openclaw - 💡(How to fix) Fix memory-core: memory_search blocks event loop for 60+ seconds — Discord gateway closes, agents hang

Bug Type

Crash (process/app hangs)

Summary

The memory-core plugin's memory_search tool blocks the Node.js event loop for 60+ seconds when processing Discord DMs, causing:

• Event loop delay spikes to 62,000ms+
• Discord gateway closes connection
• Agent becomes unresponsive to all messages
• No self-recovery — requires restart

Reproducible: 100% on Discord DMs

Environment

• OpenClaw: 2026.5.7
• memory-core: 2026.5.7 (bundled)
• OS: Ubuntu 24.04 (Linux 6.8.0-106-generic)
• Node.js: v22.x
• Plugin: @openclaw/memory-core (bundled)

Evidence from Production Logs

``` 2026-05-12T20:19:37.124+00:00 [diagnostic] liveness warning: reasons=event_loop_delay interval=80s eventLoopDelayP99Ms=100.2 eventLoopDelayMaxMs=62746.8 eventLoopUtilization=0.818

work=[active=agent:main:discord:default:direct:1493259950661697657( processing/tool_call, q=1, age=63s last=tool:memory_search:started )] ```

Key indicators:

• Tool: memory_search:started
• Age: 63 seconds and counting
• Event loop delay: 62,746.8ms
• Result: Gateway websocket closed, agent frozen

Steps to Reproduce

1. Enable memory-core plugin (default enabled)
2. Send Discord DM to agent
3. Agent begins processing, calls memory_search
4. Event loop blocks for 60+ seconds
5. Discord gateway timeout → connection closes
6. Agent hangs indefinitely

Expected Behavior

`memory_search` should:

• Complete or timeout within reasonable time (<10s)
• Not block the event loop
• Allow Discord gateway to remain responsive
• Handle embedding failures gracefully

Actual Behavior

Workaround Deployed

Agent Watchdog — Monitors every 2 minutes and auto-restarts hung agents

Related Issues

Same root cause pattern:

• #65517 - active-memory blocks Telegram polling
• #56733 - Gateway frozen (embedded run)
• #52231 - Zombie handle (embedded run timeout)
• #9751 - Gateway unresponsive (OpenAI embeddings)

Common pattern: Blocking embedded operations on single-process Node.js event loop starves I/O.

Impact

• 17 agents affected fleet-wide
• Discord DMs (primary user interaction) broken
• Agents appear "frozen", no response
• Recovery requires watchdog (2 min delay)

Suggested Fix

1. Make `memory_search` non-blocking (worker threads / child process)
2. Add configurable timeout (`memory_search.timeoutMs` with 5-10s default)
3. Fail gracefully (return empty results rather than block)
4. Async embedding (don't block event loop during vector queries)