openclaw - 💡(How to fix) Fix Gateway runtime unstable after updating to 2026.4.24 until doctor fix + restart + bonjour disable [6 comments, 7 participants]

Summary

After updating OpenClaw to 2026.4.24 via openclaw update, the CLI version updated successfully, but the running gateway became unstable for a while. The system showed mixed signals such as Runtime: running while connectivity probe failed with ECONNREFUSED / timeout, and doctor --fix surfaced a bundled Telegram channel/plugin runtime error.

The installation was eventually recovered by rebuilding runtime state (doctor --fix), restarting the gateway service, and disabling the bonjour plugin before another restart.

Environment

• Host: Linux VPS
• Install type: pnpm
• OpenClaw target version: 2026.4.24
• Gateway service: systemd user service
• Channel in use: Telegram

What happened

1. openclaw update reported success and openclaw --version became 2026.4.24.
2. Initially, openclaw status / gateway health did not fully reflect a clean running state.
3. openclaw doctor --fix showed intermittent gateway state contradictions:
   • Gateway not running
   • port 18789 already in use
   • Runtime: running
4. openclaw gateway status later showed:
   • service active
   • but connectivity probe failed (ECONNREFUSED or timeout)
5. doctor --fix also logged a bundled Telegram/plugin runtime error:

failed to load bundled channel telegram: ENOTEMPTY: directory not empty, rmdir '/root/.openclaw/plugin-runtime-deps/openclaw-2026.4.24-.../dist/extensions/node_modules/openclaw/plugin-sdk'

Recovery steps that worked

These steps eventually restored the system to a healthy 2026.4.24 runtime:

1. Run:

   openclaw doctor --fix

2. Manually restart the gateway service:

   openclaw gateway stop
   openclaw gateway start
   # later also:
   openclaw gateway restart

3. Disable bonjour plugin:

   openclaw plugins disable bonjour
   openclaw gateway restart

4. Re-check:

   openclaw status
   openclaw gateway status

Final good state

Eventually the runtime converged and both of these showed the new version correctly:

• Gateway ... app 2026.4.24
• Gateway self ... app 2026.4.24
• Telegram channel status returned to OK

Suspected cause

This looked less like a failed package update and more like a post-update runtime inconsistency involving one or more of:

• stale / conflicting plugin-runtime-deps state
• bundled plugin/channel runtime cleanup race (ENOTEMPTY during rmdir)
• gateway process restart / probe timing mismatch
• a non-essential plugin (bonjour) contributing to unstable startup after upgrade

Why this may matter

This can leave users in a confusing in-between state where:

• openclaw --version is already upgraded
• but gateway health is unstable or misleading
• Telegram / channel behavior may be inconsistent until runtime is rebuilt

Suggestion

Potential improvements that might help:

1. Make openclaw update more explicit about whether the gateway runtime has successfully switched to the new version.
2. Add better repair/cleanup for plugin-runtime-deps when bundled plugin directories cannot be removed cleanly.
3. Improve gateway status reporting when runtime is active but probe is temporarily failing during/after restart.
4. Consider warning about or isolating optional plugins (like bonjour) if they can affect recovery after update.