openclaw - 💡(How to fix) Fix [Bug]: OpenClaw 4.29 severe event loop block, API timeout, stuck session & Bonjour mDNS loop [2 comments, 2 participants]

Labels

bug, critical, gateway, windows, bonjour, performance

Description

Version: OpenClaw 4.29 (latest) OS: Windows 11 Summary: After starting OpenClaw 4.29, the Node.js event loop is severely blocked, causing extreme API timeout, long stuck sessions, and infinite Bonjour/mDNS advertisement retries that consume resources and create a vicious cycle.

Actual Behavior

1. sessions.abort API throws INVALID_REQUEST due to unexpected extra field sessionKey.
2. Massive API response timeout:
   • chat.history: response time rises from 5s up to 10s+
   • device.pair.list: extreme timeout up to 356s
   • sessions.list: common timeout 10s ~ 67s
3. Sessions stuck permanently in processing state, queue depth piling up, skip auto-recovery due to active work.
4. Critical event loop delay: eventLoopDelayP99Ms up to 77577ms, CPU core ratio 0.7~0.89, main thread blocked completely.
5. Bonjour/mDNS keeps switching between probing and unannounced, auto-disables after 3 failed restarts but still retries infinitely.
6. Model warmup timeout: ali/qwen3.6-plus warmup timed out after 5000ms.
7. WebSocket disconnect with code=1006, local connection closed before handshake.

Expected Behavior

1. All API response latency should be within 2s.
2. Event loop delay should stay below 100ms without liveness warning.
3. Bonjour/mDNS should stop retrying gracefully instead of blocking main thread.
4. Sessions should complete normally without permanent stuck state.
5. Model warmup should finish successfully without timeout.

Steps to Reproduce

1. Install OpenClaw 4.29 on Windows 11.
2. Start gateway with default config.
3. Call APIs repeatedly: chat.history, device.pair.list, sessions.abort with extra sessionKey param.
4. Check logs, observe event loop warning, huge timeout, stuck session and Bonjour loop.

Relevant Logs

sessions.abort 9726ms errorCode=INVALID_REQUEST errorMessage=invalid sessions.abort params: at root: unexpected property 'sessionKey'

liveness warning: reasons=event_loop_delay,event_loop_utilization eventLoopDelayP99Ms=50029.7 eventLoopDelayMaxMs=50029.7

bonjour: disabling advertiser after 3 failed restarts (service stuck in unannounced)

stuck session: state=processing age=773s queueDepth=1 recovery=checking

startup model warmup timed out after 5000ms; continuing without waiting

ws closed before connect code=1006

Environment

• OS: Windows 11 Pro
• OpenClaw: 4.29 Latest
• Model: ali/qwen3.6-plus
• CPU: Intel Core Ultra 9 275HX
• RAM: 192GB
• GPU: NVIDIA RTX 5090 24GB

Temporary Workaround

1. Disable mDNS/Bonjour: set env OPENCLAW_DISABLE_BONJOUR=1 or add discovery.mdns.mode: off in config.
2. Manually terminate all long-time processing stuck sessions.
3. Remove redundant sessionKey param when calling sessions.abort.
4. Increase model warmup timeout or use local cached model to avoid network warmup delay.

Proposed Fix

1. Add strict API parameter validation to reject unknown fields like sessionKey.
2. Limit Bonjour retry times and add timeout isolation to avoid blocking event loop.
3. Add automatic timeout kill for stuck sessions, optimize history session/device list query performance.
4. Extend default model warmup timeout and support local model cache strategy.