/** * Copyright (C) 2022-present MongoDB, Inc. * * This program is free software: you can redistribute it and/or modify * it under the terms of the Server Side Public License, version 1, * as published by MongoDB, Inc. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * Server Side Public License for more details. * * You should have received a copy of the Server Side Public License * along with this program. If not, see * . * * As a special exception, the copyright holders give permission to link the * code of portions of this program with the OpenSSL library under certain * conditions as described in each individual source file and distribute * linked combinations including the program with the OpenSSL library. You * must comply with the Server Side Public License in all respects for * all of the code used other than as permitted herein. If you modify file(s) * with this exception, you may extend this exception to your version of the * file(s), but you are not obligated to do so. If you do not wish to do so, * delete this exception statement from your version. If you delete this * exception statement from all source files in the program, then also delete * it in the license file. */ #include "mongo/db/shard_role.h" #include #include #include #include "mongo/db/catalog/catalog_helper.h" #include "mongo/db/catalog/collection_catalog.h" #include "mongo/db/catalog/collection_uuid_mismatch.h" #include "mongo/db/catalog/collection_uuid_mismatch_info.h" #include "mongo/db/concurrency/exception_util.h" #include "mongo/db/curop.h" #include "mongo/db/db_raii.h" #include "mongo/db/repl/read_concern_args.h" #include "mongo/db/s/collection_sharding_runtime.h" #include "mongo/db/storage/capped_snapshots.h" #include "mongo/logv2/log.h" #include "mongo/util/decorable.h" #define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kSharding namespace mongo { namespace { auto getTransactionResources = OperationContext::declareDecoration< std::unique_ptr>(); shard_role_details::TransactionResources& getOrMakeTransactionResources(OperationContext* opCtx) { auto& optTransactionResources = getTransactionResources(opCtx); if (!optTransactionResources) { optTransactionResources = std::make_unique(); } return *optTransactionResources; } struct ResolvedNamespaceOrViewAcquisitionRequest { // Populated in the first phase of collection(s) acquisition AcquisitionPrerequisites prerequisites; // Populated optionally in the second phase of collection(s) acquisition std::shared_ptr dbLock; boost::optional collLock; }; using ResolvedNamespaceOrViewAcquisitionRequestsMap = std::map; /** * Takes the input acquisitions, populates the NSS and UUID parts and returns a list, sorted by NSS, * suitable for a defined lock acquisition order. */ ResolvedNamespaceOrViewAcquisitionRequestsMap resolveNamespaceOrViewAcquisitionRequests( OperationContext* opCtx, const CollectionCatalog& catalog, const std::vector& acquisitionRequests) { ResolvedNamespaceOrViewAcquisitionRequestsMap sortedAcquisitionRequests; for (const auto& ar : acquisitionRequests) { if (ar.nss) { uassert(ErrorCodes::InvalidNamespace, str::stream() << "Namespace " << ar.nss->toStringForErrorMsg() << "is not a valid collection name", ar.nss->isValid()); auto coll = catalog.lookupCollectionByNamespace(opCtx, *ar.nss); checkCollectionUUIDMismatch(opCtx, *ar.nss, coll, ar.uuid); AcquisitionPrerequisites prerequisites(*ar.nss, ar.uuid, ar.readConcern, ar.placementConcern, ar.operationType, ar.viewMode); ResolvedNamespaceOrViewAcquisitionRequest resolvedAcquisitionRequest{ prerequisites, nullptr, boost::none}; sortedAcquisitionRequests.emplace(ResourceId(RESOURCE_COLLECTION, *ar.nss), std::move(resolvedAcquisitionRequest)); } else if (ar.dbname) { invariant(ar.uuid); uassert(ErrorCodes::InvalidNamespace, str::stream() << "Invalid db name " << ar.dbname->toStringForErrorMsg(), NamespaceString::validDBName(*ar.dbname, NamespaceString::DollarInDbNameBehavior::Allow)); auto coll = catalog.lookupCollectionByUUID(opCtx, *ar.uuid); uassert(ErrorCodes::NamespaceNotFound, str::stream() << "Namespace " << (*ar.dbname).toStringForErrorMsg() << ":" << *ar.uuid << " not found", coll); uassert(ErrorCodes::NamespaceNotFound, str::stream() << "Database name mismatch for " << (*ar.dbname).toStringForErrorMsg() << ":" << *ar.uuid << ". Expected: " << (*ar.dbname).toStringForErrorMsg() << " Actual: " << coll->ns().dbName().toStringForErrorMsg(), coll->ns().dbName() == *ar.dbname); if (ar.nss) { checkCollectionUUIDMismatch(opCtx, *ar.nss, coll, *ar.uuid); } AcquisitionPrerequisites prerequisites(coll->ns(), coll->uuid(), ar.readConcern, ar.placementConcern, ar.operationType, ar.viewMode); ResolvedNamespaceOrViewAcquisitionRequest resolvedAcquisitionRequest{ prerequisites, nullptr, boost::none}; sortedAcquisitionRequests.emplace(ResourceId(RESOURCE_COLLECTION, coll->ns()), std::move(resolvedAcquisitionRequest)); } else { MONGO_UNREACHABLE; } } return sortedAcquisitionRequests; } void verifyDbAndCollection(OperationContext* opCtx, const NamespaceString& nss, CollectionPtr& coll, AcquisitionPrerequisites::OperationType operationType) { invariant(coll); // In most cases we expect modifications for system.views to upgrade MODE_IX to MODE_X // before taking the lock. One exception is a query by UUID of system.views in a // transaction. Usual queries of system.views (by name, not UUID) within a transaction are // rejected. However, if the query is by UUID we can't determine whether the namespace is // actually system.views until we take the lock here. So we have this one last assertion. uassert(6944500, "Modifications to system.views must take an exclusive lock", !nss.isSystemDotViews() || opCtx->lockState()->isCollectionLockedForMode(nss, MODE_X)); // Verify that we are using the latest instance if we intend to perform writes. if (operationType == AcquisitionPrerequisites::OperationType::kWrite) { auto latest = CollectionCatalog::latest(opCtx); if (!latest->containsCollection(opCtx, coll.get())) { throwWriteConflictException(str::stream() << "Unable to write to collection '" << coll->ns().toStringForErrorMsg() << "' due to catalog changes; please " "retry the operation"); } if (opCtx->recoveryUnit()->isActive()) { const auto mySnapshot = opCtx->recoveryUnit()->getPointInTimeReadTimestamp(opCtx); if (mySnapshot && *mySnapshot < coll->getMinimumValidSnapshot()) { throwWriteConflictException(str::stream() << "Unable to write to collection '" << coll->ns().toStringForErrorMsg() << "' due to snapshot timestamp " << *mySnapshot << " being older than collection minimum " << *coll->getMinimumValidSnapshot() << "; please retry the operation"); } } } } void checkPlacementVersion(OperationContext* opCtx, const NamespaceString& nss, const PlacementConcern& placementConcern) { const auto& receivedDbVersion = placementConcern.dbVersion; if (receivedDbVersion) { DatabaseShardingState::assertMatchingDbVersion(opCtx, nss.dbName(), *receivedDbVersion); } const auto& receivedShardVersion = placementConcern.shardVersion; if (receivedShardVersion) { const auto scopedCSS = CollectionShardingState::acquire(opCtx, nss); scopedCSS->checkShardVersionOrThrow(opCtx, *receivedShardVersion); } } std::variant> acquireLocalCollectionOrView( OperationContext* opCtx, const CollectionCatalog& catalog, const AcquisitionPrerequisites& prerequisites) { const auto& nss = prerequisites.nss; // TODO SERVER-76415 fix secondary reads if shouldReadAtLastApplied auto readTimestamp = opCtx->recoveryUnit()->getPointInTimeReadTimestamp(opCtx); auto coll = CollectionPtr(catalog.establishConsistentCollection( opCtx, NamespaceStringOrUUID(prerequisites.nss), readTimestamp)); checkCollectionUUIDMismatch(opCtx, nss, coll, prerequisites.uuid); if (coll) { verifyDbAndCollection(opCtx, nss, coll, prerequisites.operationType); return coll; } else if (auto view = catalog.lookupView(opCtx, nss)) { uassert(ErrorCodes::CommandNotSupportedOnView, str::stream() << "Namespace " << nss.toStringForErrorMsg() << " is a view, not a collection", prerequisites.viewMode == AcquisitionPrerequisites::kCanBeView); return view; } else { return CollectionPtr(); } } struct SnapshotedServices { std::variant> collectionPtrOrView; boost::optional collectionDescription; boost::optional ownershipFilter; }; SnapshotedServices acquireServicesSnapshot(OperationContext* opCtx, const CollectionCatalog& catalog, const AcquisitionPrerequisites& prerequisites) { if (stdx::holds_alternative( prerequisites.placementConcern)) { return SnapshotedServices{ acquireLocalCollectionOrView(opCtx, catalog, prerequisites), boost::none, boost::none}; } const auto& placementConcern = stdx::get(prerequisites.placementConcern); auto collOrView = acquireLocalCollectionOrView(opCtx, catalog, prerequisites); const auto& nss = prerequisites.nss; const bool isPlacementConcernVersioned = placementConcern.dbVersion || placementConcern.shardVersion; const auto scopedCSS = CollectionShardingState::acquire(opCtx, nss); auto collectionDescription = scopedCSS->getCollectionDescription(opCtx, isPlacementConcernVersioned); invariant(!collectionDescription.isSharded() || placementConcern.shardVersion); auto optOwnershipFilter = collectionDescription.isSharded() ? boost::optional(scopedCSS->getOwnershipFilter( opCtx, prerequisites.operationType == AcquisitionPrerequisites::OperationType::kRead ? CollectionShardingState::OrphanCleanupPolicy::kDisallowOrphanCleanup : CollectionShardingState::OrphanCleanupPolicy::kAllowOrphanCleanup, *placementConcern.shardVersion)) : boost::none; // TODO: This will be removed when we no longer snapshot sharding state on CollectionPtr. if (std::holds_alternative(collOrView) && collectionDescription.isSharded()) { std::get(collOrView) .setShardKeyPattern(collectionDescription.getKeyPattern()); } return SnapshotedServices{ std::move(collOrView), std::move(collectionDescription), std::move(optOwnershipFilter)}; } std::vector acquireResolvedCollectionsOrViewsWithoutTakingLocks( OperationContext* opCtx, const CollectionCatalog& catalog, ResolvedNamespaceOrViewAcquisitionRequestsMap sortedAcquisitionRequests) { std::vector acquisitions; for (auto& acquisitionRequest : sortedAcquisitionRequests) { tassert(7328900, "Cannot acquire for write without locks", acquisitionRequest.second.prerequisites.operationType == AcquisitionPrerequisites::kRead || acquisitionRequest.second.collLock); auto& prerequisites = acquisitionRequest.second.prerequisites; auto snapshotedServices = acquireServicesSnapshot(opCtx, catalog, prerequisites); const bool isCollection = std::holds_alternative(snapshotedServices.collectionPtrOrView); if (isCollection) { shard_role_details::AcquiredCollection& acquiredCollection = getOrMakeTransactionResources(opCtx).addAcquiredCollection( {prerequisites, std::move(acquisitionRequest.second.dbLock), std::move(acquisitionRequest.second.collLock), std::move(snapshotedServices.collectionDescription), std::move(snapshotedServices.ownershipFilter), std::move(std::get(snapshotedServices.collectionPtrOrView))}); ScopedCollectionAcquisition scopedAcquisition(opCtx, acquiredCollection); acquisitions.emplace_back(std::move(scopedAcquisition)); } else { // It's a view. const shard_role_details::AcquiredView& acquiredView = getOrMakeTransactionResources(opCtx).addAcquiredView( {prerequisites, std::move(acquisitionRequest.second.dbLock), std::move(acquisitionRequest.second.collLock), std::move(std::get>( snapshotedServices.collectionPtrOrView))}); ScopedViewAcquisition scopedAcquisition(opCtx, acquiredView); acquisitions.emplace_back(std::move(scopedAcquisition)); } } return acquisitions; } /* * Establish a capped snapshot if necessary on the provided namespace. */ void establishCappedSnapshotIfNeeded(OperationContext* opCtx, const std::shared_ptr& catalog, const NamespaceStringOrUUID& nsOrUUID) { auto coll = catalog->lookupCollectionByNamespaceOrUUID(opCtx, nsOrUUID); if (coll && coll->usesCappedSnapshots()) { CappedSnapshots::get(opCtx).establish(opCtx, coll); } } bool haveAcquiredConsistentCatalogAndSnapshot(const CollectionCatalog* catalogBeforeSnapshot, const CollectionCatalog* catalogAfterSnapshot, long long replTermBeforeSnapshot, long long replTermAfterSnapshot) { return catalogBeforeSnapshot == catalogAfterSnapshot && replTermBeforeSnapshot == replTermAfterSnapshot; } std::shared_ptr getConsistentCatalogAndSnapshot( OperationContext* opCtx, const std::vector& acquisitionRequests) { while (true) { shard_role_details::SnapshotAttempt snapshotAttempt(opCtx, acquisitionRequests); snapshotAttempt.snapshotInitialState(); snapshotAttempt.openStorageSnapshot(); if (auto catalog = snapshotAttempt.getConsistentCatalog()) { return catalog; } } } std::vector toNamespaceStringOrUUIDs( const std::list& acquiredCollections) { std::vector requests; for (const auto& acquiredCollection : acquiredCollections) { const auto& prerequisites = acquiredCollection.prerequisites; requests.emplace_back(prerequisites.nss); } return requests; } std::vector toNamespaceStringOrUUIDs( const std::vector& acquisitionRequests) { std::vector requests; for (const auto& ar : acquisitionRequests) { if (ar.nss) { requests.emplace_back(*ar.nss); } else { requests.emplace_back(*ar.dbname, *ar.uuid); } } return requests; } void checkShardingPlacement( OperationContext* opCtx, const std::vector& acquisitionRequests) { for (const auto& ar : acquisitionRequests) { // We only have to check placement for collections that come from a router, which // will have the namespace set. if (ar.nss) { checkPlacementVersion(opCtx, *ar.nss, ar.placementConcern); } } } std::shared_ptr stashConsistentCatalog( OperationContext* opCtx, const std::vector& acquisitionRequests) { auto requests = toNamespaceStringOrUUIDs(acquisitionRequests); auto catalog = getConsistentCatalogAndSnapshot(opCtx, requests); // Stash the catalog, it will be automatically unstashed when the snapshot is released. CollectionCatalog::stash(opCtx, catalog); return catalog; } } // namespace CollectionOrViewAcquisitionRequest CollectionOrViewAcquisitionRequest::fromOpCtx( OperationContext* opCtx, NamespaceString nss, AcquisitionPrerequisites::OperationType operationType, AcquisitionPrerequisites::ViewMode viewMode, boost::optional expectedUUID) { auto& oss = OperationShardingState::get(opCtx); auto& readConcern = repl::ReadConcernArgs::get(opCtx); return CollectionOrViewAcquisitionRequest( nss, expectedUUID, {oss.getDbVersion(nss.db()), oss.getShardVersion(nss)}, readConcern, operationType, viewMode); } CollectionAcquisitionRequest CollectionAcquisitionRequest::fromOpCtx( OperationContext* opCtx, NamespaceString nss, AcquisitionPrerequisites::OperationType operationType, boost::optional expectedUUID) { auto& oss = OperationShardingState::get(opCtx); auto& readConcern = repl::ReadConcernArgs::get(opCtx); return CollectionAcquisitionRequest(nss, expectedUUID, {oss.getDbVersion(nss.db()), oss.getShardVersion(nss)}, readConcern, operationType); } CollectionAcquisitionRequest CollectionAcquisitionRequest::fromOpCtx( OperationContext* opCtx, NamespaceStringOrUUID nssOrUUID, AcquisitionPrerequisites::OperationType operationType) { auto& oss = OperationShardingState::get(opCtx); auto& readConcern = repl::ReadConcernArgs::get(opCtx); // Acquisitions by uuid cannot possibly have a corresponding ShardVersion attached. PlacementConcern placementConcern = nssOrUUID.nss() ? PlacementConcern{oss.getDbVersion(nssOrUUID.dbName().db()), oss.getShardVersion(*nssOrUUID.nss())} : PlacementConcern{oss.getDbVersion(nssOrUUID.dbName().db()), {}}; return CollectionAcquisitionRequest(nssOrUUID, placementConcern, readConcern, operationType); } UUID ScopedCollectionAcquisition::uuid() const { invariant(exists(), str::stream() << "Collection " << nss().toStringForErrorMsg() << " doesn't exist, so its UUID cannot be obtained"); return _acquiredCollection.collectionPtr->uuid(); } const ScopedCollectionDescription& ScopedCollectionAcquisition::getShardingDescription() const { // The collectionDescription will only not be set if the caller as acquired the acquisition // using the kLocalCatalogOnlyWithPotentialDataLoss placement concern invariant(_acquiredCollection.collectionDescription); return *_acquiredCollection.collectionDescription; } const boost::optional& ScopedCollectionAcquisition::getShardingFilter() const { // The collectionDescription will only not be set if the caller as acquired the acquisition // using the kLocalCatalogOnlyWithPotentialDataLoss placement concern invariant(_acquiredCollection.collectionDescription); return _acquiredCollection.ownershipFilter; } const CollectionPtr& ScopedCollectionAcquisition::getCollectionPtr() const { tassert(ErrorCodes::InternalError, "Collection acquisition has been invalidated", !_acquiredCollection.invalidated); return _acquiredCollection.collectionPtr; } ScopedCollectionAcquisition::~ScopedCollectionAcquisition() { if (_opCtx) { const auto& transactionResources = getTransactionResources(_opCtx); if (transactionResources) { transactionResources->acquiredCollections.remove_if( [this](const shard_role_details::AcquiredCollection& txnResourceAcquiredColl) { return &txnResourceAcquiredColl == &(this->_acquiredCollection); }); if (transactionResources->acquiredCollections.empty() && transactionResources->acquiredViews.empty()) { transactionResources->releaseAllResourcesOnCommitOrAbort(); } } } } ScopedViewAcquisition::~ScopedViewAcquisition() { if (_opCtx) { const auto& transactionResources = getTransactionResources(_opCtx); if (transactionResources) { transactionResources->acquiredViews.remove_if( [this](const shard_role_details::AcquiredView& txnResourceAcquiredView) { return &txnResourceAcquiredView == &(this->_acquiredView); }); if (transactionResources->acquiredCollections.empty() && transactionResources->acquiredViews.empty()) { transactionResources->releaseAllResourcesOnCommitOrAbort(); } } } } ScopedCollectionAcquisition acquireCollection(OperationContext* opCtx, CollectionAcquisitionRequest acquisitionRequest, LockMode mode) { return std::get( acquireCollectionOrView(opCtx, acquisitionRequest, mode)); } std::vector acquireCollections( OperationContext* opCtx, std::vector acquisitionRequests, LockMode mode) { // Transform the CollectionAcquisitionRequests to NamespaceOrViewAcquisitionRequests. std::vector namespaceOrViewAcquisitionRequests; std::move(acquisitionRequests.begin(), acquisitionRequests.end(), std::back_inserter(namespaceOrViewAcquisitionRequests)); // Acquire the collections auto acquisitions = acquireCollectionsOrViews(opCtx, namespaceOrViewAcquisitionRequests, mode); // Transform the acquisitions to ScopedCollectionAcquisitions std::vector collectionAcquisitions; for (auto& acquisition : acquisitions) { // It must be a collection, because that's what the acquisition request stated. invariant(std::holds_alternative(acquisition)); collectionAcquisitions.emplace_back( std::move(std::get(acquisition))); } return collectionAcquisitions; } ScopedCollectionOrViewAcquisition acquireCollectionOrView( OperationContext* opCtx, CollectionOrViewAcquisitionRequest acquisitionRequest, LockMode mode) { auto acquisition = acquireCollectionsOrViews(opCtx, {std::move(acquisitionRequest)}, mode); invariant(acquisition.size() == 1); return std::move(acquisition.front()); } namespace shard_role_details { void SnapshotAttempt::snapshotInitialState() { // The read source used can change depending on replication state, so we must fetch the repl // state beforehand, to compare with afterwards. _replTermBeforeSnapshot = repl::ReplicationCoordinator::get(_opCtx)->getTerm(); _catalogBeforeSnapshot = CollectionCatalog::get(_opCtx); } void SnapshotAttempt::openStorageSnapshot() { invariant(_catalogBeforeSnapshot && _replTermBeforeSnapshot); if (_opCtx->recoveryUnit()->isActive()) { // We already have an open storage snapshot, nothing to do. return; } for (auto& nssOrUUID : _acquisitionRequests) { establishCappedSnapshotIfNeeded(_opCtx, *_catalogBeforeSnapshot, nssOrUUID); } // TODO SERVER-76415 call preallocateSnapshotForOplogRead() when reading from the oplog. _opCtx->recoveryUnit()->preallocateSnapshot(); _openedSnapshot = true; } std::shared_ptr SnapshotAttempt::getConsistentCatalog() { invariant(_catalogBeforeSnapshot && _replTermBeforeSnapshot && _opCtx->recoveryUnit()->isActive()); auto catalogAfterSnapshot = CollectionCatalog::get(_opCtx); const auto replTermAfterSnapshot = repl::ReplicationCoordinator::get(_opCtx)->getTerm(); if (!haveAcquiredConsistentCatalogAndSnapshot(_catalogBeforeSnapshot->get(), catalogAfterSnapshot.get(), *_replTermBeforeSnapshot, replTermAfterSnapshot)) { return nullptr; } _successful = true; return catalogAfterSnapshot; } SnapshotAttempt::~SnapshotAttempt() { if (_successful) { // We were successful, nothing to clean up. return; } // We can't abandon the snapshot if we are in a WUOW. if (_openedSnapshot && !_opCtx->lockState()->inAWriteUnitOfWork()) { _opCtx->recoveryUnit()->abandonSnapshot(); } CurOp::get(_opCtx)->yielded(); } ResolvedNamespaceOrViewAcquisitionRequestsMap generateSortedAcquisitionRequests( OperationContext* opCtx, const CollectionCatalog& catalog, const std::vector& acquisitionRequests) { ResolvedNamespaceOrViewAcquisitionRequestsMap sortedAcquisitionRequests; auto readTimestamp = opCtx->recoveryUnit()->getPointInTimeReadTimestamp(opCtx); int counter = 0; for (const auto& ar : acquisitionRequests) { NamespaceStringOrUUID nssOrUUID = ar.nss ? NamespaceStringOrUUID(*ar.nss) : NamespaceStringOrUUID(*ar.dbname, *ar.uuid); uassertStatusOK(nssOrUUID.isNssValid()); uassert(ErrorCodes::InvalidNamespace, str::stream() << "Invalid db name " << ar.dbname->toStringForErrorMsg(), NamespaceString::validDBName(nssOrUUID.dbName(), NamespaceString::DollarInDbNameBehavior::Allow)); auto coll = CollectionPtr(catalog.establishConsistentCollection(opCtx, nssOrUUID, readTimestamp)); // If we are acquiring by UUID and no collection matches, raise NamespaceNotFound. if (!ar.nss) { uassert(ErrorCodes::NamespaceNotFound, str::stream() << "Namespace " << (*ar.dbname).toStringForErrorMsg() << ":" << *ar.uuid << " not found", coll); uassert(ErrorCodes::NamespaceNotFound, str::stream() << "Database name mismatch for " << (*ar.dbname).toStringForErrorMsg() << ":" << *ar.uuid << ". Expected: " << (*ar.dbname).toStringForErrorMsg() << " Actual: " << coll->ns().dbName().toStringForErrorMsg(), coll->ns().dbName() == *ar.dbname); } const auto& nss = ar.nss ? *ar.nss : coll->ns(); AcquisitionPrerequisites prerequisites( nss, ar.uuid, ar.readConcern, ar.placementConcern, ar.operationType, ar.viewMode); ResolvedNamespaceOrViewAcquisitionRequest resolvedAcquisitionRequest{ prerequisites, nullptr, boost::none}; // We don't care about ordering in this case, use a mock ResourceId as the key. sortedAcquisitionRequests.emplace(ResourceId(RESOURCE_COLLECTION, counter++), std::move(resolvedAcquisitionRequest)); } return sortedAcquisitionRequests; } } // namespace shard_role_details std::vector acquireCollectionsOrViews( OperationContext* opCtx, std::vector acquisitionRequests, LockMode mode) { if (acquisitionRequests.size() == 0) { return {}; } // We shouldn't have an open snapshot unless we are in a multi document transaction or a // previous acquisition opened and stashed it already. // TODO enable invariant when everything uses acquisitions. bool inMultiDocumentTransaction = opCtx->inMultiDocumentTransaction(); while (true) { // Optimistically populate the nss and uuid parts of the resolved acquisition requests and // sort them auto sortedAcquisitionRequests = resolveNamespaceOrViewAcquisitionRequests( opCtx, *CollectionCatalog::get(opCtx), acquisitionRequests); // At this point, sortedAcquisitionRequests contains fully resolved (both nss and uuid) // namespace or view requests in sorted order. However, there is still no guarantee that the // nss <-> uuid mapping won't change from underneath. // // Lock the collection locks in the sorted order and recheck the UUIDS. If it throws // CollectionUUIDMismatch, we need to start over. const auto& dbName = sortedAcquisitionRequests.begin()->second.prerequisites.nss.dbName(); Lock::DBLockSkipOptions dbLockOptions = [&]() { Lock::DBLockSkipOptions dbLockOptions; dbLockOptions.skipRSTLLock = std::all_of(sortedAcquisitionRequests.begin(), sortedAcquisitionRequests.end(), [](const auto& ar) { return AutoGetDb::canSkipRSTLLock(ar.second.prerequisites.nss); }); dbLockOptions.skipFlowControlTicket = std::all_of( sortedAcquisitionRequests.begin(), sortedAcquisitionRequests.end(), [](const auto& ar) { return AutoGetDb::canSkipFlowControlTicket(ar.second.prerequisites.nss); }); return dbLockOptions; }(); const auto dbLock = std::make_shared(opCtx, dbName, isSharedLockMode(mode) ? MODE_IS : MODE_IX, Date_t::max(), dbLockOptions); for (auto& ar : sortedAcquisitionRequests) { const auto& nss = ar.second.prerequisites.nss; tassert(7300400, str::stream() << "Cannot acquire locks for collections across different databases ('" << dbName.toStringForErrorMsg() << "' vs '" << nss.dbName().toStringForErrorMsg() << "'", dbName == nss.dbName()); ar.second.dbLock = dbLock; ar.second.collLock.emplace(opCtx, nss, mode); } // Wait for a configured amount of time after acquiring locks if the failpoint is // enabled catalog_helper::setAutoGetCollectionWaitFailpointExecute([&](const BSONObj& data) { sleepFor(Milliseconds(data["waitForMillis"].numberInt())); }); // Recheck UUIDs. For writes in multi document transactions we need to use the latest // catalog and throw a WW conflict if UUIDs don't match. In all other cases we use the // current catalog and retry the loop. const auto& currentCatalog = CollectionCatalog::get(opCtx); const auto& latestCatalog = CollectionCatalog::latest(opCtx); for (auto& ar : sortedAcquisitionRequests) { const auto& prerequisites = ar.second.prerequisites; auto changedUUID = [&](auto& catalog) { const auto coll = catalog->lookupCollectionByNamespace(opCtx, prerequisites.nss); if (prerequisites.uuid && (!coll || coll->uuid() != prerequisites.uuid)) { return true; } return false; }; bool writeInMultiDocumentTransaction = inMultiDocumentTransaction && prerequisites.operationType == AcquisitionPrerequisites::OperationType::kWrite; if (changedUUID(writeInMultiDocumentTransaction ? latestCatalog : currentCatalog)) { if (writeInMultiDocumentTransaction) { throwWriteConflictException( str::stream() << "Unable to write to collection '" << prerequisites.nss << "' due to catalog changes; please retry the operation"); } else { // Retry optimistic resolution. continue; } } } checkShardingPlacement(opCtx, acquisitionRequests); // TODO SERVER-76901 open consistent storage snapshot. auto catalog = CollectionCatalog::get(opCtx); return acquireResolvedCollectionsOrViewsWithoutTakingLocks( opCtx, *catalog, std::move(sortedAcquisitionRequests)); } } std::vector acquireCollectionsOrViewsWithoutTakingLocks( OperationContext* opCtx, std::vector acquisitionRequests) { if (acquisitionRequests.size() == 0) { return {}; } // We shouldn't have an open snapshot unless a previous lock-free acquisition opened and stashed // it already. invariant(!opCtx->recoveryUnit()->isActive() || opCtx->isLockFreeReadsOp()); // TODO SERVER-76415 take global lock. // Make sure the sharding placement is correct before opening the storage snapshot, we will // check it again after opening it to make sure it is consistent. This is specially important in // secondaries since they can be lagging and might not be aware of the latests routing changes. checkShardingPlacement(opCtx, acquisitionRequests); // Open a consistent catalog snapshot if needed. bool openSnapshot = !opCtx->recoveryUnit()->isActive(); auto catalog = openSnapshot ? stashConsistentCatalog(opCtx, acquisitionRequests) : CollectionCatalog::get(opCtx); try { // Second sharding placement check. checkShardingPlacement(opCtx, acquisitionRequests); auto sortedAcquisitionRequests = shard_role_details::generateSortedAcquisitionRequests( opCtx, *catalog, acquisitionRequests); return acquireResolvedCollectionsOrViewsWithoutTakingLocks( opCtx, *catalog, std::move(sortedAcquisitionRequests)); } catch (...) { if (openSnapshot && !opCtx->lockState()->inAWriteUnitOfWork()) opCtx->recoveryUnit()->abandonSnapshot(); throw; } } ScopedCollectionAcquisition acquireCollectionForLocalCatalogOnlyWithPotentialDataLoss( OperationContext* opCtx, const NamespaceString& nss, LockMode mode) { invariant(!OperationShardingState::isComingFromRouter(opCtx)); auto& txnResources = getOrMakeTransactionResources(opCtx); txnResources.assertNoAcquiredCollections(); auto dbLock = std::make_shared( opCtx, nss.dbName(), isSharedLockMode(mode) ? MODE_IS : MODE_IX); Lock::CollectionLock collLock(opCtx, nss, mode); const auto catalog = CollectionCatalog::get(opCtx); auto prerequisites = AcquisitionPrerequisites(nss, boost::none, repl::ReadConcernArgs::get(opCtx), AcquisitionPrerequisites::kLocalCatalogOnlyWithPotentialDataLoss, AcquisitionPrerequisites::OperationType::kWrite, AcquisitionPrerequisites::ViewMode::kMustBeCollection); auto collOrView = acquireLocalCollectionOrView(opCtx, *catalog, prerequisites); invariant(std::holds_alternative(collOrView)); auto& coll = std::get(collOrView); if (coll) prerequisites.uuid = boost::optional(coll->uuid()); shard_role_details::AcquiredCollection& acquiredCollection = txnResources.addAcquiredCollection({prerequisites, std::move(dbLock), std::move(collLock), boost::none, boost::none, std::move(coll)}); return ScopedCollectionAcquisition(opCtx, acquiredCollection); } ScopedLocalCatalogWriteFence::ScopedLocalCatalogWriteFence(OperationContext* opCtx, ScopedCollectionAcquisition* acquisition) : _opCtx(opCtx), _acquiredCollection(&acquisition->_acquiredCollection) { // Clear the collectionPtr from the acquisition to indicate that it should not be used until the // caller is done with the DDL modifications _acquiredCollection->collectionPtr = CollectionPtr(); // OnCommit, there is nothing to do because the caller is not allowed to use the collection in // the scope of the ScopedLocalCatalogWriteFence and the destructor will take care of updating // the acquisition to point to the latest changed value. std::weak_ptr sharedImplWeakPtr = _acquiredCollection->sharedImpl; opCtx->recoveryUnit()->onRollback( [acquiredCollection = _acquiredCollection, sharedImplWeakPtr = sharedImplWeakPtr](OperationContext* opCtx) mutable { // OnRollback, the acquired collection must be set to reference the previously // established catalog snapshot if (!sharedImplWeakPtr.expired()) { _updateAcquiredLocalCollection(opCtx, acquiredCollection); } }); } ScopedLocalCatalogWriteFence::~ScopedLocalCatalogWriteFence() { _updateAcquiredLocalCollection(_opCtx, _acquiredCollection); } void ScopedLocalCatalogWriteFence::_updateAcquiredLocalCollection( OperationContext* opCtx, shard_role_details::AcquiredCollection* acquiredCollection) { try { const auto catalog = CollectionCatalog::latest(opCtx); const auto& nss = acquiredCollection->prerequisites.nss; auto collection = catalog->lookupCollectionByNamespace(opCtx, acquiredCollection->prerequisites.nss); checkCollectionUUIDMismatch(opCtx, nss, collection, acquiredCollection->prerequisites.uuid); acquiredCollection->collectionPtr = CollectionPtr(collection); } catch (const DBException& ex) { LOGV2_DEBUG(7653800, 1, "Failed to update ScopedLocalCatalogWriteFence", "ex"_attr = redact(ex.toString())); acquiredCollection->invalidated = true; } } YieldedTransactionResources::~YieldedTransactionResources() { invariant(!_yieldedResources); } YieldedTransactionResources::YieldedTransactionResources( std::unique_ptr&& yieldedResources) : _yieldedResources(std::move(yieldedResources)) {} void YieldedTransactionResources::dispose() { if (_yieldedResources) { _yieldedResources->releaseAllResourcesOnCommitOrAbort(); _yieldedResources.reset(); } } boost::optional yieldTransactionResourcesFromOperationContext( OperationContext* opCtx) { auto& transactionResources = getTransactionResources(opCtx); if (!transactionResources) { return YieldedTransactionResources(); } invariant(!transactionResources->yielded); for (auto& acquisition : transactionResources->acquiredCollections) { // Yielding kLocalCatalogOnlyWithPotentialDataLoss acquisitions is not allowed. invariant( !stdx::holds_alternative( acquisition.prerequisites.placementConcern), str::stream() << "Collection " << acquisition.prerequisites.nss.toStringForErrorMsg() << " acquired with special placement concern and cannot be yielded"); // If the uuid wasn't originally set on the prerequisites, set it now so that on restore // from yield we can check we are restoring the same instance of the ns. if (!acquisition.prerequisites.uuid && acquisition.collectionPtr) { acquisition.prerequisites.uuid = acquisition.collectionPtr->uuid(); } } // Yielding view acquisitions is not supported. tassert(7300502, "Yielding view acquisitions is forbidden", transactionResources->acquiredViews.empty()); invariant(!transactionResources->yieldedLocker); Locker::LockSnapshot lockSnapshot; opCtx->lockState()->saveLockStateAndUnlock(&lockSnapshot); transactionResources->yieldedLocker.emplace(std::move(lockSnapshot)); transactionResources->yielded = true; return YieldedTransactionResources(std::move(transactionResources)); } void restoreTransactionResourcesToOperationContext(OperationContext* opCtx, YieldedTransactionResources&& yieldedResources) { if (!yieldedResources._yieldedResources) { // Nothing to restore. return; } // On failure to restore, release the yielded resources. ScopeGuard scopeGuard([&] { yieldedResources._yieldedResources->releaseAllResourcesOnCommitOrAbort(); yieldedResources._yieldedResources.reset(); }); auto restoreFn = [&]() { // Reacquire locks. if (yieldedResources._yieldedResources->yieldedLocker) { opCtx->lockState()->restoreLockState( opCtx, *yieldedResources._yieldedResources->yieldedLocker); yieldedResources._yieldedResources->yieldedLocker.reset(); } // Reestablish a consistent catalog snapshot (multi document transactions don't yield). auto requests = toNamespaceStringOrUUIDs(yieldedResources._yieldedResources->acquiredCollections); auto catalog = getConsistentCatalogAndSnapshot(opCtx, requests); // Reacquire service snapshots. Will throw if placement concern can no longer be met. for (auto& acquiredCollection : yieldedResources._yieldedResources->acquiredCollections) { const auto& prerequisites = acquiredCollection.prerequisites; auto uassertCollectionAppearedAfterRestore = [&] { uasserted( 743870, str::stream() << "Collection " << prerequisites.nss.toStringForErrorMsg() << " appeared after a restore, which violates the semantics of restore"); }; if (prerequisites.operationType == AcquisitionPrerequisites::OperationType::kRead) { // Just reacquire the CollectionPtr. Reads don't care about placement changes // because they have already established a ScopedCollectionFilter that acts as // RangePreserver. auto collOrView = acquireLocalCollectionOrView(opCtx, *catalog, prerequisites); // We do not support yielding view acquisitions. Therefore it is not possible that // upon restore 'acquireLocalCollectionOrView' snapshoted a view -- it would not // have met the prerequisite that the collection instance is still the same as the // one before yielding. invariant(std::holds_alternative(collOrView)); if (!acquiredCollection.collectionPtr != !std::get(collOrView)) { uassertCollectionAppearedAfterRestore(); } // Update the services snapshot on TransactionResources acquiredCollection.collectionPtr = std::move(std::get(collOrView)); } else { auto reacquiredServicesSnapshot = acquireServicesSnapshot(opCtx, *catalog, prerequisites); // We do not support yielding view acquisitions. Therefore it is not possible that // upon restore 'acquireLocalCollectionOrView' snapshoted a view -- it would not // have met the prerequisite that the collection instance is still the same as the // one before yielding. invariant(std::holds_alternative( reacquiredServicesSnapshot.collectionPtrOrView)); if (!acquiredCollection.collectionPtr != !std::get(reacquiredServicesSnapshot.collectionPtrOrView)) { uassertCollectionAppearedAfterRestore(); } // Update the services snapshot on TransactionResources acquiredCollection.collectionPtr = std::move( std::get(reacquiredServicesSnapshot.collectionPtrOrView)); acquiredCollection.collectionDescription = std::move(reacquiredServicesSnapshot.collectionDescription); acquiredCollection.ownershipFilter = std::move(reacquiredServicesSnapshot.ownershipFilter); } // TODO: This will be removed when we no longer snapshot sharding state on // CollectionPtr. invariant(acquiredCollection.collectionDescription); if (acquiredCollection.collectionDescription->isSharded()) { acquiredCollection.collectionPtr.setShardKeyPattern( acquiredCollection.collectionDescription->getKeyPattern()); } } return catalog; }; auto catalog = [&]() { while (true) { try { return restoreFn(); } catch (const ExceptionFor& ex) { if (ShardVersion::isPlacementVersionIgnored(ex->getVersionReceived()) && ex->getCriticalSectionSignal()) { // If ShardVersion is IGNORED and we encountered a critical section, then yield, // wait for the critical section to finish and then we'll resume the write from // the point we had left. We do this to prevent large multi-writes from // repeatedly failing due to StaleConfig and exhausting the mongos retry // attempts. Yield the locks. yieldedResources._yieldedResources->yieldedLocker.emplace(); opCtx->recoveryUnit()->abandonSnapshot(); opCtx->lockState()->saveLockStateAndUnlock( yieldedResources._yieldedResources->yieldedLocker.get_ptr()); // Wait for the critical section to finish. OperationShardingState::waitForCriticalSectionToComplete( opCtx, *ex->getCriticalSectionSignal()) .ignore(); // Try again to restore. continue; } throw; } } }(); // Restore TransactionsResource on opCtx. yieldedResources._yieldedResources->yielded = false; getTransactionResources(opCtx) = std::move(yieldedResources)._yieldedResources; if (!opCtx->inMultiDocumentTransaction()) { CollectionCatalog::stash(opCtx, catalog); } scopeGuard.dismiss(); } } // namespace mongo