Merge pull request #509 from netease-youdao/fix/client-side-timeout-watchdog

fix: add client-side timeout watchdog for stuck running state

src/main/i18n.ts

@@ -23,6 +23,9 @@ const translations: Record<LanguageType, Record<string, string>> = {
 
     // Session titles (created by ChannelSessionSync)
     cronSessionPrefix: '定时',
+
+    // Timeout hint
+    taskTimedOut: '[任务超时] 任务因超过最大允许时长而被自动停止。你可以继续对话以从中断处继续。',
   },
   en: {
     // Tray menu
@@ -33,6 +36,9 @@ const translations: Record<LanguageType, Record<string, string>> = {
 
     // Session titles
     cronSessionPrefix: 'Cron',
+
+    // Timeout hint
+    taskTimedOut: '[Task timed out] The task was automatically stopped because it exceeded the maximum allowed duration. You can continue the conversation to pick up where it left off.',
   },
 };
 

src/main/libs/agentEngine/openclawRuntimeAdapter.ts

@@ -27,6 +27,8 @@ import {
   extractGatewayMessageText,
 } from '../openclawHistory';
 import { buildOpenClawLocalTimeContextPrompt } from '../openclawLocalTimeContextPrompt';
+import { OPENCLAW_AGENT_TIMEOUT_SECONDS } from '../openclawConfigSync';
+import { t } from '../../i18n';
 
 const OPENCLAW_GATEWAY_TOOL_EVENTS_CAP = 'tool-events';
 const BRIDGE_MAX_MESSAGES = 20;
@@ -116,6 +118,8 @@ type ActiveTurn = {
   bufferedChatPayloads: BufferedChatEvent[];
   /** Agent events buffered while pendingUserSync is true. */
   bufferedAgentPayloads: BufferedAgentEvent[];
+  /** Client-side timeout watchdog timer (fallback for missing gateway abort events). */
+  timeoutTimer?: ReturnType<typeof setTimeout>;
 };
 
 type BufferedChatEvent = {
@@ -554,6 +558,14 @@ export class OpenClawRuntimeAdapter extends EventEmitter implements CoworkRuntim
   private pendingMessageUpdateTimer: Map<string, ReturnType<typeof setTimeout>> = new Map();
   private static readonly MESSAGE_UPDATE_THROTTLE_MS = 100;
 
+  /**
+   * Server-side agent timeout in seconds (mirrors agents.defaults.timeoutSeconds in openclaw config).
+   * Used to set a client-side fallback timer that fires slightly after the server timeout,
+   * so LobsterAI can recover even when the gateway fails to deliver the abort event.
+   */
+  agentTimeoutSeconds = OPENCLAW_AGENT_TIMEOUT_SECONDS;
+  private static readonly CLIENT_TIMEOUT_GRACE_MS = 30_000;
+
   constructor(store: CoworkStore, engineManager: OpenClawEngineManager) {
     super();
     this.store = store;
@@ -987,6 +999,13 @@ export class OpenClawRuntimeAdapter extends EventEmitter implements CoworkRuntim
     });
     this.sessionIdByRunId.set(runId, sessionId);
 
+    // Start client-side timeout watchdog.
+    // OpenClaw gateway has a known issue where embedded run timeouts may not
+    // produce a WS abort/final event (the subscription is torn down before the
+    // lifecycle event fires). This timer fires slightly after the server-side
+    // timeout to recover the UI from a stuck "running" state.
+    this.startTurnTimeoutWatchdog(sessionId);
+
     const client = this.requireGatewayClient();
     try {
       const attachments = options.imageAttachments?.length
@@ -2285,6 +2304,14 @@ export class OpenClawRuntimeAdapter extends EventEmitter implements CoworkRuntim
   private handleChatAborted(sessionId: string, turn: ActiveTurn): void {
     this.store.updateSession(sessionId, { status: 'idle' });
     if (!turn.stopRequested) {
+      // The run was aborted without user request — most likely a timeout.
+      // Add a visible hint so the user knows the task was interrupted.
+      const hintMessage = this.store.addMessage(sessionId, {
+        type: 'assistant',
+        content: t('taskTimedOut'),
+        metadata: { isTimeout: true },
+      });
+      this.emit('message', sessionId, hintMessage);
       this.emit('complete', sessionId, turn.runId);
     }
     const abortedSessionKey = turn.sessionKey;
@@ -3162,6 +3189,11 @@ export class OpenClawRuntimeAdapter extends EventEmitter implements CoworkRuntim
   private cleanupSessionTurn(sessionId: string): void {
     const turn = this.activeTurns.get(sessionId);
     if (turn) {
+      // Clear client-side timeout watchdog
+      if (turn.timeoutTimer) {
+        clearTimeout(turn.timeoutTimer);
+        turn.timeoutTimer = undefined;
+      }
       // Cancel any pending throttled messageUpdate timer for this turn
       if (turn.assistantMessageId) {
         this.clearPendingMessageUpdate(turn.assistantMessageId);
@@ -3182,6 +3214,27 @@ export class OpenClawRuntimeAdapter extends EventEmitter implements CoworkRuntim
     this.reCreatedChannelSessionIds.delete(sessionId);
   }
 
+  /**
+   * Start a client-side timeout watchdog for a turn.
+   * Fires after the server-side timeout + grace period, recovering the UI
+   * if the gateway fails to deliver the abort/final event.
+   */
+  private startTurnTimeoutWatchdog(sessionId: string): void {
+    const turn = this.activeTurns.get(sessionId);
+    if (!turn) return;
+    const timeoutMs = this.agentTimeoutSeconds * 1000
+      + OpenClawRuntimeAdapter.CLIENT_TIMEOUT_GRACE_MS;
+    turn.timeoutTimer = setTimeout(() => {
+      const currentTurn = this.activeTurns.get(sessionId);
+      if (!currentTurn || currentTurn.turnToken !== turn.turnToken) return;
+      console.warn(
+        `[OpenClawRuntime] Client-side timeout watchdog fired for session ${sessionId}, `
+        + `runId=${currentTurn.runId} after ${timeoutMs}ms — gateway did not deliver abort event`,
+      );
+      this.handleChatAborted(sessionId, currentTurn);
+    }, timeoutMs);
+  }
+
   /**
    * Called when a session is deleted from the store.
    * Purges all in-memory references so that new channel messages
@@ -3265,6 +3318,7 @@ export class OpenClawRuntimeAdapter extends EventEmitter implements CoworkRuntim
       this.sessionIdByRunId.set(runId, sessionId);
     }
     this.store.updateSession(sessionId, { status: 'running' });
+    this.startTurnTimeoutWatchdog(sessionId);
 
     // For channel sessions, prefetch user messages before streaming starts
     if (isChannel) {

src/main/libs/openclawConfigSync.ts

@@ -22,6 +22,12 @@ const mapExecutionModeToSandboxMode = (_mode: CoworkExecutionMode): 'off' | 'non
   return 'off';
 };
 
+/**
+ * Default agent timeout in seconds written to openclaw config.
+ * Also used by the runtime adapter's client-side timeout watchdog.
+ */
+export const OPENCLAW_AGENT_TIMEOUT_SECONDS = 3600;
+
 const mapApiTypeToOpenClawApi = (apiType: 'anthropic' | 'openai' | undefined): 'anthropic-messages' | 'openai-completions' => {
   return apiType === 'openai' ? 'openai-completions' : 'anthropic-messages';
 };
@@ -538,6 +544,7 @@ export class OpenClawConfigSync {
       },
       agents: {
         defaults: {
+          timeoutSeconds: OPENCLAW_AGENT_TIMEOUT_SECONDS,
           model: {
             primary: providerSelection.primaryModel,
           },