openclaw - 💡(How to fix) Fix pi-trajectory-flush timeout aborts entire agent run — should degrade gracefully [1 pull requests]

Problem

When the pi-trajectory-flush step exceeds its 10s hard timeout during agent run cleanup, the entire run is aborted. This happened today (2026-05-31 ~12:21–12:55 CST) across multiple sessions simultaneously, suggesting a transient I/O slowdown on macOS triggered cascading session aborts.

Observed behavior

• agent cleanup timed out: step=pi-trajectory-flush timeoutMs=10000 with very small pending data (633–2534 bytes)
• embedded abort settle timed out: timeoutMs=2000 follows
• Session enters stuck state → automatic stuck session recovery aborts it
• Multiple sessions affected in parallel → eventual Gateway stability bundle + shutdown timeout

Expected behavior

Trajectory flush is telemetry/audit data. A slow flush should not abort the primary agent run or cause session-level stuck recovery.

Suggested improvements

1. Graceful degradation — if trajectory flush times out, drop the pending trajectory batch but let the run complete normally
2. Configurable or adaptive timeout — 10s is tight when macOS has background I/O pressure (Time Machine, Spotlight, etc.)
3. Retry before abort — currently one timeout → immediate failure; a single retry with backoff would prevent cascading aborts during transient I/O spikes

Environment

• OpenClaw 2026.5.27
• macOS 12.7.6 (Intel)
• Node v24.13.0
• Disk: APFS SSD (standard MacBook Pro)

Logs (excerpt)

agent cleanup timed out: runId=dc4ff2b4 sessionId=699ab5b5 step=pi-trajectory-flush timeoutMs=10000 details=pendingWrites=1 queuedBytes=704 activeOperation=file-append
embedded abort settle timed out: runId=dc4ff2b4 sessionId=699ab5b5 timeoutMs=2000
stuck session recovery outcome: status=aborted action=abort_embedded_run sessionId=699ab5b5