openclaw - 💡(How to fix) Fix 2026.4.24 gateway crash loop: Unhandled promise rejection 'CIAO PROBING CANCELLED' from bonjour [1 comments, 2 participants]

Summary

After upgrading from 2026.4.23 to 2026.4.24 on macOS, the gateway enters a crash-loop on startup. Each gateway process exits within ~25 seconds with an unhandled promise rejection from the bonjour mDNS plugin during its announcing/probing phase. openclaw doctor cannot complete because the gateway is never up long enough to accept connections.

Environment

• Host: macOS 26.3.1 (arm64)
• Node: 25.6.1 (gateway runs under embedded node 22.22.0)
• Install: pnpm global + a parallel npm-under-nvm install used by the LaunchAgent

Reproduction

1. pnpm add -g [email protected] and npm install -g [email protected] (under nvm so ai.openclaw.node.plist resolves the new install).
2. Update ~/Library/LaunchAgents/ai.openclaw.gateway.plist and ai.openclaw.dj-runtime.plist to point at …/.pnpm/[email protected]/node_modules/openclaw/dist/index.js (note: 4.24 dropped the _@[email protected] suffix that 4.22 and 4.23 carried).
3. launchctl bootout gui/$UID/ai.openclaw.gateway && launchctl bootstrap gui/$UID ~/Library/LaunchAgents/ai.openclaw.gateway.plist.

Observed: gateway starts, enters bonjour announcing/probing, then crashes within 20–30 s. launchctl list | grep openclaw shows the gateway constantly recycling PIDs.

Logs

`~/.openclaw/logs/gateway.err.log`:

``` [bonjour] gateway name conflict resolved; newName="Joules's Mac mini (OpenClaw) (2)" [bonjour] restarting advertiser (service stuck in probing for 23162ms (gateway fqdn=Joules's Mac mini (OpenClaw)._openclaw-gw._tcp.local. host=openclaw.local. port=18789 state=probing)) [openclaw] Unhandled promise rejection: CIAO PROBING CANCELLED [openclaw] wrote stability bundle: /Users/joules/.openclaw/logs/stability/openclaw-stability-2026-04-25T20-36-18-345Z-93259-unhandled_rejection.json ```

Two stability bundles are available locally for analysis (paths: `/.openclaw/logs/stability/openclaw-stability-2026-04-25T20-35-39-595Z-92941-unhandled_rejection.json` and `/.openclaw/logs/stability/openclaw-stability-2026-04-25T20-36-18-345Z-93259-unhandled_rejection.json`). Happy to upload if useful.

Hypothesis

The bonjour advertiser's probing/announcing watchdog cancels the in-flight CIAO operation when it appears stuck. The cancellation rejects an internal promise. Under Node's strict unhandled-rejection mode this terminates the process.

A second symptom suggests a contributing cause: the bonjour layer keeps detecting a name/host conflict and renaming the service to `(2)`, suggesting the previous gateway PID's mDNS advertisement hasn't been retired before the new gateway starts probing.

Suggested fixes

• Catch the bonjour CIAO cancellation rejection so it doesn't bubble unhandled.
• Retire the prior mDNS advertisement before starting a new probing cycle (the rename-to-`(2)` suggests the old advertisement is still live when the new gateway starts).

Workaround

On the affected host, rolled back to `2026.4.23`: reverted both LaunchAgent plists to the `…/[email protected]_@[email protected]/…` path, downgraded pnpm/nvm globals, ran one bootout/bootstrap cycle. Stable since.

Not yet tested whether disabling the bonjour plugin in `openclaw.json` would also unblock 2026.4.24 startup.