Version

OpenClaw 2026.6.1

Summary

When chat.abort is triggered during a long-running model call, the embedded run's abort settle can time out (embedded abort settle timed out, 2000ms). The gateway acquires a session write lock during cleanup, but the lock is never released on the timeout path, blocking all subsequent messages to that session.

Trigger

1. Agent is executing a long-running model call
2. User triggers chat.abort
3. embedded abort settle timed out after timeoutMs=2000

Observed behavior

• Gateway process (not agent worker) holds session JSONL write lock
• Lock maxHoldMs is 1,020,000ms (17 minutes)
• All subsequent agent replies time out at 60s waiting for the lock
• Error: SessionWriteLockTimeoutError: session file locked ... pid=<gateway-pid> alive=true
• Lock is only released when gateway process exits (restart) or after 17 min expiry

Log timeline (UTC)

04:35:52 chat.abort ✓ 87ms
04:35:55 embedded abort settle timed out: runId=ba228388… sessionId=e3307808… timeoutMs=2000
04:35:55 lock file created: pid=<gateway> maxHoldMs=1020000
04:37:14 SessionWriteLockTimeoutError ageMs=79210  (×2 lanes)
04:38:25 SessionWriteLockTimeoutError ageMs=149601 (×2 lanes)

Impact

Complete session blockage — all user messages during the 17-minute lock window return errors instead of actual replies.

Suspected root cause

The abort cleanup path acquires a session write lock but the finally/cleanup only releases it on the success path. When abort settle timed out, the lock is not released.

Workaround

Restart gateway (openclaw gateway restart), but this drops all active connections.