openclaw - 💡(How to fix) Fix Gateway freezes — 'chunks_vec not updated' memory vector crash kills all channel communication

Description

Gateway goes completely unresponsive — both Feishu (Lark) and webchat stop receiving/sending messages. Only a full gateway restart (SIGUSR1 or process kill) fixes it.

Error symptoms

1. Tools start failing with non-descriptive parameter validation errors (e.g. feishu_bitable_app_table_record: Validation failed for tool — though the actual tool params were correct)
2. Memory system logs: memory chunks_vec not updated — semantic vector embeddings unavailable — no vector dimensions resolved. Vector recall degraded. Further duplicate warnings suppressed.
3. The gateway continues running (process alive) but stops processing any incoming messages
4. No error in logs, just silent deadlock

Encountered

• OpenClaw version: 2026.5.20 (e510042)
• OS: Linux 6.6.114.1-microsoft-standard-WSL2 (x64)
• Runtime: node v22.22.1
• Provider: deepseek-chat (api-key)

Config notes

• agents.defaults.memorySearch.enabled was toggled to false then back to true
• Feishu channel active (WebSocket mode)

File observations during freeze

• ~/.openclaw/memory/main.sqlite-wal → 0 bytes
• ~/.openclaw/memory/main.sqlite-shm → stale (still present, not cleaned up)
• Trajectory log shows repeated session.ended entries with no error propagation

Expected behavior

Memory vector failures should not freeze the entire gateway. A corrupted/empty memory index should degrade gracefully (e.g. fallback to no-search) rather than cause a deadlock that blocks message processing.

Workaround

Manual gateway restart.