openclaw - 💡(How to fix) Fix [Bug] Gateway receives periodic SIGTERM at exactly 10:00 AM daily + every ~448 seconds (macOS) [1 comments, 2 participants]

Environment

• macOS: Darwin 25.4.0 (arm64, Mac mini M4 Pro 64GB)
• OpenClaw: 2026.5.7
• Node: v22.22.2
• Deployment: LaunchAgent (ai.openclaw.gateway.plist) with KeepAlive: true
• Gateway PID: running via launchd (parent PID 1)

Bug Description

Gateway receives SIGTERM at exactly 10:00 AM Beijing time (02:00 UTC) every day, then continues receiving SIGTERM every ~448 seconds (7m28s) throughout the entire day. The launchd KeepAlive: true correctly restarts the gateway each time, but this is noisy.

Key characteristics:

1. Triggers precisely at 10:00 AM CST every day
2. Then repeats every ~448 seconds for the rest of the day
3. SIGTERM source is always launchd (PID 1)
4. No crontab / at jobs / launchd calendar tasks found

Confirmed SIGTERM Timestamps (Beijing Time)

2026-05-05 10:00:58 + 10:01:00 (2x rapid)
2026-05-06 10:00:38 + 10:00:40 (2x rapid)
2026-05-07 10:00:27 + 10:00:29 (2x rapid)
2026-05-08 10:00:09 + 10:00:11 (2x rapid)
2026-05-09 10:00:30 + 10:00:32 (2x rapid)
2026-05-10 10:00:35 + 10:00:36 (2x rapid)
2026-05-11 10:00:47 + 10:00:48 (2x rapid)
2026-05-12 10:00:34 + 10:00:35 (2x rapid)
2026-05-13 10:00:58 + 10:01:00 (2x rapid)

Detailed Interval Analysis (May 12)

After the 10:00 AM trigger, SIGTERM continues throughout the day at highly regular intervals:

10:00:34 → 11:51:33  (448s gap)
11:51:33 → 11:56:11  (278s - anomalous)
11:56:11 → 12:03:48  (457s)
12:03:48 → 12:11:24  (456s)
...continues until 20:47:55

Investigation Already Done

1. ✅ No crontab or at jobs
2. ✅ No launchd calendar-based scheduled tasks
3. ✅ No external scripts sending SIGTERM
4. ✅ No macOS watchdogd involvement
5. ✅ No OpenClaw internal config reload triggers at 10am
6. ✅ No Discord health-monitor correlation
7. ✅ launchd KeepAlive correctly restarts gateway each time
8. ✅ Only happens to OpenClaw gateway, not other launchd services

Preliminary Conclusions

1. Not a crash-loop: launchd successfully restarts gateway every time
2. Not config hot-reload: No reload reasons found in gateway logs at SIGTERM time
3. Not an external scheduler: No cron/launchd calendar tasks identified
4. Not watchdogd: macOS watchdogd does not send SIGTERM to launchd-managed processes

Possible causes:

• Some internal OpenClaw periodic mechanism (health check? session cleanup?)
• Conflict with a macOS system scheduled task at 02:00 UTC

Diagnostics

Full logs and diagnostics bundle available on request.

Related Issues

• #43035 - Gateway self-restart via SIGTERM loses launchd service registration (macOS)
• #78742 - task-139: audit gateway restart attribution
• #73849 - fix(daemon): use KeepAlive SuccessfulExit=false to prevent launchd crash-loop