openclaw - 💡(How to fix) Fix file_lock_stale race condition under rapid subagent spawns

Description

When spawning multiple subagents rapidly (N ≥ 10) from a single parent session, some children fail immediately with:

Error: file lock stale for /path/to/agents/<agent>/sessions/<id>.jsonl: code=file_lock_stale

The child never writes its first message — the lock is already stale before the session starts writing.

Reproduction

1. Spawn 10+ Max subagents in rapid succession (no delay between sessions_spawn calls)
2. Observe that ~10-20% fail with file_lock_stale
3. Children that survive the lock race complete normally

Environment

• OpenClaw: 2026.5.12
• OS: Linux 6.17.0-29-generic (x64)
• Node: v25.9.0
• Config: agents.defaults.subagents.maxConcurrent=20, maxChildrenPerAgent=20, maxSpawnDepth=2

Evidence

• 9 unique Max sessions hit file_lock_stale over ~48 hours of heavy swarm testing
• One dispatcher session (multi-spawn controller) hit it 9 times in a single run
• Retest with 10 Max branches: 2/10 lost to stale-lock
• c50 test (50 simultaneous Max children) had 0 stale-lock failures, suggesting timing sensitivity

Workaround

Staggering spawns with >=2s delay between branches eliminates the race in practice. But this should not be necessary for concurrent spawn patterns.

Suspected Cause

The session JSONL file lock has a TTL or contention window that is too short when multiple sessions are created in rapid succession under the same agent. The lock acquisition may race with lock creation/rotation from the parent process.