openclaw - 💡(How to fix) Fix Bug: WebSocket close before connection established causes uncaughtException crash — not covered by isBenignUncaughtExceptionError [2 pull requests]

Summary

When a Feishu/WebSocket connection is closed before the handshake completes, Node.js emits an uncaughtException with the message "WebSocket was closed before the connection was established". This error is not classified as benign by isBenignUncaughtExceptionError() in unhandled-rejections--*.js, causing the gateway process to crash.

Current Behavior (v2026.5.28)

The benign error handler in dist/unhandled-rejections--*.js covers:

• Network errors: ECONNREFUSED, EHOSTUNREACH, ENETUNREACH, EAI_AGAIN, ENOTFOUND, ETIMEDOUT, UND_ERR_CONNECT_TIMEOUT, UND_ERR_DNS_RESOLVE_FAILED, UND_ERR_CONNECT
• I/O errors: EPIPE, EIO

But it does not cover WebSocket close messages, which are a common transient failure during Feishu gateway reconnection.

The uncaughtException handler in dist/index.js (line ~42) and dist/cli/run-main.js (line ~416) calls isBenignUncaughtExceptionError() which returns false for WebSocket close errors, leading to process.exit(1).

Suggested Fix

Option A (preferred): Add WebSocket close messages to the benign error classification in unhandled-rejections--*.js:

// In isBenignUncaughtExceptionError():
const msg = typeof error === 'string' ? error : (error?.message || '');
if (msg.includes("WebSocket was closed before the connection was established")) {
  return true;
}

Option B: Add a dedicated guard in the uncaughtException handler in index.js and cli/run-main.js, between the benign check and the fatal log:

const msg = formatUncaughtError(error);
if (typeof msg === "string" && msg.includes("WebSocket was closed before the connection was established")) {
  console.warn("[openclaw] Ignored recoverable uncaught exception:", msg);
  return;
}

Environment

• OpenClaw version: 2026.5.28 (67ddc1a)
• Files: dist/index.js, dist/cli/run-main.js, dist/unhandled-rejections--*.js
• Node.js: v22+
• Related: The upstream isBenignUncaughtExceptionError was added in 2026.4.29 but doesn't cover this case

Impact

Gateway process crash, requiring automatic restart via LaunchAgent. During network instability, this can cause a crash loop, degrading all channel availability.