SERVER-20928 Heartbeat shouldn't cause stepdown in protocol version 1.

Author: visualzhou

src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp

@@ -224,6 +224,7 @@ void ReplicationCoordinatorImpl::_handleHeartbeatResponseAction(
         case HeartbeatResponseAction::StepDownSelf:
             invariant(action.getPrimaryConfigIndex() == _selfIndex);
             log() << "Stepping down from primary in response to heartbeat";
+            _topCoord->prepareForStepDown();
             _stepDownStart();
             break;
         case HeartbeatResponseAction::StepDownRemotePrimary: {

src/mongo/db/repl/topology_coordinator_impl.cpp

@@ -1118,18 +1118,9 @@ HeartbeatResponseAction TopologyCoordinatorImpl::_updatePrimaryFromHBDataV1(
 
     // We do not believe any remote to be primary.
 
-    // If we are primary, check if we can still see majority of the set; stepdown if we can't.
+    // Return if we are primary. The stepdown decision is based on liveness rather than
+    // heartbeats in pv 1.
     if (_iAmPrimary()) {
-        if (CannotSeeMajority & _getMyUnelectableReason(now, lastOpApplied)) {
-            if (_stepDownPending) {
-                return HeartbeatResponseAction::makeNoAction();
-            }
-            _stepDownPending = true;
-            log() << "can't see a majority of the set, relinquishing primary";
-            return HeartbeatResponseAction::makeStepDownSelfAction(_selfIndex);
-        }
-
-        LOG(2) << "Choosing to remain primary";
         return HeartbeatResponseAction::makeNoAction();
     }
 

src/mongo/db/repl/topology_coordinator_impl_v1_test.cpp

@@ -2665,7 +2665,7 @@ TEST_F(HeartbeatResponseTestV1, UpdateHeartbeatDataRemoteDoesNotExist) {
     ASSERT_TRUE(TopologyCoordinator::Role::follower == getTopoCoord().getRole());
 }
 
-TEST_F(HeartbeatResponseTestV1, UpdateHeartbeatDataRelinquishPrimaryDueToNodeDisappearing) {
+TEST_F(HeartbeatResponseTestV1, UpdateHeartbeatDataDoesNotRelinquishPrimary) {
     // Become PRIMARY.
     ASSERT_EQUALS(-1, getCurrentPrimaryIndex());
     makeSelfPrimary(Timestamp(2, 0));
@@ -2679,20 +2679,15 @@ TEST_F(HeartbeatResponseTestV1, UpdateHeartbeatDataRelinquishPrimaryDueToNodeDis
     heartbeatFromMember(HostAndPort("host3"), "rs0", MemberState::RS_SECONDARY, OpTime());
     heartbeatFromMember(HostAndPort("host3"), "rs0", MemberState::RS_SECONDARY, OpTime());
 
-    // Lose that awareness and be sure we are going to stepdown.
+    // Lose that awareness, but we are not going to step down, because stepdown only
+    // depends on liveness.
     HeartbeatResponseAction nextAction =
         receiveDownHeartbeat(HostAndPort("host2"), "rs0", OpTime(Timestamp(100, 0), 0));
     ASSERT_NO_ACTION(nextAction.getAction());
     nextAction = receiveDownHeartbeat(HostAndPort("host3"), "rs0", OpTime(Timestamp(100, 0), 0));
-    ASSERT_EQUALS(HeartbeatResponseAction::StepDownSelf, nextAction.getAction());
-    ASSERT_EQUALS(0, nextAction.getPrimaryConfigIndex());
+    ASSERT_NO_ACTION(nextAction.getAction());
     ASSERT_TRUE(TopologyCoordinator::Role::leader == getTopoCoord().getRole());
     ASSERT_EQUALS(0, getCurrentPrimaryIndex());
-    // Doesn't actually do the stepdown until stepDownIfPending is called.
-
-    ASSERT_TRUE(getTopoCoord().stepDownIfPending());
-    ASSERT_TRUE(TopologyCoordinator::Role::follower == getTopoCoord().getRole());
-    ASSERT_EQUALS(-1, getCurrentPrimaryIndex());
 }
 
 TEST_F(HeartbeatResponseTestV1, UpdateHeartbeatDataPriorTakeoverDueToHigherPriority) {