Bug: SSH tunnel not restarted on unexpected exit

[shanselman]

Summary

When the SSH tunnel process exits unexpectedly, SshTunnelService only logs a warning. It does not clean up _process/_lastSpec, does not raise any event, and does not attempt a restart. The gateway client keeps reconnecting to ws://127.0.0.1:<localPort> which is now unreachable.

Location

src/OpenClaw.Tray.WinUI/Services/SshTunnelService.cs lines ~123-134 (process.Exited handler)

Impact

If the SSH tunnel crashes, the entire remote connectivity feature is broken until the user manually intervenes (e.g., changes settings or restarts the app). The reconnect loop will spin forever against a dead localhost port.

Flagged by

Multi-model code review consensus (2/3 models: Opus, GPT-5.4)

Suggested Fix

1. Null out _process and _lastSpec in the Exited handler so the next EnsureStarted call creates a fresh tunnel
2. Expose a TunnelExited event that App subscribes to for triggering automatic tunnel restart and gateway reconnect
3. Consider a bounded retry with backoff for automatic tunnel restart

[github-actions]

🤖 This is an automated response from Repo Assist.

This is a solid bug report. The process.Exited handler in SshTunnelService does log a warning, but neither clears state (_process/_lastSpec) nor signals the rest of the app — so any subsequent EnsureStarted call silently skips the "already running" early-out even though the tunnel is dead.

Draft PR #133 (created by Copilot) implements the fix:

• Adds a TunnelExited event raised only on unexpected exits (guarded by a _stopping flag to avoid firing during intentional Stop() calls)
• Nulls _process and _lastSpec in the Exited handler so the next EnsureStarted creates a fresh tunnel
• App.xaml.cs subscribes to TunnelExited and retries with bounded backoff (1 s → 2 s → 5 s → 10 s → 30 s), aborting early if the app is shutting down or SSH tunnelling is disabled

The gateway client's existing reconnect loop then recovers automatically once the tunnel port is live again — no additional reconnect logic needed at the gateway layer.

Generated by Repo Assist · ◷

To install this agentic workflow, run

gh aw add githubnext/agentics/workflows/repo-assist.md@cbb46ab386962aa371045839fc9998ee4e97ca64

[shanselman]

Fixed by PR #141 — SSH tunnel now auto-restarts on unexpected exit.