SERVER-72774 Prevent a node in quiesce mode to win election.

3 files changed, 18 insertions, 4 deletions

diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index e6af5c08838..d4bd43a16ff 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -1007,6 +1007,10 @@ bool ReplicationCoordinatorImpl::enterQuiesceModeIfSecondary(Milliseconds quiesc
         return false;
     }
 
+    // Cancel any ongoing election so that the node cannot become primary once in quiesce mode,
+    // and do not wait for cancellation to complete.
+    _cancelElectionIfNeeded(lk);
+
     _inQuiesceMode = true;
     _quiesceDeadline = _replExecutor->now() + quiesceTime;
 
diff --git a/src/mongo/db/repl/replication_coordinator_impl_elect_v1.cpp b/src/mongo/db/repl/replication_coordinator_impl_elect_v1.cpp
index 57959db018b..a064fedf1e7 100644
--- a/src/mongo/db/repl/replication_coordinator_impl_elect_v1.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl_elect_v1.cpp
@@ -49,6 +49,8 @@ namespace mongo {
 namespace repl {
 
 MONGO_FAIL_POINT_DEFINE(hangInWritingLastVoteForDryRun);
+MONGO_FAIL_POINT_DEFINE(electionHangsBeforeUpdateMemberState);
+MONGO_FAIL_POINT_DEFINE(hangBeforeOnVoteRequestCompleteCallback);
 
 class ReplicationCoordinatorImpl::ElectionState::LoseElectionGuardV1 {
     LoseElectionGuardV1(const LoseElectionGuardV1&) = delete;
@@ -137,7 +139,12 @@ ReplicationCoordinatorImpl::ElectionState::getElectionDryRunFinishedEvent(WithLo
 
 void ReplicationCoordinatorImpl::ElectionState::cancel(WithLock) {
     _isCanceled = true;
-    _voteRequester->cancel();
+    // This check is necessary because _voteRequester is only initialized in _startVoteRequester.
+    // Since we don't hold mutex during the entire election process, it is possible to get here
+    // before _startVoteRequester is ever called.
+    if (_voteRequester) {
+        _voteRequester->cancel();
+    }
 }
 
 void ReplicationCoordinatorImpl::ElectionState::start(WithLock lk, StartElectionReasonEnum reason) {
@@ -396,13 +403,16 @@ void ReplicationCoordinatorImpl::ElectionState::_requestVotesForRealElection(
     _replExecutor
         ->onEvent(nextPhaseEvh.getValue(),
                   [=](const executor::TaskExecutor::CallbackArgs&) {
+                      if (MONGO_unlikely(hangBeforeOnVoteRequestCompleteCallback.shouldFail())) {
+                          LOGV2(7277400,
+                                "Hang due to hangBeforeOnVoteRequestCompleteCallback failpoint");
+                          hangBeforeOnVoteRequestCompleteCallback.pauseWhileSet();
+                      }
                       _onVoteRequestComplete(newTerm, reason);
                   })
         .status_with_transitional_ignore();
 }
 
-MONGO_FAIL_POINT_DEFINE(electionHangsBeforeUpdateMemberState);
-
 void ReplicationCoordinatorImpl::ElectionState::_onVoteRequestComplete(
     long long newTerm, StartElectionReasonEnum reason) {
     stdx::lock_guard<Latch> lk(_repl->_mutex);
diff --git a/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp b/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
index 83f39f24cd1..4330baaf036 100644
--- a/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
@@ -1327,7 +1327,7 @@ void ReplicationCoordinatorImpl::_startElectSelfIfEligibleV1(WithLock lk,
         _cancelCatchupTakeover_inlock();
         _cancelPriorityTakeover_inlock();
         _cancelAndRescheduleElectionTimeout_inlock();
-        if (_inShutdown) {
+        if (_inShutdown || _inQuiesceMode) {
             LOGV2_FOR_ELECTION(4615654, 0, "Not starting an election, since we are shutting down");
             return;
         }