diff options
author | Allison Easton <allison.easton@mongodb.com> | 2022-05-16 16:59:22 +0000 |
---|---|---|
committer | Evergreen Agent <no-reply@evergreen.mongodb.com> | 2022-05-19 12:10:12 +0000 |
commit | ea66e125bf6368da4146c1d8974c595e29787542 (patch) | |
tree | b77f80604846328130fc0226a8bd1b862d894046 /src/mongo | |
parent | 856c10ff0da11b45e1a8462e48f93cf03381485c (diff) | |
download | mongo-ea66e125bf6368da4146c1d8974c595e29787542.tar.gz |
SERVER-65930 DDL coordinators and rename participant initial checkpoint may incur in DuplicateKey error
(cherry picked from commit 1eb5a9257b3bfc0c768b342d73c3668cc6566841)
Diffstat (limited to 'src/mongo')
-rw-r--r-- | src/mongo/db/s/rename_collection_participant_service.cpp | 19 | ||||
-rw-r--r-- | src/mongo/db/s/sharding_ddl_coordinator.cpp | 7 | ||||
-rw-r--r-- | src/mongo/db/s/sharding_ddl_coordinator.h | 17 |
3 files changed, 33 insertions, 10 deletions
diff --git a/src/mongo/db/s/rename_collection_participant_service.cpp b/src/mongo/db/s/rename_collection_participant_service.cpp index 4d9efb66056..51256975393 100644 --- a/src/mongo/db/s/rename_collection_participant_service.cpp +++ b/src/mongo/db/s/rename_collection_participant_service.cpp @@ -201,12 +201,23 @@ void RenameParticipantInstance::_enterPhase(Phase newPhase) { PersistentTaskStore<StateDoc> store(NamespaceString::kShardingRenameParticipantsNamespace); if (_doc.getPhase() == Phase::kUnset) { - store.add(opCtx.get(), newDoc, WriteConcerns::kMajorityWriteConcernShardingTimeout); + try { + store.add(opCtx.get(), newDoc, WriteConcerns::kMajorityWriteConcernNoTimeout); + } catch (const ExceptionFor<ErrorCodes::DuplicateKey>&) { + // A series of step-up and step-down events can cause a node to try and insert the + // document when it has already been persisted locally, but we must still wait for + // majority commit. + const auto replCoord = repl::ReplicationCoordinator::get(opCtx.get()); + const auto lastLocalOpTime = replCoord->getMyLastAppliedOpTime(); + WaitForMajorityService::get(opCtx->getServiceContext()) + .waitUntilMajority(lastLocalOpTime, opCtx.get()->getCancellationToken()) + .get(opCtx.get()); + } } else { store.update(opCtx.get(), BSON(StateDoc::kFromNssFieldName << fromNss().ns()), newDoc.toBSON(), - WriteConcerns::kMajorityWriteConcernShardingTimeout); + WriteConcerns::kMajorityWriteConcernNoTimeout); } _doc = std::move(newDoc); @@ -222,7 +233,7 @@ void RenameParticipantInstance::_removeStateDocument(OperationContext* opCtx) { PersistentTaskStore<StateDoc> store(NamespaceString::kShardingRenameParticipantsNamespace); store.remove(opCtx, BSON(StateDoc::kFromNssFieldName << fromNss().ns()), - WriteConcerns::kMajorityWriteConcernShardingTimeout); + WriteConcerns::kMajorityWriteConcernNoTimeout); _doc = {}; } @@ -387,7 +398,7 @@ SemiFuture<void> RenameParticipantInstance::_runImpl( service->releaseRecoverableCriticalSection( opCtx, fromNss(), reason, ShardingCatalogClient::kLocalWriteConcern); service->releaseRecoverableCriticalSection( - opCtx, toNss(), reason, ShardingCatalogClient::kMajorityWriteConcern); + opCtx, toNss(), reason, WriteConcerns::kMajorityWriteConcernNoTimeout); LOGV2(5515107, "CRUD unblocked", "fromNs"_attr = fromNss(), "toNs"_attr = toNss()); })) diff --git a/src/mongo/db/s/sharding_ddl_coordinator.cpp b/src/mongo/db/s/sharding_ddl_coordinator.cpp index 28741f4503b..9ce509fe4b6 100644 --- a/src/mongo/db/s/sharding_ddl_coordinator.cpp +++ b/src/mongo/db/s/sharding_ddl_coordinator.cpp @@ -116,10 +116,9 @@ bool ShardingDDLCoordinator::_removeDocument(OperationContext* opCtx) { batchedResponse.parseBSON(commandReply, &unusedErrmsg); WriteConcernResult ignoreResult; - const WriteConcernOptions majorityWriteConcern{ - WriteConcernOptions::kMajority, - WriteConcernOptions::SyncMode::UNSET, - WriteConcernOptions::kWriteConcernTimeoutSharding}; + const WriteConcernOptions majorityWriteConcern{WriteConcernOptions::kMajority, + WriteConcernOptions::SyncMode::UNSET, + WriteConcernOptions::kNoTimeout}; auto latestOpTime = repl::ReplClientInfo::forClient(opCtx->getClient()).getLastOp(); uassertStatusOK(waitForWriteConcern(opCtx, latestOpTime, majorityWriteConcern, &ignoreResult)); diff --git a/src/mongo/db/s/sharding_ddl_coordinator.h b/src/mongo/db/s/sharding_ddl_coordinator.h index e549a13f2a4..fdb4d35014b 100644 --- a/src/mongo/db/s/sharding_ddl_coordinator.h +++ b/src/mongo/db/s/sharding_ddl_coordinator.h @@ -32,6 +32,8 @@ #include "mongo/db/namespace_string.h" #include "mongo/db/operation_context.h" #include "mongo/db/persistent_task_store.h" +#include "mongo/db/repl/replication_coordinator.h" +#include "mongo/db/repl/wait_for_majority_service.h" #include "mongo/db/s/dist_lock_manager.h" #include "mongo/db/s/forwardable_operation_metadata.h" #include "mongo/db/s/sharding_ddl_coordinator_gen.h" @@ -129,7 +131,18 @@ protected: auto opCtx = cc().makeOperationContext(); PersistentTaskStore<StateDoc> store(NamespaceString::kShardingDDLCoordinatorsNamespace); - store.add(opCtx.get(), newDoc, WriteConcerns::kMajorityWriteConcernShardingTimeout); + try { + store.add(opCtx.get(), newDoc, WriteConcerns::kMajorityWriteConcernNoTimeout); + } catch (const ExceptionFor<ErrorCodes::DuplicateKey>&) { + // A series of step-up and step-down events can cause a node to try and insert the + // document when it has already been persisted locally, but we must still wait for + // majority commit. + const auto replCoord = repl::ReplicationCoordinator::get(opCtx.get()); + const auto lastLocalOpTime = replCoord->getMyLastAppliedOpTime(); + WaitForMajorityService::get(opCtx->getServiceContext()) + .waitUntilMajority(lastLocalOpTime, opCtx.get()->getCancellationToken()) + .get(opCtx.get()); + } return std::move(newDoc); } @@ -141,7 +154,7 @@ protected: store.update(opCtx, BSON(StateDoc::kIdFieldName << newDoc.getId().toBSON()), newDoc.toBSON(), - WriteConcerns::kMajorityWriteConcernShardingTimeout); + WriteConcerns::kMajorityWriteConcernNoTimeout); return std::move(newDoc); } |