fix: harden websocket heartbeat under load

server/metrics/metrics.go

@@ -71,6 +71,14 @@ var (
 		Name: "im_ws_disconnect_total",
 		Help: "WebSocket read-loop disconnects by pod and bounded reason.",
 	}, []string{"pod", "reason"})
+	IMWSHeartbeatTotal = prometheus.NewCounterVec(prometheus.CounterOpts{
+		Name: "im_ws_heartbeat_total",
+		Help: "WebSocket heartbeat events by pod and event.",
+	}, []string{"pod", "event"})
+	IMWSHeartbeatSeconds = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "im_ws_heartbeat_seconds",
+		Help: "Effective WebSocket heartbeat timing by pod and setting.",
+	}, []string{"pod", "setting"})
 	IMWSSendQueueCurrent = prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: "im_ws_send_queue_current",
 		Help: "Current number of queued WebSocket outbound channel entries across sessions on a pod.",
@@ -216,7 +224,8 @@ var (
 
 func init() {
 	Registry.MustRegister(WSConnectionsActive, IMWSConnectionsActive, IMWSWriteTotal, IMWSWriteDuration,
-		IMWSWriteBytesTotal, IMWSDisconnectTotal, IMWSSendQueueCurrent, IMWSSendQueueCapacity,
+		IMWSWriteBytesTotal, IMWSDisconnectTotal, IMWSHeartbeatTotal, IMWSHeartbeatSeconds,
+		IMWSSendQueueCurrent, IMWSSendQueueCapacity,
 		IMWSSendQueueEnqueueTotal, IMWSSendQueueDepth, IMWSSendQueueWaitDuration,
 		IMClusterLivePeers, IMClusterExpectedPeers,
 		HTTPRequestsTotal, HTTPRequestDuration, DBQueryDuration,
@@ -332,6 +341,16 @@ func IncWSDisconnect(reason string) {
 	IMWSDisconnectTotal.WithLabelValues(metricPodName(), normalizeLabel(reason)).Inc()
 }
 
+func IncWSHeartbeat(event string) {
+	IMWSHeartbeatTotal.WithLabelValues(metricPodName(), normalizeLabel(event)).Inc()
+}
+
+func SetWSHeartbeatConfig(pingPeriod, pongWait time.Duration) {
+	pod := metricPodName()
+	IMWSHeartbeatSeconds.WithLabelValues(pod, "ping_period").Set(pingPeriod.Seconds())
+	IMWSHeartbeatSeconds.WithLabelValues(pod, "pong_wait").Set(pongWait.Seconds())
+}
+
 func ObserveWSSendQueueEnqueue(depth, capacity int, accepted bool) {
 	result := "ok"
 	if !accepted {