SERVER-48179 Allow transition to SECONDARY at the end of rollback even it was changed to REMOVED.

2 files changed, 10 insertions, 3 deletions

diff --git a/src/mongo/db/repl/rollback_impl.cpp b/src/mongo/db/repl/rollback_impl.cpp
index c1ea618a6e5..2e1493b43e9 100644
--- a/src/mongo/db/repl/rollback_impl.cpp
+++ b/src/mongo/db/repl/rollback_impl.cpp
@@ -1225,8 +1225,12 @@ Status RollbackImpl::_triggerOpObserver(OperationContext* opCtx) {
 
 void RollbackImpl::_transitionFromRollbackToSecondary(OperationContext* opCtx) {
     invariant(opCtx);
-    invariant(_replicationCoordinator->getMemberState() == MemberState(MemberState::RS_ROLLBACK));
 
+    // It is possible that this node has actually been removed due to a reconfig via
+    // heartbeat during rollback. But it should be fine to transition to SECONDARY
+    // and this won't change how the node reports its member state since topology
+    // coordinator will always check if the node exists in its local config when
+    // returning member state.
     LOGV2(21611, "Transition to SECONDARY");
 
     ReplicationStateTransitionLockGuard transitionGuard(opCtx, MODE_X);
diff --git a/src/mongo/db/repl/topology_coordinator.cpp b/src/mongo/db/repl/topology_coordinator.cpp
index ede2650481d..53ca6befcd1 100644
--- a/src/mongo/db/repl/topology_coordinator.cpp
+++ b/src/mongo/db/repl/topology_coordinator.cpp
@@ -1003,8 +1003,11 @@ HeartbeatResponseAction TopologyCoordinator::processHeartbeatResponse(
             nextAction = HeartbeatResponseAction::makeReconfigAction();
             nextAction.setNextHeartbeatStartDate(nextHeartbeatStartDate);
 
-            // TODO(SERVER-48178) Only continue processing heartbeat in primary state to avoid
-            // concurrent reconfig and rollback.
+            // Only continue processing heartbeat in primary state. In other states it is not
+            // safe to continue processing heartbeat and should start reconfig right away.
+            // e.g. if this node was removed from replSet, _selfIndex is -1, and a following
+            // check on _selfIndex will keep retrying heartbeat to fetch new config, preventing
+            // the new config to be installed.
             if (_role != Role::kLeader) {
                 return nextAction;
             }