Fix queue diff metrics for disabled clusters (#2329)

temporalio · Dec 29, 2021 · 81ce604 · 81ce604
1 parent 94ef184
commit 81ce604
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 2 deletions.
diff --git a/service/history/replicationDLQHandler_test.go b/service/history/replicationDLQHandler_test.go
@@ -109,6 +109,7 @@ func (s *replicationDLQHandlerSuite) SetupTest() {
 	s.shardManager = s.mockResource.ShardMgr
 	s.config = tests.NewDynamicConfig()
 	s.clusterMetadata.EXPECT().GetCurrentClusterName().Return(cluster.TestCurrentClusterName).AnyTimes()
+	s.clusterMetadata.EXPECT().GetAllClusterInfo().Return(cluster.TestAllClusterInfo).AnyTimes()
 	s.taskExecutors = make(map[string]replicationTaskExecutor)
 	s.taskExecutor = NewMockreplicationTaskExecutor(s.controller)
 	s.sourceCluster = cluster.TestAlternativeClusterName

diff --git a/service/history/shard/context_impl.go b/service/history/shard/context_impl.go
@@ -992,10 +992,15 @@ func (s *ContextImpl) updateShardInfoLocked() error {
 
 func (s *ContextImpl) emitShardInfoMetricsLogsLocked() {
 	currentCluster := s.GetClusterMetadata().GetCurrentClusterName()
+	clusterInfo := s.GetClusterMetadata().GetAllClusterInfo()
 
 	minTransferLevel := s.shardInfo.ClusterTransferAckLevel[currentCluster]
 	maxTransferLevel := s.shardInfo.ClusterTransferAckLevel[currentCluster]
-	for _, v := range s.shardInfo.ClusterTransferAckLevel {
+	for clusterName, v := range s.shardInfo.ClusterTransferAckLevel {
+		if !clusterInfo[clusterName].Enabled {
+			continue
+		}
+
 		if v < minTransferLevel {
 			minTransferLevel = v
 		}
@@ -1007,7 +1012,11 @@ func (s *ContextImpl) emitShardInfoMetricsLogsLocked() {
 
 	minTimerLevel := timestamp.TimeValue(s.shardInfo.ClusterTimerAckLevel[currentCluster])
 	maxTimerLevel := timestamp.TimeValue(s.shardInfo.ClusterTimerAckLevel[currentCluster])
-	for _, v := range s.shardInfo.ClusterTimerAckLevel {
+	for clusterName, v := range s.shardInfo.ClusterTimerAckLevel {
+		if !clusterInfo[clusterName].Enabled {
+			continue
+		}
+
 		t := timestamp.TimeValue(v)
 		if t.Before(minTimerLevel) {
 			minTimerLevel = t

diff --git a/service/history/timerQueueAckMgr_test.go b/service/history/timerQueueAckMgr_test.go
@@ -476,6 +476,7 @@ func (s *timerQueueAckMgrSuite) TestReadCompleteUpdateTimerTasks() {
 		NextPageToken: nil,
 	}
 	s.mockClusterMetadata.EXPECT().GetCurrentClusterName().Return(cluster.TestCurrentClusterName).AnyTimes()
+	s.mockClusterMetadata.EXPECT().GetAllClusterInfo().Return(cluster.TestAllClusterInfo).AnyTimes()
 	s.mockExecutionMgr.EXPECT().GetTimerTasks(gomock.Any()).Return(response, nil)
 	s.mockExecutionMgr.EXPECT().GetTimerTasks(gomock.Any()).Return(&persistence.GetTimerTasksResponse{}, nil)
 	filteredTasks, lookAheadTask, moreTasks, err := s.timerQueueAckMgr.readTimerTasks()
@@ -784,6 +785,7 @@ func (s *timerQueueFailoverAckMgrSuite) TestReadCompleteUpdateTimerTasks() {
 		NextPageToken: nil,
 	}
 	s.mockClusterMetadata.EXPECT().GetCurrentClusterName().Return(cluster.TestCurrentClusterName).AnyTimes()
+	s.mockClusterMetadata.EXPECT().GetAllClusterInfo().Return(cluster.TestAllClusterInfo).AnyTimes()
 	s.mockExecutionMgr.EXPECT().GetTimerTasks(gomock.Any()).Return(response, nil)
 	filteredTasks, lookAheadTask, moreTasks, err := s.timerQueueFailoverAckMgr.readTimerTasks()
 	s.Nil(err)