openclaw - 💡(How to fix) Fix OpenClaw 2026.4.24 update can leave macOS gateway in mixed-version/unhealthy plugin-runtime-deps state [4 comments, 4 participants]

Summary

After attempting to update OpenClaw to 2026.4.24 on macOS, the gateway can enter an unhealthy/mixed-version startup state.

Observed failure modes:

• openclaw update appears to partially apply 2026.4.24, but the LaunchAgent/gateway may still run 2026.4.23.
• Config/plugins can require >=2026.4.24 while the running host reports 2026.4.23.
• Gateway can start with ready (0 plugins, ...) after bundled runtime dependency staging fails.
• Bundled plugin runtime deps are repeatedly detected as missing and reinstalled across restarts.
• Under low disk space, runtime dependency staging fails with ENOSPC, but the resulting state is not cleanly rolled back or clearly blocked by a preflight.

This made Telegram/WhatsApp unusable until cleanup/restart/downgrade steps were applied.

Environment

• macOS 26.3.1 arm64
• Node 24.13.0
• OpenClaw attempted version: 2026.4.24
• OpenClaw running/remaining version after recovery: 2026.4.23
• Previous working versions: 2026.4.23 / 2026.4.22
• Install method: npm global, user prefix ~/.npm-global
• Gateway service: macOS LaunchAgent ai.openclaw.gateway
• Gateway port: 18789
• Disk situation during failure: Data volume was at ~99% capacity / low free space

Symptoms / evidence

openclaw gateway status after the incident showed the service reachable, but still running 2026.4.23 while 2026.4.24 was available:

OpenClaw 2026.4.23 (a979721)
Service: LaunchAgent (loaded)
Command: /usr/local/bin/node /Users/appel/.npm-global/lib/node_modules/openclaw/dist/index.js gateway --port 18789
Runtime: running (..., state active)
Connectivity probe: ok
Listening: 127.0.0.1:18789

During the failed/mixed state, logs included version/plugin incompatibility:

Invalid config at /Users/appel/.openclaw/openclaw.json:
- plugins.entries.feishu: plugin feishu: plugin requires OpenClaw >=2026.4.24, but this host is 2026.4.23; skipping load
- plugins.entries.whatsapp: plugin whatsapp: plugin requires OpenClaw >=2026.4.24, but this host is 2026.4.23; skipping load

The gateway then started with no plugins after dependency staging failures:

[gateway] ready (0 plugins, 46.7s)

Low disk/runtime dependency staging errors:

[plugins] failed to install bundled runtime deps: Error: ENOSPC: no space left on device, mkdir '/Users/appel/.openclaw/plugin-runtime-deps/openclaw-2026.4.24-.../.openclaw-runtime-deps.lock'
[plugins] failed to install bundled runtime deps: Error: ENOSPC: no space left on device, open '/Users/appel/.openclaw/plugin-runtime-deps/openclaw-2026.4.24-.../dist/extensions/node_modules/openclaw/package.json'
[gateway] startup model warmup failed ... ENOSPC: no space left on device
[gateway] failed to promote config last-known-good backup: Error: ENOSPC: no space left on device
[gateway] parse/handle error: Error: ENOSPC: no space left on device, open '/Users/appel/.openclaw/devices/paired.json....tmp'

Repeated bundled runtime dependency staging/install was also observed across multiple startups:

[plugins] acpx staging bundled runtime deps (...): [email protected]
[plugins] acpx installed bundled runtime deps ...
[plugins] bonjour staging bundled runtime deps (...): @homebridge/ciao@^1.3.6
[plugins] bonjour installed bundled runtime deps ...
[plugins] browser staging bundled runtime deps (...)
[plugins] browser installed bundled runtime deps ...
[plugins] telegram staging bundled runtime deps (...)
[plugins] telegram installed bundled runtime deps ...
[plugins] whatsapp staging bundled runtime deps (...)
[plugins] whatsapp installed bundled runtime deps ...

Also observed earlier in the same incident:

Cannot find package '@homebridge/ciao'
Cannot find module 'acpx/runtime'
.openclaw/plugin-runtime-deps/openclaw-unknown-*

Expected behavior

• openclaw update should preflight available disk space before starting a version/plugin/runtime-deps migration.
• If disk space is insufficient, the update should fail before changing config/plugin state.
• The gateway should not start as "ready" with 0 plugins after plugin dependency staging fails, or at least should report a clearly unhealthy state.
• The LaunchAgent, installed npm package, plugin config, and runtime deps should not be left in a mixed 2026.4.23/2026.4.24 state.
• Bundled runtime deps should not be repeatedly reinstalled if they were already staged successfully.
• Runtime dependency staging should not unexpectedly rely on or recreate ~/node_modules outside the intended plugin runtime deps directory.

Actual behavior

The update/startup flow left the system in a mixed and confusing state:

• 2026.4.24 was attempted/partially staged.
• Gateway still ran as 2026.4.23.
• Some plugins/config expected 2026.4.24.
• Runtime deps repeatedly staged/reinstalled.
• With low disk space, many plugins failed and gateway reported ready (0 plugins, ...).

Recovery / workaround

Recovery required freeing space / clearing stale runtime deps and restarting/reinstalling the gateway. Downgrading/restoring 2026.4.23 also restored healthy startup faster.

Commands involved during recovery included combinations of:

openclaw gateway stop
openclaw gateway install --force
openclaw gateway start
openclaw gateway restart

Request

Please consider adding:

1. Disk-space preflight before update/plugin runtime deps staging.
2. Atomic update/rollback for CLI + service + plugin config compatibility.
3. A hard unhealthy state when gateway starts with 0 plugins due to runtime deps failures.
4. Better diagnostics when plugin runtime deps are repeatedly detected as missing or when openclaw-unknown-* staging dirs are created.