fix: harden high-fanout delivery under load

server/hdl_websock.go

@@ -70,14 +70,23 @@ func (sess *Session) readLoop() {
 	}()
 
 	sess.ws.SetReadLimit(globals.maxMessageSize)
-	sess.ws.SetReadDeadline(time.Now().Add(pongWait))
-	sess.ws.SetPongHandler(func(string) error {
+	refreshReadDeadline := func(event, presenceSource string) {
 		atomic.StoreInt64(&sess.lastAction, types.TimeNow().UnixNano())
-		immetrics.IncWSHeartbeat("pong_received")
-		sess.refreshPresence("ws.pong")
-		sess.ws.SetReadDeadline(time.Now().Add(pongWait))
+		immetrics.IncWSHeartbeat(event)
+		if presenceSource != "" {
+			sess.refreshPresence(presenceSource)
+		}
+		_ = sess.ws.SetReadDeadline(time.Now().Add(pongWait))
+	}
+	_ = sess.ws.SetReadDeadline(time.Now().Add(pongWait))
+	sess.ws.SetPongHandler(func(string) error {
+		refreshReadDeadline("pong_received", "ws.pong")
 		return nil
 	})
+	sess.ws.SetPingHandler(func(appData string) error {
+		refreshReadDeadline("ping_received", "")
+		return sess.ws.WriteControl(websocket.PongMessage, []byte(appData), time.Now().Add(writeWait))
+	})
 
 	for {
 		// Read a ClientComMessage
@@ -90,6 +99,11 @@ func (sess *Session) readLoop() {
 			}
 			return
 		}
+		// Any successfully read client frame proves that the connection is alive.
+		// This prevents a busy client from timing out merely because its Pong was
+		// delayed behind application traffic.
+		atomic.StoreInt64(&sess.lastAction, types.TimeNow().UnixNano())
+		_ = sess.ws.SetReadDeadline(time.Now().Add(pongWait))
 		statsInc("IncomingMessagesWebsockTotal", 1)
 		sess.dispatchRaw(raw)
 	}