[Remote Tunnels] Tunnel process becomes zombie (silent, no relay reconnect) after all clients disconnect and Extension Host shuts down

Summary

When all clients disconnect from a code tunnel (systemd user service), the reconnection grace period expires (default 3h), the Extension Host shuts down, and the relay WebSocket closes — the tunnel process does not attempt to re-establish the relay connection. It stays alive silently (no log output), and the tunnel appears offline on vscode.dev indefinitely. The only recovery is a full tunnel service uninstall + tunnel service install.

Environment

• VS Code CLI version: 1.112.0 (commit 07ff9d6178ede9a1bd12ad3399074d726ebe6e43)
• OS: Ubuntu 24.04 LTS (Linux 6.17.0, x86_64)
• Deployment: code tunnel service install → systemd user service with Restart=always
• Tunnel name: managed via jpe1.rel.tunnels.api.visualstudio.com

Steps to Reproduce

1. Run code tunnel service install and connect from vscode.dev.
2. Close all browser/client connections to the tunnel.
3. Wait for the ManagementConnection reconnection grace period to expire (3 hours).
4. Observe the Extension Host shutdown: Last EH closed, shutting down.
5. Check vscode.dev — tunnel shows offline.
6. Check systemctl --user status code-tunnel — process is still running (Active: running).
7. Check journalctl --user -u code-tunnel — complete silence from this point on; no reconnection attempts, no errors.

Expected Behavior

After the relay WebSocket closes (due to all clients disconnecting / EH shutdown), the tunnel process should re-register with the relay and restore the "online" state, so new clients can connect without manual intervention.

Actual Behavior

The process enters a zombie state: alive (no exit, so Restart=always never triggers) but permanently disconnected from the relay. The tunnel remains offline on vscode.dev until the service is manually uninstalled and reinstalled.

Relevant Log Sequence

# Client disconnects, grace period starts
[07:48:11] [<unknown>][59b930de][ManagementConnection] The client has disconnected, will wait for reconnection 3h before disposing...

# 3h later — grace period expires
[10:48:19] [<unknown>][59b930de][ManagementConnection] The reconnection grace time of 3h has expired, so the connection will be disposed.

# Extension Host shuts down
[10:52:51] Last EH closed, shutting down

# ← Complete silence from this point. No relay reconnect attempt. Ever.

Note: immediately before the grace period started, a transient DNS failure was observed during a relay reconnect cycle:

[07:47:46] warn failed to check GitHub token: error sending request for url (https://api.github.com/user): error trying to connect: dns error: failed to lookup address information: Try again

This may have contributed to the relay WebSocket not being re-established properly, but the core issue is that the process never retries.

Workaround

A watchdog script (run via systemd timer every 5 minutes) that detects log silence > 10 minutes and calls systemctl --user restart code-tunnel. This works because --verbose mode produces continuous output (liveness pongs, etc.) when the relay is genuinely connected.

Additional Context

• The process is running as a systemd user service (loginctl enable-linger is set).
• Two processes are spawned: a singleton server (long-lived) and a client wrapper that reconnects to it on respawn. After the EH shutdown, neither process attempts relay re-registration.
• Restart=always / RestartSec=10 in the service unit does not help because the process never exits.