diff options
Diffstat (limited to 'src')
87 files changed, 556 insertions, 525 deletions
diff --git a/src/mongo/db/SConscript b/src/mongo/db/SConscript index 9f0bc23aa28..e028a244889 100644 --- a/src/mongo/db/SConscript +++ b/src/mongo/db/SConscript @@ -151,6 +151,7 @@ env.Library( ], LIBDEPS=[ '$BUILD_DIR/mongo/db/repl/repl_coordinator_interface', + '$BUILD_DIR/mongo/db/repl/repl_settings', '$BUILD_DIR/mongo/util/processinfo', 'startup_warnings_common', ] diff --git a/src/mongo/db/auth/authorization_checks.cpp b/src/mongo/db/auth/authorization_checks.cpp index 6beb7def53a..06fa2164fbc 100644 --- a/src/mongo/db/auth/authorization_checks.cpp +++ b/src/mongo/db/auth/authorization_checks.cpp @@ -43,7 +43,8 @@ namespace { // Checks if this connection has the privileges necessary to create or modify the view 'viewNs' // to be a view on 'viewOnNs' with pipeline 'viewPipeline'. Call this function after verifying // that the user has the 'createCollection' or 'collMod' action, respectively. -Status checkAuthForCreateOrModifyView(AuthorizationSession* authzSession, +Status checkAuthForCreateOrModifyView(OperationContext* opCtx, + AuthorizationSession* authzSession, const NamespaceString& viewNs, const NamespaceString& viewOnNs, const BSONArray& viewPipeline, @@ -54,6 +55,7 @@ Status checkAuthForCreateOrModifyView(AuthorizationSession* authzSession, } auto request = aggregation_request_helper::parseFromBSON( + opCtx, viewNs, BSON("aggregate" << viewOnNs.coll() << "pipeline" << viewPipeline << "cursor" << BSONObj() << "$db" << viewOnNs.db()), @@ -198,7 +200,8 @@ Status checkAuthForKillCursors(AuthorizationSession* authSession, str::stream() << "not authorized to kill cursor on " << ns.ns()); } -Status checkAuthForCreate(AuthorizationSession* authSession, +Status checkAuthForCreate(OperationContext* opCtx, + AuthorizationSession* authSession, const CreateCommand& cmd, bool isMongos) { auto ns = cmd.getNamespace(); @@ -229,7 +232,7 @@ Status checkAuthForCreate(AuthorizationSession* authSession, pipelineArray.append(stage); } return checkAuthForCreateOrModifyView( - authSession, ns, viewOnNs, pipelineArray.arr(), isMongos); + opCtx, authSession, ns, viewOnNs, pipelineArray.arr(), isMongos); } // To create a regular collection, ActionType::createCollection or ActionType::insert are @@ -242,7 +245,8 @@ Status checkAuthForCreate(AuthorizationSession* authSession, return Status(ErrorCodes::Unauthorized, "unauthorized"); } -Status checkAuthForCollMod(AuthorizationSession* authSession, +Status checkAuthForCollMod(OperationContext* opCtx, + AuthorizationSession* authSession, const NamespaceString& ns, const BSONObj& cmdObj, bool isMongos) { @@ -264,7 +268,8 @@ Status checkAuthForCollMod(AuthorizationSession* authSession, if (hasViewOn) { NamespaceString viewOnNs(ns.db(), cmdObj["viewOn"].checkAndGetStringData()); auto viewPipeline = BSONArray(cmdObj["pipeline"].Obj()); - return checkAuthForCreateOrModifyView(authSession, ns, viewOnNs, viewPipeline, isMongos); + return checkAuthForCreateOrModifyView( + opCtx, authSession, ns, viewOnNs, viewPipeline, isMongos); } return Status::OK(); diff --git a/src/mongo/db/auth/authorization_checks.h b/src/mongo/db/auth/authorization_checks.h index 326722c6a5b..4a211fb0ee1 100644 --- a/src/mongo/db/auth/authorization_checks.h +++ b/src/mongo/db/auth/authorization_checks.h @@ -90,13 +90,15 @@ StatusWith<PrivilegeVector> getPrivilegesForAggregate(AuthorizationSession* auth // Checks if this connection has the privileges necessary to create 'ns' with the options // supplied in 'cmdObj' either directly on mongoD or via mongoS. -Status checkAuthForCreate(AuthorizationSession* authSession, +Status checkAuthForCreate(OperationContext* opCtx, + AuthorizationSession* authSession, const CreateCommand& cmd, bool isMongos); // Checks if this connection has the privileges necessary to modify 'ns' with the options // supplied in 'cmdObj' either directly on mongoD or via mongoS. -Status checkAuthForCollMod(AuthorizationSession* authSession, +Status checkAuthForCollMod(OperationContext* opCtx, + AuthorizationSession* authSession, const NamespaceString& ns, const BSONObj& cmdObj, bool isMongos); diff --git a/src/mongo/db/catalog/drop_database.cpp b/src/mongo/db/catalog/drop_database.cpp index 85e2087a321..9445deb39d5 100644 --- a/src/mongo/db/catalog/drop_database.cpp +++ b/src/mongo/db/catalog/drop_database.cpp @@ -129,7 +129,7 @@ Status _dropDatabase(OperationContext* opCtx, const std::string& dbName, bool ab uassert(ErrorCodes::IllegalOperation, "Cannot drop a database in read-only mode", - !storageGlobalParams.readOnly); + !opCtx->readOnly()); // As of SERVER-32205, dropping the admin database is prohibited. uassert(ErrorCodes::IllegalOperation, diff --git a/src/mongo/db/commands/count_cmd.cpp b/src/mongo/db/commands/count_cmd.cpp index c548bf3ef1c..35b360eb9b9 100644 --- a/src/mongo/db/commands/count_cmd.cpp +++ b/src/mongo/db/commands/count_cmd.cpp @@ -175,6 +175,7 @@ public: auto viewAggCmd = OpMsgRequest::fromDBAndBody(nss.db(), viewAggregation.getValue()).body; auto viewAggRequest = aggregation_request_helper::parseFromBSON( + opCtx, nss, viewAggCmd, verbosity, diff --git a/src/mongo/db/commands/create_command.cpp b/src/mongo/db/commands/create_command.cpp index ebe478ca160..10160cdcc62 100644 --- a/src/mongo/db/commands/create_command.cpp +++ b/src/mongo/db/commands/create_command.cpp @@ -97,7 +97,7 @@ public: void doCheckAuthorization(OperationContext* opCtx) const final { uassertStatusOK(auth::checkAuthForCreate( - AuthorizationSession::get(opCtx->getClient()), request(), false)); + opCtx, AuthorizationSession::get(opCtx->getClient()), request(), false)); } NamespaceString ns() const final { diff --git a/src/mongo/db/commands/dbcommands.cpp b/src/mongo/db/commands/dbcommands.cpp index 0e88f993f93..0d219fe856e 100644 --- a/src/mongo/db/commands/dbcommands.cpp +++ b/src/mongo/db/commands/dbcommands.cpp @@ -542,7 +542,8 @@ public: const std::string& dbname, const BSONObj& cmdObj) const { const NamespaceString nss(parseNs(dbname, cmdObj)); - return auth::checkAuthForCollMod(AuthorizationSession::get(client), nss, cmdObj, false); + return auth::checkAuthForCollMod( + client->getOperationContext(), AuthorizationSession::get(client), nss, cmdObj, false); } bool runWithRequestParser(OperationContext* opCtx, diff --git a/src/mongo/db/commands/distinct.cpp b/src/mongo/db/commands/distinct.cpp index 57585a88401..da5b72c80e6 100644 --- a/src/mongo/db/commands/distinct.cpp +++ b/src/mongo/db/commands/distinct.cpp @@ -164,6 +164,7 @@ public: auto viewAggCmd = OpMsgRequest::fromDBAndBody(nss.db(), viewAggregation.getValue()).body; auto viewAggRequest = aggregation_request_helper::parseFromBSON( + opCtx, nss, viewAggCmd, verbosity, diff --git a/src/mongo/db/commands/feature_compatibility_version.cpp b/src/mongo/db/commands/feature_compatibility_version.cpp index b9af74e8847..7062c8ec1e4 100644 --- a/src/mongo/db/commands/feature_compatibility_version.cpp +++ b/src/mongo/db/commands/feature_compatibility_version.cpp @@ -200,10 +200,6 @@ Lock::ResourceMutex fcvLock("featureCompatibilityVersionLock"); Timestamp lastFCVUpdateTimestamp; SimpleMutex lastFCVUpdateTimestampMutex; -bool isWriteableStorageEngine() { - return !storageGlobalParams.readOnly && (storageGlobalParams.engine != "devnull"); -} - /** * Build update command for featureCompatibilityVersion document updates. */ @@ -498,8 +494,10 @@ void FeatureCompatibilityVersion::fassertInitializedAfterStartup(OperationContex // If we are part of a replica set and are started up with no data files, we do not set the // featureCompatibilityVersion until a primary is chosen. For this case, we expect the in-memory - // featureCompatibilityVersion parameter to still be uninitialized until after startup. - if (isWriteableStorageEngine() && (!usingReplication || nonLocalDatabases)) { + // featureCompatibilityVersion parameter to still be uninitialized until after startup. In + // standalone mode, FCV is initialized during startup, even in read-only mode. + bool isWriteableStorageEngine = storageGlobalParams.engine != "devnull"; + if (isWriteableStorageEngine && (!usingReplication || nonLocalDatabases)) { invariant(serverGlobalParams.featureCompatibility.isVersionInitialized()); } } diff --git a/src/mongo/db/commands/find_cmd.cpp b/src/mongo/db/commands/find_cmd.cpp index fc81cb6904d..3aabc3ddf6b 100644 --- a/src/mongo/db/commands/find_cmd.cpp +++ b/src/mongo/db/commands/find_cmd.cpp @@ -345,6 +345,7 @@ public: // Create the agg request equivalent of the find operation, with the explain // verbosity included. auto aggRequest = aggregation_request_helper::parseFromBSON( + opCtx, nss, viewAggCmd, verbosity, diff --git a/src/mongo/db/commands/list_collections.cpp b/src/mongo/db/commands/list_collections.cpp index bc144913d4f..093ff11a439 100644 --- a/src/mongo/db/commands/list_collections.cpp +++ b/src/mongo/db/commands/list_collections.cpp @@ -239,7 +239,7 @@ BSONObj buildCollectionBson(OperationContext* opCtx, b.append("options", options.toBSON(false)); BSONObjBuilder infoBuilder; - infoBuilder.append("readOnly", storageGlobalParams.readOnly); + infoBuilder.append("readOnly", opCtx->readOnly()); if (options.uuid) infoBuilder.appendElements(options.uuid->toBSON()); b.append("info", infoBuilder.obj()); diff --git a/src/mongo/db/commands/pipeline_command.cpp b/src/mongo/db/commands/pipeline_command.cpp index eb5f74d4387..1d2e0f25059 100644 --- a/src/mongo/db/commands/pipeline_command.cpp +++ b/src/mongo/db/commands/pipeline_command.cpp @@ -74,6 +74,7 @@ public: const OpMsgRequest& opMsgRequest, boost::optional<ExplainOptions::Verbosity> explainVerbosity) override { const auto aggregationRequest = aggregation_request_helper::parseFromBSON( + opCtx, opMsgRequest.getDatabase().toString(), opMsgRequest.body, explainVerbosity, diff --git a/src/mongo/db/commands/validate.cpp b/src/mongo/db/commands/validate.cpp index c6724076b9f..d17529cdaec 100644 --- a/src/mongo/db/commands/validate.cpp +++ b/src/mongo/db/commands/validate.cpp @@ -148,7 +148,7 @@ public: } const bool repair = cmdObj["repair"].trueValue(); - if (storageGlobalParams.readOnly && repair) { + if (opCtx->readOnly() && repair) { uasserted(ErrorCodes::InvalidOptions, str::stream() << "Running the validate command with { repair: true } in" << " read-only mode is not supported."); @@ -226,7 +226,7 @@ public: }(); auto repairMode = [&] { - if (storageGlobalParams.readOnly) { + if (opCtx->readOnly()) { // On read-only mode we can't make any adjustments. return CollectionValidation::RepairMode::kNone; } diff --git a/src/mongo/db/error_labels.cpp b/src/mongo/db/error_labels.cpp index b66d0df2d74..23fa5994b46 100644 --- a/src/mongo/db/error_labels.cpp +++ b/src/mongo/db/error_labels.cpp @@ -117,10 +117,10 @@ bool ErrorLabelBuilder::isResumableChangeStreamError() const { bool apiStrict = APIParameters::get(_opCtx).getAPIStrict().value_or(false); // Do enough parsing to confirm that this is a well-formed pipeline with a $changeStream. - const auto swLitePipe = [&nss, &cmdObj, apiStrict]() -> StatusWith<LiteParsedPipeline> { + const auto swLitePipe = [this, &nss, &cmdObj, apiStrict]() -> StatusWith<LiteParsedPipeline> { try { - auto aggRequest = - aggregation_request_helper::parseFromBSON(nss, cmdObj, boost::none, apiStrict); + auto aggRequest = aggregation_request_helper::parseFromBSON( + _opCtx, nss, cmdObj, boost::none, apiStrict); return LiteParsedPipeline(aggRequest); } catch (const DBException& ex) { return ex.toStatus(); diff --git a/src/mongo/db/mongod_main.cpp b/src/mongo/db/mongod_main.cpp index bce1bc1815c..94aaa958448 100644 --- a/src/mongo/db/mongod_main.cpp +++ b/src/mongo/db/mongod_main.cpp @@ -389,6 +389,13 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { auto runner = makePeriodicRunner(serviceContext); serviceContext->setPeriodicRunner(std::move(runner)); + // When starting the server with --queryableBackupMode or --recoverFromOplogAsStandalone, we are + // in read-only mode and don't allow user-originating operations to perform writes + if (storageGlobalParams.queryableBackupMode || + repl::ReplSettings::shouldRecoverFromOplogAsStandalone()) { + serviceContext->disallowUserWrites(); + } + #ifdef MONGO_CONFIG_SSL OCSPManager::start(serviceContext); CertificateExpirationMonitor::get()->start(serviceContext); @@ -650,7 +657,7 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { readWriteConcernDefaultsMongodStartupChecks(startupOpCtx.get()); // Perform replication recovery for queryable backup mode if needed. - if (storageGlobalParams.readOnly) { + if (storageGlobalParams.queryableBackupMode) { uassert(ErrorCodes::BadValue, str::stream() << "Cannot specify both queryableBackupMode and " << "recoverFromOplogAsStandalone at the same time", @@ -673,7 +680,7 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { replCoord->startup(startupOpCtx.get(), lastShutdownState); } - if (!storageGlobalParams.readOnly) { + if (!storageGlobalParams.queryableBackupMode) { if (storageEngine->supportsCappedCollections()) { logStartup(startupOpCtx.get()); diff --git a/src/mongo/db/mongod_options.cpp b/src/mongo/db/mongod_options.cpp index d474782085c..6e3772f4083 100644 --- a/src/mongo/db/mongod_options.cpp +++ b/src/mongo/db/mongod_options.cpp @@ -438,7 +438,7 @@ Status storeMongodOptions(const moe::Environment& params) { if (params.count("storage.queryableBackupMode") && params["storage.queryableBackupMode"].as<bool>()) { - storageGlobalParams.readOnly = true; + storageGlobalParams.queryableBackupMode = true; } if (params.count("storage.groupCollections")) { diff --git a/src/mongo/db/operation_context.h b/src/mongo/db/operation_context.h index be234281d8b..ec3d01f4133 100644 --- a/src/mongo/db/operation_context.h +++ b/src/mongo/db/operation_context.h @@ -588,6 +588,19 @@ public: */ void restoreMaxTimeMS(); + /** + * Returns whether this operation must run in read-only mode. + * + * If the read-only flag is set on the ServiceContext then: + * - Internal operations are allowed to perform writes. + * - User originating operations are not allowed to perform writes. + */ + bool readOnly() const { + if (!getClient()->isFromUserConnection()) + return false; + return !getServiceContext()->userWritesAllowed(); + } + private: StatusWith<stdx::cv_status> waitForConditionOrInterruptNoAssertUntil( stdx::condition_variable& cv, BasicLockableAdapter m, Date_t deadline) noexcept override; diff --git a/src/mongo/db/pipeline/aggregation_request_helper.cpp b/src/mongo/db/pipeline/aggregation_request_helper.cpp index 5bfa4b7c2a3..04fcc4023e3 100644 --- a/src/mongo/db/pipeline/aggregation_request_helper.cpp +++ b/src/mongo/db/pipeline/aggregation_request_helper.cpp @@ -49,15 +49,17 @@ namespace aggregation_request_helper { /** * Validate the aggregate command object. */ -void validate(const BSONObj& cmdObj, +void validate(OperationContext* opCtx, + const BSONObj& cmdObj, const NamespaceString& nss, boost::optional<ExplainOptions::Verbosity> explainVerbosity); -AggregateCommandRequest parseFromBSON(const std::string& dbName, +AggregateCommandRequest parseFromBSON(OperationContext* opCtx, + const std::string& dbName, const BSONObj& cmdObj, boost::optional<ExplainOptions::Verbosity> explainVerbosity, bool apiStrict) { - return parseFromBSON(parseNs(dbName, cmdObj), cmdObj, explainVerbosity, apiStrict); + return parseFromBSON(opCtx, parseNs(dbName, cmdObj), cmdObj, explainVerbosity, apiStrict); } StatusWith<AggregateCommandRequest> parseFromBSONForTests( @@ -66,7 +68,7 @@ StatusWith<AggregateCommandRequest> parseFromBSONForTests( boost::optional<ExplainOptions::Verbosity> explainVerbosity, bool apiStrict) { try { - return parseFromBSON(nss, cmdObj, explainVerbosity, apiStrict); + return parseFromBSON(/*opCtx=*/nullptr, nss, cmdObj, explainVerbosity, apiStrict); } catch (const AssertionException&) { return exceptionToStatus(); } @@ -78,13 +80,14 @@ StatusWith<AggregateCommandRequest> parseFromBSONForTests( boost::optional<ExplainOptions::Verbosity> explainVerbosity, bool apiStrict) { try { - return parseFromBSON(dbName, cmdObj, explainVerbosity, apiStrict); + return parseFromBSON(/*opCtx=*/nullptr, dbName, cmdObj, explainVerbosity, apiStrict); } catch (const AssertionException&) { return exceptionToStatus(); } } -AggregateCommandRequest parseFromBSON(NamespaceString nss, +AggregateCommandRequest parseFromBSON(OperationContext* opCtx, + NamespaceString nss, const BSONObj& cmdObj, boost::optional<ExplainOptions::Verbosity> explainVerbosity, bool apiStrict) { @@ -111,7 +114,7 @@ AggregateCommandRequest parseFromBSON(NamespaceString nss, request.setExplain(explainVerbosity); } - validate(cmdObj, nss, explainVerbosity); + validate(opCtx, cmdObj, nss, explainVerbosity); return request; } @@ -149,7 +152,8 @@ Document serializeToCommandDoc(const AggregateCommandRequest& request) { return Document(request.toBSON(BSONObj()).getOwned()); } -void validate(const BSONObj& cmdObj, +void validate(OperationContext* opCtx, + const BSONObj& cmdObj, const NamespaceString& nss, boost::optional<ExplainOptions::Verbosity> explainVerbosity) { bool hasAllowDiskUseElem = cmdObj.hasField(AggregateCommandRequest::kAllowDiskUseFieldName); @@ -177,10 +181,12 @@ void validate(const BSONObj& cmdObj, << "' without '" << AggregateCommandRequest::kFromMongosFieldName << "'", (!hasNeedsMergeElem || hasFromMongosElem)); - uassert(ErrorCodes::IllegalOperation, - str::stream() << "The '" << AggregateCommandRequest::kAllowDiskUseFieldName - << "' option is not permitted in read-only mode.", - (!hasAllowDiskUseElem || !storageGlobalParams.readOnly)); + if (opCtx) { + uassert(ErrorCodes::IllegalOperation, + str::stream() << "The '" << AggregateCommandRequest::kAllowDiskUseFieldName + << "' option is not permitted in read-only mode.", + (!hasAllowDiskUseElem || !opCtx->readOnly())); + } auto requestReshardingResumeTokenElem = cmdObj[AggregateCommandRequest::kRequestReshardingResumeTokenFieldName]; diff --git a/src/mongo/db/pipeline/aggregation_request_helper.h b/src/mongo/db/pipeline/aggregation_request_helper.h index bb447712f7f..a0af24777f5 100644 --- a/src/mongo/db/pipeline/aggregation_request_helper.h +++ b/src/mongo/db/pipeline/aggregation_request_helper.h @@ -68,7 +68,8 @@ static constexpr long long kDefaultBatchSize = 101; * then 'explainVerbosity' contains this information. In this case, 'cmdObj' may not itself * contain the explain specifier. Otherwise, 'explainVerbosity' should be boost::none. */ -AggregateCommandRequest parseFromBSON(NamespaceString nss, +AggregateCommandRequest parseFromBSON(OperationContext* opCtx, + NamespaceString nss, const BSONObj& cmdObj, boost::optional<ExplainOptions::Verbosity> explainVerbosity, bool apiStrict); @@ -83,7 +84,8 @@ StatusWith<AggregateCommandRequest> parseFromBSONForTests( * Convenience overload which constructs the request's NamespaceString from the given database * name and command object. */ -AggregateCommandRequest parseFromBSON(const std::string& dbName, +AggregateCommandRequest parseFromBSON(OperationContext* opCtx, + const std::string& dbName, const BSONObj& cmdObj, boost::optional<ExplainOptions::Verbosity> explainVerbosity, bool apiStrict); diff --git a/src/mongo/db/pipeline/document_source_change_stream_handle_topology_change.cpp b/src/mongo/db/pipeline/document_source_change_stream_handle_topology_change.cpp index 0f32b20aa47..56cd64d653c 100644 --- a/src/mongo/db/pipeline/document_source_change_stream_handle_topology_change.cpp +++ b/src/mongo/db/pipeline/document_source_change_stream_handle_topology_change.cpp @@ -214,7 +214,7 @@ BSONObj DocumentSourceChangeStreamHandleTopologyChange::createUpdatedCommandForN // Create the 'AggregateCommandRequest' object which will help in creating the parsed pipeline. auto aggCmdRequest = aggregation_request_helper::parseFromBSON( - pExpCtx->ns, shardCommand, boost::none, apiStrict); + opCtx, pExpCtx->ns, shardCommand, boost::none, apiStrict); // Parse and optimize the pipeline. auto pipeline = Pipeline::parse(aggCmdRequest.getPipeline(), pExpCtx); diff --git a/src/mongo/db/pipeline/document_source_group.cpp b/src/mongo/db/pipeline/document_source_group.cpp index 80a2dc60488..9f7197c43e2 100644 --- a/src/mongo/db/pipeline/document_source_group.cpp +++ b/src/mongo/db/pipeline/document_source_group.cpp @@ -610,7 +610,7 @@ MONGO_COMPILER_NOINLINE DocumentSource::GetNextResult DocumentSourceGroup::initi } } - if (kDebugBuild && !storageGlobalParams.readOnly) { + if (kDebugBuild && !pExpCtx->opCtx->readOnly()) { // In debug mode, spill every time we have a duplicate id to stress merge logic. if (!inserted && // is a dup !pExpCtx->inMongos && // can't spill to disk in mongos diff --git a/src/mongo/db/pipeline/pipeline.cpp b/src/mongo/db/pipeline/pipeline.cpp index d92a6fa3b29..b85a1e379ee 100644 --- a/src/mongo/db/pipeline/pipeline.cpp +++ b/src/mongo/db/pipeline/pipeline.cpp @@ -605,7 +605,7 @@ Status Pipeline::_pipelineCanRunOnMongoS() const { const bool mustWriteToDisk = (constraints.diskRequirement == DiskUseRequirement::kWritesPersistentData); const bool mayWriteTmpDataAndDiskUseIsAllowed = - (pCtx->allowDiskUse && !storageGlobalParams.readOnly && + (pCtx->allowDiskUse && !pCtx->opCtx->readOnly() && constraints.diskRequirement == DiskUseRequirement::kWritesTmpData); const bool needsDisk = (mustWriteToDisk || mayWriteTmpDataAndDiskUseIsAllowed); diff --git a/src/mongo/db/repl/replication_coordinator_external_state_impl.cpp b/src/mongo/db/repl/replication_coordinator_external_state_impl.cpp index b3faa85859f..3bbe0000e1a 100644 --- a/src/mongo/db/repl/replication_coordinator_external_state_impl.cpp +++ b/src/mongo/db/repl/replication_coordinator_external_state_impl.cpp @@ -399,8 +399,7 @@ void ReplicationCoordinatorExternalStateImpl::shutdown(OperationContext* opCtx) // primary shutdown. Stepdown is similarly safe from holes and halts updates to and clears // the truncate point. The other replication states do need truncation if the truncate point // is set: e.g. interruption mid batch application can leave oplog holes. - if (!storageGlobalParams.readOnly && - _replicationProcess->getConsistencyMarkers() + if (_replicationProcess->getConsistencyMarkers() ->isOplogTruncateAfterPointBeingUsedForPrimary()) { _stopAsyncUpdatesOfAndClearOplogTruncateAfterPoint(); } diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp index a1296d98522..3e6dca73357 100644 --- a/src/mongo/db/repl/replication_coordinator_impl.cpp +++ b/src/mongo/db/repl/replication_coordinator_impl.cpp @@ -898,7 +898,7 @@ void ReplicationCoordinatorImpl::startup(OperationContext* opCtx, _replicationProcess->getReplicationRecovery()->recoverFromOplogAsStandalone(opCtx); } - if (storageGlobalParams.readOnly && !recoverToOplogTimestamp.empty()) { + if (storageGlobalParams.queryableBackupMode && !recoverToOplogTimestamp.empty()) { BSONObj recoverToTimestampObj = fromjson(recoverToOplogTimestamp); uassert(ErrorCodes::BadValue, str::stream() << "'recoverToOplogTimestamp' needs to have a 'timestamp' field", @@ -917,9 +917,6 @@ void ReplicationCoordinatorImpl::startup(OperationContext* opCtx, cfg.isOK()); // Need to perform replication recovery up to and including the given timestamp. - // Temporarily turn off read-only mode for this procedure as we'll have to do writes. - storageGlobalParams.readOnly = false; - ON_BLOCK_EXIT([&] { storageGlobalParams.readOnly = true; }); _replicationProcess->getReplicationRecovery()->recoverFromOplogUpTo(opCtx, recoverToTimestamp); } diff --git a/src/mongo/db/repl/replication_info.cpp b/src/mongo/db/repl/replication_info.cpp index 2172307d791..110831f1663 100644 --- a/src/mongo/db/repl/replication_info.cpp +++ b/src/mongo/db/repl/replication_info.cpp @@ -493,7 +493,7 @@ public: wireSpec->incomingExternalClient.maxWireVersion); } - result.append(HelloCommandReply::kReadOnlyFieldName, storageGlobalParams.readOnly); + result.append(HelloCommandReply::kReadOnlyFieldName, opCtx->readOnly()); if (auto param = ServerParameterSet::getNodeParameterSet()->getIfExists( kAutomationServiceDescriptorFieldName)) { diff --git a/src/mongo/db/repl/replication_recovery.cpp b/src/mongo/db/repl/replication_recovery.cpp index 5cd96227393..ca20262c266 100644 --- a/src/mongo/db/repl/replication_recovery.cpp +++ b/src/mongo/db/repl/replication_recovery.cpp @@ -364,16 +364,6 @@ void ReplicationRecoveryImpl::recoverFromOplogAsStandalone(OperationContext* opC if (!_duringInitialSync) { // Initial sync will reconstruct prepared transactions when it is completely done. reconstructPreparedTransactions(opCtx, OplogApplication::Mode::kRecovering); - - // Two-phase index builds are built in the background, which may still be in-progress after - // recovering from the oplog. To prevent crashing the server, skip enabling read-only mode. - if (IndexBuildsCoordinator::get(opCtx)->noIndexBuildInProgress()) { - LOGV2_WARNING(21558, - "Setting mongod to readOnly mode as a result of specifying " - "'recoverFromOplogAsStandalone'"); - - storageGlobalParams.readOnly = true; - } } } diff --git a/src/mongo/db/repl/replication_recovery_test.cpp b/src/mongo/db/repl/replication_recovery_test.cpp index ee937d39648..fa01fb5ebb0 100644 --- a/src/mongo/db/repl/replication_recovery_test.cpp +++ b/src/mongo/db/repl/replication_recovery_test.cpp @@ -222,7 +222,6 @@ private: serverGlobalParams.enableMajorityReadConcern = _stashedEnableMajorityReadConcern; ServiceContextMongoDTest::tearDown(); - storageGlobalParams.readOnly = false; gTakeUnstableCheckpointOnShutdown = false; } @@ -1478,10 +1477,6 @@ TEST_F(ReplicationRecoveryTest, RecoverFromOplogAsStandaloneRecoversOplog) { recovery.recoverFromOplogAsStandalone(opCtx); _assertDocsInTestCollection(opCtx, {5}); - - // Test the node is readOnly. - ASSERT_THROWS(getStorageInterface()->insertDocument(opCtx, testNs, {_makeInsertDocument(2)}, 1), - AssertionException); } TEST_F(ReplicationRecoveryTest, diff --git a/src/mongo/db/repl/storage_timestamp_test.cpp b/src/mongo/db/repl/storage_timestamp_test.cpp index fead4dab5ee..e6b65302e25 100644 --- a/src/mongo/db/repl/storage_timestamp_test.cpp +++ b/src/mongo/db/repl/storage_timestamp_test.cpp @@ -320,7 +320,7 @@ public: void dumpOplog() { OneOffRead oor(_opCtx, Timestamp::min()); - _opCtx->recoveryUnit()->beginUnitOfWork(_opCtx); + _opCtx->recoveryUnit()->beginUnitOfWork(_opCtx->readOnly()); LOGV2(8423335, "Dumping oplog collection"); AutoGetCollectionForRead oplogRaii(_opCtx, NamespaceString::kRsOplogNamespace); const CollectionPtr& oplogColl = oplogRaii.getCollection(); diff --git a/src/mongo/db/repl/timestamp_block.cpp b/src/mongo/db/repl/timestamp_block.cpp index 41992c8ed58..211d61d3f9c 100644 --- a/src/mongo/db/repl/timestamp_block.cpp +++ b/src/mongo/db/repl/timestamp_block.cpp @@ -40,7 +40,7 @@ namespace mongo { TimestampBlock::TimestampBlock(OperationContext* opCtx, Timestamp ts) : _opCtx(opCtx), _ts(ts) { uassert(ErrorCodes::IllegalOperation, "Cannot timestamp a write operation in read-only mode", - !storageGlobalParams.readOnly); + !_opCtx->readOnly()); if (!_ts.isNull()) { _opCtx->recoveryUnit()->setCommitTimestamp(_ts); } diff --git a/src/mongo/db/s/cluster_pipeline_cmd_d.cpp b/src/mongo/db/s/cluster_pipeline_cmd_d.cpp index cd78fa52e18..8526a411f67 100644 --- a/src/mongo/db/s/cluster_pipeline_cmd_d.cpp +++ b/src/mongo/db/s/cluster_pipeline_cmd_d.cpp @@ -62,13 +62,15 @@ struct ClusterPipelineCommandD { } static AggregateCommandRequest parseAggregationRequest( + OperationContext* opCtx, const OpMsgRequest& opMsgRequest, boost::optional<ExplainOptions::Verbosity> explainVerbosity, bool apiStrict) { // Replace clusterAggregate in the request body because the parser doesn't recognize it. auto modifiedRequestBody = opMsgRequest.body.replaceFieldNames(BSON(AggregateCommandRequest::kCommandName << 1)); - return aggregation_request_helper::parseFromBSON(opMsgRequest.getDatabase().toString(), + return aggregation_request_helper::parseFromBSON(opCtx, + opMsgRequest.getDatabase().toString(), modifiedRequestBody, explainVerbosity, apiStrict); diff --git a/src/mongo/db/s/flush_database_cache_updates_command.cpp b/src/mongo/db/s/flush_database_cache_updates_command.cpp index de95a293333..e56515581f7 100644 --- a/src/mongo/db/s/flush_database_cache_updates_command.cpp +++ b/src/mongo/db/s/flush_database_cache_updates_command.cpp @@ -115,7 +115,7 @@ public: uassert(ErrorCodes::IllegalOperation, "Can't call _flushDatabaseCacheUpdates if in read-only mode", - !storageGlobalParams.readOnly); + !opCtx->readOnly()); boost::optional<SharedSemiFuture<void>> criticalSectionSignal; diff --git a/src/mongo/db/s/flush_resharding_state_change_command.cpp b/src/mongo/db/s/flush_resharding_state_change_command.cpp index c1049660307..9fb897dc90b 100644 --- a/src/mongo/db/s/flush_resharding_state_change_command.cpp +++ b/src/mongo/db/s/flush_resharding_state_change_command.cpp @@ -108,7 +108,7 @@ public: uassert(ErrorCodes::IllegalOperation, "Can't call _flushReshardingStateChange if in read-only mode", - !storageGlobalParams.readOnly); + !opCtx->readOnly()); ExecutorFuture<void>(Grid::get(opCtx)->getExecutorPool()->getArbitraryExecutor()) .then([svcCtx = opCtx->getServiceContext(), nss = ns()] { diff --git a/src/mongo/db/s/flush_routing_table_cache_updates_command.cpp b/src/mongo/db/s/flush_routing_table_cache_updates_command.cpp index 943b78ece6d..901a8834517 100644 --- a/src/mongo/db/s/flush_routing_table_cache_updates_command.cpp +++ b/src/mongo/db/s/flush_routing_table_cache_updates_command.cpp @@ -110,7 +110,7 @@ public: uassert(ErrorCodes::IllegalOperation, str::stream() << "Can't call " << Derived::Request::kCommandName << " if in read-only mode", - !storageGlobalParams.readOnly); + !opCtx->readOnly()); boost::optional<SharedSemiFuture<void>> criticalSectionSignal; diff --git a/src/mongo/db/s/periodic_sharded_index_consistency_checker.cpp b/src/mongo/db/s/periodic_sharded_index_consistency_checker.cpp index 16df6e4e20c..6145ddc0079 100644 --- a/src/mongo/db/s/periodic_sharded_index_consistency_checker.cpp +++ b/src/mongo/db/s/periodic_sharded_index_consistency_checker.cpp @@ -138,7 +138,7 @@ void PeriodicShardedIndexConsistencyChecker::_launchShardedIndexConsistencyCheck } auto request = aggregation_request_helper::parseFromBSON( - nss, aggRequestBSON, boost::none, false); + opCtx, nss, aggRequestBSON, boost::none, false); auto catalogCache = Grid::get(opCtx)->catalogCache(); shardVersionRetry( diff --git a/src/mongo/db/s/sharding_initialization_mongod.cpp b/src/mongo/db/s/sharding_initialization_mongod.cpp index 9d57fd74933..6b0cba8ed9b 100644 --- a/src/mongo/db/s/sharding_initialization_mongod.cpp +++ b/src/mongo/db/s/sharding_initialization_mongod.cpp @@ -289,9 +289,9 @@ void ShardingInitializationMongoD::shutDown(OperationContext* opCtx) { bool ShardingInitializationMongoD::initializeShardingAwarenessIfNeeded(OperationContext* opCtx) { invariant(!opCtx->lockState()->isLocked()); - // In sharded readOnly mode, we ignore the shardIdentity document on disk and instead *require* - // a shardIdentity document to be passed through --overrideShardIdentity - if (storageGlobalParams.readOnly) { + // In sharded queryableBackupMode mode, we ignore the shardIdentity document on disk and instead + // *require* a shardIdentity document to be passed through --overrideShardIdentity + if (storageGlobalParams.queryableBackupMode) { if (serverGlobalParams.clusterRole == ClusterRole::ShardServer) { uassert(ErrorCodes::InvalidOptions, "If started with --shardsvr in queryableBackupMode, a shardIdentity document " @@ -516,7 +516,7 @@ void initializeGlobalShardingStateForMongoD(OperationContext* opCtx, auto const service = opCtx->getServiceContext(); if (serverGlobalParams.clusterRole == ClusterRole::ShardServer) { - if (storageGlobalParams.readOnly) { + if (storageGlobalParams.queryableBackupMode) { CatalogCacheLoader::set(service, std::make_unique<ReadOnlyCatalogCacheLoader>()); } else { CatalogCacheLoader::set(service, diff --git a/src/mongo/db/s/sharding_initialization_mongod_test.cpp b/src/mongo/db/s/sharding_initialization_mongod_test.cpp index ab2bffb2ff7..da9d238f92a 100644 --- a/src/mongo/db/s/sharding_initialization_mongod_test.cpp +++ b/src/mongo/db/s/sharding_initialization_mongod_test.cpp @@ -94,7 +94,7 @@ protected: _dbDirectClient.reset(); // Restore the defaults before calling tearDown - storageGlobalParams.readOnly = false; + storageGlobalParams.queryableBackupMode = false; serverGlobalParams.overrideShardIdentity = BSONObj(); CatalogCacheLoader::clearForTests(getServiceContext()); @@ -271,14 +271,15 @@ TEST_F(ShardingInitializationMongoDTest, InitializeAgainWithMatchingReplSetNameS } // The tests below check for different combinations of the compatible startup parameters for -// --shardsvr, --overrideShardIdentity, and queryableBackup (readOnly) mode +// --shardsvr, --overrideShardIdentity, and queryableBackup mode /** - * readOnly and --shardsvr + * queryableBackupMode and --shardsvr */ -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededReadOnlyAndShardServerAndNoOverrideShardIdentity) { - storageGlobalParams.readOnly = true; +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededQueryableBackupModeAndShardServerAndNoOverrideShardIdentity) { + storageGlobalParams.queryableBackupMode = true; ASSERT_THROWS_CODE( shardingInitialization()->initializeShardingAwarenessIfNeeded(operationContext()), @@ -286,9 +287,10 @@ TEST_F(ShardingInitializationMongoDTest, ErrorCodes::InvalidOptions); } -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededReadOnlyAndShardServerAndInvalidOverrideShardIdentity) { - storageGlobalParams.readOnly = true; +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededQueryableBackupModeAndShardServerAndInvalidOverrideShardIdentity) { + storageGlobalParams.queryableBackupMode = true; serverGlobalParams.overrideShardIdentity = BSON("_id" << "shardIdentity" << ShardIdentity::kShardNameFieldName << kShardName @@ -301,9 +303,10 @@ TEST_F(ShardingInitializationMongoDTest, ErrorCodes::UnsupportedFormat); } -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededReadOnlyAndShardServerAndValidOverrideShardIdentity) { - storageGlobalParams.readOnly = true; +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededQueryableBackupModeAndShardServerAndValidOverrideShardIdentity) { + storageGlobalParams.queryableBackupMode = true; serverGlobalParams.clusterRole = ClusterRole::ShardServer; serverGlobalParams.overrideShardIdentity = [] { ShardIdentityType shardIdentity; @@ -319,11 +322,12 @@ TEST_F(ShardingInitializationMongoDTest, } /** - * readOnly and not --shardsvr + * queryableBackupMode and not --shardsvr */ -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededReadOnlyAndNotShardServerAndNoOverrideShardIdentity) { - storageGlobalParams.readOnly = true; +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededQueryableBackupModeAndNotShardServerAndNoOverrideShardIdentity) { + storageGlobalParams.queryableBackupMode = true; serverGlobalParams.clusterRole = ClusterRole::None; ASSERT(!shardingInitialization()->initializeShardingAwarenessIfNeeded(operationContext())); @@ -331,8 +335,8 @@ TEST_F(ShardingInitializationMongoDTest, TEST_F( ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededReadOnlyAndNotShardServerAndInvalidOverrideShardIdentity) { - storageGlobalParams.readOnly = true; + InitializeShardingAwarenessIfNeededQueryableBackupModeAndNotShardServerAndInvalidOverrideShardIdentity) { + storageGlobalParams.queryableBackupMode = true; serverGlobalParams.clusterRole = ClusterRole::None; serverGlobalParams.overrideShardIdentity = BSON("_id" << "shardIdentity" @@ -345,9 +349,10 @@ TEST_F( ErrorCodes::InvalidOptions); } -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededReadOnlyAndNotShardServerAndValidOverrideShardIdentity) { - storageGlobalParams.readOnly = true; +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededQueryableBackupModeAndNotShardServerAndValidOverrideShardIdentity) { + storageGlobalParams.queryableBackupMode = true; serverGlobalParams.clusterRole = ClusterRole::None; serverGlobalParams.overrideShardIdentity = [] { ShardIdentityType shardIdentity; @@ -366,10 +371,10 @@ TEST_F(ShardingInitializationMongoDTest, } /** - * not readOnly and --overrideShardIdentity + * not queryableBackupMode and --overrideShardIdentity */ TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededNotReadOnlyAndInvalidOverrideShardIdentity) { + InitializeShardingAwarenessIfNeededNotQueryableBackupModeAndInvalidOverrideShardIdentity) { serverGlobalParams.clusterRole = ClusterRole::ShardServer; serverGlobalParams.overrideShardIdentity = BSON("_id" << "shardIdentity" @@ -390,7 +395,7 @@ TEST_F(ShardingInitializationMongoDTest, } TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededNotReadOnlyAndValidOverrideShardIdentity) { + InitializeShardingAwarenessIfNeededNotQueryableBackupModeAndValidOverrideShardIdentity) { serverGlobalParams.clusterRole = ClusterRole::ShardServer; serverGlobalParams.overrideShardIdentity = [] { ShardIdentityType shardIdentity; @@ -416,15 +421,16 @@ TEST_F(ShardingInitializationMongoDTest, } /** - * not readOnly and --shardsvr + * not queryableBackupMode and --shardsvr */ TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededNotReadOnlyAndShardServerAndNoShardIdentity) { + InitializeShardingAwarenessIfNeededNotQueryableBackupModeAndShardServerAndNoShardIdentity) { ASSERT(!shardingInitialization()->initializeShardingAwarenessIfNeeded(operationContext())); } -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededNotReadOnlyAndShardServerAndInvalidShardIdentity) { +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededNotQueryableBackupModeAndShardServerAndInvalidShardIdentity) { // Insert the shardIdentity doc to disk while pretending that we are in "standalone" mode, // otherwise OpObserver for inserts will prevent the insert from occurring because the // shardIdentity doc is invalid @@ -448,8 +454,9 @@ TEST_F(ShardingInitializationMongoDTest, ErrorCodes::UnsupportedFormat); } -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededNotReadOnlyAndShardServerAndValidShardIdentity) { +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededNotQueryableBackupModeAndShardServerAndValidShardIdentity) { // Insert the shardIdentity doc to disk while pretending that we are in "standalone" mode, // otherwise OpObserver for inserts will prevent the insert from occurring because the // shardIdentity doc is invalid @@ -474,17 +481,19 @@ TEST_F(ShardingInitializationMongoDTest, } /** - * not readOnly and not --shardsvr + * not queryableBackupMode and not --shardsvr */ -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededNotReadOnlyAndNotShardServerAndNoShardIdentity) { +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededNotQueryableBackupModeAndNotShardServerAndNoShardIdentity) { ScopedSetStandaloneMode standalone(getServiceContext()); ASSERT(!shardingInitialization()->initializeShardingAwarenessIfNeeded(operationContext())); } -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededNotReadOnlyAndNotShardServerAndInvalidShardIdentity) { +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededNotQueryableBackupModeAndNotShardServerAndInvalidShardIdentity) { ScopedSetStandaloneMode standalone(getServiceContext()); _dbDirectClient->insert(NamespaceString::kServerConfigurationNamespace.toString(), @@ -498,8 +507,9 @@ TEST_F(ShardingInitializationMongoDTest, ASSERT(!shardingInitialization()->initializeShardingAwarenessIfNeeded(operationContext())); } -TEST_F(ShardingInitializationMongoDTest, - InitializeShardingAwarenessIfNeededNotReadOnlyAndNotShardServerAndValidShardIdentity) { +TEST_F( + ShardingInitializationMongoDTest, + InitializeShardingAwarenessIfNeededNotQueryableBackupModeAndNotShardServerAndValidShardIdentity) { ScopedSetStandaloneMode standalone(getServiceContext()); BSONObj validShardIdentity = [&] { diff --git a/src/mongo/db/s/shardsvr_collmod_command.cpp b/src/mongo/db/s/shardsvr_collmod_command.cpp index 97f47538531..f1736ec156c 100644 --- a/src/mongo/db/s/shardsvr_collmod_command.cpp +++ b/src/mongo/db/s/shardsvr_collmod_command.cpp @@ -82,7 +82,8 @@ public: const std::string& dbname, const BSONObj& cmdObj) const override { const NamespaceString nss(parseNs(dbname, cmdObj)); - return auth::checkAuthForCollMod(AuthorizationSession::get(client), nss, cmdObj, false); + return auth::checkAuthForCollMod( + client->getOperationContext(), AuthorizationSession::get(client), nss, cmdObj, false); } bool skipApiVersionCheck() const override { diff --git a/src/mongo/db/service_context.h b/src/mongo/db/service_context.h index 44e12901cea..10dd5e33203 100644 --- a/src/mongo/db/service_context.h +++ b/src/mongo/db/service_context.h @@ -603,6 +603,24 @@ public: _catalogGeneration.fetchAndAdd(1); } + void disallowUserWrites() { + _userWritesAllowed.store(false); + } + + /** + * Returns true if user writes are allowed. + * + * User writes are disallowed when starting with queryableBackupMode or + * recoverFromOplogAsStandalone to prevent users from writing to replicated collections in + * standalone mode. + * + * To determine whether an operation must run in read-only mode, use + * OperationContext::readOnly(). + */ + bool userWritesAllowed() const { + return _userWritesAllowed.load(); + } + LockedClient getLockedClient(OperationId id); private: @@ -749,6 +767,9 @@ private: // When the catalog is restarted, the generation goes up by one each time. AtomicWord<uint64_t> _catalogGeneration{0}; + // Server-wide flag indicating whether users' writes are allowed. + AtomicWord<bool> _userWritesAllowed{true}; + bool _startupComplete = false; stdx::condition_variable _startupCompleteCondVar; }; diff --git a/src/mongo/db/service_entry_point_common.cpp b/src/mongo/db/service_entry_point_common.cpp index 43894e827c5..aaf3d228440 100644 --- a/src/mongo/db/service_entry_point_common.cpp +++ b/src/mongo/db/service_entry_point_common.cpp @@ -2184,7 +2184,7 @@ void HandleRequest::completeOperation(DbResponse& response) { // TODO SERVER-26825: Fix race condition where fsyncLock is acquired post // lockedForWriting() call but prior to profile collection lock acquisition. LOGV2_DEBUG(21972, 1, "Note: not profiling because doing fsync+lock"); - } else if (storageGlobalParams.readOnly) { + } else if (opCtx->readOnly()) { LOGV2_DEBUG(21973, 1, "Note: not profiling because server is read-only"); } else { invariant(!opCtx->lockState()->inAWriteUnitOfWork()); diff --git a/src/mongo/db/sorter/sorter.cpp b/src/mongo/db/sorter/sorter.cpp index 3fa9ebcab32..c3a846772d0 100644 --- a/src/mongo/db/sorter/sorter.cpp +++ b/src/mongo/db/sorter/sorter.cpp @@ -1075,9 +1075,6 @@ private: << " Pass allowDiskUse:true to opt in."); } - // We should check readOnly before getting here. - invariant(!storageGlobalParams.readOnly); - sort(); updateCutoff(); diff --git a/src/mongo/db/startup_recovery.cpp b/src/mongo/db/startup_recovery.cpp index ca1b6fb94e0..f6dfd5352c6 100644 --- a/src/mongo/db/startup_recovery.cpp +++ b/src/mongo/db/startup_recovery.cpp @@ -74,7 +74,7 @@ MONGO_FAIL_POINT_DEFINE(exitBeforeRepairInvalidatesConfig); // Returns true if storage engine is writable. bool isWriteableStorageEngine() { - return !storageGlobalParams.readOnly && (storageGlobalParams.engine != "devnull"); + return storageGlobalParams.engine != "devnull"; } // Attempt to restore the featureCompatibilityVersion document if it is missing. @@ -452,7 +452,7 @@ void setReplSetMemberInStandaloneMode(OperationContext* opCtx, StartupRecoveryMo // Perform startup procedures for --repair mode. void startupRepair(OperationContext* opCtx, StorageEngine* storageEngine) { - invariant(!storageGlobalParams.readOnly); + invariant(!storageGlobalParams.queryableBackupMode); if (MONGO_unlikely(exitBeforeDataRepair.shouldFail())) { LOGV2(21006, "Exiting because 'exitBeforeDataRepair' fail point was set."); @@ -539,26 +539,12 @@ void startupRepair(OperationContext* opCtx, StorageEngine* storageEngine) { } } -// Perform startup procedures for read-only mode. -void startupRecoveryReadOnly(OperationContext* opCtx, StorageEngine* storageEngine) { - invariant(!storageGlobalParams.repair); - - setReplSetMemberInStandaloneMode(opCtx, StartupRecoveryMode::kAuto); - - FeatureCompatibilityVersion::initializeForStartup(opCtx); - - openDatabases(opCtx, storageEngine, [&](auto db) { - // Ensures all collections meet requirements such as having _id indexes. - uassertStatusOK(ensureCollectionProperties(opCtx, db, EnsureIndexPolicy::kError)); - }); -} - // Perform routine startup recovery procedure. void startupRecovery(OperationContext* opCtx, StorageEngine* storageEngine, StorageEngine::LastShutdownState lastShutdownState, StartupRecoveryMode mode) { - invariant(!storageGlobalParams.readOnly && !storageGlobalParams.repair); + invariant(!storageGlobalParams.repair); // Determine whether this is a replica set node running in standalone mode. This must be set // before determining whether to restart index builds. @@ -630,8 +616,6 @@ void repairAndRecoverDatabases(OperationContext* opCtx, if (storageGlobalParams.repair) { startupRepair(opCtx, storageEngine); - } else if (storageGlobalParams.readOnly) { - startupRecoveryReadOnly(opCtx, storageEngine); } else { startupRecovery(opCtx, storageEngine, lastShutdownState, StartupRecoveryMode::kAuto); } @@ -648,7 +632,7 @@ void runStartupRecoveryInMode(OperationContext* opCtx, auto const storageEngine = opCtx->getServiceContext()->getStorageEngine(); Lock::GlobalWrite lk(opCtx); - invariant(isWriteableStorageEngine() && !storageGlobalParams.readOnly); + invariant(isWriteableStorageEngine()); invariant(!storageGlobalParams.repair); const bool usingReplication = repl::ReplicationCoordinator::get(opCtx)->isReplEnabled(); invariant(usingReplication); diff --git a/src/mongo/db/startup_warnings_mongod.cpp b/src/mongo/db/startup_warnings_mongod.cpp index ce7bdd92c6c..60e5bc1f134 100644 --- a/src/mongo/db/startup_warnings_mongod.cpp +++ b/src/mongo/db/startup_warnings_mongod.cpp @@ -388,5 +388,12 @@ void logMongodStartupWarnings(const StorageGlobalParams& storageParams, {logv2::LogTag::kStartupWarnings}, "Running with --restore. This should only be used when restoring from a backup"); } + + if (repl::ReplSettings::shouldRecoverFromOplogAsStandalone()) { + LOGV2_WARNING_OPTIONS(21558, + {logv2::LogTag::kStartupWarnings}, + "Setting mongod to readOnly mode as a result of specifying " + "'recoverFromOplogAsStandalone'"); + } } } // namespace mongo diff --git a/src/mongo/db/storage/control/storage_control.cpp b/src/mongo/db/storage/control/storage_control.cpp index 11222448eba..8b384cbc72f 100644 --- a/src/mongo/db/storage/control/storage_control.cpp +++ b/src/mongo/db/storage/control/storage_control.cpp @@ -83,7 +83,7 @@ void startStorageControls(ServiceContext* serviceContext, bool forTestOnly) { JournalFlusher::set(serviceContext, std::move(journalFlusher)); } - if (!storageEngine->isEphemeral() && !storageGlobalParams.readOnly) { + if (!storageEngine->isEphemeral() && !storageGlobalParams.queryableBackupMode) { std::unique_ptr<Checkpointer> checkpointer = std::make_unique<Checkpointer>(storageEngine->getEngine()); checkpointer->go(); diff --git a/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_kv_engine_test.cpp b/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_kv_engine_test.cpp index 88a76d7cab6..f129c04baa2 100644 --- a/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_kv_engine_test.cpp +++ b/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_kv_engine_test.cpp @@ -77,19 +77,40 @@ class EphemeralForTestKVEngineTest : public ServiceContextTest { public: EphemeralForTestKVEngineTest() : _helper(getServiceContext()), _engine(_helper.getEngine()) {} + ServiceContext::UniqueOperationContext makeOpCtx() { + auto opCtx = makeOperationContext(); + opCtx->setRecoveryUnit(std::unique_ptr<RecoveryUnit>(_engine->newRecoveryUnit()), + WriteUnitOfWork::RecoveryUnitState::kNotInUnitOfWork); + opCtx->swapLockState(std::make_unique<LockerNoop>(), WithLock::withoutLock()); + return opCtx; + } + + std::vector<std::pair<ServiceContext::UniqueClient, ServiceContext::UniqueOperationContext>> + makeOpCtxs(unsigned num) { + std::vector<std::pair<ServiceContext::UniqueClient, ServiceContext::UniqueOperationContext>> + opCtxs; + opCtxs.reserve(num); + + for (unsigned i = 0; i < num; ++i) { + auto client = getServiceContext()->makeClient(std::to_string(i)); + + auto opCtx = client->makeOperationContext(); + opCtx->setRecoveryUnit(std::unique_ptr<RecoveryUnit>(_engine->newRecoveryUnit()), + WriteUnitOfWork::RecoveryUnitState::kNotInUnitOfWork); + opCtx->swapLockState(std::make_unique<LockerNoop>(), WithLock::withoutLock()); + + opCtxs.emplace_back(std::move(client), std::move(opCtx)); + } + + return opCtxs; + } + protected: std::unique_ptr<KVHarnessHelper> helper; KVHarnessHelper _helper; KVEngine* _engine; }; -class OperationContextFromKVEngine : public OperationContextNoop { -public: - OperationContextFromKVEngine(KVEngine* engine) - : OperationContextNoop(engine->newRecoveryUnit()) {} -}; - - TEST_F(EphemeralForTestKVEngineTest, AvailableHistoryUpdate) { NamespaceString nss("a.b"); std::string ident = "collection-1234"; @@ -98,9 +119,9 @@ TEST_F(EphemeralForTestKVEngineTest, AvailableHistoryUpdate) { std::unique_ptr<mongo::RecordStore> rs; { - OperationContextFromKVEngine opCtx(_engine); - ASSERT_OK(_engine->createRecordStore(&opCtx, nss, ident, defaultCollectionOptions)); - rs = _engine->getRecordStore(&opCtx, nss, ident, defaultCollectionOptions); + auto opCtx = makeOpCtx(); + ASSERT_OK(_engine->createRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions)); + rs = _engine->getRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions); ASSERT(rs); } @@ -112,10 +133,10 @@ TEST_F(EphemeralForTestKVEngineTest, AvailableHistoryUpdate) { ASSERT_EQ(_engine->getOldestTimestamp(), currentMaster); { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = makeOpCtx(); + WriteUnitOfWork uow(opCtx.get()); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx.get(), record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); uow.commit(); } @@ -135,9 +156,9 @@ TEST_F(EphemeralForTestKVEngineTest, PinningOldestTimestampWithReadTransaction) std::unique_ptr<mongo::RecordStore> rs; { - OperationContextFromKVEngine opCtx(_engine); - ASSERT_OK(_engine->createRecordStore(&opCtx, nss, ident, defaultCollectionOptions)); - rs = _engine->getRecordStore(&opCtx, nss, ident, defaultCollectionOptions); + auto opCtx = makeOpCtx(); + ASSERT_OK(_engine->createRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions)); + rs = _engine->getRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions); ASSERT(rs); } @@ -146,24 +167,26 @@ TEST_F(EphemeralForTestKVEngineTest, PinningOldestTimestampWithReadTransaction) RecordId loc; { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = makeOpCtx(); + WriteUnitOfWork uow(opCtx.get()); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx.get(), record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); loc = res.getValue(); uow.commit(); } - OperationContextFromKVEngine opCtxRead(_engine); + auto opCtxs = makeOpCtxs(2); + + auto opCtxRead = opCtxs[0].second.get(); RecordData rd; - ASSERT(rs->findRecord(&opCtxRead, loc, &rd)); + ASSERT(rs->findRecord(opCtxRead, loc, &rd)); { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = opCtxs[1].second.get(); + WriteUnitOfWork uow(opCtx); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx, record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); uow.commit(); } @@ -181,9 +204,9 @@ TEST_F(EphemeralForTestKVEngineTest, SettingOldestTimestampClearsHistory) { std::unique_ptr<mongo::RecordStore> rs; { - OperationContextFromKVEngine opCtx(_engine); - ASSERT_OK(_engine->createRecordStore(&opCtx, nss, ident, defaultCollectionOptions)); - rs = _engine->getRecordStore(&opCtx, nss, ident, defaultCollectionOptions); + auto opCtx = makeOpCtx(); + ASSERT_OK(_engine->createRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions)); + rs = _engine->getRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions); ASSERT(rs); } @@ -192,24 +215,26 @@ TEST_F(EphemeralForTestKVEngineTest, SettingOldestTimestampClearsHistory) { RecordId loc; { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = makeOpCtx(); + WriteUnitOfWork uow(opCtx.get()); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx.get(), record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); loc = res.getValue(); uow.commit(); } - OperationContextFromKVEngine opCtxRead(_engine); + auto opCtxs = makeOpCtxs(2); + + auto opCtxRead = opCtxs[0].second.get(); RecordData rd; - ASSERT(rs->findRecord(&opCtxRead, loc, &rd)); + ASSERT(rs->findRecord(opCtxRead, loc, &rd)); { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = opCtxs[1].second.get(); + WriteUnitOfWork uow(opCtx); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx, record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); uow.commit(); } @@ -227,17 +252,17 @@ TEST_F(EphemeralForTestKVEngineTest, SettingOldestTimestampToMax) { std::unique_ptr<mongo::RecordStore> rs; { - OperationContextFromKVEngine opCtx(_engine); - ASSERT_OK(_engine->createRecordStore(&opCtx, nss, ident, defaultCollectionOptions)); - rs = _engine->getRecordStore(&opCtx, nss, ident, defaultCollectionOptions); + auto opCtx = makeOpCtx(); + ASSERT_OK(_engine->createRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions)); + rs = _engine->getRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions); ASSERT(rs); } { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = makeOpCtx(); + WriteUnitOfWork uow(opCtx.get()); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx.get(), record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); uow.commit(); } @@ -258,9 +283,9 @@ TEST_F(EphemeralForTestKVEngineTest, CleanHistoryWithOpenTransaction) { std::unique_ptr<mongo::RecordStore> rs; { - OperationContextFromKVEngine opCtx(_engine); - ASSERT_OK(_engine->createRecordStore(&opCtx, nss, ident, defaultCollectionOptions)); - rs = _engine->getRecordStore(&opCtx, nss, ident, defaultCollectionOptions); + auto opCtx = makeOpCtx(); + ASSERT_OK(_engine->createRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions)); + rs = _engine->getRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions); ASSERT(rs); } @@ -269,25 +294,27 @@ TEST_F(EphemeralForTestKVEngineTest, CleanHistoryWithOpenTransaction) { RecordId loc; { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = makeOpCtx(); + WriteUnitOfWork uow(opCtx.get()); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx.get(), record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); loc = res.getValue(); uow.commit(); } - OperationContextFromKVEngine opCtxRead(_engine); + auto opCtxs = makeOpCtxs(2); + + auto opCtxRead = opCtxs[0].second.get(); Timestamp readTime1 = _engine->getHistory_forTest().rbegin()->first; RecordData rd; - ASSERT(rs->findRecord(&opCtxRead, loc, &rd)); + ASSERT(rs->findRecord(opCtxRead, loc, &rd)); { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = opCtxs[1].second.get(); + WriteUnitOfWork uow(opCtx); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx, record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); uow.commit(); } @@ -295,14 +322,17 @@ TEST_F(EphemeralForTestKVEngineTest, CleanHistoryWithOpenTransaction) { Timestamp readTime2 = _engine->getHistory_forTest().rbegin()->first; { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = opCtxs[1].second.get(); + WriteUnitOfWork uow(opCtx); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx, record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); uow.commit(); } + // Destruct the client used for writes prior to checking use_count(). + opCtxs.pop_back(); + Timestamp readTime3 = _engine->getHistory_forTest().rbegin()->first; _engine->cleanHistory(); @@ -321,44 +351,46 @@ TEST_F(EphemeralForTestKVEngineTest, ReadOlderSnapshotsSimple) { std::unique_ptr<mongo::RecordStore> rs; { - OperationContextFromKVEngine opCtx(_engine); - ASSERT_OK(_engine->createRecordStore(&opCtx, nss, ident, defaultCollectionOptions)); - rs = _engine->getRecordStore(&opCtx, nss, ident, defaultCollectionOptions); + auto opCtx = makeOpCtx(); + ASSERT_OK(_engine->createRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions)); + rs = _engine->getRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions); ASSERT(rs); } + auto opCtxs = makeOpCtxs(2); + // Pin oldest timestamp with a read transaction. - OperationContextFromKVEngine pinningOldest(_engine); - ASSERT(!rs->findRecord(&pinningOldest, RecordId(1), nullptr)); + auto pinningOldest = opCtxs[0].second.get(); + ASSERT(!rs->findRecord(pinningOldest, RecordId(1), nullptr)); // Set readFrom to timestamp with no committed transactions. Timestamp readFrom = _engine->getHistory_forTest().rbegin()->first; - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow1(&opCtx); + auto opCtx = opCtxs[1].second.get(); + WriteUnitOfWork uow1(opCtx); StatusWith<RecordId> res1 = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx, record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res1.getStatus()); RecordId loc1 = res1.getValue(); uow1.commit(); - WriteUnitOfWork uow2(&opCtx); + WriteUnitOfWork uow2(opCtx); StatusWith<RecordId> res2 = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx, record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res2.getStatus()); RecordId loc2 = res2.getValue(); uow2.commit(); RecordData rd; - opCtx.recoveryUnit()->abandonSnapshot(); - opCtx.recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kProvided, readFrom); - ASSERT(!rs->findRecord(&opCtx, loc1, &rd)); - ASSERT(!rs->findRecord(&opCtx, loc2, &rd)); - - opCtx.recoveryUnit()->abandonSnapshot(); - opCtx.recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp); - ASSERT(rs->findRecord(&opCtx, loc1, &rd)); - ASSERT(rs->findRecord(&opCtx, loc2, &rd)); + opCtx->recoveryUnit()->abandonSnapshot(); + opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kProvided, readFrom); + ASSERT(!rs->findRecord(opCtx, loc1, &rd)); + ASSERT(!rs->findRecord(opCtx, loc2, &rd)); + + opCtx->recoveryUnit()->abandonSnapshot(); + opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp); + ASSERT(rs->findRecord(opCtx, loc1, &rd)); + ASSERT(rs->findRecord(opCtx, loc2, &rd)); } TEST_F(EphemeralForTestKVEngineTest, ReadOutdatedSnapshot) { @@ -369,44 +401,50 @@ TEST_F(EphemeralForTestKVEngineTest, ReadOutdatedSnapshot) { std::unique_ptr<mongo::RecordStore> rs; { - OperationContextFromKVEngine opCtx(_engine); - ASSERT_OK(_engine->createRecordStore(&opCtx, nss, ident, defaultCollectionOptions)); - rs = _engine->getRecordStore(&opCtx, nss, ident, defaultCollectionOptions); + auto opCtx = makeOpCtx(); + ASSERT_OK(_engine->createRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions)); + rs = _engine->getRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions); ASSERT(rs); } RecordId loc1; { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = makeOpCtx(); + WriteUnitOfWork uow(opCtx.get()); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx.get(), record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); loc1 = res.getValue(); uow.commit(); } - OperationContextFromKVEngine opCtxRead(_engine); + auto opCtxs = makeOpCtxs(2); + + auto opCtxRead = opCtxs[0].second.get(); RecordData rd; - ASSERT(rs->findRecord(&opCtxRead, loc1, &rd)); + ASSERT(rs->findRecord(opCtxRead, loc1, &rd)); Timestamp readFrom = _engine->getHistory_forTest().rbegin()->first; RecordId loc2; { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = opCtxs[1].second.get(); + WriteUnitOfWork uow(opCtx); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx, record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); loc2 = res.getValue(); uow.commit(); } - ASSERT(rs->findRecord(&opCtxRead, loc1, &rd)); - opCtxRead.recoveryUnit()->abandonSnapshot(); - opCtxRead.recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kProvided, readFrom); + // Destruct the client used for writes prior to trying to get a snapshot too old. + opCtxs.pop_back(); + + ASSERT(rs->findRecord(opCtxRead, loc1, &rd)); + opCtxRead->recoveryUnit()->abandonSnapshot(); + opCtxRead->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kProvided, + readFrom); ASSERT_THROWS_CODE( - rs->findRecord(&opCtxRead, loc1, &rd), DBException, ErrorCodes::SnapshotTooOld); + rs->findRecord(opCtxRead, loc1, &rd), DBException, ErrorCodes::SnapshotTooOld); } TEST_F(EphemeralForTestKVEngineTest, SetReadTimestampBehindOldestTimestamp) { @@ -417,18 +455,18 @@ TEST_F(EphemeralForTestKVEngineTest, SetReadTimestampBehindOldestTimestamp) { std::unique_ptr<mongo::RecordStore> rs; { - OperationContextFromKVEngine opCtx(_engine); - ASSERT_OK(_engine->createRecordStore(&opCtx, nss, ident, defaultCollectionOptions)); - rs = _engine->getRecordStore(&opCtx, nss, ident, defaultCollectionOptions); + auto opCtx = makeOpCtx(); + ASSERT_OK(_engine->createRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions)); + rs = _engine->getRecordStore(opCtx.get(), nss, ident, defaultCollectionOptions); ASSERT(rs); } RecordId loc1; { - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = makeOpCtx(); + WriteUnitOfWork uow(opCtx.get()); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx.get(), record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); loc1 = res.getValue(); uow.commit(); @@ -436,22 +474,23 @@ TEST_F(EphemeralForTestKVEngineTest, SetReadTimestampBehindOldestTimestamp) { RecordData rd; Timestamp readFrom = _engine->getHistory_forTest().begin()->first; - OperationContextFromKVEngine opCtx(_engine); - WriteUnitOfWork uow(&opCtx); + auto opCtx = makeOpCtx(); + WriteUnitOfWork uow(opCtx.get()); StatusWith<RecordId> res = - rs->insertRecord(&opCtx, record.c_str(), record.length() + 1, Timestamp()); + rs->insertRecord(opCtx.get(), record.c_str(), record.length() + 1, Timestamp()); ASSERT_OK(res.getStatus()); RecordId loc2 = res.getValue(); uow.commit(); - opCtx.recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kProvided, readFrom); + opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kProvided, readFrom); _engine->setOldestTimestamp(Timestamp::max(), true); - ASSERT_THROWS_CODE(rs->findRecord(&opCtx, loc2, &rd), DBException, ErrorCodes::SnapshotTooOld); + ASSERT_THROWS_CODE( + rs->findRecord(opCtx.get(), loc2, &rd), DBException, ErrorCodes::SnapshotTooOld); - opCtx.recoveryUnit()->abandonSnapshot(); - opCtx.recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp); - ASSERT(rs->findRecord(&opCtx, loc1, &rd)); - ASSERT(rs->findRecord(&opCtx, loc2, &rd)); + opCtx->recoveryUnit()->abandonSnapshot(); + opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp); + ASSERT(rs->findRecord(opCtx.get(), loc1, &rd)); + ASSERT(rs->findRecord(opCtx.get(), loc2, &rd)); } } // namespace ephemeral_for_test diff --git a/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit.cpp b/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit.cpp index 252b3a0616e..e5b3752381d 100644 --- a/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit.cpp +++ b/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit.cpp @@ -53,7 +53,7 @@ RecoveryUnit::~RecoveryUnit() { _abort(); } -void RecoveryUnit::beginUnitOfWork(OperationContext* opCtx) { +void RecoveryUnit::doBeginUnitOfWork() { invariant(!_inUnitOfWork(), toString(_getState())); _setState(_isActive() ? State::kActive : State::kInactiveInUnitOfWork); } diff --git a/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit.h b/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit.h index c130c4f348f..3be769ef99b 100644 --- a/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit.h +++ b/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit.h @@ -45,8 +45,6 @@ public: RecoveryUnit(KVEngine* parentKVEngine, std::function<void()> cb = nullptr); ~RecoveryUnit(); - void beginUnitOfWork(OperationContext* opCtx) override final; - virtual bool waitUntilDurable(OperationContext* opCtx) override; virtual void setOrderedCommit(bool orderedCommit) override; @@ -109,6 +107,8 @@ public: static RecoveryUnit* get(OperationContext* opCtx); private: + void doBeginUnitOfWork() override final; + void doCommitUnitOfWork() override final; void doAbortUnitOfWork() override final; diff --git a/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit_test.cpp b/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit_test.cpp index 93781b84ed7..2b638c82c3e 100644 --- a/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit_test.cpp +++ b/src/mongo/db/storage/ephemeral_for_test/ephemeral_for_test_recovery_unit_test.cpp @@ -91,7 +91,7 @@ TEST_F(EphemeralForTestRecoveryUnitTestHarness, AbandonSnapshotAbortMode) { const auto rs = harnessHelper->createRecordStore(opCtx.get(), "table1"); opCtx->lockState()->beginWriteUnitOfWork(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); StatusWith<RecordId> rid1 = rs->insertRecord(opCtx.get(), "ABC", 3, Timestamp()); StatusWith<RecordId> rid2 = rs->insertRecord(opCtx.get(), "123", 3, Timestamp()); ASSERT_TRUE(rid1.isOK()); diff --git a/src/mongo/db/storage/record_store.cpp b/src/mongo/db/storage/record_store.cpp index 8dd22c869f6..194bf1f1641 100644 --- a/src/mongo/db/storage/record_store.cpp +++ b/src/mongo/db/storage/record_store.cpp @@ -35,22 +35,22 @@ namespace mongo { namespace { -void validateWriteAllowed() { +void validateWriteAllowed(OperationContext* opCtx) { uassert(ErrorCodes::IllegalOperation, "Cannot execute a write operation in read-only mode", - !storageGlobalParams.readOnly); + !opCtx->readOnly()); } } // namespace void RecordStore::deleteRecord(OperationContext* opCtx, const RecordId& dl) { - validateWriteAllowed(); + validateWriteAllowed(opCtx); doDeleteRecord(opCtx, dl); } Status RecordStore::insertRecords(OperationContext* opCtx, std::vector<Record>* inOutRecords, const std::vector<Timestamp>& timestamps) { - validateWriteAllowed(); + validateWriteAllowed(opCtx); return doInsertRecords(opCtx, inOutRecords, timestamps); } @@ -58,7 +58,7 @@ Status RecordStore::updateRecord(OperationContext* opCtx, const RecordId& recordId, const char* data, int len) { - validateWriteAllowed(); + validateWriteAllowed(opCtx); return doUpdateRecord(opCtx, recordId, data, len); } @@ -67,22 +67,22 @@ StatusWith<RecordData> RecordStore::updateWithDamages(OperationContext* opCtx, const RecordData& oldRec, const char* damageSource, const mutablebson::DamageVector& damages) { - validateWriteAllowed(); + validateWriteAllowed(opCtx); return doUpdateWithDamages(opCtx, loc, oldRec, damageSource, damages); } Status RecordStore::truncate(OperationContext* opCtx) { - validateWriteAllowed(); + validateWriteAllowed(opCtx); return doTruncate(opCtx); } void RecordStore::cappedTruncateAfter(OperationContext* opCtx, RecordId end, bool inclusive) { - validateWriteAllowed(); + validateWriteAllowed(opCtx); doCappedTruncateAfter(opCtx, end, inclusive); } Status RecordStore::compact(OperationContext* opCtx) { - validateWriteAllowed(); + validateWriteAllowed(opCtx); return doCompact(opCtx); } diff --git a/src/mongo/db/storage/recovery_unit.cpp b/src/mongo/db/storage/recovery_unit.cpp index 7edc0197999..4727d564daf 100644 --- a/src/mongo/db/storage/recovery_unit.cpp +++ b/src/mongo/db/storage/recovery_unit.cpp @@ -92,6 +92,29 @@ void RecoveryUnit::commitRegisteredChanges(boost::optional<Timestamp> commitTime _executeCommitHandlers(commitTimestamp); } +void RecoveryUnit::beginUnitOfWork(bool readOnly) { + _readOnly = readOnly; + if (!_readOnly) { + doBeginUnitOfWork(); + } +} + +void RecoveryUnit::commitUnitOfWork() { + invariant(!_readOnly); + doCommitUnitOfWork(); + assignNextSnapshotId(); +} + +void RecoveryUnit::abortUnitOfWork() { + invariant(!_readOnly); + doAbortUnitOfWork(); + assignNextSnapshotId(); +} + +void RecoveryUnit::endReadOnlyUnitOfWork() { + _readOnly = false; +} + void RecoveryUnit::_executeCommitHandlers(boost::optional<Timestamp> commitTimestamp) { for (auto& change : _changes) { try { @@ -158,9 +181,8 @@ void RecoveryUnit::_executeRollbackHandlers() { } void RecoveryUnit::validateInUnitOfWork() const { - invariant(_inUnitOfWork() || storageGlobalParams.readOnly, - fmt::format( - "state: {}, readOnly: {}", toString(_getState()), storageGlobalParams.readOnly)); + invariant(_inUnitOfWork() || _readOnly, + fmt::format("state: {}, readOnly: {}", toString(_getState()), _readOnly)); } } // namespace mongo diff --git a/src/mongo/db/storage/recovery_unit.h b/src/mongo/db/storage/recovery_unit.h index a67d1f090f5..04101ad66dc 100644 --- a/src/mongo/db/storage/recovery_unit.h +++ b/src/mongo/db/storage/recovery_unit.h @@ -129,9 +129,12 @@ public: * Marks the beginning of a unit of work. Each call must be matched with exactly one call to * either commitUnitOfWork or abortUnitOfWork. * + * When called with readOnly=true, no unit of work is started. Calling commitUnitOfWork or + * abortUnitOfWork will invariant. + * * Should be called through WriteUnitOfWork rather than directly. */ - virtual void beginUnitOfWork(OperationContext* opCtx) = 0; + void beginUnitOfWork(bool readOnly); /** * Marks the end of a unit of work and commits all changes registered by calls to onCommit or @@ -139,10 +142,7 @@ public: * * Should be called through WriteUnitOfWork rather than directly. */ - void commitUnitOfWork() { - doCommitUnitOfWork(); - assignNextSnapshotId(); - } + void commitUnitOfWork(); /** * Marks the end of a unit of work and rolls back all changes registered by calls to onRollback @@ -151,10 +151,14 @@ public: * * Should be called through WriteUnitOfWork rather than directly. */ - void abortUnitOfWork() { - doAbortUnitOfWork(); - assignNextSnapshotId(); - } + void abortUnitOfWork(); + + /** + * Cleans up any state set for this unit of work. + * + * Should be called through WriteUnitOfWork rather than directly. + */ + void endReadOnlyUnitOfWork(); /** * Transitions the active unit of work to the "prepared" state. Must be called after @@ -803,6 +807,7 @@ private: // Sets the snapshot associated with this RecoveryUnit to a new globally unique id number. void assignNextSnapshotId(); + virtual void doBeginUnitOfWork() = 0; virtual void doAbandonSnapshot() = 0; virtual void doCommitUnitOfWork() = 0; virtual void doAbortUnitOfWork() = 0; @@ -816,6 +821,7 @@ private: std::unique_ptr<Change> _changeForCatalogVisibility; State _state = State::kInactive; uint64_t _mySnapshotId; + bool _readOnly = false; }; /** diff --git a/src/mongo/db/storage/recovery_unit_noop.h b/src/mongo/db/storage/recovery_unit_noop.h index 4c06eeb62b2..dcc35495b02 100644 --- a/src/mongo/db/storage/recovery_unit_noop.h +++ b/src/mongo/db/storage/recovery_unit_noop.h @@ -40,8 +40,6 @@ class OperationContext; class RecoveryUnitNoop : public RecoveryUnit { public: - void beginUnitOfWork(OperationContext* opCtx) final {} - bool waitUntilDurable(OperationContext* opCtx) override { return true; } @@ -59,6 +57,8 @@ public: } private: + void doBeginUnitOfWork() final {} + void doCommitUnitOfWork() final { _executeCommitHandlers(boost::none); } diff --git a/src/mongo/db/storage/recovery_unit_test_harness.cpp b/src/mongo/db/storage/recovery_unit_test_harness.cpp index 0c4cd3b1739..ca9c0ca5f25 100644 --- a/src/mongo/db/storage/recovery_unit_test_harness.cpp +++ b/src/mongo/db/storage/recovery_unit_test_harness.cpp @@ -88,7 +88,7 @@ TEST_F(RecoveryUnitTestHarness, CommitUnitOfWork) { Lock::GlobalLock globalLk(opCtx.get(), MODE_IX); const auto rs = harnessHelper->createRecordStore(opCtx.get(), "table1"); opCtx->lockState()->beginWriteUnitOfWork(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); StatusWith<RecordId> s = rs->insertRecord(opCtx.get(), "data", 4, Timestamp()); ASSERT_TRUE(s.isOK()); ASSERT_EQUALS(1, rs->numRecords(opCtx.get())); @@ -102,7 +102,7 @@ TEST_F(RecoveryUnitTestHarness, AbortUnitOfWork) { Lock::GlobalLock globalLk(opCtx.get(), MODE_IX); const auto rs = harnessHelper->createRecordStore(opCtx.get(), "table1"); opCtx->lockState()->beginWriteUnitOfWork(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); StatusWith<RecordId> s = rs->insertRecord(opCtx.get(), "data", 4, Timestamp()); ASSERT_TRUE(s.isOK()); ASSERT_EQUALS(1, rs->numRecords(opCtx.get())); @@ -115,13 +115,13 @@ TEST_F(RecoveryUnitTestHarness, CommitAndRollbackChanges) { int count = 0; const auto rs = harnessHelper->createRecordStore(opCtx.get(), "table1"); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); ru->registerChange(std::make_unique<TestChange>(&count)); ASSERT_EQUALS(count, 0); ru->commitUnitOfWork(); ASSERT_EQUALS(count, 1); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); ru->registerChange(std::make_unique<TestChange>(&count)); ASSERT_EQUALS(count, 1); ru->abortUnitOfWork(); @@ -132,7 +132,7 @@ TEST_F(RecoveryUnitTestHarness, CheckIsActiveWithCommit) { Lock::GlobalLock globalLk(opCtx.get(), MODE_IX); const auto rs = harnessHelper->createRecordStore(opCtx.get(), "table1"); opCtx->lockState()->beginWriteUnitOfWork(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); // TODO SERVER-51787: to re-enable this. // ASSERT_TRUE(ru->isActive()); StatusWith<RecordId> s = rs->insertRecord(opCtx.get(), "data", 4, Timestamp()); @@ -145,7 +145,7 @@ TEST_F(RecoveryUnitTestHarness, CheckIsActiveWithAbort) { Lock::GlobalLock globalLk(opCtx.get(), MODE_IX); const auto rs = harnessHelper->createRecordStore(opCtx.get(), "table1"); opCtx->lockState()->beginWriteUnitOfWork(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); // TODO SERVER-51787: to re-enable this. // ASSERT_TRUE(ru->isActive()); StatusWith<RecordId> s = rs->insertRecord(opCtx.get(), "data", 4, Timestamp()); @@ -156,7 +156,7 @@ TEST_F(RecoveryUnitTestHarness, CheckIsActiveWithAbort) { TEST_F(RecoveryUnitTestHarness, BeginningUnitOfWorkDoesNotIncrementSnapshotId) { auto snapshotIdBefore = ru->getSnapshotId(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); ASSERT_EQ(snapshotIdBefore, ru->getSnapshotId()); ru->abortUnitOfWork(); } @@ -174,14 +174,14 @@ TEST_F(RecoveryUnitTestHarness, AbandonSnapshotIncrementsSnapshotId) { TEST_F(RecoveryUnitTestHarness, CommitUnitOfWorkIncrementsSnapshotId) { auto snapshotIdBefore = ru->getSnapshotId(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); ru->commitUnitOfWork(); ASSERT_NE(snapshotIdBefore, ru->getSnapshotId()); } TEST_F(RecoveryUnitTestHarness, AbortUnitOfWorkIncrementsSnapshotId) { auto snapshotIdBefore = ru->getSnapshotId(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); ru->abortUnitOfWork(); ASSERT_NE(snapshotIdBefore, ru->getSnapshotId()); } @@ -195,7 +195,7 @@ TEST_F(RecoveryUnitTestHarness, AbandonSnapshotCommitMode) { const auto rs = harnessHelper->createRecordStore(opCtx.get(), "table1"); opCtx->lockState()->beginWriteUnitOfWork(); - ru->beginUnitOfWork(opCtx.get()); + ru->beginUnitOfWork(opCtx->readOnly()); StatusWith<RecordId> rid1 = rs->insertRecord(opCtx.get(), "ABC", 3, Timestamp()); StatusWith<RecordId> rid2 = rs->insertRecord(opCtx.get(), "123", 3, Timestamp()); ASSERT_TRUE(rid1.isOK()); @@ -231,6 +231,17 @@ TEST_F(RecoveryUnitTestHarness, AbandonSnapshotCommitMode) { ASSERT_EQ(strncmp(recordAfterAbandon->data.data(), "123", 3), 0); } +TEST_F(RecoveryUnitTestHarness, FlipReadOnly) { + ru->beginUnitOfWork(/*readOnly=*/true); + ru->endReadOnlyUnitOfWork(); + + ru->beginUnitOfWork(/*readOnly=*/false); + ru->commitUnitOfWork(); + + ru->beginUnitOfWork(/*readOnly=*/false); + ru->abortUnitOfWork(); +} + DEATH_TEST_F(RecoveryUnitTestHarness, RegisterChangeMustBeInUnitOfWork, "invariant") { int count = 0; opCtx->recoveryUnit()->registerChange(std::make_unique<TestChange>(&count)); @@ -245,7 +256,7 @@ DEATH_TEST_F(RecoveryUnitTestHarness, AbortMustBeInUnitOfWork, "invariant") { } DEATH_TEST_F(RecoveryUnitTestHarness, CannotHaveUnfinishedUnitOfWorkOnExit, "invariant") { - opCtx->recoveryUnit()->beginUnitOfWork(opCtx.get()); + opCtx->recoveryUnit()->beginUnitOfWork(opCtx->readOnly()); } DEATH_TEST_F(RecoveryUnitTestHarness, PrepareMustBeInUnitOfWork, "invariant") { @@ -258,14 +269,24 @@ DEATH_TEST_F(RecoveryUnitTestHarness, PrepareMustBeInUnitOfWork, "invariant") { } DEATH_TEST_F(RecoveryUnitTestHarness, WaitUntilDurableMustBeOutOfUnitOfWork, "invariant") { - opCtx->recoveryUnit()->beginUnitOfWork(opCtx.get()); + opCtx->recoveryUnit()->beginUnitOfWork(opCtx->readOnly()); opCtx->recoveryUnit()->waitUntilDurable(opCtx.get()); } DEATH_TEST_F(RecoveryUnitTestHarness, AbandonSnapshotMustBeOutOfUnitOfWork, "invariant") { - opCtx->recoveryUnit()->beginUnitOfWork(opCtx.get()); + opCtx->recoveryUnit()->beginUnitOfWork(opCtx->readOnly()); opCtx->recoveryUnit()->abandonSnapshot(); } +DEATH_TEST_F(RecoveryUnitTestHarness, CommitInReadOnly, "invariant") { + opCtx->recoveryUnit()->beginUnitOfWork(/*readOnly=*/true); + opCtx->recoveryUnit()->commitUnitOfWork(); +} + +DEATH_TEST_F(RecoveryUnitTestHarness, AbortInReadOnly, "invariant") { + opCtx->recoveryUnit()->beginUnitOfWork(/*readOnly=*/true); + opCtx->recoveryUnit()->abortUnitOfWork(); +} + } // namespace } // namespace mongo diff --git a/src/mongo/db/storage/storage_engine.h b/src/mongo/db/storage/storage_engine.h index 7a973d9883f..313dcb8c439 100644 --- a/src/mongo/db/storage/storage_engine.h +++ b/src/mongo/db/storage/storage_engine.h @@ -100,8 +100,7 @@ public: virtual ~Factory() {} /** - * Return a new instance of the StorageEngine. The lockFile parameter may be null if - * params.readOnly is set. Caller owns the returned pointer. + * Return a new instance of the StorageEngine. Caller owns the returned pointer. */ virtual std::unique_ptr<StorageEngine> create( OperationContext* opCtx, @@ -161,13 +160,11 @@ public: virtual BSONObj createMetadataOptions(const StorageGlobalParams& params) const = 0; /** - * Returns whether the engine supports read-only mode. If read-only mode is enabled, the - * engine may be started on a read-only filesystem (either mounted read-only or with - * read-only permissions). If readOnly mode is enabled, it is undefined behavior to call - * methods that write data (e.g. insertRecord). This method is provided on the Factory - * because it must be called before the storageEngine is instantiated. + * Returns whether the engine supports queryable backup mode. If queryable backup mode is + * enabled, user writes are not permitted but internally generated writes are still + * permitted. */ - virtual bool supportsReadOnly() const { + virtual bool supportsQueryableBackupMode() const { return false; } }; diff --git a/src/mongo/db/storage/storage_engine_impl.cpp b/src/mongo/db/storage/storage_engine_impl.cpp index 5f5e0de5a9d..a70e6039c23 100644 --- a/src/mongo/db/storage/storage_engine_impl.cpp +++ b/src/mongo/db/storage/storage_engine_impl.cpp @@ -797,10 +797,6 @@ void StorageEngineImpl::cleanShutdown() { StorageEngineImpl::~StorageEngineImpl() {} void StorageEngineImpl::startTimestampMonitor() { - if (storageGlobalParams.readOnly) { - return; - } - // Unless explicitly disabled, all storage engines should create a TimestampMonitor for // drop-pending internal idents, even if they do not support pending drops for collections // and indexes. diff --git a/src/mongo/db/storage/storage_engine_init.cpp b/src/mongo/db/storage/storage_engine_init.cpp index 7c345ddcfe6..eb6f1e7dc40 100644 --- a/src/mongo/db/storage/storage_engine_init.cpp +++ b/src/mongo/db/storage/storage_engine_init.cpp @@ -85,19 +85,16 @@ StorageEngine::LastShutdownState initializeStorageEngine(OperationContext* opCtx const std::string dbpath = storageGlobalParams.dbpath; - if (!storageGlobalParams.readOnly) { - StorageRepairObserver::set(service, std::make_unique<StorageRepairObserver>(dbpath)); - auto repairObserver = StorageRepairObserver::get(service); - - if (storageGlobalParams.repair) { - repairObserver->onRepairStarted(); - } else if (repairObserver->isIncomplete()) { - LOGV2_FATAL_NOTRACE( - 50922, - "An incomplete repair has been detected! This is likely because a repair " - "operation unexpectedly failed before completing. MongoDB will not start up " - "again without --repair."); - } + StorageRepairObserver::set(service, std::make_unique<StorageRepairObserver>(dbpath)); + auto repairObserver = StorageRepairObserver::get(service); + + if (storageGlobalParams.repair) { + repairObserver->onRepairStarted(); + } else if (repairObserver->isIncomplete()) { + LOGV2_FATAL_NOTRACE(50922, + "An incomplete repair has been detected! This is likely because a " + "repair operation unexpectedly failed before completing. MongoDB will " + "not start up again without --repair."); } if (auto existingStorageEngine = StorageEngineMetadata::getStorageEngineForPath(dbpath)) { @@ -134,12 +131,12 @@ StorageEngine::LastShutdownState initializeStorageEngine(OperationContext* opCtx << storageGlobalParams.engine, factory); - if (storageGlobalParams.readOnly) { + if (storageGlobalParams.queryableBackupMode) { uassert(34368, - str::stream() - << "Server was started in read-only mode, but the configured storage engine, " - << storageGlobalParams.engine << ", does not support read-only operation", - factory->supportsReadOnly()); + str::stream() << "Server was started in queryable backup mode, but the configured " + << "storage engine, " << storageGlobalParams.engine + << ", does not support queryable backup mode", + factory->supportsQueryableBackupMode()); } std::unique_ptr<StorageEngineMetadata> metadata; @@ -147,13 +144,6 @@ StorageEngine::LastShutdownState initializeStorageEngine(OperationContext* opCtx metadata = StorageEngineMetadata::forPath(dbpath); } - if (storageGlobalParams.readOnly) { - uassert(34415, - "Server was started in read-only mode, but the storage metadata file was not" - " found.", - metadata.get()); - } - // Validate options in metadata against current startup options. if (metadata.get()) { uassertStatusOK(factory->validateMetadata(*metadata, storageGlobalParams)); @@ -221,7 +211,6 @@ StorageEngine::LastShutdownState initializeStorageEngine(OperationContext* opCtx // Write a new metadata file if it is not present. if (!metadata.get() && (initFlags & StorageEngineInitFlags::kSkipMetadataFile) == StorageEngineInitFlags{}) { - invariant(!storageGlobalParams.readOnly); metadata.reset(new StorageEngineMetadata(storageGlobalParams.dbpath)); metadata->setStorageEngine(factory->getCanonicalName().toString()); metadata->setStorageEngineOptions(factory->createMetadataOptions(storageGlobalParams)); @@ -297,18 +286,13 @@ void createLockFile(ServiceContext* service) { } const bool wasUnclean = lockFile->createdByUncleanShutdown(); const auto openStatus = lockFile->open(); - if (storageGlobalParams.readOnly && openStatus == ErrorCodes::IllegalOperation) { + if (openStatus == ErrorCodes::IllegalOperation) { lockFile = boost::none; } else { uassertStatusOK(openStatus); } if (wasUnclean) { - if (storageGlobalParams.readOnly) { - LOGV2_FATAL_NOTRACE(34416, - "Attempted to open dbpath in readOnly mode, but the server was " - "previously not shut down cleanly."); - } LOGV2_WARNING(22271, "Detected unclean shutdown - Lock file is not empty", "lockFile"_attr = lockFile->getFilespec()); diff --git a/src/mongo/db/storage/storage_init.cpp b/src/mongo/db/storage/storage_init.cpp index d2e48977c70..18fe6c43b70 100644 --- a/src/mongo/db/storage/storage_init.cpp +++ b/src/mongo/db/storage/storage_init.cpp @@ -68,7 +68,7 @@ public: bob.append("dropPendingIdents", static_cast<long long>(engine->getDropPendingIdents().size())); bob.append("supportsSnapshotReadConcern", engine->supportsReadConcernSnapshot()); - bob.append("readOnly", storageGlobalParams.readOnly); + bob.append("readOnly", !opCtx->getServiceContext()->userWritesAllowed()); bob.append("persistent", !engine->isEphemeral()); bob.append("backupCursorOpen", backupCursorHooks->isBackupCursorOpen()); diff --git a/src/mongo/db/storage/storage_options.cpp b/src/mongo/db/storage/storage_options.cpp index 69a0f2aa6dc..ca2d9de84dc 100644 --- a/src/mongo/db/storage/storage_options.cpp +++ b/src/mongo/db/storage/storage_options.cpp @@ -55,7 +55,7 @@ void StorageGlobalParams::reset() { noTableScan.store(false); directoryperdb = false; syncdelay = 60.0; - readOnly = false; + queryableBackupMode = false; groupCollections = false; oplogMinRetentionHours.store(0.0); allowOplogTruncation = true; diff --git a/src/mongo/db/storage/storage_options.h b/src/mongo/db/storage/storage_options.h index 8e95c4c934f..0455f09d245 100644 --- a/src/mongo/db/storage/storage_options.h +++ b/src/mongo/db/storage/storage_options.h @@ -105,10 +105,9 @@ struct StorageGlobalParams { AtomicDouble syncdelay; // seconds between fsyncs // --queryableBackupMode - // Puts MongoD into "read-only" mode. MongoD will not write any data to the underlying - // filesystem. Note that read operations may require writes. For example, a sort on a large - // dataset may fail if it requires spilling to disk. - bool readOnly; + // Prevents user-originating operations from performing writes to the server. Internally + // generated writes are still permitted. + bool queryableBackupMode; // --groupCollections // Dictate to the storage engine that it should attempt to create new MongoDB collections from diff --git a/src/mongo/db/storage/storage_repair_observer.cpp b/src/mongo/db/storage/storage_repair_observer.cpp index 22b76a6a39c..44bc890f3e8 100644 --- a/src/mongo/db/storage/storage_repair_observer.cpp +++ b/src/mongo/db/storage/storage_repair_observer.cpp @@ -65,8 +65,6 @@ const auto getRepairObserver = } // namespace StorageRepairObserver::StorageRepairObserver(const std::string& dbpath) { - invariant(!storageGlobalParams.readOnly); - using boost::filesystem::path; _repairIncompleteFilePath = path(dbpath) / path(kRepairIncompleteFileName); diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_index.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_index.cpp index c54af529d49..6d9686f2341 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_index.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_index.cpp @@ -233,10 +233,9 @@ WiredTigerIndex::WiredTigerIndex(OperationContext* ctx, StringData ident, KeyFormat rsKeyFormat, const IndexDescriptor* desc, - bool isLogged, - bool isReadOnly) + bool isLogged) : SortedDataInterface(ident, - _handleVersionInfo(ctx, uri, desc, isLogged, isReadOnly), + _handleVersionInfo(ctx, uri, desc, isLogged), Ordering::make(desc->keyPattern()), rsKeyFormat), _uri(uri), @@ -633,8 +632,7 @@ StatusWith<bool> WiredTigerIndex::_checkDups(OperationContext* opCtx, KeyString::Version WiredTigerIndex::_handleVersionInfo(OperationContext* ctx, const std::string& uri, const IndexDescriptor* desc, - bool isLogged, - bool isReadOnly) { + bool isLogged) { auto version = WiredTigerUtil::checkApplicationMetadataFormatVersion( ctx, uri, kMinimumIndexVersion, kMaximumIndexVersion); if (!version.isOK()) { @@ -661,9 +659,7 @@ KeyString::Version WiredTigerIndex::_handleVersionInfo(OperationContext* ctx, fassertFailedWithStatusNoTrace(31179, versionStatus); } - if (!isReadOnly) { - uassertStatusOK(WiredTigerUtil::setTableLogging(ctx, uri, isLogged)); - } + uassertStatusOK(WiredTigerUtil::setTableLogging(ctx, uri, isLogged)); /* * Index data format 6, 11, and 13 correspond to KeyString version V0 and data format 8, 12, and @@ -1484,10 +1480,8 @@ WiredTigerIndexUnique::WiredTigerIndexUnique(OperationContext* ctx, StringData ident, KeyFormat rsKeyFormat, const IndexDescriptor* desc, - bool isLogged, - bool isReadOnly) - : WiredTigerIndex(ctx, uri, ident, rsKeyFormat, desc, isLogged, isReadOnly), - _partial(desc->isPartial()) { + bool isLogged) + : WiredTigerIndex(ctx, uri, ident, rsKeyFormat, desc, isLogged), _partial(desc->isPartial()) { // _id indexes must use WiredTigerIdIndex invariant(!isIdIndex()); // All unique indexes should be in the timestamp-safe format version as of version 4.2. @@ -1573,9 +1567,8 @@ WiredTigerIdIndex::WiredTigerIdIndex(OperationContext* ctx, const std::string& uri, StringData ident, const IndexDescriptor* desc, - bool isLogged, - bool isReadOnly) - : WiredTigerIndex(ctx, uri, ident, KeyFormat::Long, desc, isLogged, isReadOnly) { + bool isLogged) + : WiredTigerIndex(ctx, uri, ident, KeyFormat::Long, desc, isLogged) { invariant(isIdIndex()); } @@ -1805,9 +1798,8 @@ WiredTigerIndexStandard::WiredTigerIndexStandard(OperationContext* ctx, StringData ident, KeyFormat rsKeyFormat, const IndexDescriptor* desc, - bool isLogged, - bool isReadOnly) - : WiredTigerIndex(ctx, uri, ident, rsKeyFormat, desc, isLogged, isReadOnly) {} + bool isLogged) + : WiredTigerIndex(ctx, uri, ident, rsKeyFormat, desc, isLogged) {} std::unique_ptr<SortedDataInterface::Cursor> WiredTigerIndexStandard::newCursor( OperationContext* opCtx, bool forward) const { diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_index.h b/src/mongo/db/storage/wiredtiger/wiredtiger_index.h index 5a94980c22c..44e1bb16ad9 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_index.h +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_index.h @@ -126,8 +126,7 @@ public: StringData ident, KeyFormat rsKeyFormat, const IndexDescriptor* desc, - bool isLogged, - bool readOnly); + bool isLogged); virtual Status insert(OperationContext* opCtx, const KeyString::Value& keyString, @@ -232,8 +231,7 @@ protected: KeyString::Version _handleVersionInfo(OperationContext* ctx, const std::string& uri, const IndexDescriptor* desc, - bool isLogged, - bool isReadOnly); + bool isLogged); RecordId _decodeRecordIdAtEnd(const void* buffer, size_t size); @@ -263,8 +261,7 @@ public: StringData ident, KeyFormat rsKeyFormat, const IndexDescriptor* desc, - bool isLogged, - bool readOnly = false); + bool isLogged); std::unique_ptr<SortedDataInterface::Cursor> newCursor(OperationContext* opCtx, bool forward) const override; @@ -305,8 +302,7 @@ public: const std::string& uri, StringData ident, const IndexDescriptor* desc, - bool isLogged, - bool readOnly = false); + bool isLogged); std::unique_ptr<Cursor> newCursor(OperationContext* opCtx, bool isForward = true) const override; @@ -357,8 +353,7 @@ public: StringData ident, KeyFormat rsKeyFormat, const IndexDescriptor* desc, - bool isLogged, - bool readOnly = false); + bool isLogged); std::unique_ptr<SortedDataInterface::Cursor> newCursor(OperationContext* opCtx, bool forward) const override; diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_init.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_init.cpp index da82969596e..b674bd24111 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_init.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_init.cpp @@ -126,8 +126,7 @@ public: wiredTigerGlobalOptions.getMaxHistoryFileSizeMB(), params.dur, ephemeral, - params.repair, - params.readOnly); + params.repair); kv->setRecordStoreExtraOptions(wiredTigerGlobalOptions.collectionConfig); kv->setSortedDataInterfaceExtraOptions(wiredTigerGlobalOptions.indexConfig); @@ -204,7 +203,7 @@ public: return builder.obj(); } - bool supportsReadOnly() const final { + bool supportsQueryableBackupMode() const final { return true; } }; diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp index 244a11ac071..66f419dd63e 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.cpp @@ -143,12 +143,7 @@ boost::filesystem::path getOngoingBackupPath() { } // namespace -bool WiredTigerFileVersion::shouldDowngrade(bool readOnly, bool hasRecoveryTimestamp) { - if (readOnly) { - // A read-only state must not have upgraded. Nor could it downgrade. - return false; - } - +bool WiredTigerFileVersion::shouldDowngrade(bool hasRecoveryTimestamp) { const auto replCoord = repl::ReplicationCoordinator::get(getGlobalServiceContext()); if (replCoord && replCoord->getMemberState().arbiter()) { // SERVER-35361: Arbiters will no longer downgrade their data files. To downgrade @@ -304,8 +299,7 @@ WiredTigerKVEngine::WiredTigerKVEngine(const std::string& canonicalName, size_t maxHistoryFileSizeMB, bool durable, bool ephemeral, - bool repair, - bool readOnly) + bool repair) : _clockSource(cs), _oplogManager(std::make_unique<WiredTigerOplogManager>()), _canonicalName(canonicalName), @@ -314,7 +308,6 @@ WiredTigerKVEngine::WiredTigerKVEngine(const std::string& canonicalName, _durable(durable), _ephemeral(ephemeral), _inRepairMode(repair), - _readOnly(readOnly), _keepDataHistory(serverGlobalParams.enableMajorityReadConcern) { _pinnedOplogTimestamp.store(Timestamp::max().asULL()); boost::filesystem::path journalPath = path; @@ -359,8 +352,7 @@ WiredTigerKVEngine::WiredTigerKVEngine(const std::string& canonicalName, // The setting may have a later setting override it if not using the journal. We make it // unconditional here because even nojournal may need this setting if it is a transition // from using the journal. - ss << "log=(enabled=true,remove=" << (_readOnly ? "false" : "true") - << ",path=journal,compressor="; + ss << "log=(enabled=true,remove=true,path=journal,compressor="; ss << wiredTigerGlobalOptions.journalCompressor << "),"; ss << "builtin_extension_config=(zstd=(compression_level=" << wiredTigerGlobalOptions.zstdCompressorLevel << ")),"; @@ -530,7 +522,7 @@ WiredTigerKVEngine::WiredTigerKVEngine(const std::string& canonicalName, setOldestActiveTransactionTimestampCallback( [](Timestamp) { return StatusWith(boost::make_optional(Timestamp::min())); }); - if (!_readOnly && !_ephemeral) { + if (!_ephemeral) { if (!_recoveryTimestamp.isNull()) { // If the oldest/initial data timestamps were unset (there was no persisted durable // history), initialize them to the recovery timestamp. @@ -575,7 +567,7 @@ WiredTigerKVEngine::WiredTigerKVEngine(const std::string& canonicalName, _sizeStorerUri = _uri("sizeStorer"); WiredTigerSession session(_conn); - if (!_readOnly && repair && _hasUri(session.getSession(), _sizeStorerUri)) { + if (repair && _hasUri(session.getSession(), _sizeStorerUri)) { LOGV2(22316, "Repairing size cache"); auto status = _salvageIfNeeded(_sizeStorerUri.c_str()); @@ -583,7 +575,7 @@ WiredTigerKVEngine::WiredTigerKVEngine(const std::string& canonicalName, fassertNoTrace(28577, status); } - _sizeStorer = std::make_unique<WiredTigerSizeStorer>(_conn, _sizeStorerUri, _readOnly); + _sizeStorer = std::make_unique<WiredTigerSizeStorer>(_conn, _sizeStorerUri); _runTimeConfigParam.reset(makeServerParameter<WiredTigerEngineRuntimeConfigParameter>( "wiredTigerEngineRuntimeConfig", ServerParameterType::kRuntimeOnly)); _runTimeConfigParam->_data.second = this; @@ -740,9 +732,7 @@ void WiredTigerKVEngine::cleanShutdown() { _sessionCache->shuttingDown(); - if (!_readOnly) { - syncSizeInfo(/*syncToDisk=*/true); - } + syncSizeInfo(/*syncToDisk=*/true); // The size storer has to be destructed after the session cache has shut down. This sets the // shutdown flag internally in the session cache. As operations get interrupted during shutdown, @@ -782,7 +772,7 @@ void WiredTigerKVEngine::cleanShutdown() { quickExit(EXIT_SUCCESS); } - if (_fileVersion.shouldDowngrade(_readOnly, !_recoveryTimestamp.isNull())) { + if (_fileVersion.shouldDowngrade(!_recoveryTimestamp.isNull())) { auto startTime = Date_t::now(); LOGV2(22324, "Closing WiredTiger in preparation for reconfiguring", @@ -1604,7 +1594,6 @@ std::unique_ptr<RecordStore> WiredTigerKVEngine::getRecordStore(OperationContext params.isLogged = isLogged; params.cappedCallback = nullptr; params.sizeStorer = _sizeStorer.get(); - params.isReadOnly = _readOnly; params.tracksSizeAdjustments = true; params.forceUpdateWithFullDocument = options.timeseries != boost::none; @@ -1713,33 +1702,21 @@ std::unique_ptr<SortedDataInterface> WiredTigerKVEngine::getSortedDataInterface( if (desc->isIdIndex()) { invariant(!collOptions.clusteredIndex); return std::make_unique<WiredTigerIdIndex>( - opCtx, _uri(ident), ident, desc, WiredTigerUtil::useTableLogging(nss), _readOnly); + opCtx, _uri(ident), ident, desc, WiredTigerUtil::useTableLogging(nss)); } auto keyFormat = (collOptions.clusteredIndex) ? KeyFormat::String : KeyFormat::Long; if (desc->unique()) { - return std::make_unique<WiredTigerIndexUnique>(opCtx, - _uri(ident), - ident, - keyFormat, - desc, - WiredTigerUtil::useTableLogging(nss), - _readOnly); + return std::make_unique<WiredTigerIndexUnique>( + opCtx, _uri(ident), ident, keyFormat, desc, WiredTigerUtil::useTableLogging(nss)); } - return std::make_unique<WiredTigerIndexStandard>(opCtx, - _uri(ident), - ident, - keyFormat, - desc, - WiredTigerUtil::useTableLogging(nss), - _readOnly); + return std::make_unique<WiredTigerIndexStandard>( + opCtx, _uri(ident), ident, keyFormat, desc, WiredTigerUtil::useTableLogging(nss)); } std::unique_ptr<RecordStore> WiredTigerKVEngine::makeTemporaryRecordStore(OperationContext* opCtx, StringData ident, KeyFormat keyFormat) { - invariant(!_readOnly || !recoverToOplogTimestamp.empty()); - _ensureIdentPath(ident); WiredTigerSession wtSession(_conn); @@ -1780,7 +1757,6 @@ std::unique_ptr<RecordStore> WiredTigerKVEngine::makeTemporaryRecordStore(Operat params.sizeStorer = nullptr; // Temporary collections do not need to reconcile collection size/counts. params.tracksSizeAdjustments = false; - params.isReadOnly = false; params.forceUpdateWithFullDocument = false; std::unique_ptr<WiredTigerRecordStore> rs; @@ -1937,7 +1913,7 @@ bool WiredTigerKVEngine::haveDropsQueued() const { Date_t now = _clockSource->now(); Milliseconds delta = now - Date_t::fromMillisSinceEpoch(_previousCheckedDropsQueued.load()); - if (!_readOnly && _sizeStorerSyncTracker.intervalHasElapsed()) { + if (_sizeStorerSyncTracker.intervalHasElapsed()) { _sizeStorerSyncTracker.resetLastTime(); syncSizeInfo(false); } @@ -2453,7 +2429,7 @@ StatusWith<Timestamp> WiredTigerKVEngine::recoverToStableTimestamp(OperationCont _highestSeenDurableTimestamp = stableTimestamp.asULL(); } - _sizeStorer = std::make_unique<WiredTigerSizeStorer>(_conn, _sizeStorerUri, _readOnly); + _sizeStorer = std::make_unique<WiredTigerSizeStorer>(_conn, _sizeStorerUri); return {stableTimestamp}; } @@ -2538,10 +2514,6 @@ boost::optional<Timestamp> WiredTigerKVEngine::getOplogNeededForCrashRecovery() return boost::none; } - if (_readOnly) { - return boost::none; - } - return Timestamp(_oplogNeededForCrashRecovery.load()); } diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h index 55f3155d338..c4cecf51f04 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine.h @@ -71,7 +71,7 @@ struct WiredTigerFileVersion { inline static const std::string kLatestWTRelease = "compatibility=(release=10.0)"; StartupVersion _startupVersion; - bool shouldDowngrade(bool readOnly, bool hasRecoveryTimestamp); + bool shouldDowngrade(bool hasRecoveryTimestamp); std::string getDowngradeString(); }; @@ -109,8 +109,7 @@ public: size_t maxHistoryFileSizeMB, bool durable, bool ephemeral, - bool repair, - bool readOnly); + bool repair); ~WiredTigerKVEngine(); @@ -472,7 +471,6 @@ private: bool _durable; bool _ephemeral; // whether we are using the in-memory mode of the WT engine const bool _inRepairMode; - bool _readOnly; // If _keepDataHistory is true, then the storage engine keeps all history after the stable // timestamp, and WiredTigerKVEngine is responsible for advancing the oldest timestamp. If diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine_test.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine_test.cpp index 9917d95e0f2..c587811c169 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine_test.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_kv_engine_test.cpp @@ -99,8 +99,7 @@ private: 0, true, false, - _forRepair, - false); + _forRepair); } const std::unique_ptr<ClockSource> _cs = std::make_unique<ClockSourceMock>(); @@ -514,29 +513,29 @@ TEST_F(WiredTigerKVEngineTest, WiredTigerDowngrade) { // (Generic FCV reference): When FCV is kLatest, no downgrade is necessary. serverGlobalParams.mutableFeatureCompatibility.setVersion(multiversion::GenericFCV::kLatest); - ASSERT_FALSE(version.shouldDowngrade(/*readOnly=*/false, /*hasRecoveryTimestamp=*/false)); + ASSERT_FALSE(version.shouldDowngrade(/*hasRecoveryTimestamp=*/false)); ASSERT_EQ(WiredTigerFileVersion::kLatestWTRelease, version.getDowngradeString()); // (Generic FCV reference): When FCV is kLastContinuous or kLastLTS, a downgrade may be needed. serverGlobalParams.mutableFeatureCompatibility.setVersion( multiversion::GenericFCV::kLastContinuous); - ASSERT_TRUE(version.shouldDowngrade(/*readOnly=*/false, /*hasRecoveryTimestamp=*/false)); + ASSERT_TRUE(version.shouldDowngrade(/*hasRecoveryTimestamp=*/false)); ASSERT_EQ(WiredTigerFileVersion::kLastContinuousWTRelease, version.getDowngradeString()); serverGlobalParams.mutableFeatureCompatibility.setVersion(multiversion::GenericFCV::kLastLTS); - ASSERT_TRUE(version.shouldDowngrade(/*readOnly=*/false, /*hasRecoveryTimestamp=*/false)); + ASSERT_TRUE(version.shouldDowngrade(/*hasRecoveryTimestamp=*/false)); ASSERT_EQ(WiredTigerFileVersion::kLastLTSWTRelease, version.getDowngradeString()); // (Generic FCV reference): While we're in a semi-downgraded state, we shouldn't try downgrading // the WiredTiger compatibility version. serverGlobalParams.mutableFeatureCompatibility.setVersion( multiversion::GenericFCV::kDowngradingFromLatestToLastContinuous); - ASSERT_FALSE(version.shouldDowngrade(/*readOnly=*/false, /*hasRecoveryTimestamp=*/false)); + ASSERT_FALSE(version.shouldDowngrade(/*hasRecoveryTimestamp=*/false)); ASSERT_EQ(WiredTigerFileVersion::kLatestWTRelease, version.getDowngradeString()); serverGlobalParams.mutableFeatureCompatibility.setVersion( multiversion::GenericFCV::kDowngradingFromLatestToLastLTS); - ASSERT_FALSE(version.shouldDowngrade(/*readOnly=*/false, /*hasRecoveryTimestamp=*/false)); + ASSERT_FALSE(version.shouldDowngrade(/*hasRecoveryTimestamp=*/false)); ASSERT_EQ(WiredTigerFileVersion::kLatestWTRelease, version.getDowngradeString()); } diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.cpp index b59a4bdaef5..014c7e6e508 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.cpp @@ -927,9 +927,7 @@ WiredTigerRecordStore::WiredTigerRecordStore(WiredTigerKVEngine* kvEngine, } } - if (!params.isReadOnly) { - uassertStatusOK(WiredTigerUtil::setTableLogging(ctx, _uri, _isLogged)); - } + uassertStatusOK(WiredTigerUtil::setTableLogging(ctx, _uri, _isLogged)); if (_isOplog) { invariant(_keyFormat == KeyFormat::Long); @@ -1006,14 +1004,10 @@ void WiredTigerRecordStore::checkSize(OperationContext* opCtx) { } void WiredTigerRecordStore::postConstructorInit(OperationContext* opCtx) { - // When starting up with recoverFromOplogAsStandalone=true, the readOnly flag is initially set - // to false to allow oplog recovery to run and perform its necessary writes. After recovery is - // complete, the readOnly flag gets flipped to true. Because of this subtlety, we avoid - // calculating the oplog stones when recoverFromOplogAsStandalone=true as the RecordStore - // construction for the oplog happens before the readOnly flag gets flipped to true. - if (NamespaceString::oplog(ns()) && - !(storageGlobalParams.repair || storageGlobalParams.readOnly || - repl::ReplSettings::shouldRecoverFromOplogAsStandalone())) { + // If the server was started in read-only mode, skip calculating the oplog stones. The + // OplogCapMaintainerThread does not get started in this instance. + if (NamespaceString::oplog(ns()) && opCtx->getServiceContext()->userWritesAllowed() && + !storageGlobalParams.repair) { _oplogStones = std::make_shared<OplogStones>(opCtx, this); } diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.h b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.h index 7630cc0900b..e00ae177f95 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.h +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store.h @@ -114,7 +114,6 @@ public: boost::optional<int64_t> oplogMaxSize; CappedCallback* cappedCallback; WiredTigerSizeStorer* sizeStorer; - bool isReadOnly; bool tracksSizeAdjustments; bool forceUpdateWithFullDocument; }; diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store_test.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store_test.cpp index a2d9c5f7e99..32e0257fc4a 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store_test.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store_test.cpp @@ -952,7 +952,7 @@ TEST(WiredTigerRecordStoreTest, GetLatestOplogTest) { // 1) Initialize the top of oplog to "1". ServiceContext::UniqueOperationContext op1(harnessHelper->newOperationContext()); - op1->recoveryUnit()->beginUnitOfWork(op1.get()); + op1->recoveryUnit()->beginUnitOfWork(op1->readOnly()); Timestamp tsOne = Timestamp( static_cast<unsigned long long>(_oplogOrderInsertOplog(op1.get(), rs, 1).getLong())); op1->recoveryUnit()->commitUnitOfWork(); @@ -960,7 +960,7 @@ TEST(WiredTigerRecordStoreTest, GetLatestOplogTest) { ASSERT_EQ(tsOne, wtrs->getLatestOplogTimestamp(op1.get())); // 2) Open a hole at time "2". - op1->recoveryUnit()->beginUnitOfWork(op1.get()); + op1->recoveryUnit()->beginUnitOfWork(op1->readOnly()); // Don't save the return value because the compiler complains about unused variables. _oplogOrderInsertOplog(op1.get(), rs, 2); // Querying with the recovery unit with a snapshot will not return the uncommitted value. @@ -971,7 +971,7 @@ TEST(WiredTigerRecordStoreTest, GetLatestOplogTest) { Client::initThread("client2"); ServiceContext::UniqueOperationContext op2(harnessHelper->newOperationContext()); - op2->recoveryUnit()->beginUnitOfWork(op2.get()); + op2->recoveryUnit()->beginUnitOfWork(op2->readOnly()); Timestamp tsThree = Timestamp( static_cast<unsigned long long>(_oplogOrderInsertOplog(op2.get(), rs, 3).getLong())); // Before committing, the query still only sees timestamp "1". @@ -1116,7 +1116,6 @@ TEST(WiredTigerRecordStoreTest, ClusteredRecordStore) { params.cappedCallback = nullptr; params.sizeStorer = nullptr; params.tracksSizeAdjustments = true; - params.isReadOnly = false; params.forceUpdateWithFullDocument = false; const auto wtKvEngine = dynamic_cast<WiredTigerKVEngine*>(harnessHelper->getEngine()); diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store_test_harness.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store_test_harness.cpp index 70912221539..0b38e30c443 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_record_store_test_harness.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_record_store_test_harness.cpp @@ -28,8 +28,6 @@ */ #include "mongo/db/storage/wiredtiger/wiredtiger_record_store_test_harness.h" - -#include "mongo/db/operation_context_noop.h" #include "mongo/db/repl/replication_coordinator_mock.h" namespace mongo { @@ -53,7 +51,6 @@ WiredTigerHarnessHelper::WiredTigerHarnessHelper(StringData extraStrings) 0, true, false, - false, false) { repl::ReplicationCoordinator::set( serviceContext(), @@ -63,8 +60,8 @@ WiredTigerHarnessHelper::WiredTigerHarnessHelper(StringData extraStrings) std::unique_ptr<RecordStore> WiredTigerHarnessHelper::newRecordStore( const std::string& ns, const CollectionOptions& collOptions, KeyFormat keyFormat) { - WiredTigerRecoveryUnit* ru = checked_cast<WiredTigerRecoveryUnit*>(_engine.newRecoveryUnit()); - OperationContextNoop opCtx(ru); + ServiceContext::UniqueOperationContext opCtx(newOperationContext()); + WiredTigerRecoveryUnit* ru = checked_cast<WiredTigerRecoveryUnit*>(opCtx->recoveryUnit()); std::string uri = WiredTigerKVEngine::kTableUriPrefix + ns; StringData ident = ns; NamespaceString nss(ns); @@ -81,7 +78,7 @@ std::unique_ptr<RecordStore> WiredTigerHarnessHelper::newRecordStore( std::string config = result.getValue(); { - WriteUnitOfWork uow(&opCtx); + WriteUnitOfWork uow(opCtx.get()); WT_SESSION* s = ru->getSession()->getSession(); invariantWTOK(s->create(s, uri.c_str(), config.c_str()), s); uow.commit(); @@ -98,26 +95,24 @@ std::unique_ptr<RecordStore> WiredTigerHarnessHelper::newRecordStore( params.isLogged = WiredTigerUtil::useTableLogging(nss); params.cappedCallback = nullptr; params.sizeStorer = nullptr; - params.isReadOnly = false; params.tracksSizeAdjustments = true; params.forceUpdateWithFullDocument = collOptions.timeseries != boost::none; - auto ret = std::make_unique<StandardWiredTigerRecordStore>(&_engine, &opCtx, params); - ret->postConstructorInit(&opCtx); + auto ret = std::make_unique<StandardWiredTigerRecordStore>(&_engine, opCtx.get(), params); + ret->postConstructorInit(opCtx.get()); return std::move(ret); } std::unique_ptr<RecordStore> WiredTigerHarnessHelper::newOplogRecordStore() { auto ret = newOplogRecordStoreNoInit(); - auto* ru = _engine.newRecoveryUnit(); - OperationContextNoop opCtx(ru); - dynamic_cast<WiredTigerRecordStore*>(ret.get())->postConstructorInit(&opCtx); + ServiceContext::UniqueOperationContext opCtx(newOperationContext()); + dynamic_cast<WiredTigerRecordStore*>(ret.get())->postConstructorInit(opCtx.get()); return ret; } std::unique_ptr<RecordStore> WiredTigerHarnessHelper::newOplogRecordStoreNoInit() { - WiredTigerRecoveryUnit* ru = dynamic_cast<WiredTigerRecoveryUnit*>(_engine.newRecoveryUnit()); - OperationContextNoop opCtx(ru); + ServiceContext::UniqueOperationContext opCtx(newOperationContext()); + WiredTigerRecoveryUnit* ru = checked_cast<WiredTigerRecoveryUnit*>(opCtx->recoveryUnit()); std::string ident = NamespaceString::kRsOplogNamespace.ns(); std::string uri = WiredTigerKVEngine::kTableUriPrefix + ident; @@ -137,7 +132,7 @@ std::unique_ptr<RecordStore> WiredTigerHarnessHelper::newOplogRecordStoreNoInit( std::string config = result.getValue(); { - WriteUnitOfWork uow(&opCtx); + WriteUnitOfWork uow(opCtx.get()); WT_SESSION* s = ru->getSession()->getSession(); invariantWTOK(s->create(s, uri.c_str(), config.c_str()), s); uow.commit(); @@ -156,10 +151,9 @@ std::unique_ptr<RecordStore> WiredTigerHarnessHelper::newOplogRecordStoreNoInit( params.oplogMaxSize = 1024 * 1024 * 1024; params.cappedCallback = nullptr; params.sizeStorer = nullptr; - params.isReadOnly = false; params.tracksSizeAdjustments = true; params.forceUpdateWithFullDocument = false; - return std::make_unique<StandardWiredTigerRecordStore>(&_engine, &opCtx, params); + return std::make_unique<StandardWiredTigerRecordStore>(&_engine, opCtx.get(), params); } std::unique_ptr<RecoveryUnit> WiredTigerHarnessHelper::newRecoveryUnit() { diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.cpp index 969db98506e..99a4bc7a195 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.cpp @@ -228,7 +228,7 @@ void WiredTigerRecoveryUnit::_abort() { _setState(State::kInactive); } -void WiredTigerRecoveryUnit::beginUnitOfWork(OperationContext* opCtx) { +void WiredTigerRecoveryUnit::doBeginUnitOfWork() { invariant(!_inUnitOfWork(), toString(_getState())); invariant(!_isCommittingOrAborting(), str::stream() << "cannot begin unit of work while commit or rollback handlers are " diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.h b/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.h index 2d75b4e54f1..a1360dcdf0d 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.h +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit.h @@ -105,7 +105,6 @@ public: WiredTigerRecoveryUnit(WiredTigerSessionCache* sc, WiredTigerOplogManager* oplogManager); ~WiredTigerRecoveryUnit(); - void beginUnitOfWork(OperationContext* opCtx) override; void prepareUnitOfWork() override; bool waitUntilDurable(OperationContext* opCtx) override; @@ -232,6 +231,7 @@ public: void storeWriteContextForDebugging(const BSONObj& info); private: + void doBeginUnitOfWork() override; void doCommitUnitOfWork() override; void doAbortUnitOfWork() override; diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit_test.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit_test.cpp index da700a414bb..6ed86d9fadc 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit_test.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_recovery_unit_test.cpp @@ -60,8 +60,7 @@ public: 0, // .maxCacheOverflowFileSizeMB false, // .durable false, // .ephemeral - false, // .repair - false // .readOnly + false // .repair ) { repl::ReplicationCoordinator::set( getGlobalServiceContext(), @@ -113,7 +112,6 @@ public: params.cappedCallback = nullptr; params.sizeStorer = nullptr; params.tracksSizeAdjustments = true; - params.isReadOnly = false; params.forceUpdateWithFullDocument = false; auto ret = std::make_unique<StandardWiredTigerRecordStore>(&_engine, opCtx, params); @@ -279,7 +277,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, NoOverlapReadSource) { TEST_F(WiredTigerRecoveryUnitTestFixture, LocalReadOnADocumentBeingPreparedWithoutIgnoringPreparedTriggersPrepareConflict) { // Prepare but don't commit a transaction - ru1->beginUnitOfWork(clientAndCtx1.second.get()); + ru1->beginUnitOfWork(clientAndCtx1.second->readOnly()); WT_CURSOR* cursor; getCursor(ru1, &cursor); cursor->set_key(cursor, "key"); @@ -289,7 +287,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, ru1->prepareUnitOfWork(); // The transaction read default enforces prepare conflicts and triggers a WT_PREPARE_CONFLICT. - ru2->beginUnitOfWork(clientAndCtx2.second.get()); + ru2->beginUnitOfWork(clientAndCtx2.second->readOnly()); getCursor(ru2, &cursor); cursor->set_key(cursor, "key"); int ret = cursor->search(cursor); @@ -302,7 +300,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, TEST_F(WiredTigerRecoveryUnitTestFixture, LocalReadOnADocumentBeingPreparedDoesntTriggerPrepareConflict) { // Prepare but don't commit a transaction - ru1->beginUnitOfWork(clientAndCtx1.second.get()); + ru1->beginUnitOfWork(clientAndCtx1.second->readOnly()); WT_CURSOR* cursor; getCursor(ru1, &cursor); cursor->set_key(cursor, "key"); @@ -313,7 +311,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, // A transaction that chooses to ignore prepare conflicts does not see the record instead of // returning a prepare conflict. - ru2->beginUnitOfWork(clientAndCtx2.second.get()); + ru2->beginUnitOfWork(clientAndCtx2.second->readOnly()); ru2->setPrepareConflictBehavior(PrepareConflictBehavior::kIgnoreConflicts); getCursor(ru2, &cursor); cursor->set_key(cursor, "key"); @@ -326,7 +324,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, TEST_F(WiredTigerRecoveryUnitTestFixture, WriteAllowedWhileIgnorePrepareFalse) { // Prepare but don't commit a transaction - ru1->beginUnitOfWork(clientAndCtx1.second.get()); + ru1->beginUnitOfWork(clientAndCtx1.second->readOnly()); WT_CURSOR* cursor; getCursor(ru1, &cursor); cursor->set_key(cursor, "key1"); @@ -337,7 +335,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, WriteAllowedWhileIgnorePrepareFalse) { // A transaction that chooses to ignore prepare conflicts with kIgnoreConflictsAllowWrites does // not see the record - ru2->beginUnitOfWork(clientAndCtx2.second.get()); + ru2->beginUnitOfWork(clientAndCtx2.second->readOnly()); ru2->setPrepareConflictBehavior(PrepareConflictBehavior::kIgnoreConflictsAllowWrites); // The prepared write is not visible. @@ -358,7 +356,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, WriteAllowedWhileIgnorePrepareFalse) { TEST_F(WiredTigerRecoveryUnitTestFixture, WriteOnADocumentBeingPreparedTriggersWTRollback) { // Prepare but don't commit a transaction - ru1->beginUnitOfWork(clientAndCtx1.second.get()); + ru1->beginUnitOfWork(clientAndCtx1.second->readOnly()); WT_CURSOR* cursor; getCursor(ru1, &cursor); cursor->set_key(cursor, "key"); @@ -368,7 +366,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, WriteOnADocumentBeingPreparedTriggersW ru1->prepareUnitOfWork(); // Another transaction with write triggers WT_ROLLBACK - ru2->beginUnitOfWork(clientAndCtx2.second.get()); + ru2->beginUnitOfWork(clientAndCtx2.second->readOnly()); getCursor(ru2, &cursor); cursor->set_key(cursor, "key"); cursor->set_value(cursor, "value2"); @@ -792,7 +790,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, MultiTimestampConstraintsInternalState Timestamp ts2(2, 2); OperationContext* opCtx = clientAndCtx1.second.get(); - ru1->beginUnitOfWork(opCtx); + ru1->beginUnitOfWork(opCtx->readOnly()); // Perform an non timestamped write. WT_CURSOR* cursor; @@ -813,7 +811,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, MultiTimestampConstraintsInternalState // Committing the unit of work should reset the internal state for the multi timestamp // constraint checks. ru1->commitUnitOfWork(); - ru1->beginUnitOfWork(opCtx); + ru1->beginUnitOfWork(opCtx->readOnly()); // Perform a write at ts2. cursor->set_key(cursor, "key3"); @@ -831,7 +829,7 @@ TEST_F(WiredTigerRecoveryUnitTestFixture, AbandonSnapshotAbortMode) { const char* const key = "key"; { - ru1->beginUnitOfWork(opCtx); + ru1->beginUnitOfWork(opCtx->readOnly()); WT_CURSOR* cursor; getCursor(ru1, &cursor); @@ -869,7 +867,7 @@ DEATH_TEST_REGEX_F(WiredTigerRecoveryUnitTestFixture, Timestamp ts2(2, 2); OperationContext* opCtx = clientAndCtx1.second.get(); - ru1->beginUnitOfWork(opCtx); + ru1->beginUnitOfWork(opCtx->readOnly()); auto writeTest = [&]() { // Perform an non timestamped write. diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_size_storer.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_size_storer.cpp index 59a4e92e6a0..d326df17ae6 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_size_storer.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_size_storer.cpp @@ -47,17 +47,8 @@ namespace mongo { -WiredTigerSizeStorer::WiredTigerSizeStorer(WT_CONNECTION* conn, - const std::string& storageUri, - bool readOnly) - : _conn(conn), - _storageUri(storageUri), - _tableId(WiredTigerSession::genTableId()), - _readOnly(readOnly) { - if (_readOnly) { - return; - } - +WiredTigerSizeStorer::WiredTigerSizeStorer(WT_CONNECTION* conn, const std::string& storageUri) + : _conn(conn), _storageUri(storageUri), _tableId(WiredTigerSession::genTableId()) { std::string config = WiredTigerCustomizationHooks::get(getGlobalServiceContext()) ->getTableCreateConfig(_storageUri); @@ -69,7 +60,7 @@ WiredTigerSizeStorer::WiredTigerSizeStorer(WT_CONNECTION* conn, void WiredTigerSizeStorer::store(StringData uri, std::shared_ptr<SizeInfo> sizeInfo) { // If the SizeInfo is still dirty, we're done. - if (sizeInfo->_dirty.load() || _readOnly) + if (sizeInfo->_dirty.load()) return; // Ordering is important: as the entry may be flushed concurrently, set the dirty flag last. diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_size_storer.h b/src/mongo/db/storage/wiredtiger/wiredtiger_size_storer.h index 50eb23324c0..077fbfe75e4 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_size_storer.h +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_size_storer.h @@ -76,7 +76,7 @@ public: AtomicWord<bool> _dirty; }; - WiredTigerSizeStorer(WT_CONNECTION* conn, const std::string& storageUri, bool readOnly = false); + WiredTigerSizeStorer(WT_CONNECTION* conn, const std::string& storageUri); ~WiredTigerSizeStorer() = default; /** @@ -96,7 +96,6 @@ private: WT_CONNECTION* _conn; const std::string _storageUri; const uint64_t _tableId; // Not persisted - const bool _readOnly; // Serializes flushes to disk. Mutex _flushMutex = MONGO_MAKE_LATCH("WiredTigerSessionStorer::_flushMutex"); diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_standard_record_store_test.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_standard_record_store_test.cpp index 1b74a9ead4e..c05cc6097a5 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_standard_record_store_test.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_standard_record_store_test.cpp @@ -72,8 +72,7 @@ TEST(WiredTigerRecordStoreTest, SizeStorer1) { string uri = checked_cast<WiredTigerRecordStore*>(rs.get())->getURI(); string indexUri = WiredTigerKVEngine::kTableUriPrefix + "myindex"; - const bool enableWtLogging = false; - WiredTigerSizeStorer ss(harnessHelper->conn(), indexUri, enableWtLogging); + WiredTigerSizeStorer ss(harnessHelper->conn(), indexUri); checked_cast<WiredTigerRecordStore*>(rs.get())->setSizeStorer(&ss); int N = 12; @@ -116,7 +115,6 @@ TEST(WiredTigerRecordStoreTest, SizeStorer1) { params.isLogged = false; params.cappedCallback = nullptr; params.sizeStorer = &ss; - params.isReadOnly = false; params.tracksSizeAdjustments = true; params.forceUpdateWithFullDocument = false; @@ -146,8 +144,7 @@ TEST(WiredTigerRecordStoreTest, SizeStorer1) { { ServiceContext::UniqueOperationContext opCtx(harnessHelper->newOperationContext()); - const bool enableWtLogging = false; - WiredTigerSizeStorer ss2(harnessHelper->conn(), indexUri, enableWtLogging); + WiredTigerSizeStorer ss2(harnessHelper->conn(), indexUri); auto info = ss2.load(opCtx.get(), uri); ASSERT_EQUALS(N, info->numRecords.load()); } @@ -159,11 +156,8 @@ class SizeStorerUpdateTest : public mongo::unittest::Test { private: virtual void setUp() { harnessHelper.reset(new WiredTigerHarnessHelper()); - const bool enableWtLogging = false; - sizeStorer.reset( - new WiredTigerSizeStorer(harnessHelper->conn(), - WiredTigerKVEngine::kTableUriPrefix + "sizeStorer", - enableWtLogging)); + sizeStorer.reset(new WiredTigerSizeStorer( + harnessHelper->conn(), WiredTigerKVEngine::kTableUriPrefix + "sizeStorer")); rs = harnessHelper->newRecordStore(); WiredTigerRecordStore* wtrs = checked_cast<WiredTigerRecordStore*>(rs.get()); wtrs->setSizeStorer(sizeStorer.get()); diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp index de31ec10751..d4e48faafe9 100644 --- a/src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp +++ b/src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp @@ -832,9 +832,7 @@ void WiredTigerUtil::notifyStartupComplete() { _tableLoggingInfo.isInitializing = false; } - if (!storageGlobalParams.readOnly) { - removeTableChecksFile(); - } + removeTableChecksFile(); } void WiredTigerUtil::resetTableLoggingInfo() { @@ -878,7 +876,6 @@ Status WiredTigerUtil::setTableLogging(OperationContext* opCtx, const std::strin WiredTigerSessionCache* sessionCache = WiredTigerRecoveryUnit::get(opCtx)->getSessionCache(); sessionCache->closeAllCursors(uri); - invariant(!storageGlobalParams.readOnly); stdx::lock_guard<Latch> lk(_tableLoggingInfoMutex); // Update the table logging settings regardless if we're no longer starting up the process. diff --git a/src/mongo/db/storage/write_unit_of_work.cpp b/src/mongo/db/storage/write_unit_of_work.cpp index 3e2bb922565..564ff4098b5 100644 --- a/src/mongo/db/storage/write_unit_of_work.cpp +++ b/src/mongo/db/storage/write_unit_of_work.cpp @@ -58,9 +58,7 @@ WriteUnitOfWork::WriteUnitOfWork(OperationContext* opCtx, bool groupOplogEntries _opCtx->lockState()->beginWriteUnitOfWork(); if (_toplevel) { - if (!storageGlobalParams.readOnly) { - _opCtx->recoveryUnit()->beginUnitOfWork(_opCtx); - } + _opCtx->recoveryUnit()->beginUnitOfWork(_opCtx->readOnly()); _opCtx->_ruState = RecoveryUnitState::kActiveUnitOfWork; } // Make sure we don't silently proceed after a previous WriteUnitOfWork under the same parent @@ -71,7 +69,7 @@ WriteUnitOfWork::WriteUnitOfWork(OperationContext* opCtx, bool groupOplogEntries WriteUnitOfWork::~WriteUnitOfWork() { if (!_released && !_committed) { invariant(_opCtx->_ruState != RecoveryUnitState::kNotInUnitOfWork); - if (!storageGlobalParams.readOnly) { + if (!_opCtx->readOnly()) { if (_toplevel) { // Abort unit of work and execute rollback handlers _opCtx->recoveryUnit()->abortUnitOfWork(); @@ -80,7 +78,8 @@ WriteUnitOfWork::~WriteUnitOfWork() { _opCtx->_ruState = RecoveryUnitState::kFailedUnitOfWork; } } else { - // Just execute rollback handlers in readOnly mode + // Clear the readOnly state and execute rollback handlers in readOnly mode. + _opCtx->recoveryUnit()->endReadOnlyUnitOfWork(); _opCtx->recoveryUnit()->abortRegisteredChanges(); } _opCtx->lockState()->endWriteUnitOfWork(); @@ -142,12 +141,12 @@ void WriteUnitOfWork::commit() { // Execute preCommit hooks before committing the transaction. This is an opportunity to // throw or do any last changes before committing. _opCtx->recoveryUnit()->runPreCommitHooks(_opCtx); - if (!storageGlobalParams.readOnly) { - // Just execute commit handlers in readOnly mode - _opCtx->recoveryUnit()->commitUnitOfWork(); - } else { + if (!_opCtx->readOnly()) { // Commit unit of work and execute commit or rollback handlers depending on whether the // commit was successful. + _opCtx->recoveryUnit()->commitUnitOfWork(); + } else { + // Just execute commit handlers in readOnly mode _opCtx->recoveryUnit()->commitRegisteredChanges(boost::none); } diff --git a/src/mongo/db/system_index.cpp b/src/mongo/db/system_index.cpp index 2283e75c9d6..8386e02710a 100644 --- a/src/mongo/db/system_index.cpp +++ b/src/mongo/db/system_index.cpp @@ -129,13 +129,6 @@ void generateSystemIndexForExistingCollection(OperationContext* opCtx, } // namespace Status verifySystemIndexes(OperationContext* opCtx) { - // Do not try and generate any system indexes in read only mode. - if (storageGlobalParams.readOnly) { - LOGV2_WARNING(22489, - "Running in queryable backup mode. Unable to create authorization indexes"); - return Status::OK(); - } - const NamespaceString& systemUsers = AuthorizationManager::usersCollectionNamespace; const NamespaceString& systemRoles = AuthorizationManager::rolesCollectionNamespace; diff --git a/src/mongo/embedded/embedded.cpp b/src/mongo/embedded/embedded.cpp index 20395926254..eb2549573b5 100644 --- a/src/mongo/embedded/embedded.cpp +++ b/src/mongo/embedded/embedded.cpp @@ -234,6 +234,13 @@ ServiceContext* initialize(const char* yaml_config) { serviceContext, serviceContext->getPreciseClockSource()); serviceContext->setPeriodicRunner(std::move(periodicRunner)); + // When starting the server with --queryableBackupMode or --recoverFromOplogAsStandalone, we are + // in read-only mode and don't allow user-originating operations to perform writes + if (storageGlobalParams.queryableBackupMode || + repl::ReplSettings::shouldRecoverFromOplogAsStandalone()) { + serviceContext->disallowUserWrites(); + } + setUpCatalog(serviceContext); // Creating the operation context before initializing the storage engine allows the storage @@ -278,15 +285,11 @@ ServiceContext* initialize(const char* yaml_config) { uassert(50677, ss.str().c_str(), boost::filesystem::exists(storageGlobalParams.dbpath)); } - if (!storageGlobalParams.readOnly) { - boost::filesystem::remove_all(storageGlobalParams.dbpath + "/_tmp/"); - } + boost::filesystem::remove_all(storageGlobalParams.dbpath + "/_tmp/"); ReadWriteConcernDefaults::create(serviceContext, readWriteConcernDefaultsCacheLookupEmbedded); - bool canCallFCVSetIfCleanStartup = - !storageGlobalParams.readOnly && !(storageGlobalParams.engine == "devnull"); - if (canCallFCVSetIfCleanStartup) { + if (storageGlobalParams.engine != "devnull") { Lock::GlobalWrite lk(startupOpCtx.get()); FeatureCompatibilityVersion::setIfCleanStartup(startupOpCtx.get(), repl::StorageInterface::get(serviceContext)); diff --git a/src/mongo/embedded/embedded_ismaster.cpp b/src/mongo/embedded/embedded_ismaster.cpp index 6b9903f35ca..a9f6763952e 100644 --- a/src/mongo/embedded/embedded_ismaster.cpp +++ b/src/mongo/embedded/embedded_ismaster.cpp @@ -87,7 +87,7 @@ public: result.append("minWireVersion", wireSpec->incomingExternalClient.minWireVersion); result.append("maxWireVersion", wireSpec->incomingExternalClient.maxWireVersion); - result.append("readOnly", storageGlobalParams.readOnly); + result.append("readOnly", opCtx->readOnly()); return true; } diff --git a/src/mongo/s/commands/cluster_collection_mod_cmd.cpp b/src/mongo/s/commands/cluster_collection_mod_cmd.cpp index d295ca18c71..73e968a618e 100644 --- a/src/mongo/s/commands/cluster_collection_mod_cmd.cpp +++ b/src/mongo/s/commands/cluster_collection_mod_cmd.cpp @@ -72,7 +72,8 @@ public: const std::string& dbname, const BSONObj& cmdObj) const override { const NamespaceString nss(CommandHelpers::parseNsCollectionRequired(dbname, cmdObj)); - return auth::checkAuthForCollMod(AuthorizationSession::get(client), nss, cmdObj, true); + return auth::checkAuthForCollMod( + client->getOperationContext(), AuthorizationSession::get(client), nss, cmdObj, true); } bool supportsWriteConcern(const BSONObj& cmd) const override { diff --git a/src/mongo/s/commands/cluster_count_cmd.cpp b/src/mongo/s/commands/cluster_count_cmd.cpp index 48545c7c0d7..04aa145b3d7 100644 --- a/src/mongo/s/commands/cluster_count_cmd.cpp +++ b/src/mongo/s/commands/cluster_count_cmd.cpp @@ -139,6 +139,7 @@ public: uassertStatusOK(countCommandAsAggregationCommand(countRequest, nss)); auto aggCmdOnViewObj = OpMsgRequest::fromDBAndBody(nss.db(), aggCmdOnView).body; auto aggRequestOnView = aggregation_request_helper::parseFromBSON( + opCtx, nss, aggCmdOnViewObj, boost::none, @@ -252,6 +253,7 @@ public: auto aggCmdOnViewObj = OpMsgRequest::fromDBAndBody(nss.db(), aggCmdOnView.getValue()).body; auto aggRequestOnView = aggregation_request_helper::parseFromBSON( + opCtx, nss, aggCmdOnViewObj, verbosity, diff --git a/src/mongo/s/commands/cluster_create_cmd.cpp b/src/mongo/s/commands/cluster_create_cmd.cpp index b7c6d5ea88b..f6496645e22 100644 --- a/src/mongo/s/commands/cluster_create_cmd.cpp +++ b/src/mongo/s/commands/cluster_create_cmd.cpp @@ -108,7 +108,7 @@ public: void doCheckAuthorization(OperationContext* opCtx) const final { uassertStatusOK(auth::checkAuthForCreate( - AuthorizationSession::get(opCtx->getClient()), request(), true)); + opCtx, AuthorizationSession::get(opCtx->getClient()), request(), true)); } CreateCommandReply typedRun(OperationContext* opCtx) final { diff --git a/src/mongo/s/commands/cluster_distinct_cmd.cpp b/src/mongo/s/commands/cluster_distinct_cmd.cpp index 70ae5f4b671..cfb7b572e2c 100644 --- a/src/mongo/s/commands/cluster_distinct_cmd.cpp +++ b/src/mongo/s/commands/cluster_distinct_cmd.cpp @@ -147,6 +147,7 @@ public: auto viewAggCmd = OpMsgRequest::fromDBAndBody(nss.db(), aggCmdOnView.getValue()).body; auto aggRequestOnView = aggregation_request_helper::parseFromBSON( + opCtx, nss, viewAggCmd, verbosity, @@ -228,6 +229,7 @@ public: auto viewAggCmd = OpMsgRequest::fromDBAndBody(nss.db(), aggCmdOnView.getValue()).body; auto aggRequestOnView = aggregation_request_helper::parseFromBSON( + opCtx, nss, viewAggCmd, boost::none, diff --git a/src/mongo/s/commands/cluster_find_cmd.h b/src/mongo/s/commands/cluster_find_cmd.h index efbc0bd4d72..b45c8e3d837 100644 --- a/src/mongo/s/commands/cluster_find_cmd.h +++ b/src/mongo/s/commands/cluster_find_cmd.h @@ -174,6 +174,7 @@ public: OpMsgRequest::fromDBAndBody(_dbName, aggCmdOnView).body; auto aggRequestOnView = aggregation_request_helper::parseFromBSON( + opCtx, ns(), viewAggregationCommand, verbosity, @@ -243,6 +244,7 @@ public: OpMsgRequest::fromDBAndBody(_dbName, aggCmdOnView).body; auto aggRequestOnView = aggregation_request_helper::parseFromBSON( + opCtx, ns(), viewAggregationCommand, boost::none, diff --git a/src/mongo/s/commands/cluster_pipeline_cmd.h b/src/mongo/s/commands/cluster_pipeline_cmd.h index 788a974803d..0554d2a97e5 100644 --- a/src/mongo/s/commands/cluster_pipeline_cmd.h +++ b/src/mongo/s/commands/cluster_pipeline_cmd.h @@ -74,7 +74,8 @@ public: const OpMsgRequest& opMsgRequest, boost::optional<ExplainOptions::Verbosity> explainVerbosity) override { const auto aggregationRequest = - Impl::parseAggregationRequest(opMsgRequest, + Impl::parseAggregationRequest(opCtx, + opMsgRequest, explainVerbosity, APIParameters::get(opCtx).getAPIStrict().value_or(false)); diff --git a/src/mongo/s/commands/cluster_pipeline_cmd_s.cpp b/src/mongo/s/commands/cluster_pipeline_cmd_s.cpp index 3009d9265e6..d099095fe96 100644 --- a/src/mongo/s/commands/cluster_pipeline_cmd_s.cpp +++ b/src/mongo/s/commands/cluster_pipeline_cmd_s.cpp @@ -54,11 +54,15 @@ struct ClusterPipelineCommandS { } static AggregateCommandRequest parseAggregationRequest( + OperationContext* opCtx, const OpMsgRequest& opMsgRequest, boost::optional<ExplainOptions::Verbosity> explainVerbosity, bool apiStrict) { - return aggregation_request_helper::parseFromBSON( - opMsgRequest.getDatabase().toString(), opMsgRequest.body, explainVerbosity, apiStrict); + return aggregation_request_helper::parseFromBSON(opCtx, + opMsgRequest.getDatabase().toString(), + opMsgRequest.body, + explainVerbosity, + apiStrict); } }; ClusterPipelineCommandBase<ClusterPipelineCommandS> clusterPipelineCmdS; |