Service: ghcr.io/openclaw/openclaw:2026.1.29 Port: 18789
Observed behavior:
- Container starts successfully
- Service binds to 0.0.0.0:18789
- Internal HTTP access returns 200 OK
- WebSocket gateway is live
- Canvas UI works internally
Problem: Zeabur startup probe hits port 18789 before OpenClaw finishes initialization. Probe fails once, service is marked UNHEALTHY. Ingress then permanently returns 502, even though the service becomes ready shortly after.
This is reproducible on every deploy.
This requires:
- Increasing startupProbe delay OR
- Switching to readinessProbe OR
- Allowing post-start recovery
User-level port configuration does not solve this.
errrorr
I'm encountering a 502 Bad Gateway error on my Zeabur-deployed service.
Error Details
- Status: 502 Bad Gateway
- Request ID: sjc1::f72b0e52-0037-4891-abcf-eff9a7cc4dbb
- URL: https://openclaw-test.zeabur.app/
Possible Causes
- The service is not listening on the correct port - I can check the Network tab in the Zeabur dashboard to see which port is being listened on.
- The service crashed or stopped responding - I can check the Logs tab to investigate the error.
Please help me troubleshoot this 502 error based on the above information.