Bug Description

OpenClaw Gateway enters an infinite restart loop after an agent run timeout, attempting to restart ~250 times over 42 minutes with a fixed 10-11 second interval, rendering the system completely unresponsive.

Environment

• OpenClaw version: 2026.2.24
• OS: macOS 24.5.0 (arm64)
• Node: v22.22.0
• Gateway mode: local (ws://127.0.0.1:18789)
• Launch method: LaunchAgent (ai.openclaw.gateway.plist)

Timeline (2026-02-26)

16:08:15 - First anomaly: embedded run timeout (runId=slug-gen-1772093280356)
16:08:47 - Restart loop begins
16:50:27 - Restart loop ends (42 minutes total)

Error Log Pattern

Gateway failed to start: gateway already running (pid 40836); lock timeout after 5000ms
Port 18789 is already in use.
- pid 40836 tingjing: openclaw-gateway (127.0.0.1:18789)
- Gateway already running locally. Stop it (openclaw gateway stop) or use a different port.

Source files: gateway-cli-CHQJpgpN.js:1009 and gateway-cli-CK1Ri8SQ.js

Root Cause Analysis

1. Watchdog detection failure: After an agent timeout at 16:08:15, the internal health check mechanism was triggered
2. State detection bug: The restart logic does not correctly detect that Gateway is already running
3. No backoff: Fixed 10-11 second interval between restart attempts, no exponential backoff
4. No max retry limit: Loop continued for 42 minutes (~250 attempts) with no upper bound

Impact

• Gateway completely unresponsive for 42 minutes
• All Feishu/Lark messages unprocessable
• Log file bloat (gateway.err.log reached 156MB)
• System resource waste

Suggested Fixes

Short-term

1. Add running-state detection in gateway start — if already running, return success immediately
2. Add exponential backoff (1s, 2s, 4s, 8s, 16s, 30s, 60s...)
3. Add max retry limit (e.g., stop after 10 attempts)

Long-term

1. Implement proper health check (HTTP endpoint or WebSocket ping/pong)
2. Use openclaw gateway status in watchdog logic
3. Consider leveraging systemd/LaunchAgent KeepAlive instead of custom restart logic

Log File Locations

• Main log: /tmp/openclaw/openclaw-YYYY-MM-DD.log
• Error log: ~/.openclaw/logs/gateway.err.log
• Gateway log: ~/.openclaw/logs/gateway.log

Workaround

I built an external watchdog script (gateway-watchdog-v2.sh) that:

• Health-checks every 10 seconds
• Detects restart loops and backs off
• Attempts auto-repair via Codex CLI for config issues
• Falls back to config rollback if repair fails