fix: robust WebSocket reconnection after daemon restart

Matthias Nott

2026-03-08 6cbe1fb2618af557262a8717c494e7958494bf2d

 services/websocket.ts
..
..
@@ -4,12 +4,14 @@
4
4
 type WebSocketMessage = Record<string, unknown>;
5
5
 type MessageCallback = (data: WebSocketMessage) => void;
6
6
 type StatusCallback = () => void;
7
+type ReconnectCallback = (attempt: number) => void;
7
8
 type ErrorCallback = (error: Event) => void;
8
9
 
9
10
 interface WebSocketClientOptions {
10
11
   onMessage?: MessageCallback;
11
12
   onOpen?: StatusCallback;
12
13
   onClose?: StatusCallback;
14
+  onReconnecting?: ReconnectCallback;
13
15
   onError?: ErrorCallback;
14
16
 }
15
17
 
..
..
@@ -17,6 +19,7 @@
17
19
 const MAX_RECONNECT_DELAY = 30000;
18
20
 const RECONNECT_MULTIPLIER = 2;
19
21
 const LOCAL_TIMEOUT = 2500;
22
+const CONNECT_TIMEOUT = 5000; // max time for any single connection attempt
20
23
 const HEARTBEAT_INTERVAL = 30000; // 30s ping to detect zombie sockets
21
24
 
22
25
 export class WebSocketClient {
..
..
@@ -25,11 +28,14 @@
25
28
   private urlIndex: number = 0;
26
29
   private reconnectDelay: number = INITIAL_RECONNECT_DELAY;
27
30
   private reconnectTimer: ReturnType<typeof setTimeout> | null = null;
31
+  private connectTimer: ReturnType<typeof setTimeout> | null = null;
28
32
   private localTimer: ReturnType<typeof setTimeout> | null = null;
29
33
   private heartbeatTimer: ReturnType<typeof setInterval> | null = null;
30
34
   private lastMessageAt: number = 0; // timestamp of last received message (any type)
31
35
   private shouldReconnect: boolean = false;
32
36
   private connected: boolean = false;
37
+  private wasConnected: boolean = false; // true if we ever connected (for reconnect logic)
38
+  private reconnectAttempt: number = 0;
33
39
   private callbacks: WebSocketClientOptions = {};
34
40
 
35
41
   constructor() {
..
..
@@ -68,6 +74,7 @@
68
74
   private cleanup() {
69
75
     if (this.localTimer) { clearTimeout(this.localTimer); this.localTimer = null; }
70
76
     if (this.reconnectTimer) { clearTimeout(this.reconnectTimer); this.reconnectTimer = null; }
77
+    if (this.connectTimer) { clearTimeout(this.connectTimer); this.connectTimer = null; }
71
78
     this.stopHeartbeat();
72
79
     if (this.ws) {
73
80
       const old = this.ws;
..
..
@@ -139,9 +146,22 @@
139
146
       }, LOCAL_TIMEOUT);
140
147
     }
141
148
 
149
+    // Timeout: if this connection attempt doesn't succeed within CONNECT_TIMEOUT,
150
+    // treat it as failed. Prevents hanging on unreachable remote URLs.
151
+    this.connectTimer = setTimeout(() => {
152
+      this.connectTimer = null;
153
+      if (ws === this.ws && !this.connected) {
154
+        // Force close and let onclose trigger reconnect
155
+        try { ws.close(); } catch { /* ignore */ }
156
+      }
157
+    }, CONNECT_TIMEOUT);
158
+
142
159
     ws.onopen = () => {
143
160
       if (ws !== this.ws) return; // stale
161
+      if (this.connectTimer) { clearTimeout(this.connectTimer); this.connectTimer = null; }
144
162
       this.connected = true;
163
+      this.wasConnected = true;
164
+      this.reconnectAttempt = 0;
145
165
       if (this.localTimer) { clearTimeout(this.localTimer); this.localTimer = null; }
146
166
       this.reconnectDelay = INITIAL_RECONNECT_DELAY;
147
167
       this.startHeartbeat();
..
..
@@ -178,15 +198,19 @@
178
198
 
179
199
   private scheduleReconnect() {
180
200
     if (this.reconnectTimer) clearTimeout(this.reconnectTimer);
201
+    this.reconnectAttempt++;
202
+    // Signal reconnecting state to UI so it shows yellow/orange, not red
203
+    this.callbacks.onReconnecting?.(this.reconnectAttempt);
181
204
     this.reconnectTimer = setTimeout(() => {
182
205
       this.reconnectTimer = null;
183
206
       this.reconnectDelay = Math.min(
184
207
         this.reconnectDelay * RECONNECT_MULTIPLIER,
185
208
         MAX_RECONNECT_DELAY
186
209
       );
187
-      // Alternate between URLs on each reconnect attempt
210
+      // Always try local first (daemon restarts are local), then remote on retry
188
211
       if (this.urls.length > 1) {
189
-        this.urlIndex = this.urlIndex === 0 ? 1 : 0;
212
+        // First attempt: local. Second: remote. Then alternate.
213
+        this.urlIndex = this.reconnectAttempt % 2 === 1 ? 0 : 1;
190
214
       }
191
215
       this.tryUrl();
192
216
     }, this.reconnectDelay);