diff --git a/pkg/l4lb/l4controller.go b/pkg/l4lb/l4controller.go
index 9ce7ff62ae..df459744cd 100644
--- a/pkg/l4lb/l4controller.go
+++ b/pkg/l4lb/l4controller.go
@@ -499,11 +499,12 @@ func (l4c *L4Controller) publishMetrics(result *loadbalancers.L4ILBSyncResult, n
 	case loadbalancers.SyncTypeCreate, loadbalancers.SyncTypeUpdate:
 		klog.V(6).Infof("Internal L4 Loadbalancer for Service %s ensured, updating its state %v in metrics cache", namespacedName, result.MetricsState)
 		l4c.ctx.ControllerMetrics.SetL4ILBService(namespacedName, result.MetricsState)
+		l4metrics.PublishILBSyncMetrics(result.Error == nil, result.SyncType, result.GCEResourceInError, utils.GetErrorType(result.Error), result.StartTime)
 		if l4c.enableDualStack {
 			klog.V(6).Infof("Internal L4 DualStack Loadbalancer for Service %s ensured, updating its state %v in metrics cache", namespacedName, result.MetricsState)
 			l4c.ctx.ControllerMetrics.SetL4ILBDualStackService(namespacedName, result.DualStackMetricsState)
+			l4metrics.PublishL4ILBDualStackSyncLatency(result.Error == nil, result.SyncType, result.DualStackMetricsState.IPFamilies, result.StartTime)
 		}
-		l4metrics.PublishILBSyncMetrics(result.Error == nil, result.SyncType, result.GCEResourceInError, utils.GetErrorType(result.Error), result.StartTime)
 
 	case loadbalancers.SyncTypeDelete:
 		// if service is successfully deleted, remove it from cache
@@ -516,6 +517,9 @@ func (l4c *L4Controller) publishMetrics(result *loadbalancers.L4ILBSyncResult, n
 			}
 		}
 		l4metrics.PublishILBSyncMetrics(result.Error == nil, result.SyncType, result.GCEResourceInError, utils.GetErrorType(result.Error), result.StartTime)
+		if l4c.enableDualStack {
+			l4metrics.PublishL4ILBDualStackSyncLatency(result.Error == nil, result.SyncType, result.DualStackMetricsState.IPFamilies, result.StartTime)
+		}
 	default:
 		klog.Warningf("Unknown sync type %q, skipping metrics", result.SyncType)
 	}
diff --git a/pkg/l4lb/metrics/metrics.go b/pkg/l4lb/metrics/metrics.go
index 274696ef48..57f4a64c19 100644
--- a/pkg/l4lb/metrics/metrics.go
+++ b/pkg/l4lb/metrics/metrics.go
@@ -27,6 +27,7 @@ const (
 	statusSuccess                                  = "success"
 	statusError                                    = "error"
 	L4ilbLatencyMetricName                         = "l4_ilb_sync_duration_seconds"
+	L4ILBDualStackLatencyMetricName                = "l4_ilb_dualstack_sync_duration_seconds"
 	L4ilbErrorMetricName                           = "l4_ilb_sync_error_count"
 	L4netlbLatencyMetricName                       = "l4_netlb_sync_duration_seconds"
 	L4netlbErrorMetricName                         = "l4_netlb_sync_error_count"
@@ -39,7 +40,8 @@ var (
 		"sync_result", // result of the sync
 		"sync_type",   // whether this is a new service, update or delete
 	}
-	l4LBSyncErrorMetricLabels = []string{
+	l4LBDualStackSyncLatencyMetricsLabels = append(l4LBSyncLatencyMetricsLabels, "ipfamilies")
+	l4LBSyncErrorMetricLabels             = []string{
 		"sync_type",    // whether this is a new service, update or delete
 		"gce_resource", // The GCE resource whose update caused the error
 		// max number of values for error_type = 18 k8s error reasons + 60 http status errors.
@@ -56,6 +58,16 @@ var (
 		},
 		l4LBSyncLatencyMetricsLabels,
 	)
+	l4ILBDualStackSyncLatency = prometheus.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Name: L4ILBDualStackLatencyMetricName,
+			Help: "Latency of an L4 ILB DualStack Sync",
+			// custom buckets - [0.9375s, 1.875s, 3.75s, 7.5s, 30s, 60s, 120s, 240s(4min), 480s(8min), 960s(16m), 3840s(64min), 7680s(128m) +Inf]
+			// using funny starter bucket, 0.9375s will only add buckets to existing metric, this is a safe operation in most time series db
+			Buckets: prometheus.ExponentialBuckets(0.9375, 2, 12),
+		},
+		l4LBDualStackSyncLatencyMetricsLabels,
+	)
 	l4ILBSyncErrorCount = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
 			Name: L4ilbErrorMetricName,
@@ -63,7 +75,7 @@ var (
 		},
 		l4LBSyncErrorMetricLabels,
 	)
-	// l4ILBSyncLatency is a metric that represents the time spent processing L4NetLB service.
+	// l4NetLBSyncLatency is a metric that represents the time spent processing L4NetLB service.
 	// The metric is labeled with synchronization type and its result.
 	l4NetLBSyncLatency = prometheus.NewHistogramVec(
 		prometheus.HistogramOpts{
@@ -104,6 +116,8 @@ var (
 func init() {
 	klog.V(3).Infof("Registering L4 ILB controller metrics %v, %v", l4ILBSyncLatency, l4ILBSyncErrorCount)
 	prometheus.MustRegister(l4ILBSyncLatency, l4ILBSyncErrorCount)
+	klog.V(3).Infof("Registering L4 ILB DualStack controller metrics %v", l4ILBDualStackSyncLatency)
+	prometheus.MustRegister(l4ILBDualStackSyncLatency)
 	klog.V(3).Infof("Registering L4 NetLB controller metrics %v, %v", l4NetLBSyncLatency, l4NetLBSyncErrorCount)
 	prometheus.MustRegister(l4NetLBSyncLatency, l4NetLBSyncErrorCount)
 	klog.V(3).Infof("Registering L4 healthcheck failures count metric: %v", l4FailedHealthCheckCount)
@@ -127,6 +141,15 @@ func publishL4ILBSyncLatency(success bool, syncType string, startTime time.Time)
 	l4ILBSyncLatency.WithLabelValues(status, syncType).Observe(time.Since(startTime).Seconds())
 }
 
+// PublishL4ILBDualStackSyncLatency exports the given sync latency datapoint.
+func PublishL4ILBDualStackSyncLatency(success bool, syncType, ipFamilies string, startTime time.Time) {
+	status := statusSuccess
+	if !success {
+		status = statusError
+	}
+	l4ILBDualStackSyncLatency.WithLabelValues(status, syncType).Observe(time.Since(startTime).Seconds())
+}
+
 // publishL4ILBSyncLatency exports the given sync latency datapoint.
 func publishL4ILBSyncErrorCount(syncType, gceResource, errorType string) {
 	l4ILBSyncErrorCount.WithLabelValues(syncType, gceResource, errorType).Inc()