chore(push-gateway): ai02 module updates - config, hub, ws, kafka, health, docs

services/push-gateway/internal/hub/hub.go

@@ -172,6 +172,15 @@ func (h *Hub) HasUser(userID string) bool {
 	return ok && len(conns) > 0
 }
 
+// IsClosing reports whether CloseAll has been invoked. The /readyz probe uses
+// this to return 503 during graceful shutdown so the load balancer stops
+// sending new WebSocket upgrades while existing connections drain.
+func (h *Hub) IsClosing() bool {
+	h.mu.RLock()
+	defer h.mu.RUnlock()
+	return h.closing
+}
+
 // Register adds a new connection for userID. Returns ErrTooManyConnections
 // when the per-user cap is exceeded and ErrHubClosing when the Hub is shutting
 // down. The returned Connection must be Unregistered by the caller on close.
@@ -262,7 +271,13 @@ func (h *Hub) SendToUser(userID string, message []byte) int {
 // Cross-instance broadcast is handled separately via Redis Pub/Sub.
 func (h *Hub) Broadcast(message []byte) int {
 	h.mu.RLock()
-	snapshot := make([]*Connection, 0, h.ActiveConnections())
+	// Count inline: calling ActiveConnections() here would re-enter RLock
+	// while we already hold it, which can deadlock if a writer is waiting.
+	total := 0
+	for _, conns := range h.clients {
+		total += len(conns)
+	}
+	snapshot := make([]*Connection, 0, total)
 	for _, conns := range h.clients {
 		for _, c := range conns {
 			snapshot = append(snapshot, c)
@@ -285,7 +300,13 @@ func (h *Hub) Broadcast(message []byte) int {
 func (h *Hub) CloseAll() {
 	h.mu.Lock()
 	h.closing = true
-	all := make([]*Connection, 0, h.ActiveConnections())
+	// Count inline instead of calling ActiveConnections() to avoid reentrant
+	// RLock while holding the write lock (Go's sync.RWMutex is non-reentrant).
+	total := 0
+	for _, conns := range h.clients {
+		total += len(conns)
+	}
+	all := make([]*Connection, 0, total)
 	for _, conns := range h.clients {
 		for _, c := range conns {
 			all = append(all, c)