fix(ws): stabilize node connectivity with ping/pong keepalive

go-backend/internal/ws/server.go

@@ -54,6 +54,12 @@ type pendingRequest struct {
 	ch     chan CommandResult
 }
 
+const (
+	wsPingPeriod = 15 * time.Second
+	wsPongWait   = 45 * time.Second
+	wsWriteWait  = 5 * time.Second
+)
+
 type CommandResult struct {
 	Type    string                 `json:"type"`
 	Success bool                   `json:"success"`
@@ -120,12 +126,19 @@ func (s *Server) handleAdmin(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	cw := &connWrap{conn: conn}
+	_ = conn.SetReadDeadline(time.Now().Add(wsPongWait))
+	conn.SetPongHandler(func(string) error {
+		return conn.SetReadDeadline(time.Now().Add(wsPongWait))
+	})
+	done := make(chan struct{})
+	go startKeepalive(cw, done)
 
 	s.mu.Lock()
 	s.admins[cw] = struct{}{}
 	s.mu.Unlock()
 
 	defer func() {
+		close(done)
 		s.mu.Lock()
 		delete(s.admins, cw)
 		s.mu.Unlock()
@@ -145,6 +158,12 @@ func (s *Server) handleNode(w http.ResponseWriter, r *http.Request, nodeID int64
 		return
 	}
 	cw := &connWrap{conn: conn}
+	_ = conn.SetReadDeadline(time.Now().Add(wsPongWait))
+	conn.SetPongHandler(func(string) error {
+		return conn.SetReadDeadline(time.Now().Add(wsPongWait))
+	})
+	done := make(chan struct{})
+	go startKeepalive(cw, done)
 
 	version := r.URL.Query().Get("version")
 	httpVal := parseIntDefault(r.URL.Query().Get("http"), 0)
@@ -165,6 +184,7 @@ func (s *Server) handleNode(w http.ResponseWriter, r *http.Request, nodeID int64
 	s.broadcastStatus(nodeID, 1)
 
 	defer func() {
+		close(done)
 		needOfflineBroadcast := false
 		s.mu.Lock()
 		current, ok := s.nodes[nodeID]
@@ -442,3 +462,27 @@ func parseIntDefault(v string, fallback int) int {
 	}
 	return x
 }
+
+func startKeepalive(cw *connWrap, done <-chan struct{}) {
+	if cw == nil || cw.conn == nil {
+		return
+	}
+	ticker := time.NewTicker(wsPingPeriod)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-done:
+			return
+		case <-ticker.C:
+			cw.mu.Lock()
+			_ = cw.conn.SetWriteDeadline(time.Now().Add(wsWriteWait))
+			err := cw.conn.WriteMessage(websocket.PingMessage, nil)
+			cw.mu.Unlock()
+			if err != nil {
+				_ = cw.conn.Close()
+				return
+			}
+		}
+	}
+}