Summary

The gateway process exits with code 1 when a single outbound connection hits ENETUNREACH. A momentary network unreachability (e.g. Telegram's IP briefly unavailable) causes the entire gateway to crash, disconnecting all agents and sessions.

Error from log

2026-04-16T18:44:37.946+07:00 [openclaw] Uncaught exception: Error: connect ENETUNREACH 149.154.167.220:443 - Local (0.0.0.0:62855)
    at internalConnect (node:net:1110:16)
    ...
    at connect (/Users/greg/codebase/openclaw/node_modules/undici/lib/core/connect.js:70:20)
ELIFECYCLE  Command failed with exit code 1.

149.154.167.220 is a Telegram API IP. The connection error bubbled up as an uncaught exception and terminated the Node process.

Impact

• All TUI clients show "gateway disconnected: closed | idle"
• All agent sessions drop — "not connected to gateway — message not sent"
• Gateway must be manually restarted
• Any in-progress agent work is lost

Expected behavior

A failed outbound network connection (even a persistent one) should be caught and handled gracefully — log the error, back off and retry, and keep the gateway running. A single provider's network being temporarily unreachable should never take down the whole process.

Steps to reproduce

1. Run gateway with Telegram connected
2. Briefly interrupt network connectivity to Telegram's IPs (or have ISP/routing hiccup)
3. Gateway exits immediately

Suggested fix

• Wrap the Telegram connection/polling loop in a try/catch or attach an error event handler on the underlying socket/agent so ENETUNREACH is caught at the connection level
• Add a top-level process.on('uncaughtException') handler that logs the error and decides whether it is fatal rather than always crashing
• Alternatively, wrap outbound requests in a circuit breaker that absorbs transient network errors