rootandfactory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
0166fb6511
CI / cargo fmt (push) Has been cancelled
CI / cargo clippy (macos-latest) (push) Has been cancelled
CI / cargo clippy (ubuntu-latest) (push) Has been cancelled
CI / cargo clippy (windows-latest) (push) Has been cancelled
CI / cargo test (macos-latest) (push) Has been cancelled
CI / cargo test (ubuntu-latest) (push) Has been cancelled
CI / cargo test (windows-latest) (push) Has been cancelled
CI / cargo build (macos-latest) (push) Has been cancelled
CI / cargo build (ubuntu-latest) (push) Has been cancelled
CI / cargo build (windows-latest) (push) Has been cancelled
CI / cargo build --release (macos-latest) (push) Has been cancelled
CI / cargo build --release (ubuntu-latest) (push) Has been cancelled
CI / cargo build --release (windows-latest) (push) Has been cancelled
CI / CLI smoke (macos-latest) (push) Has been cancelled
CI / CLI smoke (ubuntu-latest) (push) Has been cancelled
CI / CLI smoke (windows-latest) (push) Has been cancelled
CI / Minimal E2E (macos-latest) (push) Has been cancelled
CI / Minimal E2E (ubuntu-latest) (push) Has been cancelled
CI / Minimal E2E (windows-latest) (push) Has been cancelled
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
58 lines
2.2 KiB
Markdown
58 lines
2.2 KiB
Markdown
# Reconnect and shutdown
|
|
|
|
The connector is designed to recover from tunnel interruptions without dropping the SOCKS5 proxy.
|
|
|
|
## Purpose
|
|
|
|
Lab networks and development environments are unstable. The connector should tolerate listener restarts, network hiccups, and temporary failures without requiring manual intervention.
|
|
|
|
## Reconnect behavior
|
|
|
|
`run_reconnect_loop` in `src/tunnel.rs` continuously attempts to maintain a tunnel:
|
|
|
|
1. Connect to the target address via TCP.
|
|
2. Perform TLS handshake and server certificate verification.
|
|
3. Send the HTTP auth request and wait for 200 OK.
|
|
4. On success, create a `StreamMux`, publish it via the `watch::channel`, and run the framing loop.
|
|
5. When the tunnel drops (framing task ends), log the disconnection and retry.
|
|
|
|
### Backoff
|
|
|
|
- Initial delay: 1 second
|
|
- Exponential doubling up to a maximum of 30 seconds
|
|
- Auth failures are terminal (the loop exits rather than retrying with a bad token)
|
|
|
|
### SOCKS5 proxy during reconnect
|
|
|
|
The SOCKS5 proxy stays bound. If a client connects while the tunnel is down, the proxy sees `mux_receiver` is `None` and rejects the connection with a warning log. Once the reconnect loop succeeds, new SOCKS5 requests work immediately.
|
|
|
|
## Shutdown
|
|
|
|
Both listener and connector handle graceful shutdown:
|
|
|
|
- Unix: SIGINT or SIGTERM triggers shutdown via `tokio::signal::unix`.
|
|
- Windows: Ctrl+C or Ctrl+Break triggers shutdown via `tokio::signal::ctrl_c`.
|
|
|
|
### Listener shutdown
|
|
|
|
1. The shutdown signal breaks the `tokio::select!` in `run_listener`.
|
|
2. The TCP listener is dropped, releasing the bind port.
|
|
3. The SOCKS5 proxy task (if any) is aborted.
|
|
4. Active framing sessions end when their streams close.
|
|
|
|
### Connector shutdown
|
|
|
|
1. The shutdown signal breaks the `tokio::select!` in `run_socks5_proxy`.
|
|
2. The SOCKS5 listener is dropped.
|
|
3. The reconnect loop handle is aborted.
|
|
4. Existing SOCKS5 connections receive EOF when their streams close.
|
|
|
|
E2E tests `e2e_graceful_shutdown_listener` and `e2e_graceful_shutdown_connector` verify that ports are released within 3 seconds of abort.
|
|
|
|
## Key source files
|
|
|
|
| File | Purpose |
|
|
| ---- | ------- |
|
|
| `src/tunnel.rs` | `run_reconnect_loop`, `run_socks5_proxy`, `run_listener`, shutdown handling |
|
|
| `src/signal.rs` | Cross-platform shutdown signal future |
|