diff options
Diffstat (limited to 'src/mongo/db/repair_database.cpp')
| -rw-r--r-- | src/mongo/db/repair_database.cpp | 317 |
1 files changed, 159 insertions, 158 deletions
diff --git a/src/mongo/db/repair_database.cpp b/src/mongo/db/repair_database.cpp index 80963f123b7..a3de8953291 100644 --- a/src/mongo/db/repair_database.cpp +++ b/src/mongo/db/repair_database.cpp @@ -51,197 +51,198 @@ namespace mongo { - using std::endl; - using std::string; +using std::endl; +using std::string; namespace { - Status rebuildIndexesOnCollection(OperationContext* txn, - DatabaseCatalogEntry* dbce, - const std::string& collectionName) { - - CollectionCatalogEntry* cce = dbce->getCollectionCatalogEntry(collectionName); - - std::vector<string> indexNames; - std::vector<BSONObj> indexSpecs; - { - // Fetch all indexes - cce->getAllIndexes( txn, &indexNames ); - for ( size_t i = 0; i < indexNames.size(); i++ ) { - const string& name = indexNames[i]; - BSONObj spec = cce->getIndexSpec( txn, name ); - indexSpecs.push_back(spec.removeField("v").getOwned()); - - const BSONObj key = spec.getObjectField("key"); - const Status keyStatus = validateKeyPattern(key); - if (!keyStatus.isOK()) { - return Status(ErrorCodes::CannotCreateIndex, str::stream() +Status rebuildIndexesOnCollection(OperationContext* txn, + DatabaseCatalogEntry* dbce, + const std::string& collectionName) { + CollectionCatalogEntry* cce = dbce->getCollectionCatalogEntry(collectionName); + + std::vector<string> indexNames; + std::vector<BSONObj> indexSpecs; + { + // Fetch all indexes + cce->getAllIndexes(txn, &indexNames); + for (size_t i = 0; i < indexNames.size(); i++) { + const string& name = indexNames[i]; + BSONObj spec = cce->getIndexSpec(txn, name); + indexSpecs.push_back(spec.removeField("v").getOwned()); + + const BSONObj key = spec.getObjectField("key"); + const Status keyStatus = validateKeyPattern(key); + if (!keyStatus.isOK()) { + return Status( + ErrorCodes::CannotCreateIndex, + str::stream() << "Cannot rebuild index " << spec << ": " << keyStatus.reason() << " For more info see http://dochub.mongodb.org/core/index-validation"); - } } } + } - // Skip the rest if there are no indexes to rebuild. - if (indexSpecs.empty()) return Status::OK(); - - std::unique_ptr<Collection> collection; - std::unique_ptr<MultiIndexBlock> indexer; - { - // These steps are combined into a single WUOW to ensure there are no commits without - // the indexes. - // 1) Drop all indexes. - // 2) Open the Collection - // 3) Start the index build process. - - WriteUnitOfWork wuow(txn); - - { // 1 - for ( size_t i = 0; i < indexNames.size(); i++ ) { - Status s = cce->removeIndex(txn, indexNames[i]); - if (!s.isOK()) return s; - } - } + // Skip the rest if there are no indexes to rebuild. + if (indexSpecs.empty()) + return Status::OK(); - // Indexes must be dropped before we open the Collection otherwise we could attempt to - // open a bad index and fail. - // TODO see if MultiIndexBlock can be made to work without a Collection. - const StringData ns = cce->ns().ns(); - collection.reset(new Collection(txn, ns, cce, dbce->getRecordStore(ns), dbce)); - - indexer.reset(new MultiIndexBlock(txn, collection.get())); - Status status = indexer->init(indexSpecs); - if (!status.isOK()) { - // The WUOW will handle cleanup, so the indexer shouldn't do its own. - indexer->abortWithoutCleanup(); - return status; + std::unique_ptr<Collection> collection; + std::unique_ptr<MultiIndexBlock> indexer; + { + // These steps are combined into a single WUOW to ensure there are no commits without + // the indexes. + // 1) Drop all indexes. + // 2) Open the Collection + // 3) Start the index build process. + + WriteUnitOfWork wuow(txn); + + { // 1 + for (size_t i = 0; i < indexNames.size(); i++) { + Status s = cce->removeIndex(txn, indexNames[i]); + if (!s.isOK()) + return s; } - - wuow.commit(); } - // Iterate all records in the collection. Delete them if they aren't valid BSON. Index them - // if they are. - - long long numRecords = 0; - long long dataSize = 0; - - RecordStore* rs = collection->getRecordStore(); - auto cursor = rs->getCursor(txn); - while (auto record = cursor->next()) { - RecordId id = record->id; - RecordData& data = record->data; - - Status status = validateBSON(data.data(), data.size()); - if (!status.isOK()) { - log() << "Invalid BSON detected at " << id << ": " << status << ". Deleting."; - cursor->savePositioned(); // 'data' is no longer valid. - { - WriteUnitOfWork wunit(txn); - rs->deleteRecord(txn, id); - wunit.commit(); - } - cursor->restore(txn); - continue; - } + // Indexes must be dropped before we open the Collection otherwise we could attempt to + // open a bad index and fail. + // TODO see if MultiIndexBlock can be made to work without a Collection. + const StringData ns = cce->ns().ns(); + collection.reset(new Collection(txn, ns, cce, dbce->getRecordStore(ns), dbce)); + + indexer.reset(new MultiIndexBlock(txn, collection.get())); + Status status = indexer->init(indexSpecs); + if (!status.isOK()) { + // The WUOW will handle cleanup, so the indexer shouldn't do its own. + indexer->abortWithoutCleanup(); + return status; + } - numRecords++; - dataSize += data.size(); + wuow.commit(); + } - // Now index the record. - // TODO SERVER-14812 add a mode that drops duplicates rather than failing - WriteUnitOfWork wunit(txn); - status = indexer->insert(data.releaseToBson(), id); - if (!status.isOK()) return status; - wunit.commit(); + // Iterate all records in the collection. Delete them if they aren't valid BSON. Index them + // if they are. + + long long numRecords = 0; + long long dataSize = 0; + + RecordStore* rs = collection->getRecordStore(); + auto cursor = rs->getCursor(txn); + while (auto record = cursor->next()) { + RecordId id = record->id; + RecordData& data = record->data; + + Status status = validateBSON(data.data(), data.size()); + if (!status.isOK()) { + log() << "Invalid BSON detected at " << id << ": " << status << ". Deleting."; + cursor->savePositioned(); // 'data' is no longer valid. + { + WriteUnitOfWork wunit(txn); + rs->deleteRecord(txn, id); + wunit.commit(); + } + cursor->restore(txn); + continue; } - Status status = indexer->doneInserting(); - if (!status.isOK()) return status; + numRecords++; + dataSize += data.size(); - { - WriteUnitOfWork wunit(txn); - indexer->commit(); - rs->updateStatsAfterRepair(txn, numRecords, dataSize); - wunit.commit(); - } + // Now index the record. + // TODO SERVER-14812 add a mode that drops duplicates rather than failing + WriteUnitOfWork wunit(txn); + status = indexer->insert(data.releaseToBson(), id); + if (!status.isOK()) + return status; + wunit.commit(); + } - return Status::OK(); + Status status = indexer->doneInserting(); + if (!status.isOK()) + return status; + + { + WriteUnitOfWork wunit(txn); + indexer->commit(); + rs->updateStatsAfterRepair(txn, numRecords, dataSize); + wunit.commit(); } -} // namespace - Status repairDatabase(OperationContext* txn, - StorageEngine* engine, - const std::string& dbName, - bool preserveClonedFilesOnFailure, - bool backupOriginalFiles) { + return Status::OK(); +} +} // namespace - DisableDocumentValidation validationDisabler(txn); +Status repairDatabase(OperationContext* txn, + StorageEngine* engine, + const std::string& dbName, + bool preserveClonedFilesOnFailure, + bool backupOriginalFiles) { + DisableDocumentValidation validationDisabler(txn); - // We must hold some form of lock here - invariant(txn->lockState()->isLocked()); - invariant( dbName.find( '.' ) == string::npos ); + // We must hold some form of lock here + invariant(txn->lockState()->isLocked()); + invariant(dbName.find('.') == string::npos); - log() << "repairDatabase " << dbName << endl; + log() << "repairDatabase " << dbName << endl; - BackgroundOperation::assertNoBgOpInProgForDb(dbName); + BackgroundOperation::assertNoBgOpInProgForDb(dbName); - txn->checkForInterrupt(); + txn->checkForInterrupt(); - if (engine->isMmapV1()) { - // MMAPv1 is a layering violation so it implements its own repairDatabase. - return static_cast<MMAPV1Engine*>(engine)->repairDatabase(txn, - dbName, - preserveClonedFilesOnFailure, - backupOriginalFiles); - } + if (engine->isMmapV1()) { + // MMAPv1 is a layering violation so it implements its own repairDatabase. + return static_cast<MMAPV1Engine*>(engine) + ->repairDatabase(txn, dbName, preserveClonedFilesOnFailure, backupOriginalFiles); + } - // These are MMAPv1 specific - if ( preserveClonedFilesOnFailure ) { - return Status( ErrorCodes::BadValue, "preserveClonedFilesOnFailure not supported" ); - } - if ( backupOriginalFiles ) { - return Status( ErrorCodes::BadValue, "backupOriginalFiles not supported" ); - } + // These are MMAPv1 specific + if (preserveClonedFilesOnFailure) { + return Status(ErrorCodes::BadValue, "preserveClonedFilesOnFailure not supported"); + } + if (backupOriginalFiles) { + return Status(ErrorCodes::BadValue, "backupOriginalFiles not supported"); + } - // Close the db to invalidate all current users and caches. - dbHolder().close(txn, dbName); - // Open the db after everything finishes - class OpenDbInDestructor { - public: - OpenDbInDestructor(OperationContext* txn, const std::string& db) : - _dbName(db) - , _txn(txn) - {} - ~OpenDbInDestructor() { - dbHolder().openDb(_txn, _dbName); - } - private: - const std::string& _dbName; - OperationContext* _txn; - } dbOpener(txn, dbName); - DatabaseCatalogEntry* dbce = engine->getDatabaseCatalogEntry(txn, dbName); + // Close the db to invalidate all current users and caches. + dbHolder().close(txn, dbName); + // Open the db after everything finishes + class OpenDbInDestructor { + public: + OpenDbInDestructor(OperationContext* txn, const std::string& db) : _dbName(db), _txn(txn) {} + ~OpenDbInDestructor() { + dbHolder().openDb(_txn, _dbName); + } - std::list<std::string> colls; - dbce->getCollectionNamespaces(&colls); + private: + const std::string& _dbName; + OperationContext* _txn; + } dbOpener(txn, dbName); + DatabaseCatalogEntry* dbce = engine->getDatabaseCatalogEntry(txn, dbName); - for (std::list<std::string>::const_iterator it = colls.begin(); it != colls.end(); ++it) { - // Don't check for interrupt after starting to repair a collection otherwise we can - // leave data in an inconsistent state. Interrupting between collections is ok, however. - txn->checkForInterrupt(); + std::list<std::string> colls; + dbce->getCollectionNamespaces(&colls); - log() << "Repairing collection " << *it; + for (std::list<std::string>::const_iterator it = colls.begin(); it != colls.end(); ++it) { + // Don't check for interrupt after starting to repair a collection otherwise we can + // leave data in an inconsistent state. Interrupting between collections is ok, however. + txn->checkForInterrupt(); - Status status = engine->repairRecordStore(txn, *it); - if (!status.isOK()) return status; + log() << "Repairing collection " << *it; - status = rebuildIndexesOnCollection(txn, dbce, *it); - if (!status.isOK()) return status; + Status status = engine->repairRecordStore(txn, *it); + if (!status.isOK()) + return status; - // TODO: uncomment once SERVER-16869 - // engine->flushAllFiles(true); - } + status = rebuildIndexesOnCollection(txn, dbce, *it); + if (!status.isOK()) + return status; - return Status::OK(); + // TODO: uncomment once SERVER-16869 + // engine->flushAllFiles(true); } -} + return Status::OK(); +} +} |