openclaw - 💡(How to fix) Fix [Bug]: v2026.4.24 multiple instabilities — postinstall pruning, bonjour crash loop, WhatsApp startup probe [2 comments, 3 participants]

Environment: Ubuntu via nvm, Node 22.22.0, npm 10.9.4, OpenClaw 2026.4.24 (cbcfdf6), gateway running as systemd service.

Bug type

Crash (process exits or hangs) / Regression

Summary

v2026.4.24 is severely unstable on this install. Three distinct root causes combine to make the gateway unreliable: a postinstall pruning bug, a bonjour/CIAO unhandled rejection crash loop, and unexpected WhatsApp bundled runtime-deps probing on startup.

Observed failures (chronological from logs)

1. bundle-mcp failures (WARN, non-fatal)

failed to start server \"notebooklm\" (http://127.0.0.1:8888/mcp): TypeError: fetch failed
failed to start server \"openspace\" (http://127.0.0.1:8081/mcp): TypeError: fetch failed

These MCP servers are not running locally — the errors are noise but indicate bundle-MCP retries on every gateway start.

2. WhatsApp pending delivery failure (logged ~22h ago)

[tools] message failed: Error: No active WhatsApp Web listener (account: default)
[cron:financial-daily-summary] transient direct announce delivery failure, retrying 2/4 in 5s: No active WhatsApp Web listener

Financial daily summary cron tried to send to WhatsApp but the channel was disconnected. Messages were queued but not delivered.

3. bonjour/CIAO crash loop (known issue, refs below)

[bonjour] watchdog detected non-announced service; attempting re-advertise
[openclaw] Unhandled promise rejection: CIAO ANNOUNCEMENT CANCELLED

This is the crash loop described in issues #71771, #72334. Despite PR #71668, CIAO rejections still crash the gateway ~30s after startup. Workaround applied: plugins.entries.bonjour.enabled: false added to config.

4. Discord channel timeout

Discord default: ... error:channel stop timed out after 5000ms

Discord bot failed to stop cleanly within 5s.

Root cause analysis

Postinstall pruning bug

The npm tarball for 2026.4.24 contains 4116 .js files in dist/. The postinstall-bundled-plugins.mjs script uses rmSync/unlinkSync to prune files. When it encounters errors (e.g., ENOTEMPTY, unsafe dist entries), it silently deletes files that should be kept, leaving the installation broken. This is documented in issue #72042.

bonjour/CIAO crash loop

The bonjour advertiser enters a "probing" state, fails to announce, and triggers a CIAO cancellation that propagates as an unhandled promise rejection, crashing the entire gateway. PR #71668 was supposed to fix this but the rejection still propagates in v2026.4.24.

WhatsApp startup probe

Gateway probes bundled WhatsApp channel runtime-deps on startup even when channels.whatsapp is not configured (refs #71994). This causes delays and npm/git resolution attempts on every restart.

Known related issues (all pointing to v2026.4.24 being broken)

• #72042 — Gateway startup fails: ERR_MODULE_NOT_FOUND, 1617 dist files missing (postinstall pruning bug)
• #71771 — Gateway crash loop: CIAO ANNOUNCEMENT CANCELLED in bonjour
• #72334 — Bonjour/CIAO unhandled rejections still crash gateway v2026.4.24 despite PR #71668
• #71994 — Gateway startup probes bundled WhatsApp runtime deps when not configured
• #71986 — 2026.4.24 is a huge mess! whatsapp and telegram gone
• #72040 — Telegram channel crash on Windows 11 in 2026.4.24 (grammY deps missing)

Workaround applied

Added to plugins.entries in openclaw.json:

\"bonjour\": { \"enabled\": false }

Expected behavior

• Gateway should start reliably without postinstall file deletion
• bonjour CIAO rejections should be caught and not crash the gateway
• Bundled channel runtime-deps should only be staged for channels that are actually configured/enabled

Upgrade path

Rolling back to 2026.4.23 is reported to restore full functionality (#72042). Upgrading to 2026.4.25-beta (when available) is expected to resolve the bonjour issue.