openclaw - 💡(How to fix) Fix [Bug]: Gateway event-loop starvation blocks Feishu/Lark WebSocket when gh/curl/child-process hangs

Describe the bug

A hanging child process (e.g., gh auth refresh waiting for browser authorization, or any long-running exec command) can block the entire Gateway event loop, causing the Feishu/Lark WebSocket connection to become unresponsive. The agent becomes unreachable from Feishu — messages are not received, reactions are not sent, and the only fix is a manual gateway restart.

To Reproduce

1. Run a command that hangs waiting for user input, e.g.:

   gh auth refresh -h github.com -s delete_repo

   This spawns a child process that sits waiting for browser authorization indefinitely.

2. While the process is hanging, try sending a message to the agent via Feishu.

3. The message is never received by the agent. The gateway logs show no incoming Feishu events during this time.

4. After a gateway restart (openclaw gateway restart), Feishu connectivity is restored immediately.

Expected behavior

The Gateway event loop should not be blocked by a hanging child process. Feishu WebSocket messages should still be processable even when other exec tasks are stalled.

Actual behavior

The entire Gateway event loop appears to starve when a child process hangs. Feishu WebSocket becomes completely unresponsive — no messages in, no reactions out.

Environment

• OpenClaw version: 2026.4.1
• OS: Ubuntu 24.04 on WSL2
• Channel: Feishu (WebSocket mode, larksuite/openclaw-lark plugin)
• Browser: Google Chrome 146.0.7680.153

Workaround

Manual gateway restart via:

openclaw gateway restart

Or WebUI → restart gateway

Related issues

• #84903 A single stalled agent session blocks the entire Gateway event loop