openclaw - 💡(How to fix) Fix 2026.4.24: ENOTEMPTY race on plugin-sdk during parallel plugin/channel bootstrap [1 comments, 2 participants]

Summary

On [email protected] (commit cbcfdf6), the gateway fails to fully boot because multiple bundled plugin/channel loaders race on the same shared plugin-sdk directory under ~/.openclaw/plugin-runtime-deps/openclaw-2026.4.24-*/dist/extensions/node_modules/openclaw/plugin-sdk. One loader wins; the rest crash with ENOTEMPTY: directory not empty, rmdir '.../plugin-sdk'.

Net effect: Discord plugin fails to register, several bundled channels (matrix, twitch, zalo, qqbot) fail to load, and the WS endpoint reports unreachable (timeout) even though the systemd unit is active (running).

Environment

• OpenClaw: 2026.4.24 (cbcfdf6)
• Last known-good: 2026.4.14 (323493f) — unaffected, rollback restores normal operation
• OS: Linux 6.12.62+rpt-rpi-2712 (arm64), Raspberry Pi 5
• Node: 22.22.2
• Install: npm i -g openclaw into ~/.npm-global
• Service: systemctl --user openclaw-gateway (with 20-performance.conf drop-in)

Reproduction

1. From a working 2026.4.14 install: npm i -g [email protected]
2. systemctl --user daemon-reload && systemctl --user restart openclaw-gateway
3. openclaw status and inspect logs.

Observed errors

[plugins] discord failed during register from /home/admin/.npm-global/lib/node_modules/openclaw/dist/extensions/discord/index.js:
  Error: ENOTEMPTY: directory not empty, rmdir
  '/home/admin/.openclaw/plugin-runtime-deps/openclaw-unknown-94d7fff93c31/dist/extensions/node_modules/openclaw/plugin-sdk'
[plugins] 1 plugin(s) failed to initialize (register: discord). Run 'openclaw plugins list' for details.
[channels] failed to load bundled channel matrix: ENOTEMPTY: directory not empty, rmdir
  '/home/admin/.openclaw/plugin-runtime-deps/openclaw-2026.4.24-4493ea0b4ae8/dist/extensions/node_modules/openclaw/plugin-sdk'
[channels] failed to load bundled channel twitch: ENOTEMPTY ... plugin-sdk
[channels] failed to load bundled channel zalo:   ENOTEMPTY ... plugin-sdk
[channels] failed to load bundled channel qqbot:  ENOTEMPTY ... plugin-sdk

openclaw status reports:

• Gateway: local · ws://127.0.0.1:18789 (local loopback) · unreachable (timeout)
• Gateway service: systemd installed · enabled · running (pid …, state active)

Why it looks like a race

• Two distinct cache prefixes are involved in the same boot: openclaw-2026.4.24-4493ea0b4ae8 (versioned) and openclaw-unknown-94d7fff93c31 (the unknown- one is suspicious in itself — looks like a loader that didn't resolve the running version).
• All five failing loaders point at the same node_modules/openclaw/plugin-sdk path inside their respective cache root. The successful loader leaves files behind; the others attempt rmdir on a populated directory and abort.
• Wiping ~/.openclaw/plugin-runtime-deps/openclaw-2026.4.24-* and openclaw-unknown-* does not help — the race reproduces on the very next start with freshly-created directories. So this is not stale-cache cleanup; it's a concurrent extract-then-rmdir on a shared path.

Workaround

Roll back to 2026.4.14:

npm i -g [email protected]
systemctl --user daemon-reload
systemctl --user restart openclaw-gateway

Suggested fix directions

• Per-plugin cache root (e.g. …/plugin-runtime-deps/<version>/<plugin-name>/…) instead of a shared plugin-sdk directory across all bundled plugins/channels.
• Or: serialize the SDK-link step behind a mutex / lockfile so only one loader performs the rmdir+link, and the others reuse the materialized result.
• Separately worth investigating why one of the cache roots is named openclaw-unknown-* rather than version-pinned — this suggests a code path that bootstraps without a resolved version.

Logs / artifacts

Happy to attach a full journalctl --user -u openclaw-gateway excerpt or openclaw status --deep output if useful — say the word.