openclaw - 💡(How to fix) Fix Abort settle timeout causes zombie session write lock cascade

Description

When an embedded agent run times out (300s), the abort cleanup fails within the 2s settle window (embedded abort settle timed out: timeoutMs=2000), leaving the session .jsonl.lock file unreleased. This creates a zombie lock that blocks ALL subsequent messages from the same chat for 5+ minutes.

Reproduction

1. Maintain a large session context (150K+ tokens)
2. Use a reasoning model that frequently approaches the 300s timeout
3. When timeout occurs → embedded abort settle timed out → lock NOT released
4. Subsequent messages hit SessionWriteLockTimeoutError for 60s each

Root Cause

proper-lockfile considers the lock "compromised" after the abort, preventing release. The shouldReclaimContendedLockFile logic in session-write-lock-_a5O1H8L.js requires the lock to be stale (default 30min via DEFAULT_SESSION_WRITE_LOCK_STALE_MS) or the PID to be dead (5s via ORPHAN_LOCK_PAYLOAD_GRACE_MS). When the PID is the Gateway itself (alive), the lock can only be reclaimed after 30 minutes.

Observed Impact

• 5 consecutive message failures over 5 minutes on 2026-05-27
• Each failed message waited the full 60s DEFAULT_SESSION_WRITE_LOCK_ACQUIRE_TIMEOUT_MS
• User experience: "Something went wrong" errors cascade

Suggested Fix

1. Increase abort settle timeout from 2s to at least 10s
2. Force-release the lock file after abort (delete .lock file) when stopReason is timeout
3. Reduce DEFAULT_SESSION_WRITE_LOCK_STALE_MS from 30min to something more reasonable (e.g., 2min)

Workaround

Set OPENCLAW_SESSION_WRITE_LOCK_ACQUIRE_TIMEOUT_MS=10000 and OPENCLAW_SESSION_WRITE_LOCK_STALE_MS=120000 as env vars on the Gateway service.