Bug Description

Session gets stuck after compaction triggers, gateway becomes unresponsive. All channels (Control UI, Feishu, new sessions) stop responding. Only gateway restart resolves it.

Environment

• OpenClaw Version: 2026.2.9
• Platform: WSL2 (Ubuntu)
• Gateway Mode: Local (loopback)
• Model: MiniMax-M2.1
• Compaction Mode: safeguard

Reproduction Steps

1. Session runs normally for extended period (multiple hours/days)
2. Compaction triggers automatically (when context approaches limit)
3. Gateway becomes completely unresponsive:
   • Control UI: Shows connection error / "(no output)"
   • Feishu: Messages fail to send
   • New sessions: Cannot establish connection
4. Only workaround: Manual gateway restart (openclaw gateway restart)

Technical Details

Timeline from Gateway Logs

# Compaction started
{"subsystem":"agent/embedded","1":"embedded run compaction start: runId=33d6ae56-5065-45c1-8eae-13b8d669bf8e"}
[Timestamp: 2026-02-10T10:56:53.317Z]

# Compaction retry triggered
{"subsystem":"agent/embedded","1":"embedded run compaction retry: runId=33d6ae56-5065-45c1-8eae-13b8d669bf8e"}
[Timestamp: 2026-02-10T10:57:30.652Z]

# TIMEOUT after 600 seconds (10 minutes)
{"subsystem":"agent/embedded","1":"embedded run timeout: runId=33d6ae56-5065-45c1-8eae-13b8d669bf8e timeoutMs=600000"}
[Timestamp: 2026-02-10T11:04:39.975Z]

Additional Symptoms

After timeout, stale cron job running markers were found:

{"module":"cron","1":{"jobId":"9c06c09b-e9b4-40df-8626-22c43ec0cd37","runningAtMs":1770726600004},"2":"cron: clearing stale running marker on startup"}
{"module":"cron","1":{"jobId":"7d2b3ecd-5e78-4fc3-aeb2-f4d559a033f0","runningAtMs":1770726600004},"2":"cron: clearing stale running marker on startup"}

These markers are cleared on subsequent gateway restart, indicating previous shutdown was unclean.

Impact

• Severity: High - Complete service disruption
• User Experience: Gateway completely frozen, requires manual intervention
• Recovery: Manual gateway restart is the only known workaround
• Frequency: Reproduced multiple times in the same session

Suggested Investigation Areas

1. Compaction timeout handling: The 600-second timeout appears to hang rather than gracefully fail
2. State cleanup: Stale running markers not being cleared during/after timeout
3. Message queue: Incoming messages not being processed during compaction
4. Channel reconnection: WebSocket connections not recovering after compaction failure

Workaround

Manual gateway restart:

openclaw gateway restart

Logs

Full gateway logs available at: /tmp/openclaw/openclaw-2026-02-10.log

Note: Logs are overwritten on gateway restart, so capture immediately after reproduction.

Additional Context

This appears to be related to issue #11140 (HEARTBEAT_OK accumulates) as compaction is involved in session context management.