perf: batch group unread fanout

server/metrics/metrics.go

@@ -102,6 +102,19 @@ var (
 		Name: "im_cluster_proxy_response_total",
 		Help: "Cluster proxy responses by pod and result.",
 	}, []string{"pod", "result"})
+	InternalQueueEnqueueTotal = prometheus.NewCounterVec(prometheus.CounterOpts{
+		Name: "im_internal_queue_enqueue_total",
+		Help: "Internal message queue enqueue attempts by pod, queue, and result.",
+	}, []string{"pod", "queue", "result"})
+	InternalQueueDepth = prometheus.NewHistogramVec(prometheus.HistogramOpts{
+		Name:    "im_internal_queue_depth",
+		Help:    "Internal message queue depth observed after enqueue attempts by pod and queue.",
+		Buckets: []float64{0, 1, 4, 16, 32, 64, 128, 192, 256, 512, 1024, 2048, 4096, 8192},
+	}, []string{"pod", "queue"})
+	InternalQueueCapacity = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+		Name: "im_internal_queue_capacity",
+		Help: "Configured internal message queue capacity by pod and queue.",
+	}, []string{"pod", "queue"})
 	MessagePersistDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
 		Name:    "im_message_persist_duration_seconds",
 		Help:    "Kafka message persistence duration by message kind and result.",
@@ -135,7 +148,8 @@ func init() {
 		HTTPRequestsTotal, HTTPRequestDuration, DBQueryDuration,
 		RealtimeFanoutEventsTotal, RealtimeFanoutDeliveryDuration, ClusterP2MQueueLength,
 		ClusterP2MQueueCapacity, ClusterP2MEnqueueTotal, ClusterRPCInflight, ClusterRPCDuration,
-		ClusterRouteTotal, ClusterProxyResponseTotal, MessagePersistDuration, MessagePublishStageDuration,
+		ClusterRouteTotal, ClusterProxyResponseTotal, InternalQueueEnqueueTotal, InternalQueueDepth,
+		InternalQueueCapacity, MessagePersistDuration, MessagePublishStageDuration,
 		KafkaMessagePersistTotal)
 	Registry.MustRegister(KafkaConsumerLag, KafkaBacklogOldestAge, KafkaDeadLettersTotal)
 }
@@ -177,6 +191,18 @@ func ObserveRealtimeFanoutDelivery(kind, targetType string, timestamp time.Time)
 		Observe(time.Since(timestamp).Seconds())
 }
 
+func ObserveInternalQueueEnqueue(queue string, depth, capacity int, accepted bool) {
+	result := "ok"
+	if !accepted {
+		result = "full"
+	}
+	pod := metricPodName()
+	queue = normalizeLabel(queue)
+	InternalQueueEnqueueTotal.WithLabelValues(pod, queue, result).Inc()
+	InternalQueueDepth.WithLabelValues(pod, queue).Observe(float64(depth))
+	InternalQueueCapacity.WithLabelValues(pod, queue).Set(float64(capacity))
+}
+
 func normalizeLabel(value string) string {
 	if value == "" {
 		return "unknown"