openclaw - 💡(How to fix) Fix Gateway crash-loops with unhandled_rejection on v2026.4.24 (D state, 85%+ CPU) [4 comments, 5 participants]

Description

The gateway process enters an unrecoverable crash loop immediately after startup on version 2026.4.24 (cbcfdf6). The TUI shows not connected to gateway — message not sent and is unable to communicate.

Environment

• OpenClaw version: 2026.4.24 (cbcfdf6)
• OS: Linux x64
• Node: 22.22.2
• Install method: npm global (npm install -g openclaw)

Symptoms

• openclaw-gateway process enters D state (uninterruptible sleep) with 85–134% CPU usage immediately after spawning
• HTTP requests to http://127.0.0.1:18789 hang indefinitely
• Stability logs show repeated unhandled_rejection crashes every ~60 seconds:

{
  "reason": "unhandled_rejection",
  "process": { "pid": 2731, "node": "22.22.2", "uptimeMs": 59027 },
  "snapshot": { "count": 0, "events": [] }
}

• The crash log snapshot.events array is empty — no error details are captured
• Killing the gateway causes it to respawn (supervised), but the new process immediately enters the same D state
• kill -9 is required to terminate the stuck process

Reproduction

1. Install [email protected] via npm
2. Run openclaw tui
3. Observe TUI shows not connected to gateway — message not sent
4. Check ps aux — gateway is in D state at high CPU
5. Check ~/.openclaw/logs/stability/ — crash logs accumulate

Workaround

Downgrading to the stable release resolves the issue:

npm install -g [email protected]
openclaw gateway start
openclaw tui

Note: on 2026.4.15, the gateway does not auto-spawn with openclaw tui — openclaw gateway start must be run first (this sets up the systemd service).

Config

Config was last touched by version 2026.4.5 — no manual config changes were made between working and broken states. The issue appears to be a regression in the gateway binary/runtime itself.