Fix control plane metrics instrumentation gaps (#384)

1. Add duckgres_hot_idle_workers gauge — hot-idle workers were
   invisible to metrics (not counted by observeWarmPoolLifecycleGauges)

2. Add connections_open to control plane — connectionsGauge was only
   incremented in standalone mode (server.go), not in the control
   plane's handleConnection. Export Inc/Dec helpers from server pkg.

3. Update org_sessions_active on session create/destroy — was only
   updated on admin API pull (AllOrgStats), not on every connection.
   Now incremented after CreateSession and decremented in defer after
   DestroySession.

4. Add observeOrgSessionsActive stub for non-kubernetes builds.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: EDsCODE

controlplane/control.go

@@ -603,6 +603,8 @@ func sessionCreationErrorResponse(err error) (code string, message string) {
 func (cp *ControlPlane) handleConnection(conn net.Conn) {
 	remoteAddr := conn.RemoteAddr()
 	slog.Info("Connection accepted.", "remote_addr", remoteAddr)
+	server.IncrementOpenConnections()
+	defer server.DecrementOpenConnections()
 
 	releaseRateLimit, msg := server.BeginRateLimitedAuthAttempt(cp.rateLimiter, remoteAddr)
 	if msg != "" {
@@ -864,7 +866,15 @@ func (cp *ControlPlane) handleConnection(conn net.Conn) {
 		_ = writer.Flush()
 		return
 	}
-	defer sessions.DestroySession(pid)
+	if orgID != "" {
+		observeOrgSessionsActive(orgID, sessions.SessionCount())
+	}
+	defer func() {
+		sessions.DestroySession(pid)
+		if orgID != "" {
+			observeOrgSessionsActive(orgID, sessions.SessionCount())
+		}
+	}()
 
 	// Register the TCP connection so OnWorkerCrash can close it to unblock
 	// the message loop if the backing worker dies.

controlplane/flight_ingress_metrics_stub.go

@@ -0,0 +1,5 @@
+//go:build !kubernetes
+
+package controlplane
+
+func observeOrgSessionsActive(string, int) {}

controlplane/warm_pool_metrics.go

@@ -31,6 +31,11 @@ var hotWorkersGauge = promauto.NewGauge(prometheus.GaugeOpts{
 	Help: "Number of activated, tenant-bound workers serving sessions",
 })
 
+var hotIdleWorkersGauge = promauto.NewGauge(prometheus.GaugeOpts{
+	Name: "duckgres_hot_idle_workers",
+	Help: "Number of activated workers retaining org assignment between sessions",
+})
+
 var drainingWorkersGauge = promauto.NewGauge(prometheus.GaugeOpts{
 	Name: "duckgres_draining_workers",
 	Help: "Number of workers currently draining sessions before retirement",
@@ -76,7 +81,7 @@ const (
 // observeWarmPoolLifecycleGauges recalculates all lifecycle gauges from the
 // current worker map. Must be called with p.mu held (at least RLock).
 func observeWarmPoolLifecycleGauges(workers map[int]*ManagedWorker) {
-	var idle, reserved, activating, hot, draining int
+	var idle, reserved, activating, hot, hotIdle, draining int
 	for _, w := range workers {
 		select {
 		case <-w.done:
@@ -92,6 +97,8 @@ func observeWarmPoolLifecycleGauges(workers map[int]*ManagedWorker) {
 			activating++
 		case WorkerLifecycleHot:
 			hot++
+		case WorkerLifecycleHotIdle:
+			hotIdle++
 		case WorkerLifecycleDraining:
 			draining++
 		}
@@ -100,6 +107,7 @@ func observeWarmPoolLifecycleGauges(workers map[int]*ManagedWorker) {
 	reservedWorkersGauge.Set(float64(reserved))
 	activatingWorkersGauge.Set(float64(activating))
 	hotWorkersGauge.Set(float64(hot))
+	hotIdleWorkersGauge.Set(float64(hotIdle))
 	drainingWorkersGauge.Set(float64(draining))
 }
 

server/server.go

@@ -53,6 +53,13 @@ var connectionsGauge = promauto.NewGauge(prometheus.GaugeOpts{
 	Help: "Number of currently open client connections",
 })
 
+// IncrementOpenConnections increments the open connections gauge.
+// Used by the control plane which handles connections separately from the standalone server.
+func IncrementOpenConnections() { connectionsGauge.Inc() }
+
+// DecrementOpenConnections decrements the open connections gauge.
+func DecrementOpenConnections() { connectionsGauge.Dec() }
+
 var queryDurationHistogram = promauto.NewHistogram(prometheus.HistogramOpts{
 	Name:    "duckgres_query_duration_seconds",
 	Help:    "Query execution duration in seconds",