diff options
author | Randolph Tan <randolph@10gen.com> | 2021-03-18 17:51:59 +0000 |
---|---|---|
committer | Evergreen Agent <no-reply@evergreen.mongodb.com> | 2021-04-23 13:31:50 +0000 |
commit | 7f7a299321f22c0042226529fd0803292fdaf657 (patch) | |
tree | 5c314bfbb79e66c8b97092b8fc5d3a827e15eb8a | |
parent | c2295adab43675bfde8c9b2aa5795d9b7fccb6b0 (diff) | |
download | mongo-7f7a299321f22c0042226529fd0803292fdaf657.tar.gz |
SERVER-55007 Deadlock between step down and MongoDOperationContextSession
(cherry picked from commit 939178eb77d25dc234de03ac7d5dc98210514c7a)
-rw-r--r-- | src/mongo/db/repl/replication_coordinator_impl.cpp | 5 | ||||
-rw-r--r-- | src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp | 11 |
2 files changed, 16 insertions, 0 deletions
diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp index da72aaa2fb3..b0960927b8c 100644 --- a/src/mongo/db/repl/replication_coordinator_impl.cpp +++ b/src/mongo/db/repl/replication_coordinator_impl.cpp @@ -2758,6 +2758,11 @@ void ReplicationCoordinatorImpl::_finishReplSetReconfig(OperationContext* opCtx, _topCoord->prepareForUnconditionalStepDown(); lk.unlock(); + // To prevent a deadlock between session checkout and RSTL lock taking, disallow new + // sessions from being checked out. Existing sessions currently checked out will be killed + // by the killOpThread. + ScopedBlockSessionCheckouts blockSessions(opCtx); + // Primary node won't be electable or removed after the configuration change. // So, finish the reconfig under RSTL, so that the step down occurs safely. arsd.emplace(this, opCtx, ReplicationCoordinator::OpsKillingStateTransitionEnum::kStepDown); diff --git a/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp b/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp index 0b32eacc507..bcdb1b7312b 100644 --- a/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp +++ b/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp @@ -53,6 +53,7 @@ #include "mongo/db/repl/topology_coordinator.h" #include "mongo/db/repl/vote_requester.h" #include "mongo/db/service_context.h" +#include "mongo/db/session_catalog.h" #include "mongo/platform/mutex.h" #include "mongo/rpc/get_status_from_command_result.h" #include "mongo/rpc/metadata/repl_set_metadata.h" @@ -409,6 +410,11 @@ void ReplicationCoordinatorImpl::_stepDownFinish( auto opCtx = cc().makeOperationContext(); + // To prevent a deadlock between session checkout and RSTL lock taking, disallow new sessions + // from being checked out. Existing sessions currently checked out will be killed by the + // killOpThread. + ScopedBlockSessionCheckouts blockSessions(opCtx.get()); + // kill all write operations which are no longer safe to run on step down. Also, operations that // have taken global lock in S mode and operations blocked on prepare conflict will be killed to // avoid 3-way deadlock between read, prepared transaction and step down thread. @@ -644,6 +650,11 @@ void ReplicationCoordinatorImpl::_heartbeatReconfigFinish( _topCoord->prepareForUnconditionalStepDown(); lk.unlock(); + // To prevent a deadlock between session checkout and RSTL lock taking, disallow new + // sessions from being checked out. Existing sessions currently checked out will be killed + // by the killOpThread. + ScopedBlockSessionCheckouts blockSessions(opCtx.get()); + // Primary node will be either unelectable or removed after the configuration change. // So, finish the reconfig under RSTL, so that the step down occurs safely. arsd.emplace( |