataFor(chunks, &remoteMetadata); - unique_ptr<CollectionMetadata> cloned( afterMetadata.clonePlusPending( pending, &errMsg ) ); - ASSERT( cloned != NULL ); + Status status = loader().promotePendingChunks(&afterMetadata, &remoteMetadata); + ASSERT(status.isOK()); - status = loader().promotePendingChunks( cloned.get(), &remoteMetadata ); - ASSERT( status.isOK() ); - ASSERT_EQUALS( remoteMetadata.getNumPending(), 0u ); - } + string errMsg; + ChunkType pending; + pending.setMin(BSON("x" << 0)); + pending.setMax(BSON("x" << 10)); - TEST_F(MultipleMetadataFixture, PromotePendingNAVersion) { - OID epoch = OID::gen(); + unique_ptr<CollectionMetadata> cloned(afterMetadata.clonePlusPending(pending, &errMsg)); + ASSERT(cloned != NULL); - unique_ptr<ChunkType> chunk( new ChunkType() ); - chunk->setNS("foo.bar"); - chunk->setShard("shard0000"); - chunk->setMin( BSON( "x" << MINKEY ) ); - chunk->setMax( BSON( "x" << 0 ) ); - chunk->setVersion( ChunkVersion( 1, 1, epoch ) ); + status = loader().promotePendingChunks(cloned.get(), &remoteMetadata); + ASSERT(status.isOK()); + ASSERT_EQUALS(remoteMetadata.getNumPending(), 0u); +} - OwnedPointerVector<ChunkType> chunks; - chunks.mutableVector().push_back( chunk.release() ); +TEST_F(MultipleMetadataFixture, PromotePendingNAVersion) { + OID epoch = OID::gen(); - CollectionMetadata afterMetadata; - getMetadataFor( chunks, &afterMetadata ); + unique_ptr<ChunkType> chunk(new ChunkType()); + chunk->setNS("foo.bar"); + chunk->setShard("shard0000"); + chunk->setMin(BSON("x" << MINKEY)); + chunk->setMax(BSON("x" << 0)); + chunk->setVersion(ChunkVersion(1, 1, epoch)); - // Metadata of same epoch, but lower version - ( *chunks.vector().begin() )->setVersion( ChunkVersion( 1, 0, epoch ) ); + OwnedPointerVector<ChunkType> chunks; + chunks.mutableVector().push_back(chunk.release()); - CollectionMetadata remoteMetadata; - getMetadataFor( chunks, &remoteMetadata ); + CollectionMetadata afterMetadata; + getMetadataFor(chunks, &afterMetadata); - Status status = loader().promotePendingChunks( &afterMetadata, &remoteMetadata ); - ASSERT( status.isOK() ); + // Metadata of same epoch, but lower version + (*chunks.vector().begin())->setVersion(ChunkVersion(1, 0, epoch)); - string errMsg; - ChunkType pending; - pending.setMin( BSON( "x" << 0 ) ); - pending.setMax( BSON( "x" << 10 ) ); + CollectionMetadata remoteMetadata; + getMetadataFor(chunks, &remoteMetadata); - unique_ptr<CollectionMetadata> cloned( afterMetadata.clonePlusPending( pending, &errMsg ) ); - ASSERT( cloned != NULL ); + Status status = loader().promotePendingChunks(&afterMetadata, &remoteMetadata); + ASSERT(status.isOK()); - status = loader().promotePendingChunks( cloned.get(), &remoteMetadata ); - ASSERT( status.isOK() ); - ASSERT_EQUALS( remoteMetadata.getNumPending(), 0u ); + string errMsg; + ChunkType pending; + pending.setMin(BSON("x" << 0)); + pending.setMax(BSON("x" << 10)); - } + unique_ptr<CollectionMetadata> cloned(afterMetadata.clonePlusPending(pending, &errMsg)); + ASSERT(cloned != NULL); - TEST_F(MultipleMetadataFixture, PromotePendingGoodOverlap) { - OID epoch = OID::gen(); + status = loader().promotePendingChunks(cloned.get(), &remoteMetadata); + ASSERT(status.isOK()); + ASSERT_EQUALS(remoteMetadata.getNumPending(), 0u); +} - // - // Setup chunk range for remote metadata - // +TEST_F(MultipleMetadataFixture, PromotePendingGoodOverlap) { + OID epoch = OID::gen(); - OwnedPointerVector<ChunkType> chunks; + // + // Setup chunk range for remote metadata + // - unique_ptr<ChunkType> chunk( new ChunkType() ); - chunk->setNS("foo.bar"); - chunk->setShard("shard0000"); - chunk->setMin( BSON( "x" << MINKEY ) ); - chunk->setMax( BSON( "x" << 0 ) ); - chunk->setVersion( ChunkVersion( 1, 0, epoch ) ); - chunks.mutableVector().push_back( chunk.release() ); + OwnedPointerVector<ChunkType> chunks; - chunk.reset( new ChunkType() ); - chunk->setNS("foo.bar"); - chunk->setShard("shard0000"); - chunk->setMin( BSON( "x" << 10 ) ); - chunk->setMax( BSON( "x" << 20 ) ); - chunks.mutableVector().push_back( chunk.release() ); + unique_ptr<ChunkType> chunk(new ChunkType()); + chunk->setNS("foo.bar"); + chunk->setShard("shard0000"); + chunk->setMin(BSON("x" << MINKEY)); + chunk->setMax(BSON("x" << 0)); + chunk->setVersion(ChunkVersion(1, 0, epoch)); + chunks.mutableVector().push_back(chunk.release()); - chunk.reset( new ChunkType() ); - chunk->setNS("foo.bar"); - chunk->setShard("shard0000"); - chunk->setMin( BSON( "x" << 30 ) ); - chunk->setMax( BSON( "x" << MAXKEY ) );\ - chunks.mutableVector().push_back( chunk.release() ); + chunk.reset(new ChunkType()); + chunk->setNS("foo.bar"); + chunk->setShard("shard0000"); + chunk->setMin(BSON("x" << 10)); + chunk->setMax(BSON("x" << 20)); + chunks.mutableVector().push_back(chunk.release()); - CollectionMetadata remoteMetadata; - getMetadataFor( chunks, &remoteMetadata ); + chunk.reset(new ChunkType()); + chunk->setNS("foo.bar"); + chunk->setShard("shard0000"); + chunk->setMin(BSON("x" << 30)); + chunk->setMax(BSON("x" << MAXKEY)); + chunks.mutableVector().push_back(chunk.release()); - // - // Setup chunk and pending range for afterMetadata - // + CollectionMetadata remoteMetadata; + getMetadataFor(chunks, &remoteMetadata); - chunks.clear(); + // + // Setup chunk and pending range for afterMetadata + // - chunk.reset( new ChunkType() ); - chunk->setNS("foo.bar"); - chunk->setShard("shard0000"); - chunk->setMin( BSON( "x" << 0 ) ); - chunk->setMax( BSON( "x" << 10 ) ); - chunk->setVersion( ChunkVersion( 1, 0, epoch ) ); + chunks.clear(); - chunks.mutableVector().push_back( chunk.release() ); + chunk.reset(new ChunkType()); + chunk->setNS("foo.bar"); + chunk->setShard("shard0000"); + chunk->setMin(BSON("x" << 0)); + chunk->setMax(BSON("x" << 10)); + chunk->setVersion(ChunkVersion(1, 0, epoch)); - CollectionMetadata afterMetadata; - getMetadataFor( chunks, &afterMetadata ); + chunks.mutableVector().push_back(chunk.release()); - string errMsg; - ChunkType pending; - pending.setMin( BSON( "x" << MINKEY ) ); - pending.setMax( BSON( "x" << 0 ) ); + CollectionMetadata afterMetadata; + getMetadataFor(chunks, &afterMetadata); - unique_ptr<CollectionMetadata> cloned( afterMetadata.clonePlusPending( pending, &errMsg ) ); - ASSERT( cloned != NULL ); + string errMsg; + ChunkType pending; + pending.setMin(BSON("x" << MINKEY)); + pending.setMax(BSON("x" << 0)); - pending.setMin( BSON( "x" << 10 ) ); - pending.setMax( BSON( "x" << 20 ) ); + unique_ptr<CollectionMetadata> cloned(afterMetadata.clonePlusPending(pending, &errMsg)); + ASSERT(cloned != NULL); - cloned.reset( cloned->clonePlusPending( pending, &errMsg ) ); - ASSERT( cloned != NULL ); + pending.setMin(BSON("x" << 10)); + pending.setMax(BSON("x" << 20)); - pending.setMin( BSON( "x" << 20 ) ); - pending.setMax( BSON( "x" << 30 ) ); + cloned.reset(cloned->clonePlusPending(pending, &errMsg)); + ASSERT(cloned != NULL); - cloned.reset( cloned->clonePlusPending( pending, &errMsg ) ); - ASSERT( cloned != NULL ); + pending.setMin(BSON("x" << 20)); + pending.setMax(BSON("x" << 30)); - pending.setMin( BSON( "x" << 30 ) ); - pending.setMax( BSON( "x" << MAXKEY ) ); + cloned.reset(cloned->clonePlusPending(pending, &errMsg)); + ASSERT(cloned != NULL); - cloned.reset( cloned->clonePlusPending( pending, &errMsg ) ); - ASSERT( cloned != NULL ); + pending.setMin(BSON("x" << 30)); + pending.setMax(BSON("x" << MAXKEY)); - Status status = loader().promotePendingChunks( cloned.get(), &remoteMetadata ); - ASSERT( status.isOK() ); + cloned.reset(cloned->clonePlusPending(pending, &errMsg)); + ASSERT(cloned != NULL); - ASSERT_EQUALS( remoteMetadata.getNumPending(), 1u ); - ASSERT( remoteMetadata.keyIsPending( BSON( "x" << 25 ) ) ); - } + Status status = loader().promotePendingChunks(cloned.get(), &remoteMetadata); + ASSERT(status.isOK()); - TEST_F(MultipleMetadataFixture, PromotePendingBadOverlap) { - OID epoch = OID::gen(); + ASSERT_EQUALS(remoteMetadata.getNumPending(), 1u); + ASSERT(remoteMetadata.keyIsPending(BSON("x" << 25))); +} - // - // Setup chunk range for remote metadata - // +TEST_F(MultipleMetadataFixture, PromotePendingBadOverlap) { + OID epoch = OID::gen(); - OwnedPointerVector<ChunkType> chunks; + // + // Setup chunk range for remote metadata + // - unique_ptr<ChunkType> chunk( new ChunkType() ); - chunk->setNS("foo.bar"); - chunk->setShard("shard0000"); - chunk->setMin( BSON( "x" << MINKEY ) ); - chunk->setMax( BSON( "x" << 0 ) ); - chunk->setVersion( ChunkVersion( 1, 0, epoch ) ); + OwnedPointerVector<ChunkType> chunks; - chunks.mutableVector().push_back( chunk.release() ); + unique_ptr<ChunkType> chunk(new ChunkType()); + chunk->setNS("foo.bar"); + chunk->setShard("shard0000"); + chunk->setMin(BSON("x" << MINKEY)); + chunk->setMax(BSON("x" << 0)); + chunk->setVersion(ChunkVersion(1, 0, epoch)); - CollectionMetadata remoteMetadata; - getMetadataFor( chunks, &remoteMetadata ); + chunks.mutableVector().push_back(chunk.release()); - // - // Setup chunk and pending range for afterMetadata - // + CollectionMetadata remoteMetadata; + getMetadataFor(chunks, &remoteMetadata); - chunks.clear(); + // + // Setup chunk and pending range for afterMetadata + // - chunk.reset( new ChunkType() ); - chunk->setNS("foo.bar"); - chunk->setShard("shard0000"); - chunk->setMin( BSON( "x" << 15 ) ); - chunk->setMax( BSON( "x" << MAXKEY ) ); - chunk->setVersion( ChunkVersion( 1, 0, epoch ) ); + chunks.clear(); - chunks.mutableVector().push_back( chunk.release() ); + chunk.reset(new ChunkType()); + chunk->setNS("foo.bar"); + chunk->setShard("shard0000"); + chunk->setMin(BSON("x" << 15)); + chunk->setMax(BSON("x" << MAXKEY)); + chunk->setVersion(ChunkVersion(1, 0, epoch)); - CollectionMetadata afterMetadata; - getMetadataFor( chunks, &afterMetadata ); + chunks.mutableVector().push_back(chunk.release()); - string errMsg; - ChunkType pending; - pending.setMin( BSON( "x" << MINKEY ) ); - pending.setMax( BSON( "x" << 1 ) ); + CollectionMetadata afterMetadata; + getMetadataFor(chunks, &afterMetadata); - unique_ptr<CollectionMetadata> cloned( afterMetadata.clonePlusPending( pending, &errMsg ) ); - ASSERT( cloned != NULL ); + string errMsg; + ChunkType pending; + pending.setMin(BSON("x" << MINKEY)); + pending.setMax(BSON("x" << 1)); - cloned.reset( cloned->clonePlusPending( pending, &errMsg ) ); - ASSERT( cloned != NULL ); + unique_ptr<CollectionMetadata> cloned(afterMetadata.clonePlusPending(pending, &errMsg)); + ASSERT(cloned != NULL); - Status status = loader().promotePendingChunks( cloned.get(), &remoteMetadata ); - ASSERT_EQUALS( status.code(), ErrorCodes::RemoteChangeDetected ); - } + cloned.reset(cloned->clonePlusPending(pending, &errMsg)); + ASSERT(cloned != NULL); + + Status status = loader().promotePendingChunks(cloned.get(), &remoteMetadata); + ASSERT_EQUALS(status.code(), ErrorCodes::RemoteChangeDetected); +} #if 0 // TODO: MockServer functionality does not support selective query - consider diff --git a/src/mongo/s/mock_ns_targeter.h b/src/mongo/s/mock_ns_targeter.h index feb07602c69..6958c66f5a2 100644 --- a/src/mongo/s/mock_ns_targeter.h +++ b/src/mongo/s/mock_ns_targeter.h @@ -37,195 +37,180 @@ namespace mongo { - /** - * A MockRange represents a range with endpoint that a MockNSTargeter uses to direct writes to - * a particular endpoint. - */ - struct MockRange { +/** + * A MockRange represents a range with endpoint that a MockNSTargeter uses to direct writes to + * a particular endpoint. + */ +struct MockRange { + MockRange(const ShardEndpoint& endpoint, + const NamespaceString nss, + const BSONObj& minKey, + const BSONObj& maxKey) + : endpoint(endpoint), range(nss, minKey, maxKey, getKeyPattern(minKey)) {} + + MockRange(const ShardEndpoint& endpoint, const KeyRange& range) + : endpoint(endpoint), range(range) {} + + static BSONObj getKeyPattern(const BSONObj& key) { + BSONObjIterator it(key); + BSONObjBuilder objB; + while (it.more()) + objB.append(it.next().fieldName(), 1); + return objB.obj(); + } - MockRange( const ShardEndpoint& endpoint, - const NamespaceString nss, - const BSONObj& minKey, - const BSONObj& maxKey ) : - endpoint( endpoint ), range( nss, minKey, maxKey, getKeyPattern( minKey ) ) { + const ShardEndpoint endpoint; + const KeyRange range; +}; - } +/** + * A MockNSTargeter directs writes to particular endpoints based on a list of MockRanges given + * to the mock targeter on initialization. + * + * No refreshing behavior is currently supported. + */ +class MockNSTargeter : public NSTargeter { +public: + void init(const std::vector<MockRange*> mockRanges) { + ASSERT(!mockRanges.empty()); + _mockRanges.mutableVector().insert( + _mockRanges.mutableVector().end(), mockRanges.begin(), mockRanges.end()); + _nss = NamespaceString(_mockRanges.vector().front()->range.ns); + } - MockRange( const ShardEndpoint& endpoint, const KeyRange& range ) : - endpoint( endpoint ), range( range ) { - } + const NamespaceString& getNS() const { + return _nss; + } - static BSONObj getKeyPattern( const BSONObj& key ) { - BSONObjIterator it( key ); - BSONObjBuilder objB; - while ( it.more() ) - objB.append( it.next().fieldName(), 1 ); - return objB.obj(); - } + /** + * Returns a ShardEndpoint for the doc from the mock ranges + */ + Status targetInsert(const BSONObj& doc, ShardEndpoint** endpoint) const { + std::vector<ShardEndpoint*> endpoints; + Status status = targetQuery(doc, &endpoints); + if (!status.isOK()) + return status; + if (!endpoints.empty()) + *endpoint = endpoints.front(); + return Status::OK(); + } - const ShardEndpoint endpoint; - const KeyRange range; - }; + /** + * Returns the first ShardEndpoint for the query from the mock ranges. Only can handle + * queries of the form { field : { $gte : <value>, $lt : <value> } }. + */ + Status targetUpdate(const BatchedUpdateDocument& updateDoc, + std::vector<ShardEndpoint*>* endpoints) const { + return targetQuery(updateDoc.getQuery(), endpoints); + } /** - * A MockNSTargeter directs writes to particular endpoints based on a list of MockRanges given - * to the mock targeter on initialization. - * - * No refreshing behavior is currently supported. + * Returns the first ShardEndpoint for the query from the mock ranges. Only can handle + * queries of the form { field : { $gte : <value>, $lt : <value> } }. */ - class MockNSTargeter : public NSTargeter { - public: - - void init( const std::vector<MockRange*> mockRanges ) { - ASSERT( !mockRanges.empty() ); - _mockRanges.mutableVector().insert( _mockRanges.mutableVector().end(), - mockRanges.begin(), - mockRanges.end() ); - _nss = NamespaceString( _mockRanges.vector().front()->range.ns ); - } + Status targetDelete(const BatchedDeleteDocument& deleteDoc, + std::vector<ShardEndpoint*>* endpoints) const { + return targetQuery(deleteDoc.getQuery(), endpoints); + } - const NamespaceString& getNS() const { - return _nss; - } + Status targetCollection(std::vector<ShardEndpoint*>* endpoints) const { + // TODO: XXX + // No-op + return Status::OK(); + } - /** - * Returns a ShardEndpoint for the doc from the mock ranges - */ - Status targetInsert( const BSONObj& doc, ShardEndpoint** endpoint ) const { - std::vector<ShardEndpoint*> endpoints; - Status status = targetQuery( doc, &endpoints ); - if ( !status.isOK() ) - return status; - if ( !endpoints.empty() ) - *endpoint = endpoints.front(); - return Status::OK(); + Status targetAllShards(std::vector<ShardEndpoint*>* endpoints) const { + const std::vector<MockRange*>& ranges = getRanges(); + for (std::vector<MockRange*>::const_iterator it = ranges.begin(); it != ranges.end(); + ++it) { + const MockRange* range = *it; + endpoints->push_back(new ShardEndpoint(range->endpoint)); } - /** - * Returns the first ShardEndpoint for the query from the mock ranges. Only can handle - * queries of the form { field : { $gte : <value>, $lt : <value> } }. - */ - Status targetUpdate( const BatchedUpdateDocument& updateDoc, - std::vector<ShardEndpoint*>* endpoints ) const { - return targetQuery( updateDoc.getQuery(), endpoints ); - } + return Status::OK(); + } - /** - * Returns the first ShardEndpoint for the query from the mock ranges. Only can handle - * queries of the form { field : { $gte : <value>, $lt : <value> } }. - */ - Status targetDelete( const BatchedDeleteDocument& deleteDoc, - std::vector<ShardEndpoint*>* endpoints ) const { - return targetQuery( deleteDoc.getQuery(), endpoints ); - } + void noteCouldNotTarget() { + // No-op + } - Status targetCollection( std::vector<ShardEndpoint*>* endpoints ) const { - // TODO: XXX - // No-op - return Status::OK(); - } + void noteStaleResponse(const ShardEndpoint& endpoint, const BSONObj& staleInfo) { + // No-op + } - Status targetAllShards( std::vector<ShardEndpoint*>* endpoints ) const { - const std::vector<MockRange*>& ranges = getRanges(); - for ( std::vector<MockRange*>::const_iterator it = ranges.begin(); it != ranges.end(); - ++it ) { + Status refreshIfNeeded(bool* wasChanged) { + // No-op + if (wasChanged) + *wasChanged = false; + return Status::OK(); + } - const MockRange* range = *it; - endpoints->push_back( new ShardEndpoint( range->endpoint ) ); - } + const std::vector<MockRange*>& getRanges() const { + return _mockRanges.vector(); + } - return Status::OK(); - } +private: + KeyRange parseRange(const BSONObj& query) const { + std::string fieldName = query.firstElement().fieldName(); - void noteCouldNotTarget() { - // No-op - } + if (query.firstElement().isNumber()) { + return KeyRange("", + BSON(fieldName << query.firstElement().numberInt()), + BSON(fieldName << query.firstElement().numberInt() + 1), + BSON(fieldName << 1)); + } else if (query.firstElement().type() == Object) { + BSONObj queryRange = query.firstElement().Obj(); - void noteStaleResponse( const ShardEndpoint& endpoint, const BSONObj& staleInfo ) { - // No-op - } + ASSERT(!queryRange[GTE.l_].eoo()); + ASSERT(!queryRange[LT.l_].eoo()); - Status refreshIfNeeded( bool* wasChanged ) { - // No-op - if ( wasChanged ) - *wasChanged = false; - return Status::OK(); - } + BSONObjBuilder minKeyB; + minKeyB.appendAs(queryRange[GTE.l_], fieldName); + BSONObjBuilder maxKeyB; + maxKeyB.appendAs(queryRange[LT.l_], fieldName); - const std::vector<MockRange*>& getRanges() const { - return _mockRanges.vector(); + return KeyRange("", minKeyB.obj(), maxKeyB.obj(), BSON(fieldName << 1)); } - private: - - KeyRange parseRange( const BSONObj& query ) const { - - std::string fieldName = query.firstElement().fieldName(); - - if ( query.firstElement().isNumber() ) { - - return KeyRange( "", - BSON( fieldName << query.firstElement().numberInt() ), - BSON( fieldName << query.firstElement().numberInt() + 1 ), - BSON( fieldName << 1 ) ); - } - else if ( query.firstElement().type() == Object ) { - - BSONObj queryRange = query.firstElement().Obj(); - - ASSERT( !queryRange[GTE.l_].eoo() ); - ASSERT( !queryRange[LT.l_].eoo() ); - - BSONObjBuilder minKeyB; - minKeyB.appendAs( queryRange[GTE.l_], fieldName ); - BSONObjBuilder maxKeyB; - maxKeyB.appendAs( queryRange[LT.l_], fieldName ); + ASSERT(false); + return KeyRange("", BSONObj(), BSONObj(), BSONObj()); + } - return KeyRange( "", minKeyB.obj(), maxKeyB.obj(), BSON( fieldName << 1 ) ); + /** + * Returns the first ShardEndpoint for the query from the mock ranges. Only can handle + * queries of the form { field : { $gte : <value>, $lt : <value> } }. + */ + Status targetQuery(const BSONObj& query, std::vector<ShardEndpoint*>* endpoints) const { + KeyRange queryRange = parseRange(query); + + const std::vector<MockRange*>& ranges = getRanges(); + for (std::vector<MockRange*>::const_iterator it = ranges.begin(); it != ranges.end(); + ++it) { + const MockRange* range = *it; + + if (rangeOverlaps(queryRange.minKey, + queryRange.maxKey, + range->range.minKey, + range->range.maxKey)) { + endpoints->push_back(new ShardEndpoint(range->endpoint)); } - - ASSERT( false ); - return KeyRange( "", BSONObj(), BSONObj(), BSONObj() ); } - /** - * Returns the first ShardEndpoint for the query from the mock ranges. Only can handle - * queries of the form { field : { $gte : <value>, $lt : <value> } }. - */ - Status targetQuery( const BSONObj& query, std::vector<ShardEndpoint*>* endpoints ) const { - - KeyRange queryRange = parseRange( query ); - - const std::vector<MockRange*>& ranges = getRanges(); - for ( std::vector<MockRange*>::const_iterator it = ranges.begin(); it != ranges.end(); - ++it ) { - - const MockRange* range = *it; - - if ( rangeOverlaps( queryRange.minKey, - queryRange.maxKey, - range->range.minKey, - range->range.maxKey ) ) { - endpoints->push_back( new ShardEndpoint( range->endpoint ) ); - } - } - - if ( endpoints->empty() ) return Status( ErrorCodes::UnknownError, - "no mock ranges found for query" ); - return Status::OK(); - } + if (endpoints->empty()) + return Status(ErrorCodes::UnknownError, "no mock ranges found for query"); + return Status::OK(); + } - NamespaceString _nss; + NamespaceString _nss; - // Manually-stored ranges - OwnedPointerVector<MockRange> _mockRanges; - }; + // Manually-stored ranges + OwnedPointerVector<MockRange> _mockRanges; +}; - inline void assertEndpointsEqual( const ShardEndpoint& endpointA, - const ShardEndpoint& endpointB ) { - ASSERT_EQUALS( endpointA.shardName, endpointB.shardName ); - ASSERT_EQUALS( endpointA.shardVersion.toLong(), endpointB.shardVersion.toLong() ); - ASSERT_EQUALS( endpointA.shardVersion.epoch(), endpointB.shardVersion.epoch() ); - } +inline void assertEndpointsEqual(const ShardEndpoint& endpointA, const ShardEndpoint& endpointB) { + ASSERT_EQUALS(endpointA.shardName, endpointB.shardName); + ASSERT_EQUALS(endpointA.shardVersion.toLong(), endpointB.shardVersion.toLong()); + ASSERT_EQUALS(endpointA.shardVersion.epoch(), endpointB.shardVersion.epoch()); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/mock_shard_resolver.h b/src/mongo/s/mock_shard_resolver.h index a37c6cfe905..23e676e5836 100644 --- a/src/mongo/s/mock_shard_resolver.h +++ b/src/mongo/s/mock_shard_resolver.h @@ -37,19 +37,16 @@ namespace mongo { - class MockShardResolver : public ShardResolver { - public: - - virtual ~MockShardResolver() { - } - - Status chooseWriteHost( const std::string& shardName, ConnectionString* shardHost ) const { - std::string errMsg; - *shardHost = ConnectionString::parse( std::string( "$" ) + shardName + ":12345", errMsg ); - ASSERT_EQUALS( errMsg, "" ); - return Status::OK(); - } - - }; - -} // namespace mongo +class MockShardResolver : public ShardResolver { +public: + virtual ~MockShardResolver() {} + + Status chooseWriteHost(const std::string& shardName, ConnectionString* shardHost) const { + std::string errMsg; + *shardHost = ConnectionString::parse(std::string("$") + shardName + ":12345", errMsg); + ASSERT_EQUALS(errMsg, ""); + return Status::OK(); + } +}; + +} // namespace mongo diff --git a/src/mongo/s/mongo_version_range.cpp b/src/mongo/s/mongo_version_range.cpp index 82870e20418..7ac4b2c4474 100644 --- a/src/mongo/s/mongo_version_range.cpp +++ b/src/mongo/s/mongo_version_range.cpp @@ -32,147 +32,135 @@ namespace mongo { - using std::string; - using std::vector; - - bool MongoVersionRange::parseBSONArray(const BSONArray& arr, - vector<MongoVersionRange>* excludes, - std::string* errMsg) - { - string dummy; - if (!errMsg) errMsg = &dummy; - - BSONObjIterator it(arr); - - while (it.more()) { - MongoVersionRange range; - if (!range.parseBSONElement(it.next(), errMsg)) return false; - excludes->push_back(range); - } - - return true; - } - - BSONArray MongoVersionRange::toBSONArray(const vector<MongoVersionRange>& ranges) { +using std::string; +using std::vector; - BSONArrayBuilder barr; +bool MongoVersionRange::parseBSONArray(const BSONArray& arr, + vector<MongoVersionRange>* excludes, + std::string* errMsg) { + string dummy; + if (!errMsg) + errMsg = &dummy; - for (vector<MongoVersionRange>::const_iterator it = ranges.begin(); it != ranges.end(); - ++it) - { - const MongoVersionRange& range = *it; - range.toBSONElement(&barr); - } + BSONObjIterator it(arr); - return barr.arr(); + while (it.more()) { + MongoVersionRange range; + if (!range.parseBSONElement(it.next(), errMsg)) + return false; + excludes->push_back(range); } - bool MongoVersionRange::parseBSONElement(const BSONElement& el, string* errMsg) { - - string dummy; - if (!errMsg) errMsg = &dummy; + return true; +} - if (el.type() == String) { - minVersion = el.String(); - if (minVersion == "") { - *errMsg = (string) "cannot parse single empty mongo version (" + el.toString() - + ")"; - return false; - } - return true; - } - else if (el.type() == Array || el.type() == Object) { +BSONArray MongoVersionRange::toBSONArray(const vector<MongoVersionRange>& ranges) { + BSONArrayBuilder barr; - BSONObj range = el.Obj(); + for (vector<MongoVersionRange>::const_iterator it = ranges.begin(); it != ranges.end(); ++it) { + const MongoVersionRange& range = *it; + range.toBSONElement(&barr); + } - if (range.nFields() != 2) { - *errMsg = (string) "not enough fields in mongo version range (" + el.toString() - + ")"; - return false; - } + return barr.arr(); +} - BSONObjIterator it(range); +bool MongoVersionRange::parseBSONElement(const BSONElement& el, string* errMsg) { + string dummy; + if (!errMsg) + errMsg = &dummy; - BSONElement subElA = it.next(); - BSONElement subElB = it.next(); + if (el.type() == String) { + minVersion = el.String(); + if (minVersion == "") { + *errMsg = (string) "cannot parse single empty mongo version (" + el.toString() + ")"; + return false; + } + return true; + } else if (el.type() == Array || el.type() == Object) { + BSONObj range = el.Obj(); - if (subElA.type() != String || subElB.type() != String) { - *errMsg = (string) "wrong field type for mongo version range (" + el.toString() - + ")"; - return false; - } + if (range.nFields() != 2) { + *errMsg = (string) "not enough fields in mongo version range (" + el.toString() + ")"; + return false; + } - minVersion = subElA.String(); - maxVersion = subElB.String(); + BSONObjIterator it(range); - if (minVersion == "") { - *errMsg = (string) "cannot parse first empty mongo version (" + el.toString() + ")"; - return false; - } + BSONElement subElA = it.next(); + BSONElement subElB = it.next(); - if (maxVersion == "") { - *errMsg = (string) "cannot parse second empty mongo version (" + el.toString() - + ")"; - return false; - } + if (subElA.type() != String || subElB.type() != String) { + *errMsg = (string) "wrong field type for mongo version range (" + el.toString() + ")"; + return false; + } - if (versionCmp(minVersion, maxVersion) > 0) { - string swap = minVersion; - minVersion = maxVersion; - maxVersion = swap; - } + minVersion = subElA.String(); + maxVersion = subElB.String(); - return true; - } - else { - *errMsg = (string) "wrong type for mongo version range " + el.toString(); + if (minVersion == "") { + *errMsg = (string) "cannot parse first empty mongo version (" + el.toString() + ")"; return false; } - } - void MongoVersionRange::toBSONElement(BSONArrayBuilder* barr) const { if (maxVersion == "") { - barr->append(minVersion); + *errMsg = (string) "cannot parse second empty mongo version (" + el.toString() + ")"; + return false; } - else { - BSONArrayBuilder rangeB(barr->subarrayStart()); - - rangeB.append(minVersion); - rangeB.append(maxVersion); - rangeB.done(); + if (versionCmp(minVersion, maxVersion) > 0) { + string swap = minVersion; + minVersion = maxVersion; + maxVersion = swap; } + + return true; + } else { + *errMsg = (string) "wrong type for mongo version range " + el.toString(); + return false; } +} - bool MongoVersionRange::isInRange(StringData version) const { +void MongoVersionRange::toBSONElement(BSONArrayBuilder* barr) const { + if (maxVersion == "") { + barr->append(minVersion); + } else { + BSONArrayBuilder rangeB(barr->subarrayStart()); - if (maxVersion == "") { - // If a prefix of the version specified is excluded, the specified version is - // excluded - if (version.find(minVersion) == 0) return true; - } - else { - // Range is inclusive, so make sure the end and beginning prefix excludes all - // prefixed versions as above - if (version.find(minVersion) == 0) return true; - if (version.find(maxVersion) == 0) return true; - if (versionCmp(minVersion, version) <= 0 && versionCmp(maxVersion, version) >= 0) { - return true; - } - } + rangeB.append(minVersion); + rangeB.append(maxVersion); - return false; + rangeB.done(); } +} - bool isInMongoVersionRanges(StringData version, const vector<MongoVersionRange>& ranges) - { - for (vector<MongoVersionRange>::const_iterator it = ranges.begin(); it != ranges.end(); - ++it) - { - if (it->isInRange(version)) return true; +bool MongoVersionRange::isInRange(StringData version) const { + if (maxVersion == "") { + // If a prefix of the version specified is excluded, the specified version is + // excluded + if (version.find(minVersion) == 0) + return true; + } else { + // Range is inclusive, so make sure the end and beginning prefix excludes all + // prefixed versions as above + if (version.find(minVersion) == 0) + return true; + if (version.find(maxVersion) == 0) + return true; + if (versionCmp(minVersion, version) <= 0 && versionCmp(maxVersion, version) >= 0) { + return true; } + } - return false; + return false; +} + +bool isInMongoVersionRanges(StringData version, const vector<MongoVersionRange>& ranges) { + for (vector<MongoVersionRange>::const_iterator it = ranges.begin(); it != ranges.end(); ++it) { + if (it->isInRange(version)) + return true; } + return false; +} } diff --git a/src/mongo/s/mongo_version_range.h b/src/mongo/s/mongo_version_range.h index fcdc4dca4c2..2e5c443fcc9 100644 --- a/src/mongo/s/mongo_version_range.h +++ b/src/mongo/s/mongo_version_range.h @@ -36,33 +36,31 @@ namespace mongo { - /** - * The MongoVersionRange represents a min/max of MongoDB versions, useful for - * excluding/including particular versions. - * - * The ranges may be single-version, in which case maxVersion == "", where only exact prefix - * matches are included in the range. Alternately, the range may have a min and max version - * and include any version with a prefix of the min and max version as well as all versions - * between the two. - */ - struct MongoVersionRange { - - static bool parseBSONArray(const BSONArray& arr, - std::vector<MongoVersionRange>* excludes, - std::string* errMsg); +/** + * The MongoVersionRange represents a min/max of MongoDB versions, useful for + * excluding/including particular versions. + * + * The ranges may be single-version, in which case maxVersion == "", where only exact prefix + * matches are included in the range. Alternately, the range may have a min and max version + * and include any version with a prefix of the min and max version as well as all versions + * between the two. + */ +struct MongoVersionRange { + static bool parseBSONArray(const BSONArray& arr, + std::vector<MongoVersionRange>* excludes, + std::string* errMsg); - static BSONArray toBSONArray(const std::vector<MongoVersionRange>& ranges); + static BSONArray toBSONArray(const std::vector<MongoVersionRange>& ranges); - bool parseBSONElement(const BSONElement& el, std::string* errMsg); + bool parseBSONElement(const BSONElement& el, std::string* errMsg); - void toBSONElement(BSONArrayBuilder* barr) const; + void toBSONElement(BSONArrayBuilder* barr) const; - bool isInRange(StringData version) const; + bool isInRange(StringData version) const; - std::string minVersion; - std::string maxVersion; - }; + std::string minVersion; + std::string maxVersion; +}; - bool isInMongoVersionRanges(StringData version, - const std::vector<MongoVersionRange>& ranges); +bool isInMongoVersionRanges(StringData version, const std::vector<MongoVersionRange>& ranges); } diff --git a/src/mongo/s/mongo_version_range_test.cpp b/src/mongo/s/mongo_version_range_test.cpp index 6aedd0168af..35397b1d486 100644 --- a/src/mongo/s/mongo_version_range_test.cpp +++ b/src/mongo/s/mongo_version_range_test.cpp @@ -41,189 +41,183 @@ namespace mongo { namespace { - using std::string; - using std::vector; +using std::string; +using std::vector; +// +// Test mongo version ranges +// + +TEST(Excludes, Empty) { // - // Test mongo version ranges + // Tests basic empty range // - TEST(Excludes, Empty) { - - // - // Tests basic empty range - // - - BSONArray includeArr; - vector<MongoVersionRange> includes; - - string errMsg; - MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); - ASSERT_EQUALS(errMsg, ""); - - // Make sure nothing is included - ASSERT(!isInMongoVersionRanges("1.2.3", includes)); - ASSERT(!isInMongoVersionRanges("1.2.3-pre", includes)); - ASSERT(!isInMongoVersionRanges("1.2.3-rc0", includes)); - } - - TEST(Excludes, SinglePointRange) { - - // - // Tests single string range - // + BSONArray includeArr; + vector<MongoVersionRange> includes; - BSONArrayBuilder bab; - bab << "1.2.3"; - BSONArray includeArr = bab.arr(); + string errMsg; + MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); + ASSERT_EQUALS(errMsg, ""); - vector<MongoVersionRange> includes; - - string errMsg; - MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); - ASSERT_EQUALS(errMsg, ""); - - ASSERT(isInMongoVersionRanges("1.2.3", includes)); - - ASSERT(!isInMongoVersionRanges("1.2.2-rc0", includes)); - ASSERT(!isInMongoVersionRanges("1.2.2", includes)); + // Make sure nothing is included + ASSERT(!isInMongoVersionRanges("1.2.3", includes)); + ASSERT(!isInMongoVersionRanges("1.2.3-pre", includes)); + ASSERT(!isInMongoVersionRanges("1.2.3-rc0", includes)); +} - ASSERT(isInMongoVersionRanges("1.2.3-pre", includes)); - ASSERT(isInMongoVersionRanges("1.2.3-rc0", includes)); +TEST(Excludes, SinglePointRange) { + // + // Tests single string range + // - ASSERT(!isInMongoVersionRanges("1.2.4-rc0", includes)); - ASSERT(!isInMongoVersionRanges("1.2.4", includes)); - } + BSONArrayBuilder bab; + bab << "1.2.3"; + BSONArray includeArr = bab.arr(); - TEST(Excludes, BetweenRange) { + vector<MongoVersionRange> includes; - // - // Tests range with two endpoints - // + string errMsg; + MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); + ASSERT_EQUALS(errMsg, ""); - BSONArrayBuilder bab; - bab << BSON_ARRAY("7.8.9" << "10.11.12"); - BSONArray includeArr = bab.arr(); + ASSERT(isInMongoVersionRanges("1.2.3", includes)); - vector<MongoVersionRange> includes; + ASSERT(!isInMongoVersionRanges("1.2.2-rc0", includes)); + ASSERT(!isInMongoVersionRanges("1.2.2", includes)); - string errMsg; - MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); - ASSERT_EQUALS(errMsg, ""); + ASSERT(isInMongoVersionRanges("1.2.3-pre", includes)); + ASSERT(isInMongoVersionRanges("1.2.3-rc0", includes)); - ASSERT(isInMongoVersionRanges("7.8.9", includes)); - ASSERT(isInMongoVersionRanges("10.11.12", includes)); + ASSERT(!isInMongoVersionRanges("1.2.4-rc0", includes)); + ASSERT(!isInMongoVersionRanges("1.2.4", includes)); +} - // Before - ASSERT(!isInMongoVersionRanges("7.8.8-rc0", includes)); - ASSERT(!isInMongoVersionRanges("7.8.8", includes)); +TEST(Excludes, BetweenRange) { + // + // Tests range with two endpoints + // - // Boundary - ASSERT(isInMongoVersionRanges("7.8.9-pre", includes)); - ASSERT(isInMongoVersionRanges("7.8.9-rc0", includes)); + BSONArrayBuilder bab; + bab << BSON_ARRAY("7.8.9" + << "10.11.12"); + BSONArray includeArr = bab.arr(); - ASSERT(isInMongoVersionRanges("7.8.10-rc0", includes)); - ASSERT(isInMongoVersionRanges("7.8.10", includes)); + vector<MongoVersionRange> includes; - // Between - ASSERT(isInMongoVersionRanges("8.9.10", includes)); - ASSERT(isInMongoVersionRanges("9.10.11", includes)); + string errMsg; + MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); + ASSERT_EQUALS(errMsg, ""); - // Boundary - ASSERT(isInMongoVersionRanges("10.11.11-rc0", includes)); - ASSERT(isInMongoVersionRanges("10.11.11", includes)); + ASSERT(isInMongoVersionRanges("7.8.9", includes)); + ASSERT(isInMongoVersionRanges("10.11.12", includes)); - ASSERT(isInMongoVersionRanges("10.11.12-pre", includes)); - ASSERT(isInMongoVersionRanges("10.11.12-rc0", includes)); + // Before + ASSERT(!isInMongoVersionRanges("7.8.8-rc0", includes)); + ASSERT(!isInMongoVersionRanges("7.8.8", includes)); - // After - ASSERT(!isInMongoVersionRanges("10.11.13-rc0", includes)); - ASSERT(!isInMongoVersionRanges("10.11.13", includes)); + // Boundary + ASSERT(isInMongoVersionRanges("7.8.9-pre", includes)); + ASSERT(isInMongoVersionRanges("7.8.9-rc0", includes)); - } + ASSERT(isInMongoVersionRanges("7.8.10-rc0", includes)); + ASSERT(isInMongoVersionRanges("7.8.10", includes)); - TEST(Excludes, WeirdRange) { + // Between + ASSERT(isInMongoVersionRanges("8.9.10", includes)); + ASSERT(isInMongoVersionRanges("9.10.11", includes)); - // - // Tests range with rc/pre endpoints - // + // Boundary + ASSERT(isInMongoVersionRanges("10.11.11-rc0", includes)); + ASSERT(isInMongoVersionRanges("10.11.11", includes)); - BSONArrayBuilder bab; - bab << BSON_ARRAY("7.8.9-rc0" << "10.11.12-pre"); - BSONArray includeArr = bab.arr(); + ASSERT(isInMongoVersionRanges("10.11.12-pre", includes)); + ASSERT(isInMongoVersionRanges("10.11.12-rc0", includes)); - vector<MongoVersionRange> includes; + // After + ASSERT(!isInMongoVersionRanges("10.11.13-rc0", includes)); + ASSERT(!isInMongoVersionRanges("10.11.13", includes)); +} - string errMsg; - MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); - ASSERT_EQUALS(errMsg, ""); +TEST(Excludes, WeirdRange) { + // + // Tests range with rc/pre endpoints + // - // Near endpoints - ASSERT(isInMongoVersionRanges("7.8.9", includes)); - ASSERT(!isInMongoVersionRanges("10.11.12", includes)); + BSONArrayBuilder bab; + bab << BSON_ARRAY("7.8.9-rc0" + << "10.11.12-pre"); + BSONArray includeArr = bab.arr(); - // Before - ASSERT(!isInMongoVersionRanges("7.8.8-rc0", includes)); - ASSERT(!isInMongoVersionRanges("7.8.8", includes)); + vector<MongoVersionRange> includes; - // Boundary - ASSERT(!isInMongoVersionRanges("7.8.9-pre", includes)); - ASSERT(isInMongoVersionRanges("7.8.9-rc0", includes)); + string errMsg; + MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); + ASSERT_EQUALS(errMsg, ""); - ASSERT(isInMongoVersionRanges("7.8.10-rc0", includes)); - ASSERT(isInMongoVersionRanges("7.8.10", includes)); + // Near endpoints + ASSERT(isInMongoVersionRanges("7.8.9", includes)); + ASSERT(!isInMongoVersionRanges("10.11.12", includes)); - // Between - ASSERT(isInMongoVersionRanges("8.9.10", includes)); - ASSERT(isInMongoVersionRanges("9.10.11", includes)); + // Before + ASSERT(!isInMongoVersionRanges("7.8.8-rc0", includes)); + ASSERT(!isInMongoVersionRanges("7.8.8", includes)); - // Boundary - ASSERT(isInMongoVersionRanges("10.11.11-rc0", includes)); - ASSERT(isInMongoVersionRanges("10.11.11", includes)); + // Boundary + ASSERT(!isInMongoVersionRanges("7.8.9-pre", includes)); + ASSERT(isInMongoVersionRanges("7.8.9-rc0", includes)); - ASSERT(isInMongoVersionRanges("10.11.12-pre", includes)); - ASSERT(!isInMongoVersionRanges("10.11.12-rc0", includes)); + ASSERT(isInMongoVersionRanges("7.8.10-rc0", includes)); + ASSERT(isInMongoVersionRanges("7.8.10", includes)); - // After - ASSERT(!isInMongoVersionRanges("10.11.13-rc0", includes)); - ASSERT(!isInMongoVersionRanges("10.11.13", includes)); - } + // Between + ASSERT(isInMongoVersionRanges("8.9.10", includes)); + ASSERT(isInMongoVersionRanges("9.10.11", includes)); - TEST(Excludes, BadRangeType) { + // Boundary + ASSERT(isInMongoVersionRanges("10.11.11-rc0", includes)); + ASSERT(isInMongoVersionRanges("10.11.11", includes)); - // - // Tests range with bad endpoint types - // + ASSERT(isInMongoVersionRanges("10.11.12-pre", includes)); + ASSERT(!isInMongoVersionRanges("10.11.12-rc0", includes)); - BSONArrayBuilder bab; - bab << "1.2.3"; - bab << 2; // integer is not version - BSONArray includeArr = bab.arr(); + // After + ASSERT(!isInMongoVersionRanges("10.11.13-rc0", includes)); + ASSERT(!isInMongoVersionRanges("10.11.13", includes)); +} - vector<MongoVersionRange> includes; +TEST(Excludes, BadRangeType) { + // + // Tests range with bad endpoint types + // - string errMsg; - MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); - ASSERT_NOT_EQUALS(errMsg, ""); - } + BSONArrayBuilder bab; + bab << "1.2.3"; + bab << 2; // integer is not version + BSONArray includeArr = bab.arr(); - TEST(Excludes, BadRangeArray) { + vector<MongoVersionRange> includes; - // - // Tests range with bad array - // + string errMsg; + MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); + ASSERT_NOT_EQUALS(errMsg, ""); +} - BSONArrayBuilder bab; - bab << BSON_ARRAY("" << "1.2.3"); // empty bound - BSONArray includeArr = bab.arr(); +TEST(Excludes, BadRangeArray) { + // + // Tests range with bad array + // - vector<MongoVersionRange> includes; + BSONArrayBuilder bab; + bab << BSON_ARRAY("" + << "1.2.3"); // empty bound + BSONArray includeArr = bab.arr(); - string errMsg; - MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); - ASSERT_NOT_EQUALS(errMsg, ""); - } + vector<MongoVersionRange> includes; + string errMsg; + MongoVersionRange::parseBSONArray(includeArr, &includes, &errMsg); + ASSERT_NOT_EQUALS(errMsg, ""); } -} // namespace mongo - +} +} // namespace mongo diff --git a/src/mongo/s/mongos_options.cpp b/src/mongo/s/mongos_options.cpp index 02fffb8230a..489235cc749 100644 --- a/src/mongo/s/mongos_options.cpp +++ b/src/mongo/s/mongos_options.cpp @@ -52,241 +52,239 @@ namespace mongo { - MongosGlobalParams mongosGlobalParams; +MongosGlobalParams mongosGlobalParams; - Status addMongosOptions(moe::OptionSection* options) { +Status addMongosOptions(moe::OptionSection* options) { + moe::OptionSection general_options("General options"); - moe::OptionSection general_options("General options"); - - Status ret = addGeneralServerOptions(&general_options); - if (!ret.isOK()) { - return ret; - } + Status ret = addGeneralServerOptions(&general_options); + if (!ret.isOK()) { + return ret; + } #if defined(_WIN32) - moe::OptionSection windows_scm_options("Windows Service Control Manager options"); + moe::OptionSection windows_scm_options("Windows Service Control Manager options"); - ret = addWindowsServerOptions(&windows_scm_options); - if (!ret.isOK()) { - return ret; - } + ret = addWindowsServerOptions(&windows_scm_options); + if (!ret.isOK()) { + return ret; + } #endif #ifdef MONGO_CONFIG_SSL - moe::OptionSection ssl_options("SSL options"); + moe::OptionSection ssl_options("SSL options"); - ret = addSSLServerOptions(&ssl_options); - if (!ret.isOK()) { - return ret; - } + ret = addSSLServerOptions(&ssl_options); + if (!ret.isOK()) { + return ret; + } #endif - moe::OptionSection sharding_options("Sharding options"); + moe::OptionSection sharding_options("Sharding options"); - sharding_options.addOptionChaining("sharding.configDB", "configdb", moe::String, - "1 or 3 comma separated config servers"); + sharding_options.addOptionChaining( + "sharding.configDB", "configdb", moe::String, "1 or 3 comma separated config servers"); - sharding_options.addOptionChaining("replication.localPingThresholdMs", "localThreshold", - moe::Int, "ping time (in ms) for a node to be considered local (default 15ms)"); + sharding_options.addOptionChaining( + "replication.localPingThresholdMs", + "localThreshold", + moe::Int, + "ping time (in ms) for a node to be considered local (default 15ms)"); - sharding_options.addOptionChaining("test", "test", moe::Switch, "just run unit tests") - .setSources(moe::SourceAllLegacy); + sharding_options.addOptionChaining("test", "test", moe::Switch, "just run unit tests") + .setSources(moe::SourceAllLegacy); - sharding_options.addOptionChaining("upgrade", "upgrade", moe::Switch, - "upgrade meta data version") - .setSources(moe::SourceAllLegacy); + sharding_options.addOptionChaining( + "upgrade", "upgrade", moe::Switch, "upgrade meta data version") + .setSources(moe::SourceAllLegacy); - sharding_options.addOptionChaining("sharding.chunkSize", "chunkSize", moe::Int, - "maximum amount of data per chunk"); + sharding_options.addOptionChaining( + "sharding.chunkSize", "chunkSize", moe::Int, "maximum amount of data per chunk"); - sharding_options.addOptionChaining("net.http.JSONPEnabled", "jsonp", moe::Switch, - "allow JSONP access via http (has security implications)") - .setSources(moe::SourceAllLegacy); + sharding_options.addOptionChaining("net.http.JSONPEnabled", + "jsonp", + moe::Switch, + "allow JSONP access via http (has security implications)") + .setSources(moe::SourceAllLegacy); - sharding_options.addOptionChaining("noscripting", "noscripting", moe::Switch, - "disable scripting engine") - .setSources(moe::SourceAllLegacy); + sharding_options.addOptionChaining( + "noscripting", "noscripting", moe::Switch, "disable scripting engine") + .setSources(moe::SourceAllLegacy); - options->addSection(general_options); + options->addSection(general_options); #if defined(_WIN32) - options->addSection(windows_scm_options); + options->addSection(windows_scm_options); #endif - options->addSection(sharding_options); + options->addSection(sharding_options); #ifdef MONGO_CONFIG_SSL - options->addSection(ssl_options); + options->addSection(ssl_options); #endif - options->addOptionChaining("noAutoSplit", "noAutoSplit", moe::Switch, - "do not send split commands with writes") - .hidden() - .setSources(moe::SourceAllLegacy); + options->addOptionChaining("noAutoSplit", + "noAutoSplit", + moe::Switch, + "do not send split commands with writes") + .hidden() + .setSources(moe::SourceAllLegacy); - options->addOptionChaining("sharding.autoSplit", "", moe::Bool, - "send split commands with writes") - .setSources(moe::SourceYAMLConfig); + options->addOptionChaining( + "sharding.autoSplit", "", moe::Bool, "send split commands with writes") + .setSources(moe::SourceYAMLConfig); - return Status::OK(); - } + return Status::OK(); +} - void printMongosHelp(const moe::OptionSection& options) { - std::cout << options.helpString() << std::endl; - }; +void printMongosHelp(const moe::OptionSection& options) { + std::cout << options.helpString() << std::endl; +}; - bool handlePreValidationMongosOptions(const moe::Environment& params, - const std::vector<std::string>& args) { - if (params.count("help") && - params["help"].as<bool>() == true) { - printMongosHelp(moe::startupOptions); - return false; - } - if (params.count("version") && - params["version"].as<bool>() == true) { - printShardingVersionInfo(true); - return false; - } - if (params.count("test") && - params["test"].as<bool>() == true) { - ::mongo::logger::globalLogDomain()->setMinimumLoggedSeverity( - ::mongo::logger::LogSeverity::Debug(5)); - StartupTest::runTests(); - return false; - } - - return true; +bool handlePreValidationMongosOptions(const moe::Environment& params, + const std::vector<std::string>& args) { + if (params.count("help") && params["help"].as<bool>() == true) { + printMongosHelp(moe::startupOptions); + return false; + } + if (params.count("version") && params["version"].as<bool>() == true) { + printShardingVersionInfo(true); + return false; + } + if (params.count("test") && params["test"].as<bool>() == true) { + ::mongo::logger::globalLogDomain()->setMinimumLoggedSeverity( + ::mongo::logger::LogSeverity::Debug(5)); + StartupTest::runTests(); + return false; } - Status validateMongosOptions(const moe::Environment& params) { + return true; +} - Status ret = validateServerOptions(params); - if (!ret.isOK()) { - return ret; - } +Status validateMongosOptions(const moe::Environment& params) { + Status ret = validateServerOptions(params); + if (!ret.isOK()) { + return ret; + } - return Status::OK(); + return Status::OK(); +} + +Status canonicalizeMongosOptions(moe::Environment* params) { + Status ret = canonicalizeServerOptions(params); + if (!ret.isOK()) { + return ret; } - Status canonicalizeMongosOptions(moe::Environment* params) { +#ifdef MONGO_CONFIG_SSL + ret = canonicalizeSSLServerOptions(params); + if (!ret.isOK()) { + return ret; + } +#endif - Status ret = canonicalizeServerOptions(params); + // "sharding.autoSplit" comes from the config file, so override it if "noAutoSplit" is set + // since that comes from the command line. + if (params->count("noAutoSplit")) { + Status ret = + params->set("sharding.autoSplit", moe::Value(!(*params)["noAutoSplit"].as<bool>())); if (!ret.isOK()) { return ret; } - -#ifdef MONGO_CONFIG_SSL - ret = canonicalizeSSLServerOptions(params); + ret = params->remove("noAutoSplit"); if (!ret.isOK()) { return ret; } -#endif - - // "sharding.autoSplit" comes from the config file, so override it if "noAutoSplit" is set - // since that comes from the command line. - if (params->count("noAutoSplit")) { - Status ret = params->set("sharding.autoSplit", - moe::Value(!(*params)["noAutoSplit"].as<bool>())); - if (!ret.isOK()) { - return ret; - } - ret = params->remove("noAutoSplit"); - if (!ret.isOK()) { - return ret; - } - } - - return Status::OK(); } - Status storeMongosOptions(const moe::Environment& params, - const std::vector<std::string>& args) { + return Status::OK(); +} - Status ret = storeServerOptions(params, args); - if (!ret.isOK()) { - return ret; - } - - if ( params.count( "sharding.chunkSize" ) ) { - int csize = params["sharding.chunkSize"].as<int>(); +Status storeMongosOptions(const moe::Environment& params, const std::vector<std::string>& args) { + Status ret = storeServerOptions(params, args); + if (!ret.isOK()) { + return ret; + } - // validate chunksize before proceeding - if ( csize == 0 ) { - return Status(ErrorCodes::BadValue, "error: need a non-zero chunksize"); - } + if (params.count("sharding.chunkSize")) { + int csize = params["sharding.chunkSize"].as<int>(); - if ( !Chunk::setMaxChunkSizeSizeMB( csize ) ) { - return Status(ErrorCodes::BadValue, "MaxChunkSize invalid"); - } + // validate chunksize before proceeding + if (csize == 0) { + return Status(ErrorCodes::BadValue, "error: need a non-zero chunksize"); } - if (params.count( "net.port" ) ) { - int port = params["net.port"].as<int>(); - if ( port <= 0 || port > 65535 ) { - return Status(ErrorCodes::BadValue, - "error: port number must be between 1 and 65535"); - } + if (!Chunk::setMaxChunkSizeSizeMB(csize)) { + return Status(ErrorCodes::BadValue, "MaxChunkSize invalid"); } + } - if ( params.count( "replication.localPingThresholdMs" ) ) { - serverGlobalParams.defaultLocalThresholdMillis = - params["replication.localPingThresholdMs"].as<int>(); + if (params.count("net.port")) { + int port = params["net.port"].as<int>(); + if (port <= 0 || port > 65535) { + return Status(ErrorCodes::BadValue, "error: port number must be between 1 and 65535"); } + } - if (params.count("net.http.JSONPEnabled")) { - serverGlobalParams.jsonp = params["net.http.JSONPEnabled"].as<bool>(); - } + if (params.count("replication.localPingThresholdMs")) { + serverGlobalParams.defaultLocalThresholdMillis = + params["replication.localPingThresholdMs"].as<int>(); + } - if (params.count("noscripting")) { - // This option currently has no effect for mongos - } + if (params.count("net.http.JSONPEnabled")) { + serverGlobalParams.jsonp = params["net.http.JSONPEnabled"].as<bool>(); + } - if (params.count("sharding.autoSplit")) { - Chunk::ShouldAutoSplit = params["sharding.autoSplit"].as<bool>(); - if (Chunk::ShouldAutoSplit == false) { - warning() << "running with auto-splitting disabled"; - } - } + if (params.count("noscripting")) { + // This option currently has no effect for mongos + } - if ( ! params.count( "sharding.configDB" ) ) { - return Status(ErrorCodes::BadValue, "error: no args for --configdb"); + if (params.count("sharding.autoSplit")) { + Chunk::ShouldAutoSplit = params["sharding.autoSplit"].as<bool>(); + if (Chunk::ShouldAutoSplit == false) { + warning() << "running with auto-splitting disabled"; } + } - { - std::string configdbString = params["sharding.configDB"].as<std::string>(); + if (!params.count("sharding.configDB")) { + return Status(ErrorCodes::BadValue, "error: no args for --configdb"); + } - auto configdbConnectionString = ConnectionString::parse(configdbString); - if (!configdbConnectionString.isOK()) { - return Status(ErrorCodes::BadValue, - str::stream() << "Invalid configdb connection string: " - << configdbConnectionString.getStatus().toString()); - } + { + std::string configdbString = params["sharding.configDB"].as<std::string>(); - mongosGlobalParams.configdbs = configdbConnectionString.getValue(); + auto configdbConnectionString = ConnectionString::parse(configdbString); + if (!configdbConnectionString.isOK()) { + return Status(ErrorCodes::BadValue, + str::stream() << "Invalid configdb connection string: " + << configdbConnectionString.getStatus().toString()); } - std::vector<HostAndPort> configServers = mongosGlobalParams.configdbs.getServers(); + mongosGlobalParams.configdbs = configdbConnectionString.getValue(); + } - if (!(mongosGlobalParams.configdbs.type() == ConnectionString::SYNC) && - !(mongosGlobalParams.configdbs.type() == ConnectionString::SET && - configServers.size() == 1)) { - return Status(ErrorCodes::BadValue, - "Must have either 3 node old-style config servers, or a single server " - "replica set config server"); - } + std::vector<HostAndPort> configServers = mongosGlobalParams.configdbs.getServers(); - if (configServers.size() < 3) { - warning() << "running with less than 3 config servers should be done only for testing " - "purposes and is not recommended for production"; - } + if (!(mongosGlobalParams.configdbs.type() == ConnectionString::SYNC) && + !(mongosGlobalParams.configdbs.type() == ConnectionString::SET && + configServers.size() == 1)) { + return Status(ErrorCodes::BadValue, + "Must have either 3 node old-style config servers, or a single server " + "replica set config server"); + } - if (params.count("upgrade")) { - mongosGlobalParams.upgrade = params["upgrade"].as<bool>(); - } + if (configServers.size() < 3) { + warning() << "running with less than 3 config servers should be done only for testing " + "purposes and is not recommended for production"; + } - return Status::OK(); + if (params.count("upgrade")) { + mongosGlobalParams.upgrade = params["upgrade"].as<bool>(); } -} // namespace mongo + return Status::OK(); +} + +} // namespace mongo diff --git a/src/mongo/s/mongos_options.h b/src/mongo/s/mongos_options.h index 81089910629..873ab6aae33 100644 --- a/src/mongo/s/mongos_options.h +++ b/src/mongo/s/mongos_options.h @@ -36,54 +36,52 @@ namespace mongo { - namespace optionenvironment { - class OptionSection; - class Environment; - } // namespace optionenvironment +namespace optionenvironment { +class OptionSection; +class Environment; +} // namespace optionenvironment - namespace moe = mongo::optionenvironment; +namespace moe = mongo::optionenvironment; - struct MongosGlobalParams { - ConnectionString configdbs; - bool upgrade; +struct MongosGlobalParams { + ConnectionString configdbs; + bool upgrade; - MongosGlobalParams() : - upgrade(false) - { } - }; + MongosGlobalParams() : upgrade(false) {} +}; - extern MongosGlobalParams mongosGlobalParams; +extern MongosGlobalParams mongosGlobalParams; - Status addMongosOptions(moe::OptionSection* options); +Status addMongosOptions(moe::OptionSection* options); - void printMongosHelp(const moe::OptionSection& options); +void printMongosHelp(const moe::OptionSection& options); - /** - * Handle options that should come before validation, such as "help". - * - * Returns false if an option was found that implies we should prematurely exit with success. - */ - bool handlePreValidationMongosOptions(const moe::Environment& params, - const std::vector<std::string>& args); +/** + * Handle options that should come before validation, such as "help". + * + * Returns false if an option was found that implies we should prematurely exit with success. + */ +bool handlePreValidationMongosOptions(const moe::Environment& params, + const std::vector<std::string>& args); - /** - * Handle custom validation of mongos options that can not currently be done by using - * Constraints in the Environment. See the "validate" function in the Environment class for - * more details. - */ - Status validateMongosOptions(const moe::Environment& params); +/** + * Handle custom validation of mongos options that can not currently be done by using + * Constraints in the Environment. See the "validate" function in the Environment class for + * more details. + */ +Status validateMongosOptions(const moe::Environment& params); - /** - * Canonicalize mongos options for the given environment. - * - * For example, the options "dur", "nodur", "journal", "nojournal", and - * "storage.journaling.enabled" should all be merged into "storage.journaling.enabled". - */ - Status canonicalizeMongosOptions(moe::Environment* params); +/** + * Canonicalize mongos options for the given environment. + * + * For example, the options "dur", "nodur", "journal", "nojournal", and + * "storage.journaling.enabled" should all be merged into "storage.journaling.enabled". + */ +Status canonicalizeMongosOptions(moe::Environment* params); - Status storeMongosOptions(const moe::Environment& params, const std::vector<std::string>& args); +Status storeMongosOptions(const moe::Environment& params, const std::vector<std::string>& args); - // This function should eventually go away, but needs to be here now because the sorter and - // the version manager must know at runtime which binary it is in. - bool isMongos(); +// This function should eventually go away, but needs to be here now because the sorter and +// the version manager must know at runtime which binary it is in. +bool isMongos(); } diff --git a/src/mongo/s/mongos_options_init.cpp b/src/mongo/s/mongos_options_init.cpp index 0ec9bca0b91..335ddef25d8 100644 --- a/src/mongo/s/mongos_options_init.cpp +++ b/src/mongo/s/mongos_options_init.cpp @@ -33,47 +33,46 @@ #include "mongo/util/quick_exit.h" namespace mongo { - MONGO_GENERAL_STARTUP_OPTIONS_REGISTER(MongosOptions)(InitializerContext* context) { - return addMongosOptions(&moe::startupOptions); - } +MONGO_GENERAL_STARTUP_OPTIONS_REGISTER(MongosOptions)(InitializerContext* context) { + return addMongosOptions(&moe::startupOptions); +} - MONGO_STARTUP_OPTIONS_VALIDATE(MongosOptions)(InitializerContext* context) { - if (!handlePreValidationMongosOptions(moe::startupOptionsParsed, context->args())) { - quickExit(EXIT_SUCCESS); - } - // Run validation, but tell the Environment that we don't want it to be set as "valid", - // since we may be making it invalid in the canonicalization process. - Status ret = moe::startupOptionsParsed.validate(false/*setValid*/); - if (!ret.isOK()) { - return ret; - } - ret = validateMongosOptions(moe::startupOptionsParsed); - if (!ret.isOK()) { - return ret; - } - ret = canonicalizeMongosOptions(&moe::startupOptionsParsed); - if (!ret.isOK()) { - return ret; - } - ret = moe::startupOptionsParsed.validate(); - if (!ret.isOK()) { - return ret; - } - return Status::OK(); +MONGO_STARTUP_OPTIONS_VALIDATE(MongosOptions)(InitializerContext* context) { + if (!handlePreValidationMongosOptions(moe::startupOptionsParsed, context->args())) { + quickExit(EXIT_SUCCESS); + } + // Run validation, but tell the Environment that we don't want it to be set as "valid", + // since we may be making it invalid in the canonicalization process. + Status ret = moe::startupOptionsParsed.validate(false /*setValid*/); + if (!ret.isOK()) { + return ret; + } + ret = validateMongosOptions(moe::startupOptionsParsed); + if (!ret.isOK()) { + return ret; + } + ret = canonicalizeMongosOptions(&moe::startupOptionsParsed); + if (!ret.isOK()) { + return ret; + } + ret = moe::startupOptionsParsed.validate(); + if (!ret.isOK()) { + return ret; } + return Status::OK(); +} - MONGO_INITIALIZER_GENERAL(MongosOptions_Store, - ("BeginStartupOptionStorage"), - ("EndStartupOptionStorage")) - (InitializerContext* context) { - Status ret = storeMongosOptions(moe::startupOptionsParsed, context->args()); - if (!ret.isOK()) { - std::cerr << ret.toString() << std::endl; - std::cerr << "try '" << context->args()[0] << " --help' for more information" - << std::endl; - quickExit(EXIT_BADOPTIONS); - } - return Status::OK(); +MONGO_INITIALIZER_GENERAL(MongosOptions_Store, + ("BeginStartupOptionStorage"), + ("EndStartupOptionStorage")) +(InitializerContext* context) { + Status ret = storeMongosOptions(moe::startupOptionsParsed, context->args()); + if (!ret.isOK()) { + std::cerr << ret.toString() << std::endl; + std::cerr << "try '" << context->args()[0] << " --help' for more information" << std::endl; + quickExit(EXIT_BADOPTIONS); } + return Status::OK(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/ns_targeter.h b/src/mongo/s/ns_targeter.h index 9212b191370..0b8e513f7fa 100644 --- a/src/mongo/s/ns_targeter.h +++ b/src/mongo/s/ns_targeter.h @@ -40,149 +40,140 @@ namespace mongo { - struct ShardEndpoint; +struct ShardEndpoint; + +/** + * The NSTargeter interface is used by a WriteOp to generate and target child write operations + * to a particular collection. + * + * The lifecyle of a NSTargeter is: + * + * 0. targetDoc/targetQuery as many times as is required + * + * 1a. On targeting failure we may need to refresh, note that it happened. + * 1b. On stale config from a child write operation we may need to refresh, note the error. + * + * 2. RefreshIfNeeded() to get newer targeting information + * + * 3. Goto 0. + * + * The refreshIfNeeded() operation must try to make progress against noted targeting or stale + * config failures, see comments below. No functions may block for shared resources or network + * calls except refreshIfNeeded(). + * + * Implementers are free to define more specific targeting error codes to allow more complex + * error handling. + * + * Interface must be externally synchronized if used in multiple threads, for now. + * TODO: Determine if we should internally synchronize. + */ +class NSTargeter { +public: + virtual ~NSTargeter() {} /** - * The NSTargeter interface is used by a WriteOp to generate and target child write operations - * to a particular collection. - * - * The lifecyle of a NSTargeter is: + * Returns the namespace targeted. + */ + virtual const NamespaceString& getNS() const = 0; + + /** + * Returns a ShardEndpoint for a single document write. * - * 0. targetDoc/targetQuery as many times as is required + * Returns !OK with message if document could not be targeted for other reasons. + */ + virtual Status targetInsert(const BSONObj& doc, ShardEndpoint** endpoint) const = 0; + + /** + * Returns a vector of ShardEndpoints for a potentially multi-shard update. * - * 1a. On targeting failure we may need to refresh, note that it happened. - * 1b. On stale config from a child write operation we may need to refresh, note the error. + * Returns OK and fills the endpoints; returns a status describing the error otherwise. + */ + virtual Status targetUpdate(const BatchedUpdateDocument& updateDoc, + std::vector<ShardEndpoint*>* endpoints) const = 0; + + /** + * Returns a vector of ShardEndpoints for a potentially multi-shard delete. * - * 2. RefreshIfNeeded() to get newer targeting information + * Returns OK and fills the endpoints; returns a status describing the error otherwise. + */ + virtual Status targetDelete(const BatchedDeleteDocument& deleteDoc, + std::vector<ShardEndpoint*>* endpoints) const = 0; + + /** + * Returns a vector of ShardEndpoints for the entire collection. * - * 3. Goto 0. + * Returns !OK with message if the full collection could not be targeted. + */ + virtual Status targetCollection(std::vector<ShardEndpoint*>* endpoints) const = 0; + + /** + * Returns a vector of ShardEndpoints for all shards. * - * The refreshIfNeeded() operation must try to make progress against noted targeting or stale - * config failures, see comments below. No functions may block for shared resources or network - * calls except refreshIfNeeded(). + * Returns !OK with message if all shards could not be targeted. + */ + virtual Status targetAllShards(std::vector<ShardEndpoint*>* endpoints) const = 0; + + /** + * Informs the targeter that a targeting failure occurred during one of the last targeting + * operations. If this is noted, we cannot note stale responses. + */ + virtual void noteCouldNotTarget() = 0; + + /** + * Informs the targeter of stale config responses for this namespace from an endpoint, with + * further information available in the returned staleInfo. * - * Implementers are free to define more specific targeting error codes to allow more complex - * error handling. + * Any stale responses noted here will be taken into account on the next refresh. * - * Interface must be externally synchronized if used in multiple threads, for now. - * TODO: Determine if we should internally synchronize. + * If stale responses are is noted, we must not have noted that we cannot target. */ - class NSTargeter { - public: - - virtual ~NSTargeter() { - } - - /** - * Returns the namespace targeted. - */ - virtual const NamespaceString& getNS() const = 0; - - /** - * Returns a ShardEndpoint for a single document write. - * - * Returns !OK with message if document could not be targeted for other reasons. - */ - virtual Status targetInsert( const BSONObj& doc, ShardEndpoint** endpoint ) const = 0; - - /** - * Returns a vector of ShardEndpoints for a potentially multi-shard update. - * - * Returns OK and fills the endpoints; returns a status describing the error otherwise. - */ - virtual Status targetUpdate( const BatchedUpdateDocument& updateDoc, - std::vector<ShardEndpoint*>* endpoints ) const = 0; - - /** - * Returns a vector of ShardEndpoints for a potentially multi-shard delete. - * - * Returns OK and fills the endpoints; returns a status describing the error otherwise. - */ - virtual Status targetDelete( const BatchedDeleteDocument& deleteDoc, - std::vector<ShardEndpoint*>* endpoints ) const = 0; - - /** - * Returns a vector of ShardEndpoints for the entire collection. - * - * Returns !OK with message if the full collection could not be targeted. - */ - virtual Status targetCollection( std::vector<ShardEndpoint*>* endpoints ) const = 0; - - /** - * Returns a vector of ShardEndpoints for all shards. - * - * Returns !OK with message if all shards could not be targeted. - */ - virtual Status targetAllShards( std::vector<ShardEndpoint*>* endpoints ) const = 0; - - /** - * Informs the targeter that a targeting failure occurred during one of the last targeting - * operations. If this is noted, we cannot note stale responses. - */ - virtual void noteCouldNotTarget() = 0; - - /** - * Informs the targeter of stale config responses for this namespace from an endpoint, with - * further information available in the returned staleInfo. - * - * Any stale responses noted here will be taken into account on the next refresh. - * - * If stale responses are is noted, we must not have noted that we cannot target. - */ - virtual void noteStaleResponse( const ShardEndpoint& endpoint, - const BSONObj& staleInfo ) = 0; - - /** - * Refreshes the targeting metadata for the namespace if needed, based on previously-noted - * stale responses and targeting failures. - * - * After this function is called, the targeter should be in a state such that the noted - * stale responses are not seen again and if a targeting failure occurred it reloaded - - * it should try to make progress. If provided, wasChanged is set to true if the targeting - * information used here was changed. - * - * NOTE: This function may block for shared resources or network calls. - * Returns !OK with message if could not refresh - */ - virtual Status refreshIfNeeded( bool* wasChanged ) = 0; - - }; + virtual void noteStaleResponse(const ShardEndpoint& endpoint, const BSONObj& staleInfo) = 0; /** - * A ShardEndpoint represents a destination for a targeted query or document. It contains both - * the logical target (shard name/version/broadcast) and the physical target (host name). + * Refreshes the targeting metadata for the namespace if needed, based on previously-noted + * stale responses and targeting failures. + * + * After this function is called, the targeter should be in a state such that the noted + * stale responses are not seen again and if a targeting failure occurred it reloaded - + * it should try to make progress. If provided, wasChanged is set to true if the targeting + * information used here was changed. + * + * NOTE: This function may block for shared resources or network calls. + * Returns !OK with message if could not refresh */ - struct ShardEndpoint { + virtual Status refreshIfNeeded(bool* wasChanged) = 0; +}; - ShardEndpoint() { - } +/** + * A ShardEndpoint represents a destination for a targeted query or document. It contains both + * the logical target (shard name/version/broadcast) and the physical target (host name). + */ +struct ShardEndpoint { + ShardEndpoint() {} - ShardEndpoint( const ShardEndpoint& other ) : - shardName( other.shardName ), shardVersion( other.shardVersion ) { - } + ShardEndpoint(const ShardEndpoint& other) + : shardName(other.shardName), shardVersion(other.shardVersion) {} - ShardEndpoint( const std::string& shardName, - const ChunkVersion& shardVersion ) : - shardName( shardName ), shardVersion( shardVersion ) { - } + ShardEndpoint(const std::string& shardName, const ChunkVersion& shardVersion) + : shardName(shardName), shardVersion(shardVersion) {} - const std::string shardName; - const ChunkVersion shardVersion; + const std::string shardName; + const ChunkVersion shardVersion; - // - // For testing *only* - do not use as part of API - // + // + // For testing *only* - do not use as part of API + // - BSONObj toBSON() const { - BSONObjBuilder b; - appendBSON( &b ); - return b.obj(); - } + BSONObj toBSON() const { + BSONObjBuilder b; + appendBSON(&b); + return b.obj(); + } - void appendBSON( BSONObjBuilder* builder ) const { - builder->append( "shardName", shardName ); - shardVersion.addToBSON( *builder, "shardVersion" ); - } - }; + void appendBSON(BSONObjBuilder* builder) const { + builder->append("shardName", shardName); + shardVersion.addToBSON(*builder, "shardVersion"); + } +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/request.cpp b/src/mongo/s/request.cpp index 6ece1659e93..89fd65e079c 100644 --- a/src/mongo/s/request.cpp +++ b/src/mongo/s/request.cpp @@ -48,115 +48,98 @@ namespace mongo { - using std::endl; - using std::string; - - Request::Request(Message& m, AbstractMessagingPort* p) - : _clientInfo(&cc()), - _m(m), - _d(m), - _p(p), - _id(_m.header().getId()), - _didInit(false) { - - ClusterLastErrorInfo::get(_clientInfo).newRequest(); - } - - void Request::init() { - if (_didInit) { - return; - } +using std::endl; +using std::string; - _m.header().setId(_id); - LastError::get(_clientInfo).startRequest(); - ClusterLastErrorInfo::get(_clientInfo).clearRequestInfo(); +Request::Request(Message& m, AbstractMessagingPort* p) + : _clientInfo(&cc()), _m(m), _d(m), _p(p), _id(_m.header().getId()), _didInit(false) { + ClusterLastErrorInfo::get(_clientInfo).newRequest(); +} - if (_d.messageShouldHaveNs()) { - const NamespaceString nss(getns()); - - uassert(ErrorCodes::IllegalOperation, - "can't use 'local' database through mongos", - nss.db() != "local"); - - uassert(ErrorCodes::InvalidNamespace, - str::stream() << "Invalid ns [" << nss.ns() << "]", - nss.isValid()); - } - - AuthorizationSession::get(_clientInfo)->startRequest(NULL); - _didInit = true; +void Request::init() { + if (_didInit) { + return; } - void Request::process( int attempt ) { - init(); - int op = _m.operation(); - verify( op > dbMsg ); + _m.header().setId(_id); + LastError::get(_clientInfo).startRequest(); + ClusterLastErrorInfo::get(_clientInfo).clearRequestInfo(); - const MSGID msgId = _m.header().getId(); + if (_d.messageShouldHaveNs()) { + const NamespaceString nss(getns()); - Timer t; - LOG(3) << "Request::process begin ns: " << getns() - << " msg id: " << msgId - << " op: " << op - << " attempt: " << attempt - << endl; + uassert(ErrorCodes::IllegalOperation, + "can't use 'local' database through mongos", + nss.db() != "local"); - _d.markSet(); + uassert(ErrorCodes::InvalidNamespace, + str::stream() << "Invalid ns [" << nss.ns() << "]", + nss.isValid()); + } - bool iscmd = false; - if ( op == dbKillCursors ) { - cursorCache.gotKillCursors( _m ); - globalOpCounters.gotOp( op , iscmd ); - } - else if ( op == dbQuery ) { - NamespaceString nss(getns()); - iscmd = nss.isCommand() || nss.isSpecialCommand(); - - if (iscmd) { - int n = _d.getQueryNToReturn(); - uassert( 16978, str::stream() << "bad numberToReturn (" << n - << ") for $cmd type ns - can only be 1 or -1", - n == 1 || n == -1 ); - - Strategy::clientCommandOp(*this); - } - else { - Strategy::queryOp( *this ); - } - - globalOpCounters.gotOp( op , iscmd ); - } - else if ( op == dbGetMore ) { - Strategy::getMore( *this ); - globalOpCounters.gotOp( op , iscmd ); - } - else { - Strategy::writeOp( op, *this ); - // globalOpCounters are handled by write commands. + AuthorizationSession::get(_clientInfo)->startRequest(NULL); + _didInit = true; +} + +void Request::process(int attempt) { + init(); + int op = _m.operation(); + verify(op > dbMsg); + + const MSGID msgId = _m.header().getId(); + + Timer t; + LOG(3) << "Request::process begin ns: " << getns() << " msg id: " << msgId << " op: " << op + << " attempt: " << attempt << endl; + + _d.markSet(); + + bool iscmd = false; + if (op == dbKillCursors) { + cursorCache.gotKillCursors(_m); + globalOpCounters.gotOp(op, iscmd); + } else if (op == dbQuery) { + NamespaceString nss(getns()); + iscmd = nss.isCommand() || nss.isSpecialCommand(); + + if (iscmd) { + int n = _d.getQueryNToReturn(); + uassert(16978, + str::stream() << "bad numberToReturn (" << n + << ") for $cmd type ns - can only be 1 or -1", + n == 1 || n == -1); + + Strategy::clientCommandOp(*this); + } else { + Strategy::queryOp(*this); } - LOG(3) << "Request::process end ns: " << getns() - << " msg id: " << msgId - << " op: " << op - << " attempt: " << attempt - << " " << t.millis() << "ms" - << endl; + globalOpCounters.gotOp(op, iscmd); + } else if (op == dbGetMore) { + Strategy::getMore(*this); + globalOpCounters.gotOp(op, iscmd); + } else { + Strategy::writeOp(op, *this); + // globalOpCounters are handled by write commands. } - void Request::reply( Message & response , const string& fromServer ) { - verify( _didInit ); - long long cursor = response.header().getCursor(); - if ( cursor ) { - if ( fromServer.size() ) { - cursorCache.storeRef(fromServer, cursor, getns()); - } - else { - // probably a getMore - // make sure we have a ref for this - verify( cursorCache.getRef( cursor ).size() ); - } + LOG(3) << "Request::process end ns: " << getns() << " msg id: " << msgId << " op: " << op + << " attempt: " << attempt << " " << t.millis() << "ms" << endl; +} + +void Request::reply(Message& response, const string& fromServer) { + verify(_didInit); + long long cursor = response.header().getCursor(); + if (cursor) { + if (fromServer.size()) { + cursorCache.storeRef(fromServer, cursor, getns()); + } else { + // probably a getMore + // make sure we have a ref for this + verify(cursorCache.getRef(cursor).size()); } - _p->reply( _m , response , _id ); } + _p->reply(_m, response, _id); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/request.h b/src/mongo/s/request.h index 4bdf7a816bb..775b44f6511 100644 --- a/src/mongo/s/request.h +++ b/src/mongo/s/request.h @@ -35,51 +35,57 @@ namespace mongo { - class Client; +class Client; - class Request { - MONGO_DISALLOW_COPYING(Request); - public: - Request(Message& m, AbstractMessagingPort* p); +class Request { + MONGO_DISALLOW_COPYING(Request); - const char* getns() const { - return _d.getns(); - } +public: + Request(Message& m, AbstractMessagingPort* p); - int op() const { - return _m.operation(); - } + const char* getns() const { + return _d.getns(); + } - bool expectResponse() const { - return op() == dbQuery || op() == dbGetMore; - } + int op() const { + return _m.operation(); + } - bool isCommand() const; + bool expectResponse() const { + return op() == dbQuery || op() == dbGetMore; + } - MSGID id() const { - return _id; - } + bool isCommand() const; - void reply(Message & response, const std::string& fromServer); + MSGID id() const { + return _id; + } - Message& m() { return _m; } - DbMessage& d() { return _d; } - AbstractMessagingPort* p() const { return _p; } + void reply(Message& response, const std::string& fromServer); - void process( int attempt = 0 ); + Message& m() { + return _m; + } + DbMessage& d() { + return _d; + } + AbstractMessagingPort* p() const { + return _p; + } - void init(); + void process(int attempt = 0); - private: - Client* const _clientInfo; + void init(); - Message& _m; - DbMessage _d; - AbstractMessagingPort* const _p; +private: + Client* const _clientInfo; - MSGID _id; + Message& _m; + DbMessage _d; + AbstractMessagingPort* const _p; - bool _didInit; - }; + MSGID _id; + bool _didInit; +}; } diff --git a/src/mongo/s/s_only.cpp b/src/mongo/s/s_only.cpp index bd13778506a..671e4708864 100644 --- a/src/mongo/s/s_only.cpp +++ b/src/mongo/s/s_only.cpp @@ -49,130 +49,121 @@ namespace mongo { - using std::string; - using std::stringstream; - - - bool isMongos() { - return true; - } - - /** When this callback is run, we record a shard that we've used for useful work - * in an operation to be read later by getLastError() - */ - void usingAShardConnection(const std::string& addr) { - ClusterLastErrorInfo::get(cc()).addShardHost(addr); +using std::string; +using std::stringstream; + + +bool isMongos() { + return true; +} + +/** When this callback is run, we record a shard that we've used for useful work + * in an operation to be read later by getLastError() +*/ +void usingAShardConnection(const std::string& addr) { + ClusterLastErrorInfo::get(cc()).addShardHost(addr); +} + +// called into by the web server. For now we just translate the parameters +// to their old style equivalents. +void Command::execCommand(OperationContext* txn, + Command* command, + const rpc::RequestInterface& request, + rpc::ReplyBuilderInterface* replyBuilder) { + int queryFlags = 0; + BSONObj cmdObj; + + std::tie(cmdObj, queryFlags) = uassertStatusOK( + rpc::downconvertRequestMetadata(request.getCommandArgs(), request.getMetadata())); + + std::string db = request.getDatabase().rawData(); + BSONObjBuilder result; + + execCommandClientBasic(txn, + command, + *txn->getClient(), + queryFlags, + request.getDatabase().rawData(), + cmdObj, + result); + + replyBuilder->setMetadata(rpc::makeEmptyMetadata()).setCommandReply(result.done()); +} + +void Command::execCommandClientBasic(OperationContext* txn, + Command* c, + ClientBasic& client, + int queryOptions, + const char* ns, + BSONObj& cmdObj, + BSONObjBuilder& result) { + std::string dbname = nsToDatabase(ns); + + if (cmdObj.getBoolField("help")) { + stringstream help; + help << "help for: " << c->name << " "; + c->help(help); + result.append("help", help.str()); + result.append("lockType", c->isWriteCommandForConfigServer() ? 1 : 0); + appendCommandStatus(result, true, ""); + return; } - // called into by the web server. For now we just translate the parameters - // to their old style equivalents. - void Command::execCommand(OperationContext* txn, - Command* command, - const rpc::RequestInterface& request, - rpc::ReplyBuilderInterface* replyBuilder) { - - int queryFlags = 0; - BSONObj cmdObj; - - std::tie(cmdObj, queryFlags) = uassertStatusOK( - rpc::downconvertRequestMetadata(request.getCommandArgs(), - request.getMetadata()) - ); - - std::string db = request.getDatabase().rawData(); - BSONObjBuilder result; - - execCommandClientBasic(txn, - command, - *txn->getClient(), - queryFlags, - request.getDatabase().rawData(), - cmdObj, - result); - - replyBuilder - ->setMetadata(rpc::makeEmptyMetadata()) - .setCommandReply(result.done()); + Status status = _checkAuthorization(c, &client, dbname, cmdObj); + if (!status.isOK()) { + appendCommandStatus(result, status); + return; } - void Command::execCommandClientBasic(OperationContext* txn, - Command * c , - ClientBasic& client, - int queryOptions, - const char *ns, - BSONObj& cmdObj, - BSONObjBuilder& result) { - std::string dbname = nsToDatabase(ns); - - if (cmdObj.getBoolField("help")) { - stringstream help; - help << "help for: " << c->name << " "; - c->help( help ); - result.append( "help" , help.str() ); - result.append("lockType", c->isWriteCommandForConfigServer() ? 1 : 0); - appendCommandStatus(result, true, ""); - return; - } - - Status status = _checkAuthorization(c, &client, dbname, cmdObj); - if (!status.isOK()) { - appendCommandStatus(result, status); - return; - } - - c->_commandsExecuted.increment(); - - std::string errmsg; - bool ok; - try { - ok = c->run(txn, dbname , cmdObj, queryOptions, errmsg, result); - } - catch (const DBException& e) { - ok = false; - int code = e.getCode(); - if (code == RecvStaleConfigCode) { // code for StaleConfigException - throw; - } - - errmsg = e.what(); - result.append("code", code); + c->_commandsExecuted.increment(); + + std::string errmsg; + bool ok; + try { + ok = c->run(txn, dbname, cmdObj, queryOptions, errmsg, result); + } catch (const DBException& e) { + ok = false; + int code = e.getCode(); + if (code == RecvStaleConfigCode) { // code for StaleConfigException + throw; } - if ( !ok ) { - c->_commandsFailed.increment(); - } + errmsg = e.what(); + result.append("code", code); + } - appendCommandStatus(result, ok, errmsg); + if (!ok) { + c->_commandsFailed.increment(); } - void Command::runAgainstRegistered(const char *ns, - BSONObj& jsobj, - BSONObjBuilder& anObjBuilder, - int queryOptions) { + appendCommandStatus(result, ok, errmsg); +} - // It should be impossible for this uassert to fail since there should be no way to get - // into this function with any other collection name. - uassert(16618, +void Command::runAgainstRegistered(const char* ns, + BSONObj& jsobj, + BSONObjBuilder& anObjBuilder, + int queryOptions) { + // It should be impossible for this uassert to fail since there should be no way to get + // into this function with any other collection name. + uassert(16618, "Illegal attempt to run a command against a namespace other than $cmd.", nsToCollectionSubstring(ns) == "$cmd"); - BSONElement e = jsobj.firstElement(); - std::string commandName = e.fieldName(); - Command* c = e.type() ? Command::findCommand(commandName) : NULL; - if (!c) { - Command::appendCommandStatus(anObjBuilder, - false, - str::stream() << "no such cmd: " << commandName); - anObjBuilder.append("code", ErrorCodes::CommandNotFound); - Command::unknownCommands.increment(); - return; - } - - auto txn = cc().makeOperationContext(); - execCommandClientBasic(txn.get(), c, cc(), queryOptions, ns, jsobj, anObjBuilder); + BSONElement e = jsobj.firstElement(); + std::string commandName = e.fieldName(); + Command* c = e.type() ? Command::findCommand(commandName) : NULL; + if (!c) { + Command::appendCommandStatus( + anObjBuilder, false, str::stream() << "no such cmd: " << commandName); + anObjBuilder.append("code", ErrorCodes::CommandNotFound); + Command::unknownCommands.increment(); + return; } - void Command::registerError(OperationContext* txn, const DBException& exception) { - } + auto txn = cc().makeOperationContext(); + execCommandClientBasic(txn.get(), c, cc(), queryOptions, ns, jsobj, anObjBuilder); +} + +void Command::registerError(OperationContext* txn, const DBException& exception) {} -} //namespace mongo +} // namespace mongo diff --git a/src/mongo/s/server.cpp b/src/mongo/s/server.cpp index 092a7339d67..94ed60e91de 100644 --- a/src/mongo/s/server.cpp +++ b/src/mongo/s/server.cpp @@ -92,125 +92,119 @@ namespace mongo { - using std::string; - using std::vector; +using std::string; +using std::vector; - using logger::LogComponent; +using logger::LogComponent; #if defined(_WIN32) - ntservice::NtServiceDefaultStrings defaultServiceStrings = { - L"MongoS", - L"MongoDB Router", - L"MongoDB Sharding Router" - }; - static ExitCode initService(); +ntservice::NtServiceDefaultStrings defaultServiceStrings = { + L"MongoS", L"MongoDB Router", L"MongoDB Sharding Router"}; +static ExitCode initService(); #endif - bool dbexitCalled = false; +bool dbexitCalled = false; - bool inShutdown() { - return dbexitCalled; - } +bool inShutdown() { + return dbexitCalled; +} - bool haveLocalShardingInfo( Client* client, const string& ns ) { - verify( 0 ); - return false; - } +bool haveLocalShardingInfo(Client* client, const string& ns) { + verify(0); + return false; +} - static BSONObj buildErrReply( const DBException& ex ) { - BSONObjBuilder errB; - errB.append( "$err", ex.what() ); - errB.append( "code", ex.getCode() ); - if ( !ex._shard.empty() ) { - errB.append( "shard", ex._shard ); - } - return errB.obj(); +static BSONObj buildErrReply(const DBException& ex) { + BSONObjBuilder errB; + errB.append("$err", ex.what()); + errB.append("code", ex.getCode()); + if (!ex._shard.empty()) { + errB.append("shard", ex._shard); } + return errB.obj(); +} - class ShardedMessageHandler : public MessageHandler { - public: - virtual ~ShardedMessageHandler() {} - - virtual void connected(AbstractMessagingPort* p) { - Client::initThread("conn", getGlobalServiceContext(), p); - } - - virtual void process(Message& m, AbstractMessagingPort* p) { - verify( p ); - Request r( m , p ); - - try { - r.init(); - r.process(); - } - catch ( const AssertionException& ex ) { - - LOG( ex.isUserAssertion() ? 1 : 0 ) << "Assertion failed" - << " while processing " << opToString( m.operation() ) << " op" - << " for " << r.getns() << causedBy( ex ); +class ShardedMessageHandler : public MessageHandler { +public: + virtual ~ShardedMessageHandler() {} - if ( r.expectResponse() ) { - m.header().setId(r.id()); - replyToQuery( ResultFlag_ErrSet, p , m , buildErrReply( ex ) ); - } + virtual void connected(AbstractMessagingPort* p) { + Client::initThread("conn", getGlobalServiceContext(), p); + } - // We *always* populate the last error for now - LastError::get(cc()).setLastError(ex.getCode(), ex.what()); + virtual void process(Message& m, AbstractMessagingPort* p) { + verify(p); + Request r(m, p); + + try { + r.init(); + r.process(); + } catch (const AssertionException& ex) { + LOG(ex.isUserAssertion() ? 1 : 0) << "Assertion failed" + << " while processing " << opToString(m.operation()) + << " op" + << " for " << r.getns() << causedBy(ex); + + if (r.expectResponse()) { + m.header().setId(r.id()); + replyToQuery(ResultFlag_ErrSet, p, m, buildErrReply(ex)); } - catch ( const DBException& ex ) { - log() << "Exception thrown" - << " while processing " << opToString( m.operation() ) << " op" - << " for " << r.getns() << causedBy( ex ); + // We *always* populate the last error for now + LastError::get(cc()).setLastError(ex.getCode(), ex.what()); + } catch (const DBException& ex) { + log() << "Exception thrown" + << " while processing " << opToString(m.operation()) << " op" + << " for " << r.getns() << causedBy(ex); - if ( r.expectResponse() ) { - m.header().setId(r.id()); - replyToQuery( ResultFlag_ErrSet, p , m , buildErrReply( ex ) ); - } - - // We *always* populate the last error for now - LastError::get(cc()).setLastError(ex.getCode(), ex.what()); + if (r.expectResponse()) { + m.header().setId(r.id()); + replyToQuery(ResultFlag_ErrSet, p, m, buildErrReply(ex)); } - // Release connections back to pool, if any still cached - ShardConnection::releaseMyConnections(); + // We *always* populate the last error for now + LastError::get(cc()).setLastError(ex.getCode(), ex.what()); } - }; - void start( const MessageServer::Options& opts ) { - balancer.go(); - cursorCache.startTimeoutThread(); - UserCacheInvalidator cacheInvalidatorThread(getGlobalAuthorizationManager()); - cacheInvalidatorThread.go(); + // Release connections back to pool, if any still cached + ShardConnection::releaseMyConnections(); + } +}; + +void start(const MessageServer::Options& opts) { + balancer.go(); + cursorCache.startTimeoutThread(); + UserCacheInvalidator cacheInvalidatorThread(getGlobalAuthorizationManager()); + cacheInvalidatorThread.go(); - PeriodicTask::startRunningPeriodicTasks(); + PeriodicTask::startRunningPeriodicTasks(); - ShardedMessageHandler handler; - MessageServer * server = createServer( opts , &handler ); - server->setAsTimeTracker(); - server->setupSockets(); - server->run(); - } + ShardedMessageHandler handler; + MessageServer* server = createServer(opts, &handler); + server->setAsTimeTracker(); + server->setupSockets(); + server->run(); +} - DBClientBase* createDirectClient(OperationContext* txn) { - uassert( 10197 , "createDirectClient not implemented for sharding yet" , 0 ); - return 0; - } +DBClientBase* createDirectClient(OperationContext* txn) { + uassert(10197, "createDirectClient not implemented for sharding yet", 0); + return 0; +} -} // namespace mongo +} // namespace mongo using namespace mongo; -static ExitCode runMongosServer( bool doUpgrade ) { - setThreadName( "mongosMain" ); - printShardingVersionInfo( false ); +static ExitCode runMongosServer(bool doUpgrade) { + setThreadName("mongosMain"); + printShardingVersionInfo(false); // Add sharding hooks to both connection pools - ShardingConnectionHook includes auth hooks globalConnPool.addHook(new ShardingConnectionHook(false)); shardConnectionPool.addHook(new ShardingConnectionHook(true)); // Mongos shouldn't lazily kill cursors, otherwise we can end up with extras from migration - DBClientConnection::setLazyKillCursor( false ); + DBClientConnection::setLazyKillCursor(false); ReplicaSetMonitor::setConfigChangeHook(&ConfigServer::replicaSetChange); @@ -232,11 +226,11 @@ static ExitCode runMongosServer( bool doUpgrade ) { } } - auto shardRegistry = stdx::make_unique<ShardRegistry>( - stdx::make_unique<RemoteCommandTargeterFactoryImpl>(), - stdx::make_unique<RemoteCommandRunnerImpl>(0), - std::unique_ptr<executor::TaskExecutor>{nullptr}, - catalogManager.get()); + auto shardRegistry = + stdx::make_unique<ShardRegistry>(stdx::make_unique<RemoteCommandTargeterFactoryImpl>(), + stdx::make_unique<RemoteCommandRunnerImpl>(0), + std::unique_ptr<executor::TaskExecutor>{nullptr}, + catalogManager.get()); grid.init(std::move(catalogManager), std::move(shardRegistry)); @@ -259,10 +253,8 @@ static ExitCode runMongosServer( bool doUpgrade ) { #endif if (serverGlobalParams.isHttpInterfaceEnabled) { - std::shared_ptr<DbWebServer> dbWebServer( - new DbWebServer(serverGlobalParams.bind_ip, - serverGlobalParams.port + 1000, - new NoAdminAccess())); + std::shared_ptr<DbWebServer> dbWebServer(new DbWebServer( + serverGlobalParams.bind_ip, serverGlobalParams.port + 1000, new NoAdminAccess())); dbWebServer->setupSockets(); stdx::thread web(stdx::bind(&webServerListenThread, dbWebServer)); @@ -298,12 +290,9 @@ MONGO_INITIALIZER_GENERAL(ForkServer, static void startupConfigActions(const std::vector<std::string>& argv) { #if defined(_WIN32) vector<string> disallowedOptions; - disallowedOptions.push_back( "upgrade" ); - ntservice::configureService(initService, - moe::startupOptionsParsed, - defaultServiceStrings, - disallowedOptions, - argv); + disallowedOptions.push_back("upgrade"); + ntservice::configureService( + initService, moe::startupOptionsParsed, defaultServiceStrings, disallowedOptions, argv); #endif } @@ -316,15 +305,15 @@ static int _main() { // we either have a setting where all processes are in localhost or none are std::vector<HostAndPort> configServers = mongosGlobalParams.configdbs.getServers(); for (std::vector<HostAndPort>::const_iterator it = configServers.begin(); - it != configServers.end(); ++it) { - + it != configServers.end(); + ++it) { const HostAndPort& configAddr = *it; if (it == configServers.begin()) { - grid.setAllowLocalHost( configAddr.isLocalHost() ); + grid.setAllowLocalHost(configAddr.isLocalHost()); } - if ( configAddr.isLocalHost() != grid.allowLocalHost() ) { + if (configAddr.isLocalHost() != grid.allowLocalHost()) { mongo::log(LogComponent::kDefault) << "cannot mix localhost and ip addresses in configdbs"; return 10; @@ -343,7 +332,7 @@ static int _main() { // To maintain backwards compatibility, we exit with EXIT_NET_ERROR if the listener loop returns. if (exitCode == EXIT_NET_ERROR) { - dbexit( EXIT_NET_ERROR ); + dbexit(EXIT_NET_ERROR); } return (exitCode == EXIT_CLEAN) ? 0 : 1; @@ -351,43 +340,43 @@ static int _main() { #if defined(_WIN32) namespace mongo { - static ExitCode initService() { - ntservice::reportStatus( SERVICE_RUNNING ); - log() << "Service running"; +static ExitCode initService() { + ntservice::reportStatus(SERVICE_RUNNING); + log() << "Service running"; - ExitCode exitCode = runMongosServer(mongosGlobalParams.upgrade); + ExitCode exitCode = runMongosServer(mongosGlobalParams.upgrade); - // ignore EXIT_NET_ERROR on clean shutdown since we return this when the listening socket - // is closed - return (exitCode == EXIT_NET_ERROR && inShutdown()) ? EXIT_CLEAN : exitCode; - } + // ignore EXIT_NET_ERROR on clean shutdown since we return this when the listening socket + // is closed + return (exitCode == EXIT_NET_ERROR && inShutdown()) ? EXIT_CLEAN : exitCode; +} } // namespace mongo #endif namespace { - std::unique_ptr<AuthzManagerExternalState> createAuthzManagerExternalStateMongos() { - return stdx::make_unique<AuthzManagerExternalStateMongos>(); - } +std::unique_ptr<AuthzManagerExternalState> createAuthzManagerExternalStateMongos() { + return stdx::make_unique<AuthzManagerExternalStateMongos>(); +} - MONGO_INITIALIZER(CreateAuthorizationExternalStateFactory) (InitializerContext* context) { - AuthzManagerExternalState::create = &createAuthzManagerExternalStateMongos; - return Status::OK(); - } +MONGO_INITIALIZER(CreateAuthorizationExternalStateFactory)(InitializerContext* context) { + AuthzManagerExternalState::create = &createAuthzManagerExternalStateMongos; + return Status::OK(); +} - MONGO_INITIALIZER(SetGlobalEnvironment)(InitializerContext* context) { - setGlobalServiceContext(stdx::make_unique<ServiceContextNoop>()); - return Status::OK(); - } +MONGO_INITIALIZER(SetGlobalEnvironment)(InitializerContext* context) { + setGlobalServiceContext(stdx::make_unique<ServiceContextNoop>()); + return Status::OK(); +} #ifdef MONGO_CONFIG_SSL - MONGO_INITIALIZER_GENERAL(setSSLManagerType, - MONGO_NO_PREREQUISITES, - ("SSLManager"))(InitializerContext* context) { - isSSLServer = true; - return Status::OK(); - } +MONGO_INITIALIZER_GENERAL(setSSLManagerType, + MONGO_NO_PREREQUISITES, + ("SSLManager"))(InitializerContext* context) { + isSSLServer = true; + return Status::OK(); +} #endif -} // namespace +} // namespace int mongoSMain(int argc, char* argv[], char** envp) { static StaticObserver staticObserver; @@ -410,17 +399,13 @@ int mongoSMain(int argc, char* argv[], char** envp) { try { int exitCode = _main(); return exitCode; - } - catch(const SocketException& e) { + } catch (const SocketException& e) { error() << "uncaught SocketException in mongos main: " << e.toString(); - } - catch (const DBException& e) { + } catch (const DBException& e) { error() << "uncaught DBException in mongos main: " << e.toString(); - } - catch (const std::exception& e) { + } catch (const std::exception& e) { error() << "uncaught std::exception in mongos main:" << e.what(); - } - catch (...) { + } catch (...) { error() << "uncaught unknown exception in mongos main"; } @@ -453,10 +438,10 @@ void mongo::signalShutdown() { void mongo::exitCleanly(ExitCode code) { // TODO: do we need to add anything? grid.catalogManager()->shutDown(); - mongo::dbexit( code ); + mongo::dbexit(code); } -void mongo::dbexit(ExitCode rc, const char *why) { +void mongo::dbexit(ExitCode rc, const char* why) { dbexitCalled = true; audit::logShutdown(ClientBasic::getCurrent()); diff --git a/src/mongo/s/server.h b/src/mongo/s/server.h index 3594cb67b59..1b76ac71a15 100644 --- a/src/mongo/s/server.h +++ b/src/mongo/s/server.h @@ -35,8 +35,8 @@ namespace mongo { - extern OID serverID; +extern OID serverID; - // from request.cpp - void processRequest(Message& m, MessagingPort& p); +// from request.cpp +void processRequest(Message& m, MessagingPort& p); } diff --git a/src/mongo/s/shard_key_pattern.cpp b/src/mongo/s/shard_key_pattern.cpp index f48088e9837..f411461b925 100644 --- a/src/mongo/s/shard_key_pattern.cpp +++ b/src/mongo/s/shard_key_pattern.cpp @@ -40,400 +40,379 @@ namespace mongo { - using std::make_pair; - using std::pair; - using std::shared_ptr; - using std::string; - using std::unique_ptr; - using std::vector; - - using pathsupport::EqualityMatches; - using mongoutils::str::stream; - - const int ShardKeyPattern::kMaxShardKeySizeBytes = 512; - const unsigned int ShardKeyPattern::kMaxFlattenedInCombinations = 4000000; - - Status ShardKeyPattern::checkShardKeySize(const BSONObj& shardKey) { - if (shardKey.objsize() <= kMaxShardKeySizeBytes) - return Status::OK(); - - return Status(ErrorCodes::ShardKeyTooBig, - stream() << "shard keys must be less than " << kMaxShardKeySizeBytes - << " bytes, but key " << shardKey << " is " << shardKey.objsize() - << " bytes"); - } - - static bool isHashedPatternEl(const BSONElement& el) { - return el.type() == String && el.String() == IndexNames::HASHED; - } - - /** - * Currently the allowable shard keys are either - * i) a hashed single field, e.g. { a : "hashed" }, or - * ii) a compound list of ascending, potentially-nested field paths, e.g. { a : 1 , b.c : 1 } - */ - static vector<FieldRef*> parseShardKeyPattern(const BSONObj& keyPattern) { - - OwnedPointerVector<FieldRef> parsedPaths; - static const vector<FieldRef*> empty; - - BSONObjIterator patternIt(keyPattern); - while (patternIt.more()) { - - BSONElement patternEl = patternIt.next(); - parsedPaths.push_back(new FieldRef(patternEl.fieldNameStringData())); - const FieldRef& patternPath = *parsedPaths.back(); +using std::make_pair; +using std::pair; +using std::shared_ptr; +using std::string; +using std::unique_ptr; +using std::vector; + +using pathsupport::EqualityMatches; +using mongoutils::str::stream; + +const int ShardKeyPattern::kMaxShardKeySizeBytes = 512; +const unsigned int ShardKeyPattern::kMaxFlattenedInCombinations = 4000000; + +Status ShardKeyPattern::checkShardKeySize(const BSONObj& shardKey) { + if (shardKey.objsize() <= kMaxShardKeySizeBytes) + return Status::OK(); + + return Status(ErrorCodes::ShardKeyTooBig, + stream() << "shard keys must be less than " << kMaxShardKeySizeBytes + << " bytes, but key " << shardKey << " is " << shardKey.objsize() + << " bytes"); +} - // Empty path - if (patternPath.numParts() == 0) - return empty; +static bool isHashedPatternEl(const BSONElement& el) { + return el.type() == String && el.String() == IndexNames::HASHED; +} - // Extra "." in path? - if (patternPath.dottedField() != patternEl.fieldNameStringData()) +/** + * Currently the allowable shard keys are either + * i) a hashed single field, e.g. { a : "hashed" }, or + * ii) a compound list of ascending, potentially-nested field paths, e.g. { a : 1 , b.c : 1 } + */ +static vector<FieldRef*> parseShardKeyPattern(const BSONObj& keyPattern) { + OwnedPointerVector<FieldRef> parsedPaths; + static const vector<FieldRef*> empty; + + BSONObjIterator patternIt(keyPattern); + while (patternIt.more()) { + BSONElement patternEl = patternIt.next(); + parsedPaths.push_back(new FieldRef(patternEl.fieldNameStringData())); + const FieldRef& patternPath = *parsedPaths.back(); + + // Empty path + if (patternPath.numParts() == 0) + return empty; + + // Extra "." in path? + if (patternPath.dottedField() != patternEl.fieldNameStringData()) + return empty; + + // Empty parts of the path, ".."? + for (size_t i = 0; i < patternPath.numParts(); ++i) { + if (patternPath.getPart(i).size() == 0) return empty; + } - // Empty parts of the path, ".."? - for (size_t i = 0; i < patternPath.numParts(); ++i) { - if (patternPath.getPart(i).size() == 0) - return empty; - } - - // Numeric and ascending (1.0), or "hashed" and single field - if (!patternEl.isNumber()) { - if (keyPattern.nFields() != 1 || !isHashedPatternEl(patternEl)) - return empty; - } - else if (patternEl.numberInt() != 1) { + // Numeric and ascending (1.0), or "hashed" and single field + if (!patternEl.isNumber()) { + if (keyPattern.nFields() != 1 || !isHashedPatternEl(patternEl)) return empty; - } + } else if (patternEl.numberInt() != 1) { + return empty; } - - return parsedPaths.release(); } - ShardKeyPattern::ShardKeyPattern(const BSONObj& keyPattern) - : _keyPatternPaths(parseShardKeyPattern(keyPattern)), - _keyPattern(_keyPatternPaths.empty() ? BSONObj() : keyPattern) { - } - - ShardKeyPattern::ShardKeyPattern(const KeyPattern& keyPattern) - : _keyPatternPaths(parseShardKeyPattern(keyPattern.toBSON())), - _keyPattern(_keyPatternPaths.empty() ? KeyPattern(BSONObj()) : keyPattern) { - } - - bool ShardKeyPattern::isValid() const { - return !_keyPattern.toBSON().isEmpty(); - } + return parsedPaths.release(); +} - bool ShardKeyPattern::isHashedPattern() const { - return isHashedPatternEl(_keyPattern.toBSON().firstElement()); - } +ShardKeyPattern::ShardKeyPattern(const BSONObj& keyPattern) + : _keyPatternPaths(parseShardKeyPattern(keyPattern)), + _keyPattern(_keyPatternPaths.empty() ? BSONObj() : keyPattern) {} - const KeyPattern& ShardKeyPattern::getKeyPattern() const { - return _keyPattern; - } +ShardKeyPattern::ShardKeyPattern(const KeyPattern& keyPattern) + : _keyPatternPaths(parseShardKeyPattern(keyPattern.toBSON())), + _keyPattern(_keyPatternPaths.empty() ? KeyPattern(BSONObj()) : keyPattern) {} - const BSONObj& ShardKeyPattern::toBSON() const { - return _keyPattern.toBSON(); - } +bool ShardKeyPattern::isValid() const { + return !_keyPattern.toBSON().isEmpty(); +} - string ShardKeyPattern::toString() const { - return toBSON().toString(); - } +bool ShardKeyPattern::isHashedPattern() const { + return isHashedPatternEl(_keyPattern.toBSON().firstElement()); +} - static bool isShardKeyElement(const BSONElement& element, bool allowRegex) { - // TODO: Disallow regex all the time - if (element.eoo() || element.type() == Array || (!allowRegex && element.type() == RegEx) - || (element.type() == Object && !element.embeddedObject().okForStorage())) - return false; - return true; - } +const KeyPattern& ShardKeyPattern::getKeyPattern() const { + return _keyPattern; +} - bool ShardKeyPattern::isShardKey(const BSONObj& shardKey) const { +const BSONObj& ShardKeyPattern::toBSON() const { + return _keyPattern.toBSON(); +} - // Shard keys are always of the form: { 'nested.path' : value, 'nested.path2' : value } +string ShardKeyPattern::toString() const { + return toBSON().toString(); +} - if (!isValid()) - return false; +static bool isShardKeyElement(const BSONElement& element, bool allowRegex) { + // TODO: Disallow regex all the time + if (element.eoo() || element.type() == Array || (!allowRegex && element.type() == RegEx) || + (element.type() == Object && !element.embeddedObject().okForStorage())) + return false; + return true; +} - BSONObjIterator patternIt(_keyPattern.toBSON()); - while (patternIt.more()) { +bool ShardKeyPattern::isShardKey(const BSONObj& shardKey) const { + // Shard keys are always of the form: { 'nested.path' : value, 'nested.path2' : value } - BSONElement patternEl = patternIt.next(); + if (!isValid()) + return false; - BSONElement keyEl = shardKey[patternEl.fieldNameStringData()]; - if (!isShardKeyElement(keyEl, true)) - return false; - } + BSONObjIterator patternIt(_keyPattern.toBSON()); + while (patternIt.more()) { + BSONElement patternEl = patternIt.next(); - return true; + BSONElement keyEl = shardKey[patternEl.fieldNameStringData()]; + if (!isShardKeyElement(keyEl, true)) + return false; } - BSONObj ShardKeyPattern::normalizeShardKey(const BSONObj& shardKey) const { - - // Shard keys are always of the form: { 'nested.path' : value, 'nested.path2' : value } - // and in the same order as the key pattern - - if (!isValid()) - return BSONObj(); + return true; +} - // We want to return an empty key if users pass us something that's not a shard key - if (shardKey.nFields() > _keyPattern.toBSON().nFields()) - return BSONObj(); +BSONObj ShardKeyPattern::normalizeShardKey(const BSONObj& shardKey) const { + // Shard keys are always of the form: { 'nested.path' : value, 'nested.path2' : value } + // and in the same order as the key pattern - BSONObjBuilder keyBuilder; - BSONObjIterator patternIt(_keyPattern.toBSON()); - while (patternIt.more()) { + if (!isValid()) + return BSONObj(); - BSONElement patternEl = patternIt.next(); + // We want to return an empty key if users pass us something that's not a shard key + if (shardKey.nFields() > _keyPattern.toBSON().nFields()) + return BSONObj(); - BSONElement keyEl = shardKey[patternEl.fieldNameStringData()]; + BSONObjBuilder keyBuilder; + BSONObjIterator patternIt(_keyPattern.toBSON()); + while (patternIt.more()) { + BSONElement patternEl = patternIt.next(); - if (!isShardKeyElement(keyEl, true)) - return BSONObj(); + BSONElement keyEl = shardKey[patternEl.fieldNameStringData()]; - keyBuilder.appendAs(keyEl, patternEl.fieldName()); - } + if (!isShardKeyElement(keyEl, true)) + return BSONObj(); - dassert(isShardKey(keyBuilder.asTempObj())); - return keyBuilder.obj(); + keyBuilder.appendAs(keyEl, patternEl.fieldName()); } - static BSONElement extractKeyElementFromMatchable(const MatchableDocument& matchable, - StringData pathStr) { - ElementPath path; - path.init(pathStr); - path.setTraverseNonleafArrays(false); - path.setTraverseLeafArray(false); - - MatchableDocument::IteratorHolder matchIt(&matchable, &path); - if (!matchIt->more()) - return BSONElement(); + dassert(isShardKey(keyBuilder.asTempObj())); + return keyBuilder.obj(); +} - BSONElement matchEl = matchIt->next().element(); - // We shouldn't have more than one element - we don't expand arrays - dassert(!matchIt->more()); +static BSONElement extractKeyElementFromMatchable(const MatchableDocument& matchable, + StringData pathStr) { + ElementPath path; + path.init(pathStr); + path.setTraverseNonleafArrays(false); + path.setTraverseLeafArray(false); - return matchEl; - } + MatchableDocument::IteratorHolder matchIt(&matchable, &path); + if (!matchIt->more()) + return BSONElement(); - BSONObj // - ShardKeyPattern::extractShardKeyFromMatchable(const MatchableDocument& matchable) const { + BSONElement matchEl = matchIt->next().element(); + // We shouldn't have more than one element - we don't expand arrays + dassert(!matchIt->more()); - if (!isValid()) - return BSONObj(); + return matchEl; +} - BSONObjBuilder keyBuilder; +BSONObj // + ShardKeyPattern::extractShardKeyFromMatchable(const MatchableDocument& matchable) const { + if (!isValid()) + return BSONObj(); - BSONObjIterator patternIt(_keyPattern.toBSON()); - while (patternIt.more()) { + BSONObjBuilder keyBuilder; - BSONElement patternEl = patternIt.next(); - BSONElement matchEl = extractKeyElementFromMatchable(matchable, - patternEl.fieldNameStringData()); + BSONObjIterator patternIt(_keyPattern.toBSON()); + while (patternIt.more()) { + BSONElement patternEl = patternIt.next(); + BSONElement matchEl = + extractKeyElementFromMatchable(matchable, patternEl.fieldNameStringData()); - if (!isShardKeyElement(matchEl, true)) - return BSONObj(); + if (!isShardKeyElement(matchEl, true)) + return BSONObj(); - if (isHashedPatternEl(patternEl)) { - keyBuilder.append(patternEl.fieldName(), - BSONElementHasher::hash64(matchEl, - BSONElementHasher::DEFAULT_HASH_SEED)); - } - else { - // NOTE: The matched element may *not* have the same field name as the path - - // index keys don't contain field names, for example - keyBuilder.appendAs(matchEl, patternEl.fieldName()); - } + if (isHashedPatternEl(patternEl)) { + keyBuilder.append( + patternEl.fieldName(), + BSONElementHasher::hash64(matchEl, BSONElementHasher::DEFAULT_HASH_SEED)); + } else { + // NOTE: The matched element may *not* have the same field name as the path - + // index keys don't contain field names, for example + keyBuilder.appendAs(matchEl, patternEl.fieldName()); } - - dassert(isShardKey(keyBuilder.asTempObj())); - return keyBuilder.obj(); - } - - BSONObj ShardKeyPattern::extractShardKeyFromDoc(const BSONObj& doc) const { - BSONMatchableDocument matchable(doc); - return extractShardKeyFromMatchable(matchable); } - static BSONElement findEqualityElement(const EqualityMatches& equalities, - const FieldRef& path) { + dassert(isShardKey(keyBuilder.asTempObj())); + return keyBuilder.obj(); +} - int parentPathPart; - const BSONElement& parentEl = pathsupport::findParentEqualityElement(equalities, - path, - &parentPathPart); +BSONObj ShardKeyPattern::extractShardKeyFromDoc(const BSONObj& doc) const { + BSONMatchableDocument matchable(doc); + return extractShardKeyFromMatchable(matchable); +} - if (parentPathPart == static_cast<int>(path.numParts())) - return parentEl; +static BSONElement findEqualityElement(const EqualityMatches& equalities, const FieldRef& path) { + int parentPathPart; + const BSONElement& parentEl = + pathsupport::findParentEqualityElement(equalities, path, &parentPathPart); - if (parentEl.type() != Object) - return BSONElement(); + if (parentPathPart == static_cast<int>(path.numParts())) + return parentEl; - StringData suffixStr = path.dottedSubstring(parentPathPart, path.numParts()); - BSONMatchableDocument matchable(parentEl.Obj()); - return extractKeyElementFromMatchable(matchable, suffixStr); - } + if (parentEl.type() != Object) + return BSONElement(); - StatusWith<BSONObj> ShardKeyPattern::extractShardKeyFromQuery(const BSONObj& basicQuery) const { - - if (!isValid()) - return StatusWith<BSONObj>(BSONObj()); + StringData suffixStr = path.dottedSubstring(parentPathPart, path.numParts()); + BSONMatchableDocument matchable(parentEl.Obj()); + return extractKeyElementFromMatchable(matchable, suffixStr); +} - // Extract equalities from query - CanonicalQuery* rawQuery; - Status queryStatus = - CanonicalQuery::canonicalize("", basicQuery, &rawQuery, WhereCallbackNoop()); - if (!queryStatus.isOK()) - return StatusWith<BSONObj>(queryStatus); - unique_ptr<CanonicalQuery> query(rawQuery); - - EqualityMatches equalities; - // TODO: Build the path set initially? - FieldRefSet keyPatternPathSet(_keyPatternPaths.vector()); - // We only care about extracting the full key pattern paths - if they don't exist (or are - // conflicting), we don't contain the shard key. - Status eqStatus = pathsupport::extractFullEqualityMatches(*query->root(), - keyPatternPathSet, - &equalities); - // NOTE: Failure to extract equality matches just means we return no shard key - it's not - // an error we propagate - if (!eqStatus.isOK()) +StatusWith<BSONObj> ShardKeyPattern::extractShardKeyFromQuery(const BSONObj& basicQuery) const { + if (!isValid()) + return StatusWith<BSONObj>(BSONObj()); + + // Extract equalities from query + CanonicalQuery* rawQuery; + Status queryStatus = + CanonicalQuery::canonicalize("", basicQuery, &rawQuery, WhereCallbackNoop()); + if (!queryStatus.isOK()) + return StatusWith<BSONObj>(queryStatus); + unique_ptr<CanonicalQuery> query(rawQuery); + + EqualityMatches equalities; + // TODO: Build the path set initially? + FieldRefSet keyPatternPathSet(_keyPatternPaths.vector()); + // We only care about extracting the full key pattern paths - if they don't exist (or are + // conflicting), we don't contain the shard key. + Status eqStatus = + pathsupport::extractFullEqualityMatches(*query->root(), keyPatternPathSet, &equalities); + // NOTE: Failure to extract equality matches just means we return no shard key - it's not + // an error we propagate + if (!eqStatus.isOK()) + return StatusWith<BSONObj>(BSONObj()); + + // Extract key from equalities + // NOTE: The method below is equivalent to constructing a BSONObj and running + // extractShardKeyFromMatchable, but doesn't require creating the doc. + + BSONObjBuilder keyBuilder; + // Iterate the parsed paths to avoid re-parsing + for (OwnedPointerVector<FieldRef>::const_iterator it = _keyPatternPaths.begin(); + it != _keyPatternPaths.end(); + ++it) { + const FieldRef& patternPath = **it; + BSONElement equalEl = findEqualityElement(equalities, patternPath); + + if (!isShardKeyElement(equalEl, false)) return StatusWith<BSONObj>(BSONObj()); - // Extract key from equalities - // NOTE: The method below is equivalent to constructing a BSONObj and running - // extractShardKeyFromMatchable, but doesn't require creating the doc. - - BSONObjBuilder keyBuilder; - // Iterate the parsed paths to avoid re-parsing - for (OwnedPointerVector<FieldRef>::const_iterator it = _keyPatternPaths.begin(); - it != _keyPatternPaths.end(); ++it) { - - const FieldRef& patternPath = **it; - BSONElement equalEl = findEqualityElement(equalities, patternPath); - - if (!isShardKeyElement(equalEl, false)) - return StatusWith<BSONObj>(BSONObj()); - - if (isHashedPattern()) { - keyBuilder.append(patternPath.dottedField(), - BSONElementHasher::hash64(equalEl, - BSONElementHasher::DEFAULT_HASH_SEED)); - } - else { - // NOTE: The equal element may *not* have the same field name as the path - - // nested $and, $eq, for example - keyBuilder.appendAs(equalEl, patternPath.dottedField()); - } + if (isHashedPattern()) { + keyBuilder.append( + patternPath.dottedField(), + BSONElementHasher::hash64(equalEl, BSONElementHasher::DEFAULT_HASH_SEED)); + } else { + // NOTE: The equal element may *not* have the same field name as the path - + // nested $and, $eq, for example + keyBuilder.appendAs(equalEl, patternPath.dottedField()); } - - dassert(isShardKey(keyBuilder.asTempObj())); - return StatusWith<BSONObj>(keyBuilder.obj()); } - bool ShardKeyPattern::isUniqueIndexCompatible(const BSONObj& uniqueIndexPattern) const { - - dassert(!KeyPattern::isHashedKeyPattern(uniqueIndexPattern)); + dassert(isShardKey(keyBuilder.asTempObj())); + return StatusWith<BSONObj>(keyBuilder.obj()); +} - if (!uniqueIndexPattern.isEmpty() - && string("_id") == uniqueIndexPattern.firstElementFieldName()) { - return true; - } +bool ShardKeyPattern::isUniqueIndexCompatible(const BSONObj& uniqueIndexPattern) const { + dassert(!KeyPattern::isHashedKeyPattern(uniqueIndexPattern)); - return _keyPattern.toBSON().isFieldNamePrefixOf(uniqueIndexPattern); + if (!uniqueIndexPattern.isEmpty() && + string("_id") == uniqueIndexPattern.firstElementFieldName()) { + return true; } - BoundList ShardKeyPattern::flattenBounds(const IndexBounds& indexBounds) const { + return _keyPattern.toBSON().isFieldNamePrefixOf(uniqueIndexPattern); +} - invariant(indexBounds.fields.size() == (size_t)_keyPattern.toBSON().nFields()); +BoundList ShardKeyPattern::flattenBounds(const IndexBounds& indexBounds) const { + invariant(indexBounds.fields.size() == (size_t)_keyPattern.toBSON().nFields()); - // If any field is unsatisfied, return empty bound list. - for (vector<OrderedIntervalList>::const_iterator it = indexBounds.fields.begin(); - it != indexBounds.fields.end(); it++) { - if (it->intervals.size() == 0) { - return BoundList(); - } + // If any field is unsatisfied, return empty bound list. + for (vector<OrderedIntervalList>::const_iterator it = indexBounds.fields.begin(); + it != indexBounds.fields.end(); + it++) { + if (it->intervals.size() == 0) { + return BoundList(); } - // To construct our bounds we will generate intervals based on bounds for - // the first field, then compound intervals based on constraints for the first - // 2 fields, then compound intervals for the first 3 fields, etc. - // As we loop through the fields, we start generating new intervals that will later - // get extended in another iteration of the loop. We define these partially constructed - // intervals using pairs of BSONObjBuilders (shared_ptrs, since after one iteration of the - // loop they still must exist outside their scope). - typedef vector<pair<shared_ptr<BSONObjBuilder>, shared_ptr<BSONObjBuilder> > > BoundBuilders; - - BoundBuilders builders; - builders.push_back(make_pair(shared_ptr<BSONObjBuilder>(new BSONObjBuilder()), - shared_ptr<BSONObjBuilder>(new BSONObjBuilder()))); - BSONObjIterator keyIter(_keyPattern.toBSON()); - // until equalityOnly is false, we are just dealing with equality (no range or $in queries). - bool equalityOnly = true; - - for (size_t i = 0; i < indexBounds.fields.size(); i++) { - BSONElement e = keyIter.next(); - - StringData fieldName = e.fieldNameStringData(); - - // get the relevant intervals for this field, but we may have to transform the - // list of what's relevant according to the expression for this field - const OrderedIntervalList& oil = indexBounds.fields[i]; - const vector<Interval>& intervals = oil.intervals; - - if (equalityOnly) { - if (intervals.size() == 1 && intervals.front().isPoint()) { - // this field is only a single point-interval - BoundBuilders::const_iterator j; - for (j = builders.begin(); j != builders.end(); ++j) { - j->first->appendAs(intervals.front().start, fieldName); - j->second->appendAs(intervals.front().end, fieldName); - } + } + // To construct our bounds we will generate intervals based on bounds for + // the first field, then compound intervals based on constraints for the first + // 2 fields, then compound intervals for the first 3 fields, etc. + // As we loop through the fields, we start generating new intervals that will later + // get extended in another iteration of the loop. We define these partially constructed + // intervals using pairs of BSONObjBuilders (shared_ptrs, since after one iteration of the + // loop they still must exist outside their scope). + typedef vector<pair<shared_ptr<BSONObjBuilder>, shared_ptr<BSONObjBuilder>>> BoundBuilders; + + BoundBuilders builders; + builders.push_back(make_pair(shared_ptr<BSONObjBuilder>(new BSONObjBuilder()), + shared_ptr<BSONObjBuilder>(new BSONObjBuilder()))); + BSONObjIterator keyIter(_keyPattern.toBSON()); + // until equalityOnly is false, we are just dealing with equality (no range or $in queries). + bool equalityOnly = true; + + for (size_t i = 0; i < indexBounds.fields.size(); i++) { + BSONElement e = keyIter.next(); + + StringData fieldName = e.fieldNameStringData(); + + // get the relevant intervals for this field, but we may have to transform the + // list of what's relevant according to the expression for this field + const OrderedIntervalList& oil = indexBounds.fields[i]; + const vector<Interval>& intervals = oil.intervals; + + if (equalityOnly) { + if (intervals.size() == 1 && intervals.front().isPoint()) { + // this field is only a single point-interval + BoundBuilders::const_iterator j; + for (j = builders.begin(); j != builders.end(); ++j) { + j->first->appendAs(intervals.front().start, fieldName); + j->second->appendAs(intervals.front().end, fieldName); } - else { - // This clause is the first to generate more than a single point. - // We only execute this clause once. After that, we simplify the bound - // extensions to prevent combinatorial explosion. - equalityOnly = false; - - BoundBuilders newBuilders; - - for (BoundBuilders::const_iterator it = builders.begin(); it != builders.end(); - ++it) { - BSONObj first = it->first->obj(); - BSONObj second = it->second->obj(); - - for (vector<Interval>::const_iterator interval = intervals.begin(); - interval != intervals.end(); ++interval) { - uassert( 17439, - "combinatorial limit of $in partitioning of results exceeded" , - newBuilders.size() < kMaxFlattenedInCombinations ); - newBuilders.push_back( // + } else { + // This clause is the first to generate more than a single point. + // We only execute this clause once. After that, we simplify the bound + // extensions to prevent combinatorial explosion. + equalityOnly = false; + + BoundBuilders newBuilders; + + for (BoundBuilders::const_iterator it = builders.begin(); it != builders.end(); + ++it) { + BSONObj first = it->first->obj(); + BSONObj second = it->second->obj(); + + for (vector<Interval>::const_iterator interval = intervals.begin(); + interval != intervals.end(); + ++interval) { + uassert(17439, + "combinatorial limit of $in partitioning of results exceeded", + newBuilders.size() < kMaxFlattenedInCombinations); + newBuilders.push_back( // make_pair(shared_ptr<BSONObjBuilder>(new BSONObjBuilder()), shared_ptr<BSONObjBuilder>(new BSONObjBuilder()))); - newBuilders.back().first->appendElements(first); - newBuilders.back().second->appendElements(second); - newBuilders.back().first->appendAs(interval->start, fieldName); - newBuilders.back().second->appendAs(interval->end, fieldName); - } + newBuilders.back().first->appendElements(first); + newBuilders.back().second->appendElements(second); + newBuilders.back().first->appendAs(interval->start, fieldName); + newBuilders.back().second->appendAs(interval->end, fieldName); } - builders = newBuilders; } + builders = newBuilders; } - else { - // if we've already generated a range or multiple point-intervals - // just extend what we've generated with min/max bounds for this field - BoundBuilders::const_iterator j; - for (j = builders.begin(); j != builders.end(); ++j) { - j->first->appendAs(intervals.front().start, fieldName); - j->second->appendAs(intervals.back().end, fieldName); - } + } else { + // if we've already generated a range or multiple point-intervals + // just extend what we've generated with min/max bounds for this field + BoundBuilders::const_iterator j; + for (j = builders.begin(); j != builders.end(); ++j) { + j->first->appendAs(intervals.front().start, fieldName); + j->second->appendAs(intervals.back().end, fieldName); } } - BoundList ret; - for (BoundBuilders::const_iterator i = builders.begin(); i != builders.end(); ++i) - ret.push_back(make_pair(i->first->obj(), i->second->obj())); - return ret; } - + BoundList ret; + for (BoundBuilders::const_iterator i = builders.begin(); i != builders.end(); ++i) + ret.push_back(make_pair(i->first->obj(), i->second->obj())); + return ret; +} } diff --git a/src/mongo/s/shard_key_pattern.h b/src/mongo/s/shard_key_pattern.h index d6cc67f56df..de0c792c37e 100644 --- a/src/mongo/s/shard_key_pattern.h +++ b/src/mongo/s/shard_key_pattern.h @@ -38,192 +38,189 @@ namespace mongo { - class FieldRef; +class FieldRef; + +/** + * Helper struct when generating flattened bounds below + * + * A BoundList contains intervals specified by inclusive start + * and end bounds. The intervals should be nonoverlapping and occur in + * the specified direction of traversal. For example, given a simple index {i:1} + * and direction +1, one valid BoundList is: (1, 2); (4, 6). The same BoundList + * would be valid for index {i:-1} with direction -1. + */ +typedef std::vector<std::pair<BSONObj, BSONObj>> BoundList; + +/** + * A ShardKeyPattern represents the key pattern used to partition data in a collection between + * shards. Shard keys are extracted from documents, simple queries, or Matchable objects based + * on the paths within the key pattern. + * + * Shard key pattern paths may be nested, but are not traversable through arrays - this means + * a shard key pattern path always yields a single value. + */ +class ShardKeyPattern { +public: + // Maximum size of shard key + static const int kMaxShardKeySizeBytes; + + // Maximum number of intervals produced by $in queries. + static const unsigned int kMaxFlattenedInCombinations; + + /** + * Helper to check shard key size and generate an appropriate error message. + */ + static Status checkShardKeySize(const BSONObj& shardKey); + + /** + * Constructs a shard key pattern from a BSON pattern document. If the document is not a + * valid shard key pattern, !isValid() will be true and key extraction will fail. + */ + explicit ShardKeyPattern(const BSONObj& keyPattern); + + /** + * Constructs a shard key pattern from a key pattern, see above. + */ + explicit ShardKeyPattern(const KeyPattern& keyPattern); + + bool isValid() const; + + bool isHashedPattern() const; + + const KeyPattern& getKeyPattern() const; + + const BSONObj& toBSON() const; + + std::string toString() const; /** - * Helper struct when generating flattened bounds below + * Returns true if the provided document is a shard key - i.e. has the same fields as the + * shard key pattern and valid shard key values. + */ + bool isShardKey(const BSONObj& shardKey) const; + + /** + * Given a shard key, return it in normal form where the fields are in the same order as + * the shard key pattern fields. * - * A BoundList contains intervals specified by inclusive start - * and end bounds. The intervals should be nonoverlapping and occur in - * the specified direction of traversal. For example, given a simple index {i:1} - * and direction +1, one valid BoundList is: (1, 2); (4, 6). The same BoundList - * would be valid for index {i:-1} with direction -1. + * If the shard key is invalid, returns BSONObj() */ - typedef std::vector<std::pair<BSONObj, BSONObj> > BoundList; + BSONObj normalizeShardKey(const BSONObj& shardKey) const; /** - * A ShardKeyPattern represents the key pattern used to partition data in a collection between - * shards. Shard keys are extracted from documents, simple queries, or Matchable objects based - * on the paths within the key pattern. + * Given a MatchableDocument, extracts the shard key corresponding to the key pattern. + * For each path in the shard key pattern, extracts a value from the matchable document. + * + * Paths to shard key fields must not contain arrays at any level, and shard keys may not + * be array fields, undefined, or non-storable sub-documents. If the shard key pattern is + * a hashed key pattern, this method performs the hashing. + * + * If a shard key cannot be extracted, returns an empty BSONObj(). * - * Shard key pattern paths may be nested, but are not traversable through arrays - this means - * a shard key pattern path always yields a single value. + * Examples: + * If 'this' KeyPattern is { a : 1 } + * { a: "hi" , b : 4} --> returns { a : "hi" } + * { c : 4 , a : 2 } --> returns { a : 2 } + * { b : 2 } -> returns {} + * { a : [1,2] } -> returns {} + * If 'this' KeyPattern is { a : "hashed" } + * { a: 1 } --> returns { a : NumberLong("5902408780260971510") } + * If 'this' KeyPattern is { 'a.b' : 1 } + * { a : { b : "hi" } } --> returns { 'a.b' : "hi" } + * { a : [{ b : "hi" }] } --> returns {} */ - class ShardKeyPattern { - public: - - // Maximum size of shard key - static const int kMaxShardKeySizeBytes; - - // Maximum number of intervals produced by $in queries. - static const unsigned int kMaxFlattenedInCombinations; - - /** - * Helper to check shard key size and generate an appropriate error message. - */ - static Status checkShardKeySize(const BSONObj& shardKey); - - /** - * Constructs a shard key pattern from a BSON pattern document. If the document is not a - * valid shard key pattern, !isValid() will be true and key extraction will fail. - */ - explicit ShardKeyPattern(const BSONObj& keyPattern); - - /** - * Constructs a shard key pattern from a key pattern, see above. - */ - explicit ShardKeyPattern(const KeyPattern& keyPattern); - - bool isValid() const; - - bool isHashedPattern() const; - - const KeyPattern& getKeyPattern() const; - - const BSONObj& toBSON() const; - - std::string toString() const; - - /** - * Returns true if the provided document is a shard key - i.e. has the same fields as the - * shard key pattern and valid shard key values. - */ - bool isShardKey(const BSONObj& shardKey) const; - - /** - * Given a shard key, return it in normal form where the fields are in the same order as - * the shard key pattern fields. - * - * If the shard key is invalid, returns BSONObj() - */ - BSONObj normalizeShardKey(const BSONObj& shardKey) const; - - /** - * Given a MatchableDocument, extracts the shard key corresponding to the key pattern. - * For each path in the shard key pattern, extracts a value from the matchable document. - * - * Paths to shard key fields must not contain arrays at any level, and shard keys may not - * be array fields, undefined, or non-storable sub-documents. If the shard key pattern is - * a hashed key pattern, this method performs the hashing. - * - * If a shard key cannot be extracted, returns an empty BSONObj(). - * - * Examples: - * If 'this' KeyPattern is { a : 1 } - * { a: "hi" , b : 4} --> returns { a : "hi" } - * { c : 4 , a : 2 } --> returns { a : 2 } - * { b : 2 } -> returns {} - * { a : [1,2] } -> returns {} - * If 'this' KeyPattern is { a : "hashed" } - * { a: 1 } --> returns { a : NumberLong("5902408780260971510") } - * If 'this' KeyPattern is { 'a.b' : 1 } - * { a : { b : "hi" } } --> returns { 'a.b' : "hi" } - * { a : [{ b : "hi" }] } --> returns {} - */ - BSONObj extractShardKeyFromMatchable(const MatchableDocument& matchable) const; - - /** - * Given a document, extracts the shard key corresponding to the key pattern. - * See above. - */ - BSONObj extractShardKeyFromDoc(const BSONObj& doc) const; - - /** - * Given a simple BSON query, extracts the shard key corresponding to the key pattern - * from equality matches in the query. The query expression *must not* be a complex query - * with sorts or other attributes. - * - * Logically, the equalities in the BSON query can be serialized into a BSON document and - * then a shard key is extracted from this equality document. - * - * NOTE: BSON queries and BSON documents look similar but are different languages. Use the - * correct shard key extraction function. - * - * Returns !OK status if the query cannot be parsed. Returns an empty BSONObj() if there is - * no shard key found in the query equalities. - * - * Examples: - * If the key pattern is { a : 1 } - * { a : "hi", b : 4 } --> returns { a : "hi" } - * { a : { $eq : "hi" }, b : 4 } --> returns { a : "hi" } - * { $and : [{a : { $eq : "hi" }}, { b : 4 }] } --> returns { a : "hi" } - * If the key pattern is { 'a.b' : 1 } - * { a : { b : "hi" } } --> returns { 'a.b' : "hi" } - * { 'a.b' : "hi" } --> returns { 'a.b' : "hi" } - * { a : { b : { $eq : "hi" } } } --> returns {} because the query language treats this as - * a : { $eq : { b : ... } } - */ - StatusWith<BSONObj> extractShardKeyFromQuery(const BSONObj& basicQuery) const; - - /** - * Returns true if the shard key pattern can ensure that the unique index pattern is - * respected across all shards. - * - * Primarily this just checks whether the shard key pattern field names are equal to or a - * prefix of the unique index pattern field names. Since documents with the same fields in - * the shard key pattern are guaranteed to go to the same shard, and all documents must - * contain the full shard key, a unique index with a shard key pattern prefix can be sure - * when resolving duplicates that documents on other shards will have different shard keys, - * and so are not duplicates. - * - * Hashed shard key patterns are similar to ordinary patterns in that they guarantee similar - * shard keys go to the same shard. - * - * Examples: - * shard key {a : 1} is compatible with a unique index on {_id : 1} - * shard key {a : 1} is compatible with a unique index on {a : 1 , b : 1} - * shard key {a : 1} is compatible with a unique index on {a : -1 , b : 1 } - * shard key {a : "hashed"} is compatible with a unique index on {a : 1} - * shard key {a : 1} is not compatible with a unique index on {b : 1} - * shard key {a : "hashed" , b : 1 } is not compatible with unique index on { b : 1 } - * - * All unique index patterns starting with _id are assumed to be enforceable by the fact - * that _ids must be unique, and so all unique _id prefixed indexes are compatible with - * any shard key pattern. - * - * NOTE: We assume 'uniqueIndexPattern' is a valid unique index pattern - a pattern like - * { k : "hashed" } is not capable of being a unique index and is an invalid argument to - * this method. - */ - bool isUniqueIndexCompatible(const BSONObj& uniqueIndexPattern) const; - - /** - * Return an ordered list of bounds generated using this KeyPattern and the - * bounds from the IndexBounds. This function is used in sharding to - * determine where to route queries according to the shard key pattern. - * - * Examples: - * - * Key { a: 1 }, Bounds a: [0] => { a: 0 } -> { a: 0 } - * Key { a: 1 }, Bounds a: [2, 3) => { a: 2 } -> { a: 3 } // bound inclusion ignored. - * - * The bounds returned by this function may be a superset of those defined - * by the constraints. For instance, if this KeyPattern is {a : 1, b: 1} - * Bounds: { a : {$in : [1,2]} , b : {$in : [3,4,5]} } - * => {a : 1 , b : 3} -> {a : 1 , b : 5}, {a : 2 , b : 3} -> {a : 2 , b : 5} - * - * If the IndexBounds are not defined for all the fields in this keypattern, which - * means some fields are unsatisfied, an empty BoundList could return. - * - */ - BoundList flattenBounds(const IndexBounds& indexBounds) const; - - private: - - // Ordered, parsed paths - const OwnedPointerVector<FieldRef> _keyPatternPaths; - - const KeyPattern _keyPattern; - }; + BSONObj extractShardKeyFromMatchable(const MatchableDocument& matchable) const; + + /** + * Given a document, extracts the shard key corresponding to the key pattern. + * See above. + */ + BSONObj extractShardKeyFromDoc(const BSONObj& doc) const; + + /** + * Given a simple BSON query, extracts the shard key corresponding to the key pattern + * from equality matches in the query. The query expression *must not* be a complex query + * with sorts or other attributes. + * + * Logically, the equalities in the BSON query can be serialized into a BSON document and + * then a shard key is extracted from this equality document. + * + * NOTE: BSON queries and BSON documents look similar but are different languages. Use the + * correct shard key extraction function. + * + * Returns !OK status if the query cannot be parsed. Returns an empty BSONObj() if there is + * no shard key found in the query equalities. + * + * Examples: + * If the key pattern is { a : 1 } + * { a : "hi", b : 4 } --> returns { a : "hi" } + * { a : { $eq : "hi" }, b : 4 } --> returns { a : "hi" } + * { $and : [{a : { $eq : "hi" }}, { b : 4 }] } --> returns { a : "hi" } + * If the key pattern is { 'a.b' : 1 } + * { a : { b : "hi" } } --> returns { 'a.b' : "hi" } + * { 'a.b' : "hi" } --> returns { 'a.b' : "hi" } + * { a : { b : { $eq : "hi" } } } --> returns {} because the query language treats this as + * a : { $eq : { b : ... } } + */ + StatusWith<BSONObj> extractShardKeyFromQuery(const BSONObj& basicQuery) const; + + /** + * Returns true if the shard key pattern can ensure that the unique index pattern is + * respected across all shards. + * + * Primarily this just checks whether the shard key pattern field names are equal to or a + * prefix of the unique index pattern field names. Since documents with the same fields in + * the shard key pattern are guaranteed to go to the same shard, and all documents must + * contain the full shard key, a unique index with a shard key pattern prefix can be sure + * when resolving duplicates that documents on other shards will have different shard keys, + * and so are not duplicates. + * + * Hashed shard key patterns are similar to ordinary patterns in that they guarantee similar + * shard keys go to the same shard. + * + * Examples: + * shard key {a : 1} is compatible with a unique index on {_id : 1} + * shard key {a : 1} is compatible with a unique index on {a : 1 , b : 1} + * shard key {a : 1} is compatible with a unique index on {a : -1 , b : 1 } + * shard key {a : "hashed"} is compatible with a unique index on {a : 1} + * shard key {a : 1} is not compatible with a unique index on {b : 1} + * shard key {a : "hashed" , b : 1 } is not compatible with unique index on { b : 1 } + * + * All unique index patterns starting with _id are assumed to be enforceable by the fact + * that _ids must be unique, and so all unique _id prefixed indexes are compatible with + * any shard key pattern. + * + * NOTE: We assume 'uniqueIndexPattern' is a valid unique index pattern - a pattern like + * { k : "hashed" } is not capable of being a unique index and is an invalid argument to + * this method. + */ + bool isUniqueIndexCompatible(const BSONObj& uniqueIndexPattern) const; + + /** + * Return an ordered list of bounds generated using this KeyPattern and the + * bounds from the IndexBounds. This function is used in sharding to + * determine where to route queries according to the shard key pattern. + * + * Examples: + * + * Key { a: 1 }, Bounds a: [0] => { a: 0 } -> { a: 0 } + * Key { a: 1 }, Bounds a: [2, 3) => { a: 2 } -> { a: 3 } // bound inclusion ignored. + * + * The bounds returned by this function may be a superset of those defined + * by the constraints. For instance, if this KeyPattern is {a : 1, b: 1} + * Bounds: { a : {$in : [1,2]} , b : {$in : [3,4,5]} } + * => {a : 1 , b : 3} -> {a : 1 , b : 5}, {a : 2 , b : 3} -> {a : 2 , b : 5} + * + * If the IndexBounds are not defined for all the fields in this keypattern, which + * means some fields are unsatisfied, an empty BoundList could return. + * + */ + BoundList flattenBounds(const IndexBounds& indexBounds) const; + +private: + // Ordered, parsed paths + const OwnedPointerVector<FieldRef> _keyPatternPaths; + const KeyPattern _keyPattern; +}; } diff --git a/src/mongo/s/shard_key_pattern_test.cpp b/src/mongo/s/shard_key_pattern_test.cpp index c90e313ce26..57cb7bca521 100644 --- a/src/mongo/s/shard_key_pattern_test.cpp +++ b/src/mongo/s/shard_key_pattern_test.cpp @@ -33,457 +33,454 @@ namespace { - using std::string; +using std::string; + +using namespace mongo; + +TEST(ShardKeyPattern, ValidShardKeyPatternSingle) { + BSONObj empty; + ASSERT(!ShardKeyPattern(empty).isValid()); + + // + // Single field ShardKeyPatterns + // + + ASSERT(ShardKeyPattern(BSON("a" << 1)).isValid()); + ASSERT(ShardKeyPattern(BSON("a" << 1)).isValid()); + ASSERT(ShardKeyPattern(BSON("a" << 1.0f)).isValid()); + ASSERT(ShardKeyPattern(BSON("a" << (long long)1L)).isValid()); + + ASSERT(!ShardKeyPattern(BSON("a" << -1)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" << -1.0)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" + << "1")).isValid()); + + ASSERT(ShardKeyPattern(BSON("a" + << "hashed")).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" + << "hash")).isValid()); + ASSERT(!ShardKeyPattern(BSON("" << 1)).isValid()); + ASSERT(!ShardKeyPattern(BSON("." << 1)).isValid()); +} - using namespace mongo; +TEST(ShardKeyPattern, ValidShardKeyPatternComposite) { + // + // Composite ShardKeyPatterns + // - TEST(ShardKeyPattern, ValidShardKeyPatternSingle) { + ASSERT(ShardKeyPattern(BSON("a" << 1 << "b" << 1)).isValid()); + ASSERT(ShardKeyPattern(BSON("a" << 1.0f << "b" << 1.0)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" << 1 << "b" << -1)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" << 1 << "b" + << "1")).isValid()); - BSONObj empty; - ASSERT(!ShardKeyPattern(empty).isValid()); + ASSERT(ShardKeyPattern(BSON("a" << 1 << "b" << 1.0 << "c" << 1.0f)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" << 1 << "b." << 1.0)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" << 1 << "" << 1.0)).isValid()); +} - // - // Single field ShardKeyPatterns - // +TEST(ShardKeyPattern, ValidShardKeyPatternNested) { + // + // Nested ShardKeyPatterns + // - ASSERT(ShardKeyPattern(BSON("a" << 1)).isValid()); - ASSERT(ShardKeyPattern(BSON("a" << 1)).isValid()); - ASSERT(ShardKeyPattern(BSON("a" << 1.0f)).isValid()); - ASSERT(ShardKeyPattern(BSON("a" << (long long)1L)).isValid()); + ASSERT(ShardKeyPattern(BSON("a.b" << 1)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a.b" << -1)).isValid()); + ASSERT(ShardKeyPattern(BSON("a.b.c.d" << 1.0)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << -1)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << -1.0)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << "1")).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" << BSON("b" << 1))).isValid()); - ASSERT(ShardKeyPattern(BSON("a" << "hashed")).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << "hash")).isValid()); - ASSERT(!ShardKeyPattern(BSON("" << 1)).isValid()); - ASSERT(!ShardKeyPattern(BSON("." << 1)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a.b." << 1)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a.b.." << 1)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a..b" << 1)).isValid()); - } + ASSERT(ShardKeyPattern(BSON("a" << 1 << "c.d" << 1.0 << "e.f.g" << 1.0f)).isValid()); + ASSERT(ShardKeyPattern(BSON("a" << 1 << "a.b" << 1.0 << "a.b.c" << 1.0f)).isValid()); - TEST(ShardKeyPattern, ValidShardKeyPatternComposite) { + ASSERT(!ShardKeyPattern(BSON("a" << 1 << "a.b." << 1.0)).isValid()); + ASSERT(!ShardKeyPattern(BSON("a" << BSON("b" << 1) << "c.d" << 1.0)).isValid()); +} - // - // Composite ShardKeyPatterns - // +TEST(ShardKeyPattern, IsShardKey) { + ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0f)); - ASSERT(ShardKeyPattern(BSON("a" << 1 << "b" << 1)).isValid()); - ASSERT(ShardKeyPattern(BSON("a" << 1.0f << "b" << 1.0)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << 1 << "b" << -1)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << 1 << "b" << "1")).isValid()); + ASSERT(pattern.isShardKey(BSON("a.b" << 10 << "c" << 30))); + ASSERT(pattern.isShardKey(BSON("c" << 30 << "a.b" << 10))); - ASSERT(ShardKeyPattern(BSON("a" << 1 << "b" << 1.0 << "c" << 1.0f)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << 1 << "b." << 1.0)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << 1 << "" << 1.0)).isValid()); + ASSERT(!pattern.isShardKey(BSON("b" << 10))); + ASSERT(!pattern.isShardKey(BSON("a" << 10 << "c" << 30))); + ASSERT(!pattern.isShardKey(BSON("a" << BSON("b" << 10) << "c" << 30))); +} - } +static BSONObj normKey(const ShardKeyPattern& pattern, const BSONObj& doc) { + return pattern.normalizeShardKey(doc); +} + +TEST(ShardKeyPattern, NormalizeShardKey) { + ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0f)); + + ASSERT_EQUALS(normKey(pattern, BSON("a.b" << 10 << "c" << 30)), BSON("a.b" << 10 << "c" << 30)); + ASSERT_EQUALS(normKey(pattern, BSON("c" << 30 << "a.b" << 10)), BSON("a.b" << 10 << "c" << 30)); + + ASSERT_EQUALS(normKey(pattern, BSON("b" << 10)), BSONObj()); + ASSERT_EQUALS(normKey(pattern, BSON("a" << 10 << "c" << 30)), BSONObj()); + ASSERT_EQUALS(normKey(pattern, BSON("a.b" << BSON("$gt" << 10) << "c" << 30)), BSONObj()); +} + +static BSONObj docKey(const ShardKeyPattern& pattern, const BSONObj& doc) { + return pattern.extractShardKeyFromDoc(doc); +} + +TEST(ShardKeyPattern, ExtractDocShardKeySingle) { + // + // Single field ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a" << 1)); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:10}")), fromjson("{a:10}")); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:'20'}")), fromjson("{a:10}")); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:10}, c:30}")), fromjson("{a:{b:10}}")); + const BSONRegEx regex("abc"); + ASSERT_EQUALS(docKey(pattern, + BSON("a" << regex << "b" + << "20")), + BSON("a" << regex)); + const BSONObj ref = BSON("$ref" + << "coll" + << "$id" << 1); + ASSERT_EQUALS(docKey(pattern, BSON("a" << ref)), BSON("a" << ref)); + + ASSERT_EQUALS(docKey(pattern, BSONObj()), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{b:10}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, BSON("" << 10)), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:[1, 2]}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{$invalid:true}}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{$gt:10}}")), BSONObj()); + // BSONObjIterator breaks this for now + // ASSERT_EQUALS(docKey(pattern, BSON("a" << 10 << "a" << 20)), BSONObj()); +} + +TEST(ShardKeyPattern, ExtractDocShardKeyCompound) { + // + // Compound ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a" << 1 << "b" << 1.0)); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:'20'}")), fromjson("{a:10, b:'20'}")); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:'20', c:30}")), fromjson("{a:10, b:'20'}")); + ASSERT_EQUALS(docKey(pattern, + BSON("c" << 30 << "b" + << "20" + << "a" << 10)), + fromjson("{a:10, b:'20'}")); + + ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:[1, 2]}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:{$invalid:true}}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{b:20}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, + BSON("" << 10 << "b" + << "20")), + BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:{$gt:20}}")), BSONObj()); + + // Ordering + ASSERT_EQUALS(docKey(pattern, BSON("b" << 20 << "a" << 10)).firstElement().numberInt(), 10); +} + +TEST(ShardKeyPattern, ExtractDocShardKeyNested) { + // + // Nested ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0f)); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:10}, c:30}")), fromjson("{'a.b':10, c:30}")); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{d:[1,2],b:10},c:30,d:40}")), + fromjson("{'a.b':10, c:30}")); + const BSONObj ref = BSON("$ref" + << "coll" + << "$id" << 1); + ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("b" << ref) << "c" << 30)), + BSON("a.b" << ref << "c" << 30)); + + ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, c:30}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{d:40}, c:30}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:[{b:10}, {b:20}], c:30}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:[10, 20]}, c:30}")), BSONObj()); +} - TEST(ShardKeyPattern, ValidShardKeyPatternNested) { +TEST(ShardKeyPattern, ExtractDocShardKeyDeepNested) { + // + // Deeply nested ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a.b.c" << 1)); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:{c:10}}}")), fromjson("{'a.b.c':10}")); + + ASSERT_EQUALS(docKey(pattern, fromjson("{a:[{b:{c:10}}]}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:[{c:10}]}}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:{c:[10, 20]}}}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:[{c:10}, {c:20}]}}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:[{b:{c:10}},{b:{c:20}}]}")), BSONObj()); + ASSERT_EQUALS(docKey(pattern, fromjson("{a:[{b:[{c:10},{c:20}]},{b:[{c:30},{c:40}]}]}}")), + BSONObj()); +} - // - // Nested ShardKeyPatterns - // +TEST(ShardKeyPattern, ExtractDocShardKeyHashed) { + // + // Hashed ShardKeyPattern + // + + const string value = "12345"; + const BSONObj bsonValue = BSON("" << value); + const long long hashValue = + BSONElementHasher::hash64(bsonValue.firstElement(), BSONElementHasher::DEFAULT_HASH_SEED); + + ShardKeyPattern pattern(BSON("a.b" + << "hashed")); + ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("b" << value))), BSON("a.b" << hashValue)); + ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("b" << value) << "c" << 30)), + BSON("a.b" << hashValue)); + ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("c" << 30 << "b" << value))), + BSON("a.b" << hashValue)); + + ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("c" << value))), BSONObj()); + ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("b" << BSON_ARRAY(value)))), BSONObj()); + ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON_ARRAY(BSON("b" << value)))), BSONObj()); +} - ASSERT(ShardKeyPattern(BSON("a.b" << 1)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a.b" << -1)).isValid()); - ASSERT(ShardKeyPattern(BSON("a.b.c.d" << 1.0)).isValid()); +static BSONObj queryKey(const ShardKeyPattern& pattern, const BSONObj& query) { + StatusWith<BSONObj> status = pattern.extractShardKeyFromQuery(query); + if (!status.isOK()) + return BSONObj(); + return status.getValue(); +} - ASSERT(!ShardKeyPattern(BSON("a" << BSON( "b" << 1 ))).isValid()); +TEST(ShardKeyPattern, ExtractQueryShardKeySingle) { + // + // Single field ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a" << 1)); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10}")), fromjson("{a:10}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:'20'}")), fromjson("{a:10}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:10}, c:30}")), fromjson("{a:{b:10}}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:{$gt:20}}")), fromjson("{a:10}")); + + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$gt:10}}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:{$invalid:'20'}}")), BSONObj()); + + // Doc key extraction shouldn't work with query + ASSERT_EQUALS(docKey(pattern, fromjson("{a:{$eq:[10, 20]}, c:30}")), BSONObj()); + + // $eq/$or/$and/$all + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$eq:10}}")), fromjson("{a:10}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{$eq:10}}]}")), fromjson("{a:10}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:{$eq:10}},{b:'20'}]}")), + fromjson("{a:10}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$all:[10]}}")), fromjson("{a:10}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{$eq:10}},{a:10}]}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:10},{a:10}]}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$all:[10,10]}}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{$eq:10}},{b:'20'}]}")), BSONObj()); + + // Regex can't be extracted from query + const BSONRegEx regex("abc"); + ASSERT_EQUALS(queryKey(pattern, + BSON("a" << regex << "b" + << "20")), + BSONObj()); +} - ASSERT(!ShardKeyPattern(BSON("a.b." << 1)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a.b.." << 1)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a..b" << 1)).isValid()); +TEST(ShardKeyPattern, ExtractQueryShardKeyCompound) { + // + // Compound ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a" << 1 << "b" << 1.0)); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:'20'}")), fromjson("{a:10, b:'20'}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:'20', c:30}")), fromjson("{a:10, b:'20'}")); + ASSERT_EQUALS(queryKey(pattern, + BSON("c" << 30 << "b" + << "20" + << "a" << 10)), + fromjson("{a:10, b:'20'}")); + + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:[1, 2]}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:{$invalid:true}}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{b:20}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, + BSON("" << 10 << "b" + << "20")), + BSONObj()); + + // $eq/$or/$and/$all + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$eq:10}, b:{$all:['20']}}")), + fromjson("{a:10, b:'20'}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:{$eq:10},b:{$eq:'20'}}]}")), + fromjson("{a:10, b:'20'}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:{$eq:10}},{b:{$eq:'20'}}]}")), + fromjson("{a:10, b:'20'}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:{$gt:20}}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{$eq:10}},{b:'20'}]}")), BSONObj()); + + // Ordering + ASSERT_EQUALS(queryKey(pattern, BSON("b" << 20 << "a" << 10)).firstElement().numberInt(), 10); +} - ASSERT(ShardKeyPattern(BSON("a" << 1 << "c.d" << 1.0 << "e.f.g" << 1.0f)).isValid()); - ASSERT(ShardKeyPattern(BSON("a" << 1 << "a.b" << 1.0 << "a.b.c" << 1.0f)).isValid()); +TEST(ShardKeyPattern, ExtractQueryShardKeyNested) { + // + // Nested ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0f)); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:10}, c:30}")), fromjson("{'a.b':10, c:30}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b':{$eq:10}, c:30, d:40}")), + fromjson("{'a.b':10, c:30}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{'a.b':10, c:30, d:40}]}")), + fromjson("{'a.b':10, c:30}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b':{$all:[10]}, c:30, d:40}")), + fromjson("{'a.b':10, c:30}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:10,d:40}, c:30}")), + fromjson("{'a.b':10, c:30}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{'a.b':{$eq:10}}, {c:30}]}")), + fromjson("{'a.b':10, c:30}")); + + // Nested $eq is actually a document element + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:{$eq:10}}, c:30}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:{b:{$eq:10}}}, {c:30}]}")), BSONObj()); + + ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{b:{$eq:10}}}, {c:30}]}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, c:30}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:10}, c:{$gt:30}}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{d:40}, c:30}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:[{b:10}, {b:20}], c:30}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:{$eq:[10, 20]}}, c:30}")), BSONObj()); +} - ASSERT(!ShardKeyPattern(BSON("a" << 1 << "a.b." << 1.0)).isValid()); - ASSERT(!ShardKeyPattern(BSON("a" << BSON( "b" << 1 ) << "c.d" << 1.0)).isValid()); +TEST(ShardKeyPattern, ExtractQueryShardKeyDeepNested) { + // + // Deeply nested ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a.b.c" << 1)); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:{c:10}}}")), fromjson("{'a.b.c':10}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b.c':10}")), fromjson("{'a.b.c':10}")); + ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b.c':{$eq:10}}")), fromjson("{'a.b.c':10}")); + + // Arrays at any nesting level means we can't extract a shard key + ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b.c':[10]}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b':[{c:10}]}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:[{b:{c:10}}]}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:[{c:10}]}}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:{c:[10, 20]}}}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:[{c:10}, {c:20}]}}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:[{b:{c:10}},{b:{c:20}}]}")), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, fromjson("{a:[{b:[{c:10},{c:20}]},{b:[{c:30},{c:40}]}]}}")), + BSONObj()); +} - } +TEST(ShardKeyPattern, ExtractQueryShardKeyHashed) { + // + // Hashed ShardKeyPattern + // + + const string value = "12345"; + const BSONObj bsonValue = BSON("" << value); + const long long hashValue = + BSONElementHasher::hash64(bsonValue.firstElement(), BSONElementHasher::DEFAULT_HASH_SEED); + + // Hashed works basically the same as non-hashed, but applies the hash function at the end + ShardKeyPattern pattern(BSON("a.b" + << "hashed")); + ASSERT_EQUALS(queryKey(pattern, BSON("a.b" << value)), BSON("a.b" << hashValue)); + ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("b" << value))), BSON("a.b" << hashValue)); + ASSERT_EQUALS(queryKey(pattern, BSON("a.b" << BSON("$eq" << value))), BSON("a.b" << hashValue)); + ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("b" << value) << "c" << 30)), + BSON("a.b" << hashValue)); + ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("c" << 30 << "b" << value))), + BSON("a.b" << hashValue)); + ASSERT_EQUALS(queryKey(pattern, // + BSON("$and" << BSON_ARRAY(BSON("a.b" << BSON("$eq" << value))))), + BSON("a.b" << hashValue)); + + ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("b" << BSON("$eq" << value)))), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, BSON("a.b" << BSON("$gt" << value))), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("c" << value))), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("b" << BSON_ARRAY(value)))), BSONObj()); + ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON_ARRAY(BSON("b" << value)))), BSONObj()); +} - TEST(ShardKeyPattern, IsShardKey) { +static bool indexComp(const ShardKeyPattern& pattern, const BSONObj& indexPattern) { + return pattern.isUniqueIndexCompatible(indexPattern); +} - ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0f)); +TEST(ShardKeyPattern, UniqueIndexCompatibleSingle) { + // + // Single field ShardKeyPatterns + // - ASSERT(pattern.isShardKey(BSON("a.b" << 10 << "c" << 30))); - ASSERT(pattern.isShardKey(BSON("c" << 30 << "a.b" << 10))); + ShardKeyPattern pattern(BSON("a" << 1)); + ASSERT(indexComp(pattern, BSON("a" << 1))); + ASSERT(indexComp(pattern, BSON("a" << -1))); + ASSERT(indexComp(pattern, BSON("a" << 1 << "b" << 1))); + ASSERT(indexComp(pattern, BSON("a" << -1 << "b" << 1))); - ASSERT(!pattern.isShardKey(BSON("b" << 10))); - ASSERT(!pattern.isShardKey(BSON("a" << 10 << "c" << 30))); - ASSERT(!pattern.isShardKey(BSON("a" << BSON("b" << 10) << "c" << 30))); - } + ASSERT(indexComp(pattern, BSON("_id" << 1))); + ASSERT(indexComp(pattern, BSON("_id" << -1 << "b" << 1))); - static BSONObj normKey(const ShardKeyPattern& pattern, const BSONObj& doc) { - return pattern.normalizeShardKey(doc); - } + ASSERT(!indexComp(pattern, BSON("b" << 1))); + ASSERT(!indexComp(pattern, BSON("b" << -1 << "a" << 1))); +} - TEST(ShardKeyPattern, NormalizeShardKey) { +TEST(ShardKeyPattern, UniqueIndexCompatibleCompound) { + // + // Compound ShardKeyPatterns + // - ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0f)); + ShardKeyPattern pattern(BSON("a" << 1 << "b" << 1.0)); + ASSERT(indexComp(pattern, BSON("a" << 1 << "b" << 1))); + ASSERT(indexComp(pattern, BSON("a" << 1 << "b" << -1.0))); + ASSERT(indexComp(pattern, BSON("a" << 1 << "b" << -1.0 << "c" << 1))); - ASSERT_EQUALS(normKey(pattern, BSON("a.b" << 10 << "c" << 30)), - BSON("a.b" << 10 << "c" << 30)); - ASSERT_EQUALS(normKey(pattern, BSON("c" << 30 << "a.b" << 10)), - BSON("a.b" << 10 << "c" << 30)); + ASSERT(indexComp(pattern, BSON("_id" << 1))); + ASSERT(indexComp(pattern, BSON("_id" << -1 << "c" << 1))); - ASSERT_EQUALS(normKey(pattern, BSON("b" << 10)), BSONObj()); - ASSERT_EQUALS(normKey(pattern, BSON("a" << 10 << "c" << 30)), BSONObj()); - ASSERT_EQUALS(normKey(pattern, BSON("a.b" << BSON("$gt" << 10) << "c" << 30)), BSONObj()); - } + ASSERT(!indexComp(pattern, BSON("a" << 1))); + ASSERT(!indexComp(pattern, BSON("b" << 1))); + ASSERT(!indexComp(pattern, BSON("a" << 1 << "c" << 1.0f))); + ASSERT(!indexComp(pattern, BSON("b" << -1 << "a" << 1 << "c" << 1))); +} - static BSONObj docKey(const ShardKeyPattern& pattern, const BSONObj& doc) { - return pattern.extractShardKeyFromDoc(doc); - } +TEST(ShardKeyPattern, UniqueIndexCompatibleNested) { + // + // Nested ShardKeyPatterns + // - TEST(ShardKeyPattern, ExtractDocShardKeySingle) { + ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0)); + ASSERT(indexComp(pattern, BSON("a.b" << 1 << "c" << 1.0f))); - // - // Single field ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a" << 1)); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:10}")), fromjson("{a:10}")); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:'20'}")), fromjson("{a:10}")); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:10}, c:30}")), fromjson("{a:{b:10}}")); - const BSONRegEx regex("abc"); - ASSERT_EQUALS(docKey(pattern, BSON("a" << regex << "b" << "20")), BSON("a" << regex)); - const BSONObj ref = BSON("$ref" << "coll" << "$id" << 1); - ASSERT_EQUALS(docKey(pattern, BSON("a" << ref)), BSON("a" << ref)); - - ASSERT_EQUALS(docKey(pattern, BSONObj()), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{b:10}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, BSON("" << 10)), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:[1, 2]}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{$invalid:true}}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{$gt:10}}")), BSONObj()); - // BSONObjIterator breaks this for now - //ASSERT_EQUALS(docKey(pattern, BSON("a" << 10 << "a" << 20)), BSONObj()); - } - - TEST(ShardKeyPattern, ExtractDocShardKeyCompound) { - - // - // Compound ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a" << 1 << "b" << 1.0)); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:'20'}")), fromjson("{a:10, b:'20'}")); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:'20', c:30}")), - fromjson("{a:10, b:'20'}")); - ASSERT_EQUALS(docKey(pattern, BSON("c" << 30 << "b" << "20" << "a" << 10)), - fromjson("{a:10, b:'20'}")); - - ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:[1, 2]}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:{$invalid:true}}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{b:20}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, BSON("" << 10 << "b" << "20")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, b:{$gt:20}}")), BSONObj()); - - // Ordering - ASSERT_EQUALS(docKey(pattern, BSON("b" << 20 << "a" << 10)).firstElement().numberInt(), 10); - } - - TEST(ShardKeyPattern, ExtractDocShardKeyNested) { - - // - // Nested ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0f)); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:10}, c:30}")), fromjson("{'a.b':10, c:30}")); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{d:[1,2],b:10},c:30,d:40}")), - fromjson("{'a.b':10, c:30}")); - const BSONObj ref = BSON("$ref" << "coll" << "$id" << 1); - ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON( "b" << ref) << "c" << 30)), - BSON("a.b" << ref << "c" << 30)); - - ASSERT_EQUALS(docKey(pattern, fromjson("{a:10, c:30}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{d:40}, c:30}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:[{b:10}, {b:20}], c:30}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:[10, 20]}, c:30}")), BSONObj()); - } - - TEST(ShardKeyPattern, ExtractDocShardKeyDeepNested) { - - // - // Deeply nested ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a.b.c" << 1)); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:{c:10}}}")), fromjson("{'a.b.c':10}")); - - ASSERT_EQUALS(docKey(pattern, fromjson("{a:[{b:{c:10}}]}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:[{c:10}]}}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:{c:[10, 20]}}}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{b:[{c:10}, {c:20}]}}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:[{b:{c:10}},{b:{c:20}}]}")), BSONObj()); - ASSERT_EQUALS(docKey(pattern, fromjson("{a:[{b:[{c:10},{c:20}]},{b:[{c:30},{c:40}]}]}}")), - BSONObj()); - } - - TEST(ShardKeyPattern, ExtractDocShardKeyHashed) { - - // - // Hashed ShardKeyPattern - // - - const string value = "12345"; - const BSONObj bsonValue = BSON("" << value); - const long long hashValue = BSONElementHasher::hash64(bsonValue.firstElement(), - BSONElementHasher::DEFAULT_HASH_SEED); - - ShardKeyPattern pattern(BSON("a.b" << "hashed")); - ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("b" << value))), BSON("a.b" << hashValue)); - ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("b" << value) << "c" << 30)), - BSON("a.b" << hashValue)); - ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("c" << 30 << "b" << value))), - BSON("a.b" << hashValue)); - - ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("c" << value))), BSONObj()); - ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON("b" << BSON_ARRAY(value)))), BSONObj()); - ASSERT_EQUALS(docKey(pattern, BSON("a" << BSON_ARRAY(BSON("b" << value)))), BSONObj()); - } - - static BSONObj queryKey(const ShardKeyPattern& pattern, const BSONObj& query) { - StatusWith<BSONObj> status = pattern.extractShardKeyFromQuery(query); - if (!status.isOK()) - return BSONObj(); - return status.getValue(); - } - - TEST(ShardKeyPattern, ExtractQueryShardKeySingle) { - - // - // Single field ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a" << 1)); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10}")), fromjson("{a:10}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:'20'}")), fromjson("{a:10}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:10}, c:30}")), fromjson("{a:{b:10}}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:{$gt:20}}")), fromjson("{a:10}")); - - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$gt:10}}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:{$invalid:'20'}}")), BSONObj()); - - // Doc key extraction shouldn't work with query - ASSERT_EQUALS(docKey(pattern, fromjson("{a:{$eq:[10, 20]}, c:30}")), BSONObj()); - - // $eq/$or/$and/$all - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$eq:10}}")), fromjson("{a:10}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{$eq:10}}]}")), fromjson("{a:10}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:{$eq:10}},{b:'20'}]}")), - fromjson("{a:10}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$all:[10]}}")), fromjson("{a:10}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{$eq:10}},{a:10}]}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:10},{a:10}]}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$all:[10,10]}}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{$eq:10}},{b:'20'}]}")), BSONObj()); - - // Regex can't be extracted from query - const BSONRegEx regex("abc"); - ASSERT_EQUALS(queryKey(pattern, BSON("a" << regex << "b" << "20")), BSONObj()); - } - - TEST(ShardKeyPattern, ExtractQueryShardKeyCompound) { - - // - // Compound ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a" << 1 << "b" << 1.0)); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:'20'}")), fromjson("{a:10, b:'20'}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:'20', c:30}")), - fromjson("{a:10, b:'20'}")); - ASSERT_EQUALS(queryKey(pattern, BSON("c" << 30 << "b" << "20" << "a" << 10)), - fromjson("{a:10, b:'20'}")); - - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:[1, 2]}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:{$invalid:true}}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{b:20}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, BSON("" << 10 << "b" << "20")), BSONObj()); - - // $eq/$or/$and/$all - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{$eq:10}, b:{$all:['20']}}")), - fromjson("{a:10, b:'20'}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:{$eq:10},b:{$eq:'20'}}]}")), - fromjson("{a:10, b:'20'}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:{$eq:10}},{b:{$eq:'20'}}]}")), - fromjson("{a:10, b:'20'}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, b:{$gt:20}}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{$eq:10}},{b:'20'}]}")), BSONObj()); - - // Ordering - ASSERT_EQUALS(queryKey(pattern, BSON("b" << 20 << "a" << 10)).firstElement().numberInt(), - 10); - - } - - TEST(ShardKeyPattern, ExtractQueryShardKeyNested) { - - // - // Nested ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0f)); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:10}, c:30}")), - fromjson("{'a.b':10, c:30}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b':{$eq:10}, c:30, d:40}")), - fromjson("{'a.b':10, c:30}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{'a.b':10, c:30, d:40}]}")), - fromjson("{'a.b':10, c:30}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b':{$all:[10]}, c:30, d:40}")), - fromjson("{'a.b':10, c:30}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:10,d:40}, c:30}")), - fromjson("{'a.b':10, c:30}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{'a.b':{$eq:10}}, {c:30}]}")), - fromjson("{'a.b':10, c:30}")); - - // Nested $eq is actually a document element - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:{$eq:10}}, c:30}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{$and:[{a:{b:{$eq:10}}}, {c:30}]}")), BSONObj()); - - ASSERT_EQUALS(queryKey(pattern, fromjson("{$or:[{a:{b:{$eq:10}}}, {c:30}]}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:10, c:30}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:10}, c:{$gt:30}}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{d:40}, c:30}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:[{b:10}, {b:20}], c:30}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:{$eq:[10, 20]}}, c:30}")), BSONObj()); - } - - TEST(ShardKeyPattern, ExtractQueryShardKeyDeepNested) { - - // - // Deeply nested ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a.b.c" << 1)); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:{c:10}}}")), fromjson("{'a.b.c':10}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b.c':10}")), fromjson("{'a.b.c':10}")); - ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b.c':{$eq:10}}")), fromjson("{'a.b.c':10}")); - - // Arrays at any nesting level means we can't extract a shard key - ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b.c':[10]}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{'a.b':[{c:10}]}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:[{b:{c:10}}]}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:[{c:10}]}}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:{c:[10, 20]}}}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:{b:[{c:10}, {c:20}]}}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:[{b:{c:10}},{b:{c:20}}]}")), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, fromjson("{a:[{b:[{c:10},{c:20}]},{b:[{c:30},{c:40}]}]}}")), - BSONObj()); - - } - - TEST(ShardKeyPattern, ExtractQueryShardKeyHashed) { - - // - // Hashed ShardKeyPattern - // - - const string value = "12345"; - const BSONObj bsonValue = BSON("" << value); - const long long hashValue = BSONElementHasher::hash64(bsonValue.firstElement(), - BSONElementHasher::DEFAULT_HASH_SEED); - - // Hashed works basically the same as non-hashed, but applies the hash function at the end - ShardKeyPattern pattern(BSON("a.b" << "hashed")); - ASSERT_EQUALS(queryKey(pattern, BSON("a.b" << value)), BSON("a.b" << hashValue)); - ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("b" << value))), BSON("a.b" << hashValue)); - ASSERT_EQUALS(queryKey(pattern, BSON("a.b" << BSON("$eq" << value))), - BSON("a.b" << hashValue)); - ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("b" << value) << "c" << 30)), - BSON("a.b" << hashValue)); - ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("c" << 30 << "b" << value))), - BSON("a.b" << hashValue)); - ASSERT_EQUALS(queryKey(pattern, // - BSON("$and" << BSON_ARRAY(BSON("a.b" << BSON("$eq" << value))))), - BSON("a.b" << hashValue)); - - ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("b" << BSON("$eq" << value)))), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, BSON("a.b" << BSON("$gt" << value))), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("c" << value))), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON("b" << BSON_ARRAY(value)))), BSONObj()); - ASSERT_EQUALS(queryKey(pattern, BSON("a" << BSON_ARRAY(BSON("b" << value)))), BSONObj()); - } - - static bool indexComp(const ShardKeyPattern& pattern, const BSONObj& indexPattern) { - return pattern.isUniqueIndexCompatible(indexPattern); - } - - TEST(ShardKeyPattern, UniqueIndexCompatibleSingle) { - - // - // Single field ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a" << 1)); - ASSERT(indexComp(pattern, BSON("a" << 1))); - ASSERT(indexComp(pattern, BSON("a" << -1))); - ASSERT(indexComp(pattern, BSON("a" << 1 << "b" << 1))); - ASSERT(indexComp(pattern, BSON("a" << -1 << "b" << 1))); - - ASSERT(indexComp(pattern, BSON("_id" << 1))); - ASSERT(indexComp(pattern, BSON("_id" << -1 << "b" << 1))); - - ASSERT(!indexComp(pattern, BSON("b" << 1))); - ASSERT(!indexComp(pattern, BSON("b" << -1 << "a" << 1))); - } - - TEST(ShardKeyPattern, UniqueIndexCompatibleCompound) { - - // - // Compound ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a" << 1 << "b" << 1.0)); - ASSERT(indexComp(pattern, BSON("a" << 1 << "b" << 1))); - ASSERT(indexComp(pattern, BSON("a" << 1 << "b" << -1.0))); - ASSERT(indexComp(pattern, BSON("a" << 1 << "b" << -1.0 << "c" << 1))); - - ASSERT(indexComp(pattern, BSON("_id" << 1))); - ASSERT(indexComp(pattern, BSON("_id" << -1 << "c" << 1))); - - ASSERT(!indexComp(pattern, BSON("a" << 1))); - ASSERT(!indexComp(pattern, BSON("b" << 1))); - ASSERT(!indexComp(pattern, BSON("a" << 1 << "c" << 1.0f))); - ASSERT(!indexComp(pattern, BSON("b" << -1 << "a" << 1 << "c" << 1))); - } - - TEST(ShardKeyPattern, UniqueIndexCompatibleNested) { - - // - // Nested ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a.b" << 1 << "c" << 1.0)); - ASSERT(indexComp(pattern, BSON("a.b" << 1 << "c" << 1.0f))); - - ASSERT(!indexComp(pattern, BSON("a.b" << 1))); - ASSERT(!indexComp(pattern, BSON("a" << 1 << "c" << -1.0))); - ASSERT(!indexComp(pattern, BSON("c" << -1 << "a.b" << 1))); - } - - TEST(ShardKeyPattern, UniqueIndexCompatibleHashed) { - - // - // Hashed ShardKeyPatterns - // - - ShardKeyPattern pattern(BSON("a.b" << "hashed")); - - ASSERT(indexComp(pattern, BSON("a.b" << 1))); - ASSERT(indexComp(pattern, BSON("a.b" << -1))); - ASSERT(indexComp(pattern, BSON("a.b" << 1 << "c" << 1))); - ASSERT(indexComp(pattern, BSON("a.b" << -1 << "c" << 1))); - - ASSERT(indexComp(pattern, BSON("_id" << 1))); - ASSERT(indexComp(pattern, BSON("_id" << -1 << "c" << 1))); - - ASSERT(!indexComp(pattern, BSON("c" << 1))); - ASSERT(!indexComp(pattern, BSON("c" << -1 << "a.b" << 1))); - } + ASSERT(!indexComp(pattern, BSON("a.b" << 1))); + ASSERT(!indexComp(pattern, BSON("a" << 1 << "c" << -1.0))); + ASSERT(!indexComp(pattern, BSON("c" << -1 << "a.b" << 1))); } +TEST(ShardKeyPattern, UniqueIndexCompatibleHashed) { + // + // Hashed ShardKeyPatterns + // + + ShardKeyPattern pattern(BSON("a.b" + << "hashed")); + + ASSERT(indexComp(pattern, BSON("a.b" << 1))); + ASSERT(indexComp(pattern, BSON("a.b" << -1))); + ASSERT(indexComp(pattern, BSON("a.b" << 1 << "c" << 1))); + ASSERT(indexComp(pattern, BSON("a.b" << -1 << "c" << 1))); + + ASSERT(indexComp(pattern, BSON("_id" << 1))); + ASSERT(indexComp(pattern, BSON("_id" << -1 << "c" << 1))); + + ASSERT(!indexComp(pattern, BSON("c" << 1))); + ASSERT(!indexComp(pattern, BSON("c" << -1 << "a.b" << 1))); +} +} diff --git a/src/mongo/s/shard_resolver.h b/src/mongo/s/shard_resolver.h index ac51c5fcd04..6c229290805 100644 --- a/src/mongo/s/shard_resolver.h +++ b/src/mongo/s/shard_resolver.h @@ -36,26 +36,23 @@ namespace mongo { +/** + * Given a shard name, the ShardResolver resolves a particular host on that shard to contact. + * + * TODO: Internally, this is backed by a cache - do we need explicit refresh mechanisms built + * into this interface? + */ +class ShardResolver { +public: + virtual ~ShardResolver() {} + /** - * Given a shard name, the ShardResolver resolves a particular host on that shard to contact. + * Returns a host we can use for write ops to this shard. * - * TODO: Internally, this is backed by a cache - do we need explicit refresh mechanisms built - * into this interface? + * Returns !OK with message if the shard host could not be found for other reasons. */ - class ShardResolver { - public: - - virtual ~ShardResolver() { - } - - /** - * Returns a host we can use for write ops to this shard. - * - * Returns !OK with message if the shard host could not be found for other reasons. - */ - virtual Status chooseWriteHost( const std::string& shardName, - ConnectionString* shardHost ) const = 0; - - }; + virtual Status chooseWriteHost(const std::string& shardName, + ConnectionString* shardHost) const = 0; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/stale_exception.h b/src/mongo/s/stale_exception.h index 9075e6ee93e..227fced8698 100644 --- a/src/mongo/s/stale_exception.h +++ b/src/mongo/s/stale_exception.h @@ -35,143 +35,125 @@ namespace mongo { - using mongoutils::str::stream; +using mongoutils::str::stream; + +/** + * Thrown whenever your config info for a given shard/chunk is out of date. + */ +class StaleConfigException : public AssertionException { +public: + StaleConfigException(const std::string& ns, + const std::string& raw, + int code, + ChunkVersion received, + ChunkVersion wanted) + : AssertionException(stream() << raw << " ( ns : " << ns + << ", received : " << received.toString() + << ", wanted : " << wanted.toString() << ", " + << (code == SendStaleConfigCode ? "send" : "recv") << " )", + code), + _ns(ns), + _received(received), + _wanted(wanted) {} + + /** Preferred if we're rebuilding this from a thrown exception */ + StaleConfigException(const std::string& raw, int code, const BSONObj& error) + : AssertionException( + stream() << raw + << " ( ns : " << (error["ns"].type() == String ? error["ns"].String() + : std::string("<unknown>")) + << ", received : " << ChunkVersion::fromBSON(error, "vReceived").toString() + << ", wanted : " << ChunkVersion::fromBSON(error, "vWanted").toString() + << ", " << (code == SendStaleConfigCode ? "send" : "recv") << " )", + code), + // For legacy reasons, we may not always get a namespace here + _ns(error["ns"].type() == String ? error["ns"].String() : ""), + _received(ChunkVersion::fromBSON(error, "vReceived")), + _wanted(ChunkVersion::fromBSON(error, "vWanted")) {} + + /** + * Needs message so when we trace all exceptions on construction we get a useful + * message + */ + StaleConfigException() + : AssertionException("initializing empty stale config exception object", 0) {} + + virtual ~StaleConfigException() throw() {} + + virtual void appendPrefix(std::stringstream& ss) const { + ss << "stale sharding config exception: "; + } + + std::string getns() const { + return _ns; + } /** - * Thrown whenever your config info for a given shard/chunk is out of date. + * true if this exception would require a full reload of config data to resolve */ - class StaleConfigException : public AssertionException { - public: - StaleConfigException(const std::string& ns, + bool requiresFullReload() const { + return !_received.hasEqualEpoch(_wanted) || _received.isSet() != _wanted.isSet(); + } + + static bool parse(const std::string& big, std::string& ns, std::string& raw) { + std::string::size_type start = big.find('['); + if (start == std::string::npos) + return false; + std::string::size_type end = big.find(']', start); + if (end == std::string::npos) + return false; + + ns = big.substr(start + 1, (end - start) - 1); + raw = big.substr(end + 1); + return true; + } + + ChunkVersion getVersionReceived() const { + return _received; + } + + ChunkVersion getVersionWanted() const { + return _wanted; + } + + StaleConfigException& operator=(const StaleConfigException& elem) { + this->_ei.msg = elem._ei.msg; + this->_ei.code = elem._ei.code; + this->_ns = elem._ns; + this->_received = elem._received; + this->_wanted = elem._wanted; + + return *this; + } + +private: + std::string _ns; + ChunkVersion _received; + ChunkVersion _wanted; +}; + +class SendStaleConfigException : public StaleConfigException { +public: + SendStaleConfigException(const std::string& ns, const std::string& raw, - int code, ChunkVersion received, ChunkVersion wanted) - : AssertionException(stream() << raw << " ( ns : " << ns - << ", received : " << received.toString() - << ", wanted : " << wanted.toString() - << ", " << ( code == SendStaleConfigCode ? - "send" : "recv" ) << " )", - code), - _ns(ns), - _received( received ), - _wanted( wanted ) { - } - - /** Preferred if we're rebuilding this from a thrown exception */ - StaleConfigException( const std::string& raw, - int code, - const BSONObj& error) - : AssertionException( stream() << raw << " ( ns : " - << ( error["ns"].type() == String ? - error["ns"].String() : std::string("<unknown>") ) - << ", received : " - << ChunkVersion::fromBSON( error, "vReceived" ).toString() - << ", wanted : " - << ChunkVersion::fromBSON( error, "vWanted" ).toString() - << ", " - << ( code == SendStaleConfigCode ? - "send" : "recv" ) << " )", - code ), - // For legacy reasons, we may not always get a namespace here - _ns( error["ns"].type() == String ? error["ns"].String() : "" ), - _received( ChunkVersion::fromBSON( error, "vReceived" ) ), - _wanted( ChunkVersion::fromBSON( error, "vWanted" ) ) { - } - - /** - * Needs message so when we trace all exceptions on construction we get a useful - * message - */ - StaleConfigException() : - AssertionException( "initializing empty stale config exception object", 0 ) { - } - - virtual ~StaleConfigException() throw() {} - - virtual void appendPrefix( std::stringstream& ss ) const { - ss << "stale sharding config exception: "; - } - - std::string getns() const { return _ns; } - - /** - * true if this exception would require a full reload of config data to resolve - */ - bool requiresFullReload() const { - return ! _received.hasEqualEpoch( _wanted ) || - _received.isSet() != _wanted.isSet(); - } - - static bool parse( const std::string& big , std::string& ns , std::string& raw ) { - std::string::size_type start = big.find( '[' ); - if ( start == std::string::npos ) - return false; - std::string::size_type end = big.find( ']' ,start ); - if ( end == std::string::npos ) - return false; - - ns = big.substr( start + 1 , ( end - start ) - 1 ); - raw = big.substr( end + 1 ); - return true; - } - - ChunkVersion getVersionReceived() const { - return _received; - } - - ChunkVersion getVersionWanted() const { - return _wanted; - } - - StaleConfigException& operator=(const StaleConfigException& elem) { - this->_ei.msg = elem._ei.msg; - this->_ei.code = elem._ei.code; - this->_ns = elem._ns; - this->_received = elem._received; - this->_wanted = elem._wanted; - - return *this; - } - - private: - std::string _ns; - ChunkVersion _received; - ChunkVersion _wanted; - }; - - class SendStaleConfigException : public StaleConfigException { - public: - SendStaleConfigException(const std::string& ns, - const std::string& raw, - ChunkVersion received, - ChunkVersion wanted) - : StaleConfigException(ns, raw, SendStaleConfigCode, received, wanted) { - - } - - SendStaleConfigException( const std::string& raw, - const BSONObj& error) - : StaleConfigException(raw, SendStaleConfigCode, error) { - - } - }; - - class RecvStaleConfigException : public StaleConfigException { - public: - RecvStaleConfigException(const std::string& ns, - const std::string& raw, - ChunkVersion received, - ChunkVersion wanted) - : StaleConfigException(ns, raw, RecvStaleConfigCode, received, wanted) { - - } - - RecvStaleConfigException(const std::string& raw, - const BSONObj& error) - : StaleConfigException(raw, RecvStaleConfigCode, error) { - - } - }; - -} // namespace mongo + : StaleConfigException(ns, raw, SendStaleConfigCode, received, wanted) {} + + SendStaleConfigException(const std::string& raw, const BSONObj& error) + : StaleConfigException(raw, SendStaleConfigCode, error) {} +}; + +class RecvStaleConfigException : public StaleConfigException { +public: + RecvStaleConfigException(const std::string& ns, + const std::string& raw, + ChunkVersion received, + ChunkVersion wanted) + : StaleConfigException(ns, raw, RecvStaleConfigCode, received, wanted) {} + + RecvStaleConfigException(const std::string& raw, const BSONObj& error) + : StaleConfigException(raw, RecvStaleConfigCode, error) {} +}; + +} // namespace mongo diff --git a/src/mongo/s/strategy.cpp b/src/mongo/s/strategy.cpp index 061059c709d..f2d44b594c0 100644 --- a/src/mongo/s/strategy.cpp +++ b/src/mongo/s/strategy.cpp @@ -69,643 +69,620 @@ namespace mongo { - using std::unique_ptr; - using std::shared_ptr; - using std::endl; - using std::set; - using std::string; - using std::stringstream; - using std::vector; - - static bool _isSystemIndexes( const char* ns ) { - return nsToCollectionSubstring(ns) == "system.indexes"; +using std::unique_ptr; +using std::shared_ptr; +using std::endl; +using std::set; +using std::string; +using std::stringstream; +using std::vector; + +static bool _isSystemIndexes(const char* ns) { + return nsToCollectionSubstring(ns) == "system.indexes"; +} + +/** + * Returns true if request is a query for sharded indexes. + */ +static bool doShardedIndexQuery(Request& r, const QuerySpec& qSpec) { + // Extract the ns field from the query, which may be embedded within the "query" or + // "$query" field. + auto nsField = qSpec.filter()["ns"]; + if (nsField.eoo()) { + return false; } + const NamespaceString indexNSSQuery(nsField.str()); - /** - * Returns true if request is a query for sharded indexes. - */ - static bool doShardedIndexQuery(Request& r, const QuerySpec& qSpec) { - // Extract the ns field from the query, which may be embedded within the "query" or - // "$query" field. - auto nsField = qSpec.filter()["ns"]; - if (nsField.eoo()) { - return false; - } - const NamespaceString indexNSSQuery(nsField.str()); + auto status = grid.catalogCache()->getDatabase(indexNSSQuery.db().toString()); + if (!status.isOK()) { + return false; + } - auto status = grid.catalogCache()->getDatabase(indexNSSQuery.db().toString()); - if (!status.isOK()) { - return false; - } + shared_ptr<DBConfig> config = status.getValue(); + if (!config->isSharded(indexNSSQuery.ns())) { + return false; + } - shared_ptr<DBConfig> config = status.getValue(); - if (!config->isSharded(indexNSSQuery.ns())) { - return false; - } + // if you are querying on system.indexes, we need to make sure we go to a shard + // that actually has chunks. This is not a perfect solution (what if you just + // look at all indexes), but better than doing nothing. - // if you are querying on system.indexes, we need to make sure we go to a shard - // that actually has chunks. This is not a perfect solution (what if you just - // look at all indexes), but better than doing nothing. - - ShardPtr shard; - ChunkManagerPtr cm; - config->getChunkManagerOrPrimary(indexNSSQuery.ns(), cm, shard); - if ( cm ) { - set<ShardId> shardIds; - cm->getAllShardIds(&shardIds); - verify(shardIds.size() > 0); - shard = grid.shardRegistry()->getShard(*shardIds.begin()); - } + ShardPtr shard; + ChunkManagerPtr cm; + config->getChunkManagerOrPrimary(indexNSSQuery.ns(), cm, shard); + if (cm) { + set<ShardId> shardIds; + cm->getAllShardIds(&shardIds); + verify(shardIds.size() > 0); + shard = grid.shardRegistry()->getShard(*shardIds.begin()); + } - ShardConnection dbcon(shard->getConnString(), r.getns()); - DBClientBase &c = dbcon.conn(); + ShardConnection dbcon(shard->getConnString(), r.getns()); + DBClientBase& c = dbcon.conn(); - string actualServer; + string actualServer; - Message response; - bool ok = c.call( r.m(), response, true , &actualServer ); - uassert( 10200 , "mongos: error calling db", ok ); + Message response; + bool ok = c.call(r.m(), response, true, &actualServer); + uassert(10200, "mongos: error calling db", ok); - { - QueryResult::View qr = response.singleData().view2ptr(); - if ( qr.getResultFlags() & ResultFlag_ShardConfigStale ) { - dbcon.done(); - // Version is zero b/c this is deprecated codepath - throw RecvStaleConfigException( r.getns(), - "Strategy::doQuery", - ChunkVersion( 0, 0, OID() ), - ChunkVersion( 0, 0, OID() )); - } + { + QueryResult::View qr = response.singleData().view2ptr(); + if (qr.getResultFlags() & ResultFlag_ShardConfigStale) { + dbcon.done(); + // Version is zero b/c this is deprecated codepath + throw RecvStaleConfigException(r.getns(), + "Strategy::doQuery", + ChunkVersion(0, 0, OID()), + ChunkVersion(0, 0, OID())); } - - r.reply( response , actualServer.size() ? actualServer : c.getServerAddress() ); - dbcon.done(); - - return true; } - void Strategy::queryOp( Request& r ) { + r.reply(response, actualServer.size() ? actualServer : c.getServerAddress()); + dbcon.done(); - verify( !NamespaceString( r.getns() ).isCommand() ); + return true; +} - Timer queryTimer; +void Strategy::queryOp(Request& r) { + verify(!NamespaceString(r.getns()).isCommand()); - QueryMessage q( r.d() ); + Timer queryTimer; - NamespaceString ns(q.ns); - ClientBasic* client = ClientBasic::getCurrent(); - AuthorizationSession* authSession = AuthorizationSession::get(client); - Status status = authSession->checkAuthForQuery(ns, q.query); - audit::logQueryAuthzCheck(client, ns, q.query, status.code()); - uassertStatusOK(status); + QueryMessage q(r.d()); - LOG(3) << "query: " << q.ns << " " << q.query << " ntoreturn: " << q.ntoreturn - << " options: " << q.queryOptions << endl; + NamespaceString ns(q.ns); + ClientBasic* client = ClientBasic::getCurrent(); + AuthorizationSession* authSession = AuthorizationSession::get(client); + Status status = authSession->checkAuthForQuery(ns, q.query); + audit::logQueryAuthzCheck(client, ns, q.query, status.code()); + uassertStatusOK(status); - if ( q.ntoreturn == 1 && strstr(q.ns, ".$cmd") ) - throw UserException( 8010 , "something is wrong, shouldn't see a command here" ); + LOG(3) << "query: " << q.ns << " " << q.query << " ntoreturn: " << q.ntoreturn + << " options: " << q.queryOptions << endl; - if (q.queryOptions & QueryOption_Exhaust) { - uasserted(18526, - string("the 'exhaust' query option is invalid for mongos queries: ") + q.ns - + " " + q.query.toString()); - } + if (q.ntoreturn == 1 && strstr(q.ns, ".$cmd")) + throw UserException(8010, "something is wrong, shouldn't see a command here"); - QuerySpec qSpec( (string)q.ns, q.query, q.fields, q.ntoskip, q.ntoreturn, q.queryOptions ); + if (q.queryOptions & QueryOption_Exhaust) { + uasserted(18526, + string("the 'exhaust' query option is invalid for mongos queries: ") + q.ns + + " " + q.query.toString()); + } - // Parse "$maxTimeMS". - StatusWith<int> maxTimeMS = LiteParsedQuery::parseMaxTimeMSQuery( q.query ); - uassert( 17233, - maxTimeMS.getStatus().reason(), - maxTimeMS.isOK() ); + QuerySpec qSpec((string)q.ns, q.query, q.fields, q.ntoskip, q.ntoreturn, q.queryOptions); - if ( _isSystemIndexes( q.ns ) && doShardedIndexQuery( r, qSpec )) { - return; - } + // Parse "$maxTimeMS". + StatusWith<int> maxTimeMS = LiteParsedQuery::parseMaxTimeMSQuery(q.query); + uassert(17233, maxTimeMS.getStatus().reason(), maxTimeMS.isOK()); - ParallelSortClusteredCursor * cursor = new ParallelSortClusteredCursor( qSpec, CommandInfo() ); - verify( cursor ); + if (_isSystemIndexes(q.ns) && doShardedIndexQuery(r, qSpec)) { + return; + } - // TODO: Move out to Request itself, not strategy based - try { - cursor->init(); - - if ( qSpec.isExplain() ) { - BSONObjBuilder explain_builder; - cursor->explain( explain_builder ); - explain_builder.appendNumber( "executionTimeMillis", - static_cast<long long>(queryTimer.millis()) ); - BSONObj b = explain_builder.obj(); - - replyToQuery( 0 , r.p() , r.m() , b ); - delete( cursor ); - return; - } - } - catch(...) { - delete cursor; - throw; - } + ParallelSortClusteredCursor* cursor = new ParallelSortClusteredCursor(qSpec, CommandInfo()); + verify(cursor); - // TODO: Revisit all of this when we revisit the sharded cursor cache + // TODO: Move out to Request itself, not strategy based + try { + cursor->init(); - if (cursor->getNumQueryShards() != 1) { + if (qSpec.isExplain()) { + BSONObjBuilder explain_builder; + cursor->explain(explain_builder); + explain_builder.appendNumber("executionTimeMillis", + static_cast<long long>(queryTimer.millis())); + BSONObj b = explain_builder.obj(); - // More than one shard (or zero), manage with a ShardedClientCursor - // NOTE: We may also have *zero* shards here when the returnPartial flag is set. - // Currently the code in ShardedClientCursor handles this. + replyToQuery(0, r.p(), r.m(), b); + delete (cursor); + return; + } + } catch (...) { + delete cursor; + throw; + } - ShardedClientCursorPtr cc (new ShardedClientCursor( q , cursor )); + // TODO: Revisit all of this when we revisit the sharded cursor cache - BufBuilder buffer( ShardedClientCursor::INIT_REPLY_BUFFER_SIZE ); - int docCount = 0; - const int startFrom = cc->getTotalSent(); - bool hasMore = cc->sendNextBatch(q.ntoreturn, buffer, docCount); + if (cursor->getNumQueryShards() != 1) { + // More than one shard (or zero), manage with a ShardedClientCursor + // NOTE: We may also have *zero* shards here when the returnPartial flag is set. + // Currently the code in ShardedClientCursor handles this. - if ( hasMore ) { - LOG(5) << "storing cursor : " << cc->getId() << endl; + ShardedClientCursorPtr cc(new ShardedClientCursor(q, cursor)); - int cursorLeftoverMillis = maxTimeMS.getValue() - queryTimer.millis(); - if ( maxTimeMS.getValue() == 0 ) { // 0 represents "no limit". - cursorLeftoverMillis = kMaxTimeCursorNoTimeLimit; - } - else if ( cursorLeftoverMillis <= 0 ) { - cursorLeftoverMillis = kMaxTimeCursorTimeLimitExpired; - } + BufBuilder buffer(ShardedClientCursor::INIT_REPLY_BUFFER_SIZE); + int docCount = 0; + const int startFrom = cc->getTotalSent(); + bool hasMore = cc->sendNextBatch(q.ntoreturn, buffer, docCount); + + if (hasMore) { + LOG(5) << "storing cursor : " << cc->getId() << endl; - cursorCache.store( cc, cursorLeftoverMillis ); + int cursorLeftoverMillis = maxTimeMS.getValue() - queryTimer.millis(); + if (maxTimeMS.getValue() == 0) { // 0 represents "no limit". + cursorLeftoverMillis = kMaxTimeCursorNoTimeLimit; + } else if (cursorLeftoverMillis <= 0) { + cursorLeftoverMillis = kMaxTimeCursorTimeLimitExpired; } - replyToQuery( 0, r.p(), r.m(), buffer.buf(), buffer.len(), docCount, - startFrom, hasMore ? cc->getId() : 0 ); + cursorCache.store(cc, cursorLeftoverMillis); } - else{ - // Only one shard is used + replyToQuery(0, + r.p(), + r.m(), + buffer.buf(), + buffer.len(), + docCount, + startFrom, + hasMore ? cc->getId() : 0); + } else { + // Only one shard is used - // Remote cursors are stored remotely, we shouldn't need this around. - unique_ptr<ParallelSortClusteredCursor> cursorDeleter( cursor ); + // Remote cursors are stored remotely, we shouldn't need this around. + unique_ptr<ParallelSortClusteredCursor> cursorDeleter(cursor); - ShardPtr shard = cursor->getQueryShard(); - verify( shard.get() ); - DBClientCursorPtr shardCursor = cursor->getShardCursor(shard->getId()); + ShardPtr shard = cursor->getQueryShard(); + verify(shard.get()); + DBClientCursorPtr shardCursor = cursor->getShardCursor(shard->getId()); - // Implicitly stores the cursor in the cache - r.reply( *(shardCursor->getMessage()) , shardCursor->originalHost() ); + // Implicitly stores the cursor in the cache + r.reply(*(shardCursor->getMessage()), shardCursor->originalHost()); - // We don't want to kill the cursor remotely if there's still data left - shardCursor->decouple(); - } + // We don't want to kill the cursor remotely if there's still data left + shardCursor->decouple(); } +} - void Strategy::clientCommandOp( Request& r ) { - QueryMessage q( r.d() ); +void Strategy::clientCommandOp(Request& r) { + QueryMessage q(r.d()); - LOG(3) << "command: " << q.ns << " " << q.query << " ntoreturn: " << q.ntoreturn - << " options: " << q.queryOptions << endl; + LOG(3) << "command: " << q.ns << " " << q.query << " ntoreturn: " << q.ntoreturn + << " options: " << q.queryOptions << endl; - if (q.queryOptions & QueryOption_Exhaust) { - uasserted(18527, - string("the 'exhaust' query option is invalid for mongos commands: ") + q.ns - + " " + q.query.toString()); - } + if (q.queryOptions & QueryOption_Exhaust) { + uasserted(18527, + string("the 'exhaust' query option is invalid for mongos commands: ") + q.ns + + " " + q.query.toString()); + } - NamespaceString nss( r.getns() ); - // Regular queries are handled in strategy_shard.cpp - verify( nss.isCommand() || nss.isSpecialCommand() ); + NamespaceString nss(r.getns()); + // Regular queries are handled in strategy_shard.cpp + verify(nss.isCommand() || nss.isSpecialCommand()); - if ( handleSpecialNamespaces( r , q ) ) - return; + if (handleSpecialNamespaces(r, q)) + return; - int loops = 5; - while ( true ) { - BSONObjBuilder builder; - try { - BSONObj cmdObj = q.query; - { - BSONElement e = cmdObj.firstElement(); - if (e.type() == Object && (e.fieldName()[0] == '$' - ? str::equals("query", e.fieldName()+1) - : str::equals("query", e.fieldName()))) { - // Extract the embedded query object. - - if (cmdObj.hasField(Query::ReadPrefField.name())) { - // The command has a read preference setting. We don't want - // to lose this information so we copy this to a new field - // called $queryOptions.$readPreference - BSONObjBuilder finalCmdObjBuilder; - finalCmdObjBuilder.appendElements(e.embeddedObject()); - - BSONObjBuilder queryOptionsBuilder( - finalCmdObjBuilder.subobjStart("$queryOptions")); - queryOptionsBuilder.append(cmdObj[Query::ReadPrefField.name()]); - queryOptionsBuilder.done(); - - cmdObj = finalCmdObjBuilder.obj(); - } - else { - cmdObj = e.embeddedObject(); - } + int loops = 5; + while (true) { + BSONObjBuilder builder; + try { + BSONObj cmdObj = q.query; + { + BSONElement e = cmdObj.firstElement(); + if (e.type() == Object && + (e.fieldName()[0] == '$' ? str::equals("query", e.fieldName() + 1) + : str::equals("query", e.fieldName()))) { + // Extract the embedded query object. + + if (cmdObj.hasField(Query::ReadPrefField.name())) { + // The command has a read preference setting. We don't want + // to lose this information so we copy this to a new field + // called $queryOptions.$readPreference + BSONObjBuilder finalCmdObjBuilder; + finalCmdObjBuilder.appendElements(e.embeddedObject()); + + BSONObjBuilder queryOptionsBuilder( + finalCmdObjBuilder.subobjStart("$queryOptions")); + queryOptionsBuilder.append(cmdObj[Query::ReadPrefField.name()]); + queryOptionsBuilder.done(); + + cmdObj = finalCmdObjBuilder.obj(); + } else { + cmdObj = e.embeddedObject(); } } - - Command::runAgainstRegistered(q.ns, cmdObj, builder, q.queryOptions); - BSONObj x = builder.done(); - replyToQuery(0, r.p(), r.m(), x); - return; } - catch ( StaleConfigException& e ) { - if ( loops <= 0 ) - throw e; - - loops--; - log() << "retrying command: " << q.query << endl; - // For legacy reasons, ns may not actually be set in the exception :-( - string staleNS = e.getns(); - if( staleNS.size() == 0 ) staleNS = q.ns; - - ShardConnection::checkMyConnectionVersions( staleNS ); - if( loops < 4 ) versionManager.forceRemoteCheckShardVersionCB( staleNS ); - } - catch ( AssertionException& e ) { - Command::appendCommandStatus(builder, e.toStatus()); - BSONObj x = builder.done(); - replyToQuery(0, r.p(), r.m(), x); - return; - } + Command::runAgainstRegistered(q.ns, cmdObj, builder, q.queryOptions); + BSONObj x = builder.done(); + replyToQuery(0, r.p(), r.m(), x); + return; + } catch (StaleConfigException& e) { + if (loops <= 0) + throw e; + + loops--; + log() << "retrying command: " << q.query << endl; + + // For legacy reasons, ns may not actually be set in the exception :-( + string staleNS = e.getns(); + if (staleNS.size() == 0) + staleNS = q.ns; + + ShardConnection::checkMyConnectionVersions(staleNS); + if (loops < 4) + versionManager.forceRemoteCheckShardVersionCB(staleNS); + } catch (AssertionException& e) { + Command::appendCommandStatus(builder, e.toStatus()); + BSONObj x = builder.done(); + replyToQuery(0, r.p(), r.m(), x); + return; } } +} - // TODO: remove after MongoDB 3.2 - bool Strategy::handleSpecialNamespaces( Request& r , QueryMessage& q ) { - const char * ns = strstr( r.getns() , ".$cmd.sys." ); - if ( ! ns ) - return false; - ns += 10; - - BSONObjBuilder reply; - - const auto upgradeToRealCommand = [&r, &q, &reply](StringData commandName) { - BSONObjBuilder cmdBob; - cmdBob.append(commandName, 1); - cmdBob.appendElements(q.query); // fields are validated by Commands - auto interposedCmd = cmdBob.done(); - NamespaceString nss(r.getns()); - NamespaceString interposedNss(nss.db(), "$cmd"); - Command::runAgainstRegistered(interposedNss.ns().c_str(), - interposedCmd, - reply, - q.queryOptions); - }; - - if ( strcmp( ns , "inprog" ) == 0 ) { - upgradeToRealCommand("currentOp"); - } - else if ( strcmp( ns , "killop" ) == 0 ) { - upgradeToRealCommand("killOp"); - } - else if ( strcmp( ns , "unlock" ) == 0 ) { - reply.append( "err" , "can't do unlock through mongos" ); - } - else { - warning() << "unknown sys command [" << ns << "]" << endl; - return false; - } - - BSONObj x = reply.done(); - replyToQuery(0, r.p(), r.m(), x); - return true; +// TODO: remove after MongoDB 3.2 +bool Strategy::handleSpecialNamespaces(Request& r, QueryMessage& q) { + const char* ns = strstr(r.getns(), ".$cmd.sys."); + if (!ns) + return false; + ns += 10; + + BSONObjBuilder reply; + + const auto upgradeToRealCommand = [&r, &q, &reply](StringData commandName) { + BSONObjBuilder cmdBob; + cmdBob.append(commandName, 1); + cmdBob.appendElements(q.query); // fields are validated by Commands + auto interposedCmd = cmdBob.done(); + NamespaceString nss(r.getns()); + NamespaceString interposedNss(nss.db(), "$cmd"); + Command::runAgainstRegistered( + interposedNss.ns().c_str(), interposedCmd, reply, q.queryOptions); + }; + + if (strcmp(ns, "inprog") == 0) { + upgradeToRealCommand("currentOp"); + } else if (strcmp(ns, "killop") == 0) { + upgradeToRealCommand("killOp"); + } else if (strcmp(ns, "unlock") == 0) { + reply.append("err", "can't do unlock through mongos"); + } else { + warning() << "unknown sys command [" << ns << "]" << endl; + return false; } - void Strategy::commandOp( const string& db, - const BSONObj& command, - int options, - const string& versionedNS, - const BSONObj& targetingQuery, - vector<CommandResult>* results ) - { - - QuerySpec qSpec(db + ".$cmd", command, BSONObj(), 0, 1, options); + BSONObj x = reply.done(); + replyToQuery(0, r.p(), r.m(), x); + return true; +} - ParallelSortClusteredCursor cursor( qSpec, CommandInfo( versionedNS, targetingQuery ) ); +void Strategy::commandOp(const string& db, + const BSONObj& command, + int options, + const string& versionedNS, + const BSONObj& targetingQuery, + vector<CommandResult>* results) { + QuerySpec qSpec(db + ".$cmd", command, BSONObj(), 0, 1, options); - // Initialize the cursor - cursor.init(); + ParallelSortClusteredCursor cursor(qSpec, CommandInfo(versionedNS, targetingQuery)); - set<ShardId> shardIds; - cursor.getQueryShardIds(shardIds); - - for (const ShardId& shardId : shardIds) { - CommandResult result; - result.shardTargetId = shardId; - - string errMsg; // ignored, should never be invalid b/c an exception thrown earlier - result.target = - ConnectionString::parse( cursor.getShardCursor(shardId)->originalHost(), - errMsg ); - result.result = cursor.getShardCursor(shardId)->peekFirst().getOwned(); - results->push_back( result ); - } + // Initialize the cursor + cursor.init(); - } + set<ShardId> shardIds; + cursor.getQueryShardIds(shardIds); - Status Strategy::commandOpWrite(const std::string& dbName, - const BSONObj& command, - BatchItemRef targetingBatchItem, - std::vector<CommandResult>* results) { + for (const ShardId& shardId : shardIds) { + CommandResult result; + result.shardTargetId = shardId; - // Note that this implementation will not handle targeting retries and does not completely - // emulate write behavior + string errMsg; // ignored, should never be invalid b/c an exception thrown earlier + result.target = + ConnectionString::parse(cursor.getShardCursor(shardId)->originalHost(), errMsg); + result.result = cursor.getShardCursor(shardId)->peekFirst().getOwned(); + results->push_back(result); + } +} - ChunkManagerTargeter targeter(NamespaceString( - targetingBatchItem.getRequest()->getTargetingNS())); - Status status = targeter.init(); +Status Strategy::commandOpWrite(const std::string& dbName, + const BSONObj& command, + BatchItemRef targetingBatchItem, + std::vector<CommandResult>* results) { + // Note that this implementation will not handle targeting retries and does not completely + // emulate write behavior + + ChunkManagerTargeter targeter( + NamespaceString(targetingBatchItem.getRequest()->getTargetingNS())); + Status status = targeter.init(); + if (!status.isOK()) + return status; + + OwnedPointerVector<ShardEndpoint> endpointsOwned; + vector<ShardEndpoint*>& endpoints = endpointsOwned.mutableVector(); + + if (targetingBatchItem.getOpType() == BatchedCommandRequest::BatchType_Insert) { + ShardEndpoint* endpoint; + Status status = targeter.targetInsert(targetingBatchItem.getDocument(), &endpoint); + if (!status.isOK()) + return status; + endpoints.push_back(endpoint); + } else if (targetingBatchItem.getOpType() == BatchedCommandRequest::BatchType_Update) { + Status status = targeter.targetUpdate(*targetingBatchItem.getUpdate(), &endpoints); if (!status.isOK()) return status; + } else { + invariant(targetingBatchItem.getOpType() == BatchedCommandRequest::BatchType_Delete); + Status status = targeter.targetDelete(*targetingBatchItem.getDelete(), &endpoints); + if (!status.isOK()) + return status; + } - OwnedPointerVector<ShardEndpoint> endpointsOwned; - vector<ShardEndpoint*>& endpoints = endpointsOwned.mutableVector(); + DBClientShardResolver resolver; + DBClientMultiCommand dispatcher; - if (targetingBatchItem.getOpType() == BatchedCommandRequest::BatchType_Insert) { - ShardEndpoint* endpoint; - Status status = targeter.targetInsert(targetingBatchItem.getDocument(), &endpoint); - if (!status.isOK()) - return status; - endpoints.push_back(endpoint); - } - else if (targetingBatchItem.getOpType() == BatchedCommandRequest::BatchType_Update) { - Status status = targeter.targetUpdate(*targetingBatchItem.getUpdate(), &endpoints); - if (!status.isOK()) - return status; - } - else { - invariant(targetingBatchItem.getOpType() == BatchedCommandRequest::BatchType_Delete); - Status status = targeter.targetDelete(*targetingBatchItem.getDelete(), &endpoints); - if (!status.isOK()) - return status; - } + // Assemble requests + for (vector<ShardEndpoint*>::const_iterator it = endpoints.begin(); it != endpoints.end(); + ++it) { + const ShardEndpoint* endpoint = *it; - DBClientShardResolver resolver; - DBClientMultiCommand dispatcher; + ConnectionString host; + Status status = resolver.chooseWriteHost(endpoint->shardName, &host); + if (!status.isOK()) + return status; - // Assemble requests - for (vector<ShardEndpoint*>::const_iterator it = endpoints.begin(); it != endpoints.end(); - ++it) { + RawBSONSerializable request(command); + dispatcher.addCommand(host, dbName, request); + } - const ShardEndpoint* endpoint = *it; + // Errors reported when recv'ing responses + dispatcher.sendAll(); + Status dispatchStatus = Status::OK(); - ConnectionString host; - Status status = resolver.chooseWriteHost(endpoint->shardName, &host); - if (!status.isOK()) - return status; + // Recv responses + while (dispatcher.numPending() > 0) { + ConnectionString host; + RawBSONSerializable response; - RawBSONSerializable request(command); - dispatcher.addCommand(host, dbName, request); + Status status = dispatcher.recvAny(&host, &response); + if (!status.isOK()) { + // We always need to recv() all the sent operations + dispatchStatus = status; + continue; } - // Errors reported when recv'ing responses - dispatcher.sendAll(); - Status dispatchStatus = Status::OK(); - - // Recv responses - while (dispatcher.numPending() > 0) { - - ConnectionString host; - RawBSONSerializable response; - - Status status = dispatcher.recvAny(&host, &response); - if (!status.isOK()) { - // We always need to recv() all the sent operations - dispatchStatus = status; - continue; - } - - CommandResult result; - result.target = host; - { - const auto shard = grid.shardRegistry()->getShard(host.toString()); - result.shardTargetId = shard->getId(); - } - result.result = response.toBSON(); - - results->push_back(result); + CommandResult result; + result.target = host; + { + const auto shard = grid.shardRegistry()->getShard(host.toString()); + result.shardTargetId = shard->getId(); } + result.result = response.toBSON(); - return dispatchStatus; + results->push_back(result); } - Status Strategy::commandOpUnsharded(const std::string& db, - const BSONObj& command, - int options, - const std::string& versionedNS, - CommandResult* cmdResult) { + return dispatchStatus; +} - // Note that this implementation will not handle targeting retries and fails when the - // sharding metadata is too stale - auto status = grid.catalogCache()->getDatabase(db); - if (!status.isOK()) { - mongoutils::str::stream ss; - ss << "Passthrough command failed: " << command.toString() - << " on ns " << versionedNS << ". Caused by " << causedBy(status.getStatus()); - return Status(ErrorCodes::IllegalOperation, ss); - } +Status Strategy::commandOpUnsharded(const std::string& db, + const BSONObj& command, + int options, + const std::string& versionedNS, + CommandResult* cmdResult) { + // Note that this implementation will not handle targeting retries and fails when the + // sharding metadata is too stale + auto status = grid.catalogCache()->getDatabase(db); + if (!status.isOK()) { + mongoutils::str::stream ss; + ss << "Passthrough command failed: " << command.toString() << " on ns " << versionedNS + << ". Caused by " << causedBy(status.getStatus()); + return Status(ErrorCodes::IllegalOperation, ss); + } - shared_ptr<DBConfig> conf = status.getValue(); - if (conf->isSharded(versionedNS)) { - mongoutils::str::stream ss; - ss << "Passthrough command failed: " << command.toString() - << " on ns " << versionedNS << ". Cannot run on sharded namespace."; - return Status(ErrorCodes::IllegalOperation, ss); - } + shared_ptr<DBConfig> conf = status.getValue(); + if (conf->isSharded(versionedNS)) { + mongoutils::str::stream ss; + ss << "Passthrough command failed: " << command.toString() << " on ns " << versionedNS + << ". Cannot run on sharded namespace."; + return Status(ErrorCodes::IllegalOperation, ss); + } - const auto primaryShard = grid.shardRegistry()->getShard(conf->getPrimaryId()); + const auto primaryShard = grid.shardRegistry()->getShard(conf->getPrimaryId()); - BSONObj shardResult; - try { - ShardConnection conn(primaryShard->getConnString(), ""); - - // TODO: this can throw a stale config when mongos is not up-to-date -- fix. - if (!conn->runCommand(db, command, shardResult, options)) { - conn.done(); - return Status(ErrorCodes::OperationFailed, - str::stream() << "Passthrough command failed: " << command - << " on ns " << versionedNS - << "; result: " << shardResult); - } + BSONObj shardResult; + try { + ShardConnection conn(primaryShard->getConnString(), ""); + + // TODO: this can throw a stale config when mongos is not up-to-date -- fix. + if (!conn->runCommand(db, command, shardResult, options)) { conn.done(); + return Status(ErrorCodes::OperationFailed, + str::stream() << "Passthrough command failed: " << command << " on ns " + << versionedNS << "; result: " << shardResult); } - catch (const DBException& ex) { - return ex.toStatus(); - } - - // Fill out the command result. - cmdResult->shardTargetId = conf->getPrimaryId(); - cmdResult->result = shardResult; - cmdResult->target = primaryShard->getConnString(); - - return Status::OK(); + conn.done(); + } catch (const DBException& ex) { + return ex.toStatus(); } - void Strategy::getMore( Request& r ) { - Timer getMoreTimer; - - const char* ns = r.getns(); - const int ntoreturn = r.d().pullInt(); - const long long id = r.d().pullInt64(); + // Fill out the command result. + cmdResult->shardTargetId = conf->getPrimaryId(); + cmdResult->result = shardResult; + cmdResult->target = primaryShard->getConnString(); - // TODO: Handle stale config exceptions here from coll being dropped or sharded during op - // for now has same semantics as legacy request - const NamespaceString nss(ns); - auto statusGetDb = grid.catalogCache()->getDatabase(nss.db().toString()); - if (statusGetDb == ErrorCodes::DatabaseNotFound) { - cursorCache.remove(id); - replyToQuery(ResultFlag_CursorNotFound, r.p(), r.m(), 0, 0, 0); - return; - } + return Status::OK(); +} - uassertStatusOK(statusGetDb); +void Strategy::getMore(Request& r) { + Timer getMoreTimer; + + const char* ns = r.getns(); + const int ntoreturn = r.d().pullInt(); + const long long id = r.d().pullInt64(); + + // TODO: Handle stale config exceptions here from coll being dropped or sharded during op + // for now has same semantics as legacy request + const NamespaceString nss(ns); + auto statusGetDb = grid.catalogCache()->getDatabase(nss.db().toString()); + if (statusGetDb == ErrorCodes::DatabaseNotFound) { + cursorCache.remove(id); + replyToQuery(ResultFlag_CursorNotFound, r.p(), r.m(), 0, 0, 0); + return; + } - shared_ptr<DBConfig> config = statusGetDb.getValue(); + uassertStatusOK(statusGetDb); - ShardPtr primary; - ChunkManagerPtr info; - config->getChunkManagerOrPrimary( ns, info, primary ); + shared_ptr<DBConfig> config = statusGetDb.getValue(); - // - // TODO: Cleanup cursor cache, consolidate into single codepath - // - const string host = cursorCache.getRef(id); - ShardedClientCursorPtr cursor = cursorCache.get( id ); - int cursorMaxTimeMS = cursorCache.getMaxTimeMS( id ); + ShardPtr primary; + ChunkManagerPtr info; + config->getChunkManagerOrPrimary(ns, info, primary); - // Cursor ids should not overlap between sharded and unsharded cursors - massert( 17012, str::stream() << "duplicate sharded and unsharded cursor id " - << id << " detected for " << ns - << ", duplicated on host " << host, - NULL == cursorCache.get( id ).get() || host.empty() ); + // + // TODO: Cleanup cursor cache, consolidate into single codepath + // + const string host = cursorCache.getRef(id); + ShardedClientCursorPtr cursor = cursorCache.get(id); + int cursorMaxTimeMS = cursorCache.getMaxTimeMS(id); - ClientBasic* client = ClientBasic::getCurrent(); - NamespaceString nsString(ns); - AuthorizationSession* authSession = AuthorizationSession::get(client); - Status status = authSession->checkAuthForGetMore( nsString, id ); - audit::logGetMoreAuthzCheck( client, nsString, id, status.code() ); - uassertStatusOK(status); + // Cursor ids should not overlap between sharded and unsharded cursors + massert(17012, + str::stream() << "duplicate sharded and unsharded cursor id " << id << " detected for " + << ns << ", duplicated on host " << host, + NULL == cursorCache.get(id).get() || host.empty()); - if( !host.empty() ){ + ClientBasic* client = ClientBasic::getCurrent(); + NamespaceString nsString(ns); + AuthorizationSession* authSession = AuthorizationSession::get(client); + Status status = authSession->checkAuthForGetMore(nsString, id); + audit::logGetMoreAuthzCheck(client, nsString, id, status.code()); + uassertStatusOK(status); - LOG(3) << "single getmore: " << ns << endl; + if (!host.empty()) { + LOG(3) << "single getmore: " << ns << endl; - // we used ScopedDbConnection because we don't get about config versions - // not deleting data is handled elsewhere - // and we don't want to call setShardVersion - ScopedDbConnection conn(host); + // we used ScopedDbConnection because we don't get about config versions + // not deleting data is handled elsewhere + // and we don't want to call setShardVersion + ScopedDbConnection conn(host); - Message response; - bool ok = conn->callRead( r.m() , response); - uassert( 10204 , "dbgrid: getmore: error calling db", ok); + Message response; + bool ok = conn->callRead(r.m(), response); + uassert(10204, "dbgrid: getmore: error calling db", ok); - bool hasMore = (response.singleData().getCursor() != 0); + bool hasMore = (response.singleData().getCursor() != 0); - if ( !hasMore ) { - cursorCache.removeRef( id ); - } + if (!hasMore) { + cursorCache.removeRef(id); + } - r.reply( response , "" /*conn->getServerAddress() */ ); - conn.done(); - return; + r.reply(response, "" /*conn->getServerAddress() */); + conn.done(); + return; + } else if (cursor) { + if (cursorMaxTimeMS == kMaxTimeCursorTimeLimitExpired) { + cursorCache.remove(id); + uasserted(ErrorCodes::ExceededTimeLimit, "operation exceeded time limit"); } - else if ( cursor ) { - if ( cursorMaxTimeMS == kMaxTimeCursorTimeLimitExpired ) { - cursorCache.remove( id ); - uasserted( ErrorCodes::ExceededTimeLimit, "operation exceeded time limit" ); - } + // TODO: Try to match logic of mongod, where on subsequent getMore() we pull lots more data? + BufBuilder buffer(ShardedClientCursor::INIT_REPLY_BUFFER_SIZE); + int docCount = 0; + const int startFrom = cursor->getTotalSent(); + bool hasMore = cursor->sendNextBatch(ntoreturn, buffer, docCount); - // TODO: Try to match logic of mongod, where on subsequent getMore() we pull lots more data? - BufBuilder buffer( ShardedClientCursor::INIT_REPLY_BUFFER_SIZE ); - int docCount = 0; - const int startFrom = cursor->getTotalSent(); - bool hasMore = cursor->sendNextBatch(ntoreturn, buffer, docCount); - - if ( hasMore ) { - // still more data - cursor->accessed(); - - if ( cursorMaxTimeMS != kMaxTimeCursorNoTimeLimit ) { - // Update remaining amount of time in cursor cache. - int cursorLeftoverMillis = cursorMaxTimeMS - getMoreTimer.millis(); - if ( cursorLeftoverMillis <= 0 ) { - cursorLeftoverMillis = kMaxTimeCursorTimeLimitExpired; - } - cursorCache.updateMaxTimeMS( id, cursorLeftoverMillis ); + if (hasMore) { + // still more data + cursor->accessed(); + + if (cursorMaxTimeMS != kMaxTimeCursorNoTimeLimit) { + // Update remaining amount of time in cursor cache. + int cursorLeftoverMillis = cursorMaxTimeMS - getMoreTimer.millis(); + if (cursorLeftoverMillis <= 0) { + cursorLeftoverMillis = kMaxTimeCursorTimeLimitExpired; } + cursorCache.updateMaxTimeMS(id, cursorLeftoverMillis); } - else { - // we've exhausted the cursor - cursorCache.remove( id ); - } - - replyToQuery( 0, r.p(), r.m(), buffer.buf(), buffer.len(), docCount, - startFrom, hasMore ? cursor->getId() : 0 ); - return; + } else { + // we've exhausted the cursor + cursorCache.remove(id); } - else { - - LOG( 3 ) << "could not find cursor " << id << " in cache for " << ns << endl; - replyToQuery( ResultFlag_CursorNotFound , r.p() , r.m() , 0 , 0 , 0 ); - return; - } + replyToQuery(0, + r.p(), + r.m(), + buffer.buf(), + buffer.len(), + docCount, + startFrom, + hasMore ? cursor->getId() : 0); + return; + } else { + LOG(3) << "could not find cursor " << id << " in cache for " << ns << endl; + + replyToQuery(ResultFlag_CursorNotFound, r.p(), r.m(), 0, 0, 0); + return; } +} - void Strategy::writeOp( int op , Request& r ) { - - // make sure we have a last error - dassert(&LastError::get(cc())); - - OwnedPointerVector<BatchedCommandRequest> requestsOwned; - vector<BatchedCommandRequest*>& requests = requestsOwned.mutableVector(); +void Strategy::writeOp(int op, Request& r) { + // make sure we have a last error + dassert(&LastError::get(cc())); - msgToBatchRequests( r.m(), &requests ); + OwnedPointerVector<BatchedCommandRequest> requestsOwned; + vector<BatchedCommandRequest*>& requests = requestsOwned.mutableVector(); - for ( vector<BatchedCommandRequest*>::iterator it = requests.begin(); - it != requests.end(); ++it ) { + msgToBatchRequests(r.m(), &requests); - // Multiple commands registered to last error as multiple requests - if ( it != requests.begin() ) - LastError::get(cc()).startRequest(); + for (vector<BatchedCommandRequest*>::iterator it = requests.begin(); it != requests.end(); + ++it) { + // Multiple commands registered to last error as multiple requests + if (it != requests.begin()) + LastError::get(cc()).startRequest(); - BatchedCommandRequest* request = *it; + BatchedCommandRequest* request = *it; - // Adjust namespaces for command - NamespaceString fullNS( request->getNS() ); - string cmdNS = fullNS.getCommandNS(); - // We only pass in collection name to command - request->setNS( fullNS.coll() ); + // Adjust namespaces for command + NamespaceString fullNS(request->getNS()); + string cmdNS = fullNS.getCommandNS(); + // We only pass in collection name to command + request->setNS(fullNS.coll()); - BSONObjBuilder builder; - BSONObj requestBSON = request->toBSON(); + BSONObjBuilder builder; + BSONObj requestBSON = request->toBSON(); - { - // Disable the last error object for the duration of the write cmd - LastError::Disabled disableLastError(&LastError::get(cc())); - Command::runAgainstRegistered( cmdNS.c_str(), requestBSON, builder, 0 ); - } + { + // Disable the last error object for the duration of the write cmd + LastError::Disabled disableLastError(&LastError::get(cc())); + Command::runAgainstRegistered(cmdNS.c_str(), requestBSON, builder, 0); + } - BatchedCommandResponse response; - bool parsed = response.parseBSON( builder.done(), NULL ); - (void) parsed; // for compile - dassert( parsed && response.isValid( NULL ) ); + BatchedCommandResponse response; + bool parsed = response.parseBSON(builder.done(), NULL); + (void)parsed; // for compile + dassert(parsed && response.isValid(NULL)); - // Populate the lastError object based on the write response - LastError::get(cc()).reset(); - bool hadError = batchErrorToLastError(*request, response, &LastError::get(cc())); + // Populate the lastError object based on the write response + LastError::get(cc()).reset(); + bool hadError = batchErrorToLastError(*request, response, &LastError::get(cc())); - // Check if this is an ordered batch and we had an error which should stop processing - if ( request->getOrdered() && hadError ) - break; - } + // Check if this is an ordered batch and we had an error which should stop processing + if (request->getOrdered() && hadError) + break; } - +} } diff --git a/src/mongo/s/strategy.h b/src/mongo/s/strategy.h index 9d599a3f7a6..eb77b6afe42 100644 --- a/src/mongo/s/strategy.h +++ b/src/mongo/s/strategy.h @@ -35,83 +35,78 @@ namespace mongo { - class BatchItemRef; +class BatchItemRef; - /** - * Legacy interface for processing client read/write/cmd requests. - */ - class Strategy { - public: +/** + * Legacy interface for processing client read/write/cmd requests. + */ +class Strategy { +public: + static void queryOp(Request& r); - static void queryOp(Request& r); + static void getMore(Request& r); - static void getMore(Request& r); + static void writeOp(int op, Request& r); - static void writeOp(int op , Request& r); + struct CommandResult { + ShardId shardTargetId; + ConnectionString target; + BSONObj result; + }; - struct CommandResult { - ShardId shardTargetId; - ConnectionString target; - BSONObj result; - }; + /** + * Executes a command against a particular database, and targets the command based on a + * collection in that database. + * + * This version should be used by internal commands when possible. + * + * TODO: Replace these methods and all other methods of command dispatch with a more general + * command op framework. + */ + static void commandOp(const std::string& db, + const BSONObj& command, + int options, + const std::string& versionedNS, + const BSONObj& targetingQuery, + std::vector<CommandResult>* results); - /** - * Executes a command against a particular database, and targets the command based on a - * collection in that database. - * - * This version should be used by internal commands when possible. - * - * TODO: Replace these methods and all other methods of command dispatch with a more general - * command op framework. - */ - static void commandOp(const std::string& db, - const BSONObj& command, - int options, - const std::string& versionedNS, - const BSONObj& targetingQuery, - std::vector<CommandResult>* results); + /** + * Executes a write command against a particular database, and targets the command based on + * a write operation. + * + * Does *not* retry or retarget if the metadata is stale. + * + * Similar to commandOp() above, but the targeting rules are different for writes than for + * reads. + */ + static Status commandOpWrite(const std::string& db, + const BSONObj& command, + BatchItemRef targetingBatchItem, + std::vector<CommandResult>* results); - /** - * Executes a write command against a particular database, and targets the command based on - * a write operation. - * - * Does *not* retry or retarget if the metadata is stale. - * - * Similar to commandOp() above, but the targeting rules are different for writes than for - * reads. - */ - static Status commandOpWrite(const std::string& db, + /** + * Some commands can only be run in a sharded configuration against a namespace that has + * not been sharded. Use this method to execute such commands. + * + * Does *not* retry or retarget if the metadata is stale. + * + * On success, fills in 'shardResult' with output from the namespace's primary shard. This + * output may itself indicate an error status on the shard. + */ + static Status commandOpUnsharded(const std::string& db, const BSONObj& command, - BatchItemRef targetingBatchItem, - std::vector<CommandResult>* results); - - /** - * Some commands can only be run in a sharded configuration against a namespace that has - * not been sharded. Use this method to execute such commands. - * - * Does *not* retry or retarget if the metadata is stale. - * - * On success, fills in 'shardResult' with output from the namespace's primary shard. This - * output may itself indicate an error status on the shard. - */ - static Status commandOpUnsharded(const std::string& db, - const BSONObj& command, - int options, - const std::string& versionedNS, - CommandResult* shardResult); - - /** - * Executes a command represented in the Request on the sharded cluster. - * - * DEPRECATED: should not be used by new code. - */ - static void clientCommandOp( Request& r ); - - protected: - - static bool handleSpecialNamespaces( Request& r , QueryMessage& q ); + int options, + const std::string& versionedNS, + CommandResult* shardResult); - }; + /** + * Executes a command represented in the Request on the sharded cluster. + * + * DEPRECATED: should not be used by new code. + */ + static void clientCommandOp(Request& r); +protected: + static bool handleSpecialNamespaces(Request& r, QueryMessage& q); +}; } - diff --git a/src/mongo/s/type_config_version.cpp b/src/mongo/s/type_config_version.cpp index 8f9c47f8d77..78a920904ba 100644 --- a/src/mongo/s/type_config_version.cpp +++ b/src/mongo/s/type_config_version.cpp @@ -33,159 +33,164 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; +using mongoutils::str::stream; - const std::string VersionType::ConfigNS = "config.version"; +const std::string VersionType::ConfigNS = "config.version"; - const BSONField<int> VersionType::minCompatibleVersion("minCompatibleVersion"); - const BSONField<int> VersionType::currentVersion("currentVersion"); - const BSONField<BSONArray> VersionType::excludingMongoVersions("excluding"); - const BSONField<OID> VersionType::clusterId("clusterId"); - const BSONField<OID> VersionType::upgradeId("upgradeId"); - const BSONField<BSONObj> VersionType::upgradeState("upgradeState"); +const BSONField<int> VersionType::minCompatibleVersion("minCompatibleVersion"); +const BSONField<int> VersionType::currentVersion("currentVersion"); +const BSONField<BSONArray> VersionType::excludingMongoVersions("excluding"); +const BSONField<OID> VersionType::clusterId("clusterId"); +const BSONField<OID> VersionType::upgradeId("upgradeId"); +const BSONField<BSONObj> VersionType::upgradeState("upgradeState"); - VersionType::VersionType() { - clear(); - } +VersionType::VersionType() { + clear(); +} - VersionType::~VersionType() { - } +VersionType::~VersionType() {} - bool VersionType::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isMinCompatibleVersionSet) { - *errMsg = stream() << "missing " << minCompatibleVersion.name() << " field"; - return false; - } - if (!_isCurrentVersionSet) { - *errMsg = stream() << "missing " << currentVersion.name() << " field"; - return false; - } - - // Hardcoded 3 here because it's the last version without a cluster id - if (_currentVersion > 3 && !_isClusterIdSet) { - *errMsg = stream() << "missing " << clusterId.name() << " field"; - return false; - } - - return true; +bool VersionType::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BSONObj VersionType::toBSON() const { - BSONObjBuilder builder; - - builder.append("_id", 1); - if (_isMinCompatibleVersionSet) { - builder.append(minCompatibleVersion(), _minCompatibleVersion); - } - if (_isCurrentVersionSet) builder.append(currentVersion(), _currentVersion); - if (_isExcludingMongoVersionsSet) { - builder.append(excludingMongoVersions(), _excludingMongoVersions); - } - if (_isClusterIdSet) builder.append(clusterId(), _clusterId); - if (_isUpgradeIdSet) { - builder.append(upgradeId(), _upgradeId); - builder.append(upgradeState(), _upgradeState); - } - - return builder.obj(); + // All the mandatory fields must be present. + if (!_isMinCompatibleVersionSet) { + *errMsg = stream() << "missing " << minCompatibleVersion.name() << " field"; + return false; + } + if (!_isCurrentVersionSet) { + *errMsg = stream() << "missing " << currentVersion.name() << " field"; + return false; } - bool VersionType::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + // Hardcoded 3 here because it's the last version without a cluster id + if (_currentVersion > 3 && !_isClusterIdSet) { + *errMsg = stream() << "missing " << clusterId.name() << " field"; + return false; + } - std::string dummy; - if (!errMsg) errMsg = &dummy; + return true; +} - FieldParser::FieldState fieldState; - fieldState = FieldParser::extractNumber(source, minCompatibleVersion, - &_minCompatibleVersion, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isMinCompatibleVersionSet = fieldState == FieldParser::FIELD_SET; +BSONObj VersionType::toBSON() const { + BSONObjBuilder builder; - fieldState = FieldParser::extractNumber(source, currentVersion, &_currentVersion, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isCurrentVersionSet = fieldState == FieldParser::FIELD_SET; + builder.append("_id", 1); + if (_isMinCompatibleVersionSet) { + builder.append(minCompatibleVersion(), _minCompatibleVersion); + } + if (_isCurrentVersionSet) + builder.append(currentVersion(), _currentVersion); + if (_isExcludingMongoVersionsSet) { + builder.append(excludingMongoVersions(), _excludingMongoVersions); + } + if (_isClusterIdSet) + builder.append(clusterId(), _clusterId); + if (_isUpgradeIdSet) { + builder.append(upgradeId(), _upgradeId); + builder.append(upgradeState(), _upgradeState); + } - if (!_isCurrentVersionSet && _isMinCompatibleVersionSet) { - _currentVersion = _minCompatibleVersion; - _isCurrentVersionSet = true; - } + return builder.obj(); +} - fieldState = FieldParser::extract(source, excludingMongoVersions, - &_excludingMongoVersions, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isExcludingMongoVersionsSet = fieldState == FieldParser::FIELD_SET; +bool VersionType::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - fieldState = FieldParser::extract(source, clusterId, &_clusterId, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isClusterIdSet = fieldState == FieldParser::FIELD_SET; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - fieldState = FieldParser::extract(source, upgradeId, &_upgradeId, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isUpgradeIdSet = fieldState == FieldParser::FIELD_SET; + FieldParser::FieldState fieldState; + fieldState = + FieldParser::extractNumber(source, minCompatibleVersion, &_minCompatibleVersion, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isMinCompatibleVersionSet = fieldState == FieldParser::FIELD_SET; - fieldState = FieldParser::extract(source, upgradeState, &_upgradeState, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isUpgradeStateSet = fieldState == FieldParser::FIELD_SET; + fieldState = FieldParser::extractNumber(source, currentVersion, &_currentVersion, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isCurrentVersionSet = fieldState == FieldParser::FIELD_SET; - return true; + if (!_isCurrentVersionSet && _isMinCompatibleVersionSet) { + _currentVersion = _minCompatibleVersion; + _isCurrentVersionSet = true; } - void VersionType::clear() { + fieldState = + FieldParser::extract(source, excludingMongoVersions, &_excludingMongoVersions, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isExcludingMongoVersionsSet = fieldState == FieldParser::FIELD_SET; - _minCompatibleVersion = -1; - _isMinCompatibleVersionSet = false; + fieldState = FieldParser::extract(source, clusterId, &_clusterId, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isClusterIdSet = fieldState == FieldParser::FIELD_SET; - _currentVersion = -1; - _isCurrentVersionSet = false; + fieldState = FieldParser::extract(source, upgradeId, &_upgradeId, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isUpgradeIdSet = fieldState == FieldParser::FIELD_SET; - _excludingMongoVersions = BSONArray(); - _isExcludingMongoVersionsSet = false; + fieldState = FieldParser::extract(source, upgradeState, &_upgradeState, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isUpgradeStateSet = fieldState == FieldParser::FIELD_SET; - _clusterId = OID(); - _isClusterIdSet = false; + return true; +} - _upgradeId = OID(); - _isUpgradeIdSet = false; +void VersionType::clear() { + _minCompatibleVersion = -1; + _isMinCompatibleVersionSet = false; - _upgradeState = BSONObj(); - _isUpgradeStateSet = false; + _currentVersion = -1; + _isCurrentVersionSet = false; - } + _excludingMongoVersions = BSONArray(); + _isExcludingMongoVersionsSet = false; - void VersionType::cloneTo(VersionType* other) const { - other->clear(); + _clusterId = OID(); + _isClusterIdSet = false; - other->_minCompatibleVersion = _minCompatibleVersion; - other->_isMinCompatibleVersionSet = _isMinCompatibleVersionSet; + _upgradeId = OID(); + _isUpgradeIdSet = false; - other->_currentVersion = _currentVersion; - other->_isCurrentVersionSet = _isCurrentVersionSet; + _upgradeState = BSONObj(); + _isUpgradeStateSet = false; +} - other->_excludingMongoVersions = _excludingMongoVersions; - other->_isExcludingMongoVersionsSet = _isExcludingMongoVersionsSet; +void VersionType::cloneTo(VersionType* other) const { + other->clear(); - other->_clusterId = _clusterId; - other->_isClusterIdSet = _isClusterIdSet; + other->_minCompatibleVersion = _minCompatibleVersion; + other->_isMinCompatibleVersionSet = _isMinCompatibleVersionSet; - other->_upgradeId = _upgradeId; - other->_isUpgradeIdSet = _isUpgradeIdSet; + other->_currentVersion = _currentVersion; + other->_isCurrentVersionSet = _isCurrentVersionSet; - other->_upgradeState = _upgradeState; - other->_isUpgradeStateSet = _isUpgradeStateSet; + other->_excludingMongoVersions = _excludingMongoVersions; + other->_isExcludingMongoVersionsSet = _isExcludingMongoVersionsSet; - } + other->_clusterId = _clusterId; + other->_isClusterIdSet = _isClusterIdSet; - std::string VersionType::toString() const { - return toBSON().toString(); - } + other->_upgradeId = _upgradeId; + other->_isUpgradeIdSet = _isUpgradeIdSet; + + other->_upgradeState = _upgradeState; + other->_isUpgradeStateSet = _isUpgradeStateSet; +} + +std::string VersionType::toString() const { + return toBSON().toString(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/type_config_version.h b/src/mongo/s/type_config_version.h index 778efab059f..c98bf531234 100644 --- a/src/mongo/s/type_config_version.h +++ b/src/mongo/s/type_config_version.h @@ -36,215 +36,233 @@ namespace mongo { - /** - * This class represents the layout and contents of documents contained in the - * config.version collection. All manipulation of documents coming from that - * collection should be done with this class. - * - * Usage Example: - * - * // Contact the config. 'conn' has been obtained before. - * DBClientBase* conn; - * BSONObj query = QUERY(VersionType::exampleField("exampleFieldName")); - * exampleDoc = conn->findOne(VersionType::ConfigNS, query); - * - * // Process the response. - * VersionType exampleType; - * std::string errMsg; - * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { - * // Can't use 'exampleType'. Take action. - * } - * // use 'exampleType' - * - */ - class VersionType { - MONGO_DISALLOW_COPYING(VersionType); - public: - - // - // schema declarations - // - - // Name of the version collection in the config server. - static const std::string ConfigNS; - - // Field names and types in the version collection type. - static const BSONField<int> minCompatibleVersion; - static const BSONField<int> currentVersion; - static const BSONField<BSONArray> excludingMongoVersions; - static const BSONField<OID> clusterId; - static const BSONField<OID> upgradeId; - static const BSONField<BSONObj> upgradeState; - - // - // version type methods - // - - VersionType(); - ~VersionType(); - - /** - * Returns true if all the mandatory fields are present and have valid - * representations. Otherwise returns false and fills in the optional 'errMsg' string. - */ - bool isValid(std::string* errMsg) const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Clears and populates the internal state using the 'source' BSON object if the - * latter contains valid values. Otherwise sets errMsg and returns false. - */ - bool parseBSON(const BSONObj& source, std::string* errMsg); - - /** - * Clears the internal state. - */ - void clear(); - - /** - * Copies all the fields present in 'this' to 'other'. - */ - void cloneTo(VersionType* other) const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - // - // individual field accessors - // - - // Mandatory Fields - void setMinCompatibleVersion(const int minCompatibleVersion) { - _minCompatibleVersion = minCompatibleVersion; - _isMinCompatibleVersionSet = true; - } - - void unsetMinCompatibleVersion() { _isMinCompatibleVersionSet = false; } - - bool isMinCompatibleVersionSet() const { return _isMinCompatibleVersionSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - int getMinCompatibleVersion() const { - dassert(_isMinCompatibleVersionSet); - return _minCompatibleVersion; - } - - void setCurrentVersion(const int currentVersion) { - _currentVersion = currentVersion; - _isCurrentVersionSet = true; - } - - void unsetCurrentVersion() { _isCurrentVersionSet = false; } - - bool isCurrentVersionSet() const { return _isCurrentVersionSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - int getCurrentVersion() const { - dassert(_isCurrentVersionSet); - return _currentVersion; - } - - void setExcludingMongoVersions(const BSONArray& excludingMongoVersions) { - _excludingMongoVersions = excludingMongoVersions; - _isExcludingMongoVersionsSet = true; - } +/** + * This class represents the layout and contents of documents contained in the + * config.version collection. All manipulation of documents coming from that + * collection should be done with this class. + * + * Usage Example: + * + * // Contact the config. 'conn' has been obtained before. + * DBClientBase* conn; + * BSONObj query = QUERY(VersionType::exampleField("exampleFieldName")); + * exampleDoc = conn->findOne(VersionType::ConfigNS, query); + * + * // Process the response. + * VersionType exampleType; + * std::string errMsg; + * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { + * // Can't use 'exampleType'. Take action. + * } + * // use 'exampleType' + * + */ +class VersionType { + MONGO_DISALLOW_COPYING(VersionType); - void unsetExcludingMongoVersions() { _isExcludingMongoVersionsSet = false; } +public: + // + // schema declarations + // - bool isExcludingMongoVersionsSet() const { - return _isExcludingMongoVersionsSet || excludingMongoVersions.hasDefault(); - } + // Name of the version collection in the config server. + static const std::string ConfigNS; - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - const BSONArray getExcludingMongoVersions() const { - if (_isExcludingMongoVersionsSet) { - return _excludingMongoVersions; - } else { - dassert(excludingMongoVersions.hasDefault()); - return excludingMongoVersions.getDefault(); - } - } + // Field names and types in the version collection type. + static const BSONField<int> minCompatibleVersion; + static const BSONField<int> currentVersion; + static const BSONField<BSONArray> excludingMongoVersions; + static const BSONField<OID> clusterId; + static const BSONField<OID> upgradeId; + static const BSONField<BSONObj> upgradeState; - void setClusterId(const OID clusterId) { - _clusterId = clusterId; - _isClusterIdSet = true; - } + // + // version type methods + // - void unsetClusterId() { _isClusterIdSet = false; } + VersionType(); + ~VersionType(); - bool isClusterIdSet() const { return _isClusterIdSet; } + /** + * Returns true if all the mandatory fields are present and have valid + * representations. Otherwise returns false and fills in the optional 'errMsg' string. + */ + bool isValid(std::string* errMsg) const; - // Calling get*() methods when the member is not set results in undefined behavior - const OID getClusterId() const { - dassert(_isClusterIdSet); - return _clusterId; - } + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; - // Optional Fields - void setUpgradeId(OID upgradeId) { - _upgradeId = upgradeId; - _isUpgradeIdSet = true; - } + /** + * Clears and populates the internal state using the 'source' BSON object if the + * latter contains valid values. Otherwise sets errMsg and returns false. + */ + bool parseBSON(const BSONObj& source, std::string* errMsg); - void unsetUpgradeId() { _isUpgradeIdSet = false; } + /** + * Clears the internal state. + */ + void clear(); - bool isUpgradeIdSet() const { - return _isUpgradeIdSet || upgradeId.hasDefault(); - } + /** + * Copies all the fields present in 'this' to 'other'. + */ + void cloneTo(VersionType* other) const; - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - OID getUpgradeId() const { - if (_isUpgradeIdSet) { - return _upgradeId; - } else { - dassert(upgradeId.hasDefault()); - return upgradeId.getDefault(); - } - } - void setUpgradeState(const BSONObj& upgradeState) { - _upgradeState = upgradeState.getOwned(); - _isUpgradeStateSet = true; + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; + + // + // individual field accessors + // + + // Mandatory Fields + void setMinCompatibleVersion(const int minCompatibleVersion) { + _minCompatibleVersion = minCompatibleVersion; + _isMinCompatibleVersionSet = true; + } + + void unsetMinCompatibleVersion() { + _isMinCompatibleVersionSet = false; + } + + bool isMinCompatibleVersionSet() const { + return _isMinCompatibleVersionSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + int getMinCompatibleVersion() const { + dassert(_isMinCompatibleVersionSet); + return _minCompatibleVersion; + } + + void setCurrentVersion(const int currentVersion) { + _currentVersion = currentVersion; + _isCurrentVersionSet = true; + } + + void unsetCurrentVersion() { + _isCurrentVersionSet = false; + } + + bool isCurrentVersionSet() const { + return _isCurrentVersionSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + int getCurrentVersion() const { + dassert(_isCurrentVersionSet); + return _currentVersion; + } + + void setExcludingMongoVersions(const BSONArray& excludingMongoVersions) { + _excludingMongoVersions = excludingMongoVersions; + _isExcludingMongoVersionsSet = true; + } + + void unsetExcludingMongoVersions() { + _isExcludingMongoVersionsSet = false; + } + + bool isExcludingMongoVersionsSet() const { + return _isExcludingMongoVersionsSet || excludingMongoVersions.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + const BSONArray getExcludingMongoVersions() const { + if (_isExcludingMongoVersionsSet) { + return _excludingMongoVersions; + } else { + dassert(excludingMongoVersions.hasDefault()); + return excludingMongoVersions.getDefault(); } - - void unsetUpgradeState() { _isUpgradeStateSet = false; } - - bool isUpgradeStateSet() const { - return _isUpgradeStateSet || upgradeState.hasDefault(); + } + + void setClusterId(const OID clusterId) { + _clusterId = clusterId; + _isClusterIdSet = true; + } + + void unsetClusterId() { + _isClusterIdSet = false; + } + + bool isClusterIdSet() const { + return _isClusterIdSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const OID getClusterId() const { + dassert(_isClusterIdSet); + return _clusterId; + } + + // Optional Fields + void setUpgradeId(OID upgradeId) { + _upgradeId = upgradeId; + _isUpgradeIdSet = true; + } + + void unsetUpgradeId() { + _isUpgradeIdSet = false; + } + + bool isUpgradeIdSet() const { + return _isUpgradeIdSet || upgradeId.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + OID getUpgradeId() const { + if (_isUpgradeIdSet) { + return _upgradeId; + } else { + dassert(upgradeId.hasDefault()); + return upgradeId.getDefault(); } - - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - BSONObj getUpgradeState() const { - if (_isUpgradeStateSet) { - return _upgradeState; - } else { - dassert(upgradeState.hasDefault()); - return upgradeState.getDefault(); - } + } + void setUpgradeState(const BSONObj& upgradeState) { + _upgradeState = upgradeState.getOwned(); + _isUpgradeStateSet = true; + } + + void unsetUpgradeState() { + _isUpgradeStateSet = false; + } + + bool isUpgradeStateSet() const { + return _isUpgradeStateSet || upgradeState.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + BSONObj getUpgradeState() const { + if (_isUpgradeStateSet) { + return _upgradeState; + } else { + dassert(upgradeState.hasDefault()); + return upgradeState.getDefault(); } - - private: - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - int _minCompatibleVersion; // (M) minimum compatible version - bool _isMinCompatibleVersionSet; - int _currentVersion; // (M) current version - bool _isCurrentVersionSet; - BSONArray _excludingMongoVersions; // (O) range of disallowed versions to upgrade to - bool _isExcludingMongoVersionsSet; - OID _clusterId; // (M) clusterId - bool _isClusterIdSet; - OID _upgradeId; // (O) upgrade id of current or last upgrade - bool _isUpgradeIdSet; - BSONObj _upgradeState; // (O) upgrade state of current or last upgrade - bool _isUpgradeStateSet; - }; - -} // namespace mongo + } + +private: + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + int _minCompatibleVersion; // (M) minimum compatible version + bool _isMinCompatibleVersionSet; + int _currentVersion; // (M) current version + bool _isCurrentVersionSet; + BSONArray _excludingMongoVersions; // (O) range of disallowed versions to upgrade to + bool _isExcludingMongoVersionsSet; + OID _clusterId; // (M) clusterId + bool _isClusterIdSet; + OID _upgradeId; // (O) upgrade id of current or last upgrade + bool _isUpgradeIdSet; + BSONObj _upgradeState; // (O) upgrade state of current or last upgrade + bool _isUpgradeStateSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/type_config_version_test.cpp b/src/mongo/s/type_config_version_test.cpp index 6c4c8ea6a3a..618238402b9 100644 --- a/src/mongo/s/type_config_version_test.cpp +++ b/src/mongo/s/type_config_version_test.cpp @@ -37,131 +37,127 @@ namespace { - using std::string; - using mongo::VersionType; - using mongo::BSONObj; - using mongo::BSONArray; - using mongo::BSONObjBuilder; - using mongo::BSONArrayBuilder; - using mongo::OID; - - TEST(Validity, Empty) { - - // - // Tests parsing of empty document - // - - VersionType versionInfo; - BSONObj emptyObj = BSONObj(); - - // parses ok - string errMsg; - bool result = versionInfo.parseBSON(emptyObj, &errMsg); - ASSERT_EQUALS(errMsg, ""); - ASSERT(result); - - // not valid - result = versionInfo.isValid(&errMsg); - ASSERT_NOT_EQUALS(errMsg, ""); - ASSERT(!result); - } - - TEST(Validity, NewVersion) { - - // - // Tests parsing a new-style config version - // - - VersionType versionInfo; - - OID clusterId = OID::gen(); - - BSONObjBuilder bob; - bob << VersionType::minCompatibleVersion(3); - bob << VersionType::currentVersion(4); - bob << VersionType::clusterId(clusterId); - - BSONObj versionDoc = bob.obj(); - - string errMsg; - bool result = versionInfo.parseBSON(versionDoc, &errMsg); - ASSERT_EQUALS(errMsg, ""); - ASSERT(result); - ASSERT_EQUALS(versionInfo.getMinCompatibleVersion(), 3); - ASSERT_EQUALS(versionInfo.getCurrentVersion(), 4); - ASSERT_EQUALS(versionInfo.getClusterId(), clusterId); - - // Valid - result = versionInfo.isValid(&errMsg); - ASSERT_EQUALS(errMsg, ""); - ASSERT(result); - } - - TEST(Validity, NewVersionRoundTrip) { - - // - // Round-trip - // - - VersionType versionInfo; - - OID clusterId = OID::gen(); - OID upgradeId = OID::gen(); - BSONObj upgradeState = BSON("a" << 1); - - BSONObjBuilder bob; - bob << VersionType::minCompatibleVersion(3); - bob << VersionType::currentVersion(4); - bob << VersionType::clusterId(clusterId); - bob << VersionType::upgradeId(upgradeId); - bob << VersionType::upgradeState(upgradeState); - - BSONObj versionDoc = bob.obj(); - - string errMsg; - bool result = versionInfo.parseBSON(versionDoc, &errMsg); - ASSERT_EQUALS(errMsg, ""); - ASSERT(result); - - result = versionInfo.parseBSON(versionInfo.toBSON(), &errMsg); - ASSERT_EQUALS(errMsg, ""); - ASSERT(result); - ASSERT_EQUALS(versionInfo.getMinCompatibleVersion(), 3); - ASSERT_EQUALS(versionInfo.getCurrentVersion(), 4); - ASSERT_EQUALS(versionInfo.getClusterId(), clusterId); - ASSERT_EQUALS(versionInfo.getUpgradeId(), upgradeId); - ASSERT_EQUALS(versionInfo.getUpgradeState(), upgradeState); - - // Valid - result = versionInfo.isValid(&errMsg); - ASSERT_EQUALS(errMsg, ""); - ASSERT(result); - } - - TEST(Validity, NewVersionNoClusterId) { - - // - // Tests error on parsing new format with no clusterId - // - - VersionType versionInfo; - - BSONObjBuilder bob; - bob << VersionType::minCompatibleVersion(3); - bob << VersionType::currentVersion(4); - - BSONObj versionDoc = bob.obj(); - - // Parses ok - string errMsg; - bool result = versionInfo.parseBSON(versionDoc, &errMsg); - ASSERT_EQUALS(errMsg, ""); - ASSERT(result); - - // Not valid - result = versionInfo.isValid(&errMsg); - ASSERT_NOT_EQUALS(errMsg, ""); - ASSERT(!result); - } - -} // unnamed namespace +using std::string; +using mongo::VersionType; +using mongo::BSONObj; +using mongo::BSONArray; +using mongo::BSONObjBuilder; +using mongo::BSONArrayBuilder; +using mongo::OID; + +TEST(Validity, Empty) { + // + // Tests parsing of empty document + // + + VersionType versionInfo; + BSONObj emptyObj = BSONObj(); + + // parses ok + string errMsg; + bool result = versionInfo.parseBSON(emptyObj, &errMsg); + ASSERT_EQUALS(errMsg, ""); + ASSERT(result); + + // not valid + result = versionInfo.isValid(&errMsg); + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(!result); +} + +TEST(Validity, NewVersion) { + // + // Tests parsing a new-style config version + // + + VersionType versionInfo; + + OID clusterId = OID::gen(); + + BSONObjBuilder bob; + bob << VersionType::minCompatibleVersion(3); + bob << VersionType::currentVersion(4); + bob << VersionType::clusterId(clusterId); + + BSONObj versionDoc = bob.obj(); + + string errMsg; + bool result = versionInfo.parseBSON(versionDoc, &errMsg); + ASSERT_EQUALS(errMsg, ""); + ASSERT(result); + ASSERT_EQUALS(versionInfo.getMinCompatibleVersion(), 3); + ASSERT_EQUALS(versionInfo.getCurrentVersion(), 4); + ASSERT_EQUALS(versionInfo.getClusterId(), clusterId); + + // Valid + result = versionInfo.isValid(&errMsg); + ASSERT_EQUALS(errMsg, ""); + ASSERT(result); +} + +TEST(Validity, NewVersionRoundTrip) { + // + // Round-trip + // + + VersionType versionInfo; + + OID clusterId = OID::gen(); + OID upgradeId = OID::gen(); + BSONObj upgradeState = BSON("a" << 1); + + BSONObjBuilder bob; + bob << VersionType::minCompatibleVersion(3); + bob << VersionType::currentVersion(4); + bob << VersionType::clusterId(clusterId); + bob << VersionType::upgradeId(upgradeId); + bob << VersionType::upgradeState(upgradeState); + + BSONObj versionDoc = bob.obj(); + + string errMsg; + bool result = versionInfo.parseBSON(versionDoc, &errMsg); + ASSERT_EQUALS(errMsg, ""); + ASSERT(result); + + result = versionInfo.parseBSON(versionInfo.toBSON(), &errMsg); + ASSERT_EQUALS(errMsg, ""); + ASSERT(result); + ASSERT_EQUALS(versionInfo.getMinCompatibleVersion(), 3); + ASSERT_EQUALS(versionInfo.getCurrentVersion(), 4); + ASSERT_EQUALS(versionInfo.getClusterId(), clusterId); + ASSERT_EQUALS(versionInfo.getUpgradeId(), upgradeId); + ASSERT_EQUALS(versionInfo.getUpgradeState(), upgradeState); + + // Valid + result = versionInfo.isValid(&errMsg); + ASSERT_EQUALS(errMsg, ""); + ASSERT(result); +} + +TEST(Validity, NewVersionNoClusterId) { + // + // Tests error on parsing new format with no clusterId + // + + VersionType versionInfo; + + BSONObjBuilder bob; + bob << VersionType::minCompatibleVersion(3); + bob << VersionType::currentVersion(4); + + BSONObj versionDoc = bob.obj(); + + // Parses ok + string errMsg; + bool result = versionInfo.parseBSON(versionDoc, &errMsg); + ASSERT_EQUALS(errMsg, ""); + ASSERT(result); + + // Not valid + result = versionInfo.isValid(&errMsg); + ASSERT_NOT_EQUALS(errMsg, ""); + ASSERT(!result); +} + +} // unnamed namespace diff --git a/src/mongo/s/type_lockpings.cpp b/src/mongo/s/type_lockpings.cpp index 47beaea1485..c24d09a3506 100644 --- a/src/mongo/s/type_lockpings.cpp +++ b/src/mongo/s/type_lockpings.cpp @@ -32,91 +32,92 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; +using mongoutils::str::stream; - const std::string LockpingsType::ConfigNS = "config.lockpings"; +const std::string LockpingsType::ConfigNS = "config.lockpings"; - const BSONField<std::string> LockpingsType::process("_id"); - const BSONField<Date_t> LockpingsType::ping("ping"); +const BSONField<std::string> LockpingsType::process("_id"); +const BSONField<Date_t> LockpingsType::ping("ping"); - LockpingsType::LockpingsType() { - clear(); - } +LockpingsType::LockpingsType() { + clear(); +} - LockpingsType::~LockpingsType() { - } +LockpingsType::~LockpingsType() {} - bool LockpingsType::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isProcessSet) { - *errMsg = stream() << "missing " << process.name() << " field"; - return false; - } - if (!_isPingSet) { - *errMsg = stream() << "missing " << ping.name() << " field"; - return false; - } - - return true; +bool LockpingsType::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BSONObj LockpingsType::toBSON() const { - BSONObjBuilder builder; - - if (_isProcessSet) builder.append(process(), _process); - if (_isPingSet) builder.append(ping(), _ping); - - return builder.obj(); + // All the mandatory fields must be present. + if (!_isProcessSet) { + *errMsg = stream() << "missing " << process.name() << " field"; + return false; + } + if (!_isPingSet) { + *errMsg = stream() << "missing " << ping.name() << " field"; + return false; } - bool LockpingsType::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + return true; +} - std::string dummy; - if (!errMsg) errMsg = &dummy; +BSONObj LockpingsType::toBSON() const { + BSONObjBuilder builder; - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, process, &_process, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isProcessSet = fieldState == FieldParser::FIELD_SET; + if (_isProcessSet) + builder.append(process(), _process); + if (_isPingSet) + builder.append(ping(), _ping); - fieldState = FieldParser::extract(source, ping, &_ping, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isPingSet = fieldState == FieldParser::FIELD_SET; + return builder.obj(); +} - return true; - } +bool LockpingsType::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - void LockpingsType::clear() { + std::string dummy; + if (!errMsg) + errMsg = &dummy; - _process.clear(); - _isProcessSet = false; + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, process, &_process, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isProcessSet = fieldState == FieldParser::FIELD_SET; - _ping = Date_t(); - _isPingSet = false; + fieldState = FieldParser::extract(source, ping, &_ping, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isPingSet = fieldState == FieldParser::FIELD_SET; - } + return true; +} - void LockpingsType::cloneTo(LockpingsType* other) const { - other->clear(); +void LockpingsType::clear() { + _process.clear(); + _isProcessSet = false; - other->_process = _process; - other->_isProcessSet = _isProcessSet; + _ping = Date_t(); + _isPingSet = false; +} - other->_ping = _ping; - other->_isPingSet = _isPingSet; +void LockpingsType::cloneTo(LockpingsType* other) const { + other->clear(); - } + other->_process = _process; + other->_isProcessSet = _isProcessSet; - std::string LockpingsType::toString() const { - return toBSON().toString(); - } + other->_ping = _ping; + other->_isPingSet = _isPingSet; +} + +std::string LockpingsType::toString() const { + return toBSON().toString(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/type_lockpings.h b/src/mongo/s/type_lockpings.h index 8197da6b9ed..845b544d206 100644 --- a/src/mongo/s/type_lockpings.h +++ b/src/mongo/s/type_lockpings.h @@ -35,131 +35,130 @@ namespace mongo { +/** + * This class represents the layout and contents of documents contained in the + * config.lockpings collection. All manipulation of documents coming from that + * collection should be done with this class. + * + * Usage Example: + * + * // Contact the config. 'conn' has been obtained before. + * DBClientBase* conn; + * BSONObj query = QUERY(LockpingsType::exampleField("exampleFieldName")); + * exampleDoc = conn->findOne(LockpingsType::ConfigNS, query); + * + * // Process the response. + * LockpingsType exampleType; + * std::string errMsg; + * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { + * // Can't use 'exampleType'. Take action. + * } + * // use 'exampleType' + * + */ +class LockpingsType { +public: + // + // schema declarations + // + + // Name of the lockpings collection in the config server. + static const std::string ConfigNS; + + // Field names and types in the lockpings collection type. + static const BSONField<std::string> process; + static const BSONField<Date_t> ping; + + // + // lockpings type methods + // + + LockpingsType(); + ~LockpingsType(); + + /** + * Returns true if all the mandatory fields are present and have valid + * representations. Otherwise returns false and fills in the optional 'errMsg' string. + */ + bool isValid(std::string* errMsg) const; + + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; + + /** + * Clears and populates the internal state using the 'source' BSON object if the + * latter contains valid values. Otherwise sets errMsg and returns false. + */ + bool parseBSON(const BSONObj& source, std::string* errMsg); + + /** + * Clears the internal state. + */ + void clear(); + + /** + * Copies all the fields present in 'this' to 'other'. + */ + void cloneTo(LockpingsType* other) const; + /** - * This class represents the layout and contents of documents contained in the - * config.lockpings collection. All manipulation of documents coming from that - * collection should be done with this class. - * - * Usage Example: - * - * // Contact the config. 'conn' has been obtained before. - * DBClientBase* conn; - * BSONObj query = QUERY(LockpingsType::exampleField("exampleFieldName")); - * exampleDoc = conn->findOne(LockpingsType::ConfigNS, query); - * - * // Process the response. - * LockpingsType exampleType; - * std::string errMsg; - * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { - * // Can't use 'exampleType'. Take action. - * } - * // use 'exampleType' - * + * Returns a std::string representation of the current internal state. */ - class LockpingsType { - public: - - // - // schema declarations - // - - // Name of the lockpings collection in the config server. - static const std::string ConfigNS; - - // Field names and types in the lockpings collection type. - static const BSONField<std::string> process; - static const BSONField<Date_t> ping; - - // - // lockpings type methods - // - - LockpingsType(); - ~LockpingsType(); - - /** - * Returns true if all the mandatory fields are present and have valid - * representations. Otherwise returns false and fills in the optional 'errMsg' string. - */ - bool isValid(std::string* errMsg) const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Clears and populates the internal state using the 'source' BSON object if the - * latter contains valid values. Otherwise sets errMsg and returns false. - */ - bool parseBSON(const BSONObj& source, std::string* errMsg); - - /** - * Clears the internal state. - */ - void clear(); - - /** - * Copies all the fields present in 'this' to 'other'. - */ - void cloneTo(LockpingsType* other) const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - // - // individual field accessors - // - - // Mandatory Fields - void setProcess(StringData process) { - _process = process.toString(); - _isProcessSet = true; - } - - void unsetProcess() { - _isProcessSet = false; - } - - bool isProcessSet() const { - return _isProcessSet; - } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getProcess() const { - dassert(_isProcessSet); - return _process; - } - - void setPing(const Date_t ping) { - _ping = ping; - _isPingSet = true; - } - - void unsetPing() { - _isPingSet = false; - } - - bool isPingSet() const { - return _isPingSet; - } - - // Calling get*() methods when the member is not set results in undefined behavior - const Date_t getPing() const { - dassert(_isPingSet); - return _ping; - } - - // Optional Fields - - private: - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - std::string _process; // (M) std::string describing the process holding the lock - bool _isProcessSet; - Date_t _ping; // (M) last time the holding process updated this document - bool _isPingSet; - }; - -} // namespace mongo + std::string toString() const; + + // + // individual field accessors + // + + // Mandatory Fields + void setProcess(StringData process) { + _process = process.toString(); + _isProcessSet = true; + } + + void unsetProcess() { + _isProcessSet = false; + } + + bool isProcessSet() const { + return _isProcessSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getProcess() const { + dassert(_isProcessSet); + return _process; + } + + void setPing(const Date_t ping) { + _ping = ping; + _isPingSet = true; + } + + void unsetPing() { + _isPingSet = false; + } + + bool isPingSet() const { + return _isPingSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const Date_t getPing() const { + dassert(_isPingSet); + return _ping; + } + + // Optional Fields + +private: + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + std::string _process; // (M) std::string describing the process holding the lock + bool _isProcessSet; + Date_t _ping; // (M) last time the holding process updated this document + bool _isPingSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/type_lockpings_test.cpp b/src/mongo/s/type_lockpings_test.cpp index 487a8714e2c..1bf6c29048a 100644 --- a/src/mongo/s/type_lockpings_test.cpp +++ b/src/mongo/s/type_lockpings_test.cpp @@ -32,42 +32,42 @@ namespace { - using std::string; - using mongo::BSONObj; - using mongo::LockpingsType; - using mongo::Date_t; +using std::string; +using mongo::BSONObj; +using mongo::LockpingsType; +using mongo::Date_t; - TEST(Validity, MissingFields) { - LockpingsType lockPing; - BSONObj objNoProcess = BSON(LockpingsType::ping(Date_t::now())); - string errMsg; - ASSERT(lockPing.parseBSON(objNoProcess, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lockPing.isValid(NULL)); +TEST(Validity, MissingFields) { + LockpingsType lockPing; + BSONObj objNoProcess = BSON(LockpingsType::ping(Date_t::now())); + string errMsg; + ASSERT(lockPing.parseBSON(objNoProcess, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lockPing.isValid(NULL)); - BSONObj objNoPing = BSON(LockpingsType::process("host.local:27017:1352918870:16807")); - ASSERT(lockPing.parseBSON(objNoPing, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lockPing.isValid(NULL)); - } + BSONObj objNoPing = BSON(LockpingsType::process("host.local:27017:1352918870:16807")); + ASSERT(lockPing.parseBSON(objNoPing, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lockPing.isValid(NULL)); +} - TEST(Validity, Valid) { - LockpingsType lockPing; - BSONObj obj = BSON(LockpingsType::process("host.local:27017:1352918870:16807") << - LockpingsType::ping(Date_t::fromMillisSinceEpoch(1))); - string errMsg; - ASSERT(lockPing.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(lockPing.isValid(NULL)); - ASSERT_EQUALS(lockPing.getProcess(), "host.local:27017:1352918870:16807"); - ASSERT_EQUALS(lockPing.getPing(), Date_t::fromMillisSinceEpoch(1)); - } +TEST(Validity, Valid) { + LockpingsType lockPing; + BSONObj obj = BSON(LockpingsType::process("host.local:27017:1352918870:16807") + << LockpingsType::ping(Date_t::fromMillisSinceEpoch(1))); + string errMsg; + ASSERT(lockPing.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(lockPing.isValid(NULL)); + ASSERT_EQUALS(lockPing.getProcess(), "host.local:27017:1352918870:16807"); + ASSERT_EQUALS(lockPing.getPing(), Date_t::fromMillisSinceEpoch(1)); +} - TEST(Validity, BadType) { - LockpingsType lockPing; - BSONObj obj = BSON(LockpingsType::process() << 0); - string errMsg; - ASSERT((!lockPing.parseBSON(obj, &errMsg)) && (errMsg != "")); - } +TEST(Validity, BadType) { + LockpingsType lockPing; + BSONObj obj = BSON(LockpingsType::process() << 0); + string errMsg; + ASSERT((!lockPing.parseBSON(obj, &errMsg)) && (errMsg != "")); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/type_locks.cpp b/src/mongo/s/type_locks.cpp index 4fd16b636fb..af4ccda0f83 100644 --- a/src/mongo/s/type_locks.cpp +++ b/src/mongo/s/type_locks.cpp @@ -32,163 +32,172 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; +using mongoutils::str::stream; - const std::string LocksType::ConfigNS = "config.locks"; +const std::string LocksType::ConfigNS = "config.locks"; - const BSONField<std::string> LocksType::name("_id"); - const BSONField<int> LocksType::state("state"); - const BSONField<std::string> LocksType::process("process"); - const BSONField<OID> LocksType::lockID("ts"); - const BSONField<std::string> LocksType::who("who"); - const BSONField<std::string> LocksType::why("why"); - const BSONField<Date_t> LocksType::when("when"); +const BSONField<std::string> LocksType::name("_id"); +const BSONField<int> LocksType::state("state"); +const BSONField<std::string> LocksType::process("process"); +const BSONField<OID> LocksType::lockID("ts"); +const BSONField<std::string> LocksType::who("who"); +const BSONField<std::string> LocksType::why("why"); +const BSONField<Date_t> LocksType::when("when"); - LocksType::LocksType() { - clear(); +LocksType::LocksType() { + clear(); +} + +LocksType::~LocksType() {} + +bool LocksType::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - LocksType::~LocksType() { + // All the mandatory fields must be present. + if (!_isNameSet) { + *errMsg = stream() << "missing " << name.name() << " field"; + return false; + } + if (!_isStateSet) { + *errMsg = stream() << "missing " << state.name() << " field"; + return false; } - bool LocksType::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; + // If the lock is locked check the remaining fields + if (_state != UNLOCKED) { + if (!_isProcessSet) { + *errMsg = stream() << "missing " << process.name() << " field"; + return false; } - // All the mandatory fields must be present. - if (!_isNameSet) { - *errMsg = stream() << "missing " << name.name() << " field"; + if (!_isLockIDSet) { + *errMsg = stream() << "missing " << lockID.name() << " field"; return false; } - if (!_isStateSet) { - *errMsg = stream() << "missing " << state.name() << " field"; + + if (!_isWhoSet) { + *errMsg = stream() << "missing " << who.name() << " field"; return false; } - // If the lock is locked check the remaining fields - if (_state != UNLOCKED) { - if (!_isProcessSet) { - *errMsg = stream() << "missing " << process.name() << " field"; - return false; - } - - if (!_isLockIDSet) { - *errMsg = stream() << "missing " << lockID.name() << " field"; - return false; - } - - if (!_isWhoSet) { - *errMsg = stream() << "missing " << who.name() << " field"; - return false; - } - - if (!_isWhySet) { - *errMsg = stream() << "missing " << why.name() << " field"; - return false; - } + if (!_isWhySet) { + *errMsg = stream() << "missing " << why.name() << " field"; + return false; } - - return true; } - BSONObj LocksType::toBSON() const { - BSONObjBuilder builder; + return true; +} - if (_isNameSet) builder.append(name(), _name); - if (_isStateSet) builder.append(state(), _state); - if (_isProcessSet) builder.append(process(), _process); - if (_isLockIDSet) builder.append(lockID(), _lockID); - if (_isWhoSet) builder.append(who(), _who); - if (_isWhySet) builder.append(why(), _why); +BSONObj LocksType::toBSON() const { + BSONObjBuilder builder; - return builder.obj(); - } - - bool LocksType::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + if (_isNameSet) + builder.append(name(), _name); + if (_isStateSet) + builder.append(state(), _state); + if (_isProcessSet) + builder.append(process(), _process); + if (_isLockIDSet) + builder.append(lockID(), _lockID); + if (_isWhoSet) + builder.append(who(), _who); + if (_isWhySet) + builder.append(why(), _why); - std::string dummy; - if (!errMsg) errMsg = &dummy; + return builder.obj(); +} - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, name, &_name, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isNameSet = fieldState == FieldParser::FIELD_SET; +bool LocksType::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - fieldState = FieldParser::extract(source, state, &_state, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isStateSet = fieldState == FieldParser::FIELD_SET; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - fieldState = FieldParser::extract(source, process, &_process, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isProcessSet = fieldState == FieldParser::FIELD_SET; + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, name, &_name, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isNameSet = fieldState == FieldParser::FIELD_SET; - fieldState = FieldParser::extract(source, lockID, &_lockID, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isLockIDSet = fieldState == FieldParser::FIELD_SET; + fieldState = FieldParser::extract(source, state, &_state, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isStateSet = fieldState == FieldParser::FIELD_SET; - fieldState = FieldParser::extract(source, who, &_who, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isWhoSet = fieldState == FieldParser::FIELD_SET; + fieldState = FieldParser::extract(source, process, &_process, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isProcessSet = fieldState == FieldParser::FIELD_SET; - fieldState = FieldParser::extract(source, why, &_why, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isWhySet = fieldState == FieldParser::FIELD_SET; + fieldState = FieldParser::extract(source, lockID, &_lockID, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isLockIDSet = fieldState == FieldParser::FIELD_SET; - return true; - } + fieldState = FieldParser::extract(source, who, &_who, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isWhoSet = fieldState == FieldParser::FIELD_SET; - void LocksType::clear() { + fieldState = FieldParser::extract(source, why, &_why, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isWhySet = fieldState == FieldParser::FIELD_SET; - _name.clear(); - _isNameSet = false; + return true; +} - _state = UNLOCKED; - _isStateSet = false; +void LocksType::clear() { + _name.clear(); + _isNameSet = false; - _process.clear(); - _isProcessSet = false; + _state = UNLOCKED; + _isStateSet = false; - _lockID = OID(); - _isLockIDSet = false; + _process.clear(); + _isProcessSet = false; - _who.clear(); - _isWhoSet = false; + _lockID = OID(); + _isLockIDSet = false; - _why.clear(); - _isWhySet = false; + _who.clear(); + _isWhoSet = false; - } + _why.clear(); + _isWhySet = false; +} - void LocksType::cloneTo(LocksType* other) const { - other->clear(); +void LocksType::cloneTo(LocksType* other) const { + other->clear(); - other->_name = _name; - other->_isNameSet = _isNameSet; + other->_name = _name; + other->_isNameSet = _isNameSet; - other->_state = _state; - other->_isStateSet = _isStateSet; + other->_state = _state; + other->_isStateSet = _isStateSet; - other->_process = _process; - other->_isProcessSet = _isProcessSet; + other->_process = _process; + other->_isProcessSet = _isProcessSet; - other->_lockID = _lockID; - other->_isLockIDSet = _isLockIDSet; + other->_lockID = _lockID; + other->_isLockIDSet = _isLockIDSet; - other->_who = _who; - other->_isWhoSet = _isWhoSet; + other->_who = _who; + other->_isWhoSet = _isWhoSet; - other->_why = _why; - other->_isWhySet = _isWhySet; + other->_why = _why; + other->_isWhySet = _isWhySet; +} - } - - std::string LocksType::toString() const { - return toBSON().toString(); - } +std::string LocksType::toString() const { + return toBSON().toString(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/type_locks.h b/src/mongo/s/type_locks.h index 3eb6c3f86dd..e77da7c188c 100644 --- a/src/mongo/s/type_locks.h +++ b/src/mongo/s/type_locks.h @@ -35,226 +35,241 @@ namespace mongo { - /** - * This class represents the layout and contents of documents contained in the - * config.locks collection. All manipulation of documents coming from that - * collection should be done with this class. - * - * Usage Example: - * - * // Contact the config. 'conn' has been obtained before. - * DBClientBase* conn; - * BSONObj query = QUERY(LocksType::exampleField("exampleFieldName")); - * exampleDoc = conn->findOne(LocksType::ConfigNS, query); - * - * // Process the response. - * LocksType exampleType; - * std::string errMsg; - * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { - * // Can't use 'exampleType'. Take action. - * } - * // use 'exampleType' - * - */ - class LocksType { - public: - - enum State { - UNLOCKED = 0, - LOCK_PREP, // Only for legacy 3 config servers. - LOCKED, - }; - - // - // schema declarations - // - - // Name of the locks collection in the config server. - static const std::string ConfigNS; - - // Field names and types in the locks collection type. - static const BSONField<std::string> name; - static const BSONField<int> state; - static const BSONField<std::string> process; - static const BSONField<OID> lockID; - static const BSONField<std::string> who; - static const BSONField<std::string> why; - static const BSONField<Date_t> when; - - // - // locks type methods - // - - LocksType(); - ~LocksType(); - - /** - * Returns true if all the mandatory fields are present and have valid - * representations. Otherwise returns false and fills in the optional 'errMsg' string. - */ - bool isValid(std::string* errMsg) const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Clears and populates the internal state using the 'source' BSON object if the - * latter contains valid values. Otherwise sets errMsg and returns false. - */ - bool parseBSON(const BSONObj& source, std::string* errMsg); - - /** - * Clears the internal state. - */ - void clear(); - - /** - * Copies all the fields present in 'this' to 'other'. - */ - void cloneTo(LocksType* other) const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - // - // individual field accessors - // - - // Mandatory Fields - void setName(StringData name) { - _name = name.toString(); - _isNameSet = true; - } - - void unsetName() { _isNameSet = false; } - - bool isNameSet() const { return _isNameSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string getName() const { - dassert(_isNameSet); - return _name; - } - - void setState(const int state) { - _state = state; - _isStateSet = true; - } - - void unsetState() { _isStateSet = false; } - - bool isStateSet() const { return _isStateSet; } +/** + * This class represents the layout and contents of documents contained in the + * config.locks collection. All manipulation of documents coming from that + * collection should be done with this class. + * + * Usage Example: + * + * // Contact the config. 'conn' has been obtained before. + * DBClientBase* conn; + * BSONObj query = QUERY(LocksType::exampleField("exampleFieldName")); + * exampleDoc = conn->findOne(LocksType::ConfigNS, query); + * + * // Process the response. + * LocksType exampleType; + * std::string errMsg; + * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { + * // Can't use 'exampleType'. Take action. + * } + * // use 'exampleType' + * + */ +class LocksType { +public: + enum State { + UNLOCKED = 0, + LOCK_PREP, // Only for legacy 3 config servers. + LOCKED, + }; - // Calling get*() methods when the member is not set results in undefined behavior - int getState() const { - dassert(_isStateSet); - return _state; - } + // + // schema declarations + // - // Optional Fields - void setProcess(StringData process) { - _process = process.toString(); - _isProcessSet = true; - } + // Name of the locks collection in the config server. + static const std::string ConfigNS; - void unsetProcess() { _isProcessSet = false; } + // Field names and types in the locks collection type. + static const BSONField<std::string> name; + static const BSONField<int> state; + static const BSONField<std::string> process; + static const BSONField<OID> lockID; + static const BSONField<std::string> who; + static const BSONField<std::string> why; + static const BSONField<Date_t> when; - bool isProcessSet() const { - return _isProcessSet || process.hasDefault(); - } + // + // locks type methods + // - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - std::string getProcess() const { - if (_isProcessSet) { - return _process; - } else { - dassert(process.hasDefault()); - return process.getDefault(); - } - } - void setLockID(OID lockID) { - _lockID = lockID; - _isLockIDSet = true; - } + LocksType(); + ~LocksType(); - void unsetLockID() { _isLockIDSet = false; } + /** + * Returns true if all the mandatory fields are present and have valid + * representations. Otherwise returns false and fills in the optional 'errMsg' string. + */ + bool isValid(std::string* errMsg) const; - bool isLockIDSet() const { - return _isLockIDSet || lockID.hasDefault(); - } + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - OID getLockID() const { - if (_isLockIDSet) { - return _lockID; - } else { - dassert(lockID.hasDefault()); - return lockID.getDefault(); - } - } - void setWho(StringData who) { - _who = who.toString(); - _isWhoSet = true; - } + /** + * Clears and populates the internal state using the 'source' BSON object if the + * latter contains valid values. Otherwise sets errMsg and returns false. + */ + bool parseBSON(const BSONObj& source, std::string* errMsg); - void unsetWho() { _isWhoSet = false; } + /** + * Clears the internal state. + */ + void clear(); - bool isWhoSet() const { - return _isWhoSet || who.hasDefault(); - } + /** + * Copies all the fields present in 'this' to 'other'. + */ + void cloneTo(LocksType* other) const; - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - std::string getWho() const { - if (_isWhoSet) { - return _who; - } else { - dassert(who.hasDefault()); - return who.getDefault(); - } + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; + + // + // individual field accessors + // + + // Mandatory Fields + void setName(StringData name) { + _name = name.toString(); + _isNameSet = true; + } + + void unsetName() { + _isNameSet = false; + } + + bool isNameSet() const { + return _isNameSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string getName() const { + dassert(_isNameSet); + return _name; + } + + void setState(const int state) { + _state = state; + _isStateSet = true; + } + + void unsetState() { + _isStateSet = false; + } + + bool isStateSet() const { + return _isStateSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + int getState() const { + dassert(_isStateSet); + return _state; + } + + // Optional Fields + void setProcess(StringData process) { + _process = process.toString(); + _isProcessSet = true; + } + + void unsetProcess() { + _isProcessSet = false; + } + + bool isProcessSet() const { + return _isProcessSet || process.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + std::string getProcess() const { + if (_isProcessSet) { + return _process; + } else { + dassert(process.hasDefault()); + return process.getDefault(); } - void setWhy(StringData why) { - _why = why.toString(); - _isWhySet = true; + } + void setLockID(OID lockID) { + _lockID = lockID; + _isLockIDSet = true; + } + + void unsetLockID() { + _isLockIDSet = false; + } + + bool isLockIDSet() const { + return _isLockIDSet || lockID.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + OID getLockID() const { + if (_isLockIDSet) { + return _lockID; + } else { + dassert(lockID.hasDefault()); + return lockID.getDefault(); } - - void unsetWhy() { _isWhySet = false; } - - bool isWhySet() const { - return _isWhySet || why.hasDefault(); + } + void setWho(StringData who) { + _who = who.toString(); + _isWhoSet = true; + } + + void unsetWho() { + _isWhoSet = false; + } + + bool isWhoSet() const { + return _isWhoSet || who.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + std::string getWho() const { + if (_isWhoSet) { + return _who; + } else { + dassert(who.hasDefault()); + return who.getDefault(); } - - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - std::string getWhy() const { - if (_isWhySet) { - return _why; - } else { - dassert(why.hasDefault()); - return why.getDefault(); - } + } + void setWhy(StringData why) { + _why = why.toString(); + _isWhySet = true; + } + + void unsetWhy() { + _isWhySet = false; + } + + bool isWhySet() const { + return _isWhySet || why.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + std::string getWhy() const { + if (_isWhySet) { + return _why; + } else { + dassert(why.hasDefault()); + return why.getDefault(); } - - private: - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - std::string _name; // (M) name of the lock - bool _isNameSet; - int _state; // (M) 0: Unlocked | 1: Locks in contention | 2: Lock held - bool _isStateSet; - std::string _process; // (O) optional if unlocked. contains the (unique) identifier - bool _isProcessSet; - OID _lockID; // (O) optional if unlocked. a unique identifier for the instance - bool _isLockIDSet; - std::string _who; // (O) optional if unlocked. a note about why the lock is held, - bool _isWhoSet; - std::string _why; // (O) optional if unlocked. a human readable description of the - bool _isWhySet; - }; - -} // namespace mongo + } + +private: + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + std::string _name; // (M) name of the lock + bool _isNameSet; + int _state; // (M) 0: Unlocked | 1: Locks in contention | 2: Lock held + bool _isStateSet; + std::string _process; // (O) optional if unlocked. contains the (unique) identifier + bool _isProcessSet; + OID _lockID; // (O) optional if unlocked. a unique identifier for the instance + bool _isLockIDSet; + std::string _who; // (O) optional if unlocked. a note about why the lock is held, + bool _isWhoSet; + std::string _why; // (O) optional if unlocked. a human readable description of the + bool _isWhySet; +}; + +} // namespace mongo diff --git a/src/mongo/s/type_locks_test.cpp b/src/mongo/s/type_locks_test.cpp index 602a5f672e1..024be534dc8 100644 --- a/src/mongo/s/type_locks_test.cpp +++ b/src/mongo/s/type_locks_test.cpp @@ -32,197 +32,187 @@ namespace { - using std::string; - using mongo::BSONObj; - using mongo::LocksType; - using mongo::OID; - - TEST(Validity, Empty) { - LocksType lock; - BSONObj emptyObj = BSONObj(); - string errMsg; - ASSERT(lock.parseBSON(emptyObj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, UnlockedWithOptional) { - LocksType lock; - OID testLockID = OID::gen(); - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(0) << - LocksType::lockID(testLockID) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(lock.isValid(NULL)); - ASSERT_EQUALS(lock.getName(), "balancer"); - ASSERT_EQUALS(lock.getProcess(), "host.local:27017:1352918870:16807"); - ASSERT_EQUALS(lock.getState(), 0); - ASSERT_EQUALS(lock.getLockID(), testLockID); - ASSERT_EQUALS(lock.getWho(), "host.local:27017:1352918870:16807:Balancer:282475249"); - ASSERT_EQUALS(lock.getWhy(), "doing balance round"); - } - - TEST(Validity, UnlockedWithoutOptional) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::state(0)); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(lock.isValid(NULL)); - ASSERT_EQUALS(lock.getName(), "balancer"); - ASSERT_EQUALS(lock.getState(), 0); - } - - TEST(Validity, LockedValid) { - LocksType lock; - OID testLockID = OID::gen(); - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(2) << - LocksType::lockID(testLockID) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(lock.isValid(NULL)); - ASSERT_EQUALS(lock.getName(), "balancer"); - ASSERT_EQUALS(lock.getProcess(), "host.local:27017:1352918870:16807"); - ASSERT_EQUALS(lock.getState(), 2); - ASSERT_EQUALS(lock.getLockID(), testLockID); - ASSERT_EQUALS(lock.getWho(), "host.local:27017:1352918870:16807:Balancer:282475249"); - ASSERT_EQUALS(lock.getWhy(), "doing balance round"); - } - - TEST(Validity, LockedMissingProcess) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::state(2) << - LocksType::lockID(OID::gen()) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, LockedMissingLockID) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(2) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, LockedMissingWho) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(2) << - LocksType::lockID(OID::gen()) << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, LockedMissingWhy) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(2) << - LocksType::lockID(OID::gen()) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, ContestedValid) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(1) << - LocksType::lockID(OID::gen()) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(lock.isValid(NULL)); - } - - TEST(Validity, ContestedMissingProcess) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::state(1) << - LocksType::lockID(OID::gen()) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, ContestedMissingLockID) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(1) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, ContestedMissingWho) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(1) << - LocksType::lockID(OID::gen()) << - LocksType::why("doing balance round")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, ContestedMissingWhy) { - LocksType lock; - BSONObj obj = BSON(LocksType::name("balancer") << - LocksType::process("host.local:27017:1352918870:16807") << - LocksType::state(1) << - LocksType::lockID(OID::gen()) << - LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249")); - string errMsg; - ASSERT(lock.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(lock.isValid(NULL)); - } - - TEST(Validity, BadType) { - LocksType lock; - BSONObj obj = BSON(LocksType::name() << 0); - string errMsg; - ASSERT((!lock.parseBSON(obj, &errMsg)) && (errMsg != "")); - } - -} // unnamed namespace +using std::string; +using mongo::BSONObj; +using mongo::LocksType; +using mongo::OID; + +TEST(Validity, Empty) { + LocksType lock; + BSONObj emptyObj = BSONObj(); + string errMsg; + ASSERT(lock.parseBSON(emptyObj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, UnlockedWithOptional) { + LocksType lock; + OID testLockID = OID::gen(); + BSONObj obj = BSON(LocksType::name("balancer") + << LocksType::process("host.local:27017:1352918870:16807") + << LocksType::state(0) << LocksType::lockID(testLockID) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(lock.isValid(NULL)); + ASSERT_EQUALS(lock.getName(), "balancer"); + ASSERT_EQUALS(lock.getProcess(), "host.local:27017:1352918870:16807"); + ASSERT_EQUALS(lock.getState(), 0); + ASSERT_EQUALS(lock.getLockID(), testLockID); + ASSERT_EQUALS(lock.getWho(), "host.local:27017:1352918870:16807:Balancer:282475249"); + ASSERT_EQUALS(lock.getWhy(), "doing balance round"); +} + +TEST(Validity, UnlockedWithoutOptional) { + LocksType lock; + BSONObj obj = BSON(LocksType::name("balancer") << LocksType::state(0)); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(lock.isValid(NULL)); + ASSERT_EQUALS(lock.getName(), "balancer"); + ASSERT_EQUALS(lock.getState(), 0); +} + +TEST(Validity, LockedValid) { + LocksType lock; + OID testLockID = OID::gen(); + BSONObj obj = BSON(LocksType::name("balancer") + << LocksType::process("host.local:27017:1352918870:16807") + << LocksType::state(2) << LocksType::lockID(testLockID) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(lock.isValid(NULL)); + ASSERT_EQUALS(lock.getName(), "balancer"); + ASSERT_EQUALS(lock.getProcess(), "host.local:27017:1352918870:16807"); + ASSERT_EQUALS(lock.getState(), 2); + ASSERT_EQUALS(lock.getLockID(), testLockID); + ASSERT_EQUALS(lock.getWho(), "host.local:27017:1352918870:16807:Balancer:282475249"); + ASSERT_EQUALS(lock.getWhy(), "doing balance round"); +} + +TEST(Validity, LockedMissingProcess) { + LocksType lock; + BSONObj obj = BSON(LocksType::name("balancer") + << LocksType::state(2) << LocksType::lockID(OID::gen()) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, LockedMissingLockID) { + LocksType lock; + BSONObj obj = + BSON(LocksType::name("balancer") + << LocksType::process("host.local:27017:1352918870:16807") << LocksType::state(2) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, LockedMissingWho) { + LocksType lock; + BSONObj obj = + BSON(LocksType::name("balancer") << LocksType::process("host.local:27017:1352918870:16807") + << LocksType::state(2) << LocksType::lockID(OID::gen()) + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, LockedMissingWhy) { + LocksType lock; + BSONObj obj = BSON(LocksType::name("balancer") + << LocksType::process("host.local:27017:1352918870:16807") + << LocksType::state(2) << LocksType::lockID(OID::gen()) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, ContestedValid) { + LocksType lock; + BSONObj obj = BSON(LocksType::name("balancer") + << LocksType::process("host.local:27017:1352918870:16807") + << LocksType::state(1) << LocksType::lockID(OID::gen()) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(lock.isValid(NULL)); +} + +TEST(Validity, ContestedMissingProcess) { + LocksType lock; + BSONObj obj = BSON(LocksType::name("balancer") + << LocksType::state(1) << LocksType::lockID(OID::gen()) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, ContestedMissingLockID) { + LocksType lock; + BSONObj obj = + BSON(LocksType::name("balancer") + << LocksType::process("host.local:27017:1352918870:16807") << LocksType::state(1) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249") + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, ContestedMissingWho) { + LocksType lock; + BSONObj obj = + BSON(LocksType::name("balancer") << LocksType::process("host.local:27017:1352918870:16807") + << LocksType::state(1) << LocksType::lockID(OID::gen()) + << LocksType::why("doing balance round")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, ContestedMissingWhy) { + LocksType lock; + BSONObj obj = BSON(LocksType::name("balancer") + << LocksType::process("host.local:27017:1352918870:16807") + << LocksType::state(1) << LocksType::lockID(OID::gen()) + << LocksType::who("host.local:27017:1352918870:16807:Balancer:282475249")); + string errMsg; + ASSERT(lock.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(lock.isValid(NULL)); +} + +TEST(Validity, BadType) { + LocksType lock; + BSONObj obj = BSON(LocksType::name() << 0); + string errMsg; + ASSERT((!lock.parseBSON(obj, &errMsg)) && (errMsg != "")); +} + +} // unnamed namespace diff --git a/src/mongo/s/type_mongos.cpp b/src/mongo/s/type_mongos.cpp index 07009c6a864..34e9f94efa5 100644 --- a/src/mongo/s/type_mongos.cpp +++ b/src/mongo/s/type_mongos.cpp @@ -32,147 +32,156 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; +using mongoutils::str::stream; - const std::string MongosType::ConfigNS = "config.mongos"; +const std::string MongosType::ConfigNS = "config.mongos"; - const BSONField<std::string> MongosType::name("_id"); - const BSONField<Date_t> MongosType::ping("ping"); - const BSONField<int> MongosType::up("up"); - const BSONField<bool> MongosType::waiting("waiting"); - const BSONField<std::string> MongosType::mongoVersion("mongoVersion"); - const BSONField<int> MongosType::configVersion("configVersion"); +const BSONField<std::string> MongosType::name("_id"); +const BSONField<Date_t> MongosType::ping("ping"); +const BSONField<int> MongosType::up("up"); +const BSONField<bool> MongosType::waiting("waiting"); +const BSONField<std::string> MongosType::mongoVersion("mongoVersion"); +const BSONField<int> MongosType::configVersion("configVersion"); - MongosType::MongosType() { - clear(); - } +MongosType::MongosType() { + clear(); +} - MongosType::~MongosType() { - } +MongosType::~MongosType() {} - bool MongosType::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isNameSet) { - *errMsg = stream() << "missing " << name.name() << " field"; - return false; - } - if (!_isPingSet) { - *errMsg = stream() << "missing " << ping.name() << " field"; - return false; - } - if (!_isUpSet) { - *errMsg = stream() << "missing " << up.name() << " field"; - return false; - } - if (!_isWaitingSet) { - *errMsg = stream() << "missing " << waiting.name() << " field"; - return false; - } - - return true; +bool MongosType::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BSONObj MongosType::toBSON() const { - BSONObjBuilder builder; - - if (_isNameSet) builder.append(name(), _name); - if (_isPingSet) builder.append(ping(), _ping); - if (_isUpSet) builder.append(up(), _up); - if (_isWaitingSet) builder.append(waiting(), _waiting); - if (_isMongoVersionSet) builder.append(mongoVersion(), _mongoVersion); - if (_isConfigVersionSet) builder.append(configVersion(), _configVersion); - - return builder.obj(); + // All the mandatory fields must be present. + if (!_isNameSet) { + *errMsg = stream() << "missing " << name.name() << " field"; + return false; + } + if (!_isPingSet) { + *errMsg = stream() << "missing " << ping.name() << " field"; + return false; + } + if (!_isUpSet) { + *errMsg = stream() << "missing " << up.name() << " field"; + return false; + } + if (!_isWaitingSet) { + *errMsg = stream() << "missing " << waiting.name() << " field"; + return false; } - bool MongosType::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + return true; +} - std::string dummy; - if (!errMsg) errMsg = &dummy; +BSONObj MongosType::toBSON() const { + BSONObjBuilder builder; - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, name, &_name, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isNameSet = fieldState == FieldParser::FIELD_SET; + if (_isNameSet) + builder.append(name(), _name); + if (_isPingSet) + builder.append(ping(), _ping); + if (_isUpSet) + builder.append(up(), _up); + if (_isWaitingSet) + builder.append(waiting(), _waiting); + if (_isMongoVersionSet) + builder.append(mongoVersion(), _mongoVersion); + if (_isConfigVersionSet) + builder.append(configVersion(), _configVersion); - fieldState = FieldParser::extract(source, ping, &_ping, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isPingSet = fieldState == FieldParser::FIELD_SET; + return builder.obj(); +} - fieldState = FieldParser::extract(source, up, &_up, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isUpSet = fieldState == FieldParser::FIELD_SET; +bool MongosType::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - fieldState = FieldParser::extract(source, waiting, &_waiting, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isWaitingSet = fieldState == FieldParser::FIELD_SET; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - fieldState = FieldParser::extract(source, mongoVersion, &_mongoVersion, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isMongoVersionSet = fieldState == FieldParser::FIELD_SET; + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, name, &_name, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isNameSet = fieldState == FieldParser::FIELD_SET; - fieldState = FieldParser::extract(source, configVersion, &_configVersion, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isConfigVersionSet = fieldState == FieldParser::FIELD_SET; + fieldState = FieldParser::extract(source, ping, &_ping, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isPingSet = fieldState == FieldParser::FIELD_SET; - return true; - } + fieldState = FieldParser::extract(source, up, &_up, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isUpSet = fieldState == FieldParser::FIELD_SET; - void MongosType::clear() { + fieldState = FieldParser::extract(source, waiting, &_waiting, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isWaitingSet = fieldState == FieldParser::FIELD_SET; - _name.clear(); - _isNameSet = false; + fieldState = FieldParser::extract(source, mongoVersion, &_mongoVersion, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isMongoVersionSet = fieldState == FieldParser::FIELD_SET; - _ping = Date_t(); - _isPingSet = false; + fieldState = FieldParser::extract(source, configVersion, &_configVersion, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isConfigVersionSet = fieldState == FieldParser::FIELD_SET; - _up = 0; - _isUpSet = false; + return true; +} - _waiting = false; - _isWaitingSet = false; +void MongosType::clear() { + _name.clear(); + _isNameSet = false; - _mongoVersion.clear(); - _isMongoVersionSet = false; + _ping = Date_t(); + _isPingSet = false; - _configVersion = 0; - _isConfigVersionSet = false; + _up = 0; + _isUpSet = false; - } + _waiting = false; + _isWaitingSet = false; - void MongosType::cloneTo(MongosType* other) const { - other->clear(); + _mongoVersion.clear(); + _isMongoVersionSet = false; - other->_name = _name; - other->_isNameSet = _isNameSet; + _configVersion = 0; + _isConfigVersionSet = false; +} - other->_ping = _ping; - other->_isPingSet = _isPingSet; +void MongosType::cloneTo(MongosType* other) const { + other->clear(); - other->_up = _up; - other->_isUpSet = _isUpSet; + other->_name = _name; + other->_isNameSet = _isNameSet; - other->_waiting = _waiting; - other->_isWaitingSet = _isWaitingSet; + other->_ping = _ping; + other->_isPingSet = _isPingSet; - other->_mongoVersion = _mongoVersion; - other->_isMongoVersionSet = _isMongoVersionSet; + other->_up = _up; + other->_isUpSet = _isUpSet; - other->_configVersion = _configVersion; - other->_isConfigVersionSet = _isConfigVersionSet; + other->_waiting = _waiting; + other->_isWaitingSet = _isWaitingSet; - } + other->_mongoVersion = _mongoVersion; + other->_isMongoVersionSet = _isMongoVersionSet; - std::string MongosType::toString() const { - return toBSON().toString(); - } + other->_configVersion = _configVersion; + other->_isConfigVersionSet = _isConfigVersionSet; +} + +std::string MongosType::toString() const { + return toBSON().toString(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/type_mongos.h b/src/mongo/s/type_mongos.h index 9969fca4815..2bf0a372359 100644 --- a/src/mongo/s/type_mongos.h +++ b/src/mongo/s/type_mongos.h @@ -36,208 +36,228 @@ namespace mongo { - /** - * This class represents the layout and contents of documents contained in the - * config.mongos collection. All manipulation of documents coming from that - * collection should be done with this class. - * - * Usage Example: - * - * // Contact the config. 'conn' has been obtained before. - * DBClientBase* conn; - * BSONObj query = QUERY(MongosType::exampleField("exampleFieldName")); - * exampleDoc = conn->findOne(MongosType::ConfigNS, query); - * - * // Process the response. - * MongosType exampleType; - * std::string errMsg; - * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { - * // Can't use 'exampleType'. Take action. - * } - * // use 'exampleType' - * - */ - class MongosType { - MONGO_DISALLOW_COPYING(MongosType); - public: - - // - // schema declarations - // - - // Name of the mongos collection in the config server. - static const std::string ConfigNS; - - // Field names and types in the mongos collection type. - static const BSONField<std::string> name; - static const BSONField<Date_t> ping; - static const BSONField<int> up; - static const BSONField<bool> waiting; - static const BSONField<std::string> mongoVersion; - static const BSONField<int> configVersion; - - // - // mongos type methods - // - - MongosType(); - ~MongosType(); - - /** - * Returns true if all the mandatory fields are present and have valid - * representations. Otherwise returns false and fills in the optional 'errMsg' string. - */ - bool isValid(std::string* errMsg) const; - - /** - * Returns the BSON representation of the entry. - */ - BSONObj toBSON() const; - - /** - * Clears and populates the internal state using the 'source' BSON object if the - * latter contains valid values. Otherwise sets errMsg and returns false. - */ - bool parseBSON(const BSONObj& source, std::string* errMsg); - - /** - * Clears the internal state. - */ - void clear(); - - /** - * Copies all the fields present in 'this' to 'other'. - */ - void cloneTo(MongosType* other) const; - - /** - * Returns a std::string representation of the current internal state. - */ - std::string toString() const; - - // - // individual field accessors - // - - // Mandatory Fields - void setName(StringData name) { - _name = name.toString(); - _isNameSet = true; - } - - void unsetName() { _isNameSet = false; } - - bool isNameSet() const { return _isNameSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const std::string& getName() const { - dassert(_isNameSet); - return _name; - } - - void setPing(const Date_t ping) { - _ping = ping; - _isPingSet = true; - } - - void unsetPing() { _isPingSet = false; } - - bool isPingSet() const { return _isPingSet; } - - // Calling get*() methods when the member is not set results in undefined behavior - const Date_t getPing() const { - dassert(_isPingSet); - return _ping; - } - - void setUp(const int up) { - _up = up; - _isUpSet = true; - } - - void unsetUp() { _isUpSet = false; } - - bool isUpSet() const { return _isUpSet; } +/** + * This class represents the layout and contents of documents contained in the + * config.mongos collection. All manipulation of documents coming from that + * collection should be done with this class. + * + * Usage Example: + * + * // Contact the config. 'conn' has been obtained before. + * DBClientBase* conn; + * BSONObj query = QUERY(MongosType::exampleField("exampleFieldName")); + * exampleDoc = conn->findOne(MongosType::ConfigNS, query); + * + * // Process the response. + * MongosType exampleType; + * std::string errMsg; + * if (!exampleType.parseBSON(exampleDoc, &errMsg) || !exampleType.isValid(&errMsg)) { + * // Can't use 'exampleType'. Take action. + * } + * // use 'exampleType' + * + */ +class MongosType { + MONGO_DISALLOW_COPYING(MongosType); - // Calling get*() methods when the member is not set results in undefined behavior - int getUp() const { - dassert(_isUpSet); - return _up; - } +public: + // + // schema declarations + // - void setWaiting(const bool waiting) { - _waiting = waiting; - _isWaitingSet = true; - } + // Name of the mongos collection in the config server. + static const std::string ConfigNS; - void unsetWaiting() { _isWaitingSet = false; } + // Field names and types in the mongos collection type. + static const BSONField<std::string> name; + static const BSONField<Date_t> ping; + static const BSONField<int> up; + static const BSONField<bool> waiting; + static const BSONField<std::string> mongoVersion; + static const BSONField<int> configVersion; - bool isWaitingSet() const { return _isWaitingSet; } + // + // mongos type methods + // - // Calling get*() methods when the member is not set results in undefined behavior - bool getWaiting() const { - dassert(_isWaitingSet); - return _waiting; - } + MongosType(); + ~MongosType(); - // Optional Fields - void setMongoVersion(StringData mongoVersion) { - _mongoVersion = mongoVersion.toString(); - _isMongoVersionSet = true; - } + /** + * Returns true if all the mandatory fields are present and have valid + * representations. Otherwise returns false and fills in the optional 'errMsg' string. + */ + bool isValid(std::string* errMsg) const; - void unsetMongoVersion() { _isMongoVersionSet = false; } + /** + * Returns the BSON representation of the entry. + */ + BSONObj toBSON() const; - bool isMongoVersionSet() const { - return _isMongoVersionSet || mongoVersion.hasDefault(); - } + /** + * Clears and populates the internal state using the 'source' BSON object if the + * latter contains valid values. Otherwise sets errMsg and returns false. + */ + bool parseBSON(const BSONObj& source, std::string* errMsg); - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - std::string getMongoVersion() const { - if (_isMongoVersionSet) { - return _mongoVersion; - } else { - dassert(mongoVersion.hasDefault()); - return mongoVersion.getDefault(); - } - } - void setConfigVersion(const int configVersion) { - _configVersion = configVersion; - _isConfigVersionSet = true; - } + /** + * Clears the internal state. + */ + void clear(); - void unsetConfigVersion() { _isConfigVersionSet = false; } + /** + * Copies all the fields present in 'this' to 'other'. + */ + void cloneTo(MongosType* other) const; - bool isConfigVersionSet() const { - return _isConfigVersionSet || configVersion.hasDefault(); + /** + * Returns a std::string representation of the current internal state. + */ + std::string toString() const; + + // + // individual field accessors + // + + // Mandatory Fields + void setName(StringData name) { + _name = name.toString(); + _isNameSet = true; + } + + void unsetName() { + _isNameSet = false; + } + + bool isNameSet() const { + return _isNameSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const std::string& getName() const { + dassert(_isNameSet); + return _name; + } + + void setPing(const Date_t ping) { + _ping = ping; + _isPingSet = true; + } + + void unsetPing() { + _isPingSet = false; + } + + bool isPingSet() const { + return _isPingSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + const Date_t getPing() const { + dassert(_isPingSet); + return _ping; + } + + void setUp(const int up) { + _up = up; + _isUpSet = true; + } + + void unsetUp() { + _isUpSet = false; + } + + bool isUpSet() const { + return _isUpSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + int getUp() const { + dassert(_isUpSet); + return _up; + } + + void setWaiting(const bool waiting) { + _waiting = waiting; + _isWaitingSet = true; + } + + void unsetWaiting() { + _isWaitingSet = false; + } + + bool isWaitingSet() const { + return _isWaitingSet; + } + + // Calling get*() methods when the member is not set results in undefined behavior + bool getWaiting() const { + dassert(_isWaitingSet); + return _waiting; + } + + // Optional Fields + void setMongoVersion(StringData mongoVersion) { + _mongoVersion = mongoVersion.toString(); + _isMongoVersionSet = true; + } + + void unsetMongoVersion() { + _isMongoVersionSet = false; + } + + bool isMongoVersionSet() const { + return _isMongoVersionSet || mongoVersion.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + std::string getMongoVersion() const { + if (_isMongoVersionSet) { + return _mongoVersion; + } else { + dassert(mongoVersion.hasDefault()); + return mongoVersion.getDefault(); } - - // Calling get*() methods when the member is not set and has no default results in undefined - // behavior - int getConfigVersion() const { - if (_isConfigVersionSet) { - return _configVersion; - } else { - dassert(configVersion.hasDefault()); - return configVersion.getDefault(); - } + } + void setConfigVersion(const int configVersion) { + _configVersion = configVersion; + _isConfigVersionSet = true; + } + + void unsetConfigVersion() { + _isConfigVersionSet = false; + } + + bool isConfigVersionSet() const { + return _isConfigVersionSet || configVersion.hasDefault(); + } + + // Calling get*() methods when the member is not set and has no default results in undefined + // behavior + int getConfigVersion() const { + if (_isConfigVersionSet) { + return _configVersion; + } else { + dassert(configVersion.hasDefault()); + return configVersion.getDefault(); } - - private: - // Convention: (M)andatory, (O)ptional, (S)pecial rule. - std::string _name; // (M) "host:port" for this mongos - bool _isNameSet; - Date_t _ping; // (M) last time it was seen alive - bool _isPingSet; - int _up; // (M) uptime at the last ping - bool _isUpSet; - bool _waiting; // (M) for testing purposes - bool _isWaitingSet; - std::string _mongoVersion; // (O) the mongodb version of the pinging mongos - bool _isMongoVersionSet; - int _configVersion; // (O) the config version of the pinging mongos - bool _isConfigVersionSet; - }; - -} // namespace mongo + } + +private: + // Convention: (M)andatory, (O)ptional, (S)pecial rule. + std::string _name; // (M) "host:port" for this mongos + bool _isNameSet; + Date_t _ping; // (M) last time it was seen alive + bool _isPingSet; + int _up; // (M) uptime at the last ping + bool _isUpSet; + bool _waiting; // (M) for testing purposes + bool _isWaitingSet; + std::string _mongoVersion; // (O) the mongodb version of the pinging mongos + bool _isMongoVersionSet; + int _configVersion; // (O) the config version of the pinging mongos + bool _isConfigVersionSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/type_mongos_test.cpp b/src/mongo/s/type_mongos_test.cpp index 1c0ab5be56d..210c1ae4169 100644 --- a/src/mongo/s/type_mongos_test.cpp +++ b/src/mongo/s/type_mongos_test.cpp @@ -32,154 +32,141 @@ namespace { - using std::string; - using mongo::BSONObj; - using mongo::MongosType; - using mongo::Date_t; +using std::string; +using mongo::BSONObj; +using mongo::MongosType; +using mongo::Date_t; - TEST(Validity, MissingName) { - MongosType mongos; - string errMsg; - BSONObj obj = BSON(MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << - MongosType::up(100) << - MongosType::waiting(false) << - MongosType::mongoVersion("x.x.x") << - MongosType::configVersion(0)); - ASSERT(mongos.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_FALSE(mongos.isNameSet()); - ASSERT_TRUE(mongos.isPingSet()); - ASSERT_TRUE(mongos.isUpSet()); - ASSERT_TRUE(mongos.isWaitingSet()); - ASSERT_TRUE(mongos.isMongoVersionSet()); - ASSERT_TRUE(mongos.isConfigVersionSet()); - ASSERT_FALSE(mongos.isValid(NULL)); - } +TEST(Validity, MissingName) { + MongosType mongos; + string errMsg; + BSONObj obj = BSON(MongosType::ping(Date_t::fromMillisSinceEpoch(1)) + << MongosType::up(100) << MongosType::waiting(false) + << MongosType::mongoVersion("x.x.x") << MongosType::configVersion(0)); + ASSERT(mongos.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_FALSE(mongos.isNameSet()); + ASSERT_TRUE(mongos.isPingSet()); + ASSERT_TRUE(mongos.isUpSet()); + ASSERT_TRUE(mongos.isWaitingSet()); + ASSERT_TRUE(mongos.isMongoVersionSet()); + ASSERT_TRUE(mongos.isConfigVersionSet()); + ASSERT_FALSE(mongos.isValid(NULL)); +} - TEST(Validity, MissingPing) { - MongosType mongos; - string errMsg; - BSONObj obj = BSON(MongosType::name("localhost:27017") << - MongosType::up(100) << - MongosType::waiting(false) << - MongosType::mongoVersion("x.x.x") << - MongosType::configVersion(0)); - ASSERT(mongos.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(mongos.isNameSet()); - ASSERT_FALSE(mongos.isPingSet()); - ASSERT_TRUE(mongos.isUpSet()); - ASSERT_TRUE(mongos.isWaitingSet()); - ASSERT_TRUE(mongos.isMongoVersionSet()); - ASSERT_TRUE(mongos.isConfigVersionSet()); - ASSERT_FALSE(mongos.isValid(NULL)); - } +TEST(Validity, MissingPing) { + MongosType mongos; + string errMsg; + BSONObj obj = BSON(MongosType::name("localhost:27017") + << MongosType::up(100) << MongosType::waiting(false) + << MongosType::mongoVersion("x.x.x") << MongosType::configVersion(0)); + ASSERT(mongos.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(mongos.isNameSet()); + ASSERT_FALSE(mongos.isPingSet()); + ASSERT_TRUE(mongos.isUpSet()); + ASSERT_TRUE(mongos.isWaitingSet()); + ASSERT_TRUE(mongos.isMongoVersionSet()); + ASSERT_TRUE(mongos.isConfigVersionSet()); + ASSERT_FALSE(mongos.isValid(NULL)); +} - TEST(Validity, MissingUp) { - MongosType mongos; - string errMsg; - BSONObj obj = BSON(MongosType::name("localhost:27017") << - MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << - MongosType::waiting(false) << - MongosType::mongoVersion("x.x.x") << - MongosType::configVersion(0)); - ASSERT(mongos.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(mongos.isNameSet()); - ASSERT_TRUE(mongos.isPingSet()); - ASSERT_FALSE(mongos.isUpSet()); - ASSERT_TRUE(mongos.isWaitingSet()); - ASSERT_TRUE(mongos.isMongoVersionSet()); - ASSERT_TRUE(mongos.isConfigVersionSet()); - ASSERT_FALSE(mongos.isValid(NULL)); - } +TEST(Validity, MissingUp) { + MongosType mongos; + string errMsg; + BSONObj obj = + BSON(MongosType::name("localhost:27017") + << MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << MongosType::waiting(false) + << MongosType::mongoVersion("x.x.x") << MongosType::configVersion(0)); + ASSERT(mongos.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(mongos.isNameSet()); + ASSERT_TRUE(mongos.isPingSet()); + ASSERT_FALSE(mongos.isUpSet()); + ASSERT_TRUE(mongos.isWaitingSet()); + ASSERT_TRUE(mongos.isMongoVersionSet()); + ASSERT_TRUE(mongos.isConfigVersionSet()); + ASSERT_FALSE(mongos.isValid(NULL)); +} - TEST(Validity, MissingWaiting) { - MongosType mongos; - string errMsg; - BSONObj obj = BSON(MongosType::name("localhost:27017") << - MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << - MongosType::up(100) << - MongosType::mongoVersion("x.x.x") << - MongosType::configVersion(0)); - ASSERT(mongos.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(mongos.isNameSet()); - ASSERT_TRUE(mongos.isPingSet()); - ASSERT_TRUE(mongos.isUpSet()); - ASSERT_FALSE(mongos.isWaitingSet()); - ASSERT_TRUE(mongos.isMongoVersionSet()); - ASSERT_TRUE(mongos.isConfigVersionSet()); - ASSERT_FALSE(mongos.isValid(NULL)); - } +TEST(Validity, MissingWaiting) { + MongosType mongos; + string errMsg; + BSONObj obj = BSON(MongosType::name("localhost:27017") + << MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << MongosType::up(100) + << MongosType::mongoVersion("x.x.x") << MongosType::configVersion(0)); + ASSERT(mongos.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(mongos.isNameSet()); + ASSERT_TRUE(mongos.isPingSet()); + ASSERT_TRUE(mongos.isUpSet()); + ASSERT_FALSE(mongos.isWaitingSet()); + ASSERT_TRUE(mongos.isMongoVersionSet()); + ASSERT_TRUE(mongos.isConfigVersionSet()); + ASSERT_FALSE(mongos.isValid(NULL)); +} - TEST(Validity, MissingMongoVersion) { - MongosType mongos; - string errMsg; - BSONObj obj = BSON(MongosType::name("localhost:27017") << - MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << - MongosType::up(100) << - MongosType::waiting(false) << - MongosType::configVersion(0)); - ASSERT(mongos.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(mongos.isNameSet()); - ASSERT_TRUE(mongos.isPingSet()); - ASSERT_TRUE(mongos.isUpSet()); - ASSERT_TRUE(mongos.isWaitingSet()); - ASSERT_FALSE(mongos.isMongoVersionSet()); - ASSERT_TRUE(mongos.isConfigVersionSet()); - /* NOTE: mongoVersion should eventually become mandatory, but is optional now for backward - * compatibility reasons */ - ASSERT_TRUE(mongos.isValid(NULL)); - } +TEST(Validity, MissingMongoVersion) { + MongosType mongos; + string errMsg; + BSONObj obj = BSON(MongosType::name("localhost:27017") + << MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << MongosType::up(100) + << MongosType::waiting(false) << MongosType::configVersion(0)); + ASSERT(mongos.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(mongos.isNameSet()); + ASSERT_TRUE(mongos.isPingSet()); + ASSERT_TRUE(mongos.isUpSet()); + ASSERT_TRUE(mongos.isWaitingSet()); + ASSERT_FALSE(mongos.isMongoVersionSet()); + ASSERT_TRUE(mongos.isConfigVersionSet()); + /* NOTE: mongoVersion should eventually become mandatory, but is optional now for backward + * compatibility reasons */ + ASSERT_TRUE(mongos.isValid(NULL)); +} - TEST(Validity, MissingConfigVersion) { - MongosType mongos; - string errMsg; - BSONObj obj = BSON(MongosType::name("localhost:27017") << - MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << - MongosType::up(100) << - MongosType::waiting(false) << - MongosType::mongoVersion("x.x.x")); - ASSERT(mongos.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(mongos.isNameSet()); - ASSERT_TRUE(mongos.isPingSet()); - ASSERT_TRUE(mongos.isUpSet()); - ASSERT_TRUE(mongos.isWaitingSet()); - ASSERT_TRUE(mongos.isMongoVersionSet()); - ASSERT_FALSE(mongos.isConfigVersionSet()); - /* NOTE: configVersion should eventually become mandatory, but is optional now for backward - * compatibility reasons */ - ASSERT_TRUE(mongos.isValid(NULL)); - } +TEST(Validity, MissingConfigVersion) { + MongosType mongos; + string errMsg; + BSONObj obj = BSON(MongosType::name("localhost:27017") + << MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << MongosType::up(100) + << MongosType::waiting(false) << MongosType::mongoVersion("x.x.x")); + ASSERT(mongos.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(mongos.isNameSet()); + ASSERT_TRUE(mongos.isPingSet()); + ASSERT_TRUE(mongos.isUpSet()); + ASSERT_TRUE(mongos.isWaitingSet()); + ASSERT_TRUE(mongos.isMongoVersionSet()); + ASSERT_FALSE(mongos.isConfigVersionSet()); + /* NOTE: configVersion should eventually become mandatory, but is optional now for backward + * compatibility reasons */ + ASSERT_TRUE(mongos.isValid(NULL)); +} - TEST(Validity, Valid) { - MongosType mongos; - BSONObj obj = BSON(MongosType::name("localhost:27017") << - MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << - MongosType::up(100) << - MongosType::waiting(false) << - MongosType::mongoVersion("x.x.x") << - MongosType::configVersion(0)); - string errMsg; - ASSERT(mongos.parseBSON(obj, &errMsg)); - ASSERT_EQUALS(errMsg, ""); - ASSERT_TRUE(mongos.isValid(NULL)); - ASSERT_EQUALS(mongos.getName(), "localhost:27017"); - ASSERT_EQUALS(mongos.getPing(), Date_t::fromMillisSinceEpoch(1)); - ASSERT_EQUALS(mongos.getUp(), 100); - ASSERT_EQUALS(mongos.getWaiting(), false); - ASSERT_EQUALS(mongos.getMongoVersion(), "x.x.x"); - ASSERT_EQUALS(mongos.getConfigVersion(), 0); - } +TEST(Validity, Valid) { + MongosType mongos; + BSONObj obj = BSON(MongosType::name("localhost:27017") + << MongosType::ping(Date_t::fromMillisSinceEpoch(1)) << MongosType::up(100) + << MongosType::waiting(false) << MongosType::mongoVersion("x.x.x") + << MongosType::configVersion(0)); + string errMsg; + ASSERT(mongos.parseBSON(obj, &errMsg)); + ASSERT_EQUALS(errMsg, ""); + ASSERT_TRUE(mongos.isValid(NULL)); + ASSERT_EQUALS(mongos.getName(), "localhost:27017"); + ASSERT_EQUALS(mongos.getPing(), Date_t::fromMillisSinceEpoch(1)); + ASSERT_EQUALS(mongos.getUp(), 100); + ASSERT_EQUALS(mongos.getWaiting(), false); + ASSERT_EQUALS(mongos.getMongoVersion(), "x.x.x"); + ASSERT_EQUALS(mongos.getConfigVersion(), 0); +} - TEST(Validity, BadType) { - MongosType mongos; - BSONObj obj = BSON(MongosType::name() << 0); - string errMsg; - ASSERT((!mongos.parseBSON(obj, &errMsg)) && (errMsg != "")); - } +TEST(Validity, BadType) { + MongosType mongos; + BSONObj obj = BSON(MongosType::name() << 0); + string errMsg; + ASSERT((!mongos.parseBSON(obj, &errMsg)) && (errMsg != "")); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/version_manager.cpp b/src/mongo/s/version_manager.cpp index ad96e57b3b7..17c624e541d 100644 --- a/src/mongo/s/version_manager.cpp +++ b/src/mongo/s/version_manager.cpp @@ -51,396 +51,386 @@ namespace mongo { - using std::shared_ptr; - using std::endl; - using std::map; - using std::string; - - // Global version manager - VersionManager versionManager; - - /** - * Tracking information, per-connection, of the latest chunk manager iteration or sequence - * number that was used to send a shard version over this connection. - * When the chunk manager is replaced, implying new versions were loaded, the chunk manager - * sequence number is iterated by 1 and connections need to re-send shard versions. - */ - struct ConnectionShardStatus { - - bool hasAnySequenceSet(DBClientBase* conn) { - stdx::lock_guard<stdx::mutex> lk(_mutex); - - SequenceMap::const_iterator seenConnIt = _map.find(conn->getConnectionId()); - return seenConnIt != _map.end() && seenConnIt->second.size() > 0; - } +using std::shared_ptr; +using std::endl; +using std::map; +using std::string; - bool getSequence(DBClientBase * conn, - const string& ns, - unsigned long long* sequence) { +// Global version manager +VersionManager versionManager; - stdx::lock_guard<stdx::mutex> lk(_mutex); +/** + * Tracking information, per-connection, of the latest chunk manager iteration or sequence + * number that was used to send a shard version over this connection. + * When the chunk manager is replaced, implying new versions were loaded, the chunk manager + * sequence number is iterated by 1 and connections need to re-send shard versions. + */ +struct ConnectionShardStatus { + bool hasAnySequenceSet(DBClientBase* conn) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + + SequenceMap::const_iterator seenConnIt = _map.find(conn->getConnectionId()); + return seenConnIt != _map.end() && seenConnIt->second.size() > 0; + } - SequenceMap::const_iterator seenConnIt = _map.find(conn->getConnectionId()); - if (seenConnIt == _map.end()) - return false; + bool getSequence(DBClientBase* conn, const string& ns, unsigned long long* sequence) { + stdx::lock_guard<stdx::mutex> lk(_mutex); - map<string, unsigned long long>::const_iterator seenNSIt = seenConnIt->second.find(ns); - if (seenNSIt == seenConnIt->second.end()) - return false; + SequenceMap::const_iterator seenConnIt = _map.find(conn->getConnectionId()); + if (seenConnIt == _map.end()) + return false; - *sequence = seenNSIt->second; - return true; - } + map<string, unsigned long long>::const_iterator seenNSIt = seenConnIt->second.find(ns); + if (seenNSIt == seenConnIt->second.end()) + return false; - void setSequence( DBClientBase * conn , const string& ns , const unsigned long long& s ) { - stdx::lock_guard<stdx::mutex> lk( _mutex ); - _map[conn->getConnectionId()][ns] = s; - } + *sequence = seenNSIt->second; + return true; + } - void reset( DBClientBase * conn ) { - stdx::lock_guard<stdx::mutex> lk( _mutex ); - _map.erase( conn->getConnectionId() ); - } + void setSequence(DBClientBase* conn, const string& ns, const unsigned long long& s) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _map[conn->getConnectionId()][ns] = s; + } - // protects _map - stdx::mutex _mutex; + void reset(DBClientBase* conn) { + stdx::lock_guard<stdx::mutex> lk(_mutex); + _map.erase(conn->getConnectionId()); + } - // a map from a connection into ChunkManager's sequence number for each namespace - typedef map<unsigned long long, map<string,unsigned long long> > SequenceMap; - SequenceMap _map; + // protects _map + stdx::mutex _mutex; - } connectionShardStatus; + // a map from a connection into ChunkManager's sequence number for each namespace + typedef map<unsigned long long, map<string, unsigned long long>> SequenceMap; + SequenceMap _map; - void VersionManager::resetShardVersionCB( DBClientBase * conn ) { - connectionShardStatus.reset( conn ); - } +} connectionShardStatus; - bool VersionManager::isVersionableCB(DBClientBase* conn) { - // We do not version shard connections when issued from mongod - if (!isMongos()) { - return false; - } +void VersionManager::resetShardVersionCB(DBClientBase* conn) { + connectionShardStatus.reset(conn); +} - return conn->type() == ConnectionString::MASTER || conn->type() == ConnectionString::SET; +bool VersionManager::isVersionableCB(DBClientBase* conn) { + // We do not version shard connections when issued from mongod + if (!isMongos()) { + return false; } - DBClientBase* getVersionable(DBClientBase* conn) { - switch (conn->type()) { + return conn->type() == ConnectionString::MASTER || conn->type() == ConnectionString::SET; +} + +DBClientBase* getVersionable(DBClientBase* conn) { + switch (conn->type()) { case ConnectionString::INVALID: - massert(15904, str::stream() << "cannot set version on invalid connection " - << conn->toString(), false); + massert(15904, + str::stream() << "cannot set version on invalid connection " + << conn->toString(), + false); return nullptr; case ConnectionString::MASTER: return conn; case ConnectionString::SYNC: - massert(15906, str::stream() << "cannot set version or shard on sync connection " - << conn->toString(), false); + massert(15906, + str::stream() << "cannot set version or shard on sync connection " + << conn->toString(), + false); return nullptr; case ConnectionString::CUSTOM: - massert(16334, str::stream() << "cannot set version or shard on custom connection " - << conn->toString(), false); + massert(16334, + str::stream() << "cannot set version or shard on custom connection " + << conn->toString(), + false); return nullptr; case ConnectionString::SET: DBClientReplicaSet* set = (DBClientReplicaSet*)conn; return &(set->masterConn()); - } - - MONGO_UNREACHABLE; } - bool VersionManager::forceRemoteCheckShardVersionCB(const string& ns) { - const NamespaceString nss(ns); + MONGO_UNREACHABLE; +} - // This will force the database catalog entry to be reloaded - grid.catalogCache()->invalidate(nss.db().toString()); +bool VersionManager::forceRemoteCheckShardVersionCB(const string& ns) { + const NamespaceString nss(ns); - auto status = grid.catalogCache()->getDatabase(nss.db().toString()); - if (!status.isOK()) { - return false; - } + // This will force the database catalog entry to be reloaded + grid.catalogCache()->invalidate(nss.db().toString()); - shared_ptr<DBConfig> conf = status.getValue(); - - // If we don't have a collection, don't refresh the chunk manager - if (nsGetCollection(ns).size() == 0) { - return false; - } - - ChunkManagerPtr manager = conf->getChunkManagerIfExists(ns, true, true); - if (!manager) { - return false; - } + auto status = grid.catalogCache()->getDatabase(nss.db().toString()); + if (!status.isOK()) { + return false; + } - return true; + shared_ptr<DBConfig> conf = status.getValue(); + // If we don't have a collection, don't refresh the chunk manager + if (nsGetCollection(ns).size() == 0) { + return false; } - /** - * Special internal logic to run reduced version handshake for empty namespace operations to - * shards. - * - * Eventually this should go completely away, but for now many commands rely on unversioned but - * mongos-specific behavior on mongod (auditing and replication information in commands) - */ - static bool initShardVersionEmptyNS(DBClientBase * conn_in) { - - bool ok; - BSONObj result; - DBClientBase* conn = NULL; - try { - // May throw if replica set primary is down - conn = getVersionable( conn_in ); - dassert( conn ); // errors thrown above - - // Check to see if we've already initialized this connection - if (connectionShardStatus.hasAnySequenceSet(conn)) - return false; - - // Check to see if this is actually a shard and not a single config server - // NOTE: Config servers are registered only by the name "config" in the shard cache, not - // by host, so lookup by host will fail unless the host is also a shard. - const auto shard = grid.shardRegistry()->getShard(conn->getServerAddress()); - if (!shard) { - return false; - } - - LOG(1) << "initializing shard connection to " << shard->toString() << endl; - - ok = setShardVersion(*conn, - "", - grid.catalogManager()->connectionString().toString(), - ChunkVersion(), - NULL, - true, - result); - } - catch( const DBException& ) { - - // NOTE: Replica sets may fail to initShardVersion because future calls relying on - // correct versioning must later call checkShardVersion on the primary. - // Secondary queries and commands may not call checkShardVersion, but secondary ops - // aren't versioned at all. - if ( conn_in->type() != ConnectionString::SET ) { - throw; - } + ChunkManagerPtr manager = conf->getChunkManagerIfExists(ns, true, true); + if (!manager) { + return false; + } - // NOTE: Only old-style cluster operations will talk via DBClientReplicaSets - using - // checkShardVersion is required (which includes initShardVersion information) if these - // connections are used. + return true; +} - OCCASIONALLY { - warning() << "failed to initialize new replica set connection version, " - << "will initialize on first use" << endl; - } +/** + * Special internal logic to run reduced version handshake for empty namespace operations to + * shards. + * + * Eventually this should go completely away, but for now many commands rely on unversioned but + * mongos-specific behavior on mongod (auditing and replication information in commands) + */ +static bool initShardVersionEmptyNS(DBClientBase* conn_in) { + bool ok; + BSONObj result; + DBClientBase* conn = NULL; + try { + // May throw if replica set primary is down + conn = getVersionable(conn_in); + dassert(conn); // errors thrown above + + // Check to see if we've already initialized this connection + if (connectionShardStatus.hasAnySequenceSet(conn)) + return false; + // Check to see if this is actually a shard and not a single config server + // NOTE: Config servers are registered only by the name "config" in the shard cache, not + // by host, so lookup by host will fail unless the host is also a shard. + const auto shard = grid.shardRegistry()->getShard(conn->getServerAddress()); + if (!shard) { return false; } - // Record the connection wire version if sent in the response, initShardVersion is a - // handshake for mongos->mongod connections. - if ( !result["minWireVersion"].eoo() ) { + LOG(1) << "initializing shard connection to " << shard->toString() << endl; + + ok = setShardVersion(*conn, + "", + grid.catalogManager()->connectionString().toString(), + ChunkVersion(), + NULL, + true, + result); + } catch (const DBException&) { + // NOTE: Replica sets may fail to initShardVersion because future calls relying on + // correct versioning must later call checkShardVersion on the primary. + // Secondary queries and commands may not call checkShardVersion, but secondary ops + // aren't versioned at all. + if (conn_in->type() != ConnectionString::SET) { + throw; + } + + // NOTE: Only old-style cluster operations will talk via DBClientReplicaSets - using + // checkShardVersion is required (which includes initShardVersion information) if these + // connections are used. - int minWireVersion = result["minWireVersion"].numberInt(); - int maxWireVersion = result["maxWireVersion"].numberInt(); - conn->setWireVersions( minWireVersion, maxWireVersion ); + OCCASIONALLY { + warning() << "failed to initialize new replica set connection version, " + << "will initialize on first use" << endl; } - LOG(3) << "initial sharding result : " << result << endl; + return false; + } - connectionShardStatus.setSequence(conn, "", 0); - return ok; + // Record the connection wire version if sent in the response, initShardVersion is a + // handshake for mongos->mongod connections. + if (!result["minWireVersion"].eoo()) { + int minWireVersion = result["minWireVersion"].numberInt(); + int maxWireVersion = result["maxWireVersion"].numberInt(); + conn->setWireVersions(minWireVersion, maxWireVersion); } - /** - * Updates the remote cached version on the remote shard host (primary, in the case of replica - * sets) if needed with a fully-qualified shard version for the given namespace: - * config server(s) + shard name + shard version - * - * If no remote cached version has ever been set, an initial shard version is sent. - * - * If the namespace is empty and no version has ever been sent, the config server + shard name - * is sent to the remote shard host to initialize the connection as coming from mongos. - * NOTE: This initialization is *best-effort only*. Operations which wish to correctly version - * must send the namespace. - * - * Config servers are special and are not (unless otherwise a shard) kept up to date with this - * protocol. This is safe so long as config servers only contain unversioned collections. - * - * It is an error to call checkShardVersion with an unversionable connection (isVersionableCB). - * - * @return true if we contacted the remote host - */ - bool checkShardVersion(DBClientBase* conn_in, - const string& ns, - ChunkManagerPtr refManager, - bool authoritative, - int tryNumber) { - - // TODO: cache, optimize, etc... - - // Empty namespaces are special - we require initialization but not versioning - if (ns.size() == 0) { - return initShardVersionEmptyNS(conn_in); - } + LOG(3) << "initial sharding result : " << result << endl; - auto status = grid.catalogCache()->getDatabase(nsToDatabase(ns)); - if (!status.isOK()) { - return false; - } + connectionShardStatus.setSequence(conn, "", 0); + return ok; +} - shared_ptr<DBConfig> conf = status.getValue(); +/** + * Updates the remote cached version on the remote shard host (primary, in the case of replica + * sets) if needed with a fully-qualified shard version for the given namespace: + * config server(s) + shard name + shard version + * + * If no remote cached version has ever been set, an initial shard version is sent. + * + * If the namespace is empty and no version has ever been sent, the config server + shard name + * is sent to the remote shard host to initialize the connection as coming from mongos. + * NOTE: This initialization is *best-effort only*. Operations which wish to correctly version + * must send the namespace. + * + * Config servers are special and are not (unless otherwise a shard) kept up to date with this + * protocol. This is safe so long as config servers only contain unversioned collections. + * + * It is an error to call checkShardVersion with an unversionable connection (isVersionableCB). + * + * @return true if we contacted the remote host + */ +bool checkShardVersion(DBClientBase* conn_in, + const string& ns, + ChunkManagerPtr refManager, + bool authoritative, + int tryNumber) { + // TODO: cache, optimize, etc... + + // Empty namespaces are special - we require initialization but not versioning + if (ns.size() == 0) { + return initShardVersionEmptyNS(conn_in); + } - DBClientBase* conn = getVersionable( conn_in ); - verify(conn); // errors thrown above + auto status = grid.catalogCache()->getDatabase(nsToDatabase(ns)); + if (!status.isOK()) { + return false; + } - unsigned long long officialSequenceNumber = 0; + shared_ptr<DBConfig> conf = status.getValue(); - ShardPtr primary; - ChunkManagerPtr manager; - if (authoritative) - conf->getChunkManagerIfExists(ns, true); + DBClientBase* conn = getVersionable(conn_in); + verify(conn); // errors thrown above - conf->getChunkManagerOrPrimary(ns, manager, primary); + unsigned long long officialSequenceNumber = 0; - if (manager) { - officialSequenceNumber = manager->getSequenceNumber(); - } + ShardPtr primary; + ChunkManagerPtr manager; + if (authoritative) + conf->getChunkManagerIfExists(ns, true); - const auto shard = grid.shardRegistry()->getShard(conn->getServerAddress()); - uassert(ErrorCodes::ShardNotFound, - str::stream() << conn->getServerAddress() << " is not recognized as a shard", - shard); - - // Check this manager against the reference manager - if (manager) { - - if (refManager && !refManager->compatibleWith(*manager, shard->getId())) { - const ChunkVersion refVersion(refManager->getVersion(shard->getId())); - const ChunkVersion currentVersion(manager->getVersion(shard->getId())); - - string msg(str::stream() << "manager (" - << currentVersion.toString() - << " : " << manager->getSequenceNumber() << ") " - << "not compatible with reference manager (" - << refVersion.toString() - << " : " << refManager->getSequenceNumber() << ") " - << "on shard " << shard->getId() - << " (" << shard->getConnString().toString() << ")"); - - throw SendStaleConfigException(ns, - msg, - refVersion, - currentVersion); - } - } - else if (refManager) { - - string msg( str::stream() << "not sharded (" - << ( (manager.get() == 0) ? string( "<none>" ) : - str::stream() << manager->getSequenceNumber() ) - << ") but has reference manager (" - << refManager->getSequenceNumber() << ") " - << "on conn " << conn->getServerAddress() << " (" - << conn_in->getServerAddress() << ")" ); - - throw SendStaleConfigException(ns, - msg, - refManager->getVersion(shard->getId()), - ChunkVersion::UNSHARDED()); - } + conf->getChunkManagerOrPrimary(ns, manager, primary); - // Do not send setShardVersion to collections on the config servers - this causes problems - // when config servers are also shards and get SSV with conflicting names. - // TODO: Make config servers regular shards - if (primary && primary->getId() == "config") { - return false; - } + if (manager) { + officialSequenceNumber = manager->getSequenceNumber(); + } - // Has the ChunkManager been reloaded since the last time we updated the shard version over - // this connection? If we've never updated the shard version, do so now. - unsigned long long sequenceNumber = 0; - if (connectionShardStatus.getSequence(conn, ns, &sequenceNumber)) { - if (sequenceNumber == officialSequenceNumber) { - return false; - } + const auto shard = grid.shardRegistry()->getShard(conn->getServerAddress()); + uassert(ErrorCodes::ShardNotFound, + str::stream() << conn->getServerAddress() << " is not recognized as a shard", + shard); + + // Check this manager against the reference manager + if (manager) { + if (refManager && !refManager->compatibleWith(*manager, shard->getId())) { + const ChunkVersion refVersion(refManager->getVersion(shard->getId())); + const ChunkVersion currentVersion(manager->getVersion(shard->getId())); + + string msg(str::stream() + << "manager (" << currentVersion.toString() << " : " + << manager->getSequenceNumber() << ") " + << "not compatible with reference manager (" << refVersion.toString() + << " : " << refManager->getSequenceNumber() << ") " + << "on shard " << shard->getId() << " (" << shard->getConnString().toString() + << ")"); + + throw SendStaleConfigException(ns, msg, refVersion, currentVersion); } + } else if (refManager) { + string msg(str::stream() << "not sharded (" + << ((manager.get() == 0) ? string("<none>") : str::stream() + << manager->getSequenceNumber()) + << ") but has reference manager (" + << refManager->getSequenceNumber() << ") " + << "on conn " << conn->getServerAddress() << " (" + << conn_in->getServerAddress() << ")"); + + throw SendStaleConfigException( + ns, msg, refManager->getVersion(shard->getId()), ChunkVersion::UNSHARDED()); + } - ChunkVersion version = ChunkVersion(0, 0, OID()); - if (manager) { - version = manager->getVersion(shard->getId()); - } + // Do not send setShardVersion to collections on the config servers - this causes problems + // when config servers are also shards and get SSV with conflicting names. + // TODO: Make config servers regular shards + if (primary && primary->getId() == "config") { + return false; + } - LOG(1) << "setting shard version of " << version << " for " << ns << " on shard " - << shard->toString(); + // Has the ChunkManager been reloaded since the last time we updated the shard version over + // this connection? If we've never updated the shard version, do so now. + unsigned long long sequenceNumber = 0; + if (connectionShardStatus.getSequence(conn, ns, &sequenceNumber)) { + if (sequenceNumber == officialSequenceNumber) { + return false; + } + } - LOG(3) << "last version sent with chunk manager iteration " << sequenceNumber - << ", current chunk manager iteration is " << officialSequenceNumber; + ChunkVersion version = ChunkVersion(0, 0, OID()); + if (manager) { + version = manager->getVersion(shard->getId()); + } - BSONObj result; - if (setShardVersion(*conn, - ns, - grid.catalogManager()->connectionString().toString(), - version, - manager.get(), - authoritative, - result)) { + LOG(1) << "setting shard version of " << version << " for " << ns << " on shard " + << shard->toString(); + + LOG(3) << "last version sent with chunk manager iteration " << sequenceNumber + << ", current chunk manager iteration is " << officialSequenceNumber; + + BSONObj result; + if (setShardVersion(*conn, + ns, + grid.catalogManager()->connectionString().toString(), + version, + manager.get(), + authoritative, + result)) { + LOG(1) << " setShardVersion success: " << result; + connectionShardStatus.setSequence(conn, ns, officialSequenceNumber); + return true; + } - LOG(1) << " setShardVersion success: " << result; - connectionShardStatus.setSequence( conn , ns , officialSequenceNumber ); - return true; - } + LOG(1) << " setShardVersion failed!\n" << result << endl; - LOG(1) << " setShardVersion failed!\n" << result << endl; + if (result["need_authoritative"].trueValue()) + massert(10428, "need_authoritative set but in authoritative mode already", !authoritative); - if ( result["need_authoritative"].trueValue() ) - massert( 10428 , "need_authoritative set but in authoritative mode already" , ! authoritative ); + if (!authoritative) { + // use the original connection and get a fresh versionable connection + // since conn can be invalidated (or worse, freed) after the failure + checkShardVersion(conn_in, ns, refManager, 1, tryNumber + 1); + return true; + } - if ( ! authoritative ) { - // use the original connection and get a fresh versionable connection - // since conn can be invalidated (or worse, freed) after the failure - checkShardVersion(conn_in, ns, refManager, 1, tryNumber + 1); - return true; - } - - if ( result["reloadConfig"].trueValue() ) { - if( result["version"].timestampTime() == Date_t() ){ - - warning() << "reloading full configuration for " << conf->name() - << ", connection state indicates significant version changes"; - - // reload db - conf->reload(); - } - else { - // reload config - conf->getChunkManager( ns , true ); - } - } + if (result["reloadConfig"].trueValue()) { + if (result["version"].timestampTime() == Date_t()) { + warning() << "reloading full configuration for " << conf->name() + << ", connection state indicates significant version changes"; - const int maxNumTries = 7; - if ( tryNumber < maxNumTries ) { - LOG( tryNumber < ( maxNumTries / 2 ) ? 1 : 0 ) - << "going to retry checkShardVersion shard: " << shard->toString() << " " << result; - sleepmillis( 10 * tryNumber ); - // use the original connection and get a fresh versionable connection - // since conn can be invalidated (or worse, freed) after the failure - checkShardVersion(conn_in, ns, refManager, true, tryNumber + 1); - return true; + // reload db + conf->reload(); + } else { + // reload config + conf->getChunkManager(ns, true); } - - string errmsg = str::stream() << "setShardVersion failed shard: " << shard->toString() - << " " << result; - log() << " " << errmsg << endl; - massert( 10429 , errmsg , 0 ); - return true; } - bool VersionManager::checkShardVersionCB( DBClientBase* conn_in , const string& ns , bool authoritative , int tryNumber ) { - return checkShardVersion( conn_in, ns, ChunkManagerPtr(), authoritative, tryNumber ); + const int maxNumTries = 7; + if (tryNumber < maxNumTries) { + LOG(tryNumber < (maxNumTries / 2) ? 1 : 0) + << "going to retry checkShardVersion shard: " << shard->toString() << " " << result; + sleepmillis(10 * tryNumber); + // use the original connection and get a fresh versionable connection + // since conn can be invalidated (or worse, freed) after the failure + checkShardVersion(conn_in, ns, refManager, true, tryNumber + 1); + return true; } - bool VersionManager::checkShardVersionCB( ShardConnection* conn_in , bool authoritative , int tryNumber ) { - return checkShardVersion( conn_in->get(), conn_in->getNS(), conn_in->getManager(), authoritative, tryNumber ); - } + string errmsg = str::stream() << "setShardVersion failed shard: " << shard->toString() << " " + << result; + log() << " " << errmsg << endl; + massert(10429, errmsg, 0); + return true; +} + +bool VersionManager::checkShardVersionCB(DBClientBase* conn_in, + const string& ns, + bool authoritative, + int tryNumber) { + return checkShardVersion(conn_in, ns, ChunkManagerPtr(), authoritative, tryNumber); +} + +bool VersionManager::checkShardVersionCB(ShardConnection* conn_in, + bool authoritative, + int tryNumber) { + return checkShardVersion( + conn_in->get(), conn_in->getNS(), conn_in->getManager(), authoritative, tryNumber); +} } // namespace mongo diff --git a/src/mongo/s/version_manager.h b/src/mongo/s/version_manager.h index d1903b2d3eb..6dc6877d574 100644 --- a/src/mongo/s/version_manager.h +++ b/src/mongo/s/version_manager.h @@ -32,22 +32,21 @@ namespace mongo { - class DBClientBase; - class ShardConnection; +class DBClientBase; +class ShardConnection; - class VersionManager { - public: - VersionManager() { } +class VersionManager { +public: + VersionManager() {} - bool isVersionableCB(DBClientBase*); - bool forceRemoteCheckShardVersionCB(const std::string&); - bool checkShardVersionCB(DBClientBase*, const std::string&, bool, int); - bool checkShardVersionCB(ShardConnection*, bool, int); - void resetShardVersionCB(DBClientBase*); + bool isVersionableCB(DBClientBase*); + bool forceRemoteCheckShardVersionCB(const std::string&); + bool checkShardVersionCB(DBClientBase*, const std::string&, bool, int); + bool checkShardVersionCB(ShardConnection*, bool, int); + void resetShardVersionCB(DBClientBase*); +}; - }; +extern VersionManager versionManager; - extern VersionManager versionManager; - -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/version_mongos.cpp b/src/mongo/s/version_mongos.cpp index 1b974512643..9ec5d9be000 100644 --- a/src/mongo/s/version_mongos.cpp +++ b/src/mongo/s/version_mongos.cpp @@ -42,23 +42,21 @@ namespace mongo { - void printShardingVersionInfo( bool out ) { - if ( out ) { - std::cout << "MongoS version " << versionString << " starting: pid=" << - ProcessId::getCurrent() << " port=" << serverGlobalParams.port << - ( sizeof(int*) == 4 ? " 32" : " 64" ) << "-bit host=" << getHostNameCached() << - " (--help for usage)" << std::endl; - DEV std::cout << "DEBUG build" << std::endl; - std::cout << "git version: " << gitVersion() << std::endl; - std::cout << openSSLVersion("OpenSSL version: ") << std::endl; - } - else { - log() << "MongoS version " << versionString << " starting: pid=" << - ProcessId::getCurrent() << " port=" << serverGlobalParams.port << - ( sizeof( int* ) == 4 ? " 32" : " 64" ) << "-bit host=" << getHostNameCached() << - " (--help for usage)" << std::endl; - DEV log() << "DEBUG build" << std::endl; - logProcessDetails(); - } +void printShardingVersionInfo(bool out) { + if (out) { + std::cout << "MongoS version " << versionString + << " starting: pid=" << ProcessId::getCurrent() + << " port=" << serverGlobalParams.port << (sizeof(int*) == 4 ? " 32" : " 64") + << "-bit host=" << getHostNameCached() << " (--help for usage)" << std::endl; + DEV std::cout << "DEBUG build" << std::endl; + std::cout << "git version: " << gitVersion() << std::endl; + std::cout << openSSLVersion("OpenSSL version: ") << std::endl; + } else { + log() << "MongoS version " << versionString << " starting: pid=" << ProcessId::getCurrent() + << " port=" << serverGlobalParams.port << (sizeof(int*) == 4 ? " 32" : " 64") + << "-bit host=" << getHostNameCached() << " (--help for usage)" << std::endl; + DEV log() << "DEBUG build" << std::endl; + logProcessDetails(); } -} // namespace mongo +} +} // namespace mongo diff --git a/src/mongo/s/version_mongos.h b/src/mongo/s/version_mongos.h index 491c36da5da..6367d9185d0 100644 --- a/src/mongo/s/version_mongos.h +++ b/src/mongo/s/version_mongos.h @@ -27,6 +27,6 @@ */ namespace mongo { - // print mongos version info - void printShardingVersionInfo(bool out); -} // namespace mongo +// print mongos version info +void printShardingVersionInfo(bool out); +} // namespace mongo diff --git a/src/mongo/s/write_ops/batch_downconvert.cpp b/src/mongo/s/write_ops/batch_downconvert.cpp index 3bcafcf31a1..f786606d9b6 100644 --- a/src/mongo/s/write_ops/batch_downconvert.cpp +++ b/src/mongo/s/write_ops/batch_downconvert.cpp @@ -40,274 +40,254 @@ namespace mongo { - using std::endl; - using std::string; - using std::vector; - - Status extractGLEErrors( const BSONObj& gleResponse, GLEErrors* errors ) { - - // DRAGONS - // Parsing GLE responses is incredibly finicky. - // The order of testing here is extremely important. - - /////////////////////////////////////////////////////////////////////// - // IMPORTANT! - // Also update extractGLEErrors in batch_api.js for any changes made here. - - const bool isOK = gleResponse["ok"].trueValue(); - const string err = gleResponse["err"].str(); - const string errMsg = gleResponse["errmsg"].str(); - const string wNote = gleResponse["wnote"].str(); - const string jNote = gleResponse["jnote"].str(); - const int code = gleResponse["code"].numberInt(); - const bool timeout = gleResponse["wtimeout"].trueValue(); - - if ( err == "norepl" || err == "noreplset" ) { - // Know this is legacy gle and the repl not enforced - write concern error in 2.4 - errors->wcError.reset( new WCErrorDetail ); - errors->wcError->setErrCode( ErrorCodes::WriteConcernFailed ); - if ( !errMsg.empty() ) { - errors->wcError->setErrMessage( errMsg ); - } - else if ( !wNote.empty() ) { - errors->wcError->setErrMessage( wNote ); - } - else { - errors->wcError->setErrMessage( err ); - } - } - else if ( timeout ) { - // Know there was no write error - errors->wcError.reset( new WCErrorDetail ); - errors->wcError->setErrCode( ErrorCodes::WriteConcernFailed ); - if ( !errMsg.empty() ) { - errors->wcError->setErrMessage( errMsg ); - } - else { - errors->wcError->setErrMessage( err ); - } - errors->wcError->setErrInfo( BSON( "wtimeout" << true ) ); - } - else if ( code == 10990 /* no longer primary */ - || code == 16805 /* replicatedToNum no longer primary */ - || code == 14830 /* gle wmode changed / invalid */ - // 2.6 Error codes - || code == ErrorCodes::NotMaster - || code == ErrorCodes::UnknownReplWriteConcern - || code == ErrorCodes::WriteConcernFailed ) { - // Write concern errors that get returned as regular errors (result may not be ok: 1.0) - errors->wcError.reset( new WCErrorDetail ); - errors->wcError->setErrCode( code ); - errors->wcError->setErrMessage( errMsg ); +using std::endl; +using std::string; +using std::vector; + +Status extractGLEErrors(const BSONObj& gleResponse, GLEErrors* errors) { + // DRAGONS + // Parsing GLE responses is incredibly finicky. + // The order of testing here is extremely important. + + /////////////////////////////////////////////////////////////////////// + // IMPORTANT! + // Also update extractGLEErrors in batch_api.js for any changes made here. + + const bool isOK = gleResponse["ok"].trueValue(); + const string err = gleResponse["err"].str(); + const string errMsg = gleResponse["errmsg"].str(); + const string wNote = gleResponse["wnote"].str(); + const string jNote = gleResponse["jnote"].str(); + const int code = gleResponse["code"].numberInt(); + const bool timeout = gleResponse["wtimeout"].trueValue(); + + if (err == "norepl" || err == "noreplset") { + // Know this is legacy gle and the repl not enforced - write concern error in 2.4 + errors->wcError.reset(new WCErrorDetail); + errors->wcError->setErrCode(ErrorCodes::WriteConcernFailed); + if (!errMsg.empty()) { + errors->wcError->setErrMessage(errMsg); + } else if (!wNote.empty()) { + errors->wcError->setErrMessage(wNote); + } else { + errors->wcError->setErrMessage(err); } - else if ( !isOK ) { - - // - // !!! SOME GLE ERROR OCCURRED, UNKNOWN WRITE RESULT !!! - // - - return Status( DBException::convertExceptionCode( - code ? code : ErrorCodes::UnknownError ), - errMsg ); + } else if (timeout) { + // Know there was no write error + errors->wcError.reset(new WCErrorDetail); + errors->wcError->setErrCode(ErrorCodes::WriteConcernFailed); + if (!errMsg.empty()) { + errors->wcError->setErrMessage(errMsg); + } else { + errors->wcError->setErrMessage(err); } - else if ( !err.empty() ) { - // Write error - errors->writeError.reset( new WriteErrorDetail ); - int writeErrorCode = code == 0 ? ErrorCodes::UnknownError : code; - - // COMPATIBILITY - // Certain clients expect write commands to always report 11000 for duplicate key - // errors, while legacy GLE can return additional codes. - if ( writeErrorCode == 11001 /* dup key in update */ - || writeErrorCode == 12582 /* dup key capped */) { - writeErrorCode = ErrorCodes::DuplicateKey; - } - - errors->writeError->setErrCode( writeErrorCode ); - errors->writeError->setErrMessage( err ); - } - else if ( !jNote.empty() ) { - // Know this is legacy gle and the journaling not enforced - write concern error in 2.4 - errors->wcError.reset( new WCErrorDetail ); - errors->wcError->setErrCode( ErrorCodes::WriteConcernFailed ); - errors->wcError->setErrMessage( jNote ); + errors->wcError->setErrInfo(BSON("wtimeout" << true)); + } else if (code == 10990 /* no longer primary */ + || + code == 16805 /* replicatedToNum no longer primary */ + || + code == 14830 /* gle wmode changed / invalid */ + // 2.6 Error codes + || + code == ErrorCodes::NotMaster || code == ErrorCodes::UnknownReplWriteConcern || + code == ErrorCodes::WriteConcernFailed) { + // Write concern errors that get returned as regular errors (result may not be ok: 1.0) + errors->wcError.reset(new WCErrorDetail); + errors->wcError->setErrCode(code); + errors->wcError->setErrMessage(errMsg); + } else if (!isOK) { + // + // !!! SOME GLE ERROR OCCURRED, UNKNOWN WRITE RESULT !!! + // + + return Status(DBException::convertExceptionCode(code ? code : ErrorCodes::UnknownError), + errMsg); + } else if (!err.empty()) { + // Write error + errors->writeError.reset(new WriteErrorDetail); + int writeErrorCode = code == 0 ? ErrorCodes::UnknownError : code; + + // COMPATIBILITY + // Certain clients expect write commands to always report 11000 for duplicate key + // errors, while legacy GLE can return additional codes. + if (writeErrorCode == 11001 /* dup key in update */ + || + writeErrorCode == 12582 /* dup key capped */) { + writeErrorCode = ErrorCodes::DuplicateKey; } - return Status::OK(); + errors->writeError->setErrCode(writeErrorCode); + errors->writeError->setErrMessage(err); + } else if (!jNote.empty()) { + // Know this is legacy gle and the journaling not enforced - write concern error in 2.4 + errors->wcError.reset(new WCErrorDetail); + errors->wcError->setErrCode(ErrorCodes::WriteConcernFailed); + errors->wcError->setErrMessage(jNote); } - /** - * Suppress the "err" and "code" field if they are coming from a previous write error and - * are not related to write concern. Also removes any write stats information (e.g. "n") - * - * Also, In some cases, 2.4 GLE w/ wOpTime can give us duplicate "err" and "code" fields b/c of - * reporting a previous error. The later field is what we want - dedup and use later field. - * - * Returns the stripped GLE response. - */ - BSONObj stripNonWCInfo( const BSONObj& gleResponse ) { - - BSONObjIterator it( gleResponse ); - BSONObjBuilder builder; - - BSONElement codeField; // eoo - BSONElement errField; // eoo - - while ( it.more() ) { - BSONElement el = it.next(); - StringData fieldName( el.fieldName() ); - if ( fieldName.compare( "err" ) == 0 ) { - errField = el; - } - else if ( fieldName.compare( "code" ) == 0 ) { - codeField = el; - } - else if ( fieldName.compare( "n" ) == 0 || fieldName.compare( "nModified" ) == 0 - || fieldName.compare( "upserted" ) == 0 - || fieldName.compare( "updatedExisting" ) == 0 ) { - // Suppress field - } - else { - builder.append( el ); - } - } + return Status::OK(); +} - if ( !codeField.eoo() ) { - if ( !gleResponse["ok"].trueValue() ) { - // The last code will be from the write concern - builder.append( codeField ); - } - else { - // The code is from a non-wc error on this connection - suppress it - } +/** + * Suppress the "err" and "code" field if they are coming from a previous write error and + * are not related to write concern. Also removes any write stats information (e.g. "n") + * + * Also, In some cases, 2.4 GLE w/ wOpTime can give us duplicate "err" and "code" fields b/c of + * reporting a previous error. The later field is what we want - dedup and use later field. + * + * Returns the stripped GLE response. + */ +BSONObj stripNonWCInfo(const BSONObj& gleResponse) { + BSONObjIterator it(gleResponse); + BSONObjBuilder builder; + + BSONElement codeField; // eoo + BSONElement errField; // eoo + + while (it.more()) { + BSONElement el = it.next(); + StringData fieldName(el.fieldName()); + if (fieldName.compare("err") == 0) { + errField = el; + } else if (fieldName.compare("code") == 0) { + codeField = el; + } else if (fieldName.compare("n") == 0 || fieldName.compare("nModified") == 0 || + fieldName.compare("upserted") == 0 || + fieldName.compare("updatedExisting") == 0) { + // Suppress field + } else { + builder.append(el); } + } - if ( !errField.eoo() ) { - string err = errField.str(); - if ( err == "norepl" || err == "noreplset" || err == "timeout" ) { - // Append err if it's from a write concern issue - builder.append( errField ); - } - else { - // Suppress non-write concern err as null, but we need to report null err if ok - if ( gleResponse["ok"].trueValue() ) - builder.appendNull( errField.fieldName() ); - } + if (!codeField.eoo()) { + if (!gleResponse["ok"].trueValue()) { + // The last code will be from the write concern + builder.append(codeField); + } else { + // The code is from a non-wc error on this connection - suppress it } - - return builder.obj(); } - // Adds a wOpTime and a wElectionId field to a set of gle options - static BSONObj buildGLECmdWithOpTime( const BSONObj& gleOptions, - const Timestamp& opTime, - const OID& electionId ) { - BSONObjBuilder builder; - BSONObjIterator it( gleOptions ); - - for ( int i = 0; it.more(); ++i ) { - BSONElement el = it.next(); - - // Make sure first element is getLastError : 1 - if ( i == 0 ) { - StringData elName( el.fieldName() ); - if ( !elName.equalCaseInsensitive( "getLastError" ) ) { - builder.append( "getLastError", 1 ); - } - } - - builder.append( el ); + if (!errField.eoo()) { + string err = errField.str(); + if (err == "norepl" || err == "noreplset" || err == "timeout") { + // Append err if it's from a write concern issue + builder.append(errField); + } else { + // Suppress non-write concern err as null, but we need to report null err if ok + if (gleResponse["ok"].trueValue()) + builder.appendNull(errField.fieldName()); } - builder.append( "wOpTime", opTime ); - builder.appendOID( "wElectionId", const_cast<OID*>(&electionId) ); - return builder.obj(); } - Status enforceLegacyWriteConcern( MultiCommandDispatch* dispatcher, - StringData dbName, - const BSONObj& options, - const HostOpTimeMap& hostOpTimes, - vector<LegacyWCResponse>* legacyWCResponses ) { + return builder.obj(); +} - if ( hostOpTimes.empty() ) { - return Status::OK(); +// Adds a wOpTime and a wElectionId field to a set of gle options +static BSONObj buildGLECmdWithOpTime(const BSONObj& gleOptions, + const Timestamp& opTime, + const OID& electionId) { + BSONObjBuilder builder; + BSONObjIterator it(gleOptions); + + for (int i = 0; it.more(); ++i) { + BSONElement el = it.next(); + + // Make sure first element is getLastError : 1 + if (i == 0) { + StringData elName(el.fieldName()); + if (!elName.equalCaseInsensitive("getLastError")) { + builder.append("getLastError", 1); + } } - for ( HostOpTimeMap::const_iterator it = hostOpTimes.begin(); it != hostOpTimes.end(); - ++it ) { - - const ConnectionString& shardEndpoint = it->first; - const HostOpTime hot = it->second; - const Timestamp& opTime = hot.opTime; - const OID& electionId = hot.electionId; - - LOG( 3 ) << "enforcing write concern " << options << " on " << shardEndpoint.toString() - << " at opTime " << opTime.toStringPretty() << " with electionID " - << electionId; + builder.append(el); + } + builder.append("wOpTime", opTime); + builder.appendOID("wElectionId", const_cast<OID*>(&electionId)); + return builder.obj(); +} - BSONObj gleCmd = buildGLECmdWithOpTime( options, opTime, electionId ); +Status enforceLegacyWriteConcern(MultiCommandDispatch* dispatcher, + StringData dbName, + const BSONObj& options, + const HostOpTimeMap& hostOpTimes, + vector<LegacyWCResponse>* legacyWCResponses) { + if (hostOpTimes.empty()) { + return Status::OK(); + } - RawBSONSerializable gleCmdSerial( gleCmd ); - dispatcher->addCommand( shardEndpoint, dbName, gleCmdSerial ); - } + for (HostOpTimeMap::const_iterator it = hostOpTimes.begin(); it != hostOpTimes.end(); ++it) { + const ConnectionString& shardEndpoint = it->first; + const HostOpTime hot = it->second; + const Timestamp& opTime = hot.opTime; + const OID& electionId = hot.electionId; - dispatcher->sendAll(); + LOG(3) << "enforcing write concern " << options << " on " << shardEndpoint.toString() + << " at opTime " << opTime.toStringPretty() << " with electionID " << electionId; - vector<Status> failedStatuses; + BSONObj gleCmd = buildGLECmdWithOpTime(options, opTime, electionId); - while ( dispatcher->numPending() > 0 ) { + RawBSONSerializable gleCmdSerial(gleCmd); + dispatcher->addCommand(shardEndpoint, dbName, gleCmdSerial); + } - ConnectionString shardEndpoint; - RawBSONSerializable gleResponseSerial; + dispatcher->sendAll(); - Status dispatchStatus = dispatcher->recvAny( &shardEndpoint, &gleResponseSerial ); - if ( !dispatchStatus.isOK() ) { - // We need to get all responses before returning - failedStatuses.push_back( dispatchStatus ); - continue; - } + vector<Status> failedStatuses; - BSONObj gleResponse = stripNonWCInfo( gleResponseSerial.toBSON() ); + while (dispatcher->numPending() > 0) { + ConnectionString shardEndpoint; + RawBSONSerializable gleResponseSerial; - // Use the downconversion tools to determine if this GLE response is ok, a - // write concern error, or an unknown error we should immediately abort for. - GLEErrors errors; - Status extractStatus = extractGLEErrors( gleResponse, &errors ); - if ( !extractStatus.isOK() ) { - failedStatuses.push_back( extractStatus ); - continue; - } + Status dispatchStatus = dispatcher->recvAny(&shardEndpoint, &gleResponseSerial); + if (!dispatchStatus.isOK()) { + // We need to get all responses before returning + failedStatuses.push_back(dispatchStatus); + continue; + } - LegacyWCResponse wcResponse; - wcResponse.shardHost = shardEndpoint.toString(); - wcResponse.gleResponse = gleResponse; - if ( errors.wcError.get() ) { - wcResponse.errToReport = errors.wcError->getErrMessage(); - } + BSONObj gleResponse = stripNonWCInfo(gleResponseSerial.toBSON()); - legacyWCResponses->push_back( wcResponse ); + // Use the downconversion tools to determine if this GLE response is ok, a + // write concern error, or an unknown error we should immediately abort for. + GLEErrors errors; + Status extractStatus = extractGLEErrors(gleResponse, &errors); + if (!extractStatus.isOK()) { + failedStatuses.push_back(extractStatus); + continue; } - if ( failedStatuses.empty() ) { - return Status::OK(); + LegacyWCResponse wcResponse; + wcResponse.shardHost = shardEndpoint.toString(); + wcResponse.gleResponse = gleResponse; + if (errors.wcError.get()) { + wcResponse.errToReport = errors.wcError->getErrMessage(); } - StringBuilder builder; - builder << "could not enforce write concern"; + legacyWCResponses->push_back(wcResponse); + } - for ( vector<Status>::const_iterator it = failedStatuses.begin(); - it != failedStatuses.end(); ++it ) { - const Status& failedStatus = *it; - if ( it == failedStatuses.begin() ) { - builder << causedBy( failedStatus.toString() ); - } - else { - builder << ":: and ::" << failedStatus.toString(); - } - } + if (failedStatuses.empty()) { + return Status::OK(); + } + + StringBuilder builder; + builder << "could not enforce write concern"; - return Status( failedStatuses.size() == 1u ? failedStatuses.front().code() : - ErrorCodes::MultipleErrorsOccurred, - builder.str() ); + for (vector<Status>::const_iterator it = failedStatuses.begin(); it != failedStatuses.end(); + ++it) { + const Status& failedStatus = *it; + if (it == failedStatuses.begin()) { + builder << causedBy(failedStatus.toString()); + } else { + builder << ":: and ::" << failedStatus.toString(); + } } + + return Status(failedStatuses.size() == 1u ? failedStatuses.front().code() + : ErrorCodes::MultipleErrorsOccurred, + builder.str()); +} } diff --git a/src/mongo/s/write_ops/batch_downconvert.h b/src/mongo/s/write_ops/batch_downconvert.h index e3c6cc56f6a..659e59d3a9a 100644 --- a/src/mongo/s/write_ops/batch_downconvert.h +++ b/src/mongo/s/write_ops/batch_downconvert.h @@ -41,50 +41,48 @@ namespace mongo { - class MultiCommandDispatch; +class MultiCommandDispatch; - // Used for reporting legacy write concern responses - struct LegacyWCResponse { - std::string shardHost; - BSONObj gleResponse; - std::string errToReport; - }; +// Used for reporting legacy write concern responses +struct LegacyWCResponse { + std::string shardHost; + BSONObj gleResponse; + std::string errToReport; +}; - /** - * Uses GLE and the shard hosts and opTimes last written by write commands to enforce a - * write concern across the previously used shards. - * - * Returns OK with the LegacyWCResponses containing only write concern error information - * Returns !OK if there was an error getting a GLE response - */ - Status enforceLegacyWriteConcern( MultiCommandDispatch* dispatcher, - StringData dbName, - const BSONObj& options, - const HostOpTimeMap& hostOpTimes, - std::vector<LegacyWCResponse>* wcResponses ); - - // - // Below exposed for testing only - // - - // Helper that acts as an auto-ptr for write and wc errors - struct GLEErrors { - std::unique_ptr<WriteErrorDetail> writeError; - std::unique_ptr<WCErrorDetail> wcError; - }; +/** + * Uses GLE and the shard hosts and opTimes last written by write commands to enforce a + * write concern across the previously used shards. + * + * Returns OK with the LegacyWCResponses containing only write concern error information + * Returns !OK if there was an error getting a GLE response + */ +Status enforceLegacyWriteConcern(MultiCommandDispatch* dispatcher, + StringData dbName, + const BSONObj& options, + const HostOpTimeMap& hostOpTimes, + std::vector<LegacyWCResponse>* wcResponses); - /** - * Given a GLE response, extracts a write error and a write concern error for the previous - * operation. - * - * Returns !OK if the GLE itself failed in an unknown way. - */ - Status extractGLEErrors( const BSONObj& gleResponse, GLEErrors* errors ); +// +// Below exposed for testing only +// - /** - * Given a GLE response, strips out all non-write-concern related information - */ - BSONObj stripNonWCInfo( const BSONObj& gleResponse ); +// Helper that acts as an auto-ptr for write and wc errors +struct GLEErrors { + std::unique_ptr<WriteErrorDetail> writeError; + std::unique_ptr<WCErrorDetail> wcError; +}; +/** + * Given a GLE response, extracts a write error and a write concern error for the previous + * operation. + * + * Returns !OK if the GLE itself failed in an unknown way. + */ +Status extractGLEErrors(const BSONObj& gleResponse, GLEErrors* errors); +/** + * Given a GLE response, strips out all non-write-concern related information + */ +BSONObj stripNonWCInfo(const BSONObj& gleResponse); } diff --git a/src/mongo/s/write_ops/batch_downconvert_test.cpp b/src/mongo/s/write_ops/batch_downconvert_test.cpp index 8dd4b440a39..548bfb6f732 100644 --- a/src/mongo/s/write_ops/batch_downconvert_test.cpp +++ b/src/mongo/s/write_ops/batch_downconvert_test.cpp @@ -38,194 +38,179 @@ namespace { - using namespace mongo; - using std::vector; - using std::deque; - - // - // Tests for parsing GLE responses into write errors and write concern errors for write - // commands. These tests essentially document our expected 2.4 GLE behaviors. - // +using namespace mongo; +using std::vector; +using std::deque; + +// +// Tests for parsing GLE responses into write errors and write concern errors for write +// commands. These tests essentially document our expected 2.4 GLE behaviors. +// + +TEST(GLEParsing, Empty) { + const BSONObj gleResponse = fromjson("{ok: 1.0, err: null}"); + + GLEErrors errors; + ASSERT_OK(extractGLEErrors(gleResponse, &errors)); + ASSERT(!errors.writeError.get()); + ASSERT(!errors.wcError.get()); +} + +TEST(GLEParsing, WriteErr) { + const BSONObj gleResponse = fromjson("{ok: 1.0, err: 'message', code: 1000}"); + + GLEErrors errors; + ASSERT_OK(extractGLEErrors(gleResponse, &errors)); + ASSERT(errors.writeError.get()); + ASSERT_EQUALS(errors.writeError->getErrMessage(), "message"); + ASSERT_EQUALS(errors.writeError->getErrCode(), 1000); + ASSERT(!errors.wcError.get()); +} + +TEST(GLEParsing, JournalFail) { + const BSONObj gleResponse = fromjson("{ok: 1.0, err: null, jnote: 'message'}"); + + GLEErrors errors; + ASSERT_OK(extractGLEErrors(gleResponse, &errors)); + ASSERT(!errors.writeError.get()); + ASSERT(errors.wcError.get()); + ASSERT_EQUALS(errors.wcError->getErrMessage(), "message"); + ASSERT_EQUALS(errors.wcError->getErrCode(), ErrorCodes::WriteConcernFailed); +} + +TEST(GLEParsing, ReplErr) { + const BSONObj gleResponse = fromjson("{ok: 1.0, err: 'norepl', wnote: 'message'}"); + + GLEErrors errors; + ASSERT_OK(extractGLEErrors(gleResponse, &errors)); + ASSERT(!errors.writeError.get()); + ASSERT(errors.wcError.get()); + ASSERT_EQUALS(errors.wcError->getErrMessage(), "message"); + ASSERT_EQUALS(errors.wcError->getErrCode(), ErrorCodes::WriteConcernFailed); +} + +TEST(GLEParsing, ReplTimeoutErr) { + const BSONObj gleResponse = + fromjson("{ok: 1.0, err: 'timeout', errmsg: 'message', wtimeout: true}"); + + GLEErrors errors; + ASSERT_OK(extractGLEErrors(gleResponse, &errors)); + ASSERT(!errors.writeError.get()); + ASSERT(errors.wcError.get()); + ASSERT_EQUALS(errors.wcError->getErrMessage(), "message"); + ASSERT(errors.wcError->getErrInfo()["wtimeout"].trueValue()); + ASSERT_EQUALS(errors.wcError->getErrCode(), ErrorCodes::WriteConcernFailed); +} + +TEST(GLEParsing, GLEFail) { + const BSONObj gleResponse = fromjson("{ok: 0.0, err: null, errmsg: 'message', code: 1000}"); + + GLEErrors errors; + Status status = extractGLEErrors(gleResponse, &errors); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(status.reason(), "message"); + ASSERT_EQUALS(status.code(), 1000); +} + +TEST(GLEParsing, GLEFailNoCode) { + const BSONObj gleResponse = fromjson("{ok: 0.0, err: null, errmsg: 'message'}"); + + GLEErrors errors; + Status status = extractGLEErrors(gleResponse, &errors); + ASSERT_NOT_OK(status); + ASSERT_EQUALS(status.reason(), "message"); + ASSERT_EQUALS(status.code(), ErrorCodes::UnknownError); +} + +TEST(GLEParsing, NotMasterGLEFail) { + // Not master code in response + const BSONObj gleResponse = fromjson("{ok: 0.0, err: null, errmsg: 'message', code: 10990}"); - TEST(GLEParsing, Empty) { + GLEErrors errors; + ASSERT_OK(extractGLEErrors(gleResponse, &errors)); + ASSERT(!errors.writeError.get()); + ASSERT(errors.wcError.get()); + ASSERT_EQUALS(errors.wcError->getErrMessage(), "message"); + ASSERT_EQUALS(errors.wcError->getErrCode(), 10990); +} + +TEST(GLEParsing, WriteErrWithStats) { + const BSONObj gleResponse = fromjson("{ok: 1.0, n: 2, err: 'message', code: 1000}"); + + GLEErrors errors; + ASSERT_OK(extractGLEErrors(gleResponse, &errors)); + ASSERT(errors.writeError.get()); + ASSERT_EQUALS(errors.writeError->getErrMessage(), "message"); + ASSERT_EQUALS(errors.writeError->getErrCode(), 1000); + ASSERT(!errors.wcError.get()); +} + +TEST(GLEParsing, ReplTimeoutErrWithStats) { + const BSONObj gleResponse = fromjson( + "{ok: 1.0, err: 'timeout', errmsg: 'message', wtimeout: true," + " n: 1, upserted: 'abcde'}"); + + GLEErrors errors; + ASSERT_OK(extractGLEErrors(gleResponse, &errors)); + ASSERT(!errors.writeError.get()); + ASSERT(errors.wcError.get()); + ASSERT_EQUALS(errors.wcError->getErrMessage(), "message"); + ASSERT(errors.wcError->getErrInfo()["wtimeout"].trueValue()); + ASSERT_EQUALS(errors.wcError->getErrCode(), ErrorCodes::WriteConcernFailed); +} + +// +// Tests of processing and suppressing non-WC related fields from legacy GLE responses +// + +TEST(LegacyGLESuppress, Basic) { + const BSONObj gleResponse = fromjson("{ok: 1.0, err: null}"); + + BSONObj stripped = stripNonWCInfo(gleResponse); + ASSERT_EQUALS(stripped.nFields(), 2); // with err, ok : true + ASSERT(stripped["ok"].trueValue()); +} - const BSONObj gleResponse = fromjson( "{ok: 1.0, err: null}" ); - - GLEErrors errors; - ASSERT_OK( extractGLEErrors( gleResponse, &errors ) ); - ASSERT( !errors.writeError.get() ); - ASSERT( !errors.wcError.get() ); - } - - TEST(GLEParsing, WriteErr) { - - const BSONObj gleResponse = fromjson( "{ok: 1.0, err: 'message', code: 1000}" ); - - GLEErrors errors; - ASSERT_OK( extractGLEErrors( gleResponse, &errors ) ); - ASSERT( errors.writeError.get() ); - ASSERT_EQUALS( errors.writeError->getErrMessage(), "message" ); - ASSERT_EQUALS( errors.writeError->getErrCode(), 1000 ); - ASSERT( !errors.wcError.get() ); - } - - TEST(GLEParsing, JournalFail) { - - const BSONObj gleResponse = fromjson( "{ok: 1.0, err: null, jnote: 'message'}" ); - - GLEErrors errors; - ASSERT_OK( extractGLEErrors( gleResponse, &errors ) ); - ASSERT( !errors.writeError.get() ); - ASSERT( errors.wcError.get() ); - ASSERT_EQUALS( errors.wcError->getErrMessage(), "message" ); - ASSERT_EQUALS( errors.wcError->getErrCode(), ErrorCodes::WriteConcernFailed ); - } - - TEST(GLEParsing, ReplErr) { - - const BSONObj gleResponse = fromjson( "{ok: 1.0, err: 'norepl', wnote: 'message'}" ); - - GLEErrors errors; - ASSERT_OK( extractGLEErrors( gleResponse, &errors ) ); - ASSERT( !errors.writeError.get() ); - ASSERT( errors.wcError.get() ); - ASSERT_EQUALS( errors.wcError->getErrMessage(), "message" ); - ASSERT_EQUALS( errors.wcError->getErrCode(), ErrorCodes::WriteConcernFailed ); - } - - TEST(GLEParsing, ReplTimeoutErr) { - - const BSONObj gleResponse = - fromjson( "{ok: 1.0, err: 'timeout', errmsg: 'message', wtimeout: true}" ); - - GLEErrors errors; - ASSERT_OK( extractGLEErrors( gleResponse, &errors ) ); - ASSERT( !errors.writeError.get() ); - ASSERT( errors.wcError.get() ); - ASSERT_EQUALS( errors.wcError->getErrMessage(), "message" ); - ASSERT( errors.wcError->getErrInfo()["wtimeout"].trueValue() ); - ASSERT_EQUALS( errors.wcError->getErrCode(), ErrorCodes::WriteConcernFailed ); - } - - TEST(GLEParsing, GLEFail) { - - const BSONObj gleResponse = - fromjson( "{ok: 0.0, err: null, errmsg: 'message', code: 1000}" ); - - GLEErrors errors; - Status status = extractGLEErrors( gleResponse, &errors ); - ASSERT_NOT_OK( status ); - ASSERT_EQUALS( status.reason(), "message" ); - ASSERT_EQUALS( status.code(), 1000 ); - } - - TEST(GLEParsing, GLEFailNoCode) { +TEST(LegacyGLESuppress, BasicStats) { + const BSONObj gleResponse = fromjson( + "{ok: 0.0, err: 'message'," + " n: 1, nModified: 1, upserted: 'abc', updatedExisting: true}"); - const BSONObj gleResponse = fromjson( "{ok: 0.0, err: null, errmsg: 'message'}" ); - - GLEErrors errors; - Status status = extractGLEErrors( gleResponse, &errors ); - ASSERT_NOT_OK( status ); - ASSERT_EQUALS( status.reason(), "message" ); - ASSERT_EQUALS( status.code(), ErrorCodes::UnknownError ); - } - - TEST(GLEParsing, NotMasterGLEFail) { - - // Not master code in response - const BSONObj gleResponse = - fromjson( "{ok: 0.0, err: null, errmsg: 'message', code: 10990}" ); - - GLEErrors errors; - ASSERT_OK( extractGLEErrors( gleResponse, &errors ) ); - ASSERT( !errors.writeError.get() ); - ASSERT( errors.wcError.get() ); - ASSERT_EQUALS( errors.wcError->getErrMessage(), "message" ); - ASSERT_EQUALS( errors.wcError->getErrCode(), 10990 ); - } - - TEST(GLEParsing, WriteErrWithStats) { - const BSONObj gleResponse = fromjson( "{ok: 1.0, n: 2, err: 'message', code: 1000}" ); - - GLEErrors errors; - ASSERT_OK( extractGLEErrors( gleResponse, &errors ) ); - ASSERT( errors.writeError.get() ); - ASSERT_EQUALS( errors.writeError->getErrMessage(), "message" ); - ASSERT_EQUALS( errors.writeError->getErrCode(), 1000 ); - ASSERT( !errors.wcError.get() ); - } - - TEST(GLEParsing, ReplTimeoutErrWithStats) { - const BSONObj gleResponse = - fromjson( "{ok: 1.0, err: 'timeout', errmsg: 'message', wtimeout: true," - " n: 1, upserted: 'abcde'}" ); - - GLEErrors errors; - ASSERT_OK( extractGLEErrors( gleResponse, &errors ) ); - ASSERT( !errors.writeError.get() ); - ASSERT( errors.wcError.get() ); - ASSERT_EQUALS( errors.wcError->getErrMessage(), "message" ); - ASSERT( errors.wcError->getErrInfo()["wtimeout"].trueValue() ); - ASSERT_EQUALS( errors.wcError->getErrCode(), ErrorCodes::WriteConcernFailed ); - } - - // - // Tests of processing and suppressing non-WC related fields from legacy GLE responses - // - - TEST(LegacyGLESuppress, Basic) { - - const BSONObj gleResponse = fromjson( "{ok: 1.0, err: null}" ); - - BSONObj stripped = stripNonWCInfo( gleResponse ); - ASSERT_EQUALS( stripped.nFields(), 2 ); // with err, ok : true - ASSERT( stripped["ok"].trueValue() ); - } - - TEST(LegacyGLESuppress, BasicStats) { - - const BSONObj gleResponse = - fromjson( "{ok: 0.0, err: 'message'," - " n: 1, nModified: 1, upserted: 'abc', updatedExisting: true}" ); - - BSONObj stripped = stripNonWCInfo( gleResponse ); - ASSERT_EQUALS( stripped.nFields(), 1 ); - ASSERT( !stripped["ok"].trueValue() ); - } - - TEST(LegacyGLESuppress, ReplError) { - - const BSONObj gleResponse = - fromjson( "{ok: 0.0, err: 'norepl', n: 1, wcField: true}" ); - - BSONObj stripped = stripNonWCInfo( gleResponse ); - ASSERT_EQUALS( stripped.nFields(), 3 ); - ASSERT( !stripped["ok"].trueValue() ); - ASSERT_EQUALS( stripped["err"].str(), "norepl" ); - ASSERT( stripped["wcField"].trueValue() ); - } - - TEST(LegacyGLESuppress, StripCode) { - - const BSONObj gleResponse = - fromjson( "{ok: 1.0, err: 'message', code: 12345}" ); - - BSONObj stripped = stripNonWCInfo( gleResponse ); - ASSERT_EQUALS( stripped.nFields(), 2 ); // with err, ok : true - ASSERT( stripped["ok"].trueValue() ); - } - - TEST(LegacyGLESuppress, TimeoutDupError24) { - - const BSONObj gleResponse = - BSON( "ok" << 0.0 << "err" << "message" << "code" << 12345 - << "err" << "timeout" << "code" << 56789 << "wtimeout" << true ); - - BSONObj stripped = stripNonWCInfo( gleResponse ); - ASSERT_EQUALS( stripped.nFields(), 4 ); - ASSERT( !stripped["ok"].trueValue() ); - ASSERT_EQUALS( stripped["err"].str(), "timeout" ); - ASSERT_EQUALS( stripped["code"].numberInt(), 56789 ); - ASSERT( stripped["wtimeout"].trueValue() ); - } + BSONObj stripped = stripNonWCInfo(gleResponse); + ASSERT_EQUALS(stripped.nFields(), 1); + ASSERT(!stripped["ok"].trueValue()); +} + +TEST(LegacyGLESuppress, ReplError) { + const BSONObj gleResponse = fromjson("{ok: 0.0, err: 'norepl', n: 1, wcField: true}"); + + BSONObj stripped = stripNonWCInfo(gleResponse); + ASSERT_EQUALS(stripped.nFields(), 3); + ASSERT(!stripped["ok"].trueValue()); + ASSERT_EQUALS(stripped["err"].str(), "norepl"); + ASSERT(stripped["wcField"].trueValue()); +} + +TEST(LegacyGLESuppress, StripCode) { + const BSONObj gleResponse = fromjson("{ok: 1.0, err: 'message', code: 12345}"); + + BSONObj stripped = stripNonWCInfo(gleResponse); + ASSERT_EQUALS(stripped.nFields(), 2); // with err, ok : true + ASSERT(stripped["ok"].trueValue()); +} + +TEST(LegacyGLESuppress, TimeoutDupError24) { + const BSONObj gleResponse = BSON("ok" << 0.0 << "err" + << "message" + << "code" << 12345 << "err" + << "timeout" + << "code" << 56789 << "wtimeout" << true); + + BSONObj stripped = stripNonWCInfo(gleResponse); + ASSERT_EQUALS(stripped.nFields(), 4); + ASSERT(!stripped["ok"].trueValue()); + ASSERT_EQUALS(stripped["err"].str(), "timeout"); + ASSERT_EQUALS(stripped["code"].numberInt(), 56789); + ASSERT(stripped["wtimeout"].trueValue()); +} } diff --git a/src/mongo/s/write_ops/batch_upconvert.cpp b/src/mongo/s/write_ops/batch_upconvert.cpp index ee02c17744a..6c7fcacd361 100644 --- a/src/mongo/s/write_ops/batch_upconvert.cpp +++ b/src/mongo/s/write_ops/batch_upconvert.cpp @@ -43,201 +43,184 @@ namespace mongo { - using mongoutils::str::stream; - using std::string; - using std::unique_ptr; - using std::vector; - - void msgToBatchRequests( const Message& msg, vector<BatchedCommandRequest*>* requests ) { - - int opType = msg.operation(); - - unique_ptr<BatchedCommandRequest> request; - if ( opType == dbInsert ) { - msgToBatchInserts( msg, requests ); - } - else if ( opType == dbUpdate ) { - requests->push_back( msgToBatchUpdate( msg ) ); - } - else { - dassert( opType == dbDelete ); - requests->push_back( msgToBatchDelete( msg ) ); - } +using mongoutils::str::stream; +using std::string; +using std::unique_ptr; +using std::vector; + +void msgToBatchRequests(const Message& msg, vector<BatchedCommandRequest*>* requests) { + int opType = msg.operation(); + + unique_ptr<BatchedCommandRequest> request; + if (opType == dbInsert) { + msgToBatchInserts(msg, requests); + } else if (opType == dbUpdate) { + requests->push_back(msgToBatchUpdate(msg)); + } else { + dassert(opType == dbDelete); + requests->push_back(msgToBatchDelete(msg)); } +} - void msgToBatchInserts( const Message& insertMsg, - vector<BatchedCommandRequest*>* insertRequests ) { - - // Parsing DbMessage throws - DbMessage dbMsg( insertMsg ); - NamespaceString nss( dbMsg.getns() ); - - // Continue-on-error == unordered - bool coe = dbMsg.reservedField() & Reserved_InsertOption_ContinueOnError; - bool ordered = !coe; - - while ( insertRequests->empty() || dbMsg.moreJSObjs() ) { - - // Collect docs for next batch, but don't exceed maximum size - int totalInsertSize = 0; - vector<BSONObj> docs; - do { - const char* prevObjMark = dbMsg.markGet(); - BSONObj nextObj = dbMsg.nextJsObj(); - if ( totalInsertSize + nextObj.objsize() <= BSONObjMaxUserSize ) { - docs.push_back( nextObj ); - totalInsertSize += docs.back().objsize(); - } - else { - // Size limit exceeded, rollback to previous insert position - dbMsg.markReset( prevObjMark ); - break; - } +void msgToBatchInserts(const Message& insertMsg, vector<BatchedCommandRequest*>* insertRequests) { + // Parsing DbMessage throws + DbMessage dbMsg(insertMsg); + NamespaceString nss(dbMsg.getns()); + + // Continue-on-error == unordered + bool coe = dbMsg.reservedField() & Reserved_InsertOption_ContinueOnError; + bool ordered = !coe; + + while (insertRequests->empty() || dbMsg.moreJSObjs()) { + // Collect docs for next batch, but don't exceed maximum size + int totalInsertSize = 0; + vector<BSONObj> docs; + do { + const char* prevObjMark = dbMsg.markGet(); + BSONObj nextObj = dbMsg.nextJsObj(); + if (totalInsertSize + nextObj.objsize() <= BSONObjMaxUserSize) { + docs.push_back(nextObj); + totalInsertSize += docs.back().objsize(); + } else { + // Size limit exceeded, rollback to previous insert position + dbMsg.markReset(prevObjMark); + break; } - while ( docs.size() < BatchedCommandRequest::kMaxWriteBatchSize - && dbMsg.moreJSObjs() ); + } while (docs.size() < BatchedCommandRequest::kMaxWriteBatchSize && dbMsg.moreJSObjs()); - dassert( !docs.empty() ); - - // No exceptions from here on - BatchedCommandRequest* request = - new BatchedCommandRequest( BatchedCommandRequest::BatchType_Insert ); - request->setNSS( nss ); - for ( vector<BSONObj>::const_iterator it = docs.begin(); it != docs.end(); ++it ) { - request->getInsertRequest()->addToDocuments( *it ); - } - request->setOrdered( ordered ); - request->setWriteConcern( WriteConcernOptions::Acknowledged ); - - insertRequests->push_back( request ); - } - } - - BatchedCommandRequest* msgToBatchUpdate( const Message& updateMsg ) { - - // Parsing DbMessage throws - DbMessage dbMsg( updateMsg ); - NamespaceString nss( dbMsg.getns() ); - int flags = dbMsg.pullInt(); - bool upsert = flags & UpdateOption_Upsert; - bool multi = flags & UpdateOption_Multi; - const BSONObj query = dbMsg.nextJsObj(); - const BSONObj updateExpr = dbMsg.nextJsObj(); + dassert(!docs.empty()); // No exceptions from here on - BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; - updateDoc->setQuery( query ); - updateDoc->setUpdateExpr( updateExpr ); - updateDoc->setUpsert( upsert ); - updateDoc->setMulti( multi ); - BatchedCommandRequest* request = - new BatchedCommandRequest( BatchedCommandRequest::BatchType_Update ); - request->setNSS( nss ); - request->getUpdateRequest()->addToUpdates( updateDoc ); - request->setWriteConcern( WriteConcernOptions::Acknowledged ); + new BatchedCommandRequest(BatchedCommandRequest::BatchType_Insert); + request->setNSS(nss); + for (vector<BSONObj>::const_iterator it = docs.begin(); it != docs.end(); ++it) { + request->getInsertRequest()->addToDocuments(*it); + } + request->setOrdered(ordered); + request->setWriteConcern(WriteConcernOptions::Acknowledged); - return request; + insertRequests->push_back(request); } +} - BatchedCommandRequest* msgToBatchDelete( const Message& deleteMsg ) { - - // Parsing DbMessage throws - DbMessage dbMsg( deleteMsg ); - NamespaceString nss( dbMsg.getns() ); - int flags = dbMsg.pullInt(); - const BSONObj query = dbMsg.nextJsObj(); - int limit = ( flags & RemoveOption_JustOne ) ? 1 : 0; +BatchedCommandRequest* msgToBatchUpdate(const Message& updateMsg) { + // Parsing DbMessage throws + DbMessage dbMsg(updateMsg); + NamespaceString nss(dbMsg.getns()); + int flags = dbMsg.pullInt(); + bool upsert = flags & UpdateOption_Upsert; + bool multi = flags & UpdateOption_Multi; + const BSONObj query = dbMsg.nextJsObj(); + const BSONObj updateExpr = dbMsg.nextJsObj(); + + // No exceptions from here on + BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; + updateDoc->setQuery(query); + updateDoc->setUpdateExpr(updateExpr); + updateDoc->setUpsert(upsert); + updateDoc->setMulti(multi); + + BatchedCommandRequest* request = + new BatchedCommandRequest(BatchedCommandRequest::BatchType_Update); + request->setNSS(nss); + request->getUpdateRequest()->addToUpdates(updateDoc); + request->setWriteConcern(WriteConcernOptions::Acknowledged); + + return request; +} - // No exceptions from here on - BatchedDeleteDocument* deleteDoc = new BatchedDeleteDocument; - deleteDoc->setLimit( limit ); - deleteDoc->setQuery( query ); +BatchedCommandRequest* msgToBatchDelete(const Message& deleteMsg) { + // Parsing DbMessage throws + DbMessage dbMsg(deleteMsg); + NamespaceString nss(dbMsg.getns()); + int flags = dbMsg.pullInt(); + const BSONObj query = dbMsg.nextJsObj(); + int limit = (flags & RemoveOption_JustOne) ? 1 : 0; + + // No exceptions from here on + BatchedDeleteDocument* deleteDoc = new BatchedDeleteDocument; + deleteDoc->setLimit(limit); + deleteDoc->setQuery(query); + + BatchedCommandRequest* request = + new BatchedCommandRequest(BatchedCommandRequest::BatchType_Delete); + request->setNSS(nss); + request->getDeleteRequest()->addToDeletes(deleteDoc); + request->setWriteConcern(WriteConcernOptions::Acknowledged); + + return request; +} - BatchedCommandRequest* request = - new BatchedCommandRequest( BatchedCommandRequest::BatchType_Delete ); - request->setNSS( nss ); - request->getDeleteRequest()->addToDeletes( deleteDoc ); - request->setWriteConcern( WriteConcernOptions::Acknowledged ); +void buildErrorFromResponse(const BatchedCommandResponse& response, WriteErrorDetail* error) { + error->setErrCode(response.getErrCode()); + error->setErrMessage(response.getErrMessage()); +} - return request; +bool batchErrorToLastError(const BatchedCommandRequest& request, + const BatchedCommandResponse& response, + LastError* error) { + unique_ptr<WriteErrorDetail> commandError; + WriteErrorDetail* lastBatchError = NULL; + + if (!response.getOk()) { + // Command-level error, all writes failed + + commandError.reset(new WriteErrorDetail); + buildErrorFromResponse(response, commandError.get()); + lastBatchError = commandError.get(); + } else if (response.isErrDetailsSet()) { + // The last error in the batch is always reported - this matches expected COE + // semantics for insert batches. For updates and deletes, error is only reported + // if the error was on the last item. + + const bool lastOpErrored = response.getErrDetails().back()->getIndex() == + static_cast<int>(request.sizeWriteOps() - 1); + if (request.getBatchType() == BatchedCommandRequest::BatchType_Insert || lastOpErrored) { + lastBatchError = response.getErrDetails().back(); + } + } else { + // We don't care about write concern errors, these happen in legacy mode in GLE. } - void buildErrorFromResponse( const BatchedCommandResponse& response, WriteErrorDetail* error ) { - error->setErrCode( response.getErrCode() ); - error->setErrMessage( response.getErrMessage() ); + // Record an error if one exists + if (lastBatchError) { + string errMsg = lastBatchError->getErrMessage(); + error->setLastError(lastBatchError->getErrCode(), + errMsg.empty() ? "see code for details" : errMsg.c_str()); + return true; } - bool batchErrorToLastError( const BatchedCommandRequest& request, - const BatchedCommandResponse& response, - LastError* error ) { - - unique_ptr<WriteErrorDetail> commandError; - WriteErrorDetail* lastBatchError = NULL; - - if ( !response.getOk() ) { - // Command-level error, all writes failed - - commandError.reset( new WriteErrorDetail ); - buildErrorFromResponse( response, commandError.get() ); - lastBatchError = commandError.get(); - } - else if ( response.isErrDetailsSet() ) { - // The last error in the batch is always reported - this matches expected COE - // semantics for insert batches. For updates and deletes, error is only reported - // if the error was on the last item. - - const bool lastOpErrored = response.getErrDetails().back()->getIndex() == - static_cast<int>(request.sizeWriteOps() - 1); - if ( request.getBatchType() == BatchedCommandRequest::BatchType_Insert || - lastOpErrored ) { - lastBatchError = response.getErrDetails().back(); + // Record write stats otherwise + // NOTE: For multi-write batches, our semantics change a little because we don't have + // un-aggregated "n" stats. + if (request.getBatchType() == BatchedCommandRequest::BatchType_Update) { + BSONObj upsertedId; + if (response.isUpsertDetailsSet()) { + // Only report the very last item's upserted id if applicable + if (response.getUpsertDetails().back()->getIndex() + 1 == + static_cast<int>(request.sizeWriteOps())) { + upsertedId = response.getUpsertDetails().back()->getUpsertedID(); } } - else { - // We don't care about write concern errors, these happen in legacy mode in GLE. - } - // Record an error if one exists - if ( lastBatchError ) { - string errMsg = lastBatchError->getErrMessage(); - error->setLastError(lastBatchError->getErrCode(), - errMsg.empty() ? "see code for details" : errMsg.c_str()); - return true; - } - - // Record write stats otherwise - // NOTE: For multi-write batches, our semantics change a little because we don't have - // un-aggregated "n" stats. - if ( request.getBatchType() == BatchedCommandRequest::BatchType_Update ) { - - BSONObj upsertedId; - if( response.isUpsertDetailsSet() ) { - // Only report the very last item's upserted id if applicable - if ( response.getUpsertDetails().back()->getIndex() + 1 - == static_cast<int>( request.sizeWriteOps() ) ) { - upsertedId = response.getUpsertDetails().back()->getUpsertedID(); - } - } + int numUpserted = 0; + if (response.isUpsertDetailsSet()) + numUpserted = response.sizeUpsertDetails(); - int numUpserted = 0; - if ( response.isUpsertDetailsSet() ) - numUpserted = response.sizeUpsertDetails(); + int numMatched = response.getN() - numUpserted; + dassert(numMatched >= 0); - int numMatched = response.getN() - numUpserted; - dassert( numMatched >= 0 ); + // Wrap upserted id in "upserted" field + BSONObj leUpsertedId; + if (!upsertedId.isEmpty()) + leUpsertedId = upsertedId.firstElement().wrap(kUpsertedFieldName); - // Wrap upserted id in "upserted" field - BSONObj leUpsertedId; - if ( !upsertedId.isEmpty() ) - leUpsertedId = upsertedId.firstElement().wrap( kUpsertedFieldName ); - - error->recordUpdate( numMatched > 0, response.getN(), leUpsertedId ); - } - else if ( request.getBatchType() == BatchedCommandRequest::BatchType_Delete ) { - error->recordDelete( response.getN() ); - } - - return false; + error->recordUpdate(numMatched > 0, response.getN(), leUpsertedId); + } else if (request.getBatchType() == BatchedCommandRequest::BatchType_Delete) { + error->recordDelete(response.getN()); } + + return false; +} } diff --git a/src/mongo/s/write_ops/batch_upconvert.h b/src/mongo/s/write_ops/batch_upconvert.h index 46b6d4552b6..a0b4712cf96 100644 --- a/src/mongo/s/write_ops/batch_upconvert.h +++ b/src/mongo/s/write_ops/batch_upconvert.h @@ -37,29 +37,28 @@ namespace mongo { - // - // Utility functions for up-converting incoming write messages into batch write requests. - // NOTE: These functions throw on invalid message format. - // +// +// Utility functions for up-converting incoming write messages into batch write requests. +// NOTE: These functions throw on invalid message format. +// - void msgToBatchRequests( const Message& msg, std::vector<BatchedCommandRequest*>* requests ); +void msgToBatchRequests(const Message& msg, std::vector<BatchedCommandRequest*>* requests); - // Batch inserts may get mapped to multiple batch requests, to avoid spilling MaxBSONObjSize - void msgToBatchInserts( const Message& insertMsg, - std::vector<BatchedCommandRequest*>* insertRequests ); +// Batch inserts may get mapped to multiple batch requests, to avoid spilling MaxBSONObjSize +void msgToBatchInserts(const Message& insertMsg, + std::vector<BatchedCommandRequest*>* insertRequests); - BatchedCommandRequest* msgToBatchUpdate( const Message& updateMsg ); +BatchedCommandRequest* msgToBatchUpdate(const Message& updateMsg); - BatchedCommandRequest* msgToBatchDelete( const Message& deleteMsg ); - - /** - * Utility function for recording completed batch writes into the LastError object. - * (Interpreting the response requires the request object as well.) - * - * Returns true if an error occurred in the batch. - */ - bool batchErrorToLastError( const BatchedCommandRequest& request, - const BatchedCommandResponse& response, - LastError* error ); +BatchedCommandRequest* msgToBatchDelete(const Message& deleteMsg); +/** + * Utility function for recording completed batch writes into the LastError object. + * (Interpreting the response requires the request object as well.) + * + * Returns true if an error occurred in the batch. + */ +bool batchErrorToLastError(const BatchedCommandRequest& request, + const BatchedCommandResponse& response, + LastError* error); } diff --git a/src/mongo/s/write_ops/batch_upconvert_test.cpp b/src/mongo/s/write_ops/batch_upconvert_test.cpp index c802411c9e5..00d485818fe 100644 --- a/src/mongo/s/write_ops/batch_upconvert_test.cpp +++ b/src/mongo/s/write_ops/batch_upconvert_test.cpp @@ -33,7 +33,7 @@ #include "mongo/bson/bsonobj.h" #include "mongo/bson/bsonobjbuilder.h" #include "mongo/bson/util/builder.h" -#include "mongo/client/dbclientinterface.h" // for write constants +#include "mongo/client/dbclientinterface.h" // for write constants #include "mongo/db/write_concern_options.h" #include "mongo/s/write_ops/batched_command_request.h" #include "mongo/unittest/unittest.h" @@ -41,112 +41,109 @@ namespace { - using std::string; - using std::vector; - - using namespace mongo; - - TEST(WriteBatchUpconvert, BasicInsert) { - - // Tests that an insert message is correctly upconverted to a batch insert - - const string ns = "foo.bar"; - const BSONObj doc = BSON( "hello" << "world" ); - - Message insertMsg; - BufBuilder insertMsgB; - - int reservedFlags = InsertOption_ContinueOnError; - insertMsgB.appendNum( reservedFlags ); - insertMsgB.appendStr( ns ); - doc.appendSelfToBufBuilder( insertMsgB ); - insertMsg.setData( dbInsert, insertMsgB.buf(), insertMsgB.len() ); - - OwnedPointerVector<BatchedCommandRequest> requestsOwned; - vector<BatchedCommandRequest*>& requests = requestsOwned.mutableVector(); - msgToBatchRequests( insertMsg, &requests ); - - BatchedCommandRequest* request = requests.back(); - ASSERT_EQUALS( request->getBatchType(), BatchedCommandRequest::BatchType_Insert ); - string errMsg; - ASSERT( request->isValid( &errMsg ) ); - ASSERT_EQUALS( request->getNS(), ns ); - ASSERT( !request->getOrdered() ); - ASSERT_EQUALS( request->sizeWriteOps(), 1u ); - bool isSameDoc = doc.woCompare( request->getInsertRequest()->getDocumentsAt( 0 ) ) == 0; - ASSERT( isSameDoc ); - ASSERT( request->getWriteConcern().woCompare( WriteConcernOptions::Acknowledged ) == 0 ); - } - - TEST(WriteBatchUpconvert, BasicUpdate) { - - // Tests that an update message is correctly upconverted to a batch update - - const string ns = "foo.bar"; - const BSONObj query = BSON( "hello" << "world" ); - const BSONObj update = BSON( "$set" << BSON( "hello" << "world" ) ); - - Message updateMsg; - BufBuilder updateMsgB; - - int reservedFlags = 0; - updateMsgB.appendNum( reservedFlags ); - updateMsgB.appendStr( ns ); - updateMsgB.appendNum( UpdateOption_Upsert ); - query.appendSelfToBufBuilder( updateMsgB ); - update.appendSelfToBufBuilder( updateMsgB ); - updateMsg.setData( dbUpdate, updateMsgB.buf(), updateMsgB.len() ); - - OwnedPointerVector<BatchedCommandRequest> requestsOwned; - vector<BatchedCommandRequest*>& requests = requestsOwned.mutableVector(); - msgToBatchRequests( updateMsg, &requests ); - - BatchedCommandRequest* request = requests.back(); - ASSERT_EQUALS( request->getBatchType(), BatchedCommandRequest::BatchType_Update ); - string errMsg; - ASSERT( request->isValid( &errMsg ) ); - ASSERT_EQUALS( request->getNS(), ns ); - ASSERT_EQUALS( request->sizeWriteOps(), 1u ); - ASSERT( query.woCompare( - request->getUpdateRequest()->getUpdatesAt( 0 )->getQuery() ) == 0 ); - ASSERT( update.woCompare( - request->getUpdateRequest()->getUpdatesAt( 0 )->getUpdateExpr() ) == 0 ); - ASSERT( request->getUpdateRequest()->getUpdatesAt( 0 )->getUpsert() ); - ASSERT( !request->getUpdateRequest()->getUpdatesAt( 0 )->getMulti() ); - ASSERT( request->getWriteConcern().woCompare( WriteConcernOptions::Acknowledged ) == 0 ); - } - - TEST(WriteBatchUpconvert, BasicDelete) { - - // Tests that an remove message is correctly upconverted to a batch delete - - const string ns = "foo.bar"; - const BSONObj query = BSON( "hello" << "world" ); - - Message deleteMsg; - BufBuilder deleteMsgB; - - int reservedFlags = 0; - deleteMsgB.appendNum( reservedFlags ); - deleteMsgB.appendStr( ns ); - deleteMsgB.appendNum( RemoveOption_JustOne ); - query.appendSelfToBufBuilder( deleteMsgB ); - deleteMsg.setData( dbDelete, deleteMsgB.buf(), deleteMsgB.len() ); - - OwnedPointerVector<BatchedCommandRequest> requestsOwned; - vector<BatchedCommandRequest*>& requests = requestsOwned.mutableVector(); - msgToBatchRequests( deleteMsg, &requests ); - - BatchedCommandRequest* request = requests.back(); - ASSERT_EQUALS( request->getBatchType(), BatchedCommandRequest::BatchType_Delete ); - string errMsg; - ASSERT( request->isValid( &errMsg ) ); - ASSERT_EQUALS( request->getNS(), ns ); - ASSERT_EQUALS( request->sizeWriteOps(), 1u ); - ASSERT( query.woCompare( - request->getDeleteRequest()->getDeletesAt( 0 )->getQuery() ) == 0 ); - ASSERT( request->getDeleteRequest()->getDeletesAt( 0 )->getLimit() == 1 ); - ASSERT( request->getWriteConcern().woCompare( WriteConcernOptions::Acknowledged ) == 0 ); - } +using std::string; +using std::vector; + +using namespace mongo; + +TEST(WriteBatchUpconvert, BasicInsert) { + // Tests that an insert message is correctly upconverted to a batch insert + + const string ns = "foo.bar"; + const BSONObj doc = BSON("hello" + << "world"); + + Message insertMsg; + BufBuilder insertMsgB; + + int reservedFlags = InsertOption_ContinueOnError; + insertMsgB.appendNum(reservedFlags); + insertMsgB.appendStr(ns); + doc.appendSelfToBufBuilder(insertMsgB); + insertMsg.setData(dbInsert, insertMsgB.buf(), insertMsgB.len()); + + OwnedPointerVector<BatchedCommandRequest> requestsOwned; + vector<BatchedCommandRequest*>& requests = requestsOwned.mutableVector(); + msgToBatchRequests(insertMsg, &requests); + + BatchedCommandRequest* request = requests.back(); + ASSERT_EQUALS(request->getBatchType(), BatchedCommandRequest::BatchType_Insert); + string errMsg; + ASSERT(request->isValid(&errMsg)); + ASSERT_EQUALS(request->getNS(), ns); + ASSERT(!request->getOrdered()); + ASSERT_EQUALS(request->sizeWriteOps(), 1u); + bool isSameDoc = doc.woCompare(request->getInsertRequest()->getDocumentsAt(0)) == 0; + ASSERT(isSameDoc); + ASSERT(request->getWriteConcern().woCompare(WriteConcernOptions::Acknowledged) == 0); +} + +TEST(WriteBatchUpconvert, BasicUpdate) { + // Tests that an update message is correctly upconverted to a batch update + + const string ns = "foo.bar"; + const BSONObj query = BSON("hello" + << "world"); + const BSONObj update = BSON("$set" << BSON("hello" + << "world")); + + Message updateMsg; + BufBuilder updateMsgB; + + int reservedFlags = 0; + updateMsgB.appendNum(reservedFlags); + updateMsgB.appendStr(ns); + updateMsgB.appendNum(UpdateOption_Upsert); + query.appendSelfToBufBuilder(updateMsgB); + update.appendSelfToBufBuilder(updateMsgB); + updateMsg.setData(dbUpdate, updateMsgB.buf(), updateMsgB.len()); + + OwnedPointerVector<BatchedCommandRequest> requestsOwned; + vector<BatchedCommandRequest*>& requests = requestsOwned.mutableVector(); + msgToBatchRequests(updateMsg, &requests); + + BatchedCommandRequest* request = requests.back(); + ASSERT_EQUALS(request->getBatchType(), BatchedCommandRequest::BatchType_Update); + string errMsg; + ASSERT(request->isValid(&errMsg)); + ASSERT_EQUALS(request->getNS(), ns); + ASSERT_EQUALS(request->sizeWriteOps(), 1u); + ASSERT(query.woCompare(request->getUpdateRequest()->getUpdatesAt(0)->getQuery()) == 0); + ASSERT(update.woCompare(request->getUpdateRequest()->getUpdatesAt(0)->getUpdateExpr()) == 0); + ASSERT(request->getUpdateRequest()->getUpdatesAt(0)->getUpsert()); + ASSERT(!request->getUpdateRequest()->getUpdatesAt(0)->getMulti()); + ASSERT(request->getWriteConcern().woCompare(WriteConcernOptions::Acknowledged) == 0); +} +TEST(WriteBatchUpconvert, BasicDelete) { + // Tests that an remove message is correctly upconverted to a batch delete + + const string ns = "foo.bar"; + const BSONObj query = BSON("hello" + << "world"); + + Message deleteMsg; + BufBuilder deleteMsgB; + + int reservedFlags = 0; + deleteMsgB.appendNum(reservedFlags); + deleteMsgB.appendStr(ns); + deleteMsgB.appendNum(RemoveOption_JustOne); + query.appendSelfToBufBuilder(deleteMsgB); + deleteMsg.setData(dbDelete, deleteMsgB.buf(), deleteMsgB.len()); + + OwnedPointerVector<BatchedCommandRequest> requestsOwned; + vector<BatchedCommandRequest*>& requests = requestsOwned.mutableVector(); + msgToBatchRequests(deleteMsg, &requests); + + BatchedCommandRequest* request = requests.back(); + ASSERT_EQUALS(request->getBatchType(), BatchedCommandRequest::BatchType_Delete); + string errMsg; + ASSERT(request->isValid(&errMsg)); + ASSERT_EQUALS(request->getNS(), ns); + ASSERT_EQUALS(request->sizeWriteOps(), 1u); + ASSERT(query.woCompare(request->getDeleteRequest()->getDeletesAt(0)->getQuery()) == 0); + ASSERT(request->getDeleteRequest()->getDeletesAt(0)->getLimit() == 1); + ASSERT(request->getWriteConcern().woCompare(WriteConcernOptions::Acknowledged) == 0); +} } diff --git a/src/mongo/s/write_ops/batch_write_exec.cpp b/src/mongo/s/write_ops/batch_write_exec.cpp index 307c39321ee..1fa89c034bc 100644 --- a/src/mongo/s/write_ops/batch_write_exec.cpp +++ b/src/mongo/s/write_ops/batch_write_exec.cpp @@ -36,7 +36,7 @@ #include "mongo/base/owned_pointer_map.h" #include "mongo/base/status.h" #include "mongo/bson/util/builder.h" -#include "mongo/client/dbclientinterface.h" // ConnectionString (header-only) +#include "mongo/client/dbclientinterface.h" // ConnectionString (header-only) #include "mongo/s/client/multi_command_dispatch.h" #include "mongo/s/write_ops/batch_write_op.h" #include "mongo/s/write_ops/write_error_detail.h" @@ -44,354 +44,337 @@ namespace mongo { - using std::endl; - using std::make_pair; - using std::stringstream; - using std::vector; - - BatchWriteExec::BatchWriteExec( NSTargeter* targeter, - ShardResolver* resolver, - MultiCommandDispatch* dispatcher ) : - _targeter( targeter ), - _resolver( resolver ), - _dispatcher( dispatcher ), - _stats( new BatchWriteExecStats ) { - } +using std::endl; +using std::make_pair; +using std::stringstream; +using std::vector; - namespace { +BatchWriteExec::BatchWriteExec(NSTargeter* targeter, + ShardResolver* resolver, + MultiCommandDispatch* dispatcher) + : _targeter(targeter), + _resolver(resolver), + _dispatcher(dispatcher), + _stats(new BatchWriteExecStats) {} - // - // Map which allows associating ConnectionString hosts with TargetedWriteBatches - // This is needed since the dispatcher only returns hosts with responses. - // +namespace { - // TODO: Unordered map? - typedef OwnedPointerMap<ConnectionString, TargetedWriteBatch> OwnedHostBatchMap; - } +// +// Map which allows associating ConnectionString hosts with TargetedWriteBatches +// This is needed since the dispatcher only returns hosts with responses. +// - static void buildErrorFrom( const Status& status, WriteErrorDetail* error ) { - error->setErrCode( status.code() ); - error->setErrMessage( status.reason() ); - } +// TODO: Unordered map? +typedef OwnedPointerMap<ConnectionString, TargetedWriteBatch> OwnedHostBatchMap; +} - // Helper to note several stale errors from a response - static void noteStaleResponses( const vector<ShardError*>& staleErrors, NSTargeter* targeter ) { - for ( vector<ShardError*>::const_iterator it = staleErrors.begin(); it != staleErrors.end(); - ++it ) { - const ShardError* error = *it; - targeter->noteStaleResponse( error->endpoint, - error->error.isErrInfoSet() ? error->error.getErrInfo() : - BSONObj() ); - } - } +static void buildErrorFrom(const Status& status, WriteErrorDetail* error) { + error->setErrCode(status.code()); + error->setErrMessage(status.reason()); +} - static bool isShardMetadataChanging( const vector<ShardError*>& staleErrors ) { - if ( !staleErrors.empty() && staleErrors.back()->error.isErrInfoSet() ) - return staleErrors.back()->error.getErrInfo()["inCriticalSection"].trueValue(); - return false; +// Helper to note several stale errors from a response +static void noteStaleResponses(const vector<ShardError*>& staleErrors, NSTargeter* targeter) { + for (vector<ShardError*>::const_iterator it = staleErrors.begin(); it != staleErrors.end(); + ++it) { + const ShardError* error = *it; + targeter->noteStaleResponse( + error->endpoint, error->error.isErrInfoSet() ? error->error.getErrInfo() : BSONObj()); } +} - // The number of times we'll try to continue a batch op if no progress is being made - // This only applies when no writes are occurring and metadata is not changing on reload - static const int kMaxRoundsWithoutProgress( 5 ); +static bool isShardMetadataChanging(const vector<ShardError*>& staleErrors) { + if (!staleErrors.empty() && staleErrors.back()->error.isErrInfoSet()) + return staleErrors.back()->error.getErrInfo()["inCriticalSection"].trueValue(); + return false; +} - void BatchWriteExec::executeBatch( const BatchedCommandRequest& clientRequest, - BatchedCommandResponse* clientResponse ) { +// The number of times we'll try to continue a batch op if no progress is being made +// This only applies when no writes are occurring and metadata is not changing on reload +static const int kMaxRoundsWithoutProgress(5); - LOG( 4 ) << "starting execution of write batch of size " - << static_cast<int>( clientRequest.sizeWriteOps() ) - << " for " << clientRequest.getNS() << endl; +void BatchWriteExec::executeBatch(const BatchedCommandRequest& clientRequest, + BatchedCommandResponse* clientResponse) { + LOG(4) << "starting execution of write batch of size " + << static_cast<int>(clientRequest.sizeWriteOps()) << " for " << clientRequest.getNS() + << endl; - BatchWriteOp batchOp; - batchOp.initClientRequest( &clientRequest ); + BatchWriteOp batchOp; + batchOp.initClientRequest(&clientRequest); - // Current batch status - bool refreshedTargeter = false; - int rounds = 0; - int numCompletedOps = 0; - int numRoundsWithoutProgress = 0; + // Current batch status + bool refreshedTargeter = false; + int rounds = 0; + int numCompletedOps = 0; + int numRoundsWithoutProgress = 0; - while ( !batchOp.isFinished() ) { + while (!batchOp.isFinished()) { + // + // Get child batches to send using the targeter + // + // Targeting errors can be caused by remote metadata changing (the collection could have + // been dropped and recreated, for example with a new shard key). If a remote metadata + // change occurs *before* a client sends us a batch, we need to make sure that we don't + // error out just because we're staler than the client - otherwise mongos will be have + // unpredictable behavior. + // + // (If a metadata change happens *during* or *after* a client sends us a batch, however, + // we make no guarantees about delivery.) + // + // For this reason, we don't record targeting errors until we've refreshed our targeting + // metadata at least once *after* receiving the client batch - at that point, we know: + // + // 1) our new metadata is the same as the metadata when the client sent a batch, and so + // targeting errors are real. + // OR + // 2) our new metadata is a newer version than when the client sent a batch, and so + // the metadata must have changed after the client batch was sent. We don't need to + // deliver in this case, since for all the client knows we may have gotten the batch + // exactly when the metadata changed. + // - // - // Get child batches to send using the targeter - // - // Targeting errors can be caused by remote metadata changing (the collection could have - // been dropped and recreated, for example with a new shard key). If a remote metadata - // change occurs *before* a client sends us a batch, we need to make sure that we don't - // error out just because we're staler than the client - otherwise mongos will be have - // unpredictable behavior. - // - // (If a metadata change happens *during* or *after* a client sends us a batch, however, - // we make no guarantees about delivery.) - // - // For this reason, we don't record targeting errors until we've refreshed our targeting - // metadata at least once *after* receiving the client batch - at that point, we know: - // - // 1) our new metadata is the same as the metadata when the client sent a batch, and so - // targeting errors are real. - // OR - // 2) our new metadata is a newer version than when the client sent a batch, and so - // the metadata must have changed after the client batch was sent. We don't need to - // deliver in this case, since for all the client knows we may have gotten the batch - // exactly when the metadata changed. - // + OwnedPointerVector<TargetedWriteBatch> childBatchesOwned; + vector<TargetedWriteBatch*>& childBatches = childBatchesOwned.mutableVector(); + + // If we've already had a targeting error, we've refreshed the metadata once and can + // record target errors definitively. + bool recordTargetErrors = refreshedTargeter; + Status targetStatus = batchOp.targetBatch(*_targeter, recordTargetErrors, &childBatches); + if (!targetStatus.isOK()) { + // Don't do anything until a targeter refresh + _targeter->noteCouldNotTarget(); + refreshedTargeter = true; + ++_stats->numTargetErrors; + dassert(childBatches.size() == 0u); + } - OwnedPointerVector<TargetedWriteBatch> childBatchesOwned; - vector<TargetedWriteBatch*>& childBatches = childBatchesOwned.mutableVector(); - - // If we've already had a targeting error, we've refreshed the metadata once and can - // record target errors definitively. - bool recordTargetErrors = refreshedTargeter; - Status targetStatus = batchOp.targetBatch( *_targeter, - recordTargetErrors, - &childBatches ); - if ( !targetStatus.isOK() ) { - // Don't do anything until a targeter refresh - _targeter->noteCouldNotTarget(); - refreshedTargeter = true; - ++_stats->numTargetErrors; - dassert( childBatches.size() == 0u ); - } + // + // Send all child batches + // + + size_t numSent = 0; + size_t numToSend = childBatches.size(); + bool remoteMetadataChanging = false; + while (numSent != numToSend) { + // Collect batches out on the network, mapped by endpoint + OwnedHostBatchMap ownedPendingBatches; + OwnedHostBatchMap::MapType& pendingBatches = ownedPendingBatches.mutableMap(); // - // Send all child batches + // Send side // - size_t numSent = 0; - size_t numToSend = childBatches.size(); - bool remoteMetadataChanging = false; - while ( numSent != numToSend ) { - - // Collect batches out on the network, mapped by endpoint - OwnedHostBatchMap ownedPendingBatches; - OwnedHostBatchMap::MapType& pendingBatches = ownedPendingBatches.mutableMap(); - + // Get as many batches as we can at once + for (vector<TargetedWriteBatch*>::iterator it = childBatches.begin(); + it != childBatches.end(); + ++it) { // - // Send side + // Collect the info needed to dispatch our targeted batch // - // Get as many batches as we can at once - for ( vector<TargetedWriteBatch*>::iterator it = childBatches.begin(); - it != childBatches.end(); ++it ) { - - // - // Collect the info needed to dispatch our targeted batch - // - - TargetedWriteBatch* nextBatch = *it; - // If the batch is NULL, we sent it previously, so skip - if ( nextBatch == NULL ) continue; - - // Figure out what host we need to dispatch our targeted batch - ConnectionString shardHost; - Status resolveStatus = _resolver->chooseWriteHost( nextBatch->getEndpoint() - .shardName, - &shardHost ); - if ( !resolveStatus.isOK() ) { - - ++_stats->numResolveErrors; - - // Record a resolve failure - // TODO: It may be necessary to refresh the cache if stale, or maybe just - // cancel and retarget the batch - WriteErrorDetail error; - buildErrorFrom( resolveStatus, &error ); - - LOG( 4 ) << "unable to send write batch to " << shardHost.toString() - << causedBy( resolveStatus.toString() ) << endl; - - batchOp.noteBatchError( *nextBatch, error ); - - // We're done with this batch - // Clean up when we can't resolve a host - delete *it; - *it = NULL; - --numToSend; - continue; - } - - // If we already have a batch for this host, wait until the next time - OwnedHostBatchMap::MapType::iterator pendingIt = pendingBatches.find( shardHost ); - if ( pendingIt != pendingBatches.end() ) continue; - - // - // We now have all the info needed to dispatch the batch - // + TargetedWriteBatch* nextBatch = *it; + // If the batch is NULL, we sent it previously, so skip + if (nextBatch == NULL) + continue; - BatchedCommandRequest request( clientRequest.getBatchType() ); - batchOp.buildBatchRequest( *nextBatch, &request ); + // Figure out what host we need to dispatch our targeted batch + ConnectionString shardHost; + Status resolveStatus = + _resolver->chooseWriteHost(nextBatch->getEndpoint().shardName, &shardHost); + if (!resolveStatus.isOK()) { + ++_stats->numResolveErrors; - // Internally we use full namespaces for request/response, but we send the - // command to a database with the collection name in the request. - NamespaceString nss( request.getNS() ); - request.setNS( nss.coll() ); + // Record a resolve failure + // TODO: It may be necessary to refresh the cache if stale, or maybe just + // cancel and retarget the batch + WriteErrorDetail error; + buildErrorFrom(resolveStatus, &error); - LOG( 4 ) << "sending write batch to " << shardHost.toString() << ": " - << request.toString() << endl; + LOG(4) << "unable to send write batch to " << shardHost.toString() + << causedBy(resolveStatus.toString()) << endl; - _dispatcher->addCommand( shardHost, nss.db(), request ); + batchOp.noteBatchError(*nextBatch, error); - // Indicate we're done by setting the batch to NULL - // We'll only get duplicate hostEndpoints if we have broadcast and non-broadcast - // endpoints for the same host, so this should be pretty efficient without - // moving stuff around. + // We're done with this batch + // Clean up when we can't resolve a host + delete *it; *it = NULL; - - // Recv-side is responsible for cleaning up the nextBatch when used - pendingBatches.insert( make_pair( shardHost, nextBatch ) ); + --numToSend; + continue; } - // Send them all out - _dispatcher->sendAll(); - numSent += pendingBatches.size(); + // If we already have a batch for this host, wait until the next time + OwnedHostBatchMap::MapType::iterator pendingIt = pendingBatches.find(shardHost); + if (pendingIt != pendingBatches.end()) + continue; // - // Recv side + // We now have all the info needed to dispatch the batch // - while ( _dispatcher->numPending() > 0 ) { + BatchedCommandRequest request(clientRequest.getBatchType()); + batchOp.buildBatchRequest(*nextBatch, &request); - // Get the response - ConnectionString shardHost; - BatchedCommandResponse response; - Status dispatchStatus = _dispatcher->recvAny( &shardHost, &response ); + // Internally we use full namespaces for request/response, but we send the + // command to a database with the collection name in the request. + NamespaceString nss(request.getNS()); + request.setNS(nss.coll()); - // Get the TargetedWriteBatch to find where to put the response - dassert( pendingBatches.find( shardHost ) != pendingBatches.end() ); - TargetedWriteBatch* batch = pendingBatches.find( shardHost )->second; + LOG(4) << "sending write batch to " << shardHost.toString() << ": " + << request.toString() << endl; - if ( dispatchStatus.isOK() ) { + _dispatcher->addCommand(shardHost, nss.db(), request); - TrackedErrors trackedErrors; - trackedErrors.startTracking( ErrorCodes::StaleShardVersion ); + // Indicate we're done by setting the batch to NULL + // We'll only get duplicate hostEndpoints if we have broadcast and non-broadcast + // endpoints for the same host, so this should be pretty efficient without + // moving stuff around. + *it = NULL; - LOG( 4 ) << "write results received from " << shardHost.toString() << ": " - << response.toString() << endl; - - // Dispatch was ok, note response - batchOp.noteBatchResponse( *batch, response, &trackedErrors ); + // Recv-side is responsible for cleaning up the nextBatch when used + pendingBatches.insert(make_pair(shardHost, nextBatch)); + } - // Note if anything was stale - const vector<ShardError*>& staleErrors = - trackedErrors.getErrors( ErrorCodes::StaleShardVersion ); + // Send them all out + _dispatcher->sendAll(); + numSent += pendingBatches.size(); - if ( staleErrors.size() > 0 ) { - noteStaleResponses( staleErrors, _targeter ); - ++_stats->numStaleBatches; - } + // + // Recv side + // - // Remember if the shard is actively changing metadata right now - if ( isShardMetadataChanging( staleErrors ) ) { - remoteMetadataChanging = true; - } + while (_dispatcher->numPending() > 0) { + // Get the response + ConnectionString shardHost; + BatchedCommandResponse response; + Status dispatchStatus = _dispatcher->recvAny(&shardHost, &response); - // Remember that we successfully wrote to this shard - // NOTE: This will record lastOps for shards where we actually didn't update - // or delete any documents, which preserves old behavior but is conservative - _stats->noteWriteAt( shardHost, - response.isLastOpSet() ? - response.getLastOp() : Timestamp(), - response.isElectionIdSet() ? - response.getElectionId() : OID()); - } - else { + // Get the TargetedWriteBatch to find where to put the response + dassert(pendingBatches.find(shardHost) != pendingBatches.end()); + TargetedWriteBatch* batch = pendingBatches.find(shardHost)->second; - // Error occurred dispatching, note it + if (dispatchStatus.isOK()) { + TrackedErrors trackedErrors; + trackedErrors.startTracking(ErrorCodes::StaleShardVersion); - stringstream msg; - msg << "write results unavailable from " << shardHost.toString() - << causedBy( dispatchStatus.toString() ); + LOG(4) << "write results received from " << shardHost.toString() << ": " + << response.toString() << endl; - WriteErrorDetail error; - buildErrorFrom( Status( ErrorCodes::RemoteResultsUnavailable, msg.str() ), - &error ); + // Dispatch was ok, note response + batchOp.noteBatchResponse(*batch, response, &trackedErrors); - LOG( 4 ) << "unable to receive write results from " << shardHost.toString() - << causedBy( dispatchStatus.toString() ) << endl; + // Note if anything was stale + const vector<ShardError*>& staleErrors = + trackedErrors.getErrors(ErrorCodes::StaleShardVersion); - batchOp.noteBatchError( *batch, error ); + if (staleErrors.size() > 0) { + noteStaleResponses(staleErrors, _targeter); + ++_stats->numStaleBatches; } - } - } - ++rounds; - ++_stats->numRounds; + // Remember if the shard is actively changing metadata right now + if (isShardMetadataChanging(staleErrors)) { + remoteMetadataChanging = true; + } - // If we're done, get out - if ( batchOp.isFinished() ) - break; + // Remember that we successfully wrote to this shard + // NOTE: This will record lastOps for shards where we actually didn't update + // or delete any documents, which preserves old behavior but is conservative + _stats->noteWriteAt(shardHost, + response.isLastOpSet() ? response.getLastOp() : Timestamp(), + response.isElectionIdSet() ? response.getElectionId() + : OID()); + } else { + // Error occurred dispatching, note it - // MORE WORK TO DO + stringstream msg; + msg << "write results unavailable from " << shardHost.toString() + << causedBy(dispatchStatus.toString()); - // - // Refresh the targeter if we need to (no-op if nothing stale) - // + WriteErrorDetail error; + buildErrorFrom(Status(ErrorCodes::RemoteResultsUnavailable, msg.str()), &error); - bool targeterChanged = false; - Status refreshStatus = _targeter->refreshIfNeeded( &targeterChanged ); + LOG(4) << "unable to receive write results from " << shardHost.toString() + << causedBy(dispatchStatus.toString()) << endl; - if ( !refreshStatus.isOK() ) { - - // It's okay if we can't refresh, we'll just record errors for the ops if - // needed. - warning() << "could not refresh targeter" << causedBy( refreshStatus.reason() ) - << endl; + batchOp.noteBatchError(*batch, error); + } } + } - // - // Ensure progress is being made toward completing the batch op - // + ++rounds; + ++_stats->numRounds; - int currCompletedOps = batchOp.numWriteOpsIn( WriteOpState_Completed ); - if ( currCompletedOps == numCompletedOps && !targeterChanged - && !remoteMetadataChanging ) { - ++numRoundsWithoutProgress; - } - else { - numRoundsWithoutProgress = 0; - } - numCompletedOps = currCompletedOps; + // If we're done, get out + if (batchOp.isFinished()) + break; - if ( numRoundsWithoutProgress > kMaxRoundsWithoutProgress ) { + // MORE WORK TO DO - stringstream msg; - msg << "no progress was made executing batch write op in " << clientRequest.getNS() - << " after " << kMaxRoundsWithoutProgress << " rounds (" << numCompletedOps - << " ops completed in " << rounds << " rounds total)"; + // + // Refresh the targeter if we need to (no-op if nothing stale) + // - WriteErrorDetail error; - buildErrorFrom( Status( ErrorCodes::NoProgressMade, msg.str() ), &error ); - batchOp.abortBatch( error ); - break; - } + bool targeterChanged = false; + Status refreshStatus = _targeter->refreshIfNeeded(&targeterChanged); + + if (!refreshStatus.isOK()) { + // It's okay if we can't refresh, we'll just record errors for the ops if + // needed. + warning() << "could not refresh targeter" << causedBy(refreshStatus.reason()) << endl; } - batchOp.buildClientResponse( clientResponse ); + // + // Ensure progress is being made toward completing the batch op + // - LOG( 4 ) << "finished execution of write batch" - << ( clientResponse->isErrDetailsSet() ? " with write errors" : "") - << ( clientResponse->isErrDetailsSet() && - clientResponse->isWriteConcernErrorSet() ? " and" : "" ) - << ( clientResponse->isWriteConcernErrorSet() ? " with write concern error" : "" ) - << " for " << clientRequest.getNS() << endl; + int currCompletedOps = batchOp.numWriteOpsIn(WriteOpState_Completed); + if (currCompletedOps == numCompletedOps && !targeterChanged && !remoteMetadataChanging) { + ++numRoundsWithoutProgress; + } else { + numRoundsWithoutProgress = 0; + } + numCompletedOps = currCompletedOps; + + if (numRoundsWithoutProgress > kMaxRoundsWithoutProgress) { + stringstream msg; + msg << "no progress was made executing batch write op in " << clientRequest.getNS() + << " after " << kMaxRoundsWithoutProgress << " rounds (" << numCompletedOps + << " ops completed in " << rounds << " rounds total)"; + + WriteErrorDetail error; + buildErrorFrom(Status(ErrorCodes::NoProgressMade, msg.str()), &error); + batchOp.abortBatch(error); + break; + } } - const BatchWriteExecStats& BatchWriteExec::getStats() { - return *_stats; - } + batchOp.buildClientResponse(clientResponse); - BatchWriteExecStats* BatchWriteExec::releaseStats() { - return _stats.release(); - } + LOG(4) << "finished execution of write batch" + << (clientResponse->isErrDetailsSet() ? " with write errors" : "") + << (clientResponse->isErrDetailsSet() && clientResponse->isWriteConcernErrorSet() + ? " and" + : "") + << (clientResponse->isWriteConcernErrorSet() ? " with write concern error" : "") + << " for " << clientRequest.getNS() << endl; +} - void BatchWriteExecStats::noteWriteAt(const ConnectionString& host, - Timestamp opTime, - const OID& electionId) { - _writeOpTimes[host] = HostOpTime(opTime, electionId); - } +const BatchWriteExecStats& BatchWriteExec::getStats() { + return *_stats; +} - const HostOpTimeMap& BatchWriteExecStats::getWriteOpTimes() const { - return _writeOpTimes; - } +BatchWriteExecStats* BatchWriteExec::releaseStats() { + return _stats.release(); +} + +void BatchWriteExecStats::noteWriteAt(const ConnectionString& host, + Timestamp opTime, + const OID& electionId) { + _writeOpTimes[host] = HostOpTime(opTime, electionId); +} + +const HostOpTimeMap& BatchWriteExecStats::getWriteOpTimes() const { + return _writeOpTimes; +} } diff --git a/src/mongo/s/write_ops/batch_write_exec.h b/src/mongo/s/write_ops/batch_write_exec.h index b7fd6c51e43..708b9481aef 100644 --- a/src/mongo/s/write_ops/batch_write_exec.h +++ b/src/mongo/s/write_ops/batch_write_exec.h @@ -41,93 +41,87 @@ namespace mongo { - class BatchWriteExecStats; - class MultiCommandDispatch; +class BatchWriteExecStats; +class MultiCommandDispatch; + +/** + * The BatchWriteExec is able to execute client batch write requests, resulting in a batch + * response to send back to the client. + * + * There are two main interfaces the exec uses to "run" the batch: + * + * - the "targeter" used to generate child batch operations to send to particular shards + * + * - the "dispatcher" used to send child batches to several shards at once, and retrieve the + * results + * + * Both the targeter and dispatcher are assumed to be dedicated to this particular + * BatchWriteExec instance. + * + */ +class BatchWriteExec { + MONGO_DISALLOW_COPYING(BatchWriteExec); + +public: + BatchWriteExec(NSTargeter* targeter, ShardResolver* resolver, MultiCommandDispatch* dispatcher); /** - * The BatchWriteExec is able to execute client batch write requests, resulting in a batch - * response to send back to the client. - * - * There are two main interfaces the exec uses to "run" the batch: - * - * - the "targeter" used to generate child batch operations to send to particular shards - * - * - the "dispatcher" used to send child batches to several shards at once, and retrieve the - * results - * - * Both the targeter and dispatcher are assumed to be dedicated to this particular - * BatchWriteExec instance. + * Executes a client batch write request by sending child batches to several shard + * endpoints, and returns a client batch write response. * + * This function does not throw, any errors are reported via the clientResponse. */ - class BatchWriteExec { - MONGO_DISALLOW_COPYING (BatchWriteExec); - public: - - BatchWriteExec( NSTargeter* targeter, - ShardResolver* resolver, - MultiCommandDispatch* dispatcher ); - - /** - * Executes a client batch write request by sending child batches to several shard - * endpoints, and returns a client batch write response. - * - * This function does not throw, any errors are reported via the clientResponse. - */ - void executeBatch( const BatchedCommandRequest& clientRequest, - BatchedCommandResponse* clientResponse ); - - const BatchWriteExecStats& getStats(); - - BatchWriteExecStats* releaseStats(); - - private: + void executeBatch(const BatchedCommandRequest& clientRequest, + BatchedCommandResponse* clientResponse); - // Not owned here - NSTargeter* _targeter; + const BatchWriteExecStats& getStats(); - // Not owned here - ShardResolver* _resolver; + BatchWriteExecStats* releaseStats(); - // Not owned here - MultiCommandDispatch* _dispatcher; +private: + // Not owned here + NSTargeter* _targeter; - // Stats - std::unique_ptr<BatchWriteExecStats> _stats; - }; + // Not owned here + ShardResolver* _resolver; - struct HostOpTime { - HostOpTime(Timestamp ot, OID e) : opTime(ot), electionId(e) {}; - HostOpTime() {}; - Timestamp opTime; - OID electionId; - }; + // Not owned here + MultiCommandDispatch* _dispatcher; - typedef std::map<ConnectionString, HostOpTime> HostOpTimeMap; + // Stats + std::unique_ptr<BatchWriteExecStats> _stats; +}; - class BatchWriteExecStats { - public: +struct HostOpTime { + HostOpTime(Timestamp ot, OID e) : opTime(ot), electionId(e){}; + HostOpTime(){}; + Timestamp opTime; + OID electionId; +}; - BatchWriteExecStats() : - numRounds( 0 ), numTargetErrors( 0 ), numResolveErrors( 0 ), numStaleBatches( 0 ) { - } +typedef std::map<ConnectionString, HostOpTime> HostOpTimeMap; - void noteWriteAt(const ConnectionString& host, Timestamp opTime, const OID& electionId); +class BatchWriteExecStats { +public: + BatchWriteExecStats() + : numRounds(0), numTargetErrors(0), numResolveErrors(0), numStaleBatches(0) {} - const HostOpTimeMap& getWriteOpTimes() const; + void noteWriteAt(const ConnectionString& host, Timestamp opTime, const OID& electionId); - // Expose via helpers if this gets more complex + const HostOpTimeMap& getWriteOpTimes() const; - // Number of round trips required for the batch - int numRounds; - // Number of times targeting failed - int numTargetErrors; - // Number of times host resolution failed - int numResolveErrors; - // Number of stale batches - int numStaleBatches; + // Expose via helpers if this gets more complex - private: + // Number of round trips required for the batch + int numRounds; + // Number of times targeting failed + int numTargetErrors; + // Number of times host resolution failed + int numResolveErrors; + // Number of stale batches + int numStaleBatches; - HostOpTimeMap _writeOpTimes; - }; +private: + HostOpTimeMap _writeOpTimes; +}; } diff --git a/src/mongo/s/write_ops/batch_write_exec_test.cpp b/src/mongo/s/write_ops/batch_write_exec_test.cpp index 3e0b2cfb9f3..21060c79983 100644 --- a/src/mongo/s/write_ops/batch_write_exec_test.cpp +++ b/src/mongo/s/write_ops/batch_write_exec_test.cpp @@ -39,277 +39,266 @@ namespace { - using std::unique_ptr; - using std::string; - using std::vector; +using std::unique_ptr; +using std::string; +using std::vector; - using namespace mongo; +using namespace mongo; - /** - * Mimics a single shard backend for a particular collection which can be initialized with a - * set of write command results to return. - */ - class MockSingleShardBackend { - public: - - MockSingleShardBackend( const NamespaceString& nss ) { - - // Initialize targeting to a mock shard - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpoint, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << MAXKEY ) ) ); - targeter.init( mockRanges ); - - // Get the connection string for the mock shard - resolver.chooseWriteHost( mockRanges.front()->endpoint.shardName, &shardHost ); +/** + * Mimics a single shard backend for a particular collection which can be initialized with a + * set of write command results to return. + */ +class MockSingleShardBackend { +public: + MockSingleShardBackend(const NamespaceString& nss) { + // Initialize targeting to a mock shard + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + vector<MockRange*> mockRanges; + mockRanges.push_back( + new MockRange(endpoint, nss, BSON("x" << MINKEY), BSON("x" << MAXKEY))); + targeter.init(mockRanges); + + // Get the connection string for the mock shard + resolver.chooseWriteHost(mockRanges.front()->endpoint.shardName, &shardHost); + + // Executor using the mock backend + exec.reset(new BatchWriteExec(&targeter, &resolver, &dispatcher)); + } - // Executor using the mock backend - exec.reset( new BatchWriteExec( &targeter, &resolver, &dispatcher ) ); - } + void setMockResults(const vector<MockWriteResult*>& results) { + dispatcher.init(results); + } - void setMockResults( const vector<MockWriteResult*>& results ) { - dispatcher.init( results ); - } + ConnectionString shardHost; - ConnectionString shardHost; + MockNSTargeter targeter; + MockShardResolver resolver; + MockMultiWriteCommand dispatcher; - MockNSTargeter targeter; - MockShardResolver resolver; - MockMultiWriteCommand dispatcher; + unique_ptr<BatchWriteExec> exec; +}; - unique_ptr<BatchWriteExec> exec; - }; +// +// Tests for the BatchWriteExec +// +TEST(BatchWriteExecTests, SingleOp) { // - // Tests for the BatchWriteExec + // Basic execution test // - TEST(BatchWriteExecTests, SingleOp) { - - // - // Basic execution test - // + NamespaceString nss("foo.bar"); - NamespaceString nss( "foo.bar" ); + MockSingleShardBackend backend(nss); - MockSingleShardBackend backend( nss ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.setWriteConcern(BSONObj()); + // Do single-target, single doc batch write op + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.setWriteConcern( BSONObj() ); - // Do single-target, single doc batch write op - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + BatchedCommandResponse response; + backend.exec->executeBatch(request, &response); + ASSERT(response.getOk()); - BatchedCommandResponse response; - backend.exec->executeBatch( request, &response ); - ASSERT( response.getOk() ); + const BatchWriteExecStats& stats = backend.exec->getStats(); + ASSERT_EQUALS(stats.numRounds, 1); +} - const BatchWriteExecStats& stats = backend.exec->getStats(); - ASSERT_EQUALS( stats.numRounds, 1 ); - } +TEST(BatchWriteExecTests, SingleOpError) { + // + // Basic error test + // - TEST(BatchWriteExecTests, SingleOpError) { + NamespaceString nss("foo.bar"); - // - // Basic error test - // + MockSingleShardBackend backend(nss); - NamespaceString nss( "foo.bar" ); + vector<MockWriteResult*> mockResults; + BatchedCommandResponse errResponse; + errResponse.setOk(false); + errResponse.setErrCode(ErrorCodes::UnknownError); + errResponse.setErrMessage("mock error"); + mockResults.push_back(new MockWriteResult(backend.shardHost, errResponse)); - MockSingleShardBackend backend( nss ); + backend.setMockResults(mockResults); - vector<MockWriteResult*> mockResults; - BatchedCommandResponse errResponse; - errResponse.setOk( false ); - errResponse.setErrCode( ErrorCodes::UnknownError ); - errResponse.setErrMessage( "mock error" ); - mockResults.push_back( new MockWriteResult( backend.shardHost, errResponse ) ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.setWriteConcern(BSONObj()); + // Do single-target, single doc batch write op + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - backend.setMockResults( mockResults ); + BatchedCommandResponse response; + backend.exec->executeBatch(request, &response); + ASSERT(response.getOk()); + ASSERT_EQUALS(response.getN(), 0); + ASSERT(response.isErrDetailsSet()); + ASSERT_EQUALS(response.getErrDetailsAt(0)->getErrCode(), errResponse.getErrCode()); + ASSERT(response.getErrDetailsAt(0)->getErrMessage().find(errResponse.getErrMessage()) != + string::npos); - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.setWriteConcern( BSONObj() ); - // Do single-target, single doc batch write op - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + const BatchWriteExecStats& stats = backend.exec->getStats(); + ASSERT_EQUALS(stats.numRounds, 1); +} - BatchedCommandResponse response; - backend.exec->executeBatch( request, &response ); - ASSERT( response.getOk() ); - ASSERT_EQUALS( response.getN(), 0 ); - ASSERT( response.isErrDetailsSet() ); - ASSERT_EQUALS( response.getErrDetailsAt( 0 )->getErrCode(), errResponse.getErrCode() ); - ASSERT( response.getErrDetailsAt( 0 )->getErrMessage().find( errResponse.getErrMessage() ) - != string::npos ); - - const BatchWriteExecStats& stats = backend.exec->getStats(); - ASSERT_EQUALS( stats.numRounds, 1 ); - } +// +// Test retryable errors +// +TEST(BatchWriteExecTests, StaleOp) { // - // Test retryable errors + // Retry op in exec b/c of stale config // - TEST(BatchWriteExecTests, StaleOp) { - - // - // Retry op in exec b/c of stale config - // + NamespaceString nss("foo.bar"); - NamespaceString nss( "foo.bar" ); + // Insert request + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.setWriteConcern(BSONObj()); + // Do single-target, single doc batch write op + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - // Insert request - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.setWriteConcern( BSONObj() ); - // Do single-target, single doc batch write op - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + MockSingleShardBackend backend(nss); - MockSingleShardBackend backend( nss ); + vector<MockWriteResult*> mockResults; + WriteErrorDetail error; + error.setErrCode(ErrorCodes::StaleShardVersion); + error.setErrMessage("mock stale error"); + mockResults.push_back(new MockWriteResult(backend.shardHost, error)); - vector<MockWriteResult*> mockResults; - WriteErrorDetail error; - error.setErrCode( ErrorCodes::StaleShardVersion ); - error.setErrMessage( "mock stale error" ); - mockResults.push_back( new MockWriteResult( backend.shardHost, error ) ); + backend.setMockResults(mockResults); - backend.setMockResults( mockResults ); + // Execute request + BatchedCommandResponse response; + backend.exec->executeBatch(request, &response); + ASSERT(response.getOk()); - // Execute request - BatchedCommandResponse response; - backend.exec->executeBatch( request, &response ); - ASSERT( response.getOk() ); + const BatchWriteExecStats& stats = backend.exec->getStats(); + ASSERT_EQUALS(stats.numStaleBatches, 1); +} - const BatchWriteExecStats& stats = backend.exec->getStats(); - ASSERT_EQUALS( stats.numStaleBatches, 1 ); - } - - TEST(BatchWriteExecTests, MultiStaleOp) { +TEST(BatchWriteExecTests, MultiStaleOp) { + // + // Retry op in exec multiple times b/c of stale config + // - // - // Retry op in exec multiple times b/c of stale config - // + NamespaceString nss("foo.bar"); - NamespaceString nss( "foo.bar" ); + // Insert request + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.setWriteConcern(BSONObj()); + // Do single-target, single doc batch write op + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - // Insert request - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.setWriteConcern( BSONObj() ); - // Do single-target, single doc batch write op - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + MockSingleShardBackend backend(nss); - MockSingleShardBackend backend( nss ); + vector<MockWriteResult*> mockResults; + WriteErrorDetail error; + error.setErrCode(ErrorCodes::StaleShardVersion); + error.setErrMessage("mock stale error"); + for (int i = 0; i < 3; i++) { + mockResults.push_back(new MockWriteResult(backend.shardHost, error)); + } - vector<MockWriteResult*> mockResults; - WriteErrorDetail error; - error.setErrCode( ErrorCodes::StaleShardVersion ); - error.setErrMessage( "mock stale error" ); - for ( int i = 0; i < 3; i++ ) { - mockResults.push_back( new MockWriteResult( backend.shardHost, error ) ); - } + backend.setMockResults(mockResults); - backend.setMockResults( mockResults ); + // Execute request + BatchedCommandResponse response; + backend.exec->executeBatch(request, &response); + ASSERT(response.getOk()); - // Execute request - BatchedCommandResponse response; - backend.exec->executeBatch( request, &response ); - ASSERT( response.getOk() ); + const BatchWriteExecStats& stats = backend.exec->getStats(); + ASSERT_EQUALS(stats.numStaleBatches, 3); +} - const BatchWriteExecStats& stats = backend.exec->getStats(); - ASSERT_EQUALS( stats.numStaleBatches, 3 ); - } +TEST(BatchWriteExecTests, TooManyStaleOp) { + // + // Retry op in exec too many times (without refresh) b/c of stale config + // (The mock targeter doesn't report progress on refresh) + // We should report a no progress error for everything in the batch + // - TEST(BatchWriteExecTests, TooManyStaleOp) { - - // - // Retry op in exec too many times (without refresh) b/c of stale config - // (The mock targeter doesn't report progress on refresh) - // We should report a no progress error for everything in the batch - // - - NamespaceString nss( "foo.bar" ); - - // Insert request - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.setWriteConcern( BSONObj() ); - // Do single-target, single doc batch write ops - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 2 ) ); - - MockSingleShardBackend backend( nss ); - - vector<MockWriteResult*> mockResults; - WriteErrorDetail error; - error.setErrCode( ErrorCodes::StaleShardVersion ); - error.setErrMessage( "mock stale error" ); - for ( int i = 0; i < 10; i++ ) { - mockResults.push_back( new MockWriteResult( backend.shardHost, - error, - request.sizeWriteOps() ) ); - } - - backend.setMockResults( mockResults ); - - // Execute request - BatchedCommandResponse response; - backend.exec->executeBatch( request, &response ); - ASSERT( response.getOk() ); - ASSERT_EQUALS( response.getN(), 0 ); - ASSERT( response.isErrDetailsSet() ); - ASSERT_EQUALS( response.getErrDetailsAt( 0 )->getErrCode(), ErrorCodes::NoProgressMade ); - ASSERT_EQUALS( response.getErrDetailsAt( 1 )->getErrCode(), ErrorCodes::NoProgressMade ); + NamespaceString nss("foo.bar"); + + // Insert request + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.setWriteConcern(BSONObj()); + // Do single-target, single doc batch write ops + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 2)); + + MockSingleShardBackend backend(nss); + + vector<MockWriteResult*> mockResults; + WriteErrorDetail error; + error.setErrCode(ErrorCodes::StaleShardVersion); + error.setErrMessage("mock stale error"); + for (int i = 0; i < 10; i++) { + mockResults.push_back( + new MockWriteResult(backend.shardHost, error, request.sizeWriteOps())); } - TEST(BatchWriteExecTests, ManyStaleOpWithMigration) { + backend.setMockResults(mockResults); - // - // Retry op in exec many times b/c of stale config, but simulate remote migrations occurring - // + // Execute request + BatchedCommandResponse response; + backend.exec->executeBatch(request, &response); + ASSERT(response.getOk()); + ASSERT_EQUALS(response.getN(), 0); + ASSERT(response.isErrDetailsSet()); + ASSERT_EQUALS(response.getErrDetailsAt(0)->getErrCode(), ErrorCodes::NoProgressMade); + ASSERT_EQUALS(response.getErrDetailsAt(1)->getErrCode(), ErrorCodes::NoProgressMade); +} - NamespaceString nss( "foo.bar" ); - - // Insert request - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.setWriteConcern( BSONObj() ); - // Do single-target, single doc batch write op - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - - MockSingleShardBackend backend( nss ); - - vector<MockWriteResult*> mockResults; - WriteErrorDetail error; - error.setErrCode( ErrorCodes::StaleShardVersion ); - error.setErrMessage( "mock stale error" ); - for ( int i = 0; i < 10; i++ ) { - if ( i % 2 == 0 ) - error.setErrInfo( BSONObj() ); - else - error.setErrInfo( BSON( "inCriticalSection" << true ) ); +TEST(BatchWriteExecTests, ManyStaleOpWithMigration) { + // + // Retry op in exec many times b/c of stale config, but simulate remote migrations occurring + // - mockResults.push_back( new MockWriteResult( backend.shardHost, error ) ); - } + NamespaceString nss("foo.bar"); + + // Insert request + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.setWriteConcern(BSONObj()); + // Do single-target, single doc batch write op + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + + MockSingleShardBackend backend(nss); + + vector<MockWriteResult*> mockResults; + WriteErrorDetail error; + error.setErrCode(ErrorCodes::StaleShardVersion); + error.setErrMessage("mock stale error"); + for (int i = 0; i < 10; i++) { + if (i % 2 == 0) + error.setErrInfo(BSONObj()); + else + error.setErrInfo(BSON("inCriticalSection" << true)); + + mockResults.push_back(new MockWriteResult(backend.shardHost, error)); + } - backend.setMockResults( mockResults ); + backend.setMockResults(mockResults); - // Execute request - BatchedCommandResponse response; - backend.exec->executeBatch( request, &response ); - ASSERT( response.getOk() ); + // Execute request + BatchedCommandResponse response; + backend.exec->executeBatch(request, &response); + ASSERT(response.getOk()); - const BatchWriteExecStats& stats = backend.exec->getStats(); - ASSERT_EQUALS( stats.numStaleBatches, 10 ); - } + const BatchWriteExecStats& stats = backend.exec->getStats(); + ASSERT_EQUALS(stats.numStaleBatches, 10); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/write_ops/batch_write_op.cpp b/src/mongo/s/write_ops/batch_write_op.cpp index b1032410c7a..847b96ff9ee 100644 --- a/src/mongo/s/write_ops/batch_write_op.cpp +++ b/src/mongo/s/write_ops/batch_write_op.cpp @@ -32,889 +32,829 @@ namespace mongo { - using std::unique_ptr; - using std::make_pair; - using std::set; - using std::stringstream; - using std::vector; - - /** - * Returns a new write concern that has the copy of every field from the original - * document but with a w set to 1. This is intended for upgrading { w: 0 } write - * concern to { w: 1 }. - */ - static BSONObj upgradeWriteConcern ( const BSONObj& origWriteConcern ) { - BSONObjIterator iter( origWriteConcern ); - BSONObjBuilder newWriteConcern; - - while ( iter.more() ) { - BSONElement elem( iter.next() ); - - if ( strncmp( elem.fieldName(), "w", 2 ) == 0 ) { - newWriteConcern.append( "w", 1 ); - } - else { - newWriteConcern.append( elem ); - } - } - - return newWriteConcern.obj(); - } - - BatchWriteStats::BatchWriteStats() : - numInserted( 0 ), numUpserted( 0 ), numMatched( 0 ), numModified( 0 ), numDeleted( 0 ) { - } +using std::unique_ptr; +using std::make_pair; +using std::set; +using std::stringstream; +using std::vector; - BatchWriteOp::BatchWriteOp() : - _clientRequest( NULL ), _writeOps( NULL ), _stats( new BatchWriteStats ) { - } - - void BatchWriteOp::initClientRequest( const BatchedCommandRequest* clientRequest ) { - dassert( clientRequest->isValid( NULL ) ); +/** + * Returns a new write concern that has the copy of every field from the original + * document but with a w set to 1. This is intended for upgrading { w: 0 } write + * concern to { w: 1 }. + */ +static BSONObj upgradeWriteConcern(const BSONObj& origWriteConcern) { + BSONObjIterator iter(origWriteConcern); + BSONObjBuilder newWriteConcern; - size_t numWriteOps = clientRequest->sizeWriteOps(); - _writeOps = static_cast<WriteOp*>( ::operator new[]( numWriteOps * sizeof(WriteOp) ) ); + while (iter.more()) { + BSONElement elem(iter.next()); - for ( size_t i = 0; i < numWriteOps; ++i ) { - // Don't want to have to define what an empty WriteOp means, so construct in-place - new ( &_writeOps[i] ) WriteOp( BatchItemRef( clientRequest, i ) ); + if (strncmp(elem.fieldName(), "w", 2) == 0) { + newWriteConcern.append("w", 1); + } else { + newWriteConcern.append(elem); } - - _clientRequest = clientRequest; } - // Arbitrary endpoint ordering, needed for grouping by endpoint - static int compareEndpoints( const ShardEndpoint* endpointA, const ShardEndpoint* endpointB ) { - - int shardNameDiff = endpointA->shardName.compare( endpointB->shardName ); - if ( shardNameDiff != 0 ) return shardNameDiff; - - long shardVersionDiff = endpointA->shardVersion.toLong() - endpointB->shardVersion.toLong(); - if ( shardVersionDiff != 0 ) return shardVersionDiff; - - int shardEpochDiff = - endpointA->shardVersion.epoch().compare( endpointB->shardVersion.epoch() ); - return shardEpochDiff; - } - - namespace { - - // - // Types for comparing shard endpoints in a map - // - - struct EndpointComp { - bool operator()( const ShardEndpoint* endpointA, - const ShardEndpoint* endpointB ) const { - return compareEndpoints( endpointA, endpointB ) < 0; - } - }; - - typedef std::map<const ShardEndpoint*, TargetedWriteBatch*, EndpointComp> TargetedBatchMap; - - // - // Types for tracking batch sizes - // + return newWriteConcern.obj(); +} - struct BatchSize { +BatchWriteStats::BatchWriteStats() + : numInserted(0), numUpserted(0), numMatched(0), numModified(0), numDeleted(0) {} - BatchSize() : - numOps(0), sizeBytes(0) { - } +BatchWriteOp::BatchWriteOp() : _clientRequest(NULL), _writeOps(NULL), _stats(new BatchWriteStats) {} - int numOps; - int sizeBytes; - }; +void BatchWriteOp::initClientRequest(const BatchedCommandRequest* clientRequest) { + dassert(clientRequest->isValid(NULL)); - typedef std::map<const ShardEndpoint*, BatchSize, EndpointComp> TargetedBatchSizeMap; - } + size_t numWriteOps = clientRequest->sizeWriteOps(); + _writeOps = static_cast<WriteOp*>(::operator new[](numWriteOps * sizeof(WriteOp))); - static void buildTargetError( const Status& errStatus, WriteErrorDetail* details ) { - details->setErrCode( errStatus.code() ); - details->setErrMessage( errStatus.reason() ); + for (size_t i = 0; i < numWriteOps; ++i) { + // Don't want to have to define what an empty WriteOp means, so construct in-place + new (&_writeOps[i]) WriteOp(BatchItemRef(clientRequest, i)); } - // Helper to determine whether a number of targeted writes require a new targeted batch - static bool isNewBatchRequired( const vector<TargetedWrite*>& writes, - const TargetedBatchMap& batchMap ) { - - for ( vector<TargetedWrite*>::const_iterator it = writes.begin(); it != writes.end(); - ++it ) { + _clientRequest = clientRequest; +} - TargetedWrite* write = *it; - if ( batchMap.find( &write->endpoint ) == batchMap.end() ) { - return true; - } - } +// Arbitrary endpoint ordering, needed for grouping by endpoint +static int compareEndpoints(const ShardEndpoint* endpointA, const ShardEndpoint* endpointB) { + int shardNameDiff = endpointA->shardName.compare(endpointB->shardName); + if (shardNameDiff != 0) + return shardNameDiff; - return false; - } + long shardVersionDiff = endpointA->shardVersion.toLong() - endpointB->shardVersion.toLong(); + if (shardVersionDiff != 0) + return shardVersionDiff; - // MAGIC NUMBERS - // Before serializing updates/deletes, we don't know how big their fields would be, but we break - // batches before serializing. - // TODO: Revisit when we revisit command limits in general - static const int kEstUpdateOverheadBytes = (BSONObjMaxInternalSize - BSONObjMaxUserSize) / 100; - static const int kEstDeleteOverheadBytes = (BSONObjMaxInternalSize - BSONObjMaxUserSize) / 100; + int shardEpochDiff = endpointA->shardVersion.epoch().compare(endpointB->shardVersion.epoch()); + return shardEpochDiff; +} - static int getWriteSizeBytes(const WriteOp& writeOp) { +namespace { - const BatchItemRef& item = writeOp.getWriteItem(); - BatchedCommandRequest::BatchType batchType = item.getOpType(); +// +// Types for comparing shard endpoints in a map +// - if (batchType == BatchedCommandRequest::BatchType_Insert) { - return item.getDocument().objsize(); - } - else if (batchType == BatchedCommandRequest::BatchType_Update) { - // Note: Be conservative here - it's okay if we send slightly too many batches - int estSize = item.getUpdate()->getQuery().objsize() - + item.getUpdate()->getUpdateExpr().objsize() + kEstUpdateOverheadBytes; - dassert(estSize >= item.getUpdate()->toBSON().objsize()); - return estSize; - } - else { - dassert( batchType == BatchedCommandRequest::BatchType_Delete ); - // Note: Be conservative here - it's okay if we send slightly too many batches - int estSize = item.getDelete()->getQuery().objsize() + kEstDeleteOverheadBytes; - dassert(estSize >= item.getDelete()->toBSON().objsize()); - return estSize; - } +struct EndpointComp { + bool operator()(const ShardEndpoint* endpointA, const ShardEndpoint* endpointB) const { + return compareEndpoints(endpointA, endpointB) < 0; } +}; - // Helper to determine whether a number of targeted writes require a new targeted batch - static bool wouldMakeBatchesTooBig(const vector<TargetedWrite*>& writes, - int writeSizeBytes, - const TargetedBatchSizeMap& batchSizes) { +typedef std::map<const ShardEndpoint*, TargetedWriteBatch*, EndpointComp> TargetedBatchMap; - for (vector<TargetedWrite*>::const_iterator it = writes.begin(); it != writes.end(); ++it) { +// +// Types for tracking batch sizes +// - const TargetedWrite* write = *it; - TargetedBatchSizeMap::const_iterator seenIt = batchSizes.find(&write->endpoint); +struct BatchSize { + BatchSize() : numOps(0), sizeBytes(0) {} - if (seenIt == batchSizes.end()) { - // If this is the first item in the batch, it can't be too big - continue; - } + int numOps; + int sizeBytes; +}; - const BatchSize& batchSize = seenIt->second; +typedef std::map<const ShardEndpoint*, BatchSize, EndpointComp> TargetedBatchSizeMap; +} - if (batchSize.numOps >= static_cast<int>(BatchedCommandRequest::kMaxWriteBatchSize)) { - // Too many items in batch - return true; - } +static void buildTargetError(const Status& errStatus, WriteErrorDetail* details) { + details->setErrCode(errStatus.code()); + details->setErrMessage(errStatus.reason()); +} - if (batchSize.sizeBytes + writeSizeBytes > BSONObjMaxUserSize) { - // Batch would be too big - return true; - } +// Helper to determine whether a number of targeted writes require a new targeted batch +static bool isNewBatchRequired(const vector<TargetedWrite*>& writes, + const TargetedBatchMap& batchMap) { + for (vector<TargetedWrite*>::const_iterator it = writes.begin(); it != writes.end(); ++it) { + TargetedWrite* write = *it; + if (batchMap.find(&write->endpoint) == batchMap.end()) { + return true; } - - return false; } - // Helper function to cancel all the write ops of targeted batches in a map - static void cancelBatches( const WriteErrorDetail& why, - WriteOp* writeOps, - TargetedBatchMap* batchMap ) { - - set<WriteOp*> targetedWriteOps; + return false; +} - // Collect all the writeOps that are currently targeted - for ( TargetedBatchMap::iterator it = batchMap->begin(); it != batchMap->end(); ) { +// MAGIC NUMBERS +// Before serializing updates/deletes, we don't know how big their fields would be, but we break +// batches before serializing. +// TODO: Revisit when we revisit command limits in general +static const int kEstUpdateOverheadBytes = (BSONObjMaxInternalSize - BSONObjMaxUserSize) / 100; +static const int kEstDeleteOverheadBytes = (BSONObjMaxInternalSize - BSONObjMaxUserSize) / 100; + +static int getWriteSizeBytes(const WriteOp& writeOp) { + const BatchItemRef& item = writeOp.getWriteItem(); + BatchedCommandRequest::BatchType batchType = item.getOpType(); + + if (batchType == BatchedCommandRequest::BatchType_Insert) { + return item.getDocument().objsize(); + } else if (batchType == BatchedCommandRequest::BatchType_Update) { + // Note: Be conservative here - it's okay if we send slightly too many batches + int estSize = item.getUpdate()->getQuery().objsize() + + item.getUpdate()->getUpdateExpr().objsize() + kEstUpdateOverheadBytes; + dassert(estSize >= item.getUpdate()->toBSON().objsize()); + return estSize; + } else { + dassert(batchType == BatchedCommandRequest::BatchType_Delete); + // Note: Be conservative here - it's okay if we send slightly too many batches + int estSize = item.getDelete()->getQuery().objsize() + kEstDeleteOverheadBytes; + dassert(estSize >= item.getDelete()->toBSON().objsize()); + return estSize; + } +} - TargetedWriteBatch* batch = it->second; - const vector<TargetedWrite*>& writes = batch->getWrites(); +// Helper to determine whether a number of targeted writes require a new targeted batch +static bool wouldMakeBatchesTooBig(const vector<TargetedWrite*>& writes, + int writeSizeBytes, + const TargetedBatchSizeMap& batchSizes) { + for (vector<TargetedWrite*>::const_iterator it = writes.begin(); it != writes.end(); ++it) { + const TargetedWrite* write = *it; + TargetedBatchSizeMap::const_iterator seenIt = batchSizes.find(&write->endpoint); - for ( vector<TargetedWrite*>::const_iterator writeIt = writes.begin(); - writeIt != writes.end(); ++writeIt ) { + if (seenIt == batchSizes.end()) { + // If this is the first item in the batch, it can't be too big + continue; + } - TargetedWrite* write = *writeIt; + const BatchSize& batchSize = seenIt->second; - // NOTE: We may repeatedly cancel a write op here, but that's fast and we want to - // cancel before erasing the TargetedWrite* (which owns the cancelled targeting - // info) for reporting reasons. - writeOps[write->writeOpRef.first].cancelWrites( &why ); - } + if (batchSize.numOps >= static_cast<int>(BatchedCommandRequest::kMaxWriteBatchSize)) { + // Too many items in batch + return true; + } - // Note that we need to *erase* first, *then* delete, since the map keys are ptrs from - // the values - batchMap->erase( it++ ); - delete batch; + if (batchSize.sizeBytes + writeSizeBytes > BSONObjMaxUserSize) { + // Batch would be too big + return true; } - batchMap->clear(); } - Status BatchWriteOp::targetBatch( const NSTargeter& targeter, - bool recordTargetErrors, - vector<TargetedWriteBatch*>* targetedBatches ) { - - // - // Targeting of unordered batches is fairly simple - each remaining write op is targeted, - // and each of those targeted writes are grouped into a batch for a particular shard - // endpoint. - // - // Targeting of ordered batches is a bit more complex - to respect the ordering of the - // batch, we can only send: - // A) a single targeted batch to one shard endpoint - // B) multiple targeted batches, but only containing targeted writes for a single write op - // - // This means that any multi-shard write operation must be targeted and sent one-by-one. - // Subsequent single-shard write operations can be batched together if they go to the same - // place. - // - // Ex: ShardA : { skey : a->k }, ShardB : { skey : k->z } - // - // Ordered insert batch of: [{ skey : a }, { skey : b }, { skey : x }] - // broken into: - // [{ skey : a }, { skey : b }], - // [{ skey : x }] - // - // Ordered update Batch of : - // [{ skey : a }{ $push }, - // { skey : b }{ $push }, - // { skey : [c, x] }{ $push }, - // { skey : y }{ $push }, - // { skey : z }{ $push }] - // broken into: - // [{ skey : a }, { skey : b }], - // [{ skey : [c,x] }], - // [{ skey : y }, { skey : z }] - // - - const bool ordered = _clientRequest->getOrdered(); - - TargetedBatchMap batchMap; - TargetedBatchSizeMap batchSizes; + return false; +} - int numTargetErrors = 0; +// Helper function to cancel all the write ops of targeted batches in a map +static void cancelBatches(const WriteErrorDetail& why, + WriteOp* writeOps, + TargetedBatchMap* batchMap) { + set<WriteOp*> targetedWriteOps; - size_t numWriteOps = _clientRequest->sizeWriteOps(); - for ( size_t i = 0; i < numWriteOps; ++i ) { + // Collect all the writeOps that are currently targeted + for (TargetedBatchMap::iterator it = batchMap->begin(); it != batchMap->end();) { + TargetedWriteBatch* batch = it->second; + const vector<TargetedWrite*>& writes = batch->getWrites(); - WriteOp& writeOp = _writeOps[i]; + for (vector<TargetedWrite*>::const_iterator writeIt = writes.begin(); + writeIt != writes.end(); + ++writeIt) { + TargetedWrite* write = *writeIt; - // Only target _Ready ops - if ( writeOp.getWriteState() != WriteOpState_Ready ) continue; + // NOTE: We may repeatedly cancel a write op here, but that's fast and we want to + // cancel before erasing the TargetedWrite* (which owns the cancelled targeting + // info) for reporting reasons. + writeOps[write->writeOpRef.first].cancelWrites(&why); + } - // - // Get TargetedWrites from the targeter for the write operation - // + // Note that we need to *erase* first, *then* delete, since the map keys are ptrs from + // the values + batchMap->erase(it++); + delete batch; + } + batchMap->clear(); +} - // TargetedWrites need to be owned once returned - OwnedPointerVector<TargetedWrite> writesOwned; - vector<TargetedWrite*>& writes = writesOwned.mutableVector(); +Status BatchWriteOp::targetBatch(const NSTargeter& targeter, + bool recordTargetErrors, + vector<TargetedWriteBatch*>* targetedBatches) { + // + // Targeting of unordered batches is fairly simple - each remaining write op is targeted, + // and each of those targeted writes are grouped into a batch for a particular shard + // endpoint. + // + // Targeting of ordered batches is a bit more complex - to respect the ordering of the + // batch, we can only send: + // A) a single targeted batch to one shard endpoint + // B) multiple targeted batches, but only containing targeted writes for a single write op + // + // This means that any multi-shard write operation must be targeted and sent one-by-one. + // Subsequent single-shard write operations can be batched together if they go to the same + // place. + // + // Ex: ShardA : { skey : a->k }, ShardB : { skey : k->z } + // + // Ordered insert batch of: [{ skey : a }, { skey : b }, { skey : x }] + // broken into: + // [{ skey : a }, { skey : b }], + // [{ skey : x }] + // + // Ordered update Batch of : + // [{ skey : a }{ $push }, + // { skey : b }{ $push }, + // { skey : [c, x] }{ $push }, + // { skey : y }{ $push }, + // { skey : z }{ $push }] + // broken into: + // [{ skey : a }, { skey : b }], + // [{ skey : [c,x] }], + // [{ skey : y }, { skey : z }] + // - Status targetStatus = writeOp.targetWrites( targeter, &writes ); + const bool ordered = _clientRequest->getOrdered(); - if ( !targetStatus.isOK() ) { + TargetedBatchMap batchMap; + TargetedBatchSizeMap batchSizes; - WriteErrorDetail targetError; - buildTargetError( targetStatus, &targetError ); + int numTargetErrors = 0; - if ( !recordTargetErrors ) { + size_t numWriteOps = _clientRequest->sizeWriteOps(); + for (size_t i = 0; i < numWriteOps; ++i) { + WriteOp& writeOp = _writeOps[i]; - // Cancel current batch state with an error + // Only target _Ready ops + if (writeOp.getWriteState() != WriteOpState_Ready) + continue; - cancelBatches( targetError, _writeOps, &batchMap ); - dassert( batchMap.empty() ); - return targetStatus; - } - else if ( !ordered || batchMap.empty() ) { + // + // Get TargetedWrites from the targeter for the write operation + // - // Record an error for this batch + // TargetedWrites need to be owned once returned + OwnedPointerVector<TargetedWrite> writesOwned; + vector<TargetedWrite*>& writes = writesOwned.mutableVector(); - writeOp.setOpError( targetError ); - ++numTargetErrors; + Status targetStatus = writeOp.targetWrites(targeter, &writes); - if ( ordered ) - return Status::OK(); + if (!targetStatus.isOK()) { + WriteErrorDetail targetError; + buildTargetError(targetStatus, &targetError); - continue; - } - else { - dassert( ordered && !batchMap.empty() ); + if (!recordTargetErrors) { + // Cancel current batch state with an error - // Send out what we have, but don't record an error yet, since there may be an - // error in the writes before this point. + cancelBatches(targetError, _writeOps, &batchMap); + dassert(batchMap.empty()); + return targetStatus; + } else if (!ordered || batchMap.empty()) { + // Record an error for this batch - writeOp.cancelWrites( &targetError ); - break; - } - } + writeOp.setOpError(targetError); + ++numTargetErrors; - // - // If ordered and we have a previous endpoint, make sure we don't need to send these - // targeted writes to any other endpoints. - // + if (ordered) + return Status::OK(); - if ( ordered && !batchMap.empty() ) { + continue; + } else { + dassert(ordered && !batchMap.empty()); - dassert( batchMap.size() == 1u ); - if ( isNewBatchRequired( writes, batchMap ) ) { + // Send out what we have, but don't record an error yet, since there may be an + // error in the writes before this point. - writeOp.cancelWrites( NULL ); - break; - } + writeOp.cancelWrites(&targetError); + break; } + } - // - // If this write will push us over some sort of size limit, stop targeting - // + // + // If ordered and we have a previous endpoint, make sure we don't need to send these + // targeted writes to any other endpoints. + // - int writeSizeBytes = getWriteSizeBytes(writeOp); - if (wouldMakeBatchesTooBig(writes, writeSizeBytes, batchSizes)) { - invariant(!batchMap.empty()); + if (ordered && !batchMap.empty()) { + dassert(batchMap.size() == 1u); + if (isNewBatchRequired(writes, batchMap)) { writeOp.cancelWrites(NULL); break; } - - // - // Targeting went ok, add to appropriate TargetedBatch - // - - for ( vector<TargetedWrite*>::iterator it = writes.begin(); it != writes.end(); ++it ) { - - TargetedWrite* write = *it; - - TargetedBatchMap::iterator batchIt = batchMap.find( &write->endpoint ); - TargetedBatchSizeMap::iterator batchSizeIt = batchSizes.find( &write->endpoint ); - - if ( batchIt == batchMap.end() ) { - TargetedWriteBatch* newBatch = new TargetedWriteBatch( write->endpoint ); - batchIt = batchMap.insert( make_pair( &newBatch->getEndpoint(), - newBatch ) ).first; - batchSizeIt = batchSizes.insert(make_pair(&newBatch->getEndpoint(), - BatchSize())).first; - } - - TargetedWriteBatch* batch = batchIt->second; - BatchSize& batchSize = batchSizeIt->second; - - ++batchSize.numOps; - batchSize.sizeBytes += writeSizeBytes; - batch->addWrite( write ); - } - - // Relinquish ownership of TargetedWrites, now the TargetedBatches own them - writesOwned.mutableVector().clear(); - - // - // Break if we're ordered and we have more than one endpoint - later writes cannot be - // enforced as ordered across multiple shard endpoints. - // - - if ( ordered && batchMap.size() > 1u ) - break; } // - // Send back our targeted batches + // If this write will push us over some sort of size limit, stop targeting // - for ( TargetedBatchMap::iterator it = batchMap.begin(); it != batchMap.end(); ++it ) { - - TargetedWriteBatch* batch = it->second; - - if ( batch->getWrites().empty() ) - continue; - - // Remember targeted batch for reporting - _targeted.insert( batch ); - // Send the handle back to caller - targetedBatches->push_back( batch ); + int writeSizeBytes = getWriteSizeBytes(writeOp); + if (wouldMakeBatchesTooBig(writes, writeSizeBytes, batchSizes)) { + invariant(!batchMap.empty()); + writeOp.cancelWrites(NULL); + break; } - return Status::OK(); - } - - void BatchWriteOp::buildBatchRequest( const TargetedWriteBatch& targetedBatch, - BatchedCommandRequest* request ) const { - - request->setNS( _clientRequest->getNS() ); - request->setShouldBypassValidation(_clientRequest->shouldBypassValidation()); - - const vector<TargetedWrite*>& targetedWrites = targetedBatch.getWrites(); + // + // Targeting went ok, add to appropriate TargetedBatch + // - for ( vector<TargetedWrite*>::const_iterator it = targetedWrites.begin(); - it != targetedWrites.end(); ++it ) { + for (vector<TargetedWrite*>::iterator it = writes.begin(); it != writes.end(); ++it) { + TargetedWrite* write = *it; - const WriteOpRef& writeOpRef = ( *it )->writeOpRef; - BatchedCommandRequest::BatchType batchType = _clientRequest->getBatchType(); + TargetedBatchMap::iterator batchIt = batchMap.find(&write->endpoint); + TargetedBatchSizeMap::iterator batchSizeIt = batchSizes.find(&write->endpoint); - // NOTE: We copy the batch items themselves here from the client request - // TODO: This could be inefficient, maybe we want to just reference in the future - if ( batchType == BatchedCommandRequest::BatchType_Insert ) { - BatchedInsertRequest* clientInsertRequest = _clientRequest->getInsertRequest(); - BSONObj insertDoc = clientInsertRequest->getDocumentsAt( writeOpRef.first ); - request->getInsertRequest()->addToDocuments( insertDoc ); - } - else if ( batchType == BatchedCommandRequest::BatchType_Update ) { - BatchedUpdateRequest* clientUpdateRequest = _clientRequest->getUpdateRequest(); - BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; - clientUpdateRequest->getUpdatesAt( writeOpRef.first )->cloneTo( updateDoc ); - request->getUpdateRequest()->addToUpdates( updateDoc ); - } - else { - dassert( batchType == BatchedCommandRequest::BatchType_Delete ); - BatchedDeleteRequest* clientDeleteRequest = _clientRequest->getDeleteRequest(); - BatchedDeleteDocument* deleteDoc = new BatchedDeleteDocument; - clientDeleteRequest->getDeletesAt( writeOpRef.first )->cloneTo( deleteDoc ); - request->getDeleteRequest()->addToDeletes( deleteDoc ); + if (batchIt == batchMap.end()) { + TargetedWriteBatch* newBatch = new TargetedWriteBatch(write->endpoint); + batchIt = batchMap.insert(make_pair(&newBatch->getEndpoint(), newBatch)).first; + batchSizeIt = + batchSizes.insert(make_pair(&newBatch->getEndpoint(), BatchSize())).first; } - // TODO: We can add logic here to allow aborting individual ops - //if ( NULL == response ) { - // ->responses.erase( it++ ); - // continue; - //} - } + TargetedWriteBatch* batch = batchIt->second; + BatchSize& batchSize = batchSizeIt->second; - if ( _clientRequest->isWriteConcernSet() ) { - if ( _clientRequest->isVerboseWC() ) { - request->setWriteConcern( _clientRequest->getWriteConcern() ); - } - else { - // Mongos needs to send to the shard with w > 0 so it will be able to - // see the writeErrors. - request->setWriteConcern( upgradeWriteConcern( - _clientRequest->getWriteConcern() )); - } + ++batchSize.numOps; + batchSize.sizeBytes += writeSizeBytes; + batch->addWrite(write); } - if ( !request->isOrderedSet() ) { - request->setOrdered( _clientRequest->getOrdered() ); - } + // Relinquish ownership of TargetedWrites, now the TargetedBatches own them + writesOwned.mutableVector().clear(); + + // + // Break if we're ordered and we have more than one endpoint - later writes cannot be + // enforced as ordered across multiple shard endpoints. + // - unique_ptr<BatchedRequestMetadata> requestMetadata( new BatchedRequestMetadata() ); - requestMetadata->setShardName( targetedBatch.getEndpoint().shardName ); - requestMetadata->setShardVersion( targetedBatch.getEndpoint().shardVersion ); - requestMetadata->setSession( 0 ); - request->setMetadata( requestMetadata.release() ); + if (ordered && batchMap.size() > 1u) + break; } // - // Helpers for manipulating batch responses + // Send back our targeted batches // - namespace { - struct WriteErrorDetailComp { - bool operator()( const WriteErrorDetail* errorA, - const WriteErrorDetail* errorB ) const { - return errorA->getIndex() < errorB->getIndex(); - } - }; - } + for (TargetedBatchMap::iterator it = batchMap.begin(); it != batchMap.end(); ++it) { + TargetedWriteBatch* batch = it->second; + + if (batch->getWrites().empty()) + continue; - static void cloneCommandErrorTo( const BatchedCommandResponse& batchResp, - WriteErrorDetail* details ) { - details->setErrCode( batchResp.getErrCode() ); - details->setErrMessage( batchResp.getErrMessage() ); + // Remember targeted batch for reporting + _targeted.insert(batch); + // Send the handle back to caller + targetedBatches->push_back(batch); } - // Given *either* a batch error or an array of per-item errors, copies errors we're interested - // in into a TrackedErrorMap - static void trackErrors( const ShardEndpoint& endpoint, - const vector<WriteErrorDetail*> itemErrors, - TrackedErrors* trackedErrors ) { - for ( vector<WriteErrorDetail*>::const_iterator it = itemErrors.begin(); - it != itemErrors.end(); ++it ) { + return Status::OK(); +} - const WriteErrorDetail* error = *it; +void BatchWriteOp::buildBatchRequest(const TargetedWriteBatch& targetedBatch, + BatchedCommandRequest* request) const { + request->setNS(_clientRequest->getNS()); + request->setShouldBypassValidation(_clientRequest->shouldBypassValidation()); - if ( trackedErrors->isTracking( error->getErrCode() ) ) { - trackedErrors->addError( new ShardError( endpoint, *error ) ); - } - } - } + const vector<TargetedWrite*>& targetedWrites = targetedBatch.getWrites(); - static void incBatchStats( BatchedCommandRequest::BatchType batchType, - const BatchedCommandResponse& response, - BatchWriteStats* stats ) { + for (vector<TargetedWrite*>::const_iterator it = targetedWrites.begin(); + it != targetedWrites.end(); + ++it) { + const WriteOpRef& writeOpRef = (*it)->writeOpRef; + BatchedCommandRequest::BatchType batchType = _clientRequest->getBatchType(); - if ( batchType == BatchedCommandRequest::BatchType_Insert) { - stats->numInserted += response.getN(); - } - else if ( batchType == BatchedCommandRequest::BatchType_Update ) { - int numUpserted = 0; - if( response.isUpsertDetailsSet() ) { - numUpserted = response.sizeUpsertDetails(); - } - stats->numMatched += ( response.getN() - numUpserted ); - long long numModified = response.getNModified(); + // NOTE: We copy the batch items themselves here from the client request + // TODO: This could be inefficient, maybe we want to just reference in the future + if (batchType == BatchedCommandRequest::BatchType_Insert) { + BatchedInsertRequest* clientInsertRequest = _clientRequest->getInsertRequest(); + BSONObj insertDoc = clientInsertRequest->getDocumentsAt(writeOpRef.first); + request->getInsertRequest()->addToDocuments(insertDoc); + } else if (batchType == BatchedCommandRequest::BatchType_Update) { + BatchedUpdateRequest* clientUpdateRequest = _clientRequest->getUpdateRequest(); + BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; + clientUpdateRequest->getUpdatesAt(writeOpRef.first)->cloneTo(updateDoc); + request->getUpdateRequest()->addToUpdates(updateDoc); + } else { + dassert(batchType == BatchedCommandRequest::BatchType_Delete); + BatchedDeleteRequest* clientDeleteRequest = _clientRequest->getDeleteRequest(); + BatchedDeleteDocument* deleteDoc = new BatchedDeleteDocument; + clientDeleteRequest->getDeletesAt(writeOpRef.first)->cloneTo(deleteDoc); + request->getDeleteRequest()->addToDeletes(deleteDoc); + } + + // TODO: We can add logic here to allow aborting individual ops + // if ( NULL == response ) { + // ->responses.erase( it++ ); + // continue; + //} + } + + if (_clientRequest->isWriteConcernSet()) { + if (_clientRequest->isVerboseWC()) { + request->setWriteConcern(_clientRequest->getWriteConcern()); + } else { + // Mongos needs to send to the shard with w > 0 so it will be able to + // see the writeErrors. + request->setWriteConcern(upgradeWriteConcern(_clientRequest->getWriteConcern())); + } + } + + if (!request->isOrderedSet()) { + request->setOrdered(_clientRequest->getOrdered()); + } + + unique_ptr<BatchedRequestMetadata> requestMetadata(new BatchedRequestMetadata()); + requestMetadata->setShardName(targetedBatch.getEndpoint().shardName); + requestMetadata->setShardVersion(targetedBatch.getEndpoint().shardVersion); + requestMetadata->setSession(0); + request->setMetadata(requestMetadata.release()); +} - if (numModified >= 0) - stats->numModified += numModified; - else - stats->numModified = -1; // sentinel used to indicate we omit the field downstream +// +// Helpers for manipulating batch responses +// - stats->numUpserted += numUpserted; - } - else { - dassert( batchType == BatchedCommandRequest::BatchType_Delete ); - stats->numDeleted += response.getN(); - } +namespace { +struct WriteErrorDetailComp { + bool operator()(const WriteErrorDetail* errorA, const WriteErrorDetail* errorB) const { + return errorA->getIndex() < errorB->getIndex(); } +}; +} - void BatchWriteOp::noteBatchResponse( const TargetedWriteBatch& targetedBatch, - const BatchedCommandResponse& response, - TrackedErrors* trackedErrors ) { - - if ( !response.getOk() ) { +static void cloneCommandErrorTo(const BatchedCommandResponse& batchResp, + WriteErrorDetail* details) { + details->setErrCode(batchResp.getErrCode()); + details->setErrMessage(batchResp.getErrMessage()); +} - WriteErrorDetail error; - cloneCommandErrorTo( response, &error ); +// Given *either* a batch error or an array of per-item errors, copies errors we're interested +// in into a TrackedErrorMap +static void trackErrors(const ShardEndpoint& endpoint, + const vector<WriteErrorDetail*> itemErrors, + TrackedErrors* trackedErrors) { + for (vector<WriteErrorDetail*>::const_iterator it = itemErrors.begin(); it != itemErrors.end(); + ++it) { + const WriteErrorDetail* error = *it; - // Treat command errors exactly like other failures of the batch - // Note that no errors will be tracked from these failures - as-designed - noteBatchError( targetedBatch, error ); - return; + if (trackedErrors->isTracking(error->getErrCode())) { + trackedErrors->addError(new ShardError(endpoint, *error)); } + } +} - dassert( response.getOk() ); - - // Stop tracking targeted batch - _targeted.erase( &targetedBatch ); +static void incBatchStats(BatchedCommandRequest::BatchType batchType, + const BatchedCommandResponse& response, + BatchWriteStats* stats) { + if (batchType == BatchedCommandRequest::BatchType_Insert) { + stats->numInserted += response.getN(); + } else if (batchType == BatchedCommandRequest::BatchType_Update) { + int numUpserted = 0; + if (response.isUpsertDetailsSet()) { + numUpserted = response.sizeUpsertDetails(); + } + stats->numMatched += (response.getN() - numUpserted); + long long numModified = response.getNModified(); + + if (numModified >= 0) + stats->numModified += numModified; + else + stats->numModified = -1; // sentinel used to indicate we omit the field downstream + + stats->numUpserted += numUpserted; + } else { + dassert(batchType == BatchedCommandRequest::BatchType_Delete); + stats->numDeleted += response.getN(); + } +} - // Increment stats for this batch - incBatchStats( _clientRequest->getBatchType(), response, _stats.get() ); +void BatchWriteOp::noteBatchResponse(const TargetedWriteBatch& targetedBatch, + const BatchedCommandResponse& response, + TrackedErrors* trackedErrors) { + if (!response.getOk()) { + WriteErrorDetail error; + cloneCommandErrorTo(response, &error); - // - // Assign errors to particular items. - // Write Concern errors are stored and handled later. - // + // Treat command errors exactly like other failures of the batch + // Note that no errors will be tracked from these failures - as-designed + noteBatchError(targetedBatch, error); + return; + } - // Special handling for write concern errors, save for later - if ( response.isWriteConcernErrorSet() ) { - unique_ptr<ShardWCError> wcError( new ShardWCError( targetedBatch.getEndpoint(), - *response.getWriteConcernError() )); - _wcErrors.mutableVector().push_back( wcError.release() ); - } + dassert(response.getOk()); - vector<WriteErrorDetail*> itemErrors; + // Stop tracking targeted batch + _targeted.erase(&targetedBatch); - // Handle batch and per-item errors - if ( response.isErrDetailsSet() ) { + // Increment stats for this batch + incBatchStats(_clientRequest->getBatchType(), response, _stats.get()); - // Per-item errors were set - itemErrors.insert( itemErrors.begin(), - response.getErrDetails().begin(), - response.getErrDetails().end() ); + // + // Assign errors to particular items. + // Write Concern errors are stored and handled later. + // - // Sort per-item errors by index - std::sort( itemErrors.begin(), itemErrors.end(), WriteErrorDetailComp() ); - } + // Special handling for write concern errors, save for later + if (response.isWriteConcernErrorSet()) { + unique_ptr<ShardWCError> wcError( + new ShardWCError(targetedBatch.getEndpoint(), *response.getWriteConcernError())); + _wcErrors.mutableVector().push_back(wcError.release()); + } - // - // Go through all pending responses of the op and sorted remote reponses, populate errors - // This will either set all errors to the batch error or apply per-item errors as-needed - // - // If the batch is ordered, cancel all writes after the first error for retargeting. - // + vector<WriteErrorDetail*> itemErrors; - bool ordered = _clientRequest->getOrdered(); + // Handle batch and per-item errors + if (response.isErrDetailsSet()) { + // Per-item errors were set + itemErrors.insert( + itemErrors.begin(), response.getErrDetails().begin(), response.getErrDetails().end()); - vector<WriteErrorDetail*>::iterator itemErrorIt = itemErrors.begin(); - int index = 0; - WriteErrorDetail* lastError = NULL; - for ( vector<TargetedWrite*>::const_iterator it = targetedBatch.getWrites().begin(); - it != targetedBatch.getWrites().end(); ++it, ++index ) { + // Sort per-item errors by index + std::sort(itemErrors.begin(), itemErrors.end(), WriteErrorDetailComp()); + } - const TargetedWrite* write = *it; - WriteOp& writeOp = _writeOps[write->writeOpRef.first]; + // + // Go through all pending responses of the op and sorted remote reponses, populate errors + // This will either set all errors to the batch error or apply per-item errors as-needed + // + // If the batch is ordered, cancel all writes after the first error for retargeting. + // - dassert( writeOp.getWriteState() == WriteOpState_Pending ); + bool ordered = _clientRequest->getOrdered(); - // See if we have an error for the write - WriteErrorDetail* writeError = NULL; + vector<WriteErrorDetail*>::iterator itemErrorIt = itemErrors.begin(); + int index = 0; + WriteErrorDetail* lastError = NULL; + for (vector<TargetedWrite*>::const_iterator it = targetedBatch.getWrites().begin(); + it != targetedBatch.getWrites().end(); + ++it, ++index) { + const TargetedWrite* write = *it; + WriteOp& writeOp = _writeOps[write->writeOpRef.first]; - if ( itemErrorIt != itemErrors.end() && ( *itemErrorIt )->getIndex() == index ) { - // We have an per-item error for this write op's index - writeError = *itemErrorIt; - ++itemErrorIt; - } + dassert(writeOp.getWriteState() == WriteOpState_Pending); - // Finish the response (with error, if needed) - if ( NULL == writeError ) { - if ( !ordered || !lastError ){ - writeOp.noteWriteComplete( *write ); - } - else { - // We didn't actually apply this write - cancel so we can retarget - dassert( writeOp.getNumTargeted() == 1u ); - writeOp.cancelWrites( lastError ); - } - } - else { - writeOp.noteWriteError( *write, *writeError ); - lastError = writeError; - } - } + // See if we have an error for the write + WriteErrorDetail* writeError = NULL; - // Track errors we care about, whether batch or individual errors - if ( NULL != trackedErrors ) { - trackErrors( targetedBatch.getEndpoint(), itemErrors, trackedErrors ); + if (itemErrorIt != itemErrors.end() && (*itemErrorIt)->getIndex() == index) { + // We have an per-item error for this write op's index + writeError = *itemErrorIt; + ++itemErrorIt; } - // Track upserted ids if we need to - if ( response.isUpsertDetailsSet() ) { - - const vector<BatchedUpsertDetail*>& upsertedIds = response.getUpsertDetails(); - for ( vector<BatchedUpsertDetail*>::const_iterator it = upsertedIds.begin(); - it != upsertedIds.end(); ++it ) { - - // The child upserted details don't have the correct index for the full batch - const BatchedUpsertDetail* childUpsertedId = *it; - - // Work backward from the child batch item index to the batch item index - int childBatchIndex = childUpsertedId->getIndex(); - int batchIndex = targetedBatch.getWrites()[childBatchIndex]->writeOpRef.first; - - // Push the upserted id with the correct index into the batch upserted ids - BatchedUpsertDetail* upsertedId = new BatchedUpsertDetail; - upsertedId->setIndex( batchIndex ); - upsertedId->setUpsertedID( childUpsertedId->getUpsertedID() ); - _upsertedIds.mutableVector().push_back( upsertedId ); + // Finish the response (with error, if needed) + if (NULL == writeError) { + if (!ordered || !lastError) { + writeOp.noteWriteComplete(*write); + } else { + // We didn't actually apply this write - cancel so we can retarget + dassert(writeOp.getNumTargeted() == 1u); + writeOp.cancelWrites(lastError); } + } else { + writeOp.noteWriteError(*write, *writeError); + lastError = writeError; } } - static void toWriteErrorResponse( const WriteErrorDetail& error, - bool ordered, - int numWrites, - BatchedCommandResponse* writeErrResponse ) { + // Track errors we care about, whether batch or individual errors + if (NULL != trackedErrors) { + trackErrors(targetedBatch.getEndpoint(), itemErrors, trackedErrors); + } + + // Track upserted ids if we need to + if (response.isUpsertDetailsSet()) { + const vector<BatchedUpsertDetail*>& upsertedIds = response.getUpsertDetails(); + for (vector<BatchedUpsertDetail*>::const_iterator it = upsertedIds.begin(); + it != upsertedIds.end(); + ++it) { + // The child upserted details don't have the correct index for the full batch + const BatchedUpsertDetail* childUpsertedId = *it; - writeErrResponse->setOk( true ); - writeErrResponse->setN( 0 ); + // Work backward from the child batch item index to the batch item index + int childBatchIndex = childUpsertedId->getIndex(); + int batchIndex = targetedBatch.getWrites()[childBatchIndex]->writeOpRef.first; - int numErrors = ordered ? 1 : numWrites; - for ( int i = 0; i < numErrors; i++ ) { - unique_ptr<WriteErrorDetail> errorClone( new WriteErrorDetail ); - error.cloneTo( errorClone.get() ); - errorClone->setIndex( i ); - writeErrResponse->addToErrDetails( errorClone.release() ); + // Push the upserted id with the correct index into the batch upserted ids + BatchedUpsertDetail* upsertedId = new BatchedUpsertDetail; + upsertedId->setIndex(batchIndex); + upsertedId->setUpsertedID(childUpsertedId->getUpsertedID()); + _upsertedIds.mutableVector().push_back(upsertedId); } - - dassert( writeErrResponse->isValid( NULL ) ); } +} - void BatchWriteOp::noteBatchError( const TargetedWriteBatch& targetedBatch, - const WriteErrorDetail& error ) { - - // Treat errors to get a batch response as failures of the contained writes - BatchedCommandResponse emulatedResponse; - toWriteErrorResponse( error, - _clientRequest->getOrdered(), - targetedBatch.getWrites().size(), - &emulatedResponse ); +static void toWriteErrorResponse(const WriteErrorDetail& error, + bool ordered, + int numWrites, + BatchedCommandResponse* writeErrResponse) { + writeErrResponse->setOk(true); + writeErrResponse->setN(0); - noteBatchResponse( targetedBatch, emulatedResponse, NULL ); + int numErrors = ordered ? 1 : numWrites; + for (int i = 0; i < numErrors; i++) { + unique_ptr<WriteErrorDetail> errorClone(new WriteErrorDetail); + error.cloneTo(errorClone.get()); + errorClone->setIndex(i); + writeErrResponse->addToErrDetails(errorClone.release()); } - void BatchWriteOp::abortBatch( const WriteErrorDetail& error ) { + dassert(writeErrResponse->isValid(NULL)); +} - dassert( !isFinished() ); - dassert( numWriteOpsIn( WriteOpState_Pending ) == 0 ); +void BatchWriteOp::noteBatchError(const TargetedWriteBatch& targetedBatch, + const WriteErrorDetail& error) { + // Treat errors to get a batch response as failures of the contained writes + BatchedCommandResponse emulatedResponse; + toWriteErrorResponse( + error, _clientRequest->getOrdered(), targetedBatch.getWrites().size(), &emulatedResponse); - size_t numWriteOps = _clientRequest->sizeWriteOps(); - bool orderedOps = _clientRequest->getOrdered(); - for ( size_t i = 0; i < numWriteOps; ++i ) { + noteBatchResponse(targetedBatch, emulatedResponse, NULL); +} - WriteOp& writeOp = _writeOps[i]; - // Can only be called with no outstanding batches - dassert( writeOp.getWriteState() != WriteOpState_Pending ); +void BatchWriteOp::abortBatch(const WriteErrorDetail& error) { + dassert(!isFinished()); + dassert(numWriteOpsIn(WriteOpState_Pending) == 0); - if ( writeOp.getWriteState() < WriteOpState_Completed ) { + size_t numWriteOps = _clientRequest->sizeWriteOps(); + bool orderedOps = _clientRequest->getOrdered(); + for (size_t i = 0; i < numWriteOps; ++i) { + WriteOp& writeOp = _writeOps[i]; + // Can only be called with no outstanding batches + dassert(writeOp.getWriteState() != WriteOpState_Pending); - writeOp.setOpError( error ); + if (writeOp.getWriteState() < WriteOpState_Completed) { + writeOp.setOpError(error); - // Only one error if we're ordered - if ( orderedOps ) break; - } + // Only one error if we're ordered + if (orderedOps) + break; } - - dassert( isFinished() ); } - bool BatchWriteOp::isFinished() { - - size_t numWriteOps = _clientRequest->sizeWriteOps(); - bool orderedOps = _clientRequest->getOrdered(); - for ( size_t i = 0; i < numWriteOps; ++i ) { - WriteOp& writeOp = _writeOps[i]; - if ( writeOp.getWriteState() < WriteOpState_Completed ) return false; - else if ( orderedOps && writeOp.getWriteState() == WriteOpState_Error ) return true; - } + dassert(isFinished()); +} - return true; +bool BatchWriteOp::isFinished() { + size_t numWriteOps = _clientRequest->sizeWriteOps(); + bool orderedOps = _clientRequest->getOrdered(); + for (size_t i = 0; i < numWriteOps; ++i) { + WriteOp& writeOp = _writeOps[i]; + if (writeOp.getWriteState() < WriteOpState_Completed) + return false; + else if (orderedOps && writeOp.getWriteState() == WriteOpState_Error) + return true; } - // - // Aggregation functions for building the final response errors - // - - void BatchWriteOp::buildClientResponse( BatchedCommandResponse* batchResp ) { + return true; +} - dassert( isFinished() ); +// +// Aggregation functions for building the final response errors +// - // Result is OK - batchResp->setOk( true ); +void BatchWriteOp::buildClientResponse(BatchedCommandResponse* batchResp) { + dassert(isFinished()); - // For non-verbose, it's all we need. - if ( !_clientRequest->isVerboseWC() ) { - dassert( batchResp->isValid( NULL ) ); - return; - } + // Result is OK + batchResp->setOk(true); - // - // Find all the errors in the batch - // + // For non-verbose, it's all we need. + if (!_clientRequest->isVerboseWC()) { + dassert(batchResp->isValid(NULL)); + return; + } - vector<WriteOp*> errOps; + // + // Find all the errors in the batch + // - size_t numWriteOps = _clientRequest->sizeWriteOps(); - for ( size_t i = 0; i < numWriteOps; ++i ) { + vector<WriteOp*> errOps; - WriteOp& writeOp = _writeOps[i]; + size_t numWriteOps = _clientRequest->sizeWriteOps(); + for (size_t i = 0; i < numWriteOps; ++i) { + WriteOp& writeOp = _writeOps[i]; - if ( writeOp.getWriteState() == WriteOpState_Error ) { - errOps.push_back( &writeOp ); - } + if (writeOp.getWriteState() == WriteOpState_Error) { + errOps.push_back(&writeOp); } + } - // - // Build the per-item errors. - // - - if ( !errOps.empty() ) { - for ( vector<WriteOp*>::iterator it = errOps.begin(); it != errOps.end(); ++it ) { - WriteOp& writeOp = **it; - WriteErrorDetail* error = new WriteErrorDetail(); - writeOp.getOpError().cloneTo( error ); - batchResp->addToErrDetails( error ); - } - } - - // Only return a write concern error if everything succeeded (unordered or ordered) - // OR if something succeeded and we're unordered - bool reportWCError = errOps.empty() - || ( !_clientRequest->getOrdered() - && errOps.size() < _clientRequest->sizeWriteOps() ); - if ( !_wcErrors.empty() && reportWCError ) { - - WCErrorDetail* error = new WCErrorDetail; - - // Generate the multi-error message below - stringstream msg; - if ( _wcErrors.size() > 1 ) { - msg << "multiple errors reported : "; - error->setErrCode( ErrorCodes::WriteConcernFailed ); - } - else { - error->setErrCode( ( *_wcErrors.begin() )->error.getErrCode() ); - } - - for ( vector<ShardWCError*>::const_iterator it = _wcErrors.begin(); - it != _wcErrors.end(); ++it ) { - const ShardWCError* wcError = *it; - if ( it != _wcErrors.begin() ) - msg << " :: and :: "; - msg << wcError->error.getErrMessage() << " at " << wcError->endpoint.shardName; - } + // + // Build the per-item errors. + // - error->setErrMessage( msg.str() ); - batchResp->setWriteConcernError( error ); + if (!errOps.empty()) { + for (vector<WriteOp*>::iterator it = errOps.begin(); it != errOps.end(); ++it) { + WriteOp& writeOp = **it; + WriteErrorDetail* error = new WriteErrorDetail(); + writeOp.getOpError().cloneTo(error); + batchResp->addToErrDetails(error); } + } - // - // Append the upserted ids, if required - // + // Only return a write concern error if everything succeeded (unordered or ordered) + // OR if something succeeded and we're unordered + bool reportWCError = errOps.empty() || + (!_clientRequest->getOrdered() && errOps.size() < _clientRequest->sizeWriteOps()); + if (!_wcErrors.empty() && reportWCError) { + WCErrorDetail* error = new WCErrorDetail; - if ( _upsertedIds.size() != 0 ) { - batchResp->setUpsertDetails( _upsertedIds.vector() ); + // Generate the multi-error message below + stringstream msg; + if (_wcErrors.size() > 1) { + msg << "multiple errors reported : "; + error->setErrCode(ErrorCodes::WriteConcernFailed); + } else { + error->setErrCode((*_wcErrors.begin())->error.getErrCode()); } - // Stats - int nValue = _stats->numInserted + _stats->numUpserted + _stats->numMatched - + _stats->numDeleted; - batchResp->setN( nValue ); - if ( _clientRequest->getBatchType() == BatchedCommandRequest::BatchType_Update && - _stats->numModified >= 0) { - batchResp->setNModified( _stats->numModified ); + for (vector<ShardWCError*>::const_iterator it = _wcErrors.begin(); it != _wcErrors.end(); + ++it) { + const ShardWCError* wcError = *it; + if (it != _wcErrors.begin()) + msg << " :: and :: "; + msg << wcError->error.getErrMessage() << " at " << wcError->endpoint.shardName; } - dassert( batchResp->isValid( NULL ) ); + error->setErrMessage(msg.str()); + batchResp->setWriteConcernError(error); } - BatchWriteOp::~BatchWriteOp() { - // Caller's responsibility to dispose of TargetedBatches - dassert( _targeted.empty() ); - - if ( NULL != _writeOps ) { - - size_t numWriteOps = _clientRequest->sizeWriteOps(); - for ( size_t i = 0; i < numWriteOps; ++i ) { - // Placement new so manual destruct - _writeOps[i].~WriteOp(); - } + // + // Append the upserted ids, if required + // - ::operator delete[]( _writeOps ); - _writeOps = NULL; - } + if (_upsertedIds.size() != 0) { + batchResp->setUpsertDetails(_upsertedIds.vector()); } - int BatchWriteOp::numWriteOps() const { - return static_cast<int>( _clientRequest->sizeWriteOps() ); + // Stats + int nValue = + _stats->numInserted + _stats->numUpserted + _stats->numMatched + _stats->numDeleted; + batchResp->setN(nValue); + if (_clientRequest->getBatchType() == BatchedCommandRequest::BatchType_Update && + _stats->numModified >= 0) { + batchResp->setNModified(_stats->numModified); } - int BatchWriteOp::numWriteOpsIn( WriteOpState opState ) const { + dassert(batchResp->isValid(NULL)); +} + +BatchWriteOp::~BatchWriteOp() { + // Caller's responsibility to dispose of TargetedBatches + dassert(_targeted.empty()); - // TODO: This could be faster, if we tracked this info explicitly + if (NULL != _writeOps) { size_t numWriteOps = _clientRequest->sizeWriteOps(); - int count = 0; - for ( size_t i = 0; i < numWriteOps; ++i ) { - WriteOp& writeOp = _writeOps[i]; - if ( writeOp.getWriteState() == opState ) - ++count; + for (size_t i = 0; i < numWriteOps; ++i) { + // Placement new so manual destruct + _writeOps[i].~WriteOp(); } - return count; + ::operator delete[](_writeOps); + _writeOps = NULL; } +} - void TrackedErrors::startTracking( int errCode ) { - dassert( !isTracking( errCode ) ); - _errorMap.insert( make_pair( errCode, vector<ShardError*>() ) ); - } +int BatchWriteOp::numWriteOps() const { + return static_cast<int>(_clientRequest->sizeWriteOps()); +} - bool TrackedErrors::isTracking( int errCode ) const { - return _errorMap.find( errCode ) != _errorMap.end(); +int BatchWriteOp::numWriteOpsIn(WriteOpState opState) const { + // TODO: This could be faster, if we tracked this info explicitly + size_t numWriteOps = _clientRequest->sizeWriteOps(); + int count = 0; + for (size_t i = 0; i < numWriteOps; ++i) { + WriteOp& writeOp = _writeOps[i]; + if (writeOp.getWriteState() == opState) + ++count; } - void TrackedErrors::addError( ShardError* error ) { - TrackedErrorMap::iterator seenIt = _errorMap.find( error->error.getErrCode() ); - if ( seenIt == _errorMap.end() ) return; - seenIt->second.push_back( error ); - } + return count; +} - const vector<ShardError*>& TrackedErrors::getErrors( int errCode ) const { - dassert( isTracking( errCode ) ); - return _errorMap.find( errCode )->second; - } +void TrackedErrors::startTracking(int errCode) { + dassert(!isTracking(errCode)); + _errorMap.insert(make_pair(errCode, vector<ShardError*>())); +} - void TrackedErrors::clear() { - for ( TrackedErrorMap::iterator it = _errorMap.begin(); it != _errorMap.end(); ++it ) { +bool TrackedErrors::isTracking(int errCode) const { + return _errorMap.find(errCode) != _errorMap.end(); +} - vector<ShardError*>& errors = it->second; +void TrackedErrors::addError(ShardError* error) { + TrackedErrorMap::iterator seenIt = _errorMap.find(error->error.getErrCode()); + if (seenIt == _errorMap.end()) + return; + seenIt->second.push_back(error); +} - for ( vector<ShardError*>::iterator errIt = errors.begin(); errIt != errors.end(); - ++errIt ) { - delete *errIt; - } - errors.clear(); - } - } +const vector<ShardError*>& TrackedErrors::getErrors(int errCode) const { + dassert(isTracking(errCode)); + return _errorMap.find(errCode)->second; +} + +void TrackedErrors::clear() { + for (TrackedErrorMap::iterator it = _errorMap.begin(); it != _errorMap.end(); ++it) { + vector<ShardError*>& errors = it->second; - TrackedErrors::~TrackedErrors() { - clear(); + for (vector<ShardError*>::iterator errIt = errors.begin(); errIt != errors.end(); ++errIt) { + delete *errIt; + } + errors.clear(); } +} +TrackedErrors::~TrackedErrors() { + clear(); +} } diff --git a/src/mongo/s/write_ops/batch_write_op.h b/src/mongo/s/write_ops/batch_write_op.h index 224d9985ef3..0add9339268 100644 --- a/src/mongo/s/write_ops/batch_write_op.h +++ b/src/mongo/s/write_ops/batch_write_op.h @@ -44,260 +44,244 @@ namespace mongo { - class TargetedWriteBatch; - struct ShardError; - struct ShardWCError; - class TrackedErrors; - struct BatchWriteStats; +class TargetedWriteBatch; +struct ShardError; +struct ShardWCError; +class TrackedErrors; +struct BatchWriteStats; + +/** + * The BatchWriteOp class manages the lifecycle of a batched write received by mongos. Each + * item in a batch is tracked via a WriteOp, and the function of the BatchWriteOp is to + * aggregate the dispatched requests and responses for the underlying WriteOps. + * + * Overall, the BatchWriteOp lifecycle is similar to the WriteOp lifecycle, with the following + * stages: + * + * 0) Client request comes in, batch write op is initialized + * + * 1a) One or more ops in the batch are targeted using targetBatch, resulting in + * TargetedWriteBatches for these ops. + * 1b) There are targeting errors, and the batch must be retargeted after refreshing the + * NSTargeter. + * + * 2) (Child BatchCommandRequests are be built for each TargetedWriteBatch before sending) + * + * 3) Responses for sent TargetedWriteBatches are noted, errors are stored and aggregated per- + * write-op. Errors the caller is interested in are returned. + * + * 4) If the batch write is not finished, goto 0 + * + * 5) When all responses come back for all write ops, errors are aggregated and returned in + * a client response + * + */ +class BatchWriteOp { + MONGO_DISALLOW_COPYING(BatchWriteOp); + +public: + BatchWriteOp(); + + ~BatchWriteOp(); /** - * The BatchWriteOp class manages the lifecycle of a batched write received by mongos. Each - * item in a batch is tracked via a WriteOp, and the function of the BatchWriteOp is to - * aggregate the dispatched requests and responses for the underlying WriteOps. - * - * Overall, the BatchWriteOp lifecycle is similar to the WriteOp lifecycle, with the following - * stages: - * - * 0) Client request comes in, batch write op is initialized - * - * 1a) One or more ops in the batch are targeted using targetBatch, resulting in - * TargetedWriteBatches for these ops. - * 1b) There are targeting errors, and the batch must be retargeted after refreshing the - * NSTargeter. + * Initializes the BatchWriteOp from a client batch request. + */ + void initClientRequest(const BatchedCommandRequest* clientRequest); + + /** + * Targets one or more of the next write ops in this batch op using a NSTargeter. The + * resulting TargetedWrites are aggregated together in the returned TargetedWriteBatches. * - * 2) (Child BatchCommandRequests are be built for each TargetedWriteBatch before sending) + * If 'recordTargetErrors' is false, any targeting error will abort all current batches and + * the method will return the targeting error. No targetedBatches will be returned on + * error. * - * 3) Responses for sent TargetedWriteBatches are noted, errors are stored and aggregated per- - * write-op. Errors the caller is interested in are returned. + * Otherwise, if 'recordTargetErrors' is true, targeting errors will be recorded for each + * write op that fails to target, and the method will return OK. * - * 4) If the batch write is not finished, goto 0 + * (The idea here is that if we are sure our NSTargeter is up-to-date we should record + * targeting errors, but if not we should refresh once first.) * - * 5) When all responses come back for all write ops, errors are aggregated and returned in - * a client response + * Returned TargetedWriteBatches are owned by the caller. + */ + Status targetBatch(const NSTargeter& targeter, + bool recordTargetErrors, + std::vector<TargetedWriteBatch*>* targetedBatches); + + /** + * Fills a BatchCommandRequest from a TargetedWriteBatch for this BatchWriteOp. + */ + void buildBatchRequest(const TargetedWriteBatch& targetedBatch, + BatchedCommandRequest* request) const; + + /** + * Stores a response from one of the outstanding TargetedWriteBatches for this BatchWriteOp. + * The response may be in any form, error or not. * + * There is an additional optional 'trackedErrors' parameter, which can be used to return + * copies of any write errors in the response that the caller is interested in (specified by + * errCode). (This avoids external callers needing to know much about the response format.) */ - class BatchWriteOp { - MONGO_DISALLOW_COPYING(BatchWriteOp); - public: - - BatchWriteOp(); - - ~BatchWriteOp(); - - /** - * Initializes the BatchWriteOp from a client batch request. - */ - void initClientRequest( const BatchedCommandRequest* clientRequest ); - - /** - * Targets one or more of the next write ops in this batch op using a NSTargeter. The - * resulting TargetedWrites are aggregated together in the returned TargetedWriteBatches. - * - * If 'recordTargetErrors' is false, any targeting error will abort all current batches and - * the method will return the targeting error. No targetedBatches will be returned on - * error. - * - * Otherwise, if 'recordTargetErrors' is true, targeting errors will be recorded for each - * write op that fails to target, and the method will return OK. - * - * (The idea here is that if we are sure our NSTargeter is up-to-date we should record - * targeting errors, but if not we should refresh once first.) - * - * Returned TargetedWriteBatches are owned by the caller. - */ - Status targetBatch( const NSTargeter& targeter, - bool recordTargetErrors, - std::vector<TargetedWriteBatch*>* targetedBatches ); - - /** - * Fills a BatchCommandRequest from a TargetedWriteBatch for this BatchWriteOp. - */ - void buildBatchRequest( const TargetedWriteBatch& targetedBatch, - BatchedCommandRequest* request ) const; - - /** - * Stores a response from one of the outstanding TargetedWriteBatches for this BatchWriteOp. - * The response may be in any form, error or not. - * - * There is an additional optional 'trackedErrors' parameter, which can be used to return - * copies of any write errors in the response that the caller is interested in (specified by - * errCode). (This avoids external callers needing to know much about the response format.) - */ - void noteBatchResponse( const TargetedWriteBatch& targetedBatch, - const BatchedCommandResponse& response, - TrackedErrors* trackedErrors ); - - /** - * Stores an error that occurred trying to send/recv a TargetedWriteBatch for this - * BatchWriteOp. - */ - void noteBatchError( const TargetedWriteBatch& targetedBatch, - const WriteErrorDetail& error ); - - /** - * Aborts any further writes in the batch with the provided error. There must be no pending - * ops awaiting results when a batch is aborted. - * - * Batch is finished immediately after aborting. - */ - void abortBatch( const WriteErrorDetail& error ); - - /** - * Returns false if the batch write op needs more processing. - */ - bool isFinished(); - - /** - * Fills a batch response to send back to the client. - */ - void buildClientResponse( BatchedCommandResponse* batchResp ); - - // - // Accessors - // - - int numWriteOps() const; - - int numWriteOpsIn( WriteOpState state ) const; - - private: - - // Incoming client request, not owned here - const BatchedCommandRequest* _clientRequest; - - // Array of ops being processed from the client request - WriteOp* _writeOps; - - // Current outstanding batch op write requests - // Not owned here but tracked for reporting - std::set<const TargetedWriteBatch*> _targeted; - - // Write concern responses from all write batches so far - OwnedPointerVector<ShardWCError> _wcErrors; - - // Upserted ids for the whole write batch - OwnedPointerVector<BatchedUpsertDetail> _upsertedIds; - - // Stats for the entire batch op - std::unique_ptr<BatchWriteStats> _stats; - }; - - struct BatchWriteStats { - - BatchWriteStats(); - - int numInserted; - int numUpserted; - int numMatched; - int numModified; - int numDeleted; - - std::string toString() const { - StringBuilder str; - str << "numInserted: " << numInserted - << " numUpserted: " << numUpserted - << " numMatched: " << numMatched - << " numModified: " << numModified - << " numDeleted: " << numDeleted; - return str.str(); - } - - }; + void noteBatchResponse(const TargetedWriteBatch& targetedBatch, + const BatchedCommandResponse& response, + TrackedErrors* trackedErrors); /** - * Data structure representing the information needed to make a batch request, along with - * pointers to where the resulting responses should be placed. + * Stores an error that occurred trying to send/recv a TargetedWriteBatch for this + * BatchWriteOp. + */ + void noteBatchError(const TargetedWriteBatch& targetedBatch, const WriteErrorDetail& error); + + /** + * Aborts any further writes in the batch with the provided error. There must be no pending + * ops awaiting results when a batch is aborted. * - * Internal support for storage as a doubly-linked list, to allow the TargetedWriteBatch to - * efficiently be registered for reporting. + * Batch is finished immediately after aborting. */ - class TargetedWriteBatch { - MONGO_DISALLOW_COPYING(TargetedWriteBatch); - public: + void abortBatch(const WriteErrorDetail& error); - TargetedWriteBatch( const ShardEndpoint& endpoint ) : - _endpoint( endpoint ) { - } + /** + * Returns false if the batch write op needs more processing. + */ + bool isFinished(); - const ShardEndpoint& getEndpoint() const { - return _endpoint; - } + /** + * Fills a batch response to send back to the client. + */ + void buildClientResponse(BatchedCommandResponse* batchResp); - /** - * TargetedWrite is owned here once given to the TargetedWriteBatch - */ - void addWrite( TargetedWrite* targetedWrite ) { - _writes.mutableVector().push_back( targetedWrite ); - } + // + // Accessors + // - const std::vector<TargetedWrite*>& getWrites() const { - return _writes.vector(); - } + int numWriteOps() const; - private: + int numWriteOpsIn(WriteOpState state) const; - // Where to send the batch - const ShardEndpoint _endpoint; +private: + // Incoming client request, not owned here + const BatchedCommandRequest* _clientRequest; - // Where the responses go - // TargetedWrite*s are owned by the TargetedWriteBatch - OwnedPointerVector<TargetedWrite> _writes; - }; + // Array of ops being processed from the client request + WriteOp* _writeOps; - /** - * Simple struct for storing an error with an endpoint. - * - * Certain types of errors are not stored in WriteOps or must be returned to a caller. - */ - struct ShardError { + // Current outstanding batch op write requests + // Not owned here but tracked for reporting + std::set<const TargetedWriteBatch*> _targeted; - ShardError( const ShardEndpoint& endpoint, const WriteErrorDetail& error ) : - endpoint( endpoint ) { - error.cloneTo( &this->error ); - } + // Write concern responses from all write batches so far + OwnedPointerVector<ShardWCError> _wcErrors; - const ShardEndpoint endpoint; - WriteErrorDetail error; - }; + // Upserted ids for the whole write batch + OwnedPointerVector<BatchedUpsertDetail> _upsertedIds; - /** - * Simple struct for storing a write concern error with an endpoint. - * - * Certain types of errors are not stored in WriteOps or must be returned to a caller. - */ - struct ShardWCError { + // Stats for the entire batch op + std::unique_ptr<BatchWriteStats> _stats; +}; + +struct BatchWriteStats { + BatchWriteStats(); + + int numInserted; + int numUpserted; + int numMatched; + int numModified; + int numDeleted; - ShardWCError( const ShardEndpoint& endpoint, const WCErrorDetail& error ) : - endpoint( endpoint ) { - error.cloneTo( &this->error ); - } + std::string toString() const { + StringBuilder str; + str << "numInserted: " << numInserted << " numUpserted: " << numUpserted + << " numMatched: " << numMatched << " numModified: " << numModified + << " numDeleted: " << numDeleted; + return str.str(); + } +}; - const ShardEndpoint endpoint; - WCErrorDetail error; - }; +/** + * Data structure representing the information needed to make a batch request, along with + * pointers to where the resulting responses should be placed. + * + * Internal support for storage as a doubly-linked list, to allow the TargetedWriteBatch to + * efficiently be registered for reporting. + */ +class TargetedWriteBatch { + MONGO_DISALLOW_COPYING(TargetedWriteBatch); + +public: + TargetedWriteBatch(const ShardEndpoint& endpoint) : _endpoint(endpoint) {} + + const ShardEndpoint& getEndpoint() const { + return _endpoint; + } /** - * Helper class for tracking certain errors from batch operations + * TargetedWrite is owned here once given to the TargetedWriteBatch */ - class TrackedErrors { - public: + void addWrite(TargetedWrite* targetedWrite) { + _writes.mutableVector().push_back(targetedWrite); + } + + const std::vector<TargetedWrite*>& getWrites() const { + return _writes.vector(); + } + +private: + // Where to send the batch + const ShardEndpoint _endpoint; - ~TrackedErrors(); + // Where the responses go + // TargetedWrite*s are owned by the TargetedWriteBatch + OwnedPointerVector<TargetedWrite> _writes; +}; + +/** + * Simple struct for storing an error with an endpoint. + * + * Certain types of errors are not stored in WriteOps or must be returned to a caller. + */ +struct ShardError { + ShardError(const ShardEndpoint& endpoint, const WriteErrorDetail& error) : endpoint(endpoint) { + error.cloneTo(&this->error); + } - void startTracking( int errCode ); + const ShardEndpoint endpoint; + WriteErrorDetail error; +}; - bool isTracking( int errCode ) const; +/** + * Simple struct for storing a write concern error with an endpoint. + * + * Certain types of errors are not stored in WriteOps or must be returned to a caller. + */ +struct ShardWCError { + ShardWCError(const ShardEndpoint& endpoint, const WCErrorDetail& error) : endpoint(endpoint) { + error.cloneTo(&this->error); + } + + const ShardEndpoint endpoint; + WCErrorDetail error; +}; + +/** + * Helper class for tracking certain errors from batch operations + */ +class TrackedErrors { +public: + ~TrackedErrors(); - void addError( ShardError* error ); + void startTracking(int errCode); - const std::vector<ShardError*>& getErrors( int errCode ) const; + bool isTracking(int errCode) const; - void clear(); + void addError(ShardError* error); - private: + const std::vector<ShardError*>& getErrors(int errCode) const; - typedef unordered_map<int, std::vector<ShardError*> > TrackedErrorMap; - TrackedErrorMap _errorMap; - }; + void clear(); +private: + typedef unordered_map<int, std::vector<ShardError*>> TrackedErrorMap; + TrackedErrorMap _errorMap; +}; } diff --git a/src/mongo/s/write_ops/batch_write_op_test.cpp b/src/mongo/s/write_ops/batch_write_op_test.cpp index 6a00115531b..bb0e923cfb1 100644 --- a/src/mongo/s/write_ops/batch_write_op_test.cpp +++ b/src/mongo/s/write_ops/batch_write_op_test.cpp @@ -37,1801 +37,1747 @@ namespace { - using std::unique_ptr; - using std::string; - using std::vector; - - using namespace mongo; - - static void initTargeterFullRange( const NamespaceString& nss, - const ShardEndpoint& endpoint, - MockNSTargeter* targeter ) { - - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpoint, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << MAXKEY ) ) ); - targeter->init( mockRanges ); - } - - static void initTargeterSplitRange( const NamespaceString& nss, - const ShardEndpoint& endpointA, - const ShardEndpoint& endpointB, - MockNSTargeter* targeter ) { - - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpointA, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << 0 ) ) ); - mockRanges.push_back( new MockRange( endpointB, - nss, - BSON( "x" << 0 ), - BSON( "x" << MAXKEY ) ) ); - targeter->init( mockRanges ); - } - - static void initTargeterHalfRange( const NamespaceString& nss, - const ShardEndpoint& endpoint, - MockNSTargeter* targeter ) { - - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpoint, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << 0 ) ) ); - - // x >= 0 values untargetable - - targeter->init( mockRanges ); - } - - static BatchedDeleteDocument* buildDelete( const BSONObj& query, int limit ) { - BatchedDeleteDocument* deleteDoc = new BatchedDeleteDocument; - deleteDoc->setQuery( query ); - deleteDoc->setLimit( limit ); - return deleteDoc; - } +using std::unique_ptr; +using std::string; +using std::vector; + +using namespace mongo; + +static void initTargeterFullRange(const NamespaceString& nss, + const ShardEndpoint& endpoint, + MockNSTargeter* targeter) { + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpoint, nss, BSON("x" << MINKEY), BSON("x" << MAXKEY))); + targeter->init(mockRanges); +} + +static void initTargeterSplitRange(const NamespaceString& nss, + const ShardEndpoint& endpointA, + const ShardEndpoint& endpointB, + MockNSTargeter* targeter) { + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpointA, nss, BSON("x" << MINKEY), BSON("x" << 0))); + mockRanges.push_back(new MockRange(endpointB, nss, BSON("x" << 0), BSON("x" << MAXKEY))); + targeter->init(mockRanges); +} + +static void initTargeterHalfRange(const NamespaceString& nss, + const ShardEndpoint& endpoint, + MockNSTargeter* targeter) { + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpoint, nss, BSON("x" << MINKEY), BSON("x" << 0))); + + // x >= 0 values untargetable + + targeter->init(mockRanges); +} + +static BatchedDeleteDocument* buildDelete(const BSONObj& query, int limit) { + BatchedDeleteDocument* deleteDoc = new BatchedDeleteDocument; + deleteDoc->setQuery(query); + deleteDoc->setLimit(limit); + return deleteDoc; +} + +static BatchedUpdateDocument* buildUpdate(const BSONObj& query, bool multi) { + BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; + updateDoc->setUpdateExpr(BSONObj()); + updateDoc->setQuery(query); + updateDoc->setMulti(multi); + return updateDoc; +} + +static BatchedUpdateDocument* buildUpdate(const BSONObj& query, + const BSONObj& updateExpr, + bool multi) { + BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; + updateDoc->setQuery(query); + updateDoc->setUpdateExpr(updateExpr); + updateDoc->setMulti(multi); + return updateDoc; +} + +static void buildResponse(int n, BatchedCommandResponse* response) { + response->clear(); + response->setOk(true); + response->setN(n); + ASSERT(response->isValid(NULL)); +} + +static void buildErrResponse(int code, const string& message, BatchedCommandResponse* response) { + response->clear(); + response->setOk(false); + response->setN(0); + response->setErrCode(code); + response->setErrMessage(message); + ASSERT(response->isValid(NULL)); +} + +static void addError(int code, const string& message, int index, BatchedCommandResponse* response) { + unique_ptr<WriteErrorDetail> error(new WriteErrorDetail); + error->setErrCode(code); + error->setErrMessage(message); + error->setIndex(index); + + response->addToErrDetails(error.release()); +} + +static void addWCError(BatchedCommandResponse* response) { + unique_ptr<WCErrorDetail> error(new WCErrorDetail); + error->setErrCode(ErrorCodes::WriteConcernFailed); + error->setErrMessage("mock wc error"); + + response->setWriteConcernError(error.release()); +} + +TEST(WriteOpTests, SingleOp) { + // + // Single-op targeting test + // - static BatchedUpdateDocument* buildUpdate( const BSONObj& query, bool multi ) { - BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; - updateDoc->setUpdateExpr( BSONObj() ); - updateDoc->setQuery( query ); - updateDoc->setMulti( multi ); - return updateDoc; - } + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); - static BatchedUpdateDocument* buildUpdate(const BSONObj& query, - const BSONObj& updateExpr, - bool multi) { - BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; - updateDoc->setQuery( query ); - updateDoc->setUpdateExpr( updateExpr ); - updateDoc->setMulti( multi ); - return updateDoc; - } + // Do single-target, single doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - static void buildResponse( int n, BatchedCommandResponse* response ) { - response->clear(); - response->setOk( true ); - response->setN( n ); - ASSERT( response->isValid( NULL ) ); - } + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); - static void buildErrResponse( int code, - const string& message, - BatchedCommandResponse* response ) { - response->clear(); - response->setOk( false ); - response->setN( 0 ); - response->setErrCode( code ); - response->setErrMessage( message ); - ASSERT( response->isValid( NULL ) ); - } + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); - static void addError( int code, - const string& message, - int index, - BatchedCommandResponse* response ) { + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpoint); - unique_ptr<WriteErrorDetail> error( new WriteErrorDetail ); - error->setErrCode( code ); - error->setErrMessage( message ); - error->setIndex( index ); + BatchedCommandResponse response; + buildResponse(1, &response); - response->addToErrDetails( error.release() ); - } + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); - static void addWCError( BatchedCommandResponse* response ) { + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); +} - unique_ptr<WCErrorDetail> error( new WCErrorDetail ); - error->setErrCode( ErrorCodes::WriteConcernFailed ); - error->setErrMessage( "mock wc error" ); +TEST(WriteOpTests, SingleError) { + // + // Single-op error test + // - response->setWriteConcernError( error.release() ); - } + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); - TEST(WriteOpTests, SingleOp) { + // Do single-target, single doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); + request.getDeleteRequest()->addToDeletes(buildDelete(BSON("x" << 1), 1)); - // - // Single-op targeting test - // + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterFullRange( nss, endpoint, &targeter ); + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); - // Do single-target, single doc batch write op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpoint); - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); + BatchedCommandResponse response; + buildErrResponse(ErrorCodes::UnknownError, "message", &response); - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpoint ); + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); - BatchedCommandResponse response; - buildResponse( 1, &response ); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), response.getErrCode()); + ASSERT(clientResponse.getErrDetailsAt(0)->getErrMessage().find(response.getErrMessage()) != + string::npos); + ASSERT_EQUALS(clientResponse.getN(), 0); +} - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); +TEST(WriteOpTests, SingleTargetError) { + // + // Single-op targeting error test + // - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterHalfRange(nss, endpoint, &targeter); - } + // Do untargetable delete op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); + request.getDeleteRequest()->addToDeletes(buildDelete(BSON("x" << 1), 1)); - TEST(WriteOpTests, SingleError) { + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); - // - // Single-op error test - // + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterFullRange( nss, endpoint, &targeter ); + ASSERT(!status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 0u); - // Do single-target, single doc batch write op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Delete ); - request.setNS( nss.ns() ); - request.getDeleteRequest()->addToDeletes( buildDelete( BSON( "x" << 1 ), 1 ) ); + // Record targeting failures + status = batchOp.targetBatch(targeter, true, &targeted); - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); + ASSERT(status.isOK()); + ASSERT(batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 0u); - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 0); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); +} - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpoint ); +TEST(WriteOpTests, SingleWriteConcernErrorOrdered) { + // + // Write concern error test - we should pass write concern to sub-batches, and pass up the + // write concern error if one occurs + // - BatchedCommandResponse response; - buildErrResponse( ErrorCodes::UnknownError, "message", &response ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + request.setWriteConcern(BSON("w" << 3)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpoint); + + BatchedCommandRequest targetBatch(BatchedCommandRequest::BatchType_Insert); + batchOp.buildBatchRequest(*targeted.front(), &targetBatch); + ASSERT(targetBatch.getWriteConcern().woCompare(request.getWriteConcern()) == 0); + + BatchedCommandResponse response; + buildResponse(1, &response); + addWCError(&response); + + // First stale response comes back, we should retry + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(!clientResponse.isErrDetailsSet()); + ASSERT(clientResponse.isWriteConcernErrorSet()); +} + +TEST(WriteOpTests, SingleStaleError) { + // + // Single-op stale version test + // We should retry the same batch until we're not stale + // - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrCode(), response.getErrCode() ); - ASSERT( clientResponse.getErrDetailsAt( 0 )->getErrMessage() - .find( response.getErrMessage()) != string::npos ); - ASSERT_EQUALS( clientResponse.getN(), 0 ); - } + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); - TEST(WriteOpTests, SingleTargetError) { + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); - // - // Single-op targeting error test - // + BatchedCommandResponse response; + buildResponse(0, &response); + addError(ErrorCodes::StaleShardVersion, "mock stale error", 0, &response); - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterHalfRange( nss, endpoint, &targeter ); + // First stale response comes back, we should retry + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); - // Do untargetable delete op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Delete ); - request.setNS( nss.ns() ); - request.getDeleteRequest()->addToDeletes( buildDelete( BSON( "x" << 1 ), 1 ) ); + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); + // Respond again with a stale response + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); - ASSERT( !status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 0u ); + buildResponse(1, &response); - // Record targeting failures - status = batchOp.targetBatch( targeter, true, &targeted ); + // Respond with an 'ok' response + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); - ASSERT( status.isOK() ); - ASSERT( batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 0u ); + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(!clientResponse.isErrDetailsSet()); +} - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 0 ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - } +// +// Multi-operation batches +// - TEST(WriteOpTests, SingleWriteConcernErrorOrdered) { - - // - // Write concern error test - we should pass write concern to sub-batches, and pass up the - // write concern error if one occurs - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterFullRange( nss, endpoint, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - request.setWriteConcern( BSON( "w" << 3 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpoint ); - - BatchedCommandRequest targetBatch( BatchedCommandRequest::BatchType_Insert ); - batchOp.buildBatchRequest( *targeted.front(), &targetBatch ); - ASSERT( targetBatch.getWriteConcern().woCompare( request.getWriteConcern() ) == 0 ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - addWCError( &response ); - - // First stale response comes back, we should retry - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( !clientResponse.isErrDetailsSet() ); - ASSERT( clientResponse.isWriteConcernErrorSet() ); +struct EndpointComp { + bool operator()(const TargetedWriteBatch* writeA, const TargetedWriteBatch* writeB) const { + return writeA->getEndpoint().shardName.compare(writeB->getEndpoint().shardName) < 0; } +}; - TEST(WriteOpTests, SingleStaleError) { - - // - // Single-op stale version test - // We should retry the same batch until we're not stale - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterFullRange( nss, endpoint, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - BatchedCommandResponse response; - buildResponse( 0, &response ); - addError( ErrorCodes::StaleShardVersion, "mock stale error", 0, &response ); - - // First stale response comes back, we should retry - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, false, &targeted ); - - // Respond again with a stale response - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, false, &targeted ); - - buildResponse( 1, &response ); - - // Respond with an 'ok' response - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( !clientResponse.isErrDetailsSet() ); - } +inline void sortByEndpoint(vector<TargetedWriteBatch*>* writes) { + std::sort(writes->begin(), writes->end(), EndpointComp()); +} +TEST(WriteOpTests, MultiOpSameShardOrdered) { // - // Multi-operation batches + // Multi-op targeting test (ordered) // - struct EndpointComp { - bool operator()( const TargetedWriteBatch* writeA, - const TargetedWriteBatch* writeB ) const { - return writeA->getEndpoint().shardName.compare( writeB->getEndpoint().shardName ) < 0; - } - }; - - inline void sortByEndpoint( vector<TargetedWriteBatch*>* writes ) { - std::sort( writes->begin(), writes->end(), EndpointComp() ); - } - - TEST(WriteOpTests, MultiOpSameShardOrdered) { - - // - // Multi-op targeting test (ordered) - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterFullRange( nss, endpoint, &targeter ); - - // Do single-target, multi-doc batch write op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Update ); - request.setNS( nss.ns() ); - request.setOrdered( true ); - request.getUpdateRequest()->addToUpdates( buildUpdate( BSON( "x" << 1 ), false ) ); - request.getUpdateRequest()->addToUpdates( buildUpdate( BSON( "x" << 2 ), false ) ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); + + // Do single-target, multi-doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); + request.setNS(nss.ns()); + request.setOrdered(true); + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << 1), false)); + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << 2), false)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpoint); + + BatchedCommandResponse response; + buildResponse(2, &response); + + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 2); +} + +TEST(WriteOpTests, MultiOpSameShardUnordered) { + // + // Multi-op targeting test (unordered) + // - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); + + // Do single-target, multi-doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); + request.setNS(nss.ns()); + request.setOrdered(false); + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << 1), false)); + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << 2), false)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpoint); + + BatchedCommandResponse response; + buildResponse(2, &response); + + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 2); +} + +TEST(WriteOpTests, MultiOpTwoShardsOrdered) { + // + // Multi-op, multi-endpoing targeting test (ordered) + // There should be two sets of single batches (one to each shard, one-by-one) + // - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + // Do multi-target, multi-doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(true); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + + BatchedCommandResponse response; + buildResponse(1, &response); + + // Respond to first targeted batch + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointB); + + // Respond to second targeted batch + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 2); +} + +TEST(WriteOpTests, MultiOpTwoShardsUnordered) { + // + // Multi-op, multi-endpoint targeting test (unordered) + // There should be one set of two batches (one to each shard) + // - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 2u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpoint ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + // Do multi-target, multi-doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 1u); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + + BatchedCommandResponse response; + buildResponse(1, &response); + + // Respond to both targeted batches + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 2); +} + +TEST(WriteOpTests, MultiOpTwoShardsEachOrdered) { + // + // Multi-op (ordered) targeting test where each op goes to both shards + // There should be two sets of two batches to each shard (two for each delete op) + // - BatchedCommandResponse response; - buildResponse( 2, &response ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + // Do multi-target, multi-doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); + request.setOrdered(true); + BSONObj queryA = BSON("x" << GTE << -1 << LT << 2); + request.getDeleteRequest()->addToDeletes(buildDelete(queryA, 0)); + BSONObj queryB = BSON("x" << GTE << -2 << LT << 1); + request.getDeleteRequest()->addToDeletes(buildDelete(queryB, 0)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + + BatchedCommandResponse response; + buildResponse(1, &response); + + // Respond to both targeted batches for first multi-delete + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(!batchOp.isFinished()); + + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + + // Respond to second targeted batches for second multi-delete + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 4); +} + +TEST(WriteOpTests, MultiOpTwoShardsEachUnordered) { + // + // Multi-op (unaordered) targeting test where each op goes to both shards + // There should be one set of two batches to each shard (containing writes for both ops) + // - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + // Do multi-target, multi-doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); + request.setNS(nss.ns()); + request.setOrdered(false); + BSONObj queryA = BSON("x" << GTE << -1 << LT << 2); + request.getUpdateRequest()->addToUpdates(buildUpdate(queryA, true)); + BSONObj queryB = BSON("x" << GTE << -2 << LT << 1); + request.getUpdateRequest()->addToUpdates(buildUpdate(queryB, true)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 2u); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + + BatchedCommandResponse response; + buildResponse(2, &response); + + // Respond to both targeted batches, each containing two ops + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 4); +} + +TEST(WriteOpTests, MultiOpOneOrTwoShardsOrdered) { + // + // Multi-op (ordered) targeting test where first two ops go to one shard, second two ops + // go to two shards. + // Should batch the first two ops, then second ops should be batched separately, then + // last ops should be batched together + // - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 2 ); - } + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); + request.setOrdered(true); + // These go to the same shard + request.getDeleteRequest()->addToDeletes(buildDelete(BSON("x" << -1), 1)); + request.getDeleteRequest()->addToDeletes(buildDelete(BSON("x" << -2), 1)); + // These go to both shards + BSONObj queryA = BSON("x" << GTE << -1 << LT << 2); + request.getDeleteRequest()->addToDeletes(buildDelete(queryA, 0)); + BSONObj queryB = BSON("x" << GTE << -2 << LT << 1); + request.getDeleteRequest()->addToDeletes(buildDelete(queryB, 0)); + // These go to the same shard + request.getDeleteRequest()->addToDeletes(buildDelete(BSON("x" << 1), 1)); + request.getDeleteRequest()->addToDeletes(buildDelete(BSON("x" << 2), 1)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + + BatchedCommandResponse response; + // Emulate one-write-per-delete-per-host + buildResponse(2, &response); + + // Respond to first targeted batch containing the two single-host deletes + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + + // Emulate one-write-per-delete-per-host + buildResponse(1, &response); + + // Respond to two targeted batches for first multi-delete + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(!batchOp.isFinished()); + + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 1u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + + // Respond to two targeted batches for second multi-delete + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(!batchOp.isFinished()); + + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + + // Emulate one-write-per-delete-per-host + buildResponse(2, &response); + + // Respond to final targeted batch containing the last two single-host deletes + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 8); +} + +TEST(WriteOpTests, MultiOpOneOrTwoShardsUnordered) { + // + // Multi-op (unordered) targeting test where first two ops go to one shard, second two ops + // go to two shards. + // Should batch all the ops together into two batches of four ops for each shard + // - TEST(WriteOpTests, MultiOpSameShardUnordered) { + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); + request.setNS(nss.ns()); + request.setOrdered(false); + // These go to the same shard + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << -1), false)); + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << -2), false)); + // These go to both shards + BSONObj queryA = BSON("x" << GTE << -1 << LT << 2); + request.getUpdateRequest()->addToUpdates(buildUpdate(queryA, true)); + BSONObj queryB = BSON("x" << GTE << -2 << LT << 1); + request.getUpdateRequest()->addToUpdates(buildUpdate(queryB, true)); + // These go to the same shard + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << 1), false)); + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << 2), false)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 4u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 4u); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + + BatchedCommandResponse response; + // Emulate one-write-per-delete-per-host + buildResponse(4, &response); + + // Respond to first targeted batch containing the two single-host deletes + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 8); +} + +TEST(WriteOpTests, MultiOpSingleShardErrorUnordered) { + // + // Multi-op targeting test where two ops go to two separate shards and there's an error on + // one op on one shard + // There should be one set of two batches to each shard and an error reported + // - // - // Multi-op targeting test (unordered) - // + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 1u); + + BatchedCommandResponse response; + buildResponse(1, &response); + + // No error on first shard + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + buildResponse(0, &response); + addError(ErrorCodes::UnknownError, "mock error", 0, &response); + + // Error on second write on second shard + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), + response.getErrDetailsAt(0)->getErrCode()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrMessage(), + response.getErrDetailsAt(0)->getErrMessage()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 1); +} + +TEST(WriteOpTests, MultiOpTwoShardErrorsUnordered) { + // + // Multi-op targeting test where two ops go to two separate shards and there's an error on + // each op on each shard + // There should be one set of two batches to each shard and and two errors reported + // - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterFullRange( nss, endpoint, &targeter ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 1u); + + BatchedCommandResponse response; + buildResponse(0, &response); + addError(ErrorCodes::UnknownError, "mock error", 0, &response); + + // Error on first write on first shard + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + // Error on second write on second shard + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 0); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 2u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), + response.getErrDetailsAt(0)->getErrCode()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrMessage(), + response.getErrDetailsAt(0)->getErrMessage()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 0); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(1)->getErrCode(), + response.getErrDetailsAt(0)->getErrCode()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(1)->getErrMessage(), + response.getErrDetailsAt(0)->getErrMessage()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(1)->getIndex(), 1); +} + +TEST(WriteOpTests, MultiOpPartialSingleShardErrorUnordered) { + // + // Multi-op targeting test where each op goes to both shards and there's an error on + // one op on one shard + // There should be one set of two batches to each shard and an error reported + // - // Do single-target, multi-doc batch write op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Update ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.getUpdateRequest()->addToUpdates( buildUpdate( BSON( "x" << 1 ), false ) ); - request.getUpdateRequest()->addToUpdates( buildUpdate( BSON( "x" << 2 ), false ) ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); + request.setOrdered(false); + BSONObj queryA = BSON("x" << GTE << -1 << LT << 2); + request.getDeleteRequest()->addToDeletes(buildDelete(queryA, 0)); + BSONObj queryB = BSON("x" << GTE << -2 << LT << 1); + request.getDeleteRequest()->addToDeletes(buildDelete(queryB, 0)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 2u); + + BatchedCommandResponse response; + buildResponse(2, &response); + + // No errors on first shard + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + buildResponse(1, &response); + addError(ErrorCodes::UnknownError, "mock error", 1, &response); + + // Error on second write on second shard + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 3); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), + response.getErrDetailsAt(0)->getErrCode()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrMessage(), + response.getErrDetailsAt(0)->getErrMessage()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 1); +} + +TEST(WriteOpTests, MultiOpPartialSingleShardErrorOrdered) { + // + // Multi-op targeting test where each op goes to both shards and there's an error on + // one op on one shard + // There should be one set of two batches to each shard and an error reported, the second + // op should not get run + // - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); + request.setOrdered(true); + BSONObj queryA = BSON("x" << GTE << -1 << LT << 2); + request.getDeleteRequest()->addToDeletes(buildDelete(queryA, 0)); + BSONObj queryB = BSON("x" << GTE << -2 << LT << 1); + request.getDeleteRequest()->addToDeletes(buildDelete(queryB, 0)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + ASSERT(!batchOp.isFinished()); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 2u); + sortByEndpoint(&targeted); + assertEndpointsEqual(targeted.front()->getEndpoint(), endpointA); + assertEndpointsEqual(targeted.back()->getEndpoint(), endpointB); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); + ASSERT_EQUALS(targeted.back()->getWrites().size(), 1u); + + BatchedCommandResponse response; + buildResponse(1, &response); + + // No errors on first shard + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + buildResponse(0, &response); + addError(ErrorCodes::UnknownError, "mock error", 0, &response); + + // Error on second write on second shard + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), + response.getErrDetailsAt(0)->getErrCode()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrMessage(), + response.getErrDetailsAt(0)->getErrMessage()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 0); +} + +// +// Tests of edge-case functionality, lifecycle is assumed to be behaving normally +// + +TEST(WriteOpTests, MultiOpErrorAndWriteConcernErrorUnordered) { + // + // Multi-op (unordered) error and write concern error test + // We never report the write concern error for single-doc batches, since the error means + // there's no write concern applied. + // Don't suppress the error if ordered : false + // - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); + request.setWriteConcern(BSON("w" << 3)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + BatchedCommandResponse response; + buildResponse(1, &response); + addError(ErrorCodes::UnknownError, "mock error", 1, &response); + addWCError(&response); + + // First stale response comes back, we should retry + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + // Unordered reports write concern error + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT(clientResponse.isWriteConcernErrorSet()); +} + + +TEST(WriteOpTests, SingleOpErrorAndWriteConcernErrorOrdered) { + // + // Single-op (ordered) error and write concern error test + // Suppress the write concern error if ordered and we also have an error + // - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 2u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpoint ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); + request.setNS(nss.ns()); + request.setOrdered(true); + BSONObj query = BSON("x" << GTE << -1 << LT << 2); + request.getUpdateRequest()->addToUpdates(buildUpdate(query, true)); + request.setWriteConcern(BSON("w" << 3)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + BatchedCommandResponse response; + buildResponse(1, &response); + addWCError(&response); + + // First response comes back with write concern error + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + buildResponse(0, &response); + addError(ErrorCodes::UnknownError, "mock error", 0, &response); + + // Second response comes back with write error + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + // Ordered doesn't report write concern error + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT(!clientResponse.isWriteConcernErrorSet()); +} + +TEST(WriteOpTests, MultiOpFailedTargetOrdered) { + // + // Targeting failure on second op in batch op (ordered) + // - BatchedCommandResponse response; - buildResponse( 2, &response ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterHalfRange(nss, endpoint, &targeter); - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 2)); + request.getInsertRequest()->addToDocuments(BSON("x" << -2)); - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 2 ); - } + // Do single-target, multi-doc batch write op - TEST(WriteOpTests, MultiOpTwoShardsOrdered) { - - // - // Multi-op, multi-endpoing targeting test (ordered) - // There should be two sets of single batches (one to each shard, one-by-one) - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - // Do multi-target, multi-doc batch write op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( true ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - - // Respond to first targeted batch - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, false, &targeted ); - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointB ); - - // Respond to second targeted batch - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 2 ); - } + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); - TEST(WriteOpTests, MultiOpTwoShardsUnordered) { - - // - // Multi-op, multi-endpoint targeting test (unordered) - // There should be one set of two batches (one to each shard) - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - // Do multi-target, multi-doc batch write op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 1u ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - - // Respond to both targeted batches - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 2 ); - } + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); - TEST(WriteOpTests, MultiOpTwoShardsEachOrdered) { - - // - // Multi-op (ordered) targeting test where each op goes to both shards - // There should be two sets of two batches to each shard (two for each delete op) - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - // Do multi-target, multi-doc batch write op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Delete ); - request.setNS( nss.ns() ); - request.setOrdered( true ); - BSONObj queryA = BSON( "x" << GTE << -1 << LT << 2 ); - request.getDeleteRequest()->addToDeletes( buildDelete( queryA, 0 ) ); - BSONObj queryB = BSON( "x" << GTE << -2 << LT << 1 ); - request.getDeleteRequest()->addToDeletes( buildDelete( queryB, 0 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - - // Respond to both targeted batches for first multi-delete - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, false, &targeted ); - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - - // Respond to second targeted batches for second multi-delete - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 4 ); - } + // First targeting round fails since we may be stale + ASSERT(!status.isOK()); + ASSERT(!batchOp.isFinished()); - TEST(WriteOpTests, MultiOpTwoShardsEachUnordered) { - - // - // Multi-op (unaordered) targeting test where each op goes to both shards - // There should be one set of two batches to each shard (containing writes for both ops) - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - // Do multi-target, multi-doc batch write op - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Update ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - BSONObj queryA = BSON( "x" << GTE << -1 << LT << 2 ); - request.getUpdateRequest()->addToUpdates( buildUpdate( queryA, true ) ); - BSONObj queryB = BSON( "x" << GTE << -2 << LT << 1 ); - request.getUpdateRequest()->addToUpdates( buildUpdate( queryB, true ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 2u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 2u ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - - BatchedCommandResponse response; - buildResponse( 2, &response ); - - // Respond to both targeted batches, each containing two ops - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 4 ); - } + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, true, &targeted); - TEST(WriteOpTests, MultiOpOneOrTwoShardsOrdered) { - - // - // Multi-op (ordered) targeting test where first two ops go to one shard, second two ops - // go to two shards. - // Should batch the first two ops, then second ops should be batched separately, then - // last ops should be batched together - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Delete ); - request.setNS( nss.ns() ); - request.setOrdered( true ); - // These go to the same shard - request.getDeleteRequest()->addToDeletes( buildDelete( BSON( "x" << -1 ), 1 ) ); - request.getDeleteRequest()->addToDeletes( buildDelete( BSON( "x" << -2 ), 1 ) ); - // These go to both shards - BSONObj queryA = BSON( "x" << GTE << -1 << LT << 2 ); - request.getDeleteRequest()->addToDeletes( buildDelete( queryA, 0 ) ); - BSONObj queryB = BSON( "x" << GTE << -2 << LT << 1 ); - request.getDeleteRequest()->addToDeletes( buildDelete( queryB, 0 ) ); - // These go to the same shard - request.getDeleteRequest()->addToDeletes( buildDelete( BSON( "x" << 1 ), 1 ) ); - request.getDeleteRequest()->addToDeletes( buildDelete( BSON( "x" << 2 ), 1 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 2u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - - BatchedCommandResponse response; - // Emulate one-write-per-delete-per-host - buildResponse( 2, &response ); - - // Respond to first targeted batch containing the two single-host deletes - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - - // Emulate one-write-per-delete-per-host - buildResponse( 1, &response ); - - // Respond to two targeted batches for first multi-delete - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 1u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - - // Respond to two targeted batches for second multi-delete - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 2u ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - - // Emulate one-write-per-delete-per-host - buildResponse( 2, &response ); - - // Respond to final targeted batch containing the last two single-host deletes - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 8 ); - } + // Second targeting round is ok, but should stop at first write + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); - TEST(WriteOpTests, MultiOpOneOrTwoShardsUnordered) { - - // - // Multi-op (unordered) targeting test where first two ops go to one shard, second two ops - // go to two shards. - // Should batch all the ops together into two batches of four ops for each shard - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Update ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - // These go to the same shard - request.getUpdateRequest()->addToUpdates( buildUpdate( BSON( "x" << -1 ), false ) ); - request.getUpdateRequest()->addToUpdates( buildUpdate( BSON( "x" << -2 ), false ) ); - // These go to both shards - BSONObj queryA = BSON( "x" << GTE << -1 << LT << 2 ); - request.getUpdateRequest()->addToUpdates( buildUpdate( queryA, true ) ); - BSONObj queryB = BSON( "x" << GTE << -2 << LT << 1 ); - request.getUpdateRequest()->addToUpdates( buildUpdate( queryB, true ) ); - // These go to the same shard - request.getUpdateRequest()->addToUpdates( buildUpdate( BSON( "x" << 1 ), false ) ); - request.getUpdateRequest()->addToUpdates( buildUpdate( BSON( "x" << 2 ), false ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 4u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 4u ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - - BatchedCommandResponse response; - // Emulate one-write-per-delete-per-host - buildResponse( 4, &response ); - - // Respond to first targeted batch containing the two single-host deletes - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 8 ); - } + BatchedCommandResponse response; + buildResponse(1, &response); - TEST(WriteOpTests, MultiOpSingleShardErrorUnordered) { - - // - // Multi-op targeting test where two ops go to two separate shards and there's an error on - // one op on one shard - // There should be one set of two batches to each shard and an error reported - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 1u ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - - // No error on first shard - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - buildResponse( 0, &response ); - addError( ErrorCodes::UnknownError, "mock error", 0, &response ); - - // Error on second write on second shard - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrCode(), - response.getErrDetailsAt(0)->getErrCode() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrMessage(), - response.getErrDetailsAt(0)->getErrMessage() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getIndex(), 1 ); - } + // First response ok + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); - TEST(WriteOpTests, MultiOpTwoShardErrorsUnordered) { - - // - // Multi-op targeting test where two ops go to two separate shards and there's an error on - // each op on each shard - // There should be one set of two batches to each shard and and two errors reported - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 1u ); - - BatchedCommandResponse response; - buildResponse( 0, &response ); - addError( ErrorCodes::UnknownError, "mock error", 0, &response ); - - // Error on first write on first shard - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - // Error on second write on second shard - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 0 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 2u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrCode(), - response.getErrDetailsAt(0)->getErrCode() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrMessage(), - response.getErrDetailsAt(0)->getErrMessage() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getIndex(), 0 ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 1 )->getErrCode(), - response.getErrDetailsAt(0)->getErrCode() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 1 )->getErrMessage(), - response.getErrDetailsAt(0)->getErrMessage() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 1 )->getIndex(), 1 ); - } + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, true, &targeted); - TEST(WriteOpTests, MultiOpPartialSingleShardErrorUnordered) { - - // - // Multi-op targeting test where each op goes to both shards and there's an error on - // one op on one shard - // There should be one set of two batches to each shard and an error reported - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Delete ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - BSONObj queryA = BSON( "x" << GTE << -1 << LT << 2 ); - request.getDeleteRequest()->addToDeletes( buildDelete( queryA, 0 ) ); - BSONObj queryB = BSON( "x" << GTE << -2 << LT << 1 ); - request.getDeleteRequest()->addToDeletes( buildDelete( queryB, 0 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 2u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 2u ); - - BatchedCommandResponse response; - buildResponse( 2, &response ); - - // No errors on first shard - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - buildResponse( 1, &response ); - addError( ErrorCodes::UnknownError, "mock error", 1, &response ); - - // Error on second write on second shard - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 3 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrCode(), - response.getErrDetailsAt(0)->getErrCode() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrMessage(), - response.getErrDetailsAt(0)->getErrMessage() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getIndex(), 1 ); - } + // Second targeting round results in an error which finishes the batch + ASSERT(status.isOK()); + ASSERT(batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 0u); - TEST(WriteOpTests, MultiOpPartialSingleShardErrorOrdered) { - - // - // Multi-op targeting test where each op goes to both shards and there's an error on - // one op on one shard - // There should be one set of two batches to each shard and an error reported, the second - // op should not get run - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Delete ); - request.setNS( nss.ns() ); - request.setOrdered( true ); - BSONObj queryA = BSON( "x" << GTE << -1 << LT << 2 ); - request.getDeleteRequest()->addToDeletes( buildDelete( queryA, 0 ) ); - BSONObj queryB = BSON( "x" << GTE << -2 << LT << 1 ); - request.getDeleteRequest()->addToDeletes( buildDelete( queryB, 0 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - ASSERT( !batchOp.isFinished() ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 2u ); - sortByEndpoint( &targeted ); - assertEndpointsEqual( targeted.front()->getEndpoint(), endpointA ); - assertEndpointsEqual( targeted.back()->getEndpoint(), endpointB ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - ASSERT_EQUALS( targeted.back()->getWrites().size(), 1u ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - - // No errors on first shard - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - buildResponse( 0, &response ); - addError( ErrorCodes::UnknownError, "mock error", 0, &response ); - - // Error on second write on second shard - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrCode(), - response.getErrDetailsAt(0)->getErrCode() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getErrMessage(), - response.getErrDetailsAt(0)->getErrMessage() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt( 0 )->getIndex(), 0 ); - } + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 1); +} +TEST(WriteOpTests, MultiOpFailedTargetUnordered) { // - // Tests of edge-case functionality, lifecycle is assumed to be behaving normally + // Targeting failure on second op in batch op (unordered) // - TEST(WriteOpTests, MultiOpErrorAndWriteConcernErrorUnordered) { - - // - // Multi-op (unordered) error and write concern error test - // We never report the write concern error for single-doc batches, since the error means - // there's no write concern applied. - // Don't suppress the error if ordered : false - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterFullRange( nss, endpoint, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); - request.setWriteConcern( BSON( "w" << 3 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - addError( ErrorCodes::UnknownError, "mock error", 1, &response ); - addWCError( &response ); - - // First stale response comes back, we should retry - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - // Unordered reports write concern error - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT( clientResponse.isWriteConcernErrorSet() ); - } - - - TEST(WriteOpTests, SingleOpErrorAndWriteConcernErrorOrdered) { - - // - // Single-op (ordered) error and write concern error test - // Suppress the write concern error if ordered and we also have an error - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Update ); - request.setNS( nss.ns() ); - request.setOrdered( true ); - BSONObj query = BSON( "x" << GTE << -1 << LT << 2 ); - request.getUpdateRequest()->addToUpdates( buildUpdate( query, true ) ); - request.setWriteConcern( BSON( "w" << 3 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - addWCError( &response ); - - // First response comes back with write concern error - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - buildResponse( 0, &response ); - addError( ErrorCodes::UnknownError, "mock error", 0, &response ); - - // Second response comes back with write error - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - // Ordered doesn't report write concern error - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT( !clientResponse.isWriteConcernErrorSet() ); - } - - TEST(WriteOpTests, MultiOpFailedTargetOrdered) { - - // - // Targeting failure on second op in batch op (ordered) - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterHalfRange( nss, endpoint, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 2 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -2 ) ); - - // Do single-target, multi-doc batch write op - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - // First targeting round fails since we may be stale - ASSERT( !status.isOK() ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, true, &targeted ); - - // Second targeting round is ok, but should stop at first write - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 1u ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterHalfRange(nss, endpoint, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 2)); + request.getInsertRequest()->addToDocuments(BSON("x" << -2)); + + // Do single-target, multi-doc batch write op + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + // First targeting round fails since we may be stale + ASSERT(!status.isOK()); + ASSERT(!batchOp.isFinished()); + + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, true, &targeted); + + // Second targeting round is ok, and should record an error + ASSERT(status.isOK()); + ASSERT(!batchOp.isFinished()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); + + BatchedCommandResponse response; + buildResponse(2, &response); + + // Response is ok for first and third write + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 2); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 1); +} + +TEST(WriteOpTests, MultiOpFailedBatchOrdered) { + // + // Batch failure (ok : 0) reported in a multi-op batch (ordered) + // Expect this gets translated down into write errors for first affected write + // - // First response ok - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 2)); + request.getInsertRequest()->addToDocuments(BSON("x" << 3)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + BatchedCommandResponse response; + buildResponse(1, &response); + + // First shard batch is ok + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, true, &targeted); + + buildErrResponse(ErrorCodes::UnknownError, "mock error", &response); + + // Second shard batch fails + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + // We should have recorded an error for the second write + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 1); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), response.getErrCode()); +} + +TEST(WriteOpTests, MultiOpFailedBatchUnordered) { + // + // Batch failure (ok : 0) reported in a multi-op batch (unordered) + // Expect this gets translated down into write errors for all affected writes + // - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, true, &targeted ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 2)); + request.getInsertRequest()->addToDocuments(BSON("x" << 3)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + BatchedCommandResponse response; + buildResponse(1, &response); + + // First shard batch is ok + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + buildErrResponse(ErrorCodes::UnknownError, "mock error", &response); + + // Second shard batch fails + batchOp.noteBatchResponse(*targeted.back(), response, NULL); + ASSERT(batchOp.isFinished()); + + // We should have recorded an error for the second and third write + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 2u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 1); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), response.getErrCode()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(1)->getIndex(), 2); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(1)->getErrCode(), response.getErrCode()); +} + +TEST(WriteOpTests, MultiOpAbortOrdered) { + // + // Batch aborted (ordered) + // Expect this gets translated down into write error for first affected write + // - // Second targeting round results in an error which finishes the batch - ASSERT( status.isOK() ); - ASSERT( batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 0u ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 2)); + request.getInsertRequest()->addToDocuments(BSON("x" << 3)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + BatchedCommandResponse response; + buildResponse(1, &response); + + // First shard batch is ok + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + WriteErrorDetail abortError; + abortError.setErrCode(ErrorCodes::UnknownError); + abortError.setErrMessage("mock abort"); + batchOp.abortBatch(abortError); + ASSERT(batchOp.isFinished()); + + // We should have recorded an error for the second write + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 1); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 1u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 1); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), abortError.getErrCode()); +} + +TEST(WriteOpTests, MultiOpAbortUnordered) { + // + // Batch aborted (unordered) + // Expect this gets translated down into write errors for all affected writes + // - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getIndex(), 1 ); - } + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.setOrdered(false); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << -2)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + + WriteErrorDetail abortError; + abortError.setErrCode(ErrorCodes::UnknownError); + abortError.setErrMessage("mock abort"); + batchOp.abortBatch(abortError); + ASSERT(batchOp.isFinished()); + + // We should have recorded an error for the first and second write + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 0); + ASSERT(clientResponse.isErrDetailsSet()); + ASSERT_EQUALS(clientResponse.sizeErrDetails(), 2u); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getIndex(), 0); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(0)->getErrCode(), abortError.getErrCode()); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(1)->getIndex(), 1); + ASSERT_EQUALS(clientResponse.getErrDetailsAt(1)->getErrCode(), abortError.getErrCode()); +} + +TEST(WriteOpTests, MultiOpTwoWCErrors) { + // + // Multi-op targeting test where each op goes to both shards and both return a write concern + // error + // - TEST(WriteOpTests, MultiOpFailedTargetUnordered) { + NamespaceString nss("foo.bar"); + ShardEndpoint endpointA("shardA", ChunkVersion::IGNORED()); + ShardEndpoint endpointB("shardB", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterSplitRange(nss, endpointA, endpointB, &targeter); + + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << -1)); + request.getInsertRequest()->addToDocuments(BSON("x" << 2)); + request.setWriteConcern(BSON("w" << 3)); + + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); + + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + + BatchedCommandResponse response; + buildResponse(1, &response); + addWCError(&response); + + // First shard write write concern fails. + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); + + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, true, &targeted); + + // Second shard write write concern fails. + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); + + BatchedCommandResponse clientResponse; + batchOp.buildClientResponse(&clientResponse); + ASSERT(clientResponse.getOk()); + ASSERT_EQUALS(clientResponse.getN(), 2); + ASSERT(!clientResponse.isErrDetailsSet()); + ASSERT(clientResponse.isWriteConcernErrorSet()); +} + +// +// Tests of batch size limit functionality +// + +TEST(WriteOpLimitTests, OneBigDoc) { + // + // Big single operation test - should go through + // - // - // Targeting failure on second op in batch op (unordered) - // + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterHalfRange( nss, endpoint, &targeter ); + // Create a BSONObj (slightly) bigger than the maximum size by including a max-size string + string bigString(BSONObjMaxUserSize, 'x'); - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 2 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -2 ) ); + // Do single-target, single doc batch write op + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << 1 << "data" << bigString)); - // Do single-target, multi-doc batch write op + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT_EQUALS(targeted.size(), 1u); - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); + BatchedCommandResponse response; + buildResponse(1, &response); - // First targeting round fails since we may be stale - ASSERT( !status.isOK() ); - ASSERT( !batchOp.isFinished() ); + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); +} - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, true, &targeted ); +TEST(WriteOpLimitTests, OneBigOneSmall) { + // + // Big doc with smaller additional doc - should go through as two batches + // - // Second targeting round is ok, and should record an error - ASSERT( status.isOK() ); - ASSERT( !batchOp.isFinished() ); - ASSERT_EQUALS( targeted.size(), 1u ); - ASSERT_EQUALS( targeted.front()->getWrites().size(), 2u ); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); - BatchedCommandResponse response; - buildResponse( 2, &response ); + // Create a BSONObj (slightly) bigger than the maximum size by including a max-size string + string bigString(BSONObjMaxUserSize, 'x'); - // Response is ok for first and third write - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); + request.setNS(nss.ns()); + BatchedUpdateDocument* bigUpdateDoc = + buildUpdate(BSON("x" << 1), BSON("data" << bigString), false); + request.getUpdateRequest()->addToUpdates(bigUpdateDoc); + request.getUpdateRequest()->addToUpdates(buildUpdate(BSON("x" << 2), BSONObj(), false)); - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 2 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getIndex(), 1 ); - } + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); - TEST(WriteOpTests, MultiOpFailedBatchOrdered) { - - // - // Batch failure (ok : 0) reported in a multi-op batch (ordered) - // Expect this gets translated down into write errors for first affected write - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 2 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 3 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - - // First shard batch is ok - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, true, &targeted ); - - buildErrResponse( ErrorCodes::UnknownError, "mock error", &response ); - - // Second shard batch fails - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - // We should have recorded an error for the second write - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getIndex(), 1 ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getErrCode(), response.getErrCode() ); - } + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); - TEST(WriteOpTests, MultiOpFailedBatchUnordered) { - - // - // Batch failure (ok : 0) reported in a multi-op batch (unordered) - // Expect this gets translated down into write errors for all affected writes - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 2 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 3 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - - // First shard batch is ok - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - buildErrResponse( ErrorCodes::UnknownError, "mock error", &response ); - - // Second shard batch fails - batchOp.noteBatchResponse( *targeted.back(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - // We should have recorded an error for the second and third write - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 2u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getIndex(), 1 ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getErrCode(), response.getErrCode() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(1)->getIndex(), 2 ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(1)->getErrCode(), response.getErrCode() ); - } + BatchedCommandResponse response; + buildResponse(1, &response); - TEST(WriteOpTests, MultiOpAbortOrdered) { - - // - // Batch aborted (ordered) - // Expect this gets translated down into write error for first affected write - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 2 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 3 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - - // First shard batch is ok - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - WriteErrorDetail abortError; - abortError.setErrCode( ErrorCodes::UnknownError ); - abortError.setErrMessage( "mock abort" ); - batchOp.abortBatch( abortError ); - ASSERT( batchOp.isFinished() ); - - // We should have recorded an error for the second write - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 1 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 1u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getIndex(), 1 ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getErrCode(), abortError.getErrCode() ); - } + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); - TEST(WriteOpTests, MultiOpAbortUnordered) { - - // - // Batch aborted (unordered) - // Expect this gets translated down into write errors for all affected writes - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.setOrdered( false ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -2 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - WriteErrorDetail abortError; - abortError.setErrCode( ErrorCodes::UnknownError ); - abortError.setErrMessage( "mock abort" ); - batchOp.abortBatch( abortError ); - ASSERT( batchOp.isFinished() ); - - // We should have recorded an error for the first and second write - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 0 ); - ASSERT( clientResponse.isErrDetailsSet() ); - ASSERT_EQUALS( clientResponse.sizeErrDetails(), 2u ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getIndex(), 0 ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(0)->getErrCode(), abortError.getErrCode() ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(1)->getIndex(), 1 ); - ASSERT_EQUALS( clientResponse.getErrDetailsAt(1)->getErrCode(), abortError.getErrCode() ); - } + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); - TEST(WriteOpTests, MultiOpTwoWCErrors) { - - // - // Multi-op targeting test where each op goes to both shards and both return a write concern - // error - // - - NamespaceString nss( "foo.bar" ); - ShardEndpoint endpointA( "shardA", ChunkVersion::IGNORED() ); - ShardEndpoint endpointB( "shardB", ChunkVersion::IGNORED() ); - MockNSTargeter targeter; - initTargeterSplitRange( nss, endpointA, endpointB, &targeter ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << -1 ) ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 2 ) ); - request.setWriteConcern( BSON( "w" << 3 ) ); - - BatchWriteOp batchOp; - batchOp.initClientRequest( &request ); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch( targeter, false, &targeted ); - - BatchedCommandResponse response; - buildResponse( 1, &response ); - addWCError( &response ); - - // First shard write write concern fails. - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( !batchOp.isFinished() ); - - targetedOwned.clear(); - status = batchOp.targetBatch( targeter, true, &targeted ); - - // Second shard write write concern fails. - batchOp.noteBatchResponse( *targeted.front(), response, NULL ); - ASSERT( batchOp.isFinished() ); - - BatchedCommandResponse clientResponse; - batchOp.buildClientResponse( &clientResponse ); - ASSERT( clientResponse.getOk() ); - ASSERT_EQUALS( clientResponse.getN(), 2 ); - ASSERT( !clientResponse.isErrDetailsSet() ); - ASSERT( clientResponse.isWriteConcernErrorSet() ); - } + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); +} +TEST(WriteOpLimitTests, TooManyOps) { // - // Tests of batch size limit functionality + // Batch of 1002 documents // - TEST(WriteOpLimitTests, OneBigDoc) { + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); - // - // Big single operation test - should go through - // + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); - NamespaceString nss("foo.bar"); - ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); - MockNSTargeter targeter; - initTargeterFullRange(nss, endpoint, &targeter); - - // Create a BSONObj (slightly) bigger than the maximum size by including a max-size string - string bigString(BSONObjMaxUserSize, 'x'); - - // Do single-target, single doc batch write op - BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); - request.setNS(nss.ns()); - request.getInsertRequest()->addToDocuments(BSON( "x" << 1 << "data" << bigString )); - - BatchWriteOp batchOp; - batchOp.initClientRequest(&request); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch(targeter, false, &targeted); - ASSERT(status.isOK()); - ASSERT_EQUALS(targeted.size(), 1u); - - BatchedCommandResponse response; - buildResponse(1, &response); - - batchOp.noteBatchResponse(*targeted.front(), response, NULL); - ASSERT(batchOp.isFinished()); + // Add 2 more than the maximum to the batch + for (size_t i = 0; i < BatchedCommandRequest::kMaxWriteBatchSize + 2u; ++i) { + request.getDeleteRequest()->addToDeletes(buildDelete(BSON("x" << 2), 0)); } - TEST(WriteOpLimitTests, OneBigOneSmall) { - - // - // Big doc with smaller additional doc - should go through as two batches - // - - NamespaceString nss("foo.bar"); - ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); - MockNSTargeter targeter; - initTargeterFullRange(nss, endpoint, &targeter); - - // Create a BSONObj (slightly) bigger than the maximum size by including a max-size string - string bigString(BSONObjMaxUserSize, 'x'); - - BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); - request.setNS(nss.ns()); - BatchedUpdateDocument* bigUpdateDoc = buildUpdate(BSON( "x" << 1 ), - BSON( "data" << bigString ), - false); - request.getUpdateRequest()->addToUpdates(bigUpdateDoc); - request.getUpdateRequest()->addToUpdates(buildUpdate(BSON( "x" << 2 ), - BSONObj(), - false)); - - BatchWriteOp batchOp; - batchOp.initClientRequest(&request); - - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch(targeter, false, &targeted); - ASSERT(status.isOK()); - ASSERT_EQUALS(targeted.size(), 1u); - ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); - - BatchedCommandResponse response; - buildResponse(1, &response); - - batchOp.noteBatchResponse(*targeted.front(), response, NULL); - ASSERT(!batchOp.isFinished()); - - targetedOwned.clear(); - status = batchOp.targetBatch(targeter, false, &targeted); - ASSERT(status.isOK()); - ASSERT_EQUALS(targeted.size(), 1u); - ASSERT_EQUALS(targeted.front()->getWrites().size(), 1u); - - batchOp.noteBatchResponse(*targeted.front(), response, NULL); - ASSERT(batchOp.isFinished()); - } + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); - TEST(WriteOpLimitTests, TooManyOps) { + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 1000u); - // - // Batch of 1002 documents - // + BatchedCommandResponse response; + buildResponse(1, &response); - NamespaceString nss("foo.bar"); - ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); - MockNSTargeter targeter; - initTargeterFullRange(nss, endpoint, &targeter); + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); - BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); - request.setNS(nss.ns()); + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); - // Add 2 more than the maximum to the batch - for (size_t i = 0; i < BatchedCommandRequest::kMaxWriteBatchSize + 2u; ++i) { - request.getDeleteRequest()->addToDeletes(buildDelete(BSON( "x" << 2 ), 0)); - } + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); +} - BatchWriteOp batchOp; - batchOp.initClientRequest(&request); +TEST(WriteOpLimitTests, UpdateOverheadIncluded) { + // + // Tests that the overhead of the extra fields in an update x 1000 is included in our size + // calculation + // - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch(targeter, false, &targeted); - ASSERT(status.isOK()); - ASSERT_EQUALS(targeted.size(), 1u); - ASSERT_EQUALS(targeted.front()->getWrites().size(), 1000u); + NamespaceString nss("foo.bar"); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); + MockNSTargeter targeter; + initTargeterFullRange(nss, endpoint, &targeter); - BatchedCommandResponse response; - buildResponse(1, &response); + int updateDataBytes = + BSONObjMaxUserSize / static_cast<int>(BatchedCommandRequest::kMaxWriteBatchSize); - batchOp.noteBatchResponse(*targeted.front(), response, NULL); - ASSERT(!batchOp.isFinished()); + string dataString(updateDataBytes - + BSON("x" << 1 << "data" + << "").objsize(), + 'x'); - targetedOwned.clear(); - status = batchOp.targetBatch(targeter, false, &targeted); - ASSERT(status.isOK()); - ASSERT_EQUALS(targeted.size(), 1u); - ASSERT_EQUALS(targeted.front()->getWrites().size(), 2u); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); + request.setNS(nss.ns()); - batchOp.noteBatchResponse(*targeted.front(), response, NULL); - ASSERT(batchOp.isFinished()); + // Add the maximum number of updates + int estSizeBytes = 0; + for (size_t i = 0; i < BatchedCommandRequest::kMaxWriteBatchSize; ++i) { + BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; + updateDoc->setQuery(BSON("x" << 1 << "data" << dataString)); + updateDoc->setUpdateExpr(BSONObj()); + updateDoc->setMulti(false); + updateDoc->setUpsert(false); + request.getUpdateRequest()->addToUpdates(updateDoc); + estSizeBytes += updateDoc->toBSON().objsize(); } - TEST(WriteOpLimitTests, UpdateOverheadIncluded) { - - // - // Tests that the overhead of the extra fields in an update x 1000 is included in our size - // calculation - // - - NamespaceString nss("foo.bar"); - ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); - MockNSTargeter targeter; - initTargeterFullRange(nss, endpoint, &targeter); - - int updateDataBytes = BSONObjMaxUserSize - / static_cast<int>(BatchedCommandRequest::kMaxWriteBatchSize); - - string dataString(updateDataBytes - BSON( "x" << 1 << "data" << "" ).objsize(), 'x'); + ASSERT_GREATER_THAN(estSizeBytes, BSONObjMaxInternalSize); - BatchedCommandRequest request(BatchedCommandRequest::BatchType_Update); - request.setNS(nss.ns()); + BatchWriteOp batchOp; + batchOp.initClientRequest(&request); - // Add the maximum number of updates - int estSizeBytes = 0; - for (size_t i = 0; i < BatchedCommandRequest::kMaxWriteBatchSize; ++i) { - BatchedUpdateDocument* updateDoc = new BatchedUpdateDocument; - updateDoc->setQuery(BSON( "x" << 1 << "data" << dataString )); - updateDoc->setUpdateExpr(BSONObj()); - updateDoc->setMulti(false); - updateDoc->setUpsert(false); - request.getUpdateRequest()->addToUpdates(updateDoc); - estSizeBytes += updateDoc->toBSON().objsize(); - } + OwnedPointerVector<TargetedWriteBatch> targetedOwned; + vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); + Status status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_LESS_THAN(targeted.front()->getWrites().size(), 1000u); - ASSERT_GREATER_THAN(estSizeBytes, BSONObjMaxInternalSize); + BatchedCommandRequest childRequest(BatchedCommandRequest::BatchType_Update); + batchOp.buildBatchRequest(*targeted.front(), &childRequest); + ASSERT_LESS_THAN(childRequest.toBSON().objsize(), BSONObjMaxInternalSize); - BatchWriteOp batchOp; - batchOp.initClientRequest(&request); + BatchedCommandResponse response; + buildResponse(1, &response); - OwnedPointerVector<TargetedWriteBatch> targetedOwned; - vector<TargetedWriteBatch*>& targeted = targetedOwned.mutableVector(); - Status status = batchOp.targetBatch(targeter, false, &targeted); - ASSERT(status.isOK()); - ASSERT_EQUALS(targeted.size(), 1u); - ASSERT_LESS_THAN(targeted.front()->getWrites().size(), 1000u); + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(!batchOp.isFinished()); - BatchedCommandRequest childRequest(BatchedCommandRequest::BatchType_Update); - batchOp.buildBatchRequest(*targeted.front(), &childRequest); - ASSERT_LESS_THAN(childRequest.toBSON().objsize(), BSONObjMaxInternalSize); + targetedOwned.clear(); + status = batchOp.targetBatch(targeter, false, &targeted); + ASSERT(status.isOK()); + ASSERT_EQUALS(targeted.size(), 1u); + ASSERT_LESS_THAN(targeted.front()->getWrites().size(), 1000u); - BatchedCommandResponse response; - buildResponse(1, &response); + childRequest.clear(); + batchOp.buildBatchRequest(*targeted.front(), &childRequest); + ASSERT_LESS_THAN(childRequest.toBSON().objsize(), BSONObjMaxInternalSize); - batchOp.noteBatchResponse(*targeted.front(), response, NULL); - ASSERT(!batchOp.isFinished()); - - targetedOwned.clear(); - status = batchOp.targetBatch(targeter, false, &targeted); - ASSERT(status.isOK()); - ASSERT_EQUALS(targeted.size(), 1u); - ASSERT_LESS_THAN(targeted.front()->getWrites().size(), 1000u); - - childRequest.clear(); - batchOp.buildBatchRequest(*targeted.front(), &childRequest); - ASSERT_LESS_THAN(childRequest.toBSON().objsize(), BSONObjMaxInternalSize); - - batchOp.noteBatchResponse(*targeted.front(), response, NULL); - ASSERT(batchOp.isFinished()); - } + batchOp.noteBatchResponse(*targeted.front(), response, NULL); + ASSERT(batchOp.isFinished()); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/write_ops/batched_command_request.cpp b/src/mongo/s/write_ops/batched_command_request.cpp index b3f4d9911bb..99ef66e767d 100644 --- a/src/mongo/s/write_ops/batched_command_request.cpp +++ b/src/mongo/s/write_ops/batched_command_request.cpp @@ -33,368 +33,364 @@ namespace mongo { - using std::unique_ptr; - using std::string; - using std::vector; +using std::unique_ptr; +using std::string; +using std::vector; - const size_t BatchedCommandRequest::kMaxWriteBatchSize = 1000; +const size_t BatchedCommandRequest::kMaxWriteBatchSize = 1000; - BatchedCommandRequest::BatchedCommandRequest( BatchType batchType ) : - _batchType( batchType ) { - switch ( getBatchType() ) { +BatchedCommandRequest::BatchedCommandRequest(BatchType batchType) : _batchType(batchType) { + switch (getBatchType()) { case BatchedCommandRequest::BatchType_Insert: - _insertReq.reset( new BatchedInsertRequest ); + _insertReq.reset(new BatchedInsertRequest); return; case BatchedCommandRequest::BatchType_Update: - _updateReq.reset( new BatchedUpdateRequest ); + _updateReq.reset(new BatchedUpdateRequest); return; default: - dassert( getBatchType() == BatchedCommandRequest::BatchType_Delete ); - _deleteReq.reset( new BatchedDeleteRequest ); + dassert(getBatchType() == BatchedCommandRequest::BatchType_Delete); + _deleteReq.reset(new BatchedDeleteRequest); return; - } } +} // This macro just invokes a given method on one of the three types of ops with parameters -#define INVOKE(M,...) \ -{\ - switch ( getBatchType() ) {\ - case BatchedCommandRequest::BatchType_Insert:\ - return _insertReq->M(__VA_ARGS__);\ - case BatchedCommandRequest::BatchType_Update:\ - return _updateReq->M(__VA_ARGS__);\ - default:\ - dassert( getBatchType() == BatchedCommandRequest::BatchType_Delete );\ - return _deleteReq->M(__VA_ARGS__);\ - }\ -} - - BatchedCommandRequest::BatchType BatchedCommandRequest::getBatchType() const { - return _batchType; - } +#define INVOKE(M, ...) \ + { \ + switch (getBatchType()) { \ + case BatchedCommandRequest::BatchType_Insert: \ + return _insertReq->M(__VA_ARGS__); \ + case BatchedCommandRequest::BatchType_Update: \ + return _updateReq->M(__VA_ARGS__); \ + default: \ + dassert(getBatchType() == BatchedCommandRequest::BatchType_Delete); \ + return _deleteReq->M(__VA_ARGS__); \ + } \ + } + +BatchedCommandRequest::BatchType BatchedCommandRequest::getBatchType() const { + return _batchType; +} - BatchedInsertRequest* BatchedCommandRequest::getInsertRequest() const { - return _insertReq.get(); - } +BatchedInsertRequest* BatchedCommandRequest::getInsertRequest() const { + return _insertReq.get(); +} - BatchedUpdateRequest* BatchedCommandRequest::getUpdateRequest() const { - return _updateReq.get(); - } +BatchedUpdateRequest* BatchedCommandRequest::getUpdateRequest() const { + return _updateReq.get(); +} - BatchedDeleteRequest* BatchedCommandRequest::getDeleteRequest() const { - return _deleteReq.get(); - } +BatchedDeleteRequest* BatchedCommandRequest::getDeleteRequest() const { + return _deleteReq.get(); +} + +bool BatchedCommandRequest::isInsertIndexRequest() const { + if (_batchType != BatchedCommandRequest::BatchType_Insert) + return false; + return getNSS().isSystemDotIndexes(); +} - bool BatchedCommandRequest::isInsertIndexRequest() const { - if ( _batchType != BatchedCommandRequest::BatchType_Insert ) return false; - return getNSS().isSystemDotIndexes(); +static bool extractUniqueIndex(const BSONObj& indexDesc) { + return indexDesc["unique"].trueValue(); +} + +bool BatchedCommandRequest::isUniqueIndexRequest() const { + if (!isInsertIndexRequest()) + return false; + return extractUniqueIndex(getInsertRequest()->getDocumentsAt(0)); +} + +bool BatchedCommandRequest::isValidIndexRequest(string* errMsg) const { + string dummy; + if (!errMsg) + errMsg = &dummy; + dassert(isInsertIndexRequest()); + + if (sizeWriteOps() != 1) { + *errMsg = "invalid batch request for index creation"; + return false; } - static bool extractUniqueIndex( const BSONObj& indexDesc ) { - return indexDesc["unique"].trueValue(); + const NamespaceString& targetNSS = getTargetingNSS(); + if (!targetNSS.isValid()) { + *errMsg = targetNSS.ns() + " is not a valid namespace to index"; + return false; } - bool BatchedCommandRequest::isUniqueIndexRequest() const { - if ( !isInsertIndexRequest() ) return false; - return extractUniqueIndex( getInsertRequest()->getDocumentsAt( 0 ) ); + const NamespaceString& reqNSS = getNSS(); + if (reqNSS.db().compare(targetNSS.db()) != 0) { + *errMsg = + targetNSS.ns() + " namespace is not in the request database " + reqNSS.db().toString(); + return false; } - bool BatchedCommandRequest::isValidIndexRequest( string* errMsg ) const { + return true; +} - string dummy; - if ( !errMsg ) - errMsg = &dummy; - dassert( isInsertIndexRequest() ); +string BatchedCommandRequest::getTargetingNS() const { + return getTargetingNSS().toString(); +} - if ( sizeWriteOps() != 1 ) { - *errMsg = "invalid batch request for index creation"; - return false; - } +const NamespaceString& BatchedCommandRequest::getTargetingNSS() const { + if (!isInsertIndexRequest()) + return getNSS(); + INVOKE(getTargetingNSS); +} - const NamespaceString& targetNSS = getTargetingNSS(); - if ( !targetNSS.isValid() ) { - *errMsg = targetNSS.ns() + " is not a valid namespace to index"; - return false; - } +static BSONObj extractIndexKeyPattern(const BSONObj& indexDesc) { + return indexDesc["key"].Obj(); +} - const NamespaceString& reqNSS = getNSS(); - if ( reqNSS.db().compare( targetNSS.db() ) != 0 ) { - *errMsg = targetNSS.ns() + " namespace is not in the request database " - + reqNSS.db().toString(); - return false; - } +BSONObj BatchedCommandRequest::getIndexKeyPattern() const { + dassert(isInsertIndexRequest()); + return extractIndexKeyPattern(getInsertRequest()->getDocumentsAt(0)); +} +bool BatchedCommandRequest::isVerboseWC() const { + if (!isWriteConcernSet()) { return true; } - string BatchedCommandRequest::getTargetingNS() const { - return getTargetingNSS().toString(); - } - - const NamespaceString& BatchedCommandRequest::getTargetingNSS() const { - if ( !isInsertIndexRequest() ) return getNSS(); - INVOKE(getTargetingNSS); - } - - static BSONObj extractIndexKeyPattern( const BSONObj& indexDesc ) { - return indexDesc["key"].Obj(); - } - - BSONObj BatchedCommandRequest::getIndexKeyPattern() const { - dassert( isInsertIndexRequest() ); - return extractIndexKeyPattern( getInsertRequest()->getDocumentsAt( 0 ) ); + BSONObj writeConcern = getWriteConcern(); + BSONElement wElem = writeConcern["w"]; + if (!wElem.isNumber() || wElem.Number() != 0) { + return true; } - bool BatchedCommandRequest::isVerboseWC() const { - if ( !isWriteConcernSet() ) { - return true; - } - - BSONObj writeConcern = getWriteConcern(); - BSONElement wElem = writeConcern["w"]; - if ( !wElem.isNumber() || wElem.Number() != 0 ) { - return true; - } - - return false; - } + return false; +} - void BatchedCommandRequest::cloneTo( BatchedCommandRequest* other ) const { - other->_insertReq.reset(); - other->_updateReq.reset(); - other->_deleteReq.reset(); - other->_batchType = _batchType; +void BatchedCommandRequest::cloneTo(BatchedCommandRequest* other) const { + other->_insertReq.reset(); + other->_updateReq.reset(); + other->_deleteReq.reset(); + other->_batchType = _batchType; - switch ( getBatchType() ) { + switch (getBatchType()) { case BatchedCommandRequest::BatchType_Insert: - other->_insertReq.reset( new BatchedInsertRequest ); - _insertReq->cloneTo( other->_insertReq.get() ); + other->_insertReq.reset(new BatchedInsertRequest); + _insertReq->cloneTo(other->_insertReq.get()); return; case BatchedCommandRequest::BatchType_Update: - other->_updateReq.reset( new BatchedUpdateRequest ); - _updateReq->cloneTo( other->_updateReq.get() ); + other->_updateReq.reset(new BatchedUpdateRequest); + _updateReq->cloneTo(other->_updateReq.get()); return; default: - dassert( getBatchType() == BatchedCommandRequest::BatchType_Delete ); - other->_deleteReq.reset( new BatchedDeleteRequest ); - _deleteReq->cloneTo( other->_deleteReq.get() ); + dassert(getBatchType() == BatchedCommandRequest::BatchType_Delete); + other->_deleteReq.reset(new BatchedDeleteRequest); + _deleteReq->cloneTo(other->_deleteReq.get()); return; - } } +} - bool BatchedCommandRequest::isValid( std::string* errMsg ) const { - INVOKE( isValid, errMsg ); - } +bool BatchedCommandRequest::isValid(std::string* errMsg) const { + INVOKE(isValid, errMsg); +} - BSONObj BatchedCommandRequest::toBSON() const { - INVOKE( toBSON ); - } +BSONObj BatchedCommandRequest::toBSON() const { + INVOKE(toBSON); +} - bool BatchedCommandRequest::parseBSON( const BSONObj& source, std::string* errMsg ) { - INVOKE( parseBSON, source, errMsg ); - } +bool BatchedCommandRequest::parseBSON(const BSONObj& source, std::string* errMsg) { + INVOKE(parseBSON, source, errMsg); +} - void BatchedCommandRequest::clear() { - INVOKE( clear ); - } +void BatchedCommandRequest::clear() { + INVOKE(clear); +} - std::string BatchedCommandRequest::toString() const { - INVOKE( toString ); - } +std::string BatchedCommandRequest::toString() const { + INVOKE(toString); +} - void BatchedCommandRequest::setNSS( const NamespaceString& nss ) { - INVOKE( setCollNameNS, nss ); - } +void BatchedCommandRequest::setNSS(const NamespaceString& nss) { + INVOKE(setCollNameNS, nss); +} - void BatchedCommandRequest::setNS( StringData collName ) { - INVOKE( setCollName, collName ); - } +void BatchedCommandRequest::setNS(StringData collName) { + INVOKE(setCollName, collName); +} - const std::string& BatchedCommandRequest::getNS() const { - INVOKE( getCollName ); - } +const std::string& BatchedCommandRequest::getNS() const { + INVOKE(getCollName); +} - const NamespaceString& BatchedCommandRequest::getNSS() const { - INVOKE(getCollNameNS); - } +const NamespaceString& BatchedCommandRequest::getNSS() const { + INVOKE(getCollNameNS); +} - std::size_t BatchedCommandRequest::sizeWriteOps() const { - switch ( getBatchType() ) { +std::size_t BatchedCommandRequest::sizeWriteOps() const { + switch (getBatchType()) { case BatchedCommandRequest::BatchType_Insert: return _insertReq->sizeDocuments(); case BatchedCommandRequest::BatchType_Update: return _updateReq->sizeUpdates(); default: return _deleteReq->sizeDeletes(); - } } +} - void BatchedCommandRequest::setWriteConcern( const BSONObj& writeConcern ) { - INVOKE( setWriteConcern, writeConcern ); - } +void BatchedCommandRequest::setWriteConcern(const BSONObj& writeConcern) { + INVOKE(setWriteConcern, writeConcern); +} - void BatchedCommandRequest::unsetWriteConcern() { - INVOKE( unsetWriteConcern ); - } +void BatchedCommandRequest::unsetWriteConcern() { + INVOKE(unsetWriteConcern); +} - bool BatchedCommandRequest::isWriteConcernSet() const { - INVOKE( isWriteConcernSet ); - } +bool BatchedCommandRequest::isWriteConcernSet() const { + INVOKE(isWriteConcernSet); +} - const BSONObj& BatchedCommandRequest::getWriteConcern() const { - INVOKE( getWriteConcern ); - } +const BSONObj& BatchedCommandRequest::getWriteConcern() const { + INVOKE(getWriteConcern); +} - void BatchedCommandRequest::setOrdered( bool continueOnError ) { - INVOKE( setOrdered, continueOnError ); - } +void BatchedCommandRequest::setOrdered(bool continueOnError) { + INVOKE(setOrdered, continueOnError); +} - void BatchedCommandRequest::unsetOrdered() { - INVOKE( unsetOrdered ); - } +void BatchedCommandRequest::unsetOrdered() { + INVOKE(unsetOrdered); +} - bool BatchedCommandRequest::isOrderedSet() const { - INVOKE( isOrderedSet ); - } +bool BatchedCommandRequest::isOrderedSet() const { + INVOKE(isOrderedSet); +} - bool BatchedCommandRequest::getOrdered() const { - INVOKE( getOrdered ); - } +bool BatchedCommandRequest::getOrdered() const { + INVOKE(getOrdered); +} - void BatchedCommandRequest::setMetadata(BatchedRequestMetadata* metadata) { - INVOKE( setMetadata, metadata ); - } +void BatchedCommandRequest::setMetadata(BatchedRequestMetadata* metadata) { + INVOKE(setMetadata, metadata); +} - void BatchedCommandRequest::unsetMetadata() { - INVOKE( unsetMetadata ); - } +void BatchedCommandRequest::unsetMetadata() { + INVOKE(unsetMetadata); +} - bool BatchedCommandRequest::isMetadataSet() const { - INVOKE( isMetadataSet ); - } +bool BatchedCommandRequest::isMetadataSet() const { + INVOKE(isMetadataSet); +} - BatchedRequestMetadata* BatchedCommandRequest::getMetadata() const { - INVOKE( getMetadata ); - } +BatchedRequestMetadata* BatchedCommandRequest::getMetadata() const { + INVOKE(getMetadata); +} - void BatchedCommandRequest::setShouldBypassValidation(bool newVal) { - INVOKE(setShouldBypassValidation, newVal); - } +void BatchedCommandRequest::setShouldBypassValidation(bool newVal) { + INVOKE(setShouldBypassValidation, newVal); +} - bool BatchedCommandRequest::shouldBypassValidation() const { - INVOKE(shouldBypassValidation); - } +bool BatchedCommandRequest::shouldBypassValidation() const { + INVOKE(shouldBypassValidation); +} - /** - * Generates a new request with insert _ids if required. Otherwise returns NULL. - */ - BatchedCommandRequest* // +/** + * Generates a new request with insert _ids if required. Otherwise returns NULL. + */ +BatchedCommandRequest* // BatchedCommandRequest::cloneWithIds(const BatchedCommandRequest& origCmdRequest) { + if (origCmdRequest.getBatchType() != BatchedCommandRequest::BatchType_Insert || + origCmdRequest.isInsertIndexRequest()) + return NULL; - if (origCmdRequest.getBatchType() != BatchedCommandRequest::BatchType_Insert - || origCmdRequest.isInsertIndexRequest()) - return NULL; - - unique_ptr<BatchedInsertRequest> idRequest; - BatchedInsertRequest* origRequest = origCmdRequest.getInsertRequest(); + unique_ptr<BatchedInsertRequest> idRequest; + BatchedInsertRequest* origRequest = origCmdRequest.getInsertRequest(); - const vector<BSONObj>& inserts = origRequest->getDocuments(); + const vector<BSONObj>& inserts = origRequest->getDocuments(); - size_t i = 0u; - for (vector<BSONObj>::const_iterator it = inserts.begin(); it != inserts.end(); ++it, ++i) { + size_t i = 0u; + for (vector<BSONObj>::const_iterator it = inserts.begin(); it != inserts.end(); ++it, ++i) { + const BSONObj& insert = *it; + BSONObj idInsert; - const BSONObj& insert = *it; - BSONObj idInsert; - - if (insert["_id"].eoo()) { - BSONObjBuilder idInsertB; - idInsertB.append("_id", OID::gen()); - idInsertB.appendElements(insert); - idInsert = idInsertB.obj(); - } - - if (NULL == idRequest.get() && !idInsert.isEmpty()) { - idRequest.reset(new BatchedInsertRequest); - origRequest->cloneTo(idRequest.get()); - } - - if (!idInsert.isEmpty()) { - idRequest->setDocumentAt(i, idInsert); - } + if (insert["_id"].eoo()) { + BSONObjBuilder idInsertB; + idInsertB.append("_id", OID::gen()); + idInsertB.appendElements(insert); + idInsert = idInsertB.obj(); } - if (NULL == idRequest.get()) - return NULL; + if (NULL == idRequest.get() && !idInsert.isEmpty()) { + idRequest.reset(new BatchedInsertRequest); + origRequest->cloneTo(idRequest.get()); + } - // Command request owns idRequest - return new BatchedCommandRequest(idRequest.release()); + if (!idInsert.isEmpty()) { + idRequest->setDocumentAt(i, idInsert); + } } - bool BatchedCommandRequest::containsNoIDUpsert(const BatchedCommandRequest& request) { + if (NULL == idRequest.get()) + return NULL; - if (request.getBatchType() != BatchedCommandRequest::BatchType_Update) - return false; - - const vector<BatchedUpdateDocument*>& updates = - request.getUpdateRequest()->getUpdates(); - - for (vector<BatchedUpdateDocument*>::const_iterator it = updates.begin(); - it != updates.end(); ++it) { - - const BatchedUpdateDocument* updateDoc = *it; - if (updateDoc->getUpsert() && updateDoc->getQuery()["_id"].eoo()) - return true; - } + // Command request owns idRequest + return new BatchedCommandRequest(idRequest.release()); +} +bool BatchedCommandRequest::containsNoIDUpsert(const BatchedCommandRequest& request) { + if (request.getBatchType() != BatchedCommandRequest::BatchType_Update) return false; - } - bool BatchedCommandRequest::containsUpserts( const BSONObj& writeCmdObj ) { + const vector<BatchedUpdateDocument*>& updates = request.getUpdateRequest()->getUpdates(); - BSONElement updatesEl = writeCmdObj[BatchedUpdateRequest::updates()]; - if ( updatesEl.type() != Array ) { - return false; - } + for (vector<BatchedUpdateDocument*>::const_iterator it = updates.begin(); it != updates.end(); + ++it) { + const BatchedUpdateDocument* updateDoc = *it; + if (updateDoc->getUpsert() && updateDoc->getQuery()["_id"].eoo()) + return true; + } - BSONObjIterator it( updatesEl.Obj() ); - while ( it.more() ) { - BSONElement updateEl = it.next(); - if ( !updateEl.isABSONObj() ) continue; - if ( updateEl.Obj()[BatchedUpdateDocument::upsert()].trueValue() ) return true; - } + return false; +} +bool BatchedCommandRequest::containsUpserts(const BSONObj& writeCmdObj) { + BSONElement updatesEl = writeCmdObj[BatchedUpdateRequest::updates()]; + if (updatesEl.type() != Array) { return false; } - bool BatchedCommandRequest::getIndexedNS( const BSONObj& writeCmdObj, - string* nsToIndex, - string* errMsg ) { + BSONObjIterator it(updatesEl.Obj()); + while (it.more()) { + BSONElement updateEl = it.next(); + if (!updateEl.isABSONObj()) + continue; + if (updateEl.Obj()[BatchedUpdateDocument::upsert()].trueValue()) + return true; + } - BSONElement documentsEl = writeCmdObj[BatchedInsertRequest::documents()]; - if ( documentsEl.type() != Array ) { - *errMsg = "index write batch is invalid"; - return false; - } + return false; +} - BSONObjIterator it( documentsEl.Obj() ); - if ( !it.more() ) { - *errMsg = "index write batch is empty"; - return false; - } +bool BatchedCommandRequest::getIndexedNS(const BSONObj& writeCmdObj, + string* nsToIndex, + string* errMsg) { + BSONElement documentsEl = writeCmdObj[BatchedInsertRequest::documents()]; + if (documentsEl.type() != Array) { + *errMsg = "index write batch is invalid"; + return false; + } - BSONElement indexDescEl = it.next(); - *nsToIndex = indexDescEl["ns"].str(); - if ( *nsToIndex == "" ) { - *errMsg = "index write batch contains an invalid index descriptor"; - return false; - } + BSONObjIterator it(documentsEl.Obj()); + if (!it.more()) { + *errMsg = "index write batch is empty"; + return false; + } - if ( it.more() ) { - *errMsg = "index write batches may only contain a single index descriptor"; - return false; - } + BSONElement indexDescEl = it.next(); + *nsToIndex = indexDescEl["ns"].str(); + if (*nsToIndex == "") { + *errMsg = "index write batch contains an invalid index descriptor"; + return false; + } - return true; + if (it.more()) { + *errMsg = "index write batches may only contain a single index descriptor"; + return false; } -} // namespace mongo + return true; +} + +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_command_request.h b/src/mongo/s/write_ops/batched_command_request.h index 66e178dd5a8..b8e6e2f8e80 100644 --- a/src/mongo/s/write_ops/batched_command_request.h +++ b/src/mongo/s/write_ops/batched_command_request.h @@ -37,213 +37,204 @@ namespace mongo { - class NamespaceString; +class NamespaceString; - /** - * This class wraps the different kinds of command requests into a generically usable write - * command request. - * - * Designed to be a very thin wrapper that mimics the underlying requests exactly. Owns the - * wrapped request object once constructed. - */ - class BatchedCommandRequest : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedCommandRequest); - public: - - // Maximum number of write ops supported per batch - static const size_t kMaxWriteBatchSize; - - enum BatchType { - BatchType_Insert, BatchType_Update, BatchType_Delete, BatchType_Unknown - }; +/** + * This class wraps the different kinds of command requests into a generically usable write + * command request. + * + * Designed to be a very thin wrapper that mimics the underlying requests exactly. Owns the + * wrapped request object once constructed. + */ +class BatchedCommandRequest : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedCommandRequest); - // - // construction / destruction - // +public: + // Maximum number of write ops supported per batch + static const size_t kMaxWriteBatchSize; - BatchedCommandRequest( BatchType batchType ); + enum BatchType { BatchType_Insert, BatchType_Update, BatchType_Delete, BatchType_Unknown }; - /** - * insertReq ownership is transferred to here. - */ - BatchedCommandRequest( BatchedInsertRequest* insertReq ) : - _batchType( BatchType_Insert ), _insertReq( insertReq ) { - } + // + // construction / destruction + // - /** - * updateReq ownership is transferred to here. - */ - BatchedCommandRequest( BatchedUpdateRequest* updateReq ) : - _batchType( BatchType_Update ), _updateReq( updateReq ) { - } + BatchedCommandRequest(BatchType batchType); - /** - * deleteReq ownership is transferred to here. - */ - BatchedCommandRequest( BatchedDeleteRequest* deleteReq ) : - _batchType( BatchType_Delete ), _deleteReq( deleteReq ) { - } - - virtual ~BatchedCommandRequest() {}; - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo( BatchedCommandRequest* other ) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid( std::string* errMsg ) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON( const BSONObj& source, std::string* errMsg ); - virtual void clear(); - virtual std::string toString() const; - - // - // Batch type accessors - // - - BatchType getBatchType() const; - BatchedInsertRequest* getInsertRequest() const; - BatchedUpdateRequest* getUpdateRequest() const; - BatchedDeleteRequest* getDeleteRequest() const; - // Index creation is also an insert, but a weird one. - bool isInsertIndexRequest() const; - bool isUniqueIndexRequest() const; - bool isValidIndexRequest( std::string* errMsg ) const; - std::string getTargetingNS() const; - const NamespaceString& getTargetingNSS() const; - BSONObj getIndexKeyPattern() const; - - // - // individual field accessors - // - - bool isVerboseWC() const; - - void setNSS( const NamespaceString& nss ); - void setNS( StringData collName ); - const std::string& getNS() const; - const NamespaceString& getNSS() const; - - std::size_t sizeWriteOps() const; - - void setWriteConcern( const BSONObj& writeConcern ); - void unsetWriteConcern(); - bool isWriteConcernSet() const; - const BSONObj& getWriteConcern() const; - - void setOrdered( bool ordered ); - void unsetOrdered(); - bool isOrderedSet() const; - bool getOrdered() const; - - void setMetadata(BatchedRequestMetadata* metadata); - void unsetMetadata(); - bool isMetadataSet() const; - BatchedRequestMetadata* getMetadata() const; - - void setShouldBypassValidation(bool newVal); - bool shouldBypassValidation() const; - - // - // Helpers for batch pre-processing - // - - /** - * Generates a new request, the same as the old, but with insert _ids if required. - * Returns NULL if this is not an insert request or all inserts already have _ids. - */ - static BatchedCommandRequest* cloneWithIds(const BatchedCommandRequest& origCmdRequest); - - /** - * Whether or not this batch contains an upsert without an _id - these can't be sent - * to multiple hosts. - */ - static bool containsNoIDUpsert(const BatchedCommandRequest& request); - - // - // Helpers for auth pre-parsing - // - - /** - * Helper to determine whether or not there are any upserts in the batch - */ - static bool containsUpserts( const BSONObj& writeCmdObj ); - - /** - * Helper to extract the namespace being indexed from a raw BSON write command. - * - * Returns false with errMsg if the index write command seems invalid. - * TODO: Remove when we have parsing hooked before authorization - */ - static bool getIndexedNS( const BSONObj& writeCmdObj, - std::string* nsToIndex, - std::string* errMsg ); - - private: - - BatchType _batchType; - std::unique_ptr<BatchedInsertRequest> _insertReq; - std::unique_ptr<BatchedUpdateRequest> _updateReq; - std::unique_ptr<BatchedDeleteRequest> _deleteReq; - }; + /** + * insertReq ownership is transferred to here. + */ + BatchedCommandRequest(BatchedInsertRequest* insertReq) + : _batchType(BatchType_Insert), _insertReq(insertReq) {} /** - * Similar to above, this class wraps the write items of a command request into a generically - * usable type. Very thin wrapper, does not own the write item itself. - * - * TODO: Use in BatchedCommandRequest above + * updateReq ownership is transferred to here. */ - class BatchItemRef { - public: + BatchedCommandRequest(BatchedUpdateRequest* updateReq) + : _batchType(BatchType_Update), _updateReq(updateReq) {} - BatchItemRef( const BatchedCommandRequest* request, int itemIndex ) : - _request( request ), _itemIndex( itemIndex ) { - } + /** + * deleteReq ownership is transferred to here. + */ + BatchedCommandRequest(BatchedDeleteRequest* deleteReq) + : _batchType(BatchType_Delete), _deleteReq(deleteReq) {} + + virtual ~BatchedCommandRequest(){}; + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(BatchedCommandRequest* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // Batch type accessors + // + + BatchType getBatchType() const; + BatchedInsertRequest* getInsertRequest() const; + BatchedUpdateRequest* getUpdateRequest() const; + BatchedDeleteRequest* getDeleteRequest() const; + // Index creation is also an insert, but a weird one. + bool isInsertIndexRequest() const; + bool isUniqueIndexRequest() const; + bool isValidIndexRequest(std::string* errMsg) const; + std::string getTargetingNS() const; + const NamespaceString& getTargetingNSS() const; + BSONObj getIndexKeyPattern() const; + + // + // individual field accessors + // + + bool isVerboseWC() const; + + void setNSS(const NamespaceString& nss); + void setNS(StringData collName); + const std::string& getNS() const; + const NamespaceString& getNSS() const; + + std::size_t sizeWriteOps() const; + + void setWriteConcern(const BSONObj& writeConcern); + void unsetWriteConcern(); + bool isWriteConcernSet() const; + const BSONObj& getWriteConcern() const; + + void setOrdered(bool ordered); + void unsetOrdered(); + bool isOrderedSet() const; + bool getOrdered() const; + + void setMetadata(BatchedRequestMetadata* metadata); + void unsetMetadata(); + bool isMetadataSet() const; + BatchedRequestMetadata* getMetadata() const; + + void setShouldBypassValidation(bool newVal); + bool shouldBypassValidation() const; + + // + // Helpers for batch pre-processing + // - const BatchedCommandRequest* getRequest() const { - return _request; - } + /** + * Generates a new request, the same as the old, but with insert _ids if required. + * Returns NULL if this is not an insert request or all inserts already have _ids. + */ + static BatchedCommandRequest* cloneWithIds(const BatchedCommandRequest& origCmdRequest); - int getItemIndex() const { - return _itemIndex; - } + /** + * Whether or not this batch contains an upsert without an _id - these can't be sent + * to multiple hosts. + */ + static bool containsNoIDUpsert(const BatchedCommandRequest& request); - BatchedCommandRequest::BatchType getOpType() const { - return _request->getBatchType(); - } + // + // Helpers for auth pre-parsing + // - const BSONObj& getDocument() const { - dassert( _itemIndex < static_cast<int>( _request->sizeWriteOps() ) ); - return _request->getInsertRequest()->getDocumentsAt( _itemIndex ); - } + /** + * Helper to determine whether or not there are any upserts in the batch + */ + static bool containsUpserts(const BSONObj& writeCmdObj); - const BatchedUpdateDocument* getUpdate() const { - dassert( _itemIndex < static_cast<int>( _request->sizeWriteOps() ) ); - return _request->getUpdateRequest()->getUpdatesAt( _itemIndex ); - } + /** + * Helper to extract the namespace being indexed from a raw BSON write command. + * + * Returns false with errMsg if the index write command seems invalid. + * TODO: Remove when we have parsing hooked before authorization + */ + static bool getIndexedNS(const BSONObj& writeCmdObj, + std::string* nsToIndex, + std::string* errMsg); - const BatchedDeleteDocument* getDelete() const { - dassert( _itemIndex < static_cast<int>( _request->sizeWriteOps() ) ); - return _request->getDeleteRequest()->getDeletesAt( _itemIndex ); - } +private: + BatchType _batchType; + std::unique_ptr<BatchedInsertRequest> _insertReq; + std::unique_ptr<BatchedUpdateRequest> _updateReq; + std::unique_ptr<BatchedDeleteRequest> _deleteReq; +}; - BSONObj toBSON() const { - switch ( getOpType() ) { +/** + * Similar to above, this class wraps the write items of a command request into a generically + * usable type. Very thin wrapper, does not own the write item itself. + * + * TODO: Use in BatchedCommandRequest above + */ +class BatchItemRef { +public: + BatchItemRef(const BatchedCommandRequest* request, int itemIndex) + : _request(request), _itemIndex(itemIndex) {} + + const BatchedCommandRequest* getRequest() const { + return _request; + } + + int getItemIndex() const { + return _itemIndex; + } + + BatchedCommandRequest::BatchType getOpType() const { + return _request->getBatchType(); + } + + const BSONObj& getDocument() const { + dassert(_itemIndex < static_cast<int>(_request->sizeWriteOps())); + return _request->getInsertRequest()->getDocumentsAt(_itemIndex); + } + + const BatchedUpdateDocument* getUpdate() const { + dassert(_itemIndex < static_cast<int>(_request->sizeWriteOps())); + return _request->getUpdateRequest()->getUpdatesAt(_itemIndex); + } + + const BatchedDeleteDocument* getDelete() const { + dassert(_itemIndex < static_cast<int>(_request->sizeWriteOps())); + return _request->getDeleteRequest()->getDeletesAt(_itemIndex); + } + + BSONObj toBSON() const { + switch (getOpType()) { case BatchedCommandRequest::BatchType_Insert: return getDocument(); case BatchedCommandRequest::BatchType_Update: return getUpdate()->toBSON(); default: return getDelete()->toBSON(); - } } + } - private: - - const BatchedCommandRequest* _request; - const int _itemIndex; - }; +private: + const BatchedCommandRequest* _request; + const int _itemIndex; +}; -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_command_response.cpp b/src/mongo/s/write_ops/batched_command_response.cpp index cce5fa660cc..4f137d1dfe6 100644 --- a/src/mongo/s/write_ops/batched_command_response.cpp +++ b/src/mongo/s/write_ops/batched_command_response.cpp @@ -33,524 +33,537 @@ namespace mongo { - using std::unique_ptr; - using std::string; - - using mongoutils::str::stream; - - const BSONField<int> BatchedCommandResponse::ok("ok"); - const BSONField<int> BatchedCommandResponse::errCode("code", ErrorCodes::UnknownError); - const BSONField<string> BatchedCommandResponse::errMessage("errmsg"); - const BSONField<long long> BatchedCommandResponse::n("n", 0); - const BSONField<long long> BatchedCommandResponse::nModified("nModified", 0); - const BSONField<std::vector<BatchedUpsertDetail*> > - BatchedCommandResponse::upsertDetails("upserted"); - const BSONField<Timestamp> BatchedCommandResponse::lastOp("lastOp"); - const BSONField<OID> BatchedCommandResponse::electionId("electionId"); - const BSONField<std::vector<WriteErrorDetail*> > - BatchedCommandResponse::writeErrors("writeErrors"); - const BSONField<WCErrorDetail*> BatchedCommandResponse::writeConcernError("writeConcernError"); - - BatchedCommandResponse::BatchedCommandResponse() { - clear(); - } +using std::unique_ptr; +using std::string; - BatchedCommandResponse::~BatchedCommandResponse() { - unsetErrDetails(); - unsetUpsertDetails(); - } +using mongoutils::str::stream; - bool BatchedCommandResponse::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } +const BSONField<int> BatchedCommandResponse::ok("ok"); +const BSONField<int> BatchedCommandResponse::errCode("code", ErrorCodes::UnknownError); +const BSONField<string> BatchedCommandResponse::errMessage("errmsg"); +const BSONField<long long> BatchedCommandResponse::n("n", 0); +const BSONField<long long> BatchedCommandResponse::nModified("nModified", 0); +const BSONField<std::vector<BatchedUpsertDetail*>> BatchedCommandResponse::upsertDetails( + "upserted"); +const BSONField<Timestamp> BatchedCommandResponse::lastOp("lastOp"); +const BSONField<OID> BatchedCommandResponse::electionId("electionId"); +const BSONField<std::vector<WriteErrorDetail*>> BatchedCommandResponse::writeErrors("writeErrors"); +const BSONField<WCErrorDetail*> BatchedCommandResponse::writeConcernError("writeConcernError"); - // All the mandatory fields must be present. - if (!_isOkSet) { - *errMsg = stream() << "missing " << ok.name() << " field"; - return false; - } +BatchedCommandResponse::BatchedCommandResponse() { + clear(); +} + +BatchedCommandResponse::~BatchedCommandResponse() { + unsetErrDetails(); + unsetUpsertDetails(); +} - return true; +bool BatchedCommandResponse::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BSONObj BatchedCommandResponse::toBSON() const { - BSONObjBuilder builder; + // All the mandatory fields must be present. + if (!_isOkSet) { + *errMsg = stream() << "missing " << ok.name() << " field"; + return false; + } - if (_isOkSet) builder.append(ok(), _ok); + return true; +} - if (_isErrCodeSet) builder.append(errCode(), _errCode); +BSONObj BatchedCommandResponse::toBSON() const { + BSONObjBuilder builder; - if (_isErrMessageSet) builder.append(errMessage(), _errMessage); + if (_isOkSet) + builder.append(ok(), _ok); - if (_isNModifiedSet) builder.appendNumber(nModified(), _nModified); - if (_isNSet) builder.appendNumber(n(), _n); + if (_isErrCodeSet) + builder.append(errCode(), _errCode); - if (_upsertDetails.get()) { - BSONArrayBuilder upsertedBuilder(builder.subarrayStart(upsertDetails())); - for (std::vector<BatchedUpsertDetail*>::const_iterator it = _upsertDetails->begin(); - it != _upsertDetails->end(); - ++it) { - BSONObj upsertedDetailsDocument = (*it)->toBSON(); - upsertedBuilder.append(upsertedDetailsDocument); - } - upsertedBuilder.done(); - } + if (_isErrMessageSet) + builder.append(errMessage(), _errMessage); - if (_isLastOpSet) builder.append(lastOp(), _lastOp); - if (_isElectionIdSet) builder.appendOID(electionId(), const_cast<OID*>(&_electionId)); - - if (_writeErrorDetails.get()) { - BSONArrayBuilder errDetailsBuilder(builder.subarrayStart(writeErrors())); - for (std::vector<WriteErrorDetail*>::const_iterator it = _writeErrorDetails->begin(); - it != _writeErrorDetails->end(); - ++it) { - BSONObj errDetailsDocument = (*it)->toBSON(); - errDetailsBuilder.append(errDetailsDocument); - } - errDetailsBuilder.done(); - } + if (_isNModifiedSet) + builder.appendNumber(nModified(), _nModified); + if (_isNSet) + builder.appendNumber(n(), _n); - if (_wcErrDetails.get()) { - builder.append(writeConcernError(), _wcErrDetails->toBSON()); + if (_upsertDetails.get()) { + BSONArrayBuilder upsertedBuilder(builder.subarrayStart(upsertDetails())); + for (std::vector<BatchedUpsertDetail*>::const_iterator it = _upsertDetails->begin(); + it != _upsertDetails->end(); + ++it) { + BSONObj upsertedDetailsDocument = (*it)->toBSON(); + upsertedBuilder.append(upsertedDetailsDocument); } - - return builder.obj(); + upsertedBuilder.done(); } - bool BatchedCommandResponse::parseBSON(const BSONObj& source, string* errMsg) { - clear(); - - std::string dummy; - if (!errMsg) errMsg = &dummy; - - FieldParser::FieldState fieldState; - fieldState = FieldParser::extractNumber(source, ok, &_ok, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isOkSet = fieldState == FieldParser::FIELD_SET; - - fieldState = FieldParser::extract(source, errCode, &_errCode, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isErrCodeSet = fieldState == FieldParser::FIELD_SET; - - fieldState = FieldParser::extract(source, errMessage, &_errMessage, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isErrMessageSet = fieldState == FieldParser::FIELD_SET; - - // We're using appendNumber on generation so we'll try a smaller type - // (int) first and then fall back to the original type (long long). - BSONField<int> fieldN(n()); - int tempN; - fieldState = FieldParser::extract(source, fieldN, &tempN, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) { - // try falling back to a larger type - fieldState = FieldParser::extract(source, n, &_n, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isNSet = fieldState == FieldParser::FIELD_SET; - } - else if (fieldState == FieldParser::FIELD_SET) { - _isNSet = true; - _n = tempN; - } + if (_isLastOpSet) + builder.append(lastOp(), _lastOp); + if (_isElectionIdSet) + builder.appendOID(electionId(), const_cast<OID*>(&_electionId)); - // We're using appendNumber on generation so we'll try a smaller type - // (int) first and then fall back to the original type (long long). - BSONField<int> fieldNModified(nModified()); - int intNModified; - fieldState = FieldParser::extract(source, fieldNModified, &intNModified, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) { - // try falling back to a larger type - fieldState = FieldParser::extract(source, nModified, &_nModified, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isNModifiedSet = fieldState == FieldParser::FIELD_SET; - } - else if (fieldState == FieldParser::FIELD_SET) { - _isNModifiedSet = true; - _nModified = intNModified; + if (_writeErrorDetails.get()) { + BSONArrayBuilder errDetailsBuilder(builder.subarrayStart(writeErrors())); + for (std::vector<WriteErrorDetail*>::const_iterator it = _writeErrorDetails->begin(); + it != _writeErrorDetails->end(); + ++it) { + BSONObj errDetailsDocument = (*it)->toBSON(); + errDetailsBuilder.append(errDetailsDocument); } - - std::vector<BatchedUpsertDetail*>* tempUpsertDetails = NULL; - fieldState = FieldParser::extract( source, upsertDetails, &tempUpsertDetails, errMsg ); - if ( fieldState == FieldParser::FIELD_INVALID ) return false; - _upsertDetails.reset(tempUpsertDetails); - - fieldState = FieldParser::extract(source, lastOp, &_lastOp, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isLastOpSet = fieldState == FieldParser::FIELD_SET; - - fieldState = FieldParser::extract(source, electionId, &_electionId, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isElectionIdSet = fieldState == FieldParser::FIELD_SET; - - std::vector<WriteErrorDetail*>* tempErrDetails = NULL; - fieldState = FieldParser::extract(source, writeErrors, &tempErrDetails, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _writeErrorDetails.reset(tempErrDetails); - - WCErrorDetail* wcError = NULL; - fieldState = FieldParser::extract(source, writeConcernError, &wcError, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _wcErrDetails.reset(wcError); - - return true; + errDetailsBuilder.done(); } - void BatchedCommandResponse::clear() { - _ok = false; - _isOkSet = false; - - _errCode = 0; - _isErrCodeSet = false; - - _errMessage.clear(); - _isErrMessageSet = false; + if (_wcErrDetails.get()) { + builder.append(writeConcernError(), _wcErrDetails->toBSON()); + } - _nModified = 0; - _isNModifiedSet = false; + return builder.obj(); +} - _n = 0; - _isNSet = false; +bool BatchedCommandResponse::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - _singleUpserted = BSONObj(); - _isSingleUpsertedSet = false; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - if ( _upsertDetails.get() ) { - for ( std::vector<BatchedUpsertDetail*>::const_iterator it = _upsertDetails->begin(); - it != _upsertDetails->end(); ++it ) { - delete *it; - }; - _upsertDetails.reset(); - } + FieldParser::FieldState fieldState; + fieldState = FieldParser::extractNumber(source, ok, &_ok, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isOkSet = fieldState == FieldParser::FIELD_SET; - _lastOp = Timestamp(); - _isLastOpSet = false; + fieldState = FieldParser::extract(source, errCode, &_errCode, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isErrCodeSet = fieldState == FieldParser::FIELD_SET; - _electionId = OID(); - _isElectionIdSet = false; + fieldState = FieldParser::extract(source, errMessage, &_errMessage, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isErrMessageSet = fieldState == FieldParser::FIELD_SET; - if (_writeErrorDetails.get()) { - for(std::vector<WriteErrorDetail*>::const_iterator it = _writeErrorDetails->begin(); - it != _writeErrorDetails->end(); - ++it) { - delete *it; - }; - _writeErrorDetails.reset(); - } - - _wcErrDetails.reset(); + // We're using appendNumber on generation so we'll try a smaller type + // (int) first and then fall back to the original type (long long). + BSONField<int> fieldN(n()); + int tempN; + fieldState = FieldParser::extract(source, fieldN, &tempN, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) { + // try falling back to a larger type + fieldState = FieldParser::extract(source, n, &_n, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isNSet = fieldState == FieldParser::FIELD_SET; + } else if (fieldState == FieldParser::FIELD_SET) { + _isNSet = true; + _n = tempN; + } + + // We're using appendNumber on generation so we'll try a smaller type + // (int) first and then fall back to the original type (long long). + BSONField<int> fieldNModified(nModified()); + int intNModified; + fieldState = FieldParser::extract(source, fieldNModified, &intNModified, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) { + // try falling back to a larger type + fieldState = FieldParser::extract(source, nModified, &_nModified, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isNModifiedSet = fieldState == FieldParser::FIELD_SET; + } else if (fieldState == FieldParser::FIELD_SET) { + _isNModifiedSet = true; + _nModified = intNModified; } - void BatchedCommandResponse::cloneTo(BatchedCommandResponse* other) const { - other->clear(); + std::vector<BatchedUpsertDetail*>* tempUpsertDetails = NULL; + fieldState = FieldParser::extract(source, upsertDetails, &tempUpsertDetails, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _upsertDetails.reset(tempUpsertDetails); - other->_ok = _ok; - other->_isOkSet = _isOkSet; + fieldState = FieldParser::extract(source, lastOp, &_lastOp, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isLastOpSet = fieldState == FieldParser::FIELD_SET; - other->_errCode = _errCode; - other->_isErrCodeSet = _isErrCodeSet; + fieldState = FieldParser::extract(source, electionId, &_electionId, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isElectionIdSet = fieldState == FieldParser::FIELD_SET; - other->_errMessage = _errMessage; - other->_isErrMessageSet = _isErrMessageSet; + std::vector<WriteErrorDetail*>* tempErrDetails = NULL; + fieldState = FieldParser::extract(source, writeErrors, &tempErrDetails, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _writeErrorDetails.reset(tempErrDetails); - other->_nModified = _nModified; - other->_isNModifiedSet = _isNModifiedSet; + WCErrorDetail* wcError = NULL; + fieldState = FieldParser::extract(source, writeConcernError, &wcError, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _wcErrDetails.reset(wcError); - other->_n = _n; - other->_isNSet = _isNSet; + return true; +} - other->_singleUpserted = _singleUpserted; - other->_isSingleUpsertedSet = _isSingleUpsertedSet; +void BatchedCommandResponse::clear() { + _ok = false; + _isOkSet = false; - other->unsetUpsertDetails(); - if (_upsertDetails.get()) { - for (std::vector<BatchedUpsertDetail*>::const_iterator it = _upsertDetails->begin(); - it != _upsertDetails->end(); - ++it) { - BatchedUpsertDetail* upsertDetailsItem = new BatchedUpsertDetail; - (*it)->cloneTo(upsertDetailsItem); - other->addToUpsertDetails(upsertDetailsItem); - } - } + _errCode = 0; + _isErrCodeSet = false; - other->_lastOp = _lastOp; - other->_isLastOpSet = _isLastOpSet; - - other->_electionId = _electionId; - other->_isElectionIdSet = _isElectionIdSet; - - other->unsetErrDetails(); - if (_writeErrorDetails.get()) { - for(std::vector<WriteErrorDetail*>::const_iterator it = _writeErrorDetails->begin(); - it != _writeErrorDetails->end(); - ++it) { - WriteErrorDetail* errDetailsItem = new WriteErrorDetail; - (*it)->cloneTo(errDetailsItem); - other->addToErrDetails(errDetailsItem); - } - } + _errMessage.clear(); + _isErrMessageSet = false; - if (_wcErrDetails.get()) { - other->_wcErrDetails.reset(new WCErrorDetail()); - _wcErrDetails->cloneTo(other->_wcErrDetails.get()); - } - } - - std::string BatchedCommandResponse::toString() const { - return toBSON().toString(); - } + _nModified = 0; + _isNModifiedSet = false; - void BatchedCommandResponse::setOk(int ok) { - _ok = ok; - _isOkSet = true; - } + _n = 0; + _isNSet = false; - void BatchedCommandResponse::unsetOk() { - _isOkSet = false; - } + _singleUpserted = BSONObj(); + _isSingleUpsertedSet = false; - bool BatchedCommandResponse::isOkSet() const { - return _isOkSet; - } - - int BatchedCommandResponse::getOk() const { - dassert(_isOkSet); - return _ok; + if (_upsertDetails.get()) { + for (std::vector<BatchedUpsertDetail*>::const_iterator it = _upsertDetails->begin(); + it != _upsertDetails->end(); + ++it) { + delete *it; + }; + _upsertDetails.reset(); } - void BatchedCommandResponse::setErrCode(int errCode) { - _errCode = errCode; - _isErrCodeSet = true; - } + _lastOp = Timestamp(); + _isLastOpSet = false; - void BatchedCommandResponse::unsetErrCode() { - _isErrCodeSet = false; - } + _electionId = OID(); + _isElectionIdSet = false; - bool BatchedCommandResponse::isErrCodeSet() const { - return _isErrCodeSet; + if (_writeErrorDetails.get()) { + for (std::vector<WriteErrorDetail*>::const_iterator it = _writeErrorDetails->begin(); + it != _writeErrorDetails->end(); + ++it) { + delete *it; + }; + _writeErrorDetails.reset(); } - int BatchedCommandResponse::getErrCode() const { - if ( _isErrCodeSet ) { - return _errCode; - } - else { - return errCode.getDefault(); - } - } + _wcErrDetails.reset(); +} - void BatchedCommandResponse::setErrMessage(StringData errMessage) { - _errMessage = errMessage.toString(); - _isErrMessageSet = true; - } +void BatchedCommandResponse::cloneTo(BatchedCommandResponse* other) const { + other->clear(); - void BatchedCommandResponse::unsetErrMessage() { - _isErrMessageSet = false; - } + other->_ok = _ok; + other->_isOkSet = _isOkSet; - bool BatchedCommandResponse::isErrMessageSet() const { - return _isErrMessageSet; - } + other->_errCode = _errCode; + other->_isErrCodeSet = _isErrCodeSet; - const std::string& BatchedCommandResponse::getErrMessage() const { - dassert(_isErrMessageSet); - return _errMessage; - } + other->_errMessage = _errMessage; + other->_isErrMessageSet = _isErrMessageSet; - void BatchedCommandResponse::setNModified(long long n) { - _nModified = n; - _isNModifiedSet = true; - } + other->_nModified = _nModified; + other->_isNModifiedSet = _isNModifiedSet; - void BatchedCommandResponse::unsetNModified() { - _isNModifiedSet = false; - } + other->_n = _n; + other->_isNSet = _isNSet; - bool BatchedCommandResponse::isNModified() const { - return _isNModifiedSet; - } + other->_singleUpserted = _singleUpserted; + other->_isSingleUpsertedSet = _isSingleUpsertedSet; - long long BatchedCommandResponse::getNModified() const { - if ( _isNModifiedSet ) { - return _nModified; - } - else { - return nModified.getDefault(); + other->unsetUpsertDetails(); + if (_upsertDetails.get()) { + for (std::vector<BatchedUpsertDetail*>::const_iterator it = _upsertDetails->begin(); + it != _upsertDetails->end(); + ++it) { + BatchedUpsertDetail* upsertDetailsItem = new BatchedUpsertDetail; + (*it)->cloneTo(upsertDetailsItem); + other->addToUpsertDetails(upsertDetailsItem); } } - void BatchedCommandResponse::setN(long long n) { - _n = n; - _isNSet = true; - } - - void BatchedCommandResponse::unsetN() { - _isNSet = false; - } + other->_lastOp = _lastOp; + other->_isLastOpSet = _isLastOpSet; - bool BatchedCommandResponse::isNSet() const { - return _isNSet; - } + other->_electionId = _electionId; + other->_isElectionIdSet = _isElectionIdSet; - long long BatchedCommandResponse::getN() const { - if ( _isNSet ) { - return _n; - } - else { - return n.getDefault(); - } - } - - void BatchedCommandResponse::setUpsertDetails( - const std::vector<BatchedUpsertDetail*>& upsertDetails) { - unsetUpsertDetails(); - for (std::vector<BatchedUpsertDetail*>::const_iterator it = upsertDetails.begin(); - it != upsertDetails.end(); + other->unsetErrDetails(); + if (_writeErrorDetails.get()) { + for (std::vector<WriteErrorDetail*>::const_iterator it = _writeErrorDetails->begin(); + it != _writeErrorDetails->end(); ++it) { - unique_ptr<BatchedUpsertDetail> tempBatchedUpsertDetail(new BatchedUpsertDetail); - (*it)->cloneTo(tempBatchedUpsertDetail.get()); - addToUpsertDetails(tempBatchedUpsertDetail.release()); + WriteErrorDetail* errDetailsItem = new WriteErrorDetail; + (*it)->cloneTo(errDetailsItem); + other->addToErrDetails(errDetailsItem); } } - void BatchedCommandResponse::addToUpsertDetails(BatchedUpsertDetail* upsertDetails) { - if (_upsertDetails.get() == NULL) { - _upsertDetails.reset(new std::vector<BatchedUpsertDetail*>); - } - _upsertDetails->push_back(upsertDetails); - } - - void BatchedCommandResponse::unsetUpsertDetails() { - if (_upsertDetails.get() != NULL) { - for (std::vector<BatchedUpsertDetail*>::iterator it = _upsertDetails->begin(); - it != _upsertDetails->end(); - ++it) { - delete *it; - } - _upsertDetails.reset(); - } - } - - bool BatchedCommandResponse::isUpsertDetailsSet() const { - return _upsertDetails.get() != NULL; - } - - size_t BatchedCommandResponse::sizeUpsertDetails() const { - dassert(_upsertDetails.get()); - return _upsertDetails->size(); - } - - const std::vector<BatchedUpsertDetail*>& BatchedCommandResponse::getUpsertDetails() const { - dassert(_upsertDetails.get()); - return *_upsertDetails; - } - - const BatchedUpsertDetail* BatchedCommandResponse::getUpsertDetailsAt(size_t pos) const { - dassert(_upsertDetails.get()); - dassert(_upsertDetails->size() > pos); - return _upsertDetails->at(pos); - } - - void BatchedCommandResponse::setLastOp(Timestamp lastOp) { - _lastOp = lastOp; - _isLastOpSet = true; + if (_wcErrDetails.get()) { + other->_wcErrDetails.reset(new WCErrorDetail()); + _wcErrDetails->cloneTo(other->_wcErrDetails.get()); } +} + +std::string BatchedCommandResponse::toString() const { + return toBSON().toString(); +} - void BatchedCommandResponse::unsetLastOp() { - _isLastOpSet = false; - } +void BatchedCommandResponse::setOk(int ok) { + _ok = ok; + _isOkSet = true; +} + +void BatchedCommandResponse::unsetOk() { + _isOkSet = false; +} + +bool BatchedCommandResponse::isOkSet() const { + return _isOkSet; +} + +int BatchedCommandResponse::getOk() const { + dassert(_isOkSet); + return _ok; +} - bool BatchedCommandResponse::isLastOpSet() const { - return _isLastOpSet; - } - - Timestamp BatchedCommandResponse::getLastOp() const { - dassert(_isLastOpSet); - return _lastOp; - } - - void BatchedCommandResponse::setElectionId(const OID& electionId) { - _electionId = electionId; - _isElectionIdSet = true; - } - - void BatchedCommandResponse::unsetElectionId() { - _isElectionIdSet = false; - } - - bool BatchedCommandResponse::isElectionIdSet() const { - return _isElectionIdSet; - } - - OID BatchedCommandResponse::getElectionId() const { - dassert(_isElectionIdSet); - return _electionId; - } - - void BatchedCommandResponse::setErrDetails(const std::vector<WriteErrorDetail*>& errDetails) { - unsetErrDetails(); - for (std::vector<WriteErrorDetail*>::const_iterator it = errDetails.begin(); - it != errDetails.end(); +void BatchedCommandResponse::setErrCode(int errCode) { + _errCode = errCode; + _isErrCodeSet = true; +} + +void BatchedCommandResponse::unsetErrCode() { + _isErrCodeSet = false; +} + +bool BatchedCommandResponse::isErrCodeSet() const { + return _isErrCodeSet; +} + +int BatchedCommandResponse::getErrCode() const { + if (_isErrCodeSet) { + return _errCode; + } else { + return errCode.getDefault(); + } +} + +void BatchedCommandResponse::setErrMessage(StringData errMessage) { + _errMessage = errMessage.toString(); + _isErrMessageSet = true; +} + +void BatchedCommandResponse::unsetErrMessage() { + _isErrMessageSet = false; +} + +bool BatchedCommandResponse::isErrMessageSet() const { + return _isErrMessageSet; +} + +const std::string& BatchedCommandResponse::getErrMessage() const { + dassert(_isErrMessageSet); + return _errMessage; +} + +void BatchedCommandResponse::setNModified(long long n) { + _nModified = n; + _isNModifiedSet = true; +} + +void BatchedCommandResponse::unsetNModified() { + _isNModifiedSet = false; +} + +bool BatchedCommandResponse::isNModified() const { + return _isNModifiedSet; +} + +long long BatchedCommandResponse::getNModified() const { + if (_isNModifiedSet) { + return _nModified; + } else { + return nModified.getDefault(); + } +} + +void BatchedCommandResponse::setN(long long n) { + _n = n; + _isNSet = true; +} + +void BatchedCommandResponse::unsetN() { + _isNSet = false; +} + +bool BatchedCommandResponse::isNSet() const { + return _isNSet; +} + +long long BatchedCommandResponse::getN() const { + if (_isNSet) { + return _n; + } else { + return n.getDefault(); + } +} + +void BatchedCommandResponse::setUpsertDetails( + const std::vector<BatchedUpsertDetail*>& upsertDetails) { + unsetUpsertDetails(); + for (std::vector<BatchedUpsertDetail*>::const_iterator it = upsertDetails.begin(); + it != upsertDetails.end(); + ++it) { + unique_ptr<BatchedUpsertDetail> tempBatchedUpsertDetail(new BatchedUpsertDetail); + (*it)->cloneTo(tempBatchedUpsertDetail.get()); + addToUpsertDetails(tempBatchedUpsertDetail.release()); + } +} + +void BatchedCommandResponse::addToUpsertDetails(BatchedUpsertDetail* upsertDetails) { + if (_upsertDetails.get() == NULL) { + _upsertDetails.reset(new std::vector<BatchedUpsertDetail*>); + } + _upsertDetails->push_back(upsertDetails); +} + +void BatchedCommandResponse::unsetUpsertDetails() { + if (_upsertDetails.get() != NULL) { + for (std::vector<BatchedUpsertDetail*>::iterator it = _upsertDetails->begin(); + it != _upsertDetails->end(); ++it) { - unique_ptr<WriteErrorDetail> tempBatchErrorDetail(new WriteErrorDetail); - (*it)->cloneTo(tempBatchErrorDetail.get()); - addToErrDetails(tempBatchErrorDetail.release()); + delete *it; } - } - - void BatchedCommandResponse::addToErrDetails(WriteErrorDetail* errDetails) { - if (_writeErrorDetails.get() == NULL) { - _writeErrorDetails.reset(new std::vector<WriteErrorDetail*>); - } - _writeErrorDetails->push_back(errDetails); - } - - void BatchedCommandResponse::unsetErrDetails() { - if (_writeErrorDetails.get() != NULL) { - for(std::vector<WriteErrorDetail*>::iterator it = _writeErrorDetails->begin(); - it != _writeErrorDetails->end(); - ++it) { - delete *it; - } - _writeErrorDetails.reset(); + _upsertDetails.reset(); + } +} + +bool BatchedCommandResponse::isUpsertDetailsSet() const { + return _upsertDetails.get() != NULL; +} + +size_t BatchedCommandResponse::sizeUpsertDetails() const { + dassert(_upsertDetails.get()); + return _upsertDetails->size(); +} + +const std::vector<BatchedUpsertDetail*>& BatchedCommandResponse::getUpsertDetails() const { + dassert(_upsertDetails.get()); + return *_upsertDetails; +} + +const BatchedUpsertDetail* BatchedCommandResponse::getUpsertDetailsAt(size_t pos) const { + dassert(_upsertDetails.get()); + dassert(_upsertDetails->size() > pos); + return _upsertDetails->at(pos); +} + +void BatchedCommandResponse::setLastOp(Timestamp lastOp) { + _lastOp = lastOp; + _isLastOpSet = true; +} + +void BatchedCommandResponse::unsetLastOp() { + _isLastOpSet = false; +} + +bool BatchedCommandResponse::isLastOpSet() const { + return _isLastOpSet; +} + +Timestamp BatchedCommandResponse::getLastOp() const { + dassert(_isLastOpSet); + return _lastOp; +} + +void BatchedCommandResponse::setElectionId(const OID& electionId) { + _electionId = electionId; + _isElectionIdSet = true; +} + +void BatchedCommandResponse::unsetElectionId() { + _isElectionIdSet = false; +} + +bool BatchedCommandResponse::isElectionIdSet() const { + return _isElectionIdSet; +} + +OID BatchedCommandResponse::getElectionId() const { + dassert(_isElectionIdSet); + return _electionId; +} + +void BatchedCommandResponse::setErrDetails(const std::vector<WriteErrorDetail*>& errDetails) { + unsetErrDetails(); + for (std::vector<WriteErrorDetail*>::const_iterator it = errDetails.begin(); + it != errDetails.end(); + ++it) { + unique_ptr<WriteErrorDetail> tempBatchErrorDetail(new WriteErrorDetail); + (*it)->cloneTo(tempBatchErrorDetail.get()); + addToErrDetails(tempBatchErrorDetail.release()); + } +} + +void BatchedCommandResponse::addToErrDetails(WriteErrorDetail* errDetails) { + if (_writeErrorDetails.get() == NULL) { + _writeErrorDetails.reset(new std::vector<WriteErrorDetail*>); + } + _writeErrorDetails->push_back(errDetails); +} + +void BatchedCommandResponse::unsetErrDetails() { + if (_writeErrorDetails.get() != NULL) { + for (std::vector<WriteErrorDetail*>::iterator it = _writeErrorDetails->begin(); + it != _writeErrorDetails->end(); + ++it) { + delete *it; } + _writeErrorDetails.reset(); } +} - bool BatchedCommandResponse::isErrDetailsSet() const { - return _writeErrorDetails.get() != NULL; - } +bool BatchedCommandResponse::isErrDetailsSet() const { + return _writeErrorDetails.get() != NULL; +} - size_t BatchedCommandResponse::sizeErrDetails() const { - dassert(_writeErrorDetails.get()); - return _writeErrorDetails->size(); - } +size_t BatchedCommandResponse::sizeErrDetails() const { + dassert(_writeErrorDetails.get()); + return _writeErrorDetails->size(); +} - const std::vector<WriteErrorDetail*>& BatchedCommandResponse::getErrDetails() const { - dassert(_writeErrorDetails.get()); - return *_writeErrorDetails; - } +const std::vector<WriteErrorDetail*>& BatchedCommandResponse::getErrDetails() const { + dassert(_writeErrorDetails.get()); + return *_writeErrorDetails; +} - const WriteErrorDetail* BatchedCommandResponse::getErrDetailsAt(size_t pos) const { - dassert(_writeErrorDetails.get()); - dassert(_writeErrorDetails->size() > pos); - return _writeErrorDetails->at(pos); - } +const WriteErrorDetail* BatchedCommandResponse::getErrDetailsAt(size_t pos) const { + dassert(_writeErrorDetails.get()); + dassert(_writeErrorDetails->size() > pos); + return _writeErrorDetails->at(pos); +} - void BatchedCommandResponse::setWriteConcernError(WCErrorDetail* error) { - _wcErrDetails.reset(error); - } +void BatchedCommandResponse::setWriteConcernError(WCErrorDetail* error) { + _wcErrDetails.reset(error); +} - void BatchedCommandResponse::unsetWriteConcernError() { - _wcErrDetails.reset(); - } +void BatchedCommandResponse::unsetWriteConcernError() { + _wcErrDetails.reset(); +} - bool BatchedCommandResponse::isWriteConcernErrorSet() const { - return _wcErrDetails.get(); - } +bool BatchedCommandResponse::isWriteConcernErrorSet() const { + return _wcErrDetails.get(); +} - const WCErrorDetail* BatchedCommandResponse::getWriteConcernError() const { - return _wcErrDetails.get(); - } +const WCErrorDetail* BatchedCommandResponse::getWriteConcernError() const { + return _wcErrDetails.get(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_command_response.h b/src/mongo/s/write_ops/batched_command_response.h index 68cee2abaa6..370f836b921 100644 --- a/src/mongo/s/write_ops/batched_command_response.h +++ b/src/mongo/s/write_ops/batched_command_response.h @@ -40,161 +40,161 @@ namespace mongo { - /** - * This class represents the layout and content of a insert/update/delete runCommand, - * the response side. - */ - class BatchedCommandResponse : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedCommandResponse); - public: - - // - // schema declarations - // - - static const BSONField<int> ok; - static const BSONField<int> errCode; - static const BSONField<std::string> errMessage; - static const BSONField<long long> n; - static const BSONField<long long> nModified; - static const BSONField<std::vector<BatchedUpsertDetail*> > upsertDetails; - static const BSONField<Timestamp> lastOp; - static const BSONField<OID> electionId; - static const BSONField<std::vector<WriteErrorDetail*> > writeErrors; - static const BSONField<WCErrorDetail*> writeConcernError; - - // - // construction / destruction - // - - BatchedCommandResponse(); - virtual ~BatchedCommandResponse(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(BatchedCommandResponse* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setOk(int ok); - void unsetOk(); - bool isOkSet() const; - int getOk() const; - - void setErrCode(int errCode); - void unsetErrCode(); - bool isErrCodeSet() const; - int getErrCode() const; - - void setErrMessage(StringData errMessage); - void unsetErrMessage(); - bool isErrMessageSet() const; - const std::string& getErrMessage() const; - - void setNModified(long long n); - void unsetNModified(); - bool isNModified() const; - long long getNModified() const; - - void setN(long long n); - void unsetN(); - bool isNSet() const; - long long getN() const; - - void setUpsertDetails(const std::vector<BatchedUpsertDetail*>& upsertDetails); - void addToUpsertDetails(BatchedUpsertDetail* upsertDetails); - void unsetUpsertDetails(); - bool isUpsertDetailsSet() const; - std::size_t sizeUpsertDetails() const; - const std::vector<BatchedUpsertDetail*>& getUpsertDetails() const; - const BatchedUpsertDetail* getUpsertDetailsAt(std::size_t pos) const; - - void setLastOp(Timestamp lastOp); - void unsetLastOp(); - bool isLastOpSet() const; - Timestamp getLastOp() const; - - void setElectionId(const OID& electionId); - void unsetElectionId(); - bool isElectionIdSet() const; - OID getElectionId() const; - - void setErrDetails(const std::vector<WriteErrorDetail*>& errDetails); - // errDetails ownership is transferred to here. - void addToErrDetails(WriteErrorDetail* errDetails); - void unsetErrDetails(); - bool isErrDetailsSet() const; - std::size_t sizeErrDetails() const; - const std::vector<WriteErrorDetail*>& getErrDetails() const; - const WriteErrorDetail* getErrDetailsAt(std::size_t pos) const; - - void setWriteConcernError(WCErrorDetail* error); - void unsetWriteConcernError(); - bool isWriteConcernErrorSet() const; - const WCErrorDetail* getWriteConcernError() const; - - private: - // Convention: (M)andatory, (O)ptional - - // (M) 0 if batch didn't get to be applied for any reason - int _ok; - bool _isOkSet; - - // (O) whether all items in the batch applied correctly - int _errCode; - bool _isErrCodeSet; - - // (O) whether all items in the batch applied correctly - std::string _errMessage; - bool _isErrMessageSet; - - // (M) number of documents affected - long long _n; - bool _isNSet; - - // (O) number of documents updated - long long _nModified; - bool _isNModifiedSet; - - // (O) "promoted" _upserted, if the corresponding request contained only one batch item - // Should only be present if _upserted is not. - BSONObj _singleUpserted; - bool _isSingleUpsertedSet; - - // (O) Array of upserted items' _id's - // Should only be present if _singleUpserted is not. - std::unique_ptr<std::vector<BatchedUpsertDetail*> >_upsertDetails; - - // (O) Timestamp assigned to the write op when it was written to the oplog. - // Normally, getLastError can use Client::_lastOp, but this is not valid for - // mongos which loses track of the session due to RCAR. Therefore, we must - // keep track of the lastOp manually ourselves. - Timestamp _lastOp; - bool _isLastOpSet; - - // (O) In addition to keeping track of the above lastOp timestamp, we must also keep - // track of the primary we talked to. This is because if the primary moves, - // subsequent calls to getLastError are invalid. The only way we know if an - // election has occurred is to use the unique electionId. - OID _electionId; - bool _isElectionIdSet; - - // (O) Array of item-level error information - std::unique_ptr<std::vector<WriteErrorDetail*> >_writeErrorDetails; - - // (O) errors that occurred while trying to satisfy the write concern. - std::unique_ptr<WCErrorDetail> _wcErrDetails; - }; - -} // namespace mongo +/** + * This class represents the layout and content of a insert/update/delete runCommand, + * the response side. + */ +class BatchedCommandResponse : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedCommandResponse); + +public: + // + // schema declarations + // + + static const BSONField<int> ok; + static const BSONField<int> errCode; + static const BSONField<std::string> errMessage; + static const BSONField<long long> n; + static const BSONField<long long> nModified; + static const BSONField<std::vector<BatchedUpsertDetail*>> upsertDetails; + static const BSONField<Timestamp> lastOp; + static const BSONField<OID> electionId; + static const BSONField<std::vector<WriteErrorDetail*>> writeErrors; + static const BSONField<WCErrorDetail*> writeConcernError; + + // + // construction / destruction + // + + BatchedCommandResponse(); + virtual ~BatchedCommandResponse(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(BatchedCommandResponse* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setOk(int ok); + void unsetOk(); + bool isOkSet() const; + int getOk() const; + + void setErrCode(int errCode); + void unsetErrCode(); + bool isErrCodeSet() const; + int getErrCode() const; + + void setErrMessage(StringData errMessage); + void unsetErrMessage(); + bool isErrMessageSet() const; + const std::string& getErrMessage() const; + + void setNModified(long long n); + void unsetNModified(); + bool isNModified() const; + long long getNModified() const; + + void setN(long long n); + void unsetN(); + bool isNSet() const; + long long getN() const; + + void setUpsertDetails(const std::vector<BatchedUpsertDetail*>& upsertDetails); + void addToUpsertDetails(BatchedUpsertDetail* upsertDetails); + void unsetUpsertDetails(); + bool isUpsertDetailsSet() const; + std::size_t sizeUpsertDetails() const; + const std::vector<BatchedUpsertDetail*>& getUpsertDetails() const; + const BatchedUpsertDetail* getUpsertDetailsAt(std::size_t pos) const; + + void setLastOp(Timestamp lastOp); + void unsetLastOp(); + bool isLastOpSet() const; + Timestamp getLastOp() const; + + void setElectionId(const OID& electionId); + void unsetElectionId(); + bool isElectionIdSet() const; + OID getElectionId() const; + + void setErrDetails(const std::vector<WriteErrorDetail*>& errDetails); + // errDetails ownership is transferred to here. + void addToErrDetails(WriteErrorDetail* errDetails); + void unsetErrDetails(); + bool isErrDetailsSet() const; + std::size_t sizeErrDetails() const; + const std::vector<WriteErrorDetail*>& getErrDetails() const; + const WriteErrorDetail* getErrDetailsAt(std::size_t pos) const; + + void setWriteConcernError(WCErrorDetail* error); + void unsetWriteConcernError(); + bool isWriteConcernErrorSet() const; + const WCErrorDetail* getWriteConcernError() const; + +private: + // Convention: (M)andatory, (O)ptional + + // (M) 0 if batch didn't get to be applied for any reason + int _ok; + bool _isOkSet; + + // (O) whether all items in the batch applied correctly + int _errCode; + bool _isErrCodeSet; + + // (O) whether all items in the batch applied correctly + std::string _errMessage; + bool _isErrMessageSet; + + // (M) number of documents affected + long long _n; + bool _isNSet; + + // (O) number of documents updated + long long _nModified; + bool _isNModifiedSet; + + // (O) "promoted" _upserted, if the corresponding request contained only one batch item + // Should only be present if _upserted is not. + BSONObj _singleUpserted; + bool _isSingleUpsertedSet; + + // (O) Array of upserted items' _id's + // Should only be present if _singleUpserted is not. + std::unique_ptr<std::vector<BatchedUpsertDetail*>> _upsertDetails; + + // (O) Timestamp assigned to the write op when it was written to the oplog. + // Normally, getLastError can use Client::_lastOp, but this is not valid for + // mongos which loses track of the session due to RCAR. Therefore, we must + // keep track of the lastOp manually ourselves. + Timestamp _lastOp; + bool _isLastOpSet; + + // (O) In addition to keeping track of the above lastOp timestamp, we must also keep + // track of the primary we talked to. This is because if the primary moves, + // subsequent calls to getLastError are invalid. The only way we know if an + // election has occurred is to use the unique electionId. + OID _electionId; + bool _isElectionIdSet; + + // (O) Array of item-level error information + std::unique_ptr<std::vector<WriteErrorDetail*>> _writeErrorDetails; + + // (O) errors that occurred while trying to satisfy the write concern. + std::unique_ptr<WCErrorDetail> _wcErrDetails; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_command_response_test.cpp b/src/mongo/s/write_ops/batched_command_response_test.cpp index e97b567e761..ba53f201128 100644 --- a/src/mongo/s/write_ops/batched_command_response_test.cpp +++ b/src/mongo/s/write_ops/batched_command_response_test.cpp @@ -37,51 +37,42 @@ namespace { - using mongo::BSONArray; - using mongo::BSONObj; - using mongo::BatchedCommandResponse; - using mongo::WriteErrorDetail; - using mongo::WCErrorDetail; - using mongo::Date_t; - using std::string; +using mongo::BSONArray; +using mongo::BSONObj; +using mongo::BatchedCommandResponse; +using mongo::WriteErrorDetail; +using mongo::WCErrorDetail; +using mongo::Date_t; +using std::string; - TEST(RoundTrip, Normal) { +TEST(RoundTrip, Normal) { + BSONArray writeErrorsArray = BSON_ARRAY( + BSON(WriteErrorDetail::index(0) << WriteErrorDetail::errCode(-2) + << WriteErrorDetail::errInfo(BSON("more info" << 1)) + << WriteErrorDetail::errMessage("index 0 failed")) + << BSON(WriteErrorDetail::index(1) << WriteErrorDetail::errCode(-3) + << WriteErrorDetail::errInfo(BSON("more info" << 1)) + << WriteErrorDetail::errMessage("index 1 failed too"))); - BSONArray writeErrorsArray = - BSON_ARRAY( - BSON(WriteErrorDetail::index(0) << - WriteErrorDetail::errCode(-2) << - WriteErrorDetail::errInfo(BSON("more info" << 1)) << - WriteErrorDetail::errMessage("index 0 failed") - ) << - BSON(WriteErrorDetail::index(1) << - WriteErrorDetail::errCode(-3) << - WriteErrorDetail::errInfo(BSON("more info" << 1)) << - WriteErrorDetail::errMessage("index 1 failed too") - ) - ); + BSONObj writeConcernError(BSON(WCErrorDetail::errCode(8) + << WCErrorDetail::errInfo(BSON("a" << 1)) + << WCErrorDetail::errMessage("norepl"))); - BSONObj writeConcernError( - BSON(WCErrorDetail::errCode(8) << - WCErrorDetail::errInfo(BSON("a" << 1)) << - WCErrorDetail::errMessage("norepl"))); + BSONObj origResponseObj = BSON( + BatchedCommandResponse::ok(false) + << BatchedCommandResponse::errCode(-1) + << BatchedCommandResponse::errMessage("this batch didn't work") + << BatchedCommandResponse::n(0) << BatchedCommandResponse::lastOp(mongo::Timestamp(1ULL)) + << BatchedCommandResponse::writeErrors() << writeErrorsArray + << BatchedCommandResponse::writeConcernError() << writeConcernError); - BSONObj origResponseObj = - BSON(BatchedCommandResponse::ok(false) << - BatchedCommandResponse::errCode(-1) << - BatchedCommandResponse::errMessage("this batch didn't work") << - BatchedCommandResponse::n(0) << - BatchedCommandResponse::lastOp(mongo::Timestamp(1ULL)) << - BatchedCommandResponse::writeErrors() << writeErrorsArray << - BatchedCommandResponse::writeConcernError() << writeConcernError); + string errMsg; + BatchedCommandResponse response; + bool ok = response.parseBSON(origResponseObj, &errMsg); + ASSERT_TRUE(ok); - string errMsg; - BatchedCommandResponse response; - bool ok = response.parseBSON(origResponseObj, &errMsg); - ASSERT_TRUE(ok); + BSONObj genResponseObj = response.toBSON(); + ASSERT_EQUALS(0, genResponseObj.woCompare(origResponseObj)); +} - BSONObj genResponseObj = response.toBSON(); - ASSERT_EQUALS(0, genResponseObj.woCompare(origResponseObj)); - } - -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/write_ops/batched_delete_document.cpp b/src/mongo/s/write_ops/batched_delete_document.cpp index e12e5ad7a86..2cca2deea4d 100644 --- a/src/mongo/s/write_ops/batched_delete_document.cpp +++ b/src/mongo/s/write_ops/batched_delete_document.cpp @@ -33,130 +33,133 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; - const BSONField<BSONObj> BatchedDeleteDocument::query("q"); - const BSONField<int> BatchedDeleteDocument::limit("limit"); +using mongoutils::str::stream; +const BSONField<BSONObj> BatchedDeleteDocument::query("q"); +const BSONField<int> BatchedDeleteDocument::limit("limit"); - BatchedDeleteDocument::BatchedDeleteDocument() { - clear(); +BatchedDeleteDocument::BatchedDeleteDocument() { + clear(); +} + +BatchedDeleteDocument::~BatchedDeleteDocument() {} + +bool BatchedDeleteDocument::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BatchedDeleteDocument::~BatchedDeleteDocument() { + // All the mandatory fields must be present. + if (!_isQuerySet) { + *errMsg = stream() << "missing " << query.name() << " field"; + return false; } - bool BatchedDeleteDocument::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isQuerySet) { - *errMsg = stream() << "missing " << query.name() << " field"; - return false; - } - - if (!_isLimitSet) { - *errMsg = stream() << "missing " << limit.name() << " field"; - return false; - } - - if (_limit != 0 && _limit != 1) { - *errMsg = stream() << "specify either a 0 to delete all" - << "matching documents or 1 to delete a single document"; - return false; - } - - return true; + if (!_isLimitSet) { + *errMsg = stream() << "missing " << limit.name() << " field"; + return false; } - BSONObj BatchedDeleteDocument::toBSON() const { - BSONObjBuilder builder; + if (_limit != 0 && _limit != 1) { + *errMsg = stream() << "specify either a 0 to delete all" + << "matching documents or 1 to delete a single document"; + return false; + } - if (_isQuerySet) builder.append(query(), _query); + return true; +} - if (_isLimitSet) builder.append(limit(), _limit); +BSONObj BatchedDeleteDocument::toBSON() const { + BSONObjBuilder builder; - return builder.obj(); - } + if (_isQuerySet) + builder.append(query(), _query); - bool BatchedDeleteDocument::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + if (_isLimitSet) + builder.append(limit(), _limit); - std::string dummy; - if (!errMsg) errMsg = &dummy; + return builder.obj(); +} - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, query, &_query, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isQuerySet = fieldState == FieldParser::FIELD_SET; +bool BatchedDeleteDocument::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - fieldState = FieldParser::extractNumber(source, limit, &_limit, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isLimitSet = fieldState == FieldParser::FIELD_SET; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - return true; - } + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, query, &_query, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isQuerySet = fieldState == FieldParser::FIELD_SET; - void BatchedDeleteDocument::clear() { - _query = BSONObj(); - _isQuerySet = false; + fieldState = FieldParser::extractNumber(source, limit, &_limit, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isLimitSet = fieldState == FieldParser::FIELD_SET; - _limit = 0; - _isLimitSet = false; + return true; +} - } +void BatchedDeleteDocument::clear() { + _query = BSONObj(); + _isQuerySet = false; - void BatchedDeleteDocument::cloneTo(BatchedDeleteDocument* other) const { - other->clear(); + _limit = 0; + _isLimitSet = false; +} - other->_query = _query; - other->_isQuerySet = _isQuerySet; +void BatchedDeleteDocument::cloneTo(BatchedDeleteDocument* other) const { + other->clear(); - other->_limit = _limit; - other->_isLimitSet = _isLimitSet; - } + other->_query = _query; + other->_isQuerySet = _isQuerySet; - std::string BatchedDeleteDocument::toString() const { - return toBSON().toString(); - } + other->_limit = _limit; + other->_isLimitSet = _isLimitSet; +} - void BatchedDeleteDocument::setQuery(const BSONObj& query) { - _query = query.getOwned(); - _isQuerySet = true; - } +std::string BatchedDeleteDocument::toString() const { + return toBSON().toString(); +} - void BatchedDeleteDocument::unsetQuery() { - _isQuerySet = false; - } +void BatchedDeleteDocument::setQuery(const BSONObj& query) { + _query = query.getOwned(); + _isQuerySet = true; +} - bool BatchedDeleteDocument::isQuerySet() const { - return _isQuerySet; - } +void BatchedDeleteDocument::unsetQuery() { + _isQuerySet = false; +} - const BSONObj& BatchedDeleteDocument::getQuery() const { - dassert(_isQuerySet); - return _query; - } +bool BatchedDeleteDocument::isQuerySet() const { + return _isQuerySet; +} - void BatchedDeleteDocument::setLimit(int limit) { - _limit = limit; - _isLimitSet = true; - } +const BSONObj& BatchedDeleteDocument::getQuery() const { + dassert(_isQuerySet); + return _query; +} - void BatchedDeleteDocument::unsetLimit() { - _isLimitSet = false; - } +void BatchedDeleteDocument::setLimit(int limit) { + _limit = limit; + _isLimitSet = true; +} - bool BatchedDeleteDocument::isLimitSet() const { - return _isLimitSet; - } +void BatchedDeleteDocument::unsetLimit() { + _isLimitSet = false; +} - int BatchedDeleteDocument::getLimit() const { - dassert(_isLimitSet); - return _limit; - } +bool BatchedDeleteDocument::isLimitSet() const { + return _isLimitSet; +} + +int BatchedDeleteDocument::getLimit() const { + dassert(_isLimitSet); + return _limit; +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_delete_document.h b/src/mongo/s/write_ops/batched_delete_document.h index 346d2cdb77d..26ab8c44716 100644 --- a/src/mongo/s/write_ops/batched_delete_document.h +++ b/src/mongo/s/write_ops/batched_delete_document.h @@ -37,65 +37,65 @@ namespace mongo { - /** - * This class represents the layout and content of a delete document runCommand, - * in the resquest side. - */ - class BatchedDeleteDocument : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedDeleteDocument); - public: - - // - // schema declarations - // - - static const BSONField<BSONObj> query; - static const BSONField<int> limit; - - // - // construction / destruction - // - - BatchedDeleteDocument(); - virtual ~BatchedDeleteDocument(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(BatchedDeleteDocument* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setQuery(const BSONObj& query); - void unsetQuery(); - bool isQuerySet() const; - const BSONObj& getQuery() const; - - void setLimit(int limit); - void unsetLimit(); - bool isLimitSet() const; - int getLimit() const; - - private: - // Convention: (M)andatory, (O)ptional - - // (M) query whose result the delete will remove - BSONObj _query; - bool _isQuerySet; - - // (M) the maximum number of documents to be deleted - int _limit; - bool _isLimitSet; - }; - -} // namespace mongo +/** + * This class represents the layout and content of a delete document runCommand, + * in the resquest side. + */ +class BatchedDeleteDocument : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedDeleteDocument); + +public: + // + // schema declarations + // + + static const BSONField<BSONObj> query; + static const BSONField<int> limit; + + // + // construction / destruction + // + + BatchedDeleteDocument(); + virtual ~BatchedDeleteDocument(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(BatchedDeleteDocument* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setQuery(const BSONObj& query); + void unsetQuery(); + bool isQuerySet() const; + const BSONObj& getQuery() const; + + void setLimit(int limit); + void unsetLimit(); + bool isLimitSet() const; + int getLimit() const; + +private: + // Convention: (M)andatory, (O)ptional + + // (M) query whose result the delete will remove + BSONObj _query; + bool _isQuerySet; + + // (M) the maximum number of documents to be deleted + int _limit; + bool _isLimitSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_delete_request.cpp b/src/mongo/s/write_ops/batched_delete_request.cpp index b0c6a3f7a7a..d99df0ffbfc 100644 --- a/src/mongo/s/write_ops/batched_delete_request.cpp +++ b/src/mongo/s/write_ops/batched_delete_request.cpp @@ -33,279 +33,287 @@ namespace mongo { - using std::unique_ptr; - using std::string; - - using mongoutils::str::stream; - - const std::string BatchedDeleteRequest::BATCHED_DELETE_REQUEST = "delete"; - const BSONField<std::string> BatchedDeleteRequest::collName( "delete" ); - const BSONField<std::vector<BatchedDeleteDocument*> > BatchedDeleteRequest::deletes( "deletes" ); - const BSONField<BSONObj> BatchedDeleteRequest::writeConcern( "writeConcern" ); - const BSONField<bool> BatchedDeleteRequest::ordered( "ordered", true ); - const BSONField<BSONObj> BatchedDeleteRequest::metadata("metadata"); - - BatchedDeleteRequest::BatchedDeleteRequest() { - clear(); +using std::unique_ptr; +using std::string; + +using mongoutils::str::stream; + +const std::string BatchedDeleteRequest::BATCHED_DELETE_REQUEST = "delete"; +const BSONField<std::string> BatchedDeleteRequest::collName("delete"); +const BSONField<std::vector<BatchedDeleteDocument*>> BatchedDeleteRequest::deletes("deletes"); +const BSONField<BSONObj> BatchedDeleteRequest::writeConcern("writeConcern"); +const BSONField<bool> BatchedDeleteRequest::ordered("ordered", true); +const BSONField<BSONObj> BatchedDeleteRequest::metadata("metadata"); + +BatchedDeleteRequest::BatchedDeleteRequest() { + clear(); +} + +BatchedDeleteRequest::~BatchedDeleteRequest() { + unsetDeletes(); +} + +bool BatchedDeleteRequest::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BatchedDeleteRequest::~BatchedDeleteRequest() { - unsetDeletes(); + // All the mandatory fields must be present. + if (!_isCollNameSet) { + *errMsg = stream() << "missing " << collName.name() << " field"; + return false; } - bool BatchedDeleteRequest::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isCollNameSet) { - *errMsg = stream() << "missing " << collName.name() << " field"; - return false; - } - - if (!_isDeletesSet) { - *errMsg = stream() << "missing " << deletes.name() << " field"; - return false; - } - - return true; + if (!_isDeletesSet) { + *errMsg = stream() << "missing " << deletes.name() << " field"; + return false; } - BSONObj BatchedDeleteRequest::toBSON() const { - BSONObjBuilder builder; - - if (_isCollNameSet) builder.append(collName(), _collName); + return true; +} - if (_isDeletesSet) { - BSONArrayBuilder deletesBuilder(builder.subarrayStart(deletes())); - for (std::vector<BatchedDeleteDocument*>::const_iterator it = _deletes.begin(); - it != _deletes.end(); - ++it) { - BSONObj deleteDocument = (*it)->toBSON(); - deletesBuilder.append(deleteDocument); - } - deletesBuilder.done(); - } - - if (_isWriteConcernSet) builder.append(writeConcern(), _writeConcern); +BSONObj BatchedDeleteRequest::toBSON() const { + BSONObjBuilder builder; - if (_isOrderedSet) builder.append(ordered(), _ordered); + if (_isCollNameSet) + builder.append(collName(), _collName); - if (_metadata) builder.append(metadata(), _metadata->toBSON()); - - return builder.obj(); - } - - bool BatchedDeleteRequest::parseBSON(const BSONObj& source, string* errMsg) { - clear(); - - std::string dummy; - if (!errMsg) errMsg = &dummy; - - FieldParser::FieldState fieldState; - std::string collNameTemp; - fieldState = FieldParser::extract(source, collName, &collNameTemp, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _collName = NamespaceString(collNameTemp); - _isCollNameSet = fieldState == FieldParser::FIELD_SET; - - fieldState = FieldParser::extract(source, deletes, &_deletes, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isDeletesSet = fieldState == FieldParser::FIELD_SET; - - fieldState = FieldParser::extract(source, writeConcern, &_writeConcern, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isWriteConcernSet = fieldState == FieldParser::FIELD_SET; - - fieldState = FieldParser::extract(source, ordered, &_ordered, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isOrderedSet = fieldState == FieldParser::FIELD_SET; - - BSONObj metadataObj; - fieldState = FieldParser::extract(source, metadata, &metadataObj, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - - if (!metadataObj.isEmpty()) { - _metadata.reset(new BatchedRequestMetadata()); - if (!_metadata->parseBSON(metadataObj, errMsg)) { - return false; - } + if (_isDeletesSet) { + BSONArrayBuilder deletesBuilder(builder.subarrayStart(deletes())); + for (std::vector<BatchedDeleteDocument*>::const_iterator it = _deletes.begin(); + it != _deletes.end(); + ++it) { + BSONObj deleteDocument = (*it)->toBSON(); + deletesBuilder.append(deleteDocument); } - - return true; + deletesBuilder.done(); } - void BatchedDeleteRequest::clear() { - _collName = NamespaceString(); - _isCollNameSet = false; + if (_isWriteConcernSet) + builder.append(writeConcern(), _writeConcern); - unsetDeletes(); + if (_isOrderedSet) + builder.append(ordered(), _ordered); - _writeConcern = BSONObj(); - _isWriteConcernSet = false; + if (_metadata) + builder.append(metadata(), _metadata->toBSON()); - _ordered = false; - _isOrderedSet = false; + return builder.obj(); +} - _metadata.reset(); - } +bool BatchedDeleteRequest::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - void BatchedDeleteRequest::cloneTo(BatchedDeleteRequest* other) const { - other->clear(); + std::string dummy; + if (!errMsg) + errMsg = &dummy; - other->_collName = _collName; - other->_isCollNameSet = _isCollNameSet; + FieldParser::FieldState fieldState; + std::string collNameTemp; + fieldState = FieldParser::extract(source, collName, &collNameTemp, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _collName = NamespaceString(collNameTemp); + _isCollNameSet = fieldState == FieldParser::FIELD_SET; - for(std::vector<BatchedDeleteDocument*>::const_iterator it = _deletes.begin(); - it != _deletes.end(); - ++it) { - unique_ptr<BatchedDeleteDocument> tempBatchDeleteDocument(new BatchedDeleteDocument); - (*it)->cloneTo(tempBatchDeleteDocument.get()); - other->addToDeletes(tempBatchDeleteDocument.release()); - } - other->_isDeletesSet = _isDeletesSet; + fieldState = FieldParser::extract(source, deletes, &_deletes, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isDeletesSet = fieldState == FieldParser::FIELD_SET; - other->_writeConcern = _writeConcern; - other->_isWriteConcernSet = _isWriteConcernSet; + fieldState = FieldParser::extract(source, writeConcern, &_writeConcern, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isWriteConcernSet = fieldState == FieldParser::FIELD_SET; - other->_ordered = _ordered; - other->_isOrderedSet = _isOrderedSet; + fieldState = FieldParser::extract(source, ordered, &_ordered, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isOrderedSet = fieldState == FieldParser::FIELD_SET; - if (_metadata) { - other->_metadata.reset(new BatchedRequestMetadata()); - _metadata->cloneTo(other->_metadata.get()); - } - } + BSONObj metadataObj; + fieldState = FieldParser::extract(source, metadata, &metadataObj, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; - std::string BatchedDeleteRequest::toString() const { - return toBSON().toString(); - } - - void BatchedDeleteRequest::setCollName(StringData collName) { - _collName = NamespaceString(collName); - _isCollNameSet = true; - } - - const std::string& BatchedDeleteRequest::getCollName() const { - dassert(_isCollNameSet); - return _collName.ns(); + if (!metadataObj.isEmpty()) { + _metadata.reset(new BatchedRequestMetadata()); + if (!_metadata->parseBSON(metadataObj, errMsg)) { + return false; + } } - void BatchedDeleteRequest::setCollNameNS(const NamespaceString& collName) { - _collName = collName; - _isCollNameSet = true; - } + return true; +} - const NamespaceString& BatchedDeleteRequest::getCollNameNS() const { - dassert(_isCollNameSet); - return _collName; - } +void BatchedDeleteRequest::clear() { + _collName = NamespaceString(); + _isCollNameSet = false; - const NamespaceString& BatchedDeleteRequest::getTargetingNSS() const { - return getCollNameNS(); - } + unsetDeletes(); - void BatchedDeleteRequest::setDeletes(const std::vector<BatchedDeleteDocument*>& deletes) { - for (std::vector<BatchedDeleteDocument*>::const_iterator it = deletes.begin(); - it != deletes.end(); - ++it) { - unique_ptr<BatchedDeleteDocument> tempBatchDeleteDocument(new BatchedDeleteDocument); - (*it)->cloneTo(tempBatchDeleteDocument.get()); - addToDeletes(tempBatchDeleteDocument.release()); - } - _isDeletesSet = deletes.size() > 0; - } - - void BatchedDeleteRequest::addToDeletes(BatchedDeleteDocument* deletes) { - _deletes.push_back(deletes); - _isDeletesSet = true; - } - - void BatchedDeleteRequest::unsetDeletes() { - for(std::vector<BatchedDeleteDocument*>::iterator it = _deletes.begin(); - it != _deletes.end(); - ++it) { - delete *it; - } - _deletes.clear(); - _isDeletesSet = false; - } + _writeConcern = BSONObj(); + _isWriteConcernSet = false; - bool BatchedDeleteRequest::isDeletesSet() const { - return _isDeletesSet; - } + _ordered = false; + _isOrderedSet = false; - size_t BatchedDeleteRequest::sizeDeletes() const { - return _deletes.size(); - } + _metadata.reset(); +} - const std::vector<BatchedDeleteDocument*>& BatchedDeleteRequest::getDeletes() const { - dassert(_isDeletesSet); - return _deletes; - } +void BatchedDeleteRequest::cloneTo(BatchedDeleteRequest* other) const { + other->clear(); - const BatchedDeleteDocument* BatchedDeleteRequest::getDeletesAt(size_t pos) const { - dassert(_isDeletesSet); - dassert(_deletes.size() > pos); - return _deletes.at(pos); - } + other->_collName = _collName; + other->_isCollNameSet = _isCollNameSet; - void BatchedDeleteRequest::setWriteConcern(const BSONObj& writeConcern) { - _writeConcern = writeConcern.getOwned(); - _isWriteConcernSet = true; + for (std::vector<BatchedDeleteDocument*>::const_iterator it = _deletes.begin(); + it != _deletes.end(); + ++it) { + unique_ptr<BatchedDeleteDocument> tempBatchDeleteDocument(new BatchedDeleteDocument); + (*it)->cloneTo(tempBatchDeleteDocument.get()); + other->addToDeletes(tempBatchDeleteDocument.release()); } + other->_isDeletesSet = _isDeletesSet; - void BatchedDeleteRequest::unsetWriteConcern() { - _isWriteConcernSet = false; - } + other->_writeConcern = _writeConcern; + other->_isWriteConcernSet = _isWriteConcernSet; - bool BatchedDeleteRequest::isWriteConcernSet() const { - return _isWriteConcernSet; - } + other->_ordered = _ordered; + other->_isOrderedSet = _isOrderedSet; - const BSONObj& BatchedDeleteRequest::getWriteConcern() const { - dassert(_isWriteConcernSet); - return _writeConcern; + if (_metadata) { + other->_metadata.reset(new BatchedRequestMetadata()); + _metadata->cloneTo(other->_metadata.get()); } - - void BatchedDeleteRequest::setOrdered(bool ordered) { - _ordered = ordered; - _isOrderedSet = true; +} + +std::string BatchedDeleteRequest::toString() const { + return toBSON().toString(); +} + +void BatchedDeleteRequest::setCollName(StringData collName) { + _collName = NamespaceString(collName); + _isCollNameSet = true; +} + +const std::string& BatchedDeleteRequest::getCollName() const { + dassert(_isCollNameSet); + return _collName.ns(); +} + +void BatchedDeleteRequest::setCollNameNS(const NamespaceString& collName) { + _collName = collName; + _isCollNameSet = true; +} + +const NamespaceString& BatchedDeleteRequest::getCollNameNS() const { + dassert(_isCollNameSet); + return _collName; +} + +const NamespaceString& BatchedDeleteRequest::getTargetingNSS() const { + return getCollNameNS(); +} + +void BatchedDeleteRequest::setDeletes(const std::vector<BatchedDeleteDocument*>& deletes) { + for (std::vector<BatchedDeleteDocument*>::const_iterator it = deletes.begin(); + it != deletes.end(); + ++it) { + unique_ptr<BatchedDeleteDocument> tempBatchDeleteDocument(new BatchedDeleteDocument); + (*it)->cloneTo(tempBatchDeleteDocument.get()); + addToDeletes(tempBatchDeleteDocument.release()); } - - void BatchedDeleteRequest::unsetOrdered() { - _isOrderedSet = false; - } - - bool BatchedDeleteRequest::isOrderedSet() const { - return _isOrderedSet; + _isDeletesSet = deletes.size() > 0; +} + +void BatchedDeleteRequest::addToDeletes(BatchedDeleteDocument* deletes) { + _deletes.push_back(deletes); + _isDeletesSet = true; +} + +void BatchedDeleteRequest::unsetDeletes() { + for (std::vector<BatchedDeleteDocument*>::iterator it = _deletes.begin(); it != _deletes.end(); + ++it) { + delete *it; } - - bool BatchedDeleteRequest::getOrdered() const { - if (_isOrderedSet) { - return _ordered; - } - else { - return ordered.getDefault(); - } + _deletes.clear(); + _isDeletesSet = false; +} + +bool BatchedDeleteRequest::isDeletesSet() const { + return _isDeletesSet; +} + +size_t BatchedDeleteRequest::sizeDeletes() const { + return _deletes.size(); +} + +const std::vector<BatchedDeleteDocument*>& BatchedDeleteRequest::getDeletes() const { + dassert(_isDeletesSet); + return _deletes; +} + +const BatchedDeleteDocument* BatchedDeleteRequest::getDeletesAt(size_t pos) const { + dassert(_isDeletesSet); + dassert(_deletes.size() > pos); + return _deletes.at(pos); +} + +void BatchedDeleteRequest::setWriteConcern(const BSONObj& writeConcern) { + _writeConcern = writeConcern.getOwned(); + _isWriteConcernSet = true; +} + +void BatchedDeleteRequest::unsetWriteConcern() { + _isWriteConcernSet = false; +} + +bool BatchedDeleteRequest::isWriteConcernSet() const { + return _isWriteConcernSet; +} + +const BSONObj& BatchedDeleteRequest::getWriteConcern() const { + dassert(_isWriteConcernSet); + return _writeConcern; +} + +void BatchedDeleteRequest::setOrdered(bool ordered) { + _ordered = ordered; + _isOrderedSet = true; +} + +void BatchedDeleteRequest::unsetOrdered() { + _isOrderedSet = false; +} + +bool BatchedDeleteRequest::isOrderedSet() const { + return _isOrderedSet; +} + +bool BatchedDeleteRequest::getOrdered() const { + if (_isOrderedSet) { + return _ordered; + } else { + return ordered.getDefault(); } +} - void BatchedDeleteRequest::setMetadata(BatchedRequestMetadata* metadata) { - _metadata.reset(metadata); - } +void BatchedDeleteRequest::setMetadata(BatchedRequestMetadata* metadata) { + _metadata.reset(metadata); +} - void BatchedDeleteRequest::unsetMetadata() { - _metadata.reset(); - } +void BatchedDeleteRequest::unsetMetadata() { + _metadata.reset(); +} - bool BatchedDeleteRequest::isMetadataSet() const { - return _metadata.get(); - } +bool BatchedDeleteRequest::isMetadataSet() const { + return _metadata.get(); +} - BatchedRequestMetadata* BatchedDeleteRequest::getMetadata() const { - return _metadata.get(); - } +BatchedRequestMetadata* BatchedDeleteRequest::getMetadata() const { + return _metadata.get(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_delete_request.h b/src/mongo/s/write_ops/batched_delete_request.h index e3846833bec..ff73155360d 100644 --- a/src/mongo/s/write_ops/batched_delete_request.h +++ b/src/mongo/s/write_ops/batched_delete_request.h @@ -40,117 +40,119 @@ namespace mongo { +/** + * This class represents the layout and content of a batched delete runCommand, + * the request side. + */ +class BatchedDeleteRequest : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedDeleteRequest); + +public: + // + // schema declarations + // + + // Name used for the batched delete invocation. + static const std::string BATCHED_DELETE_REQUEST; + + // Field names and types in the batched delete command type. + static const BSONField<std::string> collName; + static const BSONField<std::vector<BatchedDeleteDocument*>> deletes; + static const BSONField<BSONObj> writeConcern; + static const BSONField<bool> ordered; + static const BSONField<BSONObj> metadata; + + // + // construction / destruction + // + + BatchedDeleteRequest(); + virtual ~BatchedDeleteRequest(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(BatchedDeleteRequest* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setCollName(StringData collName); + void setCollNameNS(const NamespaceString& collName); + const std::string& getCollName() const; + const NamespaceString& getCollNameNS() const; + + const NamespaceString& getTargetingNSS() const; + + void setDeletes(const std::vector<BatchedDeleteDocument*>& deletes); + /** - * This class represents the layout and content of a batched delete runCommand, - * the request side. + * deletes ownership is transferred to here. */ - class BatchedDeleteRequest : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedDeleteRequest); - public: - - // - // schema declarations - // - - // Name used for the batched delete invocation. - static const std::string BATCHED_DELETE_REQUEST; - - // Field names and types in the batched delete command type. - static const BSONField<std::string> collName; - static const BSONField<std::vector<BatchedDeleteDocument*> > deletes; - static const BSONField<BSONObj> writeConcern; - static const BSONField<bool> ordered; - static const BSONField<BSONObj> metadata; - - // - // construction / destruction - // - - BatchedDeleteRequest(); - virtual ~BatchedDeleteRequest(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(BatchedDeleteRequest* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setCollName(StringData collName); - void setCollNameNS(const NamespaceString& collName); - const std::string& getCollName() const; - const NamespaceString& getCollNameNS() const; - - const NamespaceString& getTargetingNSS() const; - - void setDeletes(const std::vector<BatchedDeleteDocument*>& deletes); - - /** - * deletes ownership is transferred to here. - */ - void addToDeletes(BatchedDeleteDocument* deletes); - void unsetDeletes(); - bool isDeletesSet() const; - std::size_t sizeDeletes() const; - const std::vector<BatchedDeleteDocument*>& getDeletes() const; - const BatchedDeleteDocument* getDeletesAt(std::size_t pos) const; - - void setWriteConcern(const BSONObj& writeConcern); - void unsetWriteConcern(); - bool isWriteConcernSet() const; - const BSONObj& getWriteConcern() const; - - void setOrdered(bool ordered); - void unsetOrdered(); - bool isOrderedSet() const; - bool getOrdered() const; - - /* - * metadata ownership will be transferred to this. - */ - void setMetadata(BatchedRequestMetadata* metadata); - void unsetMetadata(); - bool isMetadataSet() const; - BatchedRequestMetadata* getMetadata() const; - - /** - * These are no-ops since delete never validates documents. They only exist to fulfill the - * unified API. - */ - void setShouldBypassValidation(bool newVal) {} - bool shouldBypassValidation() const { return false; } - - private: - // Convention: (M)andatory, (O)ptional - - // (M) collection we're deleting from - NamespaceString _collName; - bool _isCollNameSet; - - // (M) array of individual deletes - std::vector<BatchedDeleteDocument*> _deletes; - bool _isDeletesSet; - - // (O) to be issued after the batch applied - BSONObj _writeConcern; - bool _isWriteConcernSet; - - // (O) whether batch is issued in parallel or not - bool _ordered; - bool _isOrderedSet; - - // (O) metadata associated with this request for internal use. - std::unique_ptr<BatchedRequestMetadata> _metadata; - }; - -} // namespace mongo + void addToDeletes(BatchedDeleteDocument* deletes); + void unsetDeletes(); + bool isDeletesSet() const; + std::size_t sizeDeletes() const; + const std::vector<BatchedDeleteDocument*>& getDeletes() const; + const BatchedDeleteDocument* getDeletesAt(std::size_t pos) const; + + void setWriteConcern(const BSONObj& writeConcern); + void unsetWriteConcern(); + bool isWriteConcernSet() const; + const BSONObj& getWriteConcern() const; + + void setOrdered(bool ordered); + void unsetOrdered(); + bool isOrderedSet() const; + bool getOrdered() const; + + /* + * metadata ownership will be transferred to this. + */ + void setMetadata(BatchedRequestMetadata* metadata); + void unsetMetadata(); + bool isMetadataSet() const; + BatchedRequestMetadata* getMetadata() const; + + /** + * These are no-ops since delete never validates documents. They only exist to fulfill the + * unified API. + */ + void setShouldBypassValidation(bool newVal) {} + bool shouldBypassValidation() const { + return false; + } + +private: + // Convention: (M)andatory, (O)ptional + + // (M) collection we're deleting from + NamespaceString _collName; + bool _isCollNameSet; + + // (M) array of individual deletes + std::vector<BatchedDeleteDocument*> _deletes; + bool _isDeletesSet; + + // (O) to be issued after the batch applied + BSONObj _writeConcern; + bool _isWriteConcernSet; + + // (O) whether batch is issued in parallel or not + bool _ordered; + bool _isOrderedSet; + + // (O) metadata associated with this request for internal use. + std::unique_ptr<BatchedRequestMetadata> _metadata; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_delete_request_test.cpp b/src/mongo/s/write_ops/batched_delete_request_test.cpp index 33aa331259a..3ecf8aedea4 100644 --- a/src/mongo/s/write_ops/batched_delete_request_test.cpp +++ b/src/mongo/s/write_ops/batched_delete_request_test.cpp @@ -36,53 +36,46 @@ namespace { - using mongo::BSONArray; - using mongo::BSONObj; - using mongo::BatchedDeleteRequest; - using mongo::BatchedDeleteDocument; - using mongo::BatchedRequestMetadata; - using mongo::BSONArrayBuilder; - using mongo::OID; - using mongo::Timestamp; - using std::string; +using mongo::BSONArray; +using mongo::BSONObj; +using mongo::BatchedDeleteRequest; +using mongo::BatchedDeleteDocument; +using mongo::BatchedRequestMetadata; +using mongo::BSONArrayBuilder; +using mongo::OID; +using mongo::Timestamp; +using std::string; - TEST(RoundTrip, Normal) { - BSONArray deleteArray = - BSON_ARRAY( - BSON(BatchedDeleteDocument::query(BSON("a" << 1)) << - BatchedDeleteDocument::limit(1) - ) << - BSON(BatchedDeleteDocument::query(BSON("b" << 1)) << - BatchedDeleteDocument::limit(1) - ) - ); +TEST(RoundTrip, Normal) { + BSONArray deleteArray = BSON_ARRAY( + BSON(BatchedDeleteDocument::query(BSON("a" << 1)) << BatchedDeleteDocument::limit(1)) + << BSON(BatchedDeleteDocument::query(BSON("b" << 1)) << BatchedDeleteDocument::limit(1))); - BSONObj writeConcernObj = BSON("w" << 1); + BSONObj writeConcernObj = BSON("w" << 1); - // The BSON_ARRAY macro doesn't support Timestamps. - BSONArrayBuilder arrBuilder; - arrBuilder.append(Timestamp(1,1)); - arrBuilder.append(OID::gen()); - BSONArray shardVersionArray = arrBuilder.arr(); + // The BSON_ARRAY macro doesn't support Timestamps. + BSONArrayBuilder arrBuilder; + arrBuilder.append(Timestamp(1, 1)); + arrBuilder.append(OID::gen()); + BSONArray shardVersionArray = arrBuilder.arr(); - BSONObj origDeleteRequestObj = - BSON(BatchedDeleteRequest::collName("test") << - BatchedDeleteRequest::deletes() << deleteArray << - BatchedDeleteRequest::writeConcern(writeConcernObj) << - BatchedDeleteRequest::ordered(true) << - BatchedDeleteRequest::metadata() << - BSON(BatchedRequestMetadata::shardName("shard000") << - BatchedRequestMetadata::shardVersion() << shardVersionArray << - BatchedRequestMetadata::session(0))); + BSONObj origDeleteRequestObj = + BSON(BatchedDeleteRequest::collName("test") + << BatchedDeleteRequest::deletes() << deleteArray + << BatchedDeleteRequest::writeConcern(writeConcernObj) + << BatchedDeleteRequest::ordered(true) << BatchedDeleteRequest::metadata() + << BSON(BatchedRequestMetadata::shardName("shard000") + << BatchedRequestMetadata::shardVersion() << shardVersionArray + << BatchedRequestMetadata::session(0))); - string errMsg; - BatchedDeleteRequest request; - bool ok = request.parseBSON(origDeleteRequestObj, &errMsg); - ASSERT_TRUE(ok); + string errMsg; + BatchedDeleteRequest request; + bool ok = request.parseBSON(origDeleteRequestObj, &errMsg); + ASSERT_TRUE(ok); - BSONObj genDeleteRequestObj = request.toBSON(); - ASSERT_EQUALS(0, genDeleteRequestObj.woCompare(origDeleteRequestObj)); - } + BSONObj genDeleteRequestObj = request.toBSON(); + ASSERT_EQUALS(0, genDeleteRequestObj.woCompare(origDeleteRequestObj)); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/write_ops/batched_insert_request.cpp b/src/mongo/s/write_ops/batched_insert_request.cpp index 94f7daabe0a..f5b862098c7 100644 --- a/src/mongo/s/write_ops/batched_insert_request.cpp +++ b/src/mongo/s/write_ops/batched_insert_request.cpp @@ -34,295 +34,296 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; +using mongoutils::str::stream; - const std::string BatchedInsertRequest::BATCHED_INSERT_REQUEST = "insert"; - const BSONField<std::string> BatchedInsertRequest::collName("insert"); - const BSONField<std::vector<BSONObj> > BatchedInsertRequest::documents("documents"); - const BSONField<BSONObj> BatchedInsertRequest::writeConcern("writeConcern"); - const BSONField<bool> BatchedInsertRequest::ordered("ordered", true); - const BSONField<BSONObj> BatchedInsertRequest::metadata("metadata"); +const std::string BatchedInsertRequest::BATCHED_INSERT_REQUEST = "insert"; +const BSONField<std::string> BatchedInsertRequest::collName("insert"); +const BSONField<std::vector<BSONObj>> BatchedInsertRequest::documents("documents"); +const BSONField<BSONObj> BatchedInsertRequest::writeConcern("writeConcern"); +const BSONField<bool> BatchedInsertRequest::ordered("ordered", true); +const BSONField<BSONObj> BatchedInsertRequest::metadata("metadata"); - BatchedInsertRequest::BatchedInsertRequest() { - clear(); - } +BatchedInsertRequest::BatchedInsertRequest() { + clear(); +} - BatchedInsertRequest::~BatchedInsertRequest() { - } +BatchedInsertRequest::~BatchedInsertRequest() {} - bool BatchedInsertRequest::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isCollNameSet) { - *errMsg = stream() << "missing " << collName.name() << " field"; - return false; - } - - if (!_isDocumentsSet) { - *errMsg = stream() << "missing " << documents.name() << " field"; - return false; - } - - return true; +bool BatchedInsertRequest::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BSONObj BatchedInsertRequest::toBSON() const { - BSONObjBuilder builder; - - if (_isCollNameSet) builder.append(collName(), _collName); - - if (_isDocumentsSet) { - BSONArrayBuilder documentsBuilder(builder.subarrayStart(documents())); - for (std::vector<BSONObj>::const_iterator it = _documents.begin(); - it != _documents.end(); - ++it) { - documentsBuilder.append(*it); - } - documentsBuilder.done(); - } - - if (_isWriteConcernSet) builder.append(writeConcern(), _writeConcern); - - if (_isOrderedSet) builder.append(ordered(), _ordered); - - if (_metadata) builder.append(metadata(), _metadata->toBSON()); - - if (_shouldBypassValidation) builder.append(bypassDocumentValidationCommandOption(), true); - - return builder.obj(); + // All the mandatory fields must be present. + if (!_isCollNameSet) { + *errMsg = stream() << "missing " << collName.name() << " field"; + return false; } - static void extractIndexNSS(const BSONObj& indexDesc, NamespaceString* indexNSS) { - *indexNSS = NamespaceString(indexDesc["ns"].str()); + if (!_isDocumentsSet) { + *errMsg = stream() << "missing " << documents.name() << " field"; + return false; } - bool BatchedInsertRequest::parseBSON(const BSONObj& source, string* errMsg) { - clear(); - - std::string dummy; - if (!errMsg) errMsg = &dummy; + return true; +} - BSONObjIterator sourceIt(source); +BSONObj BatchedInsertRequest::toBSON() const { + BSONObjBuilder builder; - while ( sourceIt.more() ) { + if (_isCollNameSet) + builder.append(collName(), _collName); - BSONElement sourceEl = sourceIt.next(); - - if ( collName() == sourceEl.fieldName() ) { - std::string temp; - FieldParser::FieldState fieldState = - FieldParser::extract( sourceEl, collName, &temp, errMsg ); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _collName = NamespaceString(temp); - _isCollNameSet = fieldState == FieldParser::FIELD_SET; - } - else if ( documents() == sourceEl.fieldName() ) { - FieldParser::FieldState fieldState = - FieldParser::extract( sourceEl, documents, &_documents, errMsg ); - if ( fieldState == FieldParser::FIELD_INVALID ) return false; - _isDocumentsSet = fieldState == FieldParser::FIELD_SET; - if (_documents.size() >= 1) - extractIndexNSS(_documents.at(0), &_targetNSS); - } - else if ( writeConcern() == sourceEl.fieldName() ) { - FieldParser::FieldState fieldState = - FieldParser::extract(sourceEl, writeConcern, &_writeConcern, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isWriteConcernSet = fieldState == FieldParser::FIELD_SET; - } - else if ( ordered() == sourceEl.fieldName() ) { - FieldParser::FieldState fieldState = - FieldParser::extract(sourceEl, ordered, &_ordered, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isOrderedSet = fieldState == FieldParser::FIELD_SET; - } - else if ( metadata() == sourceEl.fieldName() ) { - BSONObj metadataObj; - FieldParser::FieldState fieldState = - FieldParser::extract(sourceEl, metadata, &metadataObj, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - - if (!metadataObj.isEmpty()) { - _metadata.reset(new BatchedRequestMetadata()); - if (!_metadata->parseBSON(metadataObj, errMsg)) { - return false; - } - } - } - else if (bypassDocumentValidationCommandOption() == sourceEl.fieldNameStringData()) { - _shouldBypassValidation = sourceEl.trueValue(); - } + if (_isDocumentsSet) { + BSONArrayBuilder documentsBuilder(builder.subarrayStart(documents())); + for (std::vector<BSONObj>::const_iterator it = _documents.begin(); it != _documents.end(); + ++it) { + documentsBuilder.append(*it); } - - return true; + documentsBuilder.done(); } - void BatchedInsertRequest::clear() { - _collName = NamespaceString(); - _targetNSS = NamespaceString(); - _isCollNameSet = false; - - _documents.clear(); - _isDocumentsSet =false; - - _writeConcern = BSONObj(); - _isWriteConcernSet = false; - - _ordered = false; - _isOrderedSet = false; - - _shouldBypassValidation = false; - - _metadata.reset(); - } - - void BatchedInsertRequest::cloneTo(BatchedInsertRequest* other) const { - other->clear(); - - other->_collName = _collName; - other->_targetNSS = _targetNSS; - other->_isCollNameSet = _isCollNameSet; - - for(std::vector<BSONObj>::const_iterator it = _documents.begin(); - it != _documents.end(); - ++it) { - other->addToDocuments(*it); - } - other->_isDocumentsSet = _isDocumentsSet; - - other->_writeConcern = _writeConcern; - other->_isWriteConcernSet = _isWriteConcernSet; - - other->_ordered = _ordered; - other->_isOrderedSet = _isOrderedSet; - - if (_metadata) { - other->_metadata.reset(new BatchedRequestMetadata()); - _metadata->cloneTo(other->_metadata.get()); + if (_isWriteConcernSet) + builder.append(writeConcern(), _writeConcern); + + if (_isOrderedSet) + builder.append(ordered(), _ordered); + + if (_metadata) + builder.append(metadata(), _metadata->toBSON()); + + if (_shouldBypassValidation) + builder.append(bypassDocumentValidationCommandOption(), true); + + return builder.obj(); +} + +static void extractIndexNSS(const BSONObj& indexDesc, NamespaceString* indexNSS) { + *indexNSS = NamespaceString(indexDesc["ns"].str()); +} + +bool BatchedInsertRequest::parseBSON(const BSONObj& source, string* errMsg) { + clear(); + + std::string dummy; + if (!errMsg) + errMsg = &dummy; + + BSONObjIterator sourceIt(source); + + while (sourceIt.more()) { + BSONElement sourceEl = sourceIt.next(); + + if (collName() == sourceEl.fieldName()) { + std::string temp; + FieldParser::FieldState fieldState = + FieldParser::extract(sourceEl, collName, &temp, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _collName = NamespaceString(temp); + _isCollNameSet = fieldState == FieldParser::FIELD_SET; + } else if (documents() == sourceEl.fieldName()) { + FieldParser::FieldState fieldState = + FieldParser::extract(sourceEl, documents, &_documents, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isDocumentsSet = fieldState == FieldParser::FIELD_SET; + if (_documents.size() >= 1) + extractIndexNSS(_documents.at(0), &_targetNSS); + } else if (writeConcern() == sourceEl.fieldName()) { + FieldParser::FieldState fieldState = + FieldParser::extract(sourceEl, writeConcern, &_writeConcern, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isWriteConcernSet = fieldState == FieldParser::FIELD_SET; + } else if (ordered() == sourceEl.fieldName()) { + FieldParser::FieldState fieldState = + FieldParser::extract(sourceEl, ordered, &_ordered, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isOrderedSet = fieldState == FieldParser::FIELD_SET; + } else if (metadata() == sourceEl.fieldName()) { + BSONObj metadataObj; + FieldParser::FieldState fieldState = + FieldParser::extract(sourceEl, metadata, &metadataObj, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + + if (!metadataObj.isEmpty()) { + _metadata.reset(new BatchedRequestMetadata()); + if (!_metadata->parseBSON(metadataObj, errMsg)) { + return false; + } + } + } else if (bypassDocumentValidationCommandOption() == sourceEl.fieldNameStringData()) { + _shouldBypassValidation = sourceEl.trueValue(); } } - std::string BatchedInsertRequest::toString() const { - return toBSON().toString(); - } + return true; +} - void BatchedInsertRequest::setCollName(StringData collName) { - _collName = NamespaceString(collName); - _isCollNameSet = true; - } +void BatchedInsertRequest::clear() { + _collName = NamespaceString(); + _targetNSS = NamespaceString(); + _isCollNameSet = false; - const std::string& BatchedInsertRequest::getCollName() const { - dassert(_isCollNameSet); - return _collName.ns(); - } + _documents.clear(); + _isDocumentsSet = false; - void BatchedInsertRequest::setCollNameNS(const NamespaceString& collName) { - _collName = collName; - _isCollNameSet = true; - } + _writeConcern = BSONObj(); + _isWriteConcernSet = false; - const NamespaceString& BatchedInsertRequest::getCollNameNS() const { - dassert(_isCollNameSet); - return _collName; - } + _ordered = false; + _isOrderedSet = false; - const NamespaceString& BatchedInsertRequest::getTargetingNSS() const { - return _targetNSS; - } + _shouldBypassValidation = false; - void BatchedInsertRequest::addToDocuments(const BSONObj& documents) { - _documents.push_back(documents); - _isDocumentsSet = true; + _metadata.reset(); +} - if (_documents.size() == 1) - extractIndexNSS(_documents.at(0), &_targetNSS); - } +void BatchedInsertRequest::cloneTo(BatchedInsertRequest* other) const { + other->clear(); - bool BatchedInsertRequest::isDocumentsSet() const { - return _isDocumentsSet; - } + other->_collName = _collName; + other->_targetNSS = _targetNSS; + other->_isCollNameSet = _isCollNameSet; - size_t BatchedInsertRequest::sizeDocuments() const { - return _documents.size(); + for (std::vector<BSONObj>::const_iterator it = _documents.begin(); it != _documents.end(); + ++it) { + other->addToDocuments(*it); } + other->_isDocumentsSet = _isDocumentsSet; - const std::vector<BSONObj>& BatchedInsertRequest::getDocuments() const { - dassert(_isDocumentsSet); - return _documents; - } + other->_writeConcern = _writeConcern; + other->_isWriteConcernSet = _isWriteConcernSet; - const BSONObj& BatchedInsertRequest::getDocumentsAt(size_t pos) const { - dassert(_isDocumentsSet); - dassert(_documents.size() > pos); - return _documents.at(pos); - } + other->_ordered = _ordered; + other->_isOrderedSet = _isOrderedSet; - void BatchedInsertRequest::setDocumentAt(size_t pos, const BSONObj& doc) { - dassert(_isDocumentsSet); - dassert(_documents.size() > pos); - _documents[pos] = doc; + if (_metadata) { + other->_metadata.reset(new BatchedRequestMetadata()); + _metadata->cloneTo(other->_metadata.get()); } - - void BatchedInsertRequest::setWriteConcern(const BSONObj& writeConcern) { - _writeConcern = writeConcern.getOwned(); - _isWriteConcernSet = true; +} + +std::string BatchedInsertRequest::toString() const { + return toBSON().toString(); +} + +void BatchedInsertRequest::setCollName(StringData collName) { + _collName = NamespaceString(collName); + _isCollNameSet = true; +} + +const std::string& BatchedInsertRequest::getCollName() const { + dassert(_isCollNameSet); + return _collName.ns(); +} + +void BatchedInsertRequest::setCollNameNS(const NamespaceString& collName) { + _collName = collName; + _isCollNameSet = true; +} + +const NamespaceString& BatchedInsertRequest::getCollNameNS() const { + dassert(_isCollNameSet); + return _collName; +} + +const NamespaceString& BatchedInsertRequest::getTargetingNSS() const { + return _targetNSS; +} + +void BatchedInsertRequest::addToDocuments(const BSONObj& documents) { + _documents.push_back(documents); + _isDocumentsSet = true; + + if (_documents.size() == 1) + extractIndexNSS(_documents.at(0), &_targetNSS); +} + +bool BatchedInsertRequest::isDocumentsSet() const { + return _isDocumentsSet; +} + +size_t BatchedInsertRequest::sizeDocuments() const { + return _documents.size(); +} + +const std::vector<BSONObj>& BatchedInsertRequest::getDocuments() const { + dassert(_isDocumentsSet); + return _documents; +} + +const BSONObj& BatchedInsertRequest::getDocumentsAt(size_t pos) const { + dassert(_isDocumentsSet); + dassert(_documents.size() > pos); + return _documents.at(pos); +} + +void BatchedInsertRequest::setDocumentAt(size_t pos, const BSONObj& doc) { + dassert(_isDocumentsSet); + dassert(_documents.size() > pos); + _documents[pos] = doc; +} + +void BatchedInsertRequest::setWriteConcern(const BSONObj& writeConcern) { + _writeConcern = writeConcern.getOwned(); + _isWriteConcernSet = true; +} + +void BatchedInsertRequest::unsetWriteConcern() { + _isWriteConcernSet = false; +} + +bool BatchedInsertRequest::isWriteConcernSet() const { + return _isWriteConcernSet; +} + +const BSONObj& BatchedInsertRequest::getWriteConcern() const { + dassert(_isWriteConcernSet); + return _writeConcern; +} + +void BatchedInsertRequest::setOrdered(bool ordered) { + _ordered = ordered; + _isOrderedSet = true; +} + +void BatchedInsertRequest::unsetOrdered() { + _isOrderedSet = false; +} + +bool BatchedInsertRequest::isOrderedSet() const { + return _isOrderedSet; +} + +bool BatchedInsertRequest::getOrdered() const { + if (_isOrderedSet) { + return _ordered; + } else { + return ordered.getDefault(); } +} - void BatchedInsertRequest::unsetWriteConcern() { - _isWriteConcernSet = false; - } +void BatchedInsertRequest::setMetadata(BatchedRequestMetadata* metadata) { + _metadata.reset(metadata); +} - bool BatchedInsertRequest::isWriteConcernSet() const { - return _isWriteConcernSet; - } +void BatchedInsertRequest::unsetMetadata() { + _metadata.reset(); +} - const BSONObj& BatchedInsertRequest::getWriteConcern() const { - dassert(_isWriteConcernSet); - return _writeConcern; - } - - void BatchedInsertRequest::setOrdered(bool ordered) { - _ordered = ordered; - _isOrderedSet = true; - } +bool BatchedInsertRequest::isMetadataSet() const { + return _metadata.get(); +} - void BatchedInsertRequest::unsetOrdered() { - _isOrderedSet = false; - } - - bool BatchedInsertRequest::isOrderedSet() const { - return _isOrderedSet; - } - - bool BatchedInsertRequest::getOrdered() const { - if (_isOrderedSet) { - return _ordered; - } - else { - return ordered.getDefault(); - } - } - - void BatchedInsertRequest::setMetadata(BatchedRequestMetadata* metadata) { - _metadata.reset(metadata); - } - - void BatchedInsertRequest::unsetMetadata() { - _metadata.reset(); - } - - bool BatchedInsertRequest::isMetadataSet() const { - return _metadata.get(); - } - - BatchedRequestMetadata* BatchedInsertRequest::getMetadata() const { - return _metadata.get(); - } +BatchedRequestMetadata* BatchedInsertRequest::getMetadata() const { + return _metadata.get(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_insert_request.h b/src/mongo/s/write_ops/batched_insert_request.h index 8b5238b888e..baf0fbbbe2b 100644 --- a/src/mongo/s/write_ops/batched_insert_request.h +++ b/src/mongo/s/write_ops/batched_insert_request.h @@ -39,114 +39,118 @@ namespace mongo { - /** - * This class represents the layout and content of a batched insert runCommand, - * the request side. +/** + * This class represents the layout and content of a batched insert runCommand, + * the request side. + */ +class BatchedInsertRequest : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedInsertRequest); + +public: + // + // schema declarations + // + + // Name used for the batched insert invocation. + static const std::string BATCHED_INSERT_REQUEST; + + // Field names and types in the batched insert command type. + static const BSONField<std::string> collName; + static const BSONField<std::vector<BSONObj>> documents; + static const BSONField<BSONObj> writeConcern; + static const BSONField<bool> ordered; + static const BSONField<BSONObj> metadata; + + // + // construction / destruction + // + + BatchedInsertRequest(); + virtual ~BatchedInsertRequest(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(BatchedInsertRequest* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setCollName(StringData collName); + void setCollNameNS(const NamespaceString& collName); + const std::string& getCollName() const; + const NamespaceString& getCollNameNS() const; + + const NamespaceString& getTargetingNSS() const; + + void addToDocuments(const BSONObj& documents); + bool isDocumentsSet() const; + std::size_t sizeDocuments() const; + const std::vector<BSONObj>& getDocuments() const; + const BSONObj& getDocumentsAt(std::size_t pos) const; + void setDocumentAt(std::size_t pos, const BSONObj& doc); + + void setWriteConcern(const BSONObj& writeConcern); + void unsetWriteConcern(); + bool isWriteConcernSet() const; + const BSONObj& getWriteConcern() const; + + void setOrdered(bool ordered); + void unsetOrdered(); + bool isOrderedSet() const; + bool getOrdered() const; + + void setShouldBypassValidation(bool newVal) { + _shouldBypassValidation = newVal; + } + bool shouldBypassValidation() const { + return _shouldBypassValidation; + } + + /* + * metadata ownership will be transferred to this. */ - class BatchedInsertRequest : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedInsertRequest); - public: - - // - // schema declarations - // - - // Name used for the batched insert invocation. - static const std::string BATCHED_INSERT_REQUEST; - - // Field names and types in the batched insert command type. - static const BSONField<std::string> collName; - static const BSONField<std::vector<BSONObj> > documents; - static const BSONField<BSONObj> writeConcern; - static const BSONField<bool> ordered; - static const BSONField<BSONObj> metadata; - - // - // construction / destruction - // - - BatchedInsertRequest(); - virtual ~BatchedInsertRequest(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(BatchedInsertRequest* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setCollName(StringData collName); - void setCollNameNS(const NamespaceString& collName); - const std::string& getCollName() const; - const NamespaceString& getCollNameNS() const; - - const NamespaceString& getTargetingNSS() const; - - void addToDocuments(const BSONObj& documents); - bool isDocumentsSet() const; - std::size_t sizeDocuments() const; - const std::vector<BSONObj>& getDocuments() const; - const BSONObj& getDocumentsAt(std::size_t pos) const; - void setDocumentAt(std::size_t pos, const BSONObj& doc); - - void setWriteConcern(const BSONObj& writeConcern); - void unsetWriteConcern(); - bool isWriteConcernSet() const; - const BSONObj& getWriteConcern() const; - - void setOrdered(bool ordered); - void unsetOrdered(); - bool isOrderedSet() const; - bool getOrdered() const; - - void setShouldBypassValidation(bool newVal) { _shouldBypassValidation = newVal; } - bool shouldBypassValidation() const { return _shouldBypassValidation; } - - /* - * metadata ownership will be transferred to this. - */ - void setMetadata(BatchedRequestMetadata* metadata); - void unsetMetadata(); - bool isMetadataSet() const; - BatchedRequestMetadata* getMetadata() const; - - private: - // Convention: (M)andatory, (O)ptional - - // (M) collection we're inserting on - NamespaceString _collName; - bool _isCollNameSet; - - // (M) array of documents to be inserted - std::vector<BSONObj> _documents; - bool _isDocumentsSet; - - // (O) to be issued after the batch applied - BSONObj _writeConcern; - bool _isWriteConcernSet; - - // (O) whether batch is issued in parallel or not - bool _ordered; - bool _isOrderedSet; - - // (O) metadata associated with this request for internal use. - std::unique_ptr<BatchedRequestMetadata> _metadata; - - // (O) cached copied of target ns - NamespaceString _targetNSS; - - // (O) should document validation be bypassed (default false) - bool _shouldBypassValidation; - }; - -} // namespace mongo + void setMetadata(BatchedRequestMetadata* metadata); + void unsetMetadata(); + bool isMetadataSet() const; + BatchedRequestMetadata* getMetadata() const; + +private: + // Convention: (M)andatory, (O)ptional + + // (M) collection we're inserting on + NamespaceString _collName; + bool _isCollNameSet; + + // (M) array of documents to be inserted + std::vector<BSONObj> _documents; + bool _isDocumentsSet; + + // (O) to be issued after the batch applied + BSONObj _writeConcern; + bool _isWriteConcernSet; + + // (O) whether batch is issued in parallel or not + bool _ordered; + bool _isOrderedSet; + + // (O) metadata associated with this request for internal use. + std::unique_ptr<BatchedRequestMetadata> _metadata; + + // (O) cached copied of target ns + NamespaceString _targetNSS; + + // (O) should document validation be bypassed (default false) + bool _shouldBypassValidation; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_insert_request_test.cpp b/src/mongo/s/write_ops/batched_insert_request_test.cpp index 5be837ae17a..66f99c7ffc6 100644 --- a/src/mongo/s/write_ops/batched_insert_request_test.cpp +++ b/src/mongo/s/write_ops/batched_insert_request_test.cpp @@ -36,115 +36,111 @@ namespace { - using namespace mongo; - using std::unique_ptr; - using std::string; - - TEST(RoundTrip, Normal) { - BSONArray insertArray = BSON_ARRAY(BSON("a" << 1) << BSON("b" << 1)); - - BSONObj writeConcernObj = BSON("w" << 1); - - // The BSON_ARRAY macro doesn't support Timestamps. - BSONArrayBuilder arrBuilder; - arrBuilder.append(Timestamp(1,1)); - arrBuilder.append(OID::gen()); - BSONArray shardVersionArray = arrBuilder.arr(); - - BSONObj origInsertRequestObj = - BSON(BatchedInsertRequest::collName("test") << - BatchedInsertRequest::documents() << insertArray << - BatchedInsertRequest::writeConcern(writeConcernObj) << - BatchedInsertRequest::ordered(true) << - BatchedInsertRequest::metadata() << BSON( - BatchedRequestMetadata::shardName("shard0000") << - BatchedRequestMetadata::shardVersion() << shardVersionArray << - BatchedRequestMetadata::session(0))); - - string errMsg; - BatchedInsertRequest request; - bool ok = request.parseBSON(origInsertRequestObj, &errMsg); - ASSERT_TRUE(ok); - - BSONObj genInsertRequestObj = request.toBSON(); - ASSERT_EQUALS(0, genInsertRequestObj.woCompare(origInsertRequestObj)); - } - - TEST(GenID, All) { - - BatchedCommandRequest cmdRequest(BatchedCommandRequest::BatchType_Insert); - BatchedInsertRequest& request = *cmdRequest.getInsertRequest(); - - request.setCollName("foo.bar"); - request.setOrdered(false); - - BSONObj insertA = BSON( "a" << 1 ); - BSONObj insertB = BSON( "b" << 1 ); - request.addToDocuments(insertA); - request.addToDocuments(insertB); - - unique_ptr<BatchedCommandRequest> idCmdRequest; - idCmdRequest.reset(BatchedCommandRequest::cloneWithIds(cmdRequest)); - ASSERT(idCmdRequest.get()); - - BatchedInsertRequest* idRequest = idCmdRequest->getInsertRequest(); - ASSERT_EQUALS(idRequest->getCollName(), request.getCollName()); - ASSERT_EQUALS(idRequest->getOrdered(), request.getOrdered()); - - ASSERT(!idRequest->getDocumentsAt(0)["_id"].eoo()); - ASSERT_EQUALS(idRequest->getDocumentsAt(0).nFields(), 2); - ASSERT(!idRequest->getDocumentsAt(1)["_id"].eoo()); - ASSERT_EQUALS(idRequest->getDocumentsAt(1).nFields(), 2); - } - - TEST(GenID, Partial) { - - BatchedCommandRequest cmdRequest(BatchedCommandRequest::BatchType_Insert); - BatchedInsertRequest& request = *cmdRequest.getInsertRequest(); - - request.setCollName("foo.bar"); - request.setOrdered(false); - - BSONObj insertA = BSON( "a" << 1 ); - BSONObj insertB = BSON( "b" << 1 << "_id" << 1 ); - BSONObj insertC = BSON( "c" << 1 ); - request.addToDocuments(insertA); - request.addToDocuments(insertB); - request.addToDocuments(insertC); - - unique_ptr<BatchedCommandRequest> idCmdRequest; - idCmdRequest.reset(BatchedCommandRequest::cloneWithIds(cmdRequest)); - ASSERT(idCmdRequest.get()); - - BatchedInsertRequest* idRequest = idCmdRequest->getInsertRequest(); - ASSERT_EQUALS(idRequest->getCollName(), request.getCollName()); - ASSERT_EQUALS(idRequest->getOrdered(), request.getOrdered()); - - ASSERT(!idRequest->getDocumentsAt(0)["_id"].eoo()); - ASSERT_EQUALS(idRequest->getDocumentsAt(0).nFields(), 2); - ASSERT(!idRequest->getDocumentsAt(1)["_id"].eoo()); - ASSERT_EQUALS(idRequest->getDocumentsAt(1).nFields(), 2); - ASSERT(!idRequest->getDocumentsAt(2)["_id"].eoo()); - ASSERT_EQUALS(idRequest->getDocumentsAt(1).nFields(), 2); - } - - TEST(GenID, None) { - - BatchedCommandRequest cmdRequest(BatchedCommandRequest::BatchType_Insert); - BatchedInsertRequest& request = *cmdRequest.getInsertRequest(); - - // We need to check for system.indexes namespace - request.setCollName("foo.bar"); - - BSONObj insertA = BSON( "_id" << 0 << "a" << 1 ); - BSONObj insertB = BSON( "b" << 1 << "_id" << 1 ); - request.addToDocuments(insertA); - request.addToDocuments(insertB); - - unique_ptr<BatchedCommandRequest> idCmdRequest; - idCmdRequest.reset(BatchedCommandRequest::cloneWithIds(cmdRequest)); - ASSERT(!idCmdRequest.get()); - } - - -} // unnamed namespace +using namespace mongo; +using std::unique_ptr; +using std::string; + +TEST(RoundTrip, Normal) { + BSONArray insertArray = BSON_ARRAY(BSON("a" << 1) << BSON("b" << 1)); + + BSONObj writeConcernObj = BSON("w" << 1); + + // The BSON_ARRAY macro doesn't support Timestamps. + BSONArrayBuilder arrBuilder; + arrBuilder.append(Timestamp(1, 1)); + arrBuilder.append(OID::gen()); + BSONArray shardVersionArray = arrBuilder.arr(); + + BSONObj origInsertRequestObj = + BSON(BatchedInsertRequest::collName("test") + << BatchedInsertRequest::documents() << insertArray + << BatchedInsertRequest::writeConcern(writeConcernObj) + << BatchedInsertRequest::ordered(true) << BatchedInsertRequest::metadata() + << BSON(BatchedRequestMetadata::shardName("shard0000") + << BatchedRequestMetadata::shardVersion() << shardVersionArray + << BatchedRequestMetadata::session(0))); + + string errMsg; + BatchedInsertRequest request; + bool ok = request.parseBSON(origInsertRequestObj, &errMsg); + ASSERT_TRUE(ok); + + BSONObj genInsertRequestObj = request.toBSON(); + ASSERT_EQUALS(0, genInsertRequestObj.woCompare(origInsertRequestObj)); +} + +TEST(GenID, All) { + BatchedCommandRequest cmdRequest(BatchedCommandRequest::BatchType_Insert); + BatchedInsertRequest& request = *cmdRequest.getInsertRequest(); + + request.setCollName("foo.bar"); + request.setOrdered(false); + + BSONObj insertA = BSON("a" << 1); + BSONObj insertB = BSON("b" << 1); + request.addToDocuments(insertA); + request.addToDocuments(insertB); + + unique_ptr<BatchedCommandRequest> idCmdRequest; + idCmdRequest.reset(BatchedCommandRequest::cloneWithIds(cmdRequest)); + ASSERT(idCmdRequest.get()); + + BatchedInsertRequest* idRequest = idCmdRequest->getInsertRequest(); + ASSERT_EQUALS(idRequest->getCollName(), request.getCollName()); + ASSERT_EQUALS(idRequest->getOrdered(), request.getOrdered()); + + ASSERT(!idRequest->getDocumentsAt(0)["_id"].eoo()); + ASSERT_EQUALS(idRequest->getDocumentsAt(0).nFields(), 2); + ASSERT(!idRequest->getDocumentsAt(1)["_id"].eoo()); + ASSERT_EQUALS(idRequest->getDocumentsAt(1).nFields(), 2); +} + +TEST(GenID, Partial) { + BatchedCommandRequest cmdRequest(BatchedCommandRequest::BatchType_Insert); + BatchedInsertRequest& request = *cmdRequest.getInsertRequest(); + + request.setCollName("foo.bar"); + request.setOrdered(false); + + BSONObj insertA = BSON("a" << 1); + BSONObj insertB = BSON("b" << 1 << "_id" << 1); + BSONObj insertC = BSON("c" << 1); + request.addToDocuments(insertA); + request.addToDocuments(insertB); + request.addToDocuments(insertC); + + unique_ptr<BatchedCommandRequest> idCmdRequest; + idCmdRequest.reset(BatchedCommandRequest::cloneWithIds(cmdRequest)); + ASSERT(idCmdRequest.get()); + + BatchedInsertRequest* idRequest = idCmdRequest->getInsertRequest(); + ASSERT_EQUALS(idRequest->getCollName(), request.getCollName()); + ASSERT_EQUALS(idRequest->getOrdered(), request.getOrdered()); + + ASSERT(!idRequest->getDocumentsAt(0)["_id"].eoo()); + ASSERT_EQUALS(idRequest->getDocumentsAt(0).nFields(), 2); + ASSERT(!idRequest->getDocumentsAt(1)["_id"].eoo()); + ASSERT_EQUALS(idRequest->getDocumentsAt(1).nFields(), 2); + ASSERT(!idRequest->getDocumentsAt(2)["_id"].eoo()); + ASSERT_EQUALS(idRequest->getDocumentsAt(1).nFields(), 2); +} + +TEST(GenID, None) { + BatchedCommandRequest cmdRequest(BatchedCommandRequest::BatchType_Insert); + BatchedInsertRequest& request = *cmdRequest.getInsertRequest(); + + // We need to check for system.indexes namespace + request.setCollName("foo.bar"); + + BSONObj insertA = BSON("_id" << 0 << "a" << 1); + BSONObj insertB = BSON("b" << 1 << "_id" << 1); + request.addToDocuments(insertA); + request.addToDocuments(insertB); + + unique_ptr<BatchedCommandRequest> idCmdRequest; + idCmdRequest.reset(BatchedCommandRequest::cloneWithIds(cmdRequest)); + ASSERT(!idCmdRequest.get()); +} + + +} // unnamed namespace diff --git a/src/mongo/s/write_ops/batched_request_metadata.cpp b/src/mongo/s/write_ops/batched_request_metadata.cpp index 29d4b5e97fc..a2e21f99e4c 100644 --- a/src/mongo/s/write_ops/batched_request_metadata.cpp +++ b/src/mongo/s/write_ops/batched_request_metadata.cpp @@ -32,146 +32,147 @@ namespace mongo { - using std::unique_ptr; - using std::string; +using std::unique_ptr; +using std::string; - const BSONField<string> BatchedRequestMetadata::shardName("shardName"); - const BSONField<ChunkVersion> BatchedRequestMetadata::shardVersion("shardVersion"); - const BSONField<long long> BatchedRequestMetadata::session("session"); +const BSONField<string> BatchedRequestMetadata::shardName("shardName"); +const BSONField<ChunkVersion> BatchedRequestMetadata::shardVersion("shardVersion"); +const BSONField<long long> BatchedRequestMetadata::session("session"); - BatchedRequestMetadata::BatchedRequestMetadata(): - _isShardNameSet(false), - _session(0), - _isSessionSet(false) { - } - - BatchedRequestMetadata::~BatchedRequestMetadata() { - - } - - bool BatchedRequestMetadata::isValid(string* errMsg) const { - // all fields are mandatory. - return true; - } +BatchedRequestMetadata::BatchedRequestMetadata() + : _isShardNameSet(false), _session(0), _isSessionSet(false) {} - BSONObj BatchedRequestMetadata::toBSON() const { - BSONObjBuilder metadataBuilder; +BatchedRequestMetadata::~BatchedRequestMetadata() {} - if (_isShardNameSet) metadataBuilder << shardName(_shardName); +bool BatchedRequestMetadata::isValid(string* errMsg) const { + // all fields are mandatory. + return true; +} - if (_shardVersion.get()) { - // ChunkVersion wants to be an array. - metadataBuilder.append(shardVersion(), - static_cast<BSONArray>(_shardVersion->toBSON())); - } +BSONObj BatchedRequestMetadata::toBSON() const { + BSONObjBuilder metadataBuilder; - if (_isSessionSet) metadataBuilder << session(_session); + if (_isShardNameSet) + metadataBuilder << shardName(_shardName); - return metadataBuilder.obj(); + if (_shardVersion.get()) { + // ChunkVersion wants to be an array. + metadataBuilder.append(shardVersion(), static_cast<BSONArray>(_shardVersion->toBSON())); } - bool BatchedRequestMetadata::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + if (_isSessionSet) + metadataBuilder << session(_session); - string dummy; - if (!errMsg) errMsg = &dummy; + return metadataBuilder.obj(); +} - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, shardName, &_shardName, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isShardNameSet = fieldState == FieldParser::FIELD_SET; +bool BatchedRequestMetadata::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - { - std::unique_ptr<ChunkVersion> tempChunkVersion(new ChunkVersion); - fieldState = FieldParser::extract(source, shardVersion, - tempChunkVersion.get(), errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - if (fieldState == FieldParser::FIELD_SET) _shardVersion.swap(tempChunkVersion); - } + string dummy; + if (!errMsg) + errMsg = &dummy; - fieldState = FieldParser::extract(source, session, &_session, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isSessionSet = fieldState == FieldParser::FIELD_SET; + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, shardName, &_shardName, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isShardNameSet = fieldState == FieldParser::FIELD_SET; - return true; + { + std::unique_ptr<ChunkVersion> tempChunkVersion(new ChunkVersion); + fieldState = FieldParser::extract(source, shardVersion, tempChunkVersion.get(), errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + if (fieldState == FieldParser::FIELD_SET) + _shardVersion.swap(tempChunkVersion); } - void BatchedRequestMetadata::clear() { - _shardName.clear(); - _isShardNameSet = false; + fieldState = FieldParser::extract(source, session, &_session, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isSessionSet = fieldState == FieldParser::FIELD_SET; - _shardVersion.reset(); + return true; +} - _session = 0; - _isSessionSet = false; - } +void BatchedRequestMetadata::clear() { + _shardName.clear(); + _isShardNameSet = false; - string BatchedRequestMetadata::toString() const { - return toBSON().toString(); - } + _shardVersion.reset(); - void BatchedRequestMetadata::cloneTo(BatchedRequestMetadata* other) const { - other->_shardName = _shardName; - other->_isShardNameSet = _isShardNameSet; + _session = 0; + _isSessionSet = false; +} - if (other->_shardVersion.get()) _shardVersion->cloneTo(other->_shardVersion.get()); +string BatchedRequestMetadata::toString() const { + return toBSON().toString(); +} - other->_session = _session; - other->_isSessionSet = _isSessionSet; - } +void BatchedRequestMetadata::cloneTo(BatchedRequestMetadata* other) const { + other->_shardName = _shardName; + other->_isShardNameSet = _isShardNameSet; - void BatchedRequestMetadata::setShardName( StringData shardName ) { - _shardName = shardName.toString(); - _isShardNameSet = true; - } + if (other->_shardVersion.get()) + _shardVersion->cloneTo(other->_shardVersion.get()); - void BatchedRequestMetadata::unsetShardName() { - _isShardNameSet = false; - } + other->_session = _session; + other->_isSessionSet = _isSessionSet; +} - bool BatchedRequestMetadata::isShardNameSet() const { - return _isShardNameSet; - } +void BatchedRequestMetadata::setShardName(StringData shardName) { + _shardName = shardName.toString(); + _isShardNameSet = true; +} - const string& BatchedRequestMetadata::getShardName() const { - dassert( _isShardNameSet ); - return _shardName; - } +void BatchedRequestMetadata::unsetShardName() { + _isShardNameSet = false; +} - void BatchedRequestMetadata::setShardVersion(const ChunkVersion& shardVersion) { - unique_ptr<ChunkVersion> temp(new ChunkVersion); - shardVersion.cloneTo(temp.get()); - _shardVersion.reset(temp.release()); - } +bool BatchedRequestMetadata::isShardNameSet() const { + return _isShardNameSet; +} - void BatchedRequestMetadata::unsetShardVersion() { - _shardVersion.reset(); - } +const string& BatchedRequestMetadata::getShardName() const { + dassert(_isShardNameSet); + return _shardName; +} - bool BatchedRequestMetadata::isShardVersionSet() const { - return _shardVersion.get() != NULL; - } +void BatchedRequestMetadata::setShardVersion(const ChunkVersion& shardVersion) { + unique_ptr<ChunkVersion> temp(new ChunkVersion); + shardVersion.cloneTo(temp.get()); + _shardVersion.reset(temp.release()); +} - const ChunkVersion& BatchedRequestMetadata::getShardVersion() const { - dassert(_shardVersion.get()); - return *_shardVersion; - } +void BatchedRequestMetadata::unsetShardVersion() { + _shardVersion.reset(); +} - void BatchedRequestMetadata::setSession(long long session) { - _session = session; - _isSessionSet = true; - } +bool BatchedRequestMetadata::isShardVersionSet() const { + return _shardVersion.get() != NULL; +} - void BatchedRequestMetadata::unsetSession() { - _isSessionSet = false; - } +const ChunkVersion& BatchedRequestMetadata::getShardVersion() const { + dassert(_shardVersion.get()); + return *_shardVersion; +} - bool BatchedRequestMetadata::isSessionSet() const { - return _isSessionSet; - } +void BatchedRequestMetadata::setSession(long long session) { + _session = session; + _isSessionSet = true; +} - long long BatchedRequestMetadata::getSession() const { - dassert(_isSessionSet); - return _session; - } +void BatchedRequestMetadata::unsetSession() { + _isSessionSet = false; +} + +bool BatchedRequestMetadata::isSessionSet() const { + return _isSessionSet; +} + +long long BatchedRequestMetadata::getSession() const { + dassert(_isSessionSet); + return _session; +} } diff --git a/src/mongo/s/write_ops/batched_request_metadata.h b/src/mongo/s/write_ops/batched_request_metadata.h index afe639adfaa..6e31cb60713 100644 --- a/src/mongo/s/write_ops/batched_request_metadata.h +++ b/src/mongo/s/write_ops/batched_request_metadata.h @@ -36,59 +36,58 @@ #include "mongo/s/chunk_version.h" namespace mongo { - class BatchedRequestMetadata : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedRequestMetadata); - public: - - static const BSONField<std::string> shardName; - static const BSONField<ChunkVersion> shardVersion; - static const BSONField<long long> session; - - BatchedRequestMetadata(); - virtual ~BatchedRequestMetadata(); - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - void cloneTo(BatchedRequestMetadata* other) const; - - // - // individual field accessors - // - - void setShardName(StringData shardName); - void unsetShardName(); - bool isShardNameSet() const; - const std::string& getShardName() const; - - void setShardVersion(const ChunkVersion& shardVersion); - void unsetShardVersion(); - bool isShardVersionSet() const; - const ChunkVersion& getShardVersion() const; - - void setSession(long long session); - void unsetSession(); - bool isSessionSet() const; - long long getSession() const; - - private: - - // (O) shard name we're sending this batch to - std::string _shardName; - bool _isShardNameSet; - - // (O) version for this collection on a given shard - std::unique_ptr<ChunkVersion> _shardVersion; - - // (O) session number the inserts belong to - long long _session; - bool _isSessionSet; - }; +class BatchedRequestMetadata : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedRequestMetadata); + +public: + static const BSONField<std::string> shardName; + static const BSONField<ChunkVersion> shardVersion; + static const BSONField<long long> session; + + BatchedRequestMetadata(); + virtual ~BatchedRequestMetadata(); + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + void cloneTo(BatchedRequestMetadata* other) const; + + // + // individual field accessors + // + + void setShardName(StringData shardName); + void unsetShardName(); + bool isShardNameSet() const; + const std::string& getShardName() const; + + void setShardVersion(const ChunkVersion& shardVersion); + void unsetShardVersion(); + bool isShardVersionSet() const; + const ChunkVersion& getShardVersion() const; + + void setSession(long long session); + void unsetSession(); + bool isSessionSet() const; + long long getSession() const; + +private: + // (O) shard name we're sending this batch to + std::string _shardName; + bool _isShardNameSet; + + // (O) version for this collection on a given shard + std::unique_ptr<ChunkVersion> _shardVersion; + + // (O) session number the inserts belong to + long long _session; + bool _isSessionSet; +}; } diff --git a/src/mongo/s/write_ops/batched_request_metadata_test.cpp b/src/mongo/s/write_ops/batched_request_metadata_test.cpp index 562497b5d27..e0ca03231cc 100644 --- a/src/mongo/s/write_ops/batched_request_metadata_test.cpp +++ b/src/mongo/s/write_ops/batched_request_metadata_test.cpp @@ -22,32 +22,32 @@ namespace { - using mongo::BSONArray; - using mongo::BSONArrayBuilder; - using mongo::BSONObj; - using mongo::BatchedRequestMetadata; - using mongo::OID; - using mongo::Timestamp; - using std::string; - - TEST(RoundTrip, Normal) { - // The BSON_ARRAY macro doesn't support Timestamps. - BSONArrayBuilder arrBuilder; - arrBuilder.append(Timestamp(1,1)); - arrBuilder.append(OID::gen()); - BSONArray shardVersionArray = arrBuilder.arr(); - - BSONObj metadataObj(BSON(BatchedRequestMetadata::shardName("shard0000") << - BatchedRequestMetadata::shardVersion() << shardVersionArray << - BatchedRequestMetadata::session(100))); - - string errMsg; - BatchedRequestMetadata metadata; - bool ok = metadata.parseBSON(metadataObj, &errMsg); - ASSERT_TRUE(ok); - - BSONObj genMetadataObj = metadata.toBSON(); - ASSERT_EQUALS(0, genMetadataObj.woCompare(metadataObj)); - } - -} // unnamed namespace +using mongo::BSONArray; +using mongo::BSONArrayBuilder; +using mongo::BSONObj; +using mongo::BatchedRequestMetadata; +using mongo::OID; +using mongo::Timestamp; +using std::string; + +TEST(RoundTrip, Normal) { + // The BSON_ARRAY macro doesn't support Timestamps. + BSONArrayBuilder arrBuilder; + arrBuilder.append(Timestamp(1, 1)); + arrBuilder.append(OID::gen()); + BSONArray shardVersionArray = arrBuilder.arr(); + + BSONObj metadataObj(BSON(BatchedRequestMetadata::shardName("shard0000") + << BatchedRequestMetadata::shardVersion() << shardVersionArray + << BatchedRequestMetadata::session(100))); + + string errMsg; + BatchedRequestMetadata metadata; + bool ok = metadata.parseBSON(metadataObj, &errMsg); + ASSERT_TRUE(ok); + + BSONObj genMetadataObj = metadata.toBSON(); + ASSERT_EQUALS(0, genMetadataObj.woCompare(metadataObj)); +} + +} // unnamed namespace diff --git a/src/mongo/s/write_ops/batched_update_document.cpp b/src/mongo/s/write_ops/batched_update_document.cpp index ffebede8787..7e388a78b32 100644 --- a/src/mongo/s/write_ops/batched_update_document.cpp +++ b/src/mongo/s/write_ops/batched_update_document.cpp @@ -33,207 +33,209 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; +using mongoutils::str::stream; - const BSONField<BSONObj> BatchedUpdateDocument::query("q"); - const BSONField<BSONObj> BatchedUpdateDocument::updateExpr("u"); - const BSONField<bool> BatchedUpdateDocument::multi("multi", false); - const BSONField<bool> BatchedUpdateDocument::upsert("upsert", false); +const BSONField<BSONObj> BatchedUpdateDocument::query("q"); +const BSONField<BSONObj> BatchedUpdateDocument::updateExpr("u"); +const BSONField<bool> BatchedUpdateDocument::multi("multi", false); +const BSONField<bool> BatchedUpdateDocument::upsert("upsert", false); - BatchedUpdateDocument::BatchedUpdateDocument() { - clear(); +BatchedUpdateDocument::BatchedUpdateDocument() { + clear(); +} + +BatchedUpdateDocument::~BatchedUpdateDocument() {} + +bool BatchedUpdateDocument::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BatchedUpdateDocument::~BatchedUpdateDocument() { + // All the mandatory fields must be present. + if (!_isQuerySet) { + *errMsg = stream() << "missing " << query.name() << " field"; + return false; } - bool BatchedUpdateDocument::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } + if (!_isUpdateExprSet) { + *errMsg = stream() << "missing " << updateExpr.name() << " field"; + return false; + } - // All the mandatory fields must be present. - if (!_isQuerySet) { - *errMsg = stream() << "missing " << query.name() << " field"; - return false; - } + return true; +} - if (!_isUpdateExprSet) { - *errMsg = stream() << "missing " << updateExpr.name() << " field"; - return false; - } +BSONObj BatchedUpdateDocument::toBSON() const { + BSONObjBuilder builder; - return true; - } + if (_isQuerySet) + builder.append(query(), _query); - BSONObj BatchedUpdateDocument::toBSON() const { - BSONObjBuilder builder; + if (_isUpdateExprSet) + builder.append(updateExpr(), _updateExpr); - if (_isQuerySet) builder.append(query(), _query); + if (_isMultiSet) + builder.append(multi(), _multi); - if (_isUpdateExprSet) builder.append(updateExpr(), _updateExpr); + if (_isUpsertSet) + builder.append(upsert(), _upsert); - if (_isMultiSet) builder.append(multi(), _multi); + return builder.obj(); +} - if (_isUpsertSet) builder.append(upsert(), _upsert); +bool BatchedUpdateDocument::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - return builder.obj(); - } + std::string dummy; + if (!errMsg) + errMsg = &dummy; - bool BatchedUpdateDocument::parseBSON(const BSONObj& source, string* errMsg) { - clear(); - - std::string dummy; - if (!errMsg) errMsg = &dummy; - - FieldParser::FieldState fieldState; - - BSONObjIterator it(source); - while ( it.more() ) { - BSONElement elem = it.next(); - StringData fieldName = elem.fieldNameStringData(); - - if ( fieldName == query.name() ) { - fieldState = FieldParser::extract(elem, query, &_query, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isQuerySet = fieldState == FieldParser::FIELD_SET; - } - else if ( fieldName == updateExpr.name() ) { - fieldState = FieldParser::extract(elem, updateExpr, &_updateExpr, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isUpdateExprSet = fieldState == FieldParser::FIELD_SET; - } - else if ( fieldName == multi.name() ) { - fieldState = FieldParser::extract(elem, multi, &_multi, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isMultiSet = fieldState == FieldParser::FIELD_SET; - } - else if ( fieldName == upsert.name() ) { - fieldState = FieldParser::extract(elem, upsert, &_upsert, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isUpsertSet = fieldState == FieldParser::FIELD_SET; - } - } + FieldParser::FieldState fieldState; - return true; + BSONObjIterator it(source); + while (it.more()) { + BSONElement elem = it.next(); + StringData fieldName = elem.fieldNameStringData(); + + if (fieldName == query.name()) { + fieldState = FieldParser::extract(elem, query, &_query, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isQuerySet = fieldState == FieldParser::FIELD_SET; + } else if (fieldName == updateExpr.name()) { + fieldState = FieldParser::extract(elem, updateExpr, &_updateExpr, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isUpdateExprSet = fieldState == FieldParser::FIELD_SET; + } else if (fieldName == multi.name()) { + fieldState = FieldParser::extract(elem, multi, &_multi, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isMultiSet = fieldState == FieldParser::FIELD_SET; + } else if (fieldName == upsert.name()) { + fieldState = FieldParser::extract(elem, upsert, &_upsert, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isUpsertSet = fieldState == FieldParser::FIELD_SET; + } } - void BatchedUpdateDocument::clear() { - _query = BSONObj(); - _isQuerySet = false; + return true; +} - _updateExpr = BSONObj(); - _isUpdateExprSet = false; +void BatchedUpdateDocument::clear() { + _query = BSONObj(); + _isQuerySet = false; - _multi = false; - _isMultiSet = false; + _updateExpr = BSONObj(); + _isUpdateExprSet = false; - _upsert = false; - _isUpsertSet = false; + _multi = false; + _isMultiSet = false; - } + _upsert = false; + _isUpsertSet = false; +} - void BatchedUpdateDocument::cloneTo(BatchedUpdateDocument* other) const { - other->clear(); +void BatchedUpdateDocument::cloneTo(BatchedUpdateDocument* other) const { + other->clear(); - other->_query = _query; - other->_isQuerySet = _isQuerySet; + other->_query = _query; + other->_isQuerySet = _isQuerySet; - other->_updateExpr = _updateExpr; - other->_isUpdateExprSet = _isUpdateExprSet; + other->_updateExpr = _updateExpr; + other->_isUpdateExprSet = _isUpdateExprSet; - other->_multi = _multi; - other->_isMultiSet = _isMultiSet; + other->_multi = _multi; + other->_isMultiSet = _isMultiSet; - other->_upsert = _upsert; - other->_isUpsertSet = _isUpsertSet; - } + other->_upsert = _upsert; + other->_isUpsertSet = _isUpsertSet; +} - std::string BatchedUpdateDocument::toString() const { - return toBSON().toString(); - } +std::string BatchedUpdateDocument::toString() const { + return toBSON().toString(); +} - void BatchedUpdateDocument::setQuery(const BSONObj& query) { - _query = query.getOwned(); - _isQuerySet = true; - } +void BatchedUpdateDocument::setQuery(const BSONObj& query) { + _query = query.getOwned(); + _isQuerySet = true; +} - void BatchedUpdateDocument::unsetQuery() { - _isQuerySet = false; - } +void BatchedUpdateDocument::unsetQuery() { + _isQuerySet = false; +} - bool BatchedUpdateDocument::isQuerySet() const { - return _isQuerySet; - } +bool BatchedUpdateDocument::isQuerySet() const { + return _isQuerySet; +} - const BSONObj& BatchedUpdateDocument::getQuery() const { - dassert(_isQuerySet); - return _query; - } +const BSONObj& BatchedUpdateDocument::getQuery() const { + dassert(_isQuerySet); + return _query; +} - void BatchedUpdateDocument::setUpdateExpr(const BSONObj& updateExpr) { - _updateExpr = updateExpr.getOwned(); - _isUpdateExprSet = true; - } +void BatchedUpdateDocument::setUpdateExpr(const BSONObj& updateExpr) { + _updateExpr = updateExpr.getOwned(); + _isUpdateExprSet = true; +} - void BatchedUpdateDocument::unsetUpdateExpr() { - _isUpdateExprSet = false; - } +void BatchedUpdateDocument::unsetUpdateExpr() { + _isUpdateExprSet = false; +} - bool BatchedUpdateDocument::isUpdateExprSet() const { - return _isUpdateExprSet; - } +bool BatchedUpdateDocument::isUpdateExprSet() const { + return _isUpdateExprSet; +} - const BSONObj& BatchedUpdateDocument::getUpdateExpr() const { - dassert(_isUpdateExprSet); - return _updateExpr; - } +const BSONObj& BatchedUpdateDocument::getUpdateExpr() const { + dassert(_isUpdateExprSet); + return _updateExpr; +} - void BatchedUpdateDocument::setMulti(bool multi) { - _multi = multi; - _isMultiSet = true; - } +void BatchedUpdateDocument::setMulti(bool multi) { + _multi = multi; + _isMultiSet = true; +} - void BatchedUpdateDocument::unsetMulti() { - _isMultiSet = false; - } +void BatchedUpdateDocument::unsetMulti() { + _isMultiSet = false; +} - bool BatchedUpdateDocument::isMultiSet() const { - return _isMultiSet; - } +bool BatchedUpdateDocument::isMultiSet() const { + return _isMultiSet; +} - bool BatchedUpdateDocument::getMulti() const { - if (_isMultiSet) { - return _multi; - } - else { - return multi.getDefault(); - } +bool BatchedUpdateDocument::getMulti() const { + if (_isMultiSet) { + return _multi; + } else { + return multi.getDefault(); } +} - void BatchedUpdateDocument::setUpsert(bool upsert) { - _upsert = upsert; - _isUpsertSet = true; - } +void BatchedUpdateDocument::setUpsert(bool upsert) { + _upsert = upsert; + _isUpsertSet = true; +} - void BatchedUpdateDocument::unsetUpsert() { - _isUpsertSet = false; - } +void BatchedUpdateDocument::unsetUpsert() { + _isUpsertSet = false; +} - bool BatchedUpdateDocument::isUpsertSet() const { - return _isUpsertSet; - } +bool BatchedUpdateDocument::isUpsertSet() const { + return _isUpsertSet; +} - bool BatchedUpdateDocument::getUpsert() const { - if (_isUpsertSet) { - return _upsert; - } - else { - return upsert.getDefault(); - } +bool BatchedUpdateDocument::getUpsert() const { + if (_isUpsertSet) { + return _upsert; + } else { + return upsert.getDefault(); } +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_update_document.h b/src/mongo/s/write_ops/batched_update_document.h index 0830e6c7f3c..ed9b6bc275b 100644 --- a/src/mongo/s/write_ops/batched_update_document.h +++ b/src/mongo/s/write_ops/batched_update_document.h @@ -37,85 +37,85 @@ namespace mongo { - /** - * This class represents the layout and content of a update document runCommand, - * in the request side. - */ - class BatchedUpdateDocument : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedUpdateDocument); - public: - - // - // schema declarations - // - - static const BSONField<BSONObj> query; - static const BSONField<BSONObj> updateExpr; - static const BSONField<bool> multi; - static const BSONField<bool> upsert; - - // - // construction / destruction - // - - BatchedUpdateDocument(); - virtual ~BatchedUpdateDocument(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(BatchedUpdateDocument* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setQuery(const BSONObj& query); - void unsetQuery(); - bool isQuerySet() const; - const BSONObj& getQuery() const; - - void setUpdateExpr(const BSONObj& updateExpr); - void unsetUpdateExpr(); - bool isUpdateExprSet() const; - const BSONObj& getUpdateExpr() const; - - void setMulti(bool multi); - void unsetMulti(); - bool isMultiSet() const; - bool getMulti() const; - - void setUpsert(bool upsert); - void unsetUpsert(); - bool isUpsertSet() const; - bool getUpsert() const; - - private: - // Convention: (M)andatory, (O)ptional - - // (M) query whose result the update will manipulate - BSONObj _query; - bool _isQuerySet; - - // (M) the update expression itself - BSONObj _updateExpr; - bool _isUpdateExprSet; - - // (O) whether multiple documents are to be updated - bool _multi; - bool _isMultiSet; - - // (O) whether upserts are allowed - bool _upsert; - bool _isUpsertSet; - }; - -} // namespace mongo +/** + * This class represents the layout and content of a update document runCommand, + * in the request side. + */ +class BatchedUpdateDocument : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedUpdateDocument); + +public: + // + // schema declarations + // + + static const BSONField<BSONObj> query; + static const BSONField<BSONObj> updateExpr; + static const BSONField<bool> multi; + static const BSONField<bool> upsert; + + // + // construction / destruction + // + + BatchedUpdateDocument(); + virtual ~BatchedUpdateDocument(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(BatchedUpdateDocument* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setQuery(const BSONObj& query); + void unsetQuery(); + bool isQuerySet() const; + const BSONObj& getQuery() const; + + void setUpdateExpr(const BSONObj& updateExpr); + void unsetUpdateExpr(); + bool isUpdateExprSet() const; + const BSONObj& getUpdateExpr() const; + + void setMulti(bool multi); + void unsetMulti(); + bool isMultiSet() const; + bool getMulti() const; + + void setUpsert(bool upsert); + void unsetUpsert(); + bool isUpsertSet() const; + bool getUpsert() const; + +private: + // Convention: (M)andatory, (O)ptional + + // (M) query whose result the update will manipulate + BSONObj _query; + bool _isQuerySet; + + // (M) the update expression itself + BSONObj _updateExpr; + bool _isUpdateExprSet; + + // (O) whether multiple documents are to be updated + bool _multi; + bool _isMultiSet; + + // (O) whether upserts are allowed + bool _upsert; + bool _isUpsertSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_update_request.cpp b/src/mongo/s/write_ops/batched_update_request.cpp index f2dda0be1de..0994cb04c6f 100644 --- a/src/mongo/s/write_ops/batched_update_request.cpp +++ b/src/mongo/s/write_ops/batched_update_request.cpp @@ -34,299 +34,303 @@ namespace mongo { - using std::unique_ptr; - using std::string; - - using mongoutils::str::stream; - - const std::string BatchedUpdateRequest::BATCHED_UPDATE_REQUEST = "update"; - const BSONField<std::string> BatchedUpdateRequest::collName("update"); - const BSONField<std::vector<BatchedUpdateDocument*> > BatchedUpdateRequest::updates("updates"); - const BSONField<BSONObj> BatchedUpdateRequest::writeConcern("writeConcern"); - const BSONField<bool> BatchedUpdateRequest::ordered("ordered", true); - const BSONField<BSONObj> BatchedUpdateRequest::metadata("metadata"); - - BatchedUpdateRequest::BatchedUpdateRequest() { - clear(); +using std::unique_ptr; +using std::string; + +using mongoutils::str::stream; + +const std::string BatchedUpdateRequest::BATCHED_UPDATE_REQUEST = "update"; +const BSONField<std::string> BatchedUpdateRequest::collName("update"); +const BSONField<std::vector<BatchedUpdateDocument*>> BatchedUpdateRequest::updates("updates"); +const BSONField<BSONObj> BatchedUpdateRequest::writeConcern("writeConcern"); +const BSONField<bool> BatchedUpdateRequest::ordered("ordered", true); +const BSONField<BSONObj> BatchedUpdateRequest::metadata("metadata"); + +BatchedUpdateRequest::BatchedUpdateRequest() { + clear(); +} + +BatchedUpdateRequest::~BatchedUpdateRequest() { + unsetUpdates(); +} + +bool BatchedUpdateRequest::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BatchedUpdateRequest::~BatchedUpdateRequest() { - unsetUpdates(); + // All the mandatory fields must be present. + if (!_isCollNameSet) { + *errMsg = stream() << "missing " << collName.name() << " field"; + return false; } - bool BatchedUpdateRequest::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } - - // All the mandatory fields must be present. - if (!_isCollNameSet) { - *errMsg = stream() << "missing " << collName.name() << " field"; - return false; - } - - if (!_isUpdatesSet) { - *errMsg = stream() << "missing " << updates.name() << " field"; - return false; - } - - return true; + if (!_isUpdatesSet) { + *errMsg = stream() << "missing " << updates.name() << " field"; + return false; } - BSONObj BatchedUpdateRequest::toBSON() const { - BSONObjBuilder builder; + return true; +} - if (_isCollNameSet) builder.append(collName(), _collName); +BSONObj BatchedUpdateRequest::toBSON() const { + BSONObjBuilder builder; - if (_isUpdatesSet) { - BSONArrayBuilder updatesBuilder(builder.subarrayStart(updates())); - for (std::vector<BatchedUpdateDocument*>::const_iterator it = _updates.begin(); - it != _updates.end(); - ++it) { - BSONObj updateDocument = (*it)->toBSON(); - updatesBuilder.append(updateDocument); - } - updatesBuilder.done(); - } - - if (_isWriteConcernSet) builder.append(writeConcern(), _writeConcern); - - if (_isOrderedSet) builder.append(ordered(), _ordered); - - if (_metadata) builder.append(metadata(), _metadata->toBSON()); - - if (_shouldBypassValidation) builder.append(bypassDocumentValidationCommandOption(), true); + if (_isCollNameSet) + builder.append(collName(), _collName); - return builder.obj(); + if (_isUpdatesSet) { + BSONArrayBuilder updatesBuilder(builder.subarrayStart(updates())); + for (std::vector<BatchedUpdateDocument*>::const_iterator it = _updates.begin(); + it != _updates.end(); + ++it) { + BSONObj updateDocument = (*it)->toBSON(); + updatesBuilder.append(updateDocument); + } + updatesBuilder.done(); } - bool BatchedUpdateRequest::parseBSON(const BSONObj& source, string* errMsg) { - clear(); - - std::string dummy; - if (!errMsg) errMsg = &dummy; - - FieldParser::FieldState fieldState; - - BSONObjIterator it( source ); - while ( it.more() ) { - const BSONElement& elem = it.next(); - StringData fieldName = elem.fieldNameStringData(); - - if ( fieldName == collName.name() ) { - std::string collNameTemp; - fieldState = FieldParser::extract(elem, collName, &collNameTemp, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _collName = NamespaceString(collNameTemp); - _isCollNameSet = fieldState == FieldParser::FIELD_SET; - } - else if ( fieldName == updates.name() ) { - fieldState = FieldParser::extract(elem, updates, &_updates, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isUpdatesSet = fieldState == FieldParser::FIELD_SET; - } - else if ( fieldName == writeConcern.name() ) { - fieldState = FieldParser::extract(elem, writeConcern, &_writeConcern, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isWriteConcernSet = fieldState == FieldParser::FIELD_SET; - } - else if ( fieldName == ordered.name() ) { - fieldState = FieldParser::extract(elem, ordered, &_ordered, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isOrderedSet = fieldState == FieldParser::FIELD_SET; - } - else if ( fieldName == metadata.name() ) { - BSONObj metadataObj; - fieldState = FieldParser::extract(elem, metadata, &metadataObj, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - - if (!metadataObj.isEmpty()) { - _metadata.reset(new BatchedRequestMetadata()); - if (!_metadata->parseBSON(metadataObj, errMsg)) { - return false; - } + if (_isWriteConcernSet) + builder.append(writeConcern(), _writeConcern); + + if (_isOrderedSet) + builder.append(ordered(), _ordered); + + if (_metadata) + builder.append(metadata(), _metadata->toBSON()); + + if (_shouldBypassValidation) + builder.append(bypassDocumentValidationCommandOption(), true); + + return builder.obj(); +} + +bool BatchedUpdateRequest::parseBSON(const BSONObj& source, string* errMsg) { + clear(); + + std::string dummy; + if (!errMsg) + errMsg = &dummy; + + FieldParser::FieldState fieldState; + + BSONObjIterator it(source); + while (it.more()) { + const BSONElement& elem = it.next(); + StringData fieldName = elem.fieldNameStringData(); + + if (fieldName == collName.name()) { + std::string collNameTemp; + fieldState = FieldParser::extract(elem, collName, &collNameTemp, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _collName = NamespaceString(collNameTemp); + _isCollNameSet = fieldState == FieldParser::FIELD_SET; + } else if (fieldName == updates.name()) { + fieldState = FieldParser::extract(elem, updates, &_updates, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isUpdatesSet = fieldState == FieldParser::FIELD_SET; + } else if (fieldName == writeConcern.name()) { + fieldState = FieldParser::extract(elem, writeConcern, &_writeConcern, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isWriteConcernSet = fieldState == FieldParser::FIELD_SET; + } else if (fieldName == ordered.name()) { + fieldState = FieldParser::extract(elem, ordered, &_ordered, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isOrderedSet = fieldState == FieldParser::FIELD_SET; + } else if (fieldName == metadata.name()) { + BSONObj metadataObj; + fieldState = FieldParser::extract(elem, metadata, &metadataObj, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + + if (!metadataObj.isEmpty()) { + _metadata.reset(new BatchedRequestMetadata()); + if (!_metadata->parseBSON(metadataObj, errMsg)) { + return false; } } - else if (fieldName == bypassDocumentValidationCommandOption()) { - _shouldBypassValidation = elem.trueValue(); - } - } - return true; - } - - void BatchedUpdateRequest::clear() { - _collName = NamespaceString(); - _isCollNameSet = false; - - unsetUpdates(); - - _writeConcern = BSONObj(); - _isWriteConcernSet = false; - - _ordered = false; - _isOrderedSet = false; - - _shouldBypassValidation = false; - - _metadata.reset(); - } - - void BatchedUpdateRequest::cloneTo(BatchedUpdateRequest* other) const { - other->clear(); - - other->_collName = _collName; - other->_isCollNameSet = _isCollNameSet; - - for(std::vector<BatchedUpdateDocument*>::const_iterator it = _updates.begin(); - it != _updates.end(); - ++it) { - unique_ptr<BatchedUpdateDocument> tempBatchUpdateDocument(new BatchedUpdateDocument); - (*it)->cloneTo(tempBatchUpdateDocument.get()); - other->addToUpdates(tempBatchUpdateDocument.release()); - } - other->_isUpdatesSet = _isUpdatesSet; - - other->_writeConcern = _writeConcern; - other->_isWriteConcernSet = _isWriteConcernSet; - - other->_ordered = _ordered; - other->_isOrderedSet = _isOrderedSet; - - if (_metadata) { - other->_metadata.reset(new BatchedRequestMetadata()); - _metadata->cloneTo(other->_metadata.get()); + } else if (fieldName == bypassDocumentValidationCommandOption()) { + _shouldBypassValidation = elem.trueValue(); } } + return true; +} - std::string BatchedUpdateRequest::toString() const { - return toBSON().toString(); - } +void BatchedUpdateRequest::clear() { + _collName = NamespaceString(); + _isCollNameSet = false; - void BatchedUpdateRequest::setCollName(StringData collName) { - _collName = NamespaceString(collName); - _isCollNameSet = true; - } + unsetUpdates(); - const std::string& BatchedUpdateRequest::getCollName() const { - dassert(_isCollNameSet); - return _collName.ns(); - } + _writeConcern = BSONObj(); + _isWriteConcernSet = false; - void BatchedUpdateRequest::setCollNameNS(const NamespaceString& collName) { - _collName = collName; - _isCollNameSet = true; - } + _ordered = false; + _isOrderedSet = false; - const NamespaceString& BatchedUpdateRequest::getCollNameNS() const { - dassert(_isCollNameSet); - return _collName; - } + _shouldBypassValidation = false; - const NamespaceString& BatchedUpdateRequest::getTargetingNSS() const { - return getCollNameNS(); - } + _metadata.reset(); +} - void BatchedUpdateRequest::setUpdates(const std::vector<BatchedUpdateDocument*>& updates) { - unsetUpdates(); - for (std::vector<BatchedUpdateDocument*>::const_iterator it = updates.begin(); - it != updates.end(); - ++it) { - unique_ptr<BatchedUpdateDocument> tempBatchUpdateDocument(new BatchedUpdateDocument); - (*it)->cloneTo(tempBatchUpdateDocument.get()); - addToUpdates(tempBatchUpdateDocument.release()); - } - _isUpdatesSet = updates.size() > 0; - } +void BatchedUpdateRequest::cloneTo(BatchedUpdateRequest* other) const { + other->clear(); - void BatchedUpdateRequest::addToUpdates(BatchedUpdateDocument* updates) { - _updates.push_back(updates); - _isUpdatesSet = true; - } + other->_collName = _collName; + other->_isCollNameSet = _isCollNameSet; - void BatchedUpdateRequest::unsetUpdates() { - for(std::vector<BatchedUpdateDocument*>::iterator it = _updates.begin(); - it != _updates.end(); - ++it) { - delete *it; - } - _updates.clear(); - _isUpdatesSet = false; + for (std::vector<BatchedUpdateDocument*>::const_iterator it = _updates.begin(); + it != _updates.end(); + ++it) { + unique_ptr<BatchedUpdateDocument> tempBatchUpdateDocument(new BatchedUpdateDocument); + (*it)->cloneTo(tempBatchUpdateDocument.get()); + other->addToUpdates(tempBatchUpdateDocument.release()); } + other->_isUpdatesSet = _isUpdatesSet; - bool BatchedUpdateRequest::isUpdatesSet() const { - return _isUpdatesSet; - } + other->_writeConcern = _writeConcern; + other->_isWriteConcernSet = _isWriteConcernSet; - size_t BatchedUpdateRequest::sizeUpdates() const { - return _updates.size(); - } + other->_ordered = _ordered; + other->_isOrderedSet = _isOrderedSet; - const std::vector<BatchedUpdateDocument*>& BatchedUpdateRequest::getUpdates() const { - dassert(_isUpdatesSet); - return _updates; + if (_metadata) { + other->_metadata.reset(new BatchedRequestMetadata()); + _metadata->cloneTo(other->_metadata.get()); } - - const BatchedUpdateDocument* BatchedUpdateRequest::getUpdatesAt(size_t pos) const { - dassert(_isUpdatesSet); - dassert(_updates.size() > pos); - return _updates.at(pos); +} + +std::string BatchedUpdateRequest::toString() const { + return toBSON().toString(); +} + +void BatchedUpdateRequest::setCollName(StringData collName) { + _collName = NamespaceString(collName); + _isCollNameSet = true; +} + +const std::string& BatchedUpdateRequest::getCollName() const { + dassert(_isCollNameSet); + return _collName.ns(); +} + +void BatchedUpdateRequest::setCollNameNS(const NamespaceString& collName) { + _collName = collName; + _isCollNameSet = true; +} + +const NamespaceString& BatchedUpdateRequest::getCollNameNS() const { + dassert(_isCollNameSet); + return _collName; +} + +const NamespaceString& BatchedUpdateRequest::getTargetingNSS() const { + return getCollNameNS(); +} + +void BatchedUpdateRequest::setUpdates(const std::vector<BatchedUpdateDocument*>& updates) { + unsetUpdates(); + for (std::vector<BatchedUpdateDocument*>::const_iterator it = updates.begin(); + it != updates.end(); + ++it) { + unique_ptr<BatchedUpdateDocument> tempBatchUpdateDocument(new BatchedUpdateDocument); + (*it)->cloneTo(tempBatchUpdateDocument.get()); + addToUpdates(tempBatchUpdateDocument.release()); } - - void BatchedUpdateRequest::setWriteConcern(const BSONObj& writeConcern) { - _writeConcern = writeConcern.getOwned(); - _isWriteConcernSet = true; - } - - void BatchedUpdateRequest::unsetWriteConcern() { - _isWriteConcernSet = false; - } - - bool BatchedUpdateRequest::isWriteConcernSet() const { - return _isWriteConcernSet; - } - - const BSONObj& BatchedUpdateRequest::getWriteConcern() const { - dassert(_isWriteConcernSet); - return _writeConcern; - } - - void BatchedUpdateRequest::setOrdered(bool ordered) { - _ordered = ordered; - _isOrderedSet = true; - } - - void BatchedUpdateRequest::unsetOrdered() { - _isOrderedSet = false; - } - - bool BatchedUpdateRequest::isOrderedSet() const { - return _isOrderedSet; + _isUpdatesSet = updates.size() > 0; +} + +void BatchedUpdateRequest::addToUpdates(BatchedUpdateDocument* updates) { + _updates.push_back(updates); + _isUpdatesSet = true; +} + +void BatchedUpdateRequest::unsetUpdates() { + for (std::vector<BatchedUpdateDocument*>::iterator it = _updates.begin(); it != _updates.end(); + ++it) { + delete *it; } - - bool BatchedUpdateRequest::getOrdered() const { - if (_isOrderedSet) { - return _ordered; - } - else { - return ordered.getDefault(); - } + _updates.clear(); + _isUpdatesSet = false; +} + +bool BatchedUpdateRequest::isUpdatesSet() const { + return _isUpdatesSet; +} + +size_t BatchedUpdateRequest::sizeUpdates() const { + return _updates.size(); +} + +const std::vector<BatchedUpdateDocument*>& BatchedUpdateRequest::getUpdates() const { + dassert(_isUpdatesSet); + return _updates; +} + +const BatchedUpdateDocument* BatchedUpdateRequest::getUpdatesAt(size_t pos) const { + dassert(_isUpdatesSet); + dassert(_updates.size() > pos); + return _updates.at(pos); +} + +void BatchedUpdateRequest::setWriteConcern(const BSONObj& writeConcern) { + _writeConcern = writeConcern.getOwned(); + _isWriteConcernSet = true; +} + +void BatchedUpdateRequest::unsetWriteConcern() { + _isWriteConcernSet = false; +} + +bool BatchedUpdateRequest::isWriteConcernSet() const { + return _isWriteConcernSet; +} + +const BSONObj& BatchedUpdateRequest::getWriteConcern() const { + dassert(_isWriteConcernSet); + return _writeConcern; +} + +void BatchedUpdateRequest::setOrdered(bool ordered) { + _ordered = ordered; + _isOrderedSet = true; +} + +void BatchedUpdateRequest::unsetOrdered() { + _isOrderedSet = false; +} + +bool BatchedUpdateRequest::isOrderedSet() const { + return _isOrderedSet; +} + +bool BatchedUpdateRequest::getOrdered() const { + if (_isOrderedSet) { + return _ordered; + } else { + return ordered.getDefault(); } +} - void BatchedUpdateRequest::setMetadata(BatchedRequestMetadata* metadata) { - _metadata.reset(metadata); - } +void BatchedUpdateRequest::setMetadata(BatchedRequestMetadata* metadata) { + _metadata.reset(metadata); +} - void BatchedUpdateRequest::unsetMetadata() { - _metadata.reset(); - } +void BatchedUpdateRequest::unsetMetadata() { + _metadata.reset(); +} - bool BatchedUpdateRequest::isMetadataSet() const { - return _metadata.get(); - } +bool BatchedUpdateRequest::isMetadataSet() const { + return _metadata.get(); +} - BatchedRequestMetadata* BatchedUpdateRequest::getMetadata() const { - return _metadata.get(); - } +BatchedRequestMetadata* BatchedUpdateRequest::getMetadata() const { + return _metadata.get(); +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_update_request.h b/src/mongo/s/write_ops/batched_update_request.h index db99188b7b3..6042f4c6cf2 100644 --- a/src/mongo/s/write_ops/batched_update_request.h +++ b/src/mongo/s/write_ops/batched_update_request.h @@ -40,116 +40,120 @@ namespace mongo { +/** + * This class represents the layout and content of a batched update runCommand, + * the request side. + */ +class BatchedUpdateRequest : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedUpdateRequest); + +public: + // + // schema declarations + // + + // Name used for the batched update invocation. + static const std::string BATCHED_UPDATE_REQUEST; + + // Field names and types in the batched update command type. + static const BSONField<std::string> collName; + static const BSONField<std::vector<BatchedUpdateDocument*>> updates; + static const BSONField<BSONObj> writeConcern; + static const BSONField<bool> ordered; + static const BSONField<BSONObj> metadata; + + // + // construction / destruction + // + + BatchedUpdateRequest(); + virtual ~BatchedUpdateRequest(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(BatchedUpdateRequest* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setCollName(StringData collName); + void setCollNameNS(const NamespaceString& collName); + const std::string& getCollName() const; + const NamespaceString& getCollNameNS() const; + + const NamespaceString& getTargetingNSS() const; + + void setUpdates(const std::vector<BatchedUpdateDocument*>& updates); + /** - * This class represents the layout and content of a batched update runCommand, - * the request side. + * updates ownership is transferred to here. */ - class BatchedUpdateRequest : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedUpdateRequest); - public: - - // - // schema declarations - // - - // Name used for the batched update invocation. - static const std::string BATCHED_UPDATE_REQUEST; - - // Field names and types in the batched update command type. - static const BSONField<std::string> collName; - static const BSONField<std::vector<BatchedUpdateDocument*> > updates; - static const BSONField<BSONObj> writeConcern; - static const BSONField<bool> ordered; - static const BSONField<BSONObj> metadata; - - // - // construction / destruction - // - - BatchedUpdateRequest(); - virtual ~BatchedUpdateRequest(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(BatchedUpdateRequest* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setCollName(StringData collName); - void setCollNameNS(const NamespaceString& collName); - const std::string& getCollName() const; - const NamespaceString& getCollNameNS() const; - - const NamespaceString& getTargetingNSS() const; - - void setUpdates(const std::vector<BatchedUpdateDocument*>& updates); - - /** - * updates ownership is transferred to here. - */ - void addToUpdates(BatchedUpdateDocument* updates); - void unsetUpdates(); - bool isUpdatesSet() const; - std::size_t sizeUpdates() const; - const std::vector<BatchedUpdateDocument*>& getUpdates() const; - const BatchedUpdateDocument* getUpdatesAt(std::size_t pos) const; - - void setWriteConcern(const BSONObj& writeConcern); - void unsetWriteConcern(); - bool isWriteConcernSet() const; - const BSONObj& getWriteConcern() const; - - void setOrdered(bool ordered); - void unsetOrdered(); - bool isOrderedSet() const; - bool getOrdered() const; - - void setShouldBypassValidation(bool newVal) { _shouldBypassValidation = newVal; } - bool shouldBypassValidation() const { return _shouldBypassValidation; } - - /* - * metadata ownership will be transferred to this. - */ - void setMetadata(BatchedRequestMetadata* metadata); - void unsetMetadata(); - bool isMetadataSet() const; - BatchedRequestMetadata* getMetadata() const; - - private: - // Convention: (M)andatory, (O)ptional - - // (M) collection we're updating from - NamespaceString _collName; - bool _isCollNameSet; - - // (M) array of individual updates - std::vector<BatchedUpdateDocument*> _updates; - bool _isUpdatesSet; - - // (O) to be issued after the batch applied - BSONObj _writeConcern; - bool _isWriteConcernSet; - - // (O) whether batch is issued in parallel or not - bool _ordered; - bool _isOrderedSet; - - // (O) should document validation be bypassed (default false) - bool _shouldBypassValidation; - - // (O) metadata associated with this request for internal use. - std::unique_ptr<BatchedRequestMetadata> _metadata; - }; - -} // namespace mongo + void addToUpdates(BatchedUpdateDocument* updates); + void unsetUpdates(); + bool isUpdatesSet() const; + std::size_t sizeUpdates() const; + const std::vector<BatchedUpdateDocument*>& getUpdates() const; + const BatchedUpdateDocument* getUpdatesAt(std::size_t pos) const; + + void setWriteConcern(const BSONObj& writeConcern); + void unsetWriteConcern(); + bool isWriteConcernSet() const; + const BSONObj& getWriteConcern() const; + + void setOrdered(bool ordered); + void unsetOrdered(); + bool isOrderedSet() const; + bool getOrdered() const; + + void setShouldBypassValidation(bool newVal) { + _shouldBypassValidation = newVal; + } + bool shouldBypassValidation() const { + return _shouldBypassValidation; + } + + /* + * metadata ownership will be transferred to this. + */ + void setMetadata(BatchedRequestMetadata* metadata); + void unsetMetadata(); + bool isMetadataSet() const; + BatchedRequestMetadata* getMetadata() const; + +private: + // Convention: (M)andatory, (O)ptional + + // (M) collection we're updating from + NamespaceString _collName; + bool _isCollNameSet; + + // (M) array of individual updates + std::vector<BatchedUpdateDocument*> _updates; + bool _isUpdatesSet; + + // (O) to be issued after the batch applied + BSONObj _writeConcern; + bool _isWriteConcernSet; + + // (O) whether batch is issued in parallel or not + bool _ordered; + bool _isOrderedSet; + + // (O) should document validation be bypassed (default false) + bool _shouldBypassValidation; + + // (O) metadata associated with this request for internal use. + std::unique_ptr<BatchedRequestMetadata> _metadata; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_update_request_test.cpp b/src/mongo/s/write_ops/batched_update_request_test.cpp index c8e69b92b06..662bf079b9f 100644 --- a/src/mongo/s/write_ops/batched_update_request_test.cpp +++ b/src/mongo/s/write_ops/batched_update_request_test.cpp @@ -36,56 +36,49 @@ namespace { - using std::string; - using mongo::BatchedUpdateDocument; - using mongo::BatchedUpdateRequest; - using mongo::BatchedRequestMetadata; - using mongo::BSONArray; - using mongo::BSONArrayBuilder; - using mongo::BSONObj; - using mongo::OID; - using mongo::Timestamp; +using std::string; +using mongo::BatchedUpdateDocument; +using mongo::BatchedUpdateRequest; +using mongo::BatchedRequestMetadata; +using mongo::BSONArray; +using mongo::BSONArrayBuilder; +using mongo::BSONObj; +using mongo::OID; +using mongo::Timestamp; - TEST(RoundTrip, Normal) { - BSONArray updateArray = - BSON_ARRAY( - BSON(BatchedUpdateDocument::query(BSON("a" << 1)) << - BatchedUpdateDocument::updateExpr(BSON("$set" << BSON("a" << 1))) << - BatchedUpdateDocument::multi(false) << - BatchedUpdateDocument::upsert(false) - ) << - BSON(BatchedUpdateDocument::query(BSON("b" << 1)) << - BatchedUpdateDocument::updateExpr(BSON("$set" << BSON("b" << 2))) << - BatchedUpdateDocument::multi(false) << - BatchedUpdateDocument::upsert(false) - ) - ); +TEST(RoundTrip, Normal) { + BSONArray updateArray = BSON_ARRAY( + BSON(BatchedUpdateDocument::query(BSON("a" << 1)) + << BatchedUpdateDocument::updateExpr(BSON("$set" << BSON("a" << 1))) + << BatchedUpdateDocument::multi(false) << BatchedUpdateDocument::upsert(false)) + << BSON(BatchedUpdateDocument::query(BSON("b" << 1)) + << BatchedUpdateDocument::updateExpr(BSON("$set" << BSON("b" << 2))) + << BatchedUpdateDocument::multi(false) << BatchedUpdateDocument::upsert(false))); - BSONObj writeConcernObj = BSON("w" << 1); + BSONObj writeConcernObj = BSON("w" << 1); - // The BSON_ARRAY macro doesn't support Timestamps. - BSONArrayBuilder arrBuilder; - arrBuilder.append(Timestamp(1,1)); - arrBuilder.append(OID::gen()); - BSONArray shardVersionArray = arrBuilder.arr(); + // The BSON_ARRAY macro doesn't support Timestamps. + BSONArrayBuilder arrBuilder; + arrBuilder.append(Timestamp(1, 1)); + arrBuilder.append(OID::gen()); + BSONArray shardVersionArray = arrBuilder.arr(); - BSONObj origUpdateRequestObj = - BSON(BatchedUpdateRequest::collName("test") << - BatchedUpdateRequest::updates() << updateArray << - BatchedUpdateRequest::writeConcern(writeConcernObj) << - BatchedUpdateRequest::ordered(true) << - BatchedUpdateRequest::metadata() << BSON( - BatchedRequestMetadata::shardName("shard0000") << - BatchedRequestMetadata::shardVersion() << shardVersionArray << - BatchedRequestMetadata::session(0))); + BSONObj origUpdateRequestObj = + BSON(BatchedUpdateRequest::collName("test") + << BatchedUpdateRequest::updates() << updateArray + << BatchedUpdateRequest::writeConcern(writeConcernObj) + << BatchedUpdateRequest::ordered(true) << BatchedUpdateRequest::metadata() + << BSON(BatchedRequestMetadata::shardName("shard0000") + << BatchedRequestMetadata::shardVersion() << shardVersionArray + << BatchedRequestMetadata::session(0))); - string errMsg; - BatchedUpdateRequest request; - bool ok = request.parseBSON(origUpdateRequestObj, &errMsg); - ASSERT_TRUE(ok); + string errMsg; + BatchedUpdateRequest request; + bool ok = request.parseBSON(origUpdateRequestObj, &errMsg); + ASSERT_TRUE(ok); - BSONObj genUpdateRequestObj = request.toBSON(); - ASSERT_EQUALS(0, genUpdateRequestObj.woCompare(origUpdateRequestObj)); - } + BSONObj genUpdateRequestObj = request.toBSON(); + ASSERT_EQUALS(0, genUpdateRequestObj.woCompare(origUpdateRequestObj)); +} -} // unnamed namespace +} // unnamed namespace diff --git a/src/mongo/s/write_ops/batched_upsert_detail.cpp b/src/mongo/s/write_ops/batched_upsert_detail.cpp index d6764a22c39..cbc26eadcef 100644 --- a/src/mongo/s/write_ops/batched_upsert_detail.cpp +++ b/src/mongo/s/write_ops/batched_upsert_detail.cpp @@ -33,128 +33,130 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; +using mongoutils::str::stream; - const BSONField<int> BatchedUpsertDetail::index("index"); - const BSONField<BSONObj> BatchedUpsertDetail::upsertedID("_id"); +const BSONField<int> BatchedUpsertDetail::index("index"); +const BSONField<BSONObj> BatchedUpsertDetail::upsertedID("_id"); - BatchedUpsertDetail::BatchedUpsertDetail() { - clear(); - } - - BatchedUpsertDetail::~BatchedUpsertDetail() { - } +BatchedUpsertDetail::BatchedUpsertDetail() { + clear(); +} - bool BatchedUpsertDetail::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } +BatchedUpsertDetail::~BatchedUpsertDetail() {} - // All the mandatory fields must be present. - if (!_isIndexSet) { - *errMsg = stream() << "missing " << index.name() << " field"; - return false; - } +bool BatchedUpsertDetail::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; + } - if (!_isUpsertedIDSet) { - *errMsg = stream() << "missing " << upsertedID.name() << " field"; - return false; - } + // All the mandatory fields must be present. + if (!_isIndexSet) { + *errMsg = stream() << "missing " << index.name() << " field"; + return false; + } - return true; + if (!_isUpsertedIDSet) { + *errMsg = stream() << "missing " << upsertedID.name() << " field"; + return false; } - BSONObj BatchedUpsertDetail::toBSON() const { - BSONObjBuilder builder; + return true; +} - if (_isIndexSet) builder.append(index(), _index); +BSONObj BatchedUpsertDetail::toBSON() const { + BSONObjBuilder builder; - // We're using the BSONObj to store the _id value. - if (_isUpsertedIDSet) { - builder.appendAs(_upsertedID.firstElement(), upsertedID()); - } + if (_isIndexSet) + builder.append(index(), _index); - return builder.obj(); + // We're using the BSONObj to store the _id value. + if (_isUpsertedIDSet) { + builder.appendAs(_upsertedID.firstElement(), upsertedID()); } - bool BatchedUpsertDetail::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + return builder.obj(); +} - std::string dummy; - if (!errMsg) errMsg = &dummy; +bool BatchedUpsertDetail::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, index, &_index, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isIndexSet = fieldState == FieldParser::FIELD_SET; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - fieldState = FieldParser::extractID(source, upsertedID, &_upsertedID, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isUpsertedIDSet = fieldState == FieldParser::FIELD_SET; + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, index, &_index, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isIndexSet = fieldState == FieldParser::FIELD_SET; - return true; - } + fieldState = FieldParser::extractID(source, upsertedID, &_upsertedID, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isUpsertedIDSet = fieldState == FieldParser::FIELD_SET; - void BatchedUpsertDetail::clear() { - _index = 0; - _isIndexSet = false; + return true; +} - _upsertedID = BSONObj(); - _isUpsertedIDSet = false; +void BatchedUpsertDetail::clear() { + _index = 0; + _isIndexSet = false; - } + _upsertedID = BSONObj(); + _isUpsertedIDSet = false; +} - void BatchedUpsertDetail::cloneTo(BatchedUpsertDetail* other) const { - other->clear(); +void BatchedUpsertDetail::cloneTo(BatchedUpsertDetail* other) const { + other->clear(); - other->_index = _index; - other->_isIndexSet = _isIndexSet; + other->_index = _index; + other->_isIndexSet = _isIndexSet; - other->_upsertedID = _upsertedID; - other->_isUpsertedIDSet = _isUpsertedIDSet; - } + other->_upsertedID = _upsertedID; + other->_isUpsertedIDSet = _isUpsertedIDSet; +} - std::string BatchedUpsertDetail::toString() const { - return "implement me"; - } +std::string BatchedUpsertDetail::toString() const { + return "implement me"; +} - void BatchedUpsertDetail::setIndex(int index) { - _index = index; - _isIndexSet = true; - } +void BatchedUpsertDetail::setIndex(int index) { + _index = index; + _isIndexSet = true; +} - void BatchedUpsertDetail::unsetIndex() { - _isIndexSet = false; - } +void BatchedUpsertDetail::unsetIndex() { + _isIndexSet = false; +} - bool BatchedUpsertDetail::isIndexSet() const { - return _isIndexSet; - } +bool BatchedUpsertDetail::isIndexSet() const { + return _isIndexSet; +} - int BatchedUpsertDetail::getIndex() const { - dassert(_isIndexSet); - return _index; - } +int BatchedUpsertDetail::getIndex() const { + dassert(_isIndexSet); + return _index; +} - void BatchedUpsertDetail::setUpsertedID(const BSONObj& upsertedID) { - _upsertedID = upsertedID.firstElement().wrap( "" ).getOwned(); - _isUpsertedIDSet = true; - } +void BatchedUpsertDetail::setUpsertedID(const BSONObj& upsertedID) { + _upsertedID = upsertedID.firstElement().wrap("").getOwned(); + _isUpsertedIDSet = true; +} - void BatchedUpsertDetail::unsetUpsertedID() { - _isUpsertedIDSet = false; - } +void BatchedUpsertDetail::unsetUpsertedID() { + _isUpsertedIDSet = false; +} - bool BatchedUpsertDetail::isUpsertedIDSet() const { - return _isUpsertedIDSet; - } +bool BatchedUpsertDetail::isUpsertedIDSet() const { + return _isUpsertedIDSet; +} - const BSONObj& BatchedUpsertDetail::getUpsertedID() const { - dassert(_isUpsertedIDSet); - return _upsertedID; - } +const BSONObj& BatchedUpsertDetail::getUpsertedID() const { + dassert(_isUpsertedIDSet); + return _upsertedID; +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/batched_upsert_detail.h b/src/mongo/s/write_ops/batched_upsert_detail.h index cff8122f7ee..0f065ce6984 100644 --- a/src/mongo/s/write_ops/batched_upsert_detail.h +++ b/src/mongo/s/write_ops/batched_upsert_detail.h @@ -37,65 +37,65 @@ namespace mongo { - /** - * This class represents the layout and content of an idem inside the 'upserted' array - * of a write command's response (see batched_command_response.h) - */ - class BatchedUpsertDetail : public BSONSerializable { - MONGO_DISALLOW_COPYING(BatchedUpsertDetail); - public: - - // - // schema declarations - // - - static const BSONField<int> index; - static const BSONField<BSONObj> upsertedID; // ID type - - // - // construction / destruction - // - - BatchedUpsertDetail(); - virtual ~BatchedUpsertDetail(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(BatchedUpsertDetail* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setIndex(int index); - void unsetIndex(); - bool isIndexSet() const; - int getIndex() const; - - void setUpsertedID(const BSONObj& upsertedID); - void unsetUpsertedID(); - bool isUpsertedIDSet() const; - const BSONObj& getUpsertedID() const; - - private: - // Convention: (M)andatory, (O)ptional - - // (M) number of the batch item the upsert refers to - int _index; - bool _isIndexSet; - - // (M) _id for the upserted document - BSONObj _upsertedID; - bool _isUpsertedIDSet; - }; - -} // namespace mongo +/** + * This class represents the layout and content of an idem inside the 'upserted' array + * of a write command's response (see batched_command_response.h) + */ +class BatchedUpsertDetail : public BSONSerializable { + MONGO_DISALLOW_COPYING(BatchedUpsertDetail); + +public: + // + // schema declarations + // + + static const BSONField<int> index; + static const BSONField<BSONObj> upsertedID; // ID type + + // + // construction / destruction + // + + BatchedUpsertDetail(); + virtual ~BatchedUpsertDetail(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(BatchedUpsertDetail* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setIndex(int index); + void unsetIndex(); + bool isIndexSet() const; + int getIndex() const; + + void setUpsertedID(const BSONObj& upsertedID); + void unsetUpsertedID(); + bool isUpsertedIDSet() const; + const BSONObj& getUpsertedID() const; + +private: + // Convention: (M)andatory, (O)ptional + + // (M) number of the batch item the upsert refers to + int _index; + bool _isIndexSet; + + // (M) _id for the upserted document + BSONObj _upsertedID; + bool _isUpsertedIDSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/wc_error_detail.cpp b/src/mongo/s/write_ops/wc_error_detail.cpp index 0f7536639ba..40474c819c7 100644 --- a/src/mongo/s/write_ops/wc_error_detail.cpp +++ b/src/mongo/s/write_ops/wc_error_detail.cpp @@ -33,150 +33,155 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; - const BSONField<int> WCErrorDetail::errCode("code"); - const BSONField<BSONObj> WCErrorDetail::errInfo("errInfo"); - const BSONField<std::string> WCErrorDetail::errMessage("errmsg"); +using mongoutils::str::stream; +const BSONField<int> WCErrorDetail::errCode("code"); +const BSONField<BSONObj> WCErrorDetail::errInfo("errInfo"); +const BSONField<std::string> WCErrorDetail::errMessage("errmsg"); - WCErrorDetail::WCErrorDetail() { - clear(); - } - - WCErrorDetail::~WCErrorDetail() { - } +WCErrorDetail::WCErrorDetail() { + clear(); +} - bool WCErrorDetail::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } +WCErrorDetail::~WCErrorDetail() {} - // All the mandatory fields must be present. - if (!_isErrCodeSet) { - *errMsg = stream() << "missing " << errCode.name() << " field"; - return false; - } +bool WCErrorDetail::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; + } - return true; + // All the mandatory fields must be present. + if (!_isErrCodeSet) { + *errMsg = stream() << "missing " << errCode.name() << " field"; + return false; } - BSONObj WCErrorDetail::toBSON() const { - BSONObjBuilder builder; + return true; +} - if (_isErrCodeSet) builder.append(errCode(), _errCode); +BSONObj WCErrorDetail::toBSON() const { + BSONObjBuilder builder; - if (_isErrInfoSet) builder.append(errInfo(), _errInfo); + if (_isErrCodeSet) + builder.append(errCode(), _errCode); - if (_isErrMessageSet) builder.append(errMessage(), _errMessage); + if (_isErrInfoSet) + builder.append(errInfo(), _errInfo); - return builder.obj(); - } + if (_isErrMessageSet) + builder.append(errMessage(), _errMessage); - bool WCErrorDetail::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + return builder.obj(); +} - std::string dummy; - if (!errMsg) errMsg = &dummy; +bool WCErrorDetail::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, errCode, &_errCode, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isErrCodeSet = fieldState == FieldParser::FIELD_SET; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - fieldState = FieldParser::extract(source, errInfo, &_errInfo, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isErrInfoSet = fieldState == FieldParser::FIELD_SET; + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, errCode, &_errCode, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isErrCodeSet = fieldState == FieldParser::FIELD_SET; - fieldState = FieldParser::extract(source, errMessage, &_errMessage, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isErrMessageSet = fieldState == FieldParser::FIELD_SET; + fieldState = FieldParser::extract(source, errInfo, &_errInfo, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isErrInfoSet = fieldState == FieldParser::FIELD_SET; - return true; - } + fieldState = FieldParser::extract(source, errMessage, &_errMessage, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isErrMessageSet = fieldState == FieldParser::FIELD_SET; - void WCErrorDetail::clear() { - _errCode = 0; - _isErrCodeSet = false; + return true; +} - _errInfo = BSONObj(); - _isErrInfoSet = false; +void WCErrorDetail::clear() { + _errCode = 0; + _isErrCodeSet = false; - _errMessage.clear(); - _isErrMessageSet = false; + _errInfo = BSONObj(); + _isErrInfoSet = false; - } + _errMessage.clear(); + _isErrMessageSet = false; +} - void WCErrorDetail::cloneTo(WCErrorDetail* other) const { - other->clear(); +void WCErrorDetail::cloneTo(WCErrorDetail* other) const { + other->clear(); - other->_errCode = _errCode; - other->_isErrCodeSet = _isErrCodeSet; + other->_errCode = _errCode; + other->_isErrCodeSet = _isErrCodeSet; - other->_errInfo = _errInfo; - other->_isErrInfoSet = _isErrInfoSet; + other->_errInfo = _errInfo; + other->_isErrInfoSet = _isErrInfoSet; - other->_errMessage = _errMessage; - other->_isErrMessageSet = _isErrMessageSet; - } + other->_errMessage = _errMessage; + other->_isErrMessageSet = _isErrMessageSet; +} - std::string WCErrorDetail::toString() const { - return "implement me"; - } +std::string WCErrorDetail::toString() const { + return "implement me"; +} - void WCErrorDetail::setErrCode(int errCode) { - _errCode = errCode; - _isErrCodeSet = true; - } +void WCErrorDetail::setErrCode(int errCode) { + _errCode = errCode; + _isErrCodeSet = true; +} - void WCErrorDetail::unsetErrCode() { - _isErrCodeSet = false; - } +void WCErrorDetail::unsetErrCode() { + _isErrCodeSet = false; +} - bool WCErrorDetail::isErrCodeSet() const { - return _isErrCodeSet; - } +bool WCErrorDetail::isErrCodeSet() const { + return _isErrCodeSet; +} - int WCErrorDetail::getErrCode() const { - dassert(_isErrCodeSet); - return _errCode; - } +int WCErrorDetail::getErrCode() const { + dassert(_isErrCodeSet); + return _errCode; +} - void WCErrorDetail::setErrInfo(const BSONObj& errInfo) { - _errInfo = errInfo.getOwned(); - _isErrInfoSet = true; - } +void WCErrorDetail::setErrInfo(const BSONObj& errInfo) { + _errInfo = errInfo.getOwned(); + _isErrInfoSet = true; +} - void WCErrorDetail::unsetErrInfo() { - _isErrInfoSet = false; - } +void WCErrorDetail::unsetErrInfo() { + _isErrInfoSet = false; +} - bool WCErrorDetail::isErrInfoSet() const { - return _isErrInfoSet; - } +bool WCErrorDetail::isErrInfoSet() const { + return _isErrInfoSet; +} - const BSONObj& WCErrorDetail::getErrInfo() const { - dassert(_isErrInfoSet); - return _errInfo; - } +const BSONObj& WCErrorDetail::getErrInfo() const { + dassert(_isErrInfoSet); + return _errInfo; +} - void WCErrorDetail::setErrMessage(StringData errMessage) { - _errMessage = errMessage.toString(); - _isErrMessageSet = true; - } - - void WCErrorDetail::unsetErrMessage() { - _isErrMessageSet = false; - } +void WCErrorDetail::setErrMessage(StringData errMessage) { + _errMessage = errMessage.toString(); + _isErrMessageSet = true; +} - bool WCErrorDetail::isErrMessageSet() const { - return _isErrMessageSet; - } +void WCErrorDetail::unsetErrMessage() { + _isErrMessageSet = false; +} - const std::string& WCErrorDetail::getErrMessage() const { - dassert(_isErrMessageSet); - return _errMessage; - } +bool WCErrorDetail::isErrMessageSet() const { + return _isErrMessageSet; +} -} // namespace mongo +const std::string& WCErrorDetail::getErrMessage() const { + dassert(_isErrMessageSet); + return _errMessage; +} + +} // namespace mongo diff --git a/src/mongo/s/write_ops/wc_error_detail.h b/src/mongo/s/write_ops/wc_error_detail.h index 2f07008b02b..6bf1a757997 100644 --- a/src/mongo/s/write_ops/wc_error_detail.h +++ b/src/mongo/s/write_ops/wc_error_detail.h @@ -37,75 +37,75 @@ namespace mongo { - /** - * This class represents the layout and content of the error that occurs while trying - * to satisfy the write concern after executing the insert/update/delete runCommand. - */ - class WCErrorDetail : public BSONSerializable { - MONGO_DISALLOW_COPYING(WCErrorDetail); - public: - - // - // schema declarations - // - - static const BSONField<int> errCode; - static const BSONField<BSONObj> errInfo; - static const BSONField<std::string> errMessage; - - // - // construction / destruction - // - - WCErrorDetail(); - virtual ~WCErrorDetail(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(WCErrorDetail* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setErrCode(int errCode); - void unsetErrCode(); - bool isErrCodeSet() const; - int getErrCode() const; - - void setErrInfo(const BSONObj& errInfo); - void unsetErrInfo(); - bool isErrInfoSet() const; - const BSONObj& getErrInfo() const; - - void setErrMessage(StringData errMessage); - void unsetErrMessage(); - bool isErrMessageSet() const; - const std::string& getErrMessage() const; - - private: - // Convention: (M)andatory, (O)ptional - - // (M) error code for the write concern error. - int _errCode; - bool _isErrCodeSet; - - // (O) further details about the write concern error. - BSONObj _errInfo; - bool _isErrInfoSet; - - // (O) user readable explanation about the write concern error. - std::string _errMessage; - bool _isErrMessageSet; - }; - -} // namespace mongo +/** + * This class represents the layout and content of the error that occurs while trying + * to satisfy the write concern after executing the insert/update/delete runCommand. + */ +class WCErrorDetail : public BSONSerializable { + MONGO_DISALLOW_COPYING(WCErrorDetail); + +public: + // + // schema declarations + // + + static const BSONField<int> errCode; + static const BSONField<BSONObj> errInfo; + static const BSONField<std::string> errMessage; + + // + // construction / destruction + // + + WCErrorDetail(); + virtual ~WCErrorDetail(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(WCErrorDetail* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setErrCode(int errCode); + void unsetErrCode(); + bool isErrCodeSet() const; + int getErrCode() const; + + void setErrInfo(const BSONObj& errInfo); + void unsetErrInfo(); + bool isErrInfoSet() const; + const BSONObj& getErrInfo() const; + + void setErrMessage(StringData errMessage); + void unsetErrMessage(); + bool isErrMessageSet() const; + const std::string& getErrMessage() const; + +private: + // Convention: (M)andatory, (O)ptional + + // (M) error code for the write concern error. + int _errCode; + bool _isErrCodeSet; + + // (O) further details about the write concern error. + BSONObj _errInfo; + bool _isErrInfoSet; + + // (O) user readable explanation about the write concern error. + std::string _errMessage; + bool _isErrMessageSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/write_error_detail.cpp b/src/mongo/s/write_ops/write_error_detail.cpp index 3d5b9725fdf..b374ff47638 100644 --- a/src/mongo/s/write_ops/write_error_detail.cpp +++ b/src/mongo/s/write_ops/write_error_detail.cpp @@ -33,186 +33,193 @@ namespace mongo { - using std::string; +using std::string; - using mongoutils::str::stream; - const BSONField<int> WriteErrorDetail::index("index"); - const BSONField<int> WriteErrorDetail::errCode("code"); - const BSONField<BSONObj> WriteErrorDetail::errInfo("errInfo"); - const BSONField<std::string> WriteErrorDetail::errMessage("errmsg"); +using mongoutils::str::stream; +const BSONField<int> WriteErrorDetail::index("index"); +const BSONField<int> WriteErrorDetail::errCode("code"); +const BSONField<BSONObj> WriteErrorDetail::errInfo("errInfo"); +const BSONField<std::string> WriteErrorDetail::errMessage("errmsg"); - WriteErrorDetail::WriteErrorDetail() { - clear(); - } - - WriteErrorDetail::~WriteErrorDetail() { - } - - bool WriteErrorDetail::isValid(std::string* errMsg) const { - std::string dummy; - if (errMsg == NULL) { - errMsg = &dummy; - } +WriteErrorDetail::WriteErrorDetail() { + clear(); +} - // All the mandatory fields must be present. - if (!_isIndexSet) { - *errMsg = stream() << "missing " << index.name() << " field"; - return false; - } +WriteErrorDetail::~WriteErrorDetail() {} - if (!_isErrCodeSet) { - *errMsg = stream() << "missing " << errCode.name() << " field"; - return false; - } - - return true; +bool WriteErrorDetail::isValid(std::string* errMsg) const { + std::string dummy; + if (errMsg == NULL) { + errMsg = &dummy; } - BSONObj WriteErrorDetail::toBSON() const { - BSONObjBuilder builder; - - if (_isIndexSet) builder.append(index(), _index); - - if (_isErrCodeSet) builder.append(errCode(), _errCode); - - if (_isErrInfoSet) builder.append(errInfo(), _errInfo); - - if (_isErrMessageSet) builder.append(errMessage(), _errMessage); + // All the mandatory fields must be present. + if (!_isIndexSet) { + *errMsg = stream() << "missing " << index.name() << " field"; + return false; + } - return builder.obj(); + if (!_isErrCodeSet) { + *errMsg = stream() << "missing " << errCode.name() << " field"; + return false; } - bool WriteErrorDetail::parseBSON(const BSONObj& source, string* errMsg) { - clear(); + return true; +} - std::string dummy; - if (!errMsg) errMsg = &dummy; +BSONObj WriteErrorDetail::toBSON() const { + BSONObjBuilder builder; - FieldParser::FieldState fieldState; - fieldState = FieldParser::extract(source, index, &_index, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isIndexSet = fieldState == FieldParser::FIELD_SET; + if (_isIndexSet) + builder.append(index(), _index); - fieldState = FieldParser::extract(source, errCode, &_errCode, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isErrCodeSet = fieldState == FieldParser::FIELD_SET; + if (_isErrCodeSet) + builder.append(errCode(), _errCode); - fieldState = FieldParser::extract(source, errInfo, &_errInfo, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isErrInfoSet = fieldState == FieldParser::FIELD_SET; + if (_isErrInfoSet) + builder.append(errInfo(), _errInfo); - fieldState = FieldParser::extract(source, errMessage, &_errMessage, errMsg); - if (fieldState == FieldParser::FIELD_INVALID) return false; - _isErrMessageSet = fieldState == FieldParser::FIELD_SET; + if (_isErrMessageSet) + builder.append(errMessage(), _errMessage); - return true; - } + return builder.obj(); +} - void WriteErrorDetail::clear() { - _index = 0; - _isIndexSet = false; +bool WriteErrorDetail::parseBSON(const BSONObj& source, string* errMsg) { + clear(); - _errCode = 0; - _isErrCodeSet = false; + std::string dummy; + if (!errMsg) + errMsg = &dummy; - _errInfo = BSONObj(); - _isErrInfoSet = false; + FieldParser::FieldState fieldState; + fieldState = FieldParser::extract(source, index, &_index, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isIndexSet = fieldState == FieldParser::FIELD_SET; - _errMessage.clear(); - _isErrMessageSet = false; + fieldState = FieldParser::extract(source, errCode, &_errCode, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isErrCodeSet = fieldState == FieldParser::FIELD_SET; - } + fieldState = FieldParser::extract(source, errInfo, &_errInfo, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isErrInfoSet = fieldState == FieldParser::FIELD_SET; - void WriteErrorDetail::cloneTo(WriteErrorDetail* other) const { - other->clear(); + fieldState = FieldParser::extract(source, errMessage, &_errMessage, errMsg); + if (fieldState == FieldParser::FIELD_INVALID) + return false; + _isErrMessageSet = fieldState == FieldParser::FIELD_SET; - other->_index = _index; - other->_isIndexSet = _isIndexSet; + return true; +} - other->_errCode = _errCode; - other->_isErrCodeSet = _isErrCodeSet; +void WriteErrorDetail::clear() { + _index = 0; + _isIndexSet = false; - other->_errInfo = _errInfo; - other->_isErrInfoSet = _isErrInfoSet; + _errCode = 0; + _isErrCodeSet = false; - other->_errMessage = _errMessage; - other->_isErrMessageSet = _isErrMessageSet; - } + _errInfo = BSONObj(); + _isErrInfoSet = false; - std::string WriteErrorDetail::toString() const { - return "implement me"; - } + _errMessage.clear(); + _isErrMessageSet = false; +} - void WriteErrorDetail::setIndex(int index) { - _index = index; - _isIndexSet = true; - } +void WriteErrorDetail::cloneTo(WriteErrorDetail* other) const { + other->clear(); - void WriteErrorDetail::unsetIndex() { - _isIndexSet = false; - } + other->_index = _index; + other->_isIndexSet = _isIndexSet; - bool WriteErrorDetail::isIndexSet() const { - return _isIndexSet; - } + other->_errCode = _errCode; + other->_isErrCodeSet = _isErrCodeSet; - int WriteErrorDetail::getIndex() const { - dassert(_isIndexSet); - return _index; - } + other->_errInfo = _errInfo; + other->_isErrInfoSet = _isErrInfoSet; - void WriteErrorDetail::setErrCode(int errCode) { - _errCode = errCode; - _isErrCodeSet = true; - } + other->_errMessage = _errMessage; + other->_isErrMessageSet = _isErrMessageSet; +} - void WriteErrorDetail::unsetErrCode() { - _isErrCodeSet = false; - } +std::string WriteErrorDetail::toString() const { + return "implement me"; +} - bool WriteErrorDetail::isErrCodeSet() const { - return _isErrCodeSet; - } +void WriteErrorDetail::setIndex(int index) { + _index = index; + _isIndexSet = true; +} - int WriteErrorDetail::getErrCode() const { - dassert(_isErrCodeSet); - return _errCode; - } +void WriteErrorDetail::unsetIndex() { + _isIndexSet = false; +} - void WriteErrorDetail::setErrInfo(const BSONObj& errInfo) { - _errInfo = errInfo.getOwned(); - _isErrInfoSet = true; - } +bool WriteErrorDetail::isIndexSet() const { + return _isIndexSet; +} - void WriteErrorDetail::unsetErrInfo() { - _isErrInfoSet = false; - } +int WriteErrorDetail::getIndex() const { + dassert(_isIndexSet); + return _index; +} - bool WriteErrorDetail::isErrInfoSet() const { - return _isErrInfoSet; - } +void WriteErrorDetail::setErrCode(int errCode) { + _errCode = errCode; + _isErrCodeSet = true; +} - const BSONObj& WriteErrorDetail::getErrInfo() const { - dassert(_isErrInfoSet); - return _errInfo; - } +void WriteErrorDetail::unsetErrCode() { + _isErrCodeSet = false; +} - void WriteErrorDetail::setErrMessage(StringData errMessage) { - _errMessage = errMessage.toString(); - _isErrMessageSet = true; - } +bool WriteErrorDetail::isErrCodeSet() const { + return _isErrCodeSet; +} - void WriteErrorDetail::unsetErrMessage() { - _isErrMessageSet = false; - } +int WriteErrorDetail::getErrCode() const { + dassert(_isErrCodeSet); + return _errCode; +} + +void WriteErrorDetail::setErrInfo(const BSONObj& errInfo) { + _errInfo = errInfo.getOwned(); + _isErrInfoSet = true; +} + +void WriteErrorDetail::unsetErrInfo() { + _isErrInfoSet = false; +} + +bool WriteErrorDetail::isErrInfoSet() const { + return _isErrInfoSet; +} + +const BSONObj& WriteErrorDetail::getErrInfo() const { + dassert(_isErrInfoSet); + return _errInfo; +} + +void WriteErrorDetail::setErrMessage(StringData errMessage) { + _errMessage = errMessage.toString(); + _isErrMessageSet = true; +} - bool WriteErrorDetail::isErrMessageSet() const { - return _isErrMessageSet; - } +void WriteErrorDetail::unsetErrMessage() { + _isErrMessageSet = false; +} - const std::string& WriteErrorDetail::getErrMessage() const { - dassert(_isErrMessageSet); - return _errMessage; - } +bool WriteErrorDetail::isErrMessageSet() const { + return _isErrMessageSet; +} + +const std::string& WriteErrorDetail::getErrMessage() const { + dassert(_isErrMessageSet); + return _errMessage; +} -} // namespace mongo +} // namespace mongo diff --git a/src/mongo/s/write_ops/write_error_detail.h b/src/mongo/s/write_ops/write_error_detail.h index 49f9b2d8384..9ac64a85b5d 100644 --- a/src/mongo/s/write_ops/write_error_detail.h +++ b/src/mongo/s/write_ops/write_error_detail.h @@ -37,85 +37,85 @@ namespace mongo { - /** - * This class represents the layout and content of a insert/update/delete runCommand, - * the response side. - */ - class WriteErrorDetail : public BSONSerializable { - MONGO_DISALLOW_COPYING(WriteErrorDetail); - public: - - // - // schema declarations - // - - static const BSONField<int> index; - static const BSONField<int> errCode; - static const BSONField<BSONObj> errInfo; - static const BSONField<std::string> errMessage; - - // - // construction / destruction - // - - WriteErrorDetail(); - virtual ~WriteErrorDetail(); - - /** Copies all the fields present in 'this' to 'other'. */ - void cloneTo(WriteErrorDetail* other) const; - - // - // bson serializable interface implementation - // - - virtual bool isValid(std::string* errMsg) const; - virtual BSONObj toBSON() const; - virtual bool parseBSON(const BSONObj& source, std::string* errMsg); - virtual void clear(); - virtual std::string toString() const; - - // - // individual field accessors - // - - void setIndex(int index); - void unsetIndex(); - bool isIndexSet() const; - int getIndex() const; - - void setErrCode(int errCode); - void unsetErrCode(); - bool isErrCodeSet() const; - int getErrCode() const; - - void setErrInfo(const BSONObj& errInfo); - void unsetErrInfo(); - bool isErrInfoSet() const; - const BSONObj& getErrInfo() const; - - void setErrMessage(StringData errMessage); - void unsetErrMessage(); - bool isErrMessageSet() const; - const std::string& getErrMessage() const; - - private: - // Convention: (M)andatory, (O)ptional - - // (M) number of the batch item the error refers to - int _index; - bool _isIndexSet; - - // (M) whether all items in the batch applied correctly - int _errCode; - bool _isErrCodeSet; - - // (O) further details about the batch item error - BSONObj _errInfo; - bool _isErrInfoSet; - - // (O) user readable explanation about the batch item error - std::string _errMessage; - bool _isErrMessageSet; - }; - -} // namespace mongo +/** + * This class represents the layout and content of a insert/update/delete runCommand, + * the response side. + */ +class WriteErrorDetail : public BSONSerializable { + MONGO_DISALLOW_COPYING(WriteErrorDetail); + +public: + // + // schema declarations + // + + static const BSONField<int> index; + static const BSONField<int> errCode; + static const BSONField<BSONObj> errInfo; + static const BSONField<std::string> errMessage; + + // + // construction / destruction + // + + WriteErrorDetail(); + virtual ~WriteErrorDetail(); + + /** Copies all the fields present in 'this' to 'other'. */ + void cloneTo(WriteErrorDetail* other) const; + + // + // bson serializable interface implementation + // + + virtual bool isValid(std::string* errMsg) const; + virtual BSONObj toBSON() const; + virtual bool parseBSON(const BSONObj& source, std::string* errMsg); + virtual void clear(); + virtual std::string toString() const; + + // + // individual field accessors + // + + void setIndex(int index); + void unsetIndex(); + bool isIndexSet() const; + int getIndex() const; + + void setErrCode(int errCode); + void unsetErrCode(); + bool isErrCodeSet() const; + int getErrCode() const; + + void setErrInfo(const BSONObj& errInfo); + void unsetErrInfo(); + bool isErrInfoSet() const; + const BSONObj& getErrInfo() const; + + void setErrMessage(StringData errMessage); + void unsetErrMessage(); + bool isErrMessageSet() const; + const std::string& getErrMessage() const; + +private: + // Convention: (M)andatory, (O)ptional + + // (M) number of the batch item the error refers to + int _index; + bool _isIndexSet; + + // (M) whether all items in the batch applied correctly + int _errCode; + bool _isErrCodeSet; + + // (O) further details about the batch item error + BSONObj _errInfo; + bool _isErrInfoSet; + + // (O) user readable explanation about the batch item error + std::string _errMessage; + bool _isErrMessageSet; +}; + +} // namespace mongo diff --git a/src/mongo/s/write_ops/write_op.cpp b/src/mongo/s/write_ops/write_op.cpp index 9532922237f..f630802121d 100644 --- a/src/mongo/s/write_ops/write_op.cpp +++ b/src/mongo/s/write_ops/write_op.cpp @@ -34,258 +34,239 @@ namespace mongo { - using std::stringstream; - using std::vector; +using std::stringstream; +using std::vector; - static void clear( vector<ChildWriteOp*>* childOps ) { - for ( vector<ChildWriteOp*>::const_iterator it = childOps->begin(); it != childOps->end(); - ++it ) { - delete *it; - } - childOps->clear(); - } - - WriteOp::~WriteOp() { - clear( &_childOps ); - clear( &_history ); - } - - const BatchItemRef& WriteOp::getWriteItem() const { - return _itemRef; - } - - WriteOpState WriteOp::getWriteState() const { - return _state; - } - - const WriteErrorDetail& WriteOp::getOpError() const { - dassert( _state == WriteOpState_Error ); - return *_error; +static void clear(vector<ChildWriteOp*>* childOps) { + for (vector<ChildWriteOp*>::const_iterator it = childOps->begin(); it != childOps->end(); + ++it) { + delete *it; } + childOps->clear(); +} - Status WriteOp::targetWrites( const NSTargeter& targeter, - std::vector<TargetedWrite*>* targetedWrites ) { - - bool isUpdate = _itemRef.getOpType() == BatchedCommandRequest::BatchType_Update; - bool isDelete = _itemRef.getOpType() == BatchedCommandRequest::BatchType_Delete; - bool isIndexInsert = _itemRef.getRequest()->isInsertIndexRequest(); - - Status targetStatus = Status::OK(); - OwnedPointerVector<ShardEndpoint> endpointsOwned; - vector<ShardEndpoint*>& endpoints = endpointsOwned.mutableVector(); +WriteOp::~WriteOp() { + clear(&_childOps); + clear(&_history); +} - if ( isUpdate ) { - targetStatus = targeter.targetUpdate( *_itemRef.getUpdate(), &endpoints ); - } - else if ( isDelete ) { - targetStatus = targeter.targetDelete( *_itemRef.getDelete(), &endpoints ); - } - else { - dassert( _itemRef.getOpType() == BatchedCommandRequest::BatchType_Insert ); +const BatchItemRef& WriteOp::getWriteItem() const { + return _itemRef; +} - ShardEndpoint* endpoint = NULL; - // TODO: Remove the index targeting stuff once there is a command for it - if ( !isIndexInsert ) { - targetStatus = targeter.targetInsert( _itemRef.getDocument(), &endpoint ); - } - else { - // TODO: Retry index writes with stale version? - targetStatus = targeter.targetCollection( &endpoints ); - } +WriteOpState WriteOp::getWriteState() const { + return _state; +} - if ( !targetStatus.isOK() ) { - dassert( NULL == endpoint ); - return targetStatus; - } +const WriteErrorDetail& WriteOp::getOpError() const { + dassert(_state == WriteOpState_Error); + return *_error; +} - // Store single endpoint result if we targeted a single endpoint - if ( endpoint ) endpoints.push_back( endpoint ); +Status WriteOp::targetWrites(const NSTargeter& targeter, + std::vector<TargetedWrite*>* targetedWrites) { + bool isUpdate = _itemRef.getOpType() == BatchedCommandRequest::BatchType_Update; + bool isDelete = _itemRef.getOpType() == BatchedCommandRequest::BatchType_Delete; + bool isIndexInsert = _itemRef.getRequest()->isInsertIndexRequest(); + + Status targetStatus = Status::OK(); + OwnedPointerVector<ShardEndpoint> endpointsOwned; + vector<ShardEndpoint*>& endpoints = endpointsOwned.mutableVector(); + + if (isUpdate) { + targetStatus = targeter.targetUpdate(*_itemRef.getUpdate(), &endpoints); + } else if (isDelete) { + targetStatus = targeter.targetDelete(*_itemRef.getDelete(), &endpoints); + } else { + dassert(_itemRef.getOpType() == BatchedCommandRequest::BatchType_Insert); + + ShardEndpoint* endpoint = NULL; + // TODO: Remove the index targeting stuff once there is a command for it + if (!isIndexInsert) { + targetStatus = targeter.targetInsert(_itemRef.getDocument(), &endpoint); + } else { + // TODO: Retry index writes with stale version? + targetStatus = targeter.targetCollection(&endpoints); } - // If we're targeting more than one endpoint with an update/delete, we have to target - // everywhere since we cannot currently retry partial results. - // NOTE: Index inserts are currently specially targeted only at the current collection to - // avoid creating collections everywhere. - if ( targetStatus.isOK() && endpoints.size() > 1u && !isIndexInsert ) { - endpointsOwned.clear(); - invariant( endpoints.empty() ); - targetStatus = targeter.targetAllShards( &endpoints ); + if (!targetStatus.isOK()) { + dassert(NULL == endpoint); + return targetStatus; } - // If we had an error, stop here - if ( !targetStatus.isOK() ) return targetStatus; + // Store single endpoint result if we targeted a single endpoint + if (endpoint) + endpoints.push_back(endpoint); + } - for ( vector<ShardEndpoint*>::iterator it = endpoints.begin(); it != endpoints.end(); - ++it ) { + // If we're targeting more than one endpoint with an update/delete, we have to target + // everywhere since we cannot currently retry partial results. + // NOTE: Index inserts are currently specially targeted only at the current collection to + // avoid creating collections everywhere. + if (targetStatus.isOK() && endpoints.size() > 1u && !isIndexInsert) { + endpointsOwned.clear(); + invariant(endpoints.empty()); + targetStatus = targeter.targetAllShards(&endpoints); + } - ShardEndpoint* endpoint = *it; + // If we had an error, stop here + if (!targetStatus.isOK()) + return targetStatus; - _childOps.push_back( new ChildWriteOp( this ) ); + for (vector<ShardEndpoint*>::iterator it = endpoints.begin(); it != endpoints.end(); ++it) { + ShardEndpoint* endpoint = *it; - WriteOpRef ref( _itemRef.getItemIndex(), _childOps.size() - 1 ); + _childOps.push_back(new ChildWriteOp(this)); - // For now, multiple endpoints imply no versioning - we can't retry half a multi-write - if ( endpoints.size() == 1u ) { - targetedWrites->push_back( new TargetedWrite( *endpoint, ref ) ); - } - else { - ShardEndpoint broadcastEndpoint( endpoint->shardName, - ChunkVersion::IGNORED() ); - targetedWrites->push_back( new TargetedWrite( broadcastEndpoint, ref ) ); - } + WriteOpRef ref(_itemRef.getItemIndex(), _childOps.size() - 1); - _childOps.back()->pendingWrite = targetedWrites->back(); - _childOps.back()->state = WriteOpState_Pending; + // For now, multiple endpoints imply no versioning - we can't retry half a multi-write + if (endpoints.size() == 1u) { + targetedWrites->push_back(new TargetedWrite(*endpoint, ref)); + } else { + ShardEndpoint broadcastEndpoint(endpoint->shardName, ChunkVersion::IGNORED()); + targetedWrites->push_back(new TargetedWrite(broadcastEndpoint, ref)); } - _state = WriteOpState_Pending; - return Status::OK(); - } - - size_t WriteOp::getNumTargeted() { - return _childOps.size(); + _childOps.back()->pendingWrite = targetedWrites->back(); + _childOps.back()->state = WriteOpState_Pending; } - static bool isRetryErrCode( int errCode ) { - return errCode == ErrorCodes::StaleShardVersion; - } + _state = WriteOpState_Pending; + return Status::OK(); +} - // Aggregate a bunch of errors for a single op together - static void combineOpErrors( const vector<ChildWriteOp*>& errOps, WriteErrorDetail* error ) { +size_t WriteOp::getNumTargeted() { + return _childOps.size(); +} - // Special case single response - if ( errOps.size() == 1 ) { - errOps.front()->error->cloneTo( error ); - return; - } +static bool isRetryErrCode(int errCode) { + return errCode == ErrorCodes::StaleShardVersion; +} - error->setErrCode( ErrorCodes::MultipleErrorsOccurred ); +// Aggregate a bunch of errors for a single op together +static void combineOpErrors(const vector<ChildWriteOp*>& errOps, WriteErrorDetail* error) { + // Special case single response + if (errOps.size() == 1) { + errOps.front()->error->cloneTo(error); + return; + } - // Generate the multi-error message below - stringstream msg; - msg << "multiple errors for op : "; + error->setErrCode(ErrorCodes::MultipleErrorsOccurred); - BSONArrayBuilder errB; - for ( vector<ChildWriteOp*>::const_iterator it = errOps.begin(); it != errOps.end(); - ++it ) { - const ChildWriteOp* errOp = *it; - if ( it != errOps.begin() ) msg << " :: and :: "; - msg << errOp->error->getErrMessage(); - errB.append( errOp->error->toBSON() ); - } + // Generate the multi-error message below + stringstream msg; + msg << "multiple errors for op : "; - error->setErrInfo( BSON( "causedBy" << errB.arr() ) ); - error->setIndex( errOps.front()->error->getIndex() ); - error->setErrMessage( msg.str() ); + BSONArrayBuilder errB; + for (vector<ChildWriteOp*>::const_iterator it = errOps.begin(); it != errOps.end(); ++it) { + const ChildWriteOp* errOp = *it; + if (it != errOps.begin()) + msg << " :: and :: "; + msg << errOp->error->getErrMessage(); + errB.append(errOp->error->toBSON()); } - /** - * This is the core function which aggregates all the results of a write operation on multiple - * shards and updates the write operation's state. - */ - void WriteOp::updateOpState() { - - vector<ChildWriteOp*> childErrors; - - bool isRetryError = true; - for ( vector<ChildWriteOp*>::iterator it = _childOps.begin(); it != _childOps.end(); - it++ ) { + error->setErrInfo(BSON("causedBy" << errB.arr())); + error->setIndex(errOps.front()->error->getIndex()); + error->setErrMessage(msg.str()); +} - ChildWriteOp* childOp = *it; +/** + * This is the core function which aggregates all the results of a write operation on multiple + * shards and updates the write operation's state. + */ +void WriteOp::updateOpState() { + vector<ChildWriteOp*> childErrors; - // Don't do anything till we have all the info - if ( childOp->state != WriteOpState_Completed - && childOp->state != WriteOpState_Error ) { - return; - } + bool isRetryError = true; + for (vector<ChildWriteOp*>::iterator it = _childOps.begin(); it != _childOps.end(); it++) { + ChildWriteOp* childOp = *it; - if ( childOp->state == WriteOpState_Error ) { - childErrors.push_back( childOp ); - // Any non-retry error aborts all - if ( !isRetryErrCode( childOp->error->getErrCode() ) ) isRetryError = false; - } + // Don't do anything till we have all the info + if (childOp->state != WriteOpState_Completed && childOp->state != WriteOpState_Error) { + return; } - if ( !childErrors.empty() && isRetryError ) { - // Since we're using broadcast mode for multi-shard writes, which cannot SCE - dassert( childErrors.size() == 1u ); - _state = WriteOpState_Ready; - } - else if ( !childErrors.empty() ) { - _error.reset( new WriteErrorDetail ); - combineOpErrors( childErrors, _error.get() ); - _state = WriteOpState_Error; + if (childOp->state == WriteOpState_Error) { + childErrors.push_back(childOp); + // Any non-retry error aborts all + if (!isRetryErrCode(childOp->error->getErrCode())) + isRetryError = false; } - else { - _state = WriteOpState_Completed; - } - - // Now that we're done with the child ops, do something with them - // TODO: Don't store unlimited history? - dassert( _state != WriteOpState_Pending ); - _history.insert( _history.end(), _childOps.begin(), _childOps.end() ); - _childOps.clear(); } - void WriteOp::cancelWrites( const WriteErrorDetail* why ) { + if (!childErrors.empty() && isRetryError) { + // Since we're using broadcast mode for multi-shard writes, which cannot SCE + dassert(childErrors.size() == 1u); + _state = WriteOpState_Ready; + } else if (!childErrors.empty()) { + _error.reset(new WriteErrorDetail); + combineOpErrors(childErrors, _error.get()); + _state = WriteOpState_Error; + } else { + _state = WriteOpState_Completed; + } - dassert( _state == WriteOpState_Pending || _state == WriteOpState_Ready ); - for ( vector<ChildWriteOp*>::iterator it = _childOps.begin(); it != _childOps.end(); - ++it ) { + // Now that we're done with the child ops, do something with them + // TODO: Don't store unlimited history? + dassert(_state != WriteOpState_Pending); + _history.insert(_history.end(), _childOps.begin(), _childOps.end()); + _childOps.clear(); +} - ChildWriteOp* childOp = *it; +void WriteOp::cancelWrites(const WriteErrorDetail* why) { + dassert(_state == WriteOpState_Pending || _state == WriteOpState_Ready); + for (vector<ChildWriteOp*>::iterator it = _childOps.begin(); it != _childOps.end(); ++it) { + ChildWriteOp* childOp = *it; - if ( childOp->state == WriteOpState_Pending ) { - childOp->endpoint.reset( new ShardEndpoint( childOp->pendingWrite->endpoint ) ); - if ( why ) { - childOp->error.reset( new WriteErrorDetail ); - why->cloneTo( childOp->error.get() ); - } - childOp->state = WriteOpState_Cancelled; + if (childOp->state == WriteOpState_Pending) { + childOp->endpoint.reset(new ShardEndpoint(childOp->pendingWrite->endpoint)); + if (why) { + childOp->error.reset(new WriteErrorDetail); + why->cloneTo(childOp->error.get()); } + childOp->state = WriteOpState_Cancelled; } - - _history.insert( _history.end(), _childOps.begin(), _childOps.end() ); - _childOps.clear(); - - _state = WriteOpState_Ready; } - void WriteOp::noteWriteComplete( const TargetedWrite& targetedWrite ) { - - const WriteOpRef& ref = targetedWrite.writeOpRef; - dassert( static_cast<size_t>( ref.second ) < _childOps.size() ); - ChildWriteOp& childOp = *_childOps.at( ref.second ); - - childOp.pendingWrite = NULL; - childOp.endpoint.reset( new ShardEndpoint( targetedWrite.endpoint ) ); - childOp.state = WriteOpState_Completed; - updateOpState(); - } + _history.insert(_history.end(), _childOps.begin(), _childOps.end()); + _childOps.clear(); - void WriteOp::noteWriteError( const TargetedWrite& targetedWrite, - const WriteErrorDetail& error ) { + _state = WriteOpState_Ready; +} - const WriteOpRef& ref = targetedWrite.writeOpRef; - ChildWriteOp& childOp = *_childOps.at( ref.second ); +void WriteOp::noteWriteComplete(const TargetedWrite& targetedWrite) { + const WriteOpRef& ref = targetedWrite.writeOpRef; + dassert(static_cast<size_t>(ref.second) < _childOps.size()); + ChildWriteOp& childOp = *_childOps.at(ref.second); - childOp.pendingWrite = NULL; - childOp.endpoint.reset( new ShardEndpoint( targetedWrite.endpoint ) ); - childOp.error.reset( new WriteErrorDetail ); - error.cloneTo( childOp.error.get() ); - dassert( ref.first == _itemRef.getItemIndex() ); - childOp.error->setIndex( _itemRef.getItemIndex() ); - childOp.state = WriteOpState_Error; - updateOpState(); - } + childOp.pendingWrite = NULL; + childOp.endpoint.reset(new ShardEndpoint(targetedWrite.endpoint)); + childOp.state = WriteOpState_Completed; + updateOpState(); +} - void WriteOp::setOpError( const WriteErrorDetail& error ) { - dassert( _state == WriteOpState_Ready ); - _error.reset( new WriteErrorDetail ); - error.cloneTo( _error.get() ); - _error->setIndex( _itemRef.getItemIndex() ); - _state = WriteOpState_Error; - // No need to updateOpState, set directly - } +void WriteOp::noteWriteError(const TargetedWrite& targetedWrite, const WriteErrorDetail& error) { + const WriteOpRef& ref = targetedWrite.writeOpRef; + ChildWriteOp& childOp = *_childOps.at(ref.second); + + childOp.pendingWrite = NULL; + childOp.endpoint.reset(new ShardEndpoint(targetedWrite.endpoint)); + childOp.error.reset(new WriteErrorDetail); + error.cloneTo(childOp.error.get()); + dassert(ref.first == _itemRef.getItemIndex()); + childOp.error->setIndex(_itemRef.getItemIndex()); + childOp.state = WriteOpState_Error; + updateOpState(); +} +void WriteOp::setOpError(const WriteErrorDetail& error) { + dassert(_state == WriteOpState_Ready); + _error.reset(new WriteErrorDetail); + error.cloneTo(_error.get()); + _error->setIndex(_itemRef.getItemIndex()); + _state = WriteOpState_Error; + // No need to updateOpState, set directly +} } diff --git a/src/mongo/s/write_ops/write_op.h b/src/mongo/s/write_ops/write_op.h index 98ccf0d691d..fb189edfffb 100644 --- a/src/mongo/s/write_ops/write_op.h +++ b/src/mongo/s/write_ops/write_op.h @@ -38,206 +38,196 @@ namespace mongo { - struct TargetedWrite; - struct ChildWriteOp; +struct TargetedWrite; +struct ChildWriteOp; - enum WriteOpState { +enum WriteOpState { - // Item is ready to be targeted - WriteOpState_Ready, + // Item is ready to be targeted + WriteOpState_Ready, - // Item is targeted and we're waiting for outstanding shard requests to populate - // responses - WriteOpState_Pending, + // Item is targeted and we're waiting for outstanding shard requests to populate + // responses + WriteOpState_Pending, - // Op was successful, write completed - // We assume all states higher than this one are *final* - WriteOpState_Completed, + // Op was successful, write completed + // We assume all states higher than this one are *final* + WriteOpState_Completed, - // Op failed with some error - WriteOpState_Error, + // Op failed with some error + WriteOpState_Error, - // Op was cancelled before sending (only child write ops can be cancelled) - WriteOpState_Cancelled, + // Op was cancelled before sending (only child write ops can be cancelled) + WriteOpState_Cancelled, - // Catch-all error state. - WriteOpState_Unknown - }; + // Catch-all error state. + WriteOpState_Unknown +}; + +/** + * State of a single write item in-progress from a client request. + * + * The lifecyle of a write op: + * + * 0. Begins at _Ready, + * + * 1a. Targeted, and a ChildWriteOp created to track the state of each returned TargetedWrite. + * The state is changed to _Pending. + * 1b. If the op cannot be targeted, the error is set directly (_Error), and the write op is + * completed. + * + * 2a. The current TargetedWrites are cancelled, and the op state is reset to _Ready + * 2b. TargetedWrites finish successfully and unsuccessfully. + * + * On the last error arriving... + * + * 3a. If the errors allow for retry, the WriteOp is reset to _Ready, previous ChildWriteOps + * are placed in the history, and goto 0. + * 3b. If the errors don't allow for retry, they are combined into a single error and the + * state is changed to _Error. + * 3c. If there are no errors, the state is changed to _Completed. + * + * WriteOps finish in a _Completed or _Error state. + */ +class WriteOp { +public: + WriteOp(const BatchItemRef& itemRef) : _itemRef(itemRef), _state(WriteOpState_Ready) {} + + ~WriteOp(); /** - * State of a single write item in-progress from a client request. - * - * The lifecyle of a write op: + * Returns the write item for this operation + */ + const BatchItemRef& getWriteItem() const; + + /** + * Returns the op's current state. + */ + WriteOpState getWriteState() const; + + /** + * Returns the op's error. * - * 0. Begins at _Ready, + * Can only be used in state _Error + */ + const WriteErrorDetail& getOpError() const; + + /** + * Creates TargetedWrite operations for every applicable shard, which contain the + * information needed to send the child writes generated from this write item. * - * 1a. Targeted, and a ChildWriteOp created to track the state of each returned TargetedWrite. - * The state is changed to _Pending. - * 1b. If the op cannot be targeted, the error is set directly (_Error), and the write op is - * completed. + * The ShardTargeter determines the ShardEndpoints to send child writes to, but is not + * modified by this operation. * - * 2a. The current TargetedWrites are cancelled, and the op state is reset to _Ready - * 2b. TargetedWrites finish successfully and unsuccessfully. + * Returns !OK if the targeting process itself fails + * (no TargetedWrites will be added, state unchanged) + */ + Status targetWrites(const NSTargeter& targeter, std::vector<TargetedWrite*>* targetedWrites); + + /** + * Returns the number of child writes that were last targeted. + */ + size_t getNumTargeted(); + + /** + * Resets the state of this write op to _Ready and stops waiting for any outstanding + * TargetedWrites. Optional error can be provided for reporting. * - * On the last error arriving... + * Can only be called when state is _Pending, or is a no-op if called when the state + * is still _Ready (and therefore no writes are pending). + */ + void cancelWrites(const WriteErrorDetail* why); + + /** + * Marks the targeted write as finished for this write op. * - * 3a. If the errors allow for retry, the WriteOp is reset to _Ready, previous ChildWriteOps - * are placed in the history, and goto 0. - * 3b. If the errors don't allow for retry, they are combined into a single error and the - * state is changed to _Error. - * 3c. If there are no errors, the state is changed to _Completed. + * One of noteWriteComplete or noteWriteError should be called exactly once for every + * TargetedWrite. + */ + void noteWriteComplete(const TargetedWrite& targetedWrite); + + /** + * Stores the error response of a TargetedWrite for later use, marks the write as finished. * - * WriteOps finish in a _Completed or _Error state. + * As above, one of noteWriteComplete or noteWriteError should be called exactly once for + * every TargetedWrite. */ - class WriteOp { - public: - - WriteOp( const BatchItemRef& itemRef ) : - _itemRef( itemRef ), _state( WriteOpState_Ready ) { - } - - ~WriteOp(); - - /** - * Returns the write item for this operation - */ - const BatchItemRef& getWriteItem() const; - - /** - * Returns the op's current state. - */ - WriteOpState getWriteState() const; - - /** - * Returns the op's error. - * - * Can only be used in state _Error - */ - const WriteErrorDetail& getOpError() const; - - /** - * Creates TargetedWrite operations for every applicable shard, which contain the - * information needed to send the child writes generated from this write item. - * - * The ShardTargeter determines the ShardEndpoints to send child writes to, but is not - * modified by this operation. - * - * Returns !OK if the targeting process itself fails - * (no TargetedWrites will be added, state unchanged) - */ - Status targetWrites( const NSTargeter& targeter, - std::vector<TargetedWrite*>* targetedWrites ); - - /** - * Returns the number of child writes that were last targeted. - */ - size_t getNumTargeted(); - - /** - * Resets the state of this write op to _Ready and stops waiting for any outstanding - * TargetedWrites. Optional error can be provided for reporting. - * - * Can only be called when state is _Pending, or is a no-op if called when the state - * is still _Ready (and therefore no writes are pending). - */ - void cancelWrites( const WriteErrorDetail* why ); - - /** - * Marks the targeted write as finished for this write op. - * - * One of noteWriteComplete or noteWriteError should be called exactly once for every - * TargetedWrite. - */ - void noteWriteComplete( const TargetedWrite& targetedWrite ); - - /** - * Stores the error response of a TargetedWrite for later use, marks the write as finished. - * - * As above, one of noteWriteComplete or noteWriteError should be called exactly once for - * every TargetedWrite. - */ - void noteWriteError( const TargetedWrite& targetedWrite, const WriteErrorDetail& error ); - - /** - * Sets the error for this write op directly, and forces the state to _Error. - * - * Should only be used when in state _Ready. - */ - void setOpError( const WriteErrorDetail& error ); - - private: - - /** - * Updates the op state after new information is received. - */ - void updateOpState(); - - // Owned elsewhere, reference to a batch with a write item - const BatchItemRef _itemRef; - - // What stage of the operation we are at - WriteOpState _state; - - // filled when state == _Pending - std::vector<ChildWriteOp*> _childOps; - - // filled when state == _Error - std::unique_ptr<WriteErrorDetail> _error; - - // Finished child operations, for debugging - std::vector<ChildWriteOp*> _history; - }; + void noteWriteError(const TargetedWrite& targetedWrite, const WriteErrorDetail& error); /** - * State of a write in-progress (to a single shard) which is one part of a larger write - * operation. + * Sets the error for this write op directly, and forces the state to _Error. * - * As above, the write op may finish in either a successful (_Completed) or unsuccessful - * (_Error) state. + * Should only be used when in state _Ready. */ - struct ChildWriteOp { + void setOpError(const WriteErrorDetail& error); - ChildWriteOp( WriteOp* const parent ) : - parentOp( parent ), state( WriteOpState_Ready ), pendingWrite( NULL ) { - } +private: + /** + * Updates the op state after new information is received. + */ + void updateOpState(); - const WriteOp* const parentOp; - WriteOpState state; + // Owned elsewhere, reference to a batch with a write item + const BatchItemRef _itemRef; - // non-zero when state == _Pending - // Not owned here but tracked for reporting - TargetedWrite* pendingWrite; + // What stage of the operation we are at + WriteOpState _state; - // filled when state > _Pending - std::unique_ptr<ShardEndpoint> endpoint; + // filled when state == _Pending + std::vector<ChildWriteOp*> _childOps; - // filled when state == _Error or (optionally) when state == _Cancelled - std::unique_ptr<WriteErrorDetail> error; - }; + // filled when state == _Error + std::unique_ptr<WriteErrorDetail> _error; - // First value is write item index in the batch, second value is child write op index - typedef std::pair<int, int> WriteOpRef; + // Finished child operations, for debugging + std::vector<ChildWriteOp*> _history; +}; - /** - * A write with A) a request targeted at a particular shard endpoint, and B) a response targeted - * at a particular WriteOp. - * - * TargetedWrites are the link between the RPC layer and the in-progress write - * operation. - */ - struct TargetedWrite { +/** + * State of a write in-progress (to a single shard) which is one part of a larger write + * operation. + * + * As above, the write op may finish in either a successful (_Completed) or unsuccessful + * (_Error) state. + */ +struct ChildWriteOp { + ChildWriteOp(WriteOp* const parent) + : parentOp(parent), state(WriteOpState_Ready), pendingWrite(NULL) {} - TargetedWrite( const ShardEndpoint& endpoint, WriteOpRef writeOpRef ) : - endpoint( endpoint ), writeOpRef( writeOpRef ) { - } + const WriteOp* const parentOp; + WriteOpState state; - // Where to send the write - ShardEndpoint endpoint; + // non-zero when state == _Pending + // Not owned here but tracked for reporting + TargetedWrite* pendingWrite; - // Where to find the write item and put the response - // TODO: Could be a more complex handle, shared between write state and networking code if - // we need to be able to cancel ops. - WriteOpRef writeOpRef; - }; + // filled when state > _Pending + std::unique_ptr<ShardEndpoint> endpoint; + // filled when state == _Error or (optionally) when state == _Cancelled + std::unique_ptr<WriteErrorDetail> error; +}; + +// First value is write item index in the batch, second value is child write op index +typedef std::pair<int, int> WriteOpRef; + +/** + * A write with A) a request targeted at a particular shard endpoint, and B) a response targeted + * at a particular WriteOp. + * + * TargetedWrites are the link between the RPC layer and the in-progress write + * operation. + */ +struct TargetedWrite { + TargetedWrite(const ShardEndpoint& endpoint, WriteOpRef writeOpRef) + : endpoint(endpoint), writeOpRef(writeOpRef) {} + + // Where to send the write + ShardEndpoint endpoint; + + // Where to find the write item and put the response + // TODO: Could be a more complex handle, shared between write state and networking code if + // we need to be able to cancel ops. + WriteOpRef writeOpRef; +}; } diff --git a/src/mongo/s/write_ops/write_op_test.cpp b/src/mongo/s/write_ops/write_op_test.cpp index fc5caa05dfb..b836caae80c 100644 --- a/src/mongo/s/write_ops/write_op_test.cpp +++ b/src/mongo/s/write_ops/write_op_test.cpp @@ -39,365 +39,322 @@ namespace { - using std::unique_ptr; - using std::string; - using std::vector; +using std::unique_ptr; +using std::string; +using std::vector; - using namespace mongo; +using namespace mongo; - WriteErrorDetail* buildError( int code, const BSONObj& info, const string& message ) { +WriteErrorDetail* buildError(int code, const BSONObj& info, const string& message) { + WriteErrorDetail* error = new WriteErrorDetail(); + error->setErrCode(code); + error->setErrInfo(info); + error->setErrMessage(message); - WriteErrorDetail* error = new WriteErrorDetail(); - error->setErrCode( code ); - error->setErrInfo( info ); - error->setErrMessage( message ); + return error; +} - return error; - } +TEST(WriteOpTests, BasicError) { + // + // Test of basic error-setting on write op + // - TEST(WriteOpTests, BasicError) { + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS("foo.bar"); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - // - // Test of basic error-setting on write op - // + WriteOp writeOp(BatchItemRef(&request, 0)); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( "foo.bar" ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + unique_ptr<WriteErrorDetail> error( + buildError(ErrorCodes::UnknownError, BSON("data" << 12345), "some message")); - WriteOp writeOp( BatchItemRef( &request, 0 ) ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); + writeOp.setOpError(*error); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Error); + ASSERT_EQUALS(writeOp.getOpError().getErrCode(), error->getErrCode()); + ASSERT_EQUALS(writeOp.getOpError().getErrInfo()["data"].Int(), + error->getErrInfo()["data"].Int()); + ASSERT_EQUALS(writeOp.getOpError().getErrMessage(), error->getErrMessage()); +} - unique_ptr<WriteErrorDetail> error( buildError( ErrorCodes::UnknownError, - BSON( "data" << 12345 ), - "some message" ) ); +TEST(WriteOpTests, TargetSingle) { + // + // Basic targeting test + // - writeOp.setOpError( *error ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Error ); - ASSERT_EQUALS( writeOp.getOpError().getErrCode(), error->getErrCode() ); - ASSERT_EQUALS( writeOp.getOpError().getErrInfo()["data"].Int(), - error->getErrInfo()["data"].Int() ); - ASSERT_EQUALS( writeOp.getOpError().getErrMessage(), error->getErrMessage() ); - } + NamespaceString nss("foo.bar"); - TEST(WriteOpTests, TargetSingle) { + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); - // - // Basic targeting test - // + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpoint, nss, BSON("x" << MINKEY), BSON("x" << MAXKEY))); - NamespaceString nss( "foo.bar" ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); + // Do single-target write op - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpoint, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << MAXKEY ) ) ); + WriteOp writeOp(BatchItemRef(&request, 0)); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + MockNSTargeter targeter; + targeter.init(mockRanges); - // Do single-target write op + OwnedPointerVector<TargetedWrite> targetedOwned; + vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); + Status status = writeOp.targetWrites(targeter, &targeted); - WriteOp writeOp( BatchItemRef( &request, 0 ) ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); + ASSERT(status.isOK()); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Pending); + ASSERT_EQUALS(targeted.size(), 1u); + assertEndpointsEqual(targeted.front()->endpoint, endpoint); - MockNSTargeter targeter; - targeter.init( mockRanges ); + writeOp.noteWriteComplete(*targeted.front()); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Completed); +} - OwnedPointerVector<TargetedWrite> targetedOwned; - vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); - Status status = writeOp.targetWrites( targeter, &targeted ); +BatchedDeleteDocument* buildDeleteDoc(const BSONObj& doc) { + BatchedDeleteDocument* deleteDoc = new BatchedDeleteDocument(); - ASSERT( status.isOK() ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Pending ); - ASSERT_EQUALS( targeted.size(), 1u ); - assertEndpointsEqual( targeted.front()->endpoint, endpoint ); + string errMsg; + bool ok = deleteDoc->parseBSON(doc, &errMsg); + ASSERT_EQUALS(errMsg, ""); + ASSERT(ok); + return deleteDoc; +} - writeOp.noteWriteComplete( *targeted.front() ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Completed ); +struct EndpointComp { + bool operator()(const TargetedWrite* writeA, const TargetedWrite* writeB) const { + return writeA->endpoint.shardName.compare(writeB->endpoint.shardName) < 0; } +}; - BatchedDeleteDocument* buildDeleteDoc( const BSONObj& doc ) { +inline void sortByEndpoint(vector<TargetedWrite*>* writes) { + std::sort(writes->begin(), writes->end(), EndpointComp()); +} - BatchedDeleteDocument* deleteDoc = new BatchedDeleteDocument(); +TEST(WriteOpTests, TargetMultiOneShard) { + // + // Multi-write targeting test where our query goes to one shard + // - string errMsg; - bool ok = deleteDoc->parseBSON( doc, &errMsg ); - ASSERT_EQUALS( errMsg, "" ); - ASSERT( ok ); - return deleteDoc; - } + NamespaceString nss("foo.bar"); - struct EndpointComp { - bool operator()( const TargetedWrite* writeA, const TargetedWrite* writeB ) const { - return writeA->endpoint.shardName.compare( writeB->endpoint.shardName ) < 0; - } - }; + ShardEndpoint endpointA("shardA", ChunkVersion(10, 0, OID())); + ShardEndpoint endpointB("shardB", ChunkVersion(20, 0, OID())); + ShardEndpoint endpointC("shardB", ChunkVersion(20, 0, OID())); - inline void sortByEndpoint( vector<TargetedWrite*>* writes ) { - std::sort( writes->begin(), writes->end(), EndpointComp() ); - } + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpointA, nss, BSON("x" << MINKEY), BSON("x" << 0))); + mockRanges.push_back(new MockRange(endpointB, nss, BSON("x" << 0), BSON("x" << 10))); + mockRanges.push_back(new MockRange(endpointC, nss, BSON("x" << 10), BSON("x" << MAXKEY))); - TEST(WriteOpTests, TargetMultiOneShard) { + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); + // Only hits first shard + BSONObj query = BSON("x" << GTE << -2 << LT << -1); + request.getDeleteRequest()->addToDeletes(buildDeleteDoc(BSON("q" << query))); - // - // Multi-write targeting test where our query goes to one shard - // + WriteOp writeOp(BatchItemRef(&request, 0)); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); - NamespaceString nss( "foo.bar" ); + MockNSTargeter targeter; + targeter.init(mockRanges); - ShardEndpoint endpointA( "shardA", ChunkVersion(10, 0, OID()) ); - ShardEndpoint endpointB( "shardB", ChunkVersion(20, 0, OID()) ); - ShardEndpoint endpointC( "shardB", ChunkVersion(20, 0, OID()) ); + OwnedPointerVector<TargetedWrite> targetedOwned; + vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); + Status status = writeOp.targetWrites(targeter, &targeted); - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpointA, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << 0 ) ) ); - mockRanges.push_back( new MockRange( endpointB, - nss, - BSON( "x" << 0 ), - BSON( "x" << 10 ) ) ); - mockRanges.push_back( new MockRange( endpointC, - nss, - BSON( "x" << 10 ), - BSON( "x" << MAXKEY ) ) ); + ASSERT(status.isOK()); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Pending); + ASSERT_EQUALS(targeted.size(), 1u); + assertEndpointsEqual(targeted.front()->endpoint, endpointA); - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Delete ); - request.setNS( nss.ns() ); - // Only hits first shard - BSONObj query = BSON( "x" << GTE << -2 << LT << -1 ); - request.getDeleteRequest()->addToDeletes( buildDeleteDoc( BSON( "q" << query ) ) ); + writeOp.noteWriteComplete(*targeted.front()); - WriteOp writeOp( BatchItemRef( &request, 0 ) ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Completed); +} - MockNSTargeter targeter; - targeter.init( mockRanges ); +TEST(WriteOpTests, TargetMultiAllShards) { + // + // Multi-write targeting test where our write goes to more than one shard + // - OwnedPointerVector<TargetedWrite> targetedOwned; - vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); - Status status = writeOp.targetWrites( targeter, &targeted ); + NamespaceString nss("foo.bar"); - ASSERT( status.isOK() ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Pending ); - ASSERT_EQUALS( targeted.size(), 1u ); - assertEndpointsEqual( targeted.front()->endpoint, endpointA ); + ShardEndpoint endpointA("shardA", ChunkVersion(10, 0, OID())); + ShardEndpoint endpointB("shardB", ChunkVersion(20, 0, OID())); + ShardEndpoint endpointC("shardB", ChunkVersion(20, 0, OID())); - writeOp.noteWriteComplete( *targeted.front() ); + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpointA, nss, BSON("x" << MINKEY), BSON("x" << 0))); + mockRanges.push_back(new MockRange(endpointB, nss, BSON("x" << 0), BSON("x" << 10))); + mockRanges.push_back(new MockRange(endpointC, nss, BSON("x" << 10), BSON("x" << MAXKEY))); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Completed ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Delete); + request.setNS(nss.ns()); + BSONObj query = BSON("x" << GTE << -1 << LT << 1); + request.getDeleteRequest()->addToDeletes(buildDeleteDoc(BSON("q" << query))); - } + // Do multi-target write op - TEST(WriteOpTests, TargetMultiAllShards) { - - // - // Multi-write targeting test where our write goes to more than one shard - // - - NamespaceString nss( "foo.bar" ); - - ShardEndpoint endpointA( "shardA", ChunkVersion(10, 0, OID()) ); - ShardEndpoint endpointB( "shardB", ChunkVersion(20, 0, OID()) ); - ShardEndpoint endpointC( "shardB", ChunkVersion(20, 0, OID()) ); - - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpointA, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << 0 ) ) ); - mockRanges.push_back( new MockRange( endpointB, - nss, - BSON( "x" << 0 ), - BSON( "x" << 10 ) ) ); - mockRanges.push_back( new MockRange( endpointC, - nss, - BSON( "x" << 10 ), - BSON( "x" << MAXKEY ) ) ); - - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Delete ); - request.setNS( nss.ns() ); - BSONObj query = BSON( "x" << GTE << -1 << LT << 1 ); - request.getDeleteRequest()->addToDeletes( buildDeleteDoc( BSON( "q" << query ) ) ); - - // Do multi-target write op - - WriteOp writeOp( BatchItemRef( &request, 0 ) ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); - - MockNSTargeter targeter; - targeter.init( mockRanges ); - - OwnedPointerVector<TargetedWrite> targetedOwned; - vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); - Status status = writeOp.targetWrites( targeter, &targeted ); - - ASSERT( status.isOK() ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Pending ); - ASSERT_EQUALS( targeted.size(), 3u ); - sortByEndpoint( &targeted ); - ASSERT_EQUALS( targeted[0]->endpoint.shardName, endpointA.shardName ); - ASSERT( ChunkVersion::isIgnoredVersion( targeted[0]->endpoint.shardVersion ) ); - ASSERT_EQUALS( targeted[1]->endpoint.shardName, endpointB.shardName ); - ASSERT( ChunkVersion::isIgnoredVersion( targeted[1]->endpoint.shardVersion ) ); - ASSERT_EQUALS( targeted[2]->endpoint.shardName, endpointC.shardName ); - ASSERT( ChunkVersion::isIgnoredVersion( targeted[2]->endpoint.shardVersion ) ); - - writeOp.noteWriteComplete( *targeted[0] ); - writeOp.noteWriteComplete( *targeted[1] ); - writeOp.noteWriteComplete( *targeted[2] ); - - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Completed ); - } + WriteOp writeOp(BatchItemRef(&request, 0)); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); - TEST(WriteOpTests, ErrorSingle) { + MockNSTargeter targeter; + targeter.init(mockRanges); - // - // Single error after targeting test - // + OwnedPointerVector<TargetedWrite> targetedOwned; + vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); + Status status = writeOp.targetWrites(targeter, &targeted); - NamespaceString nss( "foo.bar" ); + ASSERT(status.isOK()); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Pending); + ASSERT_EQUALS(targeted.size(), 3u); + sortByEndpoint(&targeted); + ASSERT_EQUALS(targeted[0]->endpoint.shardName, endpointA.shardName); + ASSERT(ChunkVersion::isIgnoredVersion(targeted[0]->endpoint.shardVersion)); + ASSERT_EQUALS(targeted[1]->endpoint.shardName, endpointB.shardName); + ASSERT(ChunkVersion::isIgnoredVersion(targeted[1]->endpoint.shardVersion)); + ASSERT_EQUALS(targeted[2]->endpoint.shardName, endpointC.shardName); + ASSERT(ChunkVersion::isIgnoredVersion(targeted[2]->endpoint.shardVersion)); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); + writeOp.noteWriteComplete(*targeted[0]); + writeOp.noteWriteComplete(*targeted[1]); + writeOp.noteWriteComplete(*targeted[2]); - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpoint, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << MAXKEY ) ) ); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Completed); +} - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); +TEST(WriteOpTests, ErrorSingle) { + // + // Single error after targeting test + // - // Do single-target write op + NamespaceString nss("foo.bar"); - WriteOp writeOp( BatchItemRef( &request, 0 ) ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); - MockNSTargeter targeter; - targeter.init( mockRanges ); + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpoint, nss, BSON("x" << MINKEY), BSON("x" << MAXKEY))); - OwnedPointerVector<TargetedWrite> targetedOwned; - vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); - Status status = writeOp.targetWrites( targeter, &targeted ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - ASSERT( status.isOK() ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Pending ); - ASSERT_EQUALS( targeted.size(), 1u ); - assertEndpointsEqual( targeted.front()->endpoint, endpoint ); + // Do single-target write op - unique_ptr<WriteErrorDetail> error( buildError( ErrorCodes::UnknownError, - BSON( "data" << 12345 ), - "some message" ) ); + WriteOp writeOp(BatchItemRef(&request, 0)); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); - writeOp.noteWriteError( *targeted.front(), *error ); + MockNSTargeter targeter; + targeter.init(mockRanges); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Error ); - ASSERT_EQUALS( writeOp.getOpError().getErrCode(), error->getErrCode() ); - ASSERT_EQUALS( writeOp.getOpError().getErrInfo()["data"].Int(), - error->getErrInfo()["data"].Int() ); - ASSERT_EQUALS( writeOp.getOpError().getErrMessage(), error->getErrMessage() ); - } + OwnedPointerVector<TargetedWrite> targetedOwned; + vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); + Status status = writeOp.targetWrites(targeter, &targeted); - TEST(WriteOpTests, CancelSingle) { + ASSERT(status.isOK()); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Pending); + ASSERT_EQUALS(targeted.size(), 1u); + assertEndpointsEqual(targeted.front()->endpoint, endpoint); - // - // Cancel single targeting test - // + unique_ptr<WriteErrorDetail> error( + buildError(ErrorCodes::UnknownError, BSON("data" << 12345), "some message")); - NamespaceString nss( "foo.bar" ); + writeOp.noteWriteError(*targeted.front(), *error); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Error); + ASSERT_EQUALS(writeOp.getOpError().getErrCode(), error->getErrCode()); + ASSERT_EQUALS(writeOp.getOpError().getErrInfo()["data"].Int(), + error->getErrInfo()["data"].Int()); + ASSERT_EQUALS(writeOp.getOpError().getErrMessage(), error->getErrMessage()); +} - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpoint, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << MAXKEY ) ) ); +TEST(WriteOpTests, CancelSingle) { + // + // Cancel single targeting test + // - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + NamespaceString nss("foo.bar"); - // Do single-target write op + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); - WriteOp writeOp( BatchItemRef( &request, 0 ) ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpoint, nss, BSON("x" << MINKEY), BSON("x" << MAXKEY))); - MockNSTargeter targeter; - targeter.init( mockRanges ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - OwnedPointerVector<TargetedWrite> targetedOwned; - vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); - Status status = writeOp.targetWrites( targeter, &targeted ); + // Do single-target write op - ASSERT( status.isOK() ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Pending ); - ASSERT_EQUALS( targeted.size(), 1u ); - assertEndpointsEqual( targeted.front()->endpoint, endpoint ); + WriteOp writeOp(BatchItemRef(&request, 0)); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); - writeOp.cancelWrites( NULL ); + MockNSTargeter targeter; + targeter.init(mockRanges); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); - } + OwnedPointerVector<TargetedWrite> targetedOwned; + vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); + Status status = writeOp.targetWrites(targeter, &targeted); - // - // Test retryable errors - // + ASSERT(status.isOK()); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Pending); + ASSERT_EQUALS(targeted.size(), 1u); + assertEndpointsEqual(targeted.front()->endpoint, endpoint); + + writeOp.cancelWrites(NULL); + + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); +} - TEST(WriteOpTests, RetrySingleOp) { +// +// Test retryable errors +// - // - // Retry single targeting test - // +TEST(WriteOpTests, RetrySingleOp) { + // + // Retry single targeting test + // - NamespaceString nss( "foo.bar" ); + NamespaceString nss("foo.bar"); - ShardEndpoint endpoint( "shard", ChunkVersion::IGNORED() ); + ShardEndpoint endpoint("shard", ChunkVersion::IGNORED()); - vector<MockRange*> mockRanges; - mockRanges.push_back( new MockRange( endpoint, - nss, - BSON( "x" << MINKEY ), - BSON( "x" << MAXKEY ) ) ); + vector<MockRange*> mockRanges; + mockRanges.push_back(new MockRange(endpoint, nss, BSON("x" << MINKEY), BSON("x" << MAXKEY))); - BatchedCommandRequest request( BatchedCommandRequest::BatchType_Insert ); - request.setNS( nss.ns() ); - request.getInsertRequest()->addToDocuments( BSON( "x" << 1 ) ); + BatchedCommandRequest request(BatchedCommandRequest::BatchType_Insert); + request.setNS(nss.ns()); + request.getInsertRequest()->addToDocuments(BSON("x" << 1)); - // Do single-target write op + // Do single-target write op - WriteOp writeOp( BatchItemRef( &request, 0 ) ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); + WriteOp writeOp(BatchItemRef(&request, 0)); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); - MockNSTargeter targeter; - targeter.init( mockRanges ); + MockNSTargeter targeter; + targeter.init(mockRanges); - OwnedPointerVector<TargetedWrite> targetedOwned; - vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); - Status status = writeOp.targetWrites( targeter, &targeted ); + OwnedPointerVector<TargetedWrite> targetedOwned; + vector<TargetedWrite*>& targeted = targetedOwned.mutableVector(); + Status status = writeOp.targetWrites(targeter, &targeted); - ASSERT( status.isOK() ); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Pending ); - ASSERT_EQUALS( targeted.size(), 1u ); - assertEndpointsEqual( targeted.front()->endpoint, endpoint ); + ASSERT(status.isOK()); + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Pending); + ASSERT_EQUALS(targeted.size(), 1u); + assertEndpointsEqual(targeted.front()->endpoint, endpoint); - // Stale exception + // Stale exception - unique_ptr<WriteErrorDetail> error( buildError( ErrorCodes::StaleShardVersion, - BSON( "data" << 12345 ), - "some message" ) ); + unique_ptr<WriteErrorDetail> error( + buildError(ErrorCodes::StaleShardVersion, BSON("data" << 12345), "some message")); - writeOp.noteWriteError( *targeted.front(), *error ); + writeOp.noteWriteError(*targeted.front(), *error); - ASSERT_EQUALS( writeOp.getWriteState(), WriteOpState_Ready ); - } + ASSERT_EQUALS(writeOp.getWriteState(), WriteOpState_Ready); +} -} // unnamed namespace +} // unnamed namespace |