Add replication task latency histogram

Signed-off-by: Diana Zawadzki <dzawa@live.de>

Author: zawadzkidiana

common/metrics/defs.go

@@ -2226,6 +2226,7 @@ const (
 	PersistenceFailures
 	PersistenceLatency
 	PersistenceLatencyHistogram
+	PersistenceQuota
 	PersistenceErrShardExistsCounter
 	PersistenceErrShardOwnershipLostCounter
 	PersistenceErrConditionFailedCounter
@@ -2480,6 +2481,7 @@ const (
 const (
 	TaskRequests = iota + NumCommonMetrics
 	TaskLatency
+	ExponentialTaskLatency
 	TaskFailures
 	TaskDiscarded
 	TaskAttemptTimer
@@ -2650,6 +2652,7 @@ const (
 	CacheRequests
 	CacheFailures
 	CacheLatency
+	ExponentialCacheLatency
 	CacheHitCounter
 	CacheMissCounter
 	CacheFullCounter
@@ -3045,6 +3048,7 @@ var MetricDefs = map[ServiceIdx]map[MetricIdx]metricDefinition{
 		PersistenceFailures:                                          {metricName: "persistence_errors", metricType: Counter},
 		PersistenceLatency:                                           {metricName: "persistence_latency", metricType: Timer},
 		PersistenceLatencyHistogram:                                  {metricName: "persistence_latency_histogram", metricType: Histogram, buckets: PersistenceLatencyBuckets},
+		PersistenceQuota:                                             {metricName: "persistence_quota", metricType: Gauge},
 		PersistenceErrShardExistsCounter:                             {metricName: "persistence_errors_shard_exists", metricType: Counter},
 		PersistenceErrShardOwnershipLostCounter:                      {metricName: "persistence_errors_shard_ownership_lost", metricType: Counter},
 		PersistenceErrConditionFailedCounter:                         {metricName: "persistence_errors_condition_failed", metricType: Counter},
@@ -3309,6 +3313,7 @@ var MetricDefs = map[ServiceIdx]map[MetricIdx]metricDefinition{
 	History: {
 		TaskRequests:                     {metricName: "task_requests", metricType: Counter},
 		TaskLatency:                      {metricName: "task_latency", metricType: Timer},
+		ExponentialTaskLatency:           {metricName: "task_latency_ns", metricType: Histogram, exponentialBuckets: Low1ms100s},
 		TaskAttemptTimer:                 {metricName: "task_attempt", metricType: Timer},
 		TaskFailures:                     {metricName: "task_errors", metricType: Counter},
 		TaskDiscarded:                    {metricName: "task_errors_discarded", metricType: Counter},
@@ -3470,10 +3475,11 @@ var MetricDefs = map[ServiceIdx]map[MetricIdx]metricDefinition{
 		UnbufferReplicationTaskTimer:                                 {metricName: "unbuffer_replication_tasks", metricType: Timer},
 		HistoryConflictsCounter:                                      {metricName: "history_conflicts", metricType: Counter},
 		CompleteTaskFailedCounter:                                    {metricName: "complete_task_fail_count", metricType: Counter},
-		CacheSize:                                                    {metricName: "cache_size", metricType: Timer},
+		CacheSize:                                                    {metricName: "cache_size", metricType: Gauge},
 		CacheRequests:                                                {metricName: "cache_requests", metricType: Counter},
 		CacheFailures:                                                {metricName: "cache_errors", metricType: Counter},
 		CacheLatency:                                                 {metricName: "cache_latency", metricType: Timer},
+		ExponentialCacheLatency:                                      {metricName: "cache_latency_ns", metricType: Histogram, exponentialBuckets: Low1ms100s},
 		CacheHitCounter:                                              {metricName: "cache_hit", metricType: Counter},
 		CacheMissCounter:                                             {metricName: "cache_miss", metricType: Counter},
 		CacheFullCounter:                                             {metricName: "cache_full", metricType: Counter},

service/history/replication/task_ack_manager.go

@@ -131,8 +131,13 @@ func (t *TaskAckManager) getTasks(ctx context.Context, pollingCluster string, la
 		lastReadTaskID = previousReadTaskID
 	}
 
+	// Keep timer (backwards compatible), dual-emit exponential histogram for migration.
+	taskGeneratedStart := t.timeSource.Now()
 	taskGeneratedTimer := t.scope.StartTimer(metrics.TaskLatency)
 	defer taskGeneratedTimer.Stop()
+	defer func() {
+		t.scope.ExponentialHistogram(metrics.ExponentialTaskLatency, t.timeSource.Since(taskGeneratedStart))
+	}()
 
 	batchSize := t.dynamicTaskBatchSizer.value()
 	t.scope.UpdateGauge(metrics.ReplicationTasksBatchSize, float64(batchSize))