Summary

We occasionally see the agent fail to respond with an error like:

session file locked (timeout 10000ms): pid=<gateway-pid> .../sessions/<session>.jsonl.lock
When this happens, all models fail because the session store can’t acquire the lock.

Environment

• OpenClaw: 2026.2.26 (cli banner shows commit 5987afc)
• Deployment: Docker container (1Panel)
• Session store path: ~/.openclaw/agents/main/sessions/

Observed behavior

• A *.jsonl.lock file exists (JSON like {"pid": 7, "createdAt": "..."}).
• The referenced PID is still alive (openclaw-gateway).
• Attempts to write to the session time out after 10000ms.
• Workaround: restarting gateway/container restores service; in some cases moving/removing the *.jsonl.lock file immediately restores, but that’s risky while the process is alive.

Expected behavior

• Session writes should not deadlock or time out due to a lock file while the gateway process is healthy.
• If a lock can’t be acquired quickly, the system should recover (e.g., detect stale lock, or provide safer lock recovery).

Notes / Hypothesis

This looks like stale-lock or lock-release failure. It may be related to PID reuse after restart, or a lock file persisting longer than intended.

Timestamp

• Report created: 2026-03-02 08:51 UTC