Bug type

Crash (process/app exits or hangs)

Beta release blocker

No

Summary

OpenClaw enters a crash loop for ~6 minutes (≈25 restarts at 15s intervals) whenever the plugins.allow or plugins.entries.* configuration is changed at runtime. The container self-recovers thanks to restart: unless-stopped, but during the loop the gateway is unreachable.

Environment

• Image: ghcr.io/hostinger/hvps-openclaw:latest
• Image revision: 69fa0d8 (created 2026-03-02)
• Host OS: Ubuntu 24.04.4 LTS
• Docker compose: v5.0.2
• Reverse proxy: Traefik (HTTPS via Let's Encrypt)
• Active plugins: telegram only

Steps to reproduce

Steps to Reproduce

1. Have a running OpenClaw instance with multiple plugins listed in plugins.allow (e.g. discord, slack, googlechat, nostr, whatsapp).
2. Edit /data/.openclaw/openclaw.json and remove unused plugins from plugins.allow and disable them in plugins.entries.*.
3. OpenClaw detects the config change and triggers an automatic gateway reload via SIGUSR1.
4. The reload sequence fails with an unhandled WebSocket error.

Expected behavior

Expected Behaviour

• A WebSocket client connecting to the local gateway control port should wait for the gateway to be ready, or
• The error event should be handled gracefully (retry/backoff) rather than bubbling up as an unhandled exception that kills the process.

Related Symptom: Deep Security Audit Probe Timeout

openclaw security audit --deep reports:

WARN gateway.probe_failed Gateway probe failed (deep) timeout

This appears to share the same root cause: the loopback port 18789 intermittently refuses connections or fails to respond to the local probe.

Actual behavior

Observed Behaviour

After the reload signal, the process exits with an unhandled error event:

[gateway] received SIGUSR1; restarting node:events:497 throw er; // Unhandled 'error' event ^ Error: connect ECONNREFUSED 127.0.0.1:18789 at TCPConnectWrap.afterConnect [as oncomplete] (node:net:1637:16) Emitted 'error' event on WebSocket instance at: at emitErrorAndClose (/hostinger/node_modules/ws/lib/websocket.js:1046:13) at ClientRequest. (/hostinger/node_modules/ws/lib/websocket.js:886:5) errno: -111, code: 'ECONNREFUSED', syscall: 'connect', address: '127.0.0.1', port: 18789

Docker then restarts the container per its restart policy. The new instance hits the same race condition during startup and crashes again. After roughly 25 restart cycles (~6 minutes) the timing happens to align and the gateway comes up cleanly.

docker events confirms ~25 die/start cycles during the loop window.

OpenClaw version

2026.4.9 (build 0512059)

Operating system

Ubuntu 24.04.4 LTS (kernel 6.8.0-107-generic, x86_64)

Install method

docker (image: ghcr.io/hostinger/hvps-openclaw:latest, revision 69fa0d8, image created 2026-03-02)

Model

openrouter/moonshotai/kimi-k2.5

Provider / routing chain

openclaw -> openrouter -> moonshotai/kimi-k2.5

Additional provider/model setup details

• Default agent model: openrouter/moonshotai/kimi-k2.5 (configured in agents.defaults.model.primary)
• 11 agents registered (main, Router, Chef, Codee, Mail, Memory, Ops, PM, Search, Shares, Travel)
• Mix of openrouter/moonshotai/kimi-k2.5, openrouter/anthropic/claude-sonnet-4-5, and openrouter/anthropic/claude-opus-4-6 per agent
• Single channel: telegram (allowlist groups, pairing for DMs)
• Reverse proxy: Traefik (HTTPS via Let's Encrypt) in front of port 60413
• The bug is unrelated to model/provider — it occurs purely on the gateway control loop during a config reload, before any model call.

Logs, screenshots, and evidence

## Trigger event in logs (Berlin time, UTC+02:00)

  Config change detected at 10:12:57 (`plugins.allow` modified — unused channel
  plugins removed). Reload deferred until in-flight ops complete, then SIGUSR1
  sent at 10:17:17:

  2026-04-10T10:12:57.766+02:00 [reload] config change detected; evaluating reload (plugins.allow)
  2026-04-10T10:12:57.791+02:00 [reload] config change requires gateway restart (plugins.allow) — deferring until 2 operation(s), 1 reply(ies), 1 embedded run(s) complete
  2026-04-10T10:13:58.876+02:00 [reload] config change detected; evaluating reload (plugins.entries.whatsapp.enabled, plugins.entries.discord.enabled,
  plugins.entries.slack.enabled, plugins.entries.nostr.enabled, plugins.entries.googlechat.enabled)
  2026-04-10T10:17:17.891+02:00 [reload] all operations and replies completed; restarting gateway now
  2026-04-10T10:17:17.893+02:00 [gateway] signal SIGUSR1 received
  2026-04-10T10:17:17.895+02:00 [gateway] received SIGUSR1; restarting

  ## Crash on every restart attempt

  [gateway] Starting OpenClaw gateway...
  node:events:497
        throw er; // Unhandled 'error' event
        ^
  Error: connect ECONNREFUSED 127.0.0.1:18789
      at TCPConnectWrap.afterConnect [as oncomplete] (node:net:1637:16)
  Emitted 'error' event on WebSocket instance at:
      at emitErrorAndClose (/hostinger/node_modules/ws/lib/websocket.js:1046:13)
      at ClientRequest. (/hostinger/node_modules/ws/lib/websocket.js:886:5)
      at ClientRequest.emit (node:events:519:28)
      at emitErrorEvent (node:_http_client:107:11)
      at Socket.socketErrorListener (node:_http_client:574:5)
      at Socket.emit (node:events:519:28)
      at emitErrorNT (node:internal/streams/destroy:170:8)
      at emitErrorCloseNT (node:internal/streams/destroy:129:3)
      at process.processTicksAndRejections (node:internal/process/task_queues:90:21) {
    errno: -111,
    code: 'ECONNREFUSED',
    syscall: 'connect',
    address: '127.0.0.1',
    port: 18789
  }
  Node.js v22.22.0

  ## Container restart loop (`docker events`)

  25 die/start cycles between 10:17 and 10:23, ~15s apart:

  2026-04-10T08:17:20Z die
  2026-04-10T08:17:20Z start
  2026-04-10T08:17:36Z die
  2026-04-10T08:17:37Z start
  2026-04-10T08:17:48Z die
  2026-04-10T08:17:49Z start
  ... (21 more cycles)
  2026-04-10T08:23:12Z die
  2026-04-10T08:23:27Z start  ← finally stable

  ## Related symptom

  `openclaw security audit --deep` reports the same loopback port as unreachable:

  WARN  gateway.probe_failed  Gateway probe failed (deep)
    timeout
    Fix: Run "openclaw status --all" to debug connectivity/auth, then re-run "openclaw security audit --deep".

Impact and severity

Affected: any OpenClaw deployment that edits plugins.allow or plugins.entries.* while the gateway is running (observed on 2026.4.9 docker).

Severity: High during the loop window — gateway is unreachable, no Telegram messages can be sent or received. Self-recovers after ~6 minutes thanks to docker restart: unless-stopped.

Frequency: 1/1 reproducible — happened immediately after a single plugin config edit. Did not happen on cold container restarts before/after.

Consequence: ~6 minutes of downtime per config change, missed scheduled cron job runs that fall in that window, no observed data loss.

Additional information

Workaround

Avoid letting OpenClaw auto-reload after config changes. Instead, edit the config and immediately restart the container manually:

docker restart openclaw-jhps-openclaw-1

A clean cold-restart goes through without entering the crash loop.

Impact

- Severity: Medium — gateway is unreachable for several minutes after
any plugin config change.
- Self-recovery: Yes (eventually), thanks to restart: unless-stopped.
- Data loss: None observed.