openclaw - 💡(How to fix) Fix Bonjour plugin crash-loops gateway on hosts without working multicast (2026.4.24) [1 comments, 2 participants]

Environment

• OpenClaw: 2026.4.24 (cbcfdf6), upgraded from 2026.4.5
• Host: Ubuntu 24.04.4 LTS on a Hostinger VPS (no real LAN multicast)
• Node: v22.22.1
• Install method: npm global

Symptom

Gateway enters a restart loop with restart counter climbing every ~45 seconds. Stability bundle written for each crash; same unhandled_rejection reason in every bundle (17 bundles total before disabling the plugin).

Repro

1. Install OpenClaw 2026.4.24 on a host where mDNS multicast doesn't propagate (any standard VPS).
2. Start the gateway with the bonjour plugin enabled (default).
3. Within ~30 seconds, observe the watchdog logs and crash.

Log fingerprint

[plugins] bonjour: watchdog detected non-announced service; attempting re-advertise
 (gateway fqdn=<host> (OpenClaw)._openclaw-gw._tcp.local. host=openclaw.local.
 port=18789 state=probing)
[plugins] bonjour: restarting advertiser (service stuck in probing for 28593ms ...)
[plugins] bonjour: advertised gateway ... state=unannounced
[openclaw] Unhandled promise rejection: CIAO PROBING CANCELLED
[openclaw] wrote stability bundle: /root/.openclaw/logs/stability/openclaw-stability-...-unhandled_rejection.json
systemd: openclaw-gateway.service: Main process exited, code=exited, status=1/FAILURE

The same pattern occurs with CIAO ANNOUNCEMENT CANCELLED when the service makes it to announcing state before the watchdog cancels.

Root cause hypothesis

The watchdog's restart-advertiser action cancels an in-flight CIAO promise. That cancellation rejection isn't caught by the bonjour plugin, so it propagates as an unhandled rejection at the Node level, which exits the process.

Workaround

Disable the plugin:

openclaw config set plugins.entries.bonjour.enabled false
systemctl --user restart openclaw-gateway

This works for VPS hosts that don't need mDNS but is not a fix for users who do want local network discovery.

Stability bundle (redacted)

{
  "version": 1,
  "generatedAt": "2026-04-26T20:40:16.906Z",
  "reason": "unhandled_rejection",
  "process": {
    "pid": 4181711,
    "platform": "linux",
    "arch": "x64",
    "node": "22.22.1",
    "uptimeMs": 99736
  },
  "host": {
    "hostname": "<redacted-hostname>"
  },
  "snapshot": {
    "generatedAt": "2026-04-26T20:40:16.906Z",
    "capacity": 1000,
    "count": 0,
    "dropped": 0,
    "events": [],
    "summary": {
      "byType": {}
    }
  }
}

Suggested fix

Either catch the rejection inside the watchdog's restart path, or skip the watchdog entirely on hosts where multicast is detected as non-functional after N probing attempts.