diff options
Diffstat (limited to 'src/mongo/db/mongod_main.cpp')
| -rw-r--r-- | src/mongo/db/mongod_main.cpp | 477 |
1 files changed, 88 insertions, 389 deletions
diff --git a/src/mongo/db/mongod_main.cpp b/src/mongo/db/mongod_main.cpp index 4f46f4ceb71..52a9c4651d6 100644 --- a/src/mongo/db/mongod_main.cpp +++ b/src/mongo/db/mongod_main.cpp @@ -74,6 +74,7 @@ #include "mongo/db/concurrency/flow_control_ticketholder.h" #include "mongo/db/concurrency/lock_state.h" #include "mongo/db/concurrency/replication_state_transition_lock_guard.h" +#include "mongo/db/concurrency/write_conflict_exception.h" #include "mongo/db/db_raii.h" #include "mongo/db/dbdirectclient.h" #include "mongo/db/dbhelpers.h" @@ -81,9 +82,11 @@ #include "mongo/db/exec/working_set_common.h" #include "mongo/db/fcv_op_observer.h" #include "mongo/db/fle_crud.h" +#include "mongo/db/free_mon/free_mon_mongod.h" #include "mongo/db/ftdc/ftdc_mongod.h" #include "mongo/db/ftdc/util.h" #include "mongo/db/global_settings.h" +#include "mongo/db/index/index_access_method_factory_impl.h" #include "mongo/db/index_builds_coordinator_mongod.h" #include "mongo/db/index_names.h" #include "mongo/db/initialize_server_global_state.h" @@ -102,7 +105,6 @@ #include "mongo/db/logical_time_validator.h" #include "mongo/db/mirror_maestro.h" #include "mongo/db/mongod_options.h" -#include "mongo/db/mongod_options_storage_gen.h" #include "mongo/db/namespace_string.h" #include "mongo/db/op_observer_impl.h" #include "mongo/db/op_observer_registry.h" @@ -232,10 +234,6 @@ #include <sys/file.h> #endif -#ifdef __linux__ -#include <sys/prctl.h> -#endif - namespace mongo { using logv2::LogComponent; @@ -351,38 +349,11 @@ void registerPrimaryOnlyServices(ServiceContext* serviceContext) { MONGO_FAIL_POINT_DEFINE(shutdownAtStartup); -void logMongodStartupTimeElapsedStatistics(ServiceContext* serviceContext, - Date_t beginInitAndListen, - BSONObjBuilder* startupTimeElapsedBuilder, - BSONObjBuilder* startupInfoBuilder, - StorageEngine::LastShutdownState lastShutdownState) { - mongo::Milliseconds elapsedInitAndListen = - serviceContext->getFastClockSource()->now() - beginInitAndListen; - startupTimeElapsedBuilder->append("_initAndListen total elapsed time", - elapsedInitAndListen.toString()); - startupInfoBuilder->append("Startup from clean shutdown?", - lastShutdownState == StorageEngine::LastShutdownState::kClean); - startupInfoBuilder->append("Statistics", startupTimeElapsedBuilder->obj()); - LOGV2_INFO(8423403, - "mongod startup complete", - "Summary of time elapsed"_attr = startupInfoBuilder->obj()); -} - -// Important: -// _initAndListen among its other tasks initializes the storage subsystem. -// File Copy Based Initial Sync will restart the storage subsystem and may need to repeat some -// of the initialization steps within. If you add or change any of these steps, make sure -// any necessary changes are also made to File Copy Based Initial Sync. ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { Client::initThread("initandlisten"); serviceContext->setFastClockSource(FastClockSourceFactory::create(Milliseconds(10))); - BSONObjBuilder startupTimeElapsedBuilder; - BSONObjBuilder startupInfoBuilder; - - Date_t beginInitAndListen = serviceContext->getFastClockSource()->now(); - DBDirectClientFactory::get(serviceContext).registerImplementation([](OperationContext* opCtx) { return std::unique_ptr<DBClientBase>(new DBDirectClient(opCtx)); }); @@ -426,9 +397,6 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { #endif if (!storageGlobalParams.repair) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Transport layer setup", - &startupTimeElapsedBuilder); auto tl = transport::TransportLayerManager::createWithConfig(&serverGlobalParams, serviceContext); auto res = tl->setup(); @@ -447,34 +415,16 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { serviceContext, repl::ReplicationCoordinator::get(serviceContext))); // If a crash occurred during file-copy based initial sync, we may need to finish or clean up. - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Run initial syncer crash recovery", - &startupTimeElapsedBuilder); - repl::InitialSyncerFactory::get(serviceContext)->runCrashRecovery(); - } + repl::InitialSyncerFactory::get(serviceContext)->runCrashRecovery(); // Creating the operation context before initializing the storage engine allows the storage // engine initialization to make use of the lock manager. As the storage engine is not yet // initialized, a noop recovery unit is used until the initialization is complete. auto startupOpCtx = serviceContext->makeOperationContext(&cc()); - auto lastShutdownState = initializeStorageEngine( - startupOpCtx.get(), StorageEngineInitFlags{}, &startupTimeElapsedBuilder); + auto lastShutdownState = initializeStorageEngine(startupOpCtx.get(), StorageEngineInitFlags{}); StorageControl::startStorageControls(serviceContext); - ScopeGuard logStartupStats([serviceContext, - beginInitAndListen, - &startupTimeElapsedBuilder, - &startupInfoBuilder, - lastShutdownState] { - logMongodStartupTimeElapsedStatistics(serviceContext, - beginInitAndListen, - &startupTimeElapsedBuilder, - &startupInfoBuilder, - lastShutdownState); - }); - #ifdef MONGO_CONFIG_WIREDTIGER_ENABLED if (EncryptionHooks::get(serviceContext)->restartRequired()) { exitCleanly(EXIT_CLEAN); @@ -533,12 +483,6 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { exitCleanly(EXIT_BADOPTIONS); } - if (gAllowDocumentsGreaterThanMaxUserSize && replSettings.usingReplSets()) { - LOGV2_ERROR(8472200, - "allowDocumentsGreaterThanMaxUserSize can only be used in standalone mode"); - exitCleanly(EXIT_BADOPTIONS); - } - logMongodStartupWarnings(storageGlobalParams, serverGlobalParams, serviceContext); { @@ -555,8 +499,7 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { startWatchdog(serviceContext); try { - startup_recovery::repairAndRecoverDatabases( - startupOpCtx.get(), lastShutdownState, &startupTimeElapsedBuilder); + startup_recovery::repairAndRecoverDatabases(startupOpCtx.get(), lastShutdownState); } catch (const ExceptionFor<ErrorCodes::MustDowngrade>& error) { LOGV2_FATAL_OPTIONS( 20573, @@ -602,16 +545,10 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { } // Start up health log writer thread. - HealthLogInterface::set(serviceContext, std::make_unique<HealthLog>()); - HealthLogInterface::get(startupOpCtx.get())->startup(); + HealthLog::get(startupOpCtx.get()).startup(); auto const globalAuthzManager = AuthorizationManager::get(serviceContext); - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Build user and roles graph", - &startupTimeElapsedBuilder); - uassertStatusOK(globalAuthzManager->initialize(startupOpCtx.get())); - } + uassertStatusOK(globalAuthzManager->initialize(startupOpCtx.get())); if (audit::initializeManager) { audit::initializeManager(startupOpCtx.get()); @@ -621,7 +558,7 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { srand((unsigned)(curTimeMicros64()) ^ (unsigned(uintptr_t(&startupOpCtx)))); if (globalAuthzManager->shouldValidateAuthSchemaOnStartup()) { - Status status = verifySystemIndexes(startupOpCtx.get(), &startupTimeElapsedBuilder); + Status status = verifySystemIndexes(startupOpCtx.get()); if (!status.isOK()) { LOGV2_WARNING(20538, "Unable to verify system indexes: {error}", @@ -682,14 +619,7 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { "** This mode should only be used to manually repair corrupted auth data"); } - { - TimeElapsedBuilderScopedTimer scopedTimer( - serviceContext->getFastClockSource(), - "Set up the background thread pool responsible for " - "waiting for opTimes to be majority committed", - &startupTimeElapsedBuilder); - WaitForMajorityService::get(serviceContext).startup(serviceContext); - } + WaitForMajorityService::get(serviceContext).startup(serviceContext); // This function may take the global lock. auto shardingInitialized = ShardingInitializationMongoD::get(startupOpCtx.get()) @@ -715,7 +645,7 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { "Error loading read and write concern defaults at startup", "error"_attr = redact(ex)); } - readWriteConcernDefaultsMongodStartupChecks(startupOpCtx.get(), replSettings.usingReplSets()); + readWriteConcernDefaultsMongodStartupChecks(startupOpCtx.get()); // Perform replication recovery for queryable backup mode if needed. if (storageGlobalParams.readOnly) { @@ -738,10 +668,6 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { uassert(ErrorCodes::BadValue, str::stream() << "Cannot use queryableBackupMode in a replica set", !replCoord->isReplEnabled()); - TimeElapsedBuilderScopedTimer scopedTimer( - serviceContext->getFastClockSource(), - "Start up the replication coordinator for queryable backup mode", - &startupTimeElapsedBuilder); replCoord->startup(startupOpCtx.get(), lastShutdownState); } @@ -751,6 +677,8 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { logStartup(startupOpCtx.get()); } + startFreeMonitoring(serviceContext); + auto replCoord = repl::ReplicationCoordinator::get(startupOpCtx.get()); invariant(replCoord); @@ -762,11 +690,6 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { } } } else if (serverGlobalParams.clusterRole == ClusterRole::ConfigServer) { - TimeElapsedBuilderScopedTimer scopedTimer( - serviceContext->getFastClockSource(), - "Initialize the sharding components for a config server", - &startupTimeElapsedBuilder); - initializeGlobalShardingStateForMongoD( startupOpCtx.get(), ShardId::kConfigServerId, ConnectionString::forLocal()); @@ -775,22 +698,7 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { makeShardingTaskExecutor(executor::makeNetworkInterface("AddShard-TaskExecutor"))); Grid::get(startupOpCtx.get())->setShardingInitialized(); - } - - if (replSettings.usingReplSets() && - (serverGlobalParams.clusterRole == ClusterRole::None || - !Grid::get(startupOpCtx.get())->isShardingInitialized())) { - // If this is a mongod in a standalone replica set or a shardsvr replica set that has - // not initialized its sharding identity, start up the cluster time keys manager with a - // local/direct keys client. The keys client must use local read concern if the storage - // engine can't support majority read concern. If this is a mongod in a configsvr or - // shardsvr replica set that has initialized its sharding identity, the keys manager is - // by design initialized separately with a sharded keys client when the sharding state - // is initialized. - TimeElapsedBuilderScopedTimer scopedTimer( - serviceContext->getFastClockSource(), - "Start up cluster time keys manager with a local/direct keys client", - &startupTimeElapsedBuilder); + } else if (replSettings.usingReplSets()) { // standalone replica set auto keysCollectionClient = std::make_unique<KeysCollectionClientDirect>(); auto keyManager = std::make_shared<KeysCollectionManager>( KeysCollectionManager::kKeyManagerPurposeString, @@ -800,19 +708,11 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { LogicalTimeValidator::set(startupOpCtx->getServiceContext(), std::make_unique<LogicalTimeValidator>(keyManager)); - } - if (replSettings.usingReplSets() && serverGlobalParams.clusterRole == ClusterRole::None) { ReplicaSetNodeProcessInterface::getReplicaSetNodeExecutor(serviceContext)->startup(); } - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Start up the replication coordinator", - &startupTimeElapsedBuilder); - replCoord->startup(startupOpCtx.get(), lastShutdownState); - } - + replCoord->startup(startupOpCtx.get(), lastShutdownState); // 'getOldestActiveTimestamp', which is called in the background by the checkpoint thread, // requires a read on 'config.transactions' at the stableTimestamp. If this read occurs // while applying prepared transactions at the end of replication recovery, it's possible to @@ -843,9 +743,6 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { } if (replSettings.usingReplSets()) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Create an oplog view for tenant migrations", - &startupTimeElapsedBuilder); Lock::GlobalWrite lk(startupOpCtx.get()); OldClientContext ctx(startupOpCtx.get(), NamespaceString::kRsOplogNamespace.ns()); tenant_migration_util::createOplogViewForTenantMigrations(startupOpCtx.get(), ctx.db()); @@ -876,14 +773,6 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { LOGV2_WARNING(4747501, "Not starting periodic jobs as shutdown is in progress"); // Shutdown has already started before initialization is complete. Wait for the // shutdown task to complete and return. - - logStartupStats.dismiss(); - logMongodStartupTimeElapsedStatistics(serviceContext, - beginInitAndListen, - &startupTimeElapsedBuilder, - &startupInfoBuilder, - lastShutdownState); - MONGO_IDLE_THREAD_BLOCK; return waitForShutdown(); } @@ -926,9 +815,6 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { } if (!storageGlobalParams.repair) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Start transport layer", - &startupTimeElapsedBuilder); start = serviceContext->getTransportLayer()->start(); if (!start.isOK()) { LOGV2_ERROR(20572, @@ -955,13 +841,6 @@ ExitCode _initAndListen(ServiceContext* serviceContext, int listenPort) { exitCleanly(EXIT_CLEAN); } - logStartupStats.dismiss(); - logMongodStartupTimeElapsedStatistics(serviceContext, - beginInitAndListen, - &startupTimeElapsedBuilder, - &startupInfoBuilder, - lastShutdownState); - MONGO_IDLE_THREAD_BLOCK; return waitForShutdown(); } @@ -1136,6 +1015,7 @@ void setUpCollectionShardingState(ServiceContext* serviceContext) { void setUpCatalog(ServiceContext* serviceContext) { DatabaseHolder::set(serviceContext, std::make_unique<DatabaseHolderImpl>()); + IndexAccessMethodFactory::set(serviceContext, std::make_unique<IndexAccessMethodFactoryImpl>()); Collection::Factory::set(serviceContext, std::make_unique<CollectionImpl::FactoryImpl>()); } @@ -1251,6 +1131,8 @@ void setUpObservers(ServiceContext* serviceContext) { opObserverRegistry->addObserver(std::make_unique<ClusterServerParameterOpObserver>()); } + setupFreeMonitoringOpObserver(opObserverRegistry.get()); + if (audit::opObserverRegistrar) { audit::opObserverRegistrar(opObserverRegistry.get()); } @@ -1269,20 +1151,6 @@ MONGO_INITIALIZER_GENERAL(setSSLManagerType, (), ("SSLManager")) #define __has_feature(x) 0 #endif -void logShutdownTimeElapsedStatistics(ServiceContext* serviceContext, - Date_t beginShutdownTask, - BSONObjBuilder* shutdownTimeElapsedBuilder, - BSONObjBuilder* shutdownInfoBuilder) { - mongo::Milliseconds elapsedInitAndListen = - serviceContext->getFastClockSource()->now() - beginShutdownTask; - shutdownTimeElapsedBuilder->append("shutdownTask total elapsed time", - elapsedInitAndListen.toString()); - shutdownInfoBuilder->append("Statistics", shutdownTimeElapsedBuilder->obj()); - LOGV2_INFO(8423404, - "mongod shutdown complete", - "Summary of time elapsed"_attr = shutdownInfoBuilder->obj()); -} - // NOTE: This function may be called at any time after registerShutdownTask is called below. It // must not depend on the prior execution of mongo initializers or the existence of threads. void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { @@ -1307,40 +1175,20 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { hangBeforeShutdown.pauseWhileSet(); } - BSONObjBuilder shutdownTimeElapsedBuilder; - BSONObjBuilder shutdownInfoBuilder; - - Date_t beginShutdownTask = serviceContext->getFastClockSource()->now(); - ScopeGuard logShutdownStats( - [serviceContext, beginShutdownTask, &shutdownTimeElapsedBuilder, &shutdownInfoBuilder] { - logShutdownTimeElapsedStatistics(serviceContext, - beginShutdownTask, - &shutdownTimeElapsedBuilder, - &shutdownInfoBuilder); - }); - // If we don't have shutdownArgs, we're shutting down from a signal, or other clean shutdown // path. // // In that case, do a default step down, still shutting down if stepDown fails. if (auto replCoord = repl::ReplicationCoordinator::get(serviceContext); replCoord && !shutdownArgs.isUserInitiated) { - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Enter terminal shutdown", - &shutdownTimeElapsedBuilder); - replCoord->enterTerminalShutdown(); - } + replCoord->enterTerminalShutdown(); ServiceContext::UniqueOperationContext uniqueOpCtx; OperationContext* opCtx = client->getOperationContext(); if (!opCtx) { uniqueOpCtx = client->makeOperationContext(); opCtx = uniqueOpCtx.get(); } - TimeElapsedBuilderScopedTimer scopedTimer( - serviceContext->getFastClockSource(), - "Step down the replication coordinator for shutdown", - &shutdownTimeElapsedBuilder); + const auto forceShutdown = true; auto stepDownStartTime = opCtx->getServiceContext()->getPreciseClockSource()->now(); // stepDown should never return an error during force shutdown. @@ -1355,104 +1203,62 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { (opCtx->getServiceContext()->getPreciseClockSource()->now() - stepDownStartTime)); } - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Time spent in quiesce mode", - &shutdownTimeElapsedBuilder); - if (auto replCoord = repl::ReplicationCoordinator::get(serviceContext); - replCoord && replCoord->enterQuiesceModeIfSecondary(shutdownTimeout)) { - ServiceContext::UniqueOperationContext uniqueOpCtx; - OperationContext* opCtx = client->getOperationContext(); - if (!opCtx) { - uniqueOpCtx = client->makeOperationContext(); - opCtx = uniqueOpCtx.get(); - } - if (MONGO_unlikely(hangDuringQuiesceMode.shouldFail())) { - LOGV2_OPTIONS(4695101, - {LogComponent::kReplication}, - "hangDuringQuiesceMode failpoint enabled"); - hangDuringQuiesceMode.pauseWhileSet(opCtx); - } - - LOGV2_OPTIONS(4695102, - {LogComponent::kReplication}, - "Entering quiesce mode for shutdown", - "quiesceTime"_attr = shutdownTimeout); - opCtx->sleepFor(shutdownTimeout); + if (auto replCoord = repl::ReplicationCoordinator::get(serviceContext); + replCoord && replCoord->enterQuiesceModeIfSecondary(shutdownTimeout)) { + ServiceContext::UniqueOperationContext uniqueOpCtx; + OperationContext* opCtx = client->getOperationContext(); + if (!opCtx) { + uniqueOpCtx = client->makeOperationContext(); + opCtx = uniqueOpCtx.get(); + } + if (MONGO_unlikely(hangDuringQuiesceMode.shouldFail())) { LOGV2_OPTIONS( - 4695103, {LogComponent::kReplication}, "Exiting quiesce mode for shutdown"); + 4695101, {LogComponent::kReplication}, "hangDuringQuiesceMode failpoint enabled"); + hangDuringQuiesceMode.pauseWhileSet(opCtx); } - } - - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down FLE Crud subsystem", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS(6371601, {LogComponent::kDefault}, "Shutting down the FLE Crud thread pool"); - stopFLECrud(); + LOGV2_OPTIONS(4695102, + {LogComponent::kReplication}, + "Entering quiesce mode for shutdown", + "quiesceTime"_attr = shutdownTimeout); + opCtx->sleepFor(shutdownTimeout); + LOGV2_OPTIONS(4695103, {LogComponent::kReplication}, "Exiting quiesce mode for shutdown"); } - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down MirrorMaestro", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS(4784901, {LogComponent::kCommand}, "Shutting down the MirrorMaestro"); - MirrorMaestro::shutdown(serviceContext); - } + LOGV2_OPTIONS(6371601, {LogComponent::kDefault}, "Shutting down the FLE Crud thread pool"); + stopFLECrud(); - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down WaitForMajorityService", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS( - 4784902, {LogComponent::kSharding}, "Shutting down the WaitForMajorityService"); - WaitForMajorityService::get(serviceContext).shutDown(); - } + LOGV2_OPTIONS(4784901, {LogComponent::kCommand}, "Shutting down the MirrorMaestro"); + MirrorMaestro::shutdown(serviceContext); + + LOGV2_OPTIONS(4784902, {LogComponent::kSharding}, "Shutting down the WaitForMajorityService"); + WaitForMajorityService::get(serviceContext).shutDown(); // Join the logical session cache before the transport layer. if (auto lsc = LogicalSessionCache::get(serviceContext)) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the logical session cache", - &shutdownTimeElapsedBuilder); LOGV2(4784903, "Shutting down the LogicalSessionCache"); lsc->joinOnShutDown(); } // Shutdown the TransportLayer so that new connections aren't accepted if (auto tl = serviceContext->getTransportLayer()) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the transport layer", - &shutdownTimeElapsedBuilder); LOGV2_OPTIONS( 20562, {LogComponent::kNetwork}, "Shutdown: going to close listening sockets"); tl->shutdown(); } // Shut down the global dbclient pool so callers stop waiting for connections. - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the global connection pool", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS( - 4784905, {LogComponent::kNetwork}, "Shutting down the global connection pool"); - globalConnPool.shutdown(); - } + LOGV2_OPTIONS(4784905, {LogComponent::kNetwork}, "Shutting down the global connection pool"); + globalConnPool.shutdown(); // Inform Flow Control to stop gating writes on ticket admission. This must be done before the // Periodic Runner is shut down (see SERVER-41751). if (auto flowControlTicketholder = FlowControlTicketholder::get(serviceContext)) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the flow control ticket holder", - &shutdownTimeElapsedBuilder); LOGV2(4784906, "Shutting down the FlowControlTicketholder"); flowControlTicketholder->setInShutdown(); } if (auto exec = ReplicaSetNodeProcessInterface::getReplicaSetNodeExecutor(serviceContext)) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the replica set node executor", - &shutdownTimeElapsedBuilder); LOGV2_OPTIONS( 4784907, {LogComponent::kReplication}, "Shutting down the replica set node executor"); exec->shutdown(); @@ -1471,23 +1277,16 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { uniqueOpCtx = client->makeOperationContext(); opCtx = uniqueOpCtx.get(); } - { - stdx::lock_guard lg(*client); - opCtx->setIsExecutingShutdown(); - } + opCtx->setIsExecutingShutdown(); // This can wait a long time while we drain the secondary's apply queue, especially if // it is building an index. LOGV2_OPTIONS( 4784909, {LogComponent::kReplication}, "Shutting down the ReplicationCoordinator"); - repl::ReplicationCoordinator::get(serviceContext) - ->shutdown(opCtx, &shutdownTimeElapsedBuilder); + repl::ReplicationCoordinator::get(serviceContext)->shutdown(opCtx); // Terminate the index consistency check. if (serverGlobalParams.clusterRole == ClusterRole::ConfigServer) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the index consistency checker", - &shutdownTimeElapsedBuilder); LOGV2_OPTIONS(4784904, {LogComponent::kSharding}, "Shutting down the PeriodicShardedIndexConsistencyChecker"); @@ -1512,53 +1311,32 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { // marked as killed and will not be usable other than to kill all transactions directly // below. LOGV2_OPTIONS(4784912, {LogComponent::kDefault}, "Killing all operations for shutdown"); - { - const std::set<std::string> excludedClients = {std::string(kFTDCThreadName)}; - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Kill all operations for shutdown", - &shutdownTimeElapsedBuilder); - serviceContext->setKillAllOperations(excludedClients); - - if (MONGO_unlikely(pauseWhileKillingOperationsAtShutdown.shouldFail())) { - LOGV2_OPTIONS(4701700, - {LogComponent::kDefault}, - "pauseWhileKillingOperationsAtShutdown failpoint enabled"); - sleepsecs(1); - } - } - - { - // Clear tenant migration access blockers after killing all operation contexts to ensure - // that no operation context cancellation token continuation holds the last reference to - // the TenantMigrationAccessBlockerExecutor. - TimeElapsedBuilderScopedTimer scopedTimer( - serviceContext->getFastClockSource(), - "Shut down all tenant migration access blockers on global shutdown", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS(5093807, - {LogComponent::kTenantMigration}, - "Shutting down all TenantMigrationAccessBlockers on global shutdown"); - TenantMigrationAccessBlockerRegistry::get(serviceContext).shutDown(); + const std::set<std::string> excludedClients = {std::string(kFTDCThreadName)}; + serviceContext->setKillAllOperations(excludedClients); + + // Clear tenant migration access blockers after killing all operation contexts to ensure + // that no operation context cancellation token continuation holds the last reference to the + // TenantMigrationAccessBlockerExecutor. + LOGV2_OPTIONS(5093807, + {LogComponent::kTenantMigration}, + "Shutting down all TenantMigrationAccessBlockers on global shutdown"); + TenantMigrationAccessBlockerRegistry::get(serviceContext).shutDown(); + + if (MONGO_unlikely(pauseWhileKillingOperationsAtShutdown.shouldFail())) { + LOGV2_OPTIONS(4701700, + {LogComponent::kDefault}, + "pauseWhileKillingOperationsAtShutdown failpoint enabled"); + sleepsecs(1); } // Destroy all stashed transaction resources, in order to release locks. - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down all open transactions", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS(4784913, {LogComponent::kCommand}, "Shutting down all open transactions"); - killSessionsLocalShutdownAllTransactions(opCtx); - } + LOGV2_OPTIONS(4784913, {LogComponent::kCommand}, "Shutting down all open transactions"); + killSessionsLocalShutdownAllTransactions(opCtx); - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Acquire the RSTL for shutdown", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS(4784914, - {LogComponent::kReplication}, - "Acquiring the ReplicationStateTransitionLock for shutdown"); - rstl.waitForLockUntil(Date_t::max()); - } + LOGV2_OPTIONS(4784914, + {LogComponent::kReplication}, + "Acquiring the ReplicationStateTransitionLock for shutdown"); + rstl.waitForLockUntil(Date_t::max()); // Release the rstl before waiting for the index build threads to join as index build // reacquires rstl in uninterruptible lock guard to finish their cleanup process. @@ -1566,83 +1344,38 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { // Shuts down the thread pool and waits for index builds to finish. // Depends on setKillAllOperations() above to interrupt the index build operations. - { - TimeElapsedBuilderScopedTimer scopedTimer( - serviceContext->getFastClockSource(), - "Shut down the IndexBuildsCoordinator and wait for index builds to finish", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS( - 4784915, {LogComponent::kIndex}, "Shutting down the IndexBuildsCoordinator"); - IndexBuildsCoordinator::get(serviceContext)->shutdown(opCtx); - } + LOGV2_OPTIONS(4784915, {LogComponent::kIndex}, "Shutting down the IndexBuildsCoordinator"); + IndexBuildsCoordinator::get(serviceContext)->shutdown(opCtx); } - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the replica set monitor", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS(4784918, {LogComponent::kNetwork}, "Shutting down the ReplicaSetMonitor"); - ReplicaSetMonitor::shutdown(); - } + LOGV2_OPTIONS(4784918, {LogComponent::kNetwork}, "Shutting down the ReplicaSetMonitor"); + ReplicaSetMonitor::shutdown(); if (auto sr = Grid::get(serviceContext)->shardRegistry()) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the shard registry", - &shutdownTimeElapsedBuilder); LOGV2_OPTIONS(4784919, {LogComponent::kSharding}, "Shutting down the shard registry"); sr->shutdown(); } if (ShardingState::get(serviceContext)->enabled()) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the transaction coordinator service", - &shutdownTimeElapsedBuilder); TransactionCoordinatorService::get(serviceContext)->shutdown(); } // Validator shutdown must be called after setKillAllOperations is called. Otherwise, this can // deadlock. if (auto validator = LogicalTimeValidator::get(serviceContext)) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the logical time validator", - &shutdownTimeElapsedBuilder); LOGV2_OPTIONS( 4784920, {LogComponent::kReplication}, "Shutting down the LogicalTimeValidator"); validator->shutDown(); } - if (TestingProctor::instance().isEnabled()) { - if (auto pool = Grid::get(serviceContext)->getExecutorPool()) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the executor pool", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS(6773200, {LogComponent::kSharding}, "Shutting down the ExecutorPool"); - pool->shutdownAndJoin(); - } - } - // The migrationutil executor must be shut down before shutting down the CatalogCacheLoader. // Otherwise, it may try to schedule work on the CatalogCacheLoader and fail. LOGV2_OPTIONS(4784921, {LogComponent::kSharding}, "Shutting down the MigrationUtilExecutor"); auto migrationUtilExecutor = migrationutil::getMigrationUtilExecutor(serviceContext); - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the migration util executor", - &shutdownTimeElapsedBuilder); - migrationUtilExecutor->shutdown(); - migrationUtilExecutor->join(); - } - - if (Grid::get(serviceContext)->isShardingInitialized()) { - // The CatalogCache must be shuted down before shutting down the CatalogCacheLoader as the - // CatalogCache may try to schedule work on CatalogCacheLoader and fail. - TimeElapsedBuilderScopedTimer scopedTimer( - serviceContext->getFastClockSource(), - "Shut down the catalog cache and catalog cache loader", - &shutdownTimeElapsedBuilder); - LOGV2_OPTIONS(6773201, {LogComponent::kSharding}, "Shutting down the CatalogCache"); - Grid::get(serviceContext)->catalogCache()->shutDownAndJoin(); + migrationUtilExecutor->shutdown(); + migrationUtilExecutor->join(); + if (ShardingState::get(serviceContext)->enabled()) { LOGV2_OPTIONS(4784922, {LogComponent::kSharding}, "Shutting down the CatalogCacheLoader"); CatalogCacheLoader::get(serviceContext).shutDown(); } @@ -1650,36 +1383,24 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { // Shutdown the Service Entry Point and its sessions and give it a grace period to complete. if (auto sep = serviceContext->getServiceEntryPoint()) { LOGV2_OPTIONS(4784923, {LogComponent::kCommand}, "Shutting down the ServiceEntryPoint"); - if (!sep->shutdown(Seconds(30))) { + if (!sep->shutdown(Seconds(10))) { LOGV2_OPTIONS(20563, {LogComponent::kNetwork}, "Service entry point did not shutdown within the time limit"); } } - if (auto* healthLog = HealthLogInterface::get(serviceContext)) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the health log", - &shutdownTimeElapsedBuilder); - LOGV2(4784927, "Shutting down the HealthLog"); - healthLog->shutdown(); - } + LOGV2(4784925, "Shutting down free monitoring"); + stopFreeMonitoring(); - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the TTL monitor", - &shutdownTimeElapsedBuilder); - LOGV2(4784928, "Shutting down the TTL monitor"); - shutdownTTLMonitor(serviceContext); - } + LOGV2(4784927, "Shutting down the HealthLog"); + HealthLog::get(serviceContext).shutdown(); - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down expired pre-images remover", - &shutdownTimeElapsedBuilder); - LOGV2(6278511, "Shutting down the Change Stream Expired Pre-images Remover"); - shutdownChangeStreamExpiredPreImagesRemover(serviceContext); - } + LOGV2(4784928, "Shutting down the TTL monitor"); + shutdownTTLMonitor(serviceContext); + + LOGV2(6278511, "Shutting down the Change Stream Expired Pre-images Remover"); + shutdownChangeStreamExpiredPreImagesRemover(serviceContext); // We should always be able to acquire the global lock at shutdown. // An OperationContext is not necessary to call lockGlobal() during shutdown, as it's only used @@ -1694,9 +1415,6 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { // Global storage engine may not be started in all cases before we exit if (serviceContext->getStorageEngine()) { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down the storage engine", - &shutdownTimeElapsedBuilder); LOGV2(4784930, "Shutting down the storage engine"); shutdownGlobalStorageEngineCleanly(serviceContext); } @@ -1708,12 +1426,7 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { ScriptEngine::dropScopeCache(); // Shutdown Full-Time Data Capture - { - TimeElapsedBuilderScopedTimer scopedTimer(serviceContext->getFastClockSource(), - "Shut down full-time data capture", - &shutdownTimeElapsedBuilder); - stopMongoDFTDC(); - } + stopMongoDFTDC(); LOGV2(20565, "Now exiting"); @@ -1735,18 +1448,6 @@ void shutdownTask(const ShutdownTaskArgs& shutdownArgs) { #endif } -void disableMongodTHPUnderTestingEnvironment() { -#ifdef __linux__ - if (TestingProctor::instance().isEnabled()) { - if (prctl(PR_SET_THP_DISABLE, 1, 0, 0, 0) == -1) { - LOGV2_WARNING(8751800, "Could not disable THP on mongod"); - } else { - LOGV2_INFO(8751801, "Successfully disabled THP on mongod"); - } - } -#endif -} - } // namespace int mongod_main(int argc, char* argv[]) { @@ -1769,8 +1470,6 @@ int mongod_main(int argc, char* argv[]) { quickExit(EXIT_FAILURE); } - disableMongodTHPUnderTestingEnvironment(); - auto* service = [] { try { auto serviceContextHolder = ServiceContext::make(); |