Compare commits

..

18 Commits

Author SHA1 Message Date
579e553da0 v4.5.3 2026-03-15 19:26:39 +00:00
a8ee0b33d7 fix(remoteingress-core): prioritize control frames over data in edge and hub tunnel writers 2026-03-15 19:26:39 +00:00
43e320a36d v4.5.2 2026-03-15 18:16:10 +00:00
6ac4b37532 fix(remoteingress-core): improve stream flow control retries and increase channel buffer capacity 2026-03-15 18:16:10 +00:00
f456b0ba4f v4.5.1 2026-03-15 17:52:45 +00:00
69530f73aa fix(protocol): increase per-stream flow control window and channel buffers to improve high-RTT throughput 2026-03-15 17:52:45 +00:00
207b4a5cec v4.5.0 2026-03-15 17:33:59 +00:00
761551596b feat(remoteingress-core): add per-stream flow control for edge and hub tunnel data transfer 2026-03-15 17:33:59 +00:00
cf2d32bfe7 v4.4.1 2026-03-15 17:01:27 +00:00
4e9041c6a7 fix(remoteingress-core): prevent stream data loss by applying backpressure and closing saturated channels 2026-03-15 17:01:27 +00:00
86d4e9889a v4.4.0 2026-03-03 11:47:50 +00:00
45a2811f3e feat(remoteingress): add heartbeat PING/PONG and liveness timeouts; implement fast-reconnect/backoff reset and JS crash-recovery auto-restart 2026-03-03 11:47:50 +00:00
d6a07c28a0 v4.3.0 2026-02-26 23:47:16 +00:00
56a14aa7c5 feat(hub): add optional TLS certificate/key support to hub start config and bridge 2026-02-26 23:47:16 +00:00
417f62e646 v4.2.0 2026-02-26 23:02:23 +00:00
bda82f32ca feat(core): expose edge peer address in hub events and migrate writers to channel-based, non-blocking framing with stream limits and timeouts 2026-02-26 23:02:23 +00:00
4b06cb1b24 v4.1.0 2026-02-26 17:39:40 +00:00
1aae4b8c8e feat(remoteingress-bin): use mimalloc as the global allocator to reduce memory overhead and improve allocation performance 2026-02-26 17:39:40 +00:00
11 changed files with 678 additions and 72 deletions

View File

@@ -1,5 +1,71 @@
# Changelog
## 2026-03-15 - 4.5.3 - fix(remoteingress-core)
prioritize control frames over data in edge and hub tunnel writers
- Split tunnel/frame writers into separate control and data channels in edge and hub
- Use biased select loops so PING, PONG, WINDOW_UPDATE, OPEN, and CLOSE frames are sent before data frames
- Route stream data through dedicated data channels while keeping OPEN, CLOSE, and flow-control updates on control channels to prevent keepalive starvation under load
## 2026-03-15 - 4.5.2 - fix(remoteingress-core)
improve stream flow control retries and increase channel buffer capacity
- increase per-stream mpsc channel capacity from 128 to 256 on both edge and hub paths
- only reset accumulated window update bytes after a successful try_send to avoid dropping flow-control credits when the update channel is busy
## 2026-03-15 - 4.5.1 - fix(protocol)
increase per-stream flow control window and channel buffers to improve high-RTT throughput
- raise the initial stream window from 256 KB to 4 MB to allow more in-flight data per stream
- increase edge and hub mpsc channel capacities from 16 to 128 to better absorb throughput under flow control
## 2026-03-15 - 4.5.0 - feat(remoteingress-core)
add per-stream flow control for edge and hub tunnel data transfer
- introduce WINDOW_UPDATE frame types and protocol helpers for per-stream flow control
- track per-stream send windows on both edge and hub to limit reads based on available capacity
- send window updates after downstream writes to reduce channel pressure during large transfers
## 2026-03-15 - 4.4.1 - fix(remoteingress-core)
prevent stream data loss by applying backpressure and closing saturated channels
- replace non-blocking frame writes with awaited sends in per-stream tasks so large transfers respect backpressure instead of dropping data
- close and remove streams when back-channel or data channels fill up to avoid TCP stream corruption from silently dropped frames
## 2026-03-03 - 4.4.0 - feat(remoteingress)
add heartbeat PING/PONG and liveness timeouts; implement fast-reconnect/backoff reset and JS crash-recovery auto-restart
- protocol: add FRAME_PING and FRAME_PONG and unit tests for ping/pong frames
- edge (Rust): reset backoff after successful connection, respond to PING with PONG, track liveness via deadline and reconnect on timeout, use Duration/Instant helpers
- hub (Rust): send periodic PING to edges, handle PONGs, enforce liveness timeout and disconnect inactive edges, use tokio interval and time utilities
- ts: RemoteIngressEdge and RemoteIngressHub: add crash-recovery auto-restart with exponential backoff and max attempts, save/restore config and allowed edges, register/remove exit handlers, ensure stop() marks stopping and cleans up listeners
- minor API/typing: introduce TAllowedEdge alias and persist allowed edges for restart recovery
## 2026-02-26 - 4.3.0 - feat(hub)
add optional TLS certificate/key support to hub start config and bridge
- TypeScript: add tls.certPem and tls.keyPem to IHubConfig and include tlsCertPem/tlsKeyPem in startHub bridge command when both are provided
- TypeScript: extend startHub params with tlsCertPem and tlsKeyPem and conditionally send them
- Rust: change HubConfig serde attributes for tls_cert_pem and tls_key_pem from skip to default so absent PEM fields deserialize as None
- Enables optional provisioning of TLS certificate and key to the hub when provided from the JS side
## 2026-02-26 - 4.2.0 - feat(core)
expose edge peer address in hub events and migrate writers to channel-based, non-blocking framing with stream limits and timeouts
- Add peerAddr to ConnectedEdgeStatus and HubEvent::EdgeConnected and surface it to the TS frontend event (management:edgeConnected).
- Replace Arc<Mutex<WriteHalf>> writers with dedicated mpsc channel writer tasks in both hub and edge crates to serialize writes off the main tasks.
- Use non-blocking try_send for data frames to avoid head-of-line blocking and drop frames with warnings when channels are full.
- Introduce MAX_STREAMS_PER_EDGE semaphore to limit concurrent streams per edge and reject excess opens with a CLOSE_BACK frame.
- Add a 10s timeout when connecting to SmartProxy to avoid hanging connections.
- Ensure writer tasks are aborted on shutdown/cleanup and propagate cancellation tokens appropriately.
## 2026-02-26 - 4.1.0 - feat(remoteingress-bin)
use mimalloc as the global allocator to reduce memory overhead and improve allocation performance
- added mimalloc = "0.1" dependency to rust/crates/remoteingress-bin/Cargo.toml
- registered mimalloc as the #[global_allocator] in rust/crates/remoteingress-bin/src/main.rs
- updated Cargo.lock with libmimalloc-sys and mimalloc package entries
## 2026-02-26 - 4.0.1 - fix(hub)
cancel per-stream tokens on stream close and avoid duplicate StreamClosed events; bump @types/node devDependency to ^25.3.0

View File

@@ -1,6 +1,6 @@
{
"name": "@serve.zone/remoteingress",
"version": "4.0.1",
"version": "4.5.3",
"private": false,
"description": "Edge ingress tunnel for DcRouter - accepts incoming TCP connections at network edge and tunnels them to DcRouter SmartProxy preserving client IP via PROXY protocol v1.",
"main": "dist_ts/index.js",

20
rust/Cargo.lock generated
View File

@@ -327,6 +327,16 @@ version = "0.2.182"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "6800badb6cb2082ffd7b6a67e6125bb39f18782f793520caee8cb8846be06112"
[[package]]
name = "libmimalloc-sys"
version = "0.1.44"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "667f4fec20f29dfc6bc7357c582d91796c169ad7e2fce709468aefeb2c099870"
dependencies = [
"cc",
"libc",
]
[[package]]
name = "lock_api"
version = "0.4.14"
@@ -348,6 +358,15 @@ version = "2.8.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f8ca58f447f06ed17d5fc4043ce1b10dd205e060fb3ce5b979b8ed8e59ff3f79"
[[package]]
name = "mimalloc"
version = "0.1.48"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e1ee66a4b64c74f4ef288bcbb9192ad9c3feaad75193129ac8509af543894fd8"
dependencies = [
"libmimalloc-sys",
]
[[package]]
name = "mio"
version = "1.1.1"
@@ -519,6 +538,7 @@ dependencies = [
"clap",
"env_logger",
"log",
"mimalloc",
"remoteingress-core",
"remoteingress-protocol",
"rustls",

View File

@@ -17,3 +17,4 @@ serde_json = "1"
log = "0.4"
env_logger = "0.11"
rustls = { version = "0.23", default-features = false, features = ["ring"] }
mimalloc = "0.1"

View File

@@ -1,3 +1,6 @@
#[global_allocator]
static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;
use clap::Parser;
use serde::{Deserialize, Serialize};
use std::sync::Arc;
@@ -164,10 +167,10 @@ async fn handle_request(
tokio::spawn(async move {
while let Some(event) = event_rx.recv().await {
match &event {
HubEvent::EdgeConnected { edge_id } => {
HubEvent::EdgeConnected { edge_id, peer_addr } => {
send_event(
"edgeConnected",
serde_json::json!({ "edgeId": edge_id }),
serde_json::json!({ "edgeId": edge_id, "peerAddr": peer_addr }),
);
}
HubEvent::EdgeDisconnected { edge_id } => {

View File

@@ -1,16 +1,29 @@
use std::collections::HashMap;
use std::sync::atomic::{AtomicU32, Ordering};
use std::sync::Arc;
use std::time::Duration;
use tokio::io::{AsyncBufReadExt, AsyncReadExt, AsyncWriteExt, BufReader};
use tokio::net::{TcpListener, TcpStream};
use tokio::sync::{mpsc, Mutex, RwLock};
use tokio::sync::{mpsc, Mutex, Notify, RwLock};
use tokio::task::JoinHandle;
use tokio::time::{Instant, sleep_until};
use tokio_rustls::TlsConnector;
use tokio_util::sync::CancellationToken;
use serde::{Deserialize, Serialize};
use remoteingress_protocol::*;
/// Per-stream state tracked in the edge's client_writers map.
struct EdgeStreamState {
/// Channel to deliver FRAME_DATA_BACK payloads to the hub_to_client task.
back_tx: mpsc::Sender<Vec<u8>>,
/// Send window for FRAME_DATA (upload direction).
/// Decremented by the client reader, incremented by FRAME_WINDOW_UPDATE_BACK from hub.
send_window: Arc<AtomicU32>,
/// Notifier to wake the client reader when the window opens.
window_notify: Arc<Notify>,
}
/// Edge configuration (hub-host + credentials only; ports come from hub).
#[derive(Debug, Clone, Deserialize, Serialize)]
#[serde(rename_all = "camelCase")]
@@ -202,6 +215,13 @@ async fn edge_main_loop(
// Cancel connection token to kill all orphaned tasks from this cycle
connection_token.cancel();
// Reset backoff after a successful connection for fast reconnect
let was_connected = *connected.read().await;
if was_connected {
backoff_ms = 1000;
log::info!("Was connected; resetting backoff to {}ms for fast reconnect", backoff_ms);
}
*connected.write().await = false;
let _ = event_tx.try_send(EdgeEvent::TunnelDisconnected);
active_streams.store(0, Ordering::Relaxed);
@@ -214,7 +234,7 @@ async fn edge_main_loop(
EdgeLoopResult::Reconnect => {
log::info!("Reconnecting in {}ms...", backoff_ms);
tokio::select! {
_ = tokio::time::sleep(std::time::Duration::from_millis(backoff_ms)) => {}
_ = tokio::time::sleep(Duration::from_millis(backoff_ms)) => {}
_ = cancel_token.cancelled() => break,
_ = shutdown_rx.recv() => break,
}
@@ -336,25 +356,59 @@ async fn connect_to_hub_and_run(
_ = stun_token.cancelled() => break,
}
tokio::select! {
_ = tokio::time::sleep(std::time::Duration::from_secs(stun_interval)) => {}
_ = tokio::time::sleep(Duration::from_secs(stun_interval)) => {}
_ = stun_token.cancelled() => break,
}
}
});
// Client socket map: stream_id -> sender for writing data back to client
let client_writers: Arc<Mutex<HashMap<u32, mpsc::Sender<Vec<u8>>>>> =
// Client socket map: stream_id -> per-stream state (back channel + flow control)
let client_writers: Arc<Mutex<HashMap<u32, EdgeStreamState>>> =
Arc::new(Mutex::new(HashMap::new()));
// Shared tunnel writer
let tunnel_writer = Arc::new(Mutex::new(write_half));
// QoS dual-channel tunnel writer: control frames (PONG/WINDOW_UPDATE/CLOSE/OPEN)
// have priority over data frames (DATA). Prevents PING starvation under load.
let (tunnel_ctrl_tx, mut tunnel_ctrl_rx) = mpsc::channel::<Vec<u8>>(64);
let (tunnel_data_tx, mut tunnel_data_rx) = mpsc::channel::<Vec<u8>>(4096);
// Legacy alias — control channel for PONG, CLOSE, WINDOW_UPDATE, OPEN
let tunnel_writer_tx = tunnel_ctrl_tx.clone();
let tw_token = connection_token.clone();
let tunnel_writer_handle = tokio::spawn(async move {
loop {
tokio::select! {
biased; // control frames always take priority over data
ctrl = tunnel_ctrl_rx.recv() => {
match ctrl {
Some(frame_data) => {
if write_half.write_all(&frame_data).await.is_err() {
break;
}
}
None => break,
}
}
data = tunnel_data_rx.recv() => {
match data {
Some(frame_data) => {
if write_half.write_all(&frame_data).await.is_err() {
break;
}
}
None => break,
}
}
_ = tw_token.cancelled() => break,
}
}
});
// Start TCP listeners for initial ports (hot-reloadable)
let mut port_listeners: HashMap<u16, JoinHandle<()>> = HashMap::new();
apply_port_config(
&handshake.listen_ports,
&mut port_listeners,
&tunnel_writer,
&tunnel_writer_tx,
&tunnel_data_tx,
&client_writers,
active_streams,
next_stream_id,
@@ -362,6 +416,11 @@ async fn connect_to_hub_and_run(
connection_token,
);
// Heartbeat: liveness timeout detects silent hub failures
let liveness_timeout_dur = Duration::from_secs(45);
let mut last_activity = Instant::now();
let mut liveness_deadline = Box::pin(sleep_until(last_activity + liveness_timeout_dur));
// Read frames from hub
let mut frame_reader = FrameReader::new(buf_reader);
let result = loop {
@@ -369,11 +428,35 @@ async fn connect_to_hub_and_run(
frame_result = frame_reader.next_frame() => {
match frame_result {
Ok(Some(frame)) => {
// Reset liveness on any received frame
last_activity = Instant::now();
liveness_deadline.as_mut().reset(last_activity + liveness_timeout_dur);
match frame.frame_type {
FRAME_DATA_BACK => {
let writers = client_writers.lock().await;
if let Some(tx) = writers.get(&frame.stream_id) {
let _ = tx.send(frame.payload).await;
// Non-blocking dispatch to per-stream channel.
// With flow control, the sender should rarely exceed the channel capacity.
let mut writers = client_writers.lock().await;
if let Some(state) = writers.get(&frame.stream_id) {
if state.back_tx.try_send(frame.payload).is_err() {
log::warn!("Stream {} back-channel full, closing stream", frame.stream_id);
writers.remove(&frame.stream_id);
}
}
}
FRAME_WINDOW_UPDATE_BACK => {
// Hub consumed data — increase our send window for this stream (upload direction)
if let Some(increment) = decode_window_update(&frame.payload) {
if increment > 0 {
let writers = client_writers.lock().await;
if let Some(state) = writers.get(&frame.stream_id) {
let prev = state.send_window.fetch_add(increment, Ordering::Release);
if prev + increment > MAX_WINDOW_SIZE {
state.send_window.store(MAX_WINDOW_SIZE, Ordering::Release);
}
state.window_notify.notify_one();
}
}
}
}
FRAME_CLOSE_BACK => {
@@ -390,7 +473,8 @@ async fn connect_to_hub_and_run(
apply_port_config(
&update.listen_ports,
&mut port_listeners,
&tunnel_writer,
&tunnel_writer_tx,
&tunnel_data_tx,
&client_writers,
active_streams,
next_stream_id,
@@ -399,6 +483,14 @@ async fn connect_to_hub_and_run(
);
}
}
FRAME_PING => {
let pong_frame = encode_frame(0, FRAME_PONG, &[]);
if tunnel_writer_tx.try_send(pong_frame).is_err() {
log::warn!("Failed to send PONG, writer channel full/closed");
break EdgeLoopResult::Reconnect;
}
log::trace!("Received PING from hub, sent PONG");
}
_ => {
log::warn!("Unexpected frame type {} from hub", frame.frame_type);
}
@@ -414,6 +506,11 @@ async fn connect_to_hub_and_run(
}
}
}
_ = &mut liveness_deadline => {
log::warn!("Hub liveness timeout (no frames for {}s), reconnecting",
liveness_timeout_dur.as_secs());
break EdgeLoopResult::Reconnect;
}
_ = connection_token.cancelled() => {
log::info!("Connection cancelled");
break EdgeLoopResult::Shutdown;
@@ -427,6 +524,7 @@ async fn connect_to_hub_and_run(
// Cancel connection token to propagate to all child tasks BEFORE aborting
connection_token.cancel();
stun_handle.abort();
tunnel_writer_handle.abort();
for (_, h) in port_listeners.drain() {
h.abort();
}
@@ -438,8 +536,9 @@ async fn connect_to_hub_and_run(
fn apply_port_config(
new_ports: &[u16],
port_listeners: &mut HashMap<u16, JoinHandle<()>>,
tunnel_writer: &Arc<Mutex<tokio::io::WriteHalf<tokio_rustls::client::TlsStream<TcpStream>>>>,
client_writers: &Arc<Mutex<HashMap<u32, mpsc::Sender<Vec<u8>>>>>,
tunnel_ctrl_tx: &mpsc::Sender<Vec<u8>>,
tunnel_data_tx: &mpsc::Sender<Vec<u8>>,
client_writers: &Arc<Mutex<HashMap<u32, EdgeStreamState>>>,
active_streams: &Arc<AtomicU32>,
next_stream_id: &Arc<AtomicU32>,
edge_id: &str,
@@ -458,7 +557,8 @@ fn apply_port_config(
// Add new ports
for &port in new_set.difference(&old_set) {
let tunnel_writer = tunnel_writer.clone();
let tunnel_ctrl_tx = tunnel_ctrl_tx.clone();
let tunnel_data_tx = tunnel_data_tx.clone();
let client_writers = client_writers.clone();
let active_streams = active_streams.clone();
let next_stream_id = next_stream_id.clone();
@@ -481,7 +581,8 @@ fn apply_port_config(
match accept_result {
Ok((client_stream, client_addr)) => {
let stream_id = next_stream_id.fetch_add(1, Ordering::Relaxed);
let tunnel_writer = tunnel_writer.clone();
let tunnel_ctrl_tx = tunnel_ctrl_tx.clone();
let tunnel_data_tx = tunnel_data_tx.clone();
let client_writers = client_writers.clone();
let active_streams = active_streams.clone();
let edge_id = edge_id.clone();
@@ -496,7 +597,8 @@ fn apply_port_config(
stream_id,
port,
&edge_id,
tunnel_writer,
tunnel_ctrl_tx,
tunnel_data_tx,
client_writers,
client_token,
)
@@ -526,8 +628,9 @@ async fn handle_client_connection(
stream_id: u32,
dest_port: u16,
edge_id: &str,
tunnel_writer: Arc<Mutex<tokio::io::WriteHalf<tokio_rustls::client::TlsStream<TcpStream>>>>,
client_writers: Arc<Mutex<HashMap<u32, mpsc::Sender<Vec<u8>>>>>,
tunnel_ctrl_tx: mpsc::Sender<Vec<u8>>,
tunnel_data_tx: mpsc::Sender<Vec<u8>>,
client_writers: Arc<Mutex<HashMap<u32, EdgeStreamState>>>,
client_token: CancellationToken,
) {
let client_ip = client_addr.ip().to_string();
@@ -536,36 +639,52 @@ async fn handle_client_connection(
// Determine edge IP (use 0.0.0.0 as placeholder — hub doesn't use it for routing)
let edge_ip = "0.0.0.0";
// Send OPEN frame with PROXY v1 header
// Send OPEN frame with PROXY v1 header via control channel
let proxy_header = build_proxy_v1_header(&client_ip, edge_ip, client_port, dest_port);
let open_frame = encode_frame(stream_id, FRAME_OPEN, proxy_header.as_bytes());
{
let mut w = tunnel_writer.lock().await;
if w.write_all(&open_frame).await.is_err() {
return;
}
if tunnel_ctrl_tx.send(open_frame).await.is_err() {
return;
}
// Set up channel for data coming back from hub
// Set up channel for data coming back from hub (capacity 16 is sufficient with flow control)
let (back_tx, mut back_rx) = mpsc::channel::<Vec<u8>>(256);
let send_window = Arc::new(AtomicU32::new(INITIAL_STREAM_WINDOW));
let window_notify = Arc::new(Notify::new());
{
let mut writers = client_writers.lock().await;
writers.insert(stream_id, back_tx);
writers.insert(stream_id, EdgeStreamState {
back_tx,
send_window: Arc::clone(&send_window),
window_notify: Arc::clone(&window_notify),
});
}
let (mut client_read, mut client_write) = client_stream.into_split();
// Task: hub -> client
// Task: hub -> client (download direction)
// After writing to client TCP, send WINDOW_UPDATE to hub so it can send more
let hub_to_client_token = client_token.clone();
let wu_tx = tunnel_ctrl_tx.clone();
let hub_to_client = tokio::spawn(async move {
let mut consumed_since_update: u32 = 0;
loop {
tokio::select! {
data = back_rx.recv() => {
match data {
Some(data) => {
let len = data.len() as u32;
if client_write.write_all(&data).await.is_err() {
break;
}
// Track consumption for flow control
consumed_since_update += len;
if consumed_since_update >= WINDOW_UPDATE_THRESHOLD {
let frame = encode_window_update(stream_id, FRAME_WINDOW_UPDATE, consumed_since_update);
if wu_tx.try_send(frame).is_ok() {
consumed_since_update = 0;
}
// If try_send fails, keep accumulating — retry on next threshold
}
}
None => break,
}
@@ -573,20 +692,41 @@ async fn handle_client_connection(
_ = hub_to_client_token.cancelled() => break,
}
}
// Send final window update for any remaining consumed bytes
if consumed_since_update > 0 {
let frame = encode_window_update(stream_id, FRAME_WINDOW_UPDATE, consumed_since_update);
let _ = wu_tx.try_send(frame);
}
let _ = client_write.shutdown().await;
});
// Task: client -> hub
// Task: client -> hub (upload direction) with per-stream flow control
let mut buf = vec![0u8; 32768];
loop {
// Wait for send window to have capacity
loop {
let w = send_window.load(Ordering::Acquire);
if w > 0 { break; }
tokio::select! {
_ = window_notify.notified() => continue,
_ = client_token.cancelled() => break,
}
}
if client_token.is_cancelled() { break; }
// Limit read size to available window
let w = send_window.load(Ordering::Acquire) as usize;
let max_read = w.min(buf.len());
tokio::select! {
read_result = client_read.read(&mut buf) => {
read_result = client_read.read(&mut buf[..max_read]) => {
match read_result {
Ok(0) => break,
Ok(n) => {
send_window.fetch_sub(n as u32, Ordering::Release);
let data_frame = encode_frame(stream_id, FRAME_DATA, &buf[..n]);
let mut w = tunnel_writer.lock().await;
if w.write_all(&data_frame).await.is_err() {
if tunnel_data_tx.send(data_frame).await.is_err() {
log::warn!("Stream {} data channel closed, closing", stream_id);
break;
}
}
@@ -597,11 +737,10 @@ async fn handle_client_connection(
}
}
// Send CLOSE frame (only if not cancelled)
// Send CLOSE frame via control channel (only if not cancelled)
if !client_token.is_cancelled() {
let close_frame = encode_frame(stream_id, FRAME_CLOSE, &[]);
let mut w = tunnel_writer.lock().await;
let _ = w.write_all(&close_frame).await;
let _ = tunnel_ctrl_tx.try_send(close_frame);
}
// Cleanup

View File

@@ -1,23 +1,39 @@
use std::collections::HashMap;
use std::sync::Arc;
use std::sync::atomic::{AtomicU32, Ordering};
use std::time::Duration;
use tokio::io::{AsyncBufReadExt, AsyncReadExt, AsyncWriteExt, BufReader};
use tokio::net::{TcpListener, TcpStream};
use tokio::sync::{mpsc, Mutex, RwLock};
use tokio::sync::{mpsc, Mutex, Notify, RwLock, Semaphore};
use tokio::time::{interval, sleep_until, Instant};
use tokio_rustls::TlsAcceptor;
use tokio_util::sync::CancellationToken;
use serde::{Deserialize, Serialize};
use remoteingress_protocol::*;
/// Per-stream state tracked in the hub's stream map.
struct HubStreamState {
/// Channel to deliver FRAME_DATA payloads to the upstream writer task.
data_tx: mpsc::Sender<Vec<u8>>,
/// Cancellation token for this stream.
cancel_token: CancellationToken,
/// Send window for FRAME_DATA_BACK (download direction).
/// Decremented by the upstream reader, incremented by FRAME_WINDOW_UPDATE from edge.
send_window: Arc<AtomicU32>,
/// Notifier to wake the upstream reader when the window opens.
window_notify: Arc<Notify>,
}
/// Hub configuration.
#[derive(Debug, Clone, Deserialize, Serialize)]
#[serde(rename_all = "camelCase")]
pub struct HubConfig {
pub tunnel_port: u16,
pub target_host: Option<String>,
#[serde(skip)]
#[serde(default)]
pub tls_cert_pem: Option<String>,
#[serde(skip)]
#[serde(default)]
pub tls_key_pem: Option<String>,
}
@@ -65,6 +81,7 @@ pub struct ConnectedEdgeStatus {
pub edge_id: String,
pub connected_at: u64,
pub active_streams: usize,
pub peer_addr: String,
}
/// Events emitted by the hub.
@@ -73,7 +90,7 @@ pub struct ConnectedEdgeStatus {
#[serde(tag = "type")]
pub enum HubEvent {
#[serde(rename_all = "camelCase")]
EdgeConnected { edge_id: String },
EdgeConnected { edge_id: String, peer_addr: String },
#[serde(rename_all = "camelCase")]
EdgeDisconnected { edge_id: String },
#[serde(rename_all = "camelCase")]
@@ -105,7 +122,8 @@ pub struct TunnelHub {
struct ConnectedEdgeInfo {
connected_at: u64,
active_streams: Arc<Mutex<HashMap<u32, (mpsc::Sender<Vec<u8>>, CancellationToken)>>>,
peer_addr: String,
active_streams: Arc<Mutex<HashMap<u32, HubStreamState>>>,
config_tx: mpsc::Sender<EdgeConfigUpdate>,
#[allow(dead_code)] // kept alive for Drop — cancels child tokens when edge is removed
cancel_token: CancellationToken,
@@ -176,6 +194,7 @@ impl TunnelHub {
edge_id: id.clone(),
connected_at: info.connected_at,
active_streams: streams.len(),
peer_addr: info.peer_addr.clone(),
});
}
@@ -218,9 +237,10 @@ impl TunnelHub {
let event_tx = event_tx.clone();
let target = target_host.clone();
let edge_token = hub_token.child_token();
let peer_addr = addr.ip().to_string();
tokio::spawn(async move {
if let Err(e) = handle_edge_connection(
stream, acceptor, allowed, connected, event_tx, target, edge_token,
stream, acceptor, allowed, connected, event_tx, target, edge_token, peer_addr,
).await {
log::error!("Edge connection error: {}", e);
}
@@ -264,6 +284,9 @@ impl Drop for TunnelHub {
}
}
/// Maximum concurrent streams per edge connection.
const MAX_STREAMS_PER_EDGE: usize = 1024;
/// Handle a single edge connection: authenticate, then enter frame loop.
async fn handle_edge_connection(
stream: TcpStream,
@@ -273,6 +296,7 @@ async fn handle_edge_connection(
event_tx: mpsc::Sender<HubEvent>,
target_host: String,
edge_token: CancellationToken,
peer_addr: String,
) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
let tls_stream = acceptor.accept(stream).await?;
let (read_half, mut write_half) = tokio::io::split(tls_stream);
@@ -307,9 +331,10 @@ async fn handle_edge_connection(
}
};
log::info!("Edge {} authenticated", edge_id);
log::info!("Edge {} authenticated from {}", edge_id, peer_addr);
let _ = event_tx.try_send(HubEvent::EdgeConnected {
edge_id: edge_id.clone(),
peer_addr: peer_addr.clone(),
});
// Send handshake response with initial config before frame protocol begins
@@ -322,7 +347,7 @@ async fn handle_edge_connection(
write_half.write_all(handshake_json.as_bytes()).await?;
// Track this edge
let streams: Arc<Mutex<HashMap<u32, (mpsc::Sender<Vec<u8>>, CancellationToken)>>> =
let streams: Arc<Mutex<HashMap<u32, HubStreamState>>> =
Arc::new(Mutex::new(HashMap::new()));
let now = std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
@@ -338,6 +363,7 @@ async fn handle_edge_connection(
edge_id.clone(),
ConnectedEdgeInfo {
connected_at: now,
peer_addr,
active_streams: streams.clone(),
config_tx,
cancel_token: edge_token.clone(),
@@ -345,11 +371,44 @@ async fn handle_edge_connection(
);
}
// Shared writer for sending frames back to edge
let write_half = Arc::new(Mutex::new(write_half));
// QoS dual-channel tunnel writer: control frames (PING/PONG/WINDOW_UPDATE/CLOSE)
// have priority over data frames (DATA_BACK). This prevents PING starvation under load.
let (ctrl_tx, mut ctrl_rx) = mpsc::channel::<Vec<u8>>(64);
let (data_tx, mut data_rx) = mpsc::channel::<Vec<u8>>(4096);
// Legacy alias for code that sends both control and data (will be migrated)
let frame_writer_tx = ctrl_tx.clone();
let writer_token = edge_token.clone();
let writer_handle = tokio::spawn(async move {
loop {
tokio::select! {
biased; // control frames always take priority over data
ctrl = ctrl_rx.recv() => {
match ctrl {
Some(frame_data) => {
if write_half.write_all(&frame_data).await.is_err() {
break;
}
}
None => break,
}
}
data = data_rx.recv() => {
match data {
Some(frame_data) => {
if write_half.write_all(&frame_data).await.is_err() {
break;
}
}
None => break,
}
}
_ = writer_token.cancelled() => break,
}
}
});
// Spawn task to forward config updates as FRAME_CONFIG frames
let config_writer = write_half.clone();
let config_writer_tx = frame_writer_tx.clone();
let config_edge_id = edge_id.clone();
let config_token = edge_token.clone();
let config_handle = tokio::spawn(async move {
@@ -360,8 +419,7 @@ async fn handle_edge_connection(
Some(update) => {
if let Ok(payload) = serde_json::to_vec(&update) {
let frame = encode_frame(0, FRAME_CONFIG, &payload);
let mut w = config_writer.lock().await;
if w.write_all(&frame).await.is_err() {
if config_writer_tx.send(frame).await.is_err() {
log::error!("Failed to send config update to edge {}", config_edge_id);
break;
}
@@ -376,6 +434,17 @@ async fn handle_edge_connection(
}
});
// A4: Semaphore to limit concurrent streams per edge
let stream_semaphore = Arc::new(Semaphore::new(MAX_STREAMS_PER_EDGE));
// Heartbeat: periodic PING and liveness timeout
let ping_interval_dur = Duration::from_secs(15);
let liveness_timeout_dur = Duration::from_secs(45);
let mut ping_ticker = interval(ping_interval_dur);
ping_ticker.tick().await; // consume the immediate first tick
let mut last_activity = Instant::now();
let mut liveness_deadline = Box::pin(sleep_until(last_activity + liveness_timeout_dur));
// Frame reading loop
let mut frame_reader = FrameReader::new(buf_reader);
@@ -384,8 +453,24 @@ async fn handle_edge_connection(
frame_result = frame_reader.next_frame() => {
match frame_result {
Ok(Some(frame)) => {
// Reset liveness on any received frame
last_activity = Instant::now();
liveness_deadline.as_mut().reset(last_activity + liveness_timeout_dur);
match frame.frame_type {
FRAME_OPEN => {
// A4: Check stream limit before processing
let permit = match stream_semaphore.clone().try_acquire_owned() {
Ok(p) => p,
Err(_) => {
log::warn!("Edge {} exceeded max streams ({}), rejecting stream {}",
edge_id, MAX_STREAMS_PER_EDGE, frame.stream_id);
let close_frame = encode_frame(frame.stream_id, FRAME_CLOSE_BACK, &[]);
let _ = frame_writer_tx.try_send(close_frame);
continue;
}
};
// Payload is PROXY v1 header line
let proxy_header = String::from_utf8_lossy(&frame.payload).to_string();
@@ -396,7 +481,8 @@ async fn handle_edge_connection(
let edge_id_clone = edge_id.clone();
let event_tx_clone = event_tx.clone();
let streams_clone = streams.clone();
let writer_clone = write_half.clone();
let writer_tx = ctrl_tx.clone(); // control: CLOSE_BACK, WINDOW_UPDATE_BACK
let data_writer_tx = data_tx.clone(); // data: DATA_BACK
let target = target_host.clone();
let stream_token = edge_token.child_token();
@@ -405,34 +491,64 @@ async fn handle_edge_connection(
stream_id,
});
// Create channel for data from edge to this stream
// Create channel for data from edge to this stream (capacity 16 is sufficient with flow control)
let (data_tx, mut data_rx) = mpsc::channel::<Vec<u8>>(256);
let send_window = Arc::new(AtomicU32::new(INITIAL_STREAM_WINDOW));
let window_notify = Arc::new(Notify::new());
{
let mut s = streams.lock().await;
s.insert(stream_id, (data_tx, stream_token.clone()));
s.insert(stream_id, HubStreamState {
data_tx,
cancel_token: stream_token.clone(),
send_window: Arc::clone(&send_window),
window_notify: Arc::clone(&window_notify),
});
}
// Spawn task: connect to SmartProxy, send PROXY header, pipe data
tokio::spawn(async move {
let _permit = permit; // hold semaphore permit until stream completes
let result = async {
let mut upstream =
TcpStream::connect((target.as_str(), dest_port)).await?;
// A2: Connect to SmartProxy with timeout
let mut upstream = tokio::time::timeout(
Duration::from_secs(10),
TcpStream::connect((target.as_str(), dest_port)),
)
.await
.map_err(|_| -> Box<dyn std::error::Error + Send + Sync> {
format!("connect to SmartProxy {}:{} timed out (10s)", target, dest_port).into()
})??;
upstream.write_all(proxy_header.as_bytes()).await?;
let (mut up_read, mut up_write) =
upstream.into_split();
// Forward data from edge (via channel) to SmartProxy
// After writing to upstream, send WINDOW_UPDATE_BACK to edge
let writer_token = stream_token.clone();
let wub_tx = writer_tx.clone();
let writer_for_edge_data = tokio::spawn(async move {
let mut consumed_since_update: u32 = 0;
loop {
tokio::select! {
data = data_rx.recv() => {
match data {
Some(data) => {
let len = data.len() as u32;
if up_write.write_all(&data).await.is_err() {
break;
}
// Track consumption for flow control
consumed_since_update += len;
if consumed_since_update >= WINDOW_UPDATE_THRESHOLD {
let frame = encode_window_update(stream_id, FRAME_WINDOW_UPDATE_BACK, consumed_since_update);
if wub_tx.try_send(frame).is_ok() {
consumed_since_update = 0;
}
// If try_send fails, keep accumulating — retry on next threshold
}
}
None => break,
}
@@ -440,21 +556,43 @@ async fn handle_edge_connection(
_ = writer_token.cancelled() => break,
}
}
// Send final window update for remaining consumed bytes
if consumed_since_update > 0 {
let frame = encode_window_update(stream_id, FRAME_WINDOW_UPDATE_BACK, consumed_since_update);
let _ = wub_tx.try_send(frame);
}
let _ = up_write.shutdown().await;
});
// Forward data from SmartProxy back to edge
// Forward data from SmartProxy back to edge via writer channel
// with per-stream flow control (check send_window before reading)
let mut buf = vec![0u8; 32768];
loop {
// Wait for send window to have capacity
loop {
let w = send_window.load(Ordering::Acquire);
if w > 0 { break; }
tokio::select! {
_ = window_notify.notified() => continue,
_ = stream_token.cancelled() => break,
}
}
if stream_token.is_cancelled() { break; }
// Limit read size to available window
let w = send_window.load(Ordering::Acquire) as usize;
let max_read = w.min(buf.len());
tokio::select! {
read_result = up_read.read(&mut buf) => {
read_result = up_read.read(&mut buf[..max_read]) => {
match read_result {
Ok(0) => break,
Ok(n) => {
send_window.fetch_sub(n as u32, Ordering::Release);
let frame =
encode_frame(stream_id, FRAME_DATA_BACK, &buf[..n]);
let mut w = writer_clone.lock().await;
if w.write_all(&frame).await.is_err() {
if data_writer_tx.send(frame).await.is_err() {
log::warn!("Stream {} data channel closed, closing", stream_id);
break;
}
}
@@ -468,8 +606,7 @@ async fn handle_edge_connection(
// Send CLOSE_BACK to edge (only if not cancelled)
if !stream_token.is_cancelled() {
let close_frame = encode_frame(stream_id, FRAME_CLOSE_BACK, &[]);
let mut w = writer_clone.lock().await;
let _ = w.write_all(&close_frame).await;
let _ = writer_tx.try_send(close_frame);
}
writer_for_edge_data.abort();
@@ -482,8 +619,7 @@ async fn handle_edge_connection(
// Send CLOSE_BACK on error (only if not cancelled)
if !stream_token.is_cancelled() {
let close_frame = encode_frame(stream_id, FRAME_CLOSE_BACK, &[]);
let mut w = writer_clone.lock().await;
let _ = w.write_all(&close_frame).await;
let _ = writer_tx.try_send(close_frame);
}
}
@@ -501,21 +637,46 @@ async fn handle_edge_connection(
});
}
FRAME_DATA => {
let s = streams.lock().await;
if let Some((tx, _)) = s.get(&frame.stream_id) {
let _ = tx.send(frame.payload).await;
// Non-blocking dispatch to per-stream channel.
// With flow control, the sender should rarely exceed the channel capacity.
let mut s = streams.lock().await;
if let Some(state) = s.get(&frame.stream_id) {
if state.data_tx.try_send(frame.payload).is_err() {
log::warn!("Stream {} data channel full, closing stream", frame.stream_id);
if let Some(state) = s.remove(&frame.stream_id) {
state.cancel_token.cancel();
}
}
}
}
FRAME_WINDOW_UPDATE => {
// Edge consumed data — increase our send window for this stream
if let Some(increment) = decode_window_update(&frame.payload) {
if increment > 0 {
let s = streams.lock().await;
if let Some(state) = s.get(&frame.stream_id) {
let prev = state.send_window.fetch_add(increment, Ordering::Release);
if prev + increment > MAX_WINDOW_SIZE {
state.send_window.store(MAX_WINDOW_SIZE, Ordering::Release);
}
state.window_notify.notify_one();
}
}
}
}
FRAME_CLOSE => {
let mut s = streams.lock().await;
if let Some((_, token)) = s.remove(&frame.stream_id) {
token.cancel();
if let Some(state) = s.remove(&frame.stream_id) {
state.cancel_token.cancel();
let _ = event_tx.try_send(HubEvent::StreamClosed {
edge_id: edge_id.clone(),
stream_id: frame.stream_id,
});
}
}
FRAME_PONG => {
log::debug!("Received PONG from edge {}", edge_id);
}
_ => {
log::warn!("Unexpected frame type {} from edge", frame.frame_type);
}
@@ -531,6 +692,19 @@ async fn handle_edge_connection(
}
}
}
_ = ping_ticker.tick() => {
let ping_frame = encode_frame(0, FRAME_PING, &[]);
if frame_writer_tx.try_send(ping_frame).is_err() {
log::warn!("Failed to send PING to edge {}, writer channel full/closed", edge_id);
break;
}
log::trace!("Sent PING to edge {}", edge_id);
}
_ = &mut liveness_deadline => {
log::warn!("Edge {} liveness timeout (no frames for {}s), disconnecting",
edge_id, liveness_timeout_dur.as_secs());
break;
}
_ = edge_token.cancelled() => {
log::info!("Edge {} cancelled by hub", edge_id);
break;
@@ -541,6 +715,7 @@ async fn handle_edge_connection(
// Cleanup: cancel edge token to propagate to all child tasks
edge_token.cancel();
config_handle.abort();
writer_handle.abort();
{
let mut edges = connected.lock().await;
edges.remove(&edge_id);
@@ -757,10 +932,12 @@ mod tests {
fn test_hub_event_edge_connected_serialize() {
let event = HubEvent::EdgeConnected {
edge_id: "edge-1".to_string(),
peer_addr: "203.0.113.5".to_string(),
};
let json = serde_json::to_value(&event).unwrap();
assert_eq!(json["type"], "edgeConnected");
assert_eq!(json["edgeId"], "edge-1");
assert_eq!(json["peerAddr"], "203.0.113.5");
}
#[test]

View File

@@ -7,6 +7,10 @@ pub const FRAME_CLOSE: u8 = 0x03;
pub const FRAME_DATA_BACK: u8 = 0x04;
pub const FRAME_CLOSE_BACK: u8 = 0x05;
pub const FRAME_CONFIG: u8 = 0x06; // Hub -> Edge: configuration update
pub const FRAME_PING: u8 = 0x07; // Hub -> Edge: heartbeat probe
pub const FRAME_PONG: u8 = 0x08; // Edge -> Hub: heartbeat response
pub const FRAME_WINDOW_UPDATE: u8 = 0x09; // Edge -> Hub: per-stream flow control
pub const FRAME_WINDOW_UPDATE_BACK: u8 = 0x0A; // Hub -> Edge: per-stream flow control
// Frame header size: 4 (stream_id) + 1 (type) + 4 (length) = 9 bytes
pub const FRAME_HEADER_SIZE: usize = 9;
@@ -14,6 +18,28 @@ pub const FRAME_HEADER_SIZE: usize = 9;
// Maximum payload size (16 MB)
pub const MAX_PAYLOAD_SIZE: u32 = 16 * 1024 * 1024;
// Per-stream flow control constants
/// Initial per-stream window size (4 MB). Sized for full throughput at high RTT:
/// at 100ms RTT, this sustains ~40 MB/s per stream.
pub const INITIAL_STREAM_WINDOW: u32 = 4 * 1024 * 1024;
/// Send WINDOW_UPDATE after consuming this many bytes (half the initial window).
pub const WINDOW_UPDATE_THRESHOLD: u32 = INITIAL_STREAM_WINDOW / 2;
/// Maximum window size to prevent overflow.
pub const MAX_WINDOW_SIZE: u32 = 16 * 1024 * 1024;
/// Encode a WINDOW_UPDATE frame for a specific stream.
pub fn encode_window_update(stream_id: u32, frame_type: u8, increment: u32) -> Vec<u8> {
encode_frame(stream_id, frame_type, &increment.to_be_bytes())
}
/// Decode a WINDOW_UPDATE payload into a byte increment. Returns None if payload is malformed.
pub fn decode_window_update(payload: &[u8]) -> Option<u32> {
if payload.len() != 4 {
return None;
}
Some(u32::from_be_bytes([payload[0], payload[1], payload[2], payload[3]]))
}
/// A single multiplexed frame.
#[derive(Debug, Clone)]
pub struct Frame {
@@ -261,6 +287,8 @@ mod tests {
FRAME_DATA_BACK,
FRAME_CLOSE_BACK,
FRAME_CONFIG,
FRAME_PING,
FRAME_PONG,
];
let mut data = Vec::new();
@@ -293,4 +321,19 @@ mod tests {
assert_eq!(frame.frame_type, FRAME_CLOSE);
assert!(frame.payload.is_empty());
}
#[test]
fn test_encode_frame_ping_pong() {
// PING: stream_id=0, empty payload (control frame)
let ping = encode_frame(0, FRAME_PING, &[]);
assert_eq!(ping[4], FRAME_PING);
assert_eq!(&ping[0..4], &0u32.to_be_bytes());
assert_eq!(ping.len(), FRAME_HEADER_SIZE);
// PONG: stream_id=0, empty payload (control frame)
let pong = encode_frame(0, FRAME_PONG, &[]);
assert_eq!(pong[4], FRAME_PONG);
assert_eq!(&pong[0..4], &0u32.to_be_bytes());
assert_eq!(pong.len(), FRAME_HEADER_SIZE);
}
}

View File

@@ -3,6 +3,6 @@
*/
export const commitinfo = {
name: '@serve.zone/remoteingress',
version: '4.0.1',
version: '4.5.3',
description: 'Edge ingress tunnel for DcRouter - accepts incoming TCP connections at network edge and tunnels them to DcRouter SmartProxy preserving client IP via PROXY protocol v1.'
}

View File

@@ -40,9 +40,16 @@ export interface IEdgeConfig {
secret: string;
}
const MAX_RESTART_ATTEMPTS = 10;
const MAX_RESTART_BACKOFF_MS = 30_000;
export class RemoteIngressEdge extends EventEmitter {
private bridge: InstanceType<typeof plugins.smartrust.RustBridge<TEdgeCommands>>;
private started = false;
private stopping = false;
private savedConfig: IEdgeConfig | null = null;
private restartBackoffMs = 1000;
private restartAttempts = 0;
private statusInterval: ReturnType<typeof setInterval> | undefined;
constructor() {
@@ -109,11 +116,17 @@ export class RemoteIngressEdge extends EventEmitter {
edgeConfig = config;
}
this.savedConfig = edgeConfig;
this.stopping = false;
const spawned = await this.bridge.spawn();
if (!spawned) {
throw new Error('Failed to spawn remoteingress-bin');
}
// Register crash recovery handler
this.bridge.on('exit', this.handleCrashRecovery);
await this.bridge.sendCommand('startEdge', {
hubHost: edgeConfig.hubHost,
hubPort: edgeConfig.hubPort ?? 8443,
@@ -122,6 +135,8 @@ export class RemoteIngressEdge extends EventEmitter {
});
this.started = true;
this.restartAttempts = 0;
this.restartBackoffMs = 1000;
// Start periodic status logging
this.statusInterval = setInterval(async () => {
@@ -142,6 +157,7 @@ export class RemoteIngressEdge extends EventEmitter {
* Stop the edge and kill the Rust process.
*/
public async stop(): Promise<void> {
this.stopping = true;
if (this.statusInterval) {
clearInterval(this.statusInterval);
this.statusInterval = undefined;
@@ -152,6 +168,7 @@ export class RemoteIngressEdge extends EventEmitter {
} catch {
// Process may already be dead
}
this.bridge.removeListener('exit', this.handleCrashRecovery);
this.bridge.kill();
this.started = false;
}
@@ -170,4 +187,55 @@ export class RemoteIngressEdge extends EventEmitter {
public get running(): boolean {
return this.bridge.running;
}
/**
* Handle unexpected Rust binary crash — auto-restart with backoff.
*/
private handleCrashRecovery = async (code: number | null, signal: string | null) => {
if (this.stopping || !this.started || !this.savedConfig) {
return;
}
console.error(
`[RemoteIngressEdge] Rust binary crashed (code=${code}, signal=${signal}), ` +
`attempt ${this.restartAttempts + 1}/${MAX_RESTART_ATTEMPTS}`
);
this.started = false;
if (this.restartAttempts >= MAX_RESTART_ATTEMPTS) {
console.error('[RemoteIngressEdge] Max restart attempts reached, giving up');
this.emit('crashRecoveryFailed');
return;
}
await new Promise(resolve => setTimeout(resolve, this.restartBackoffMs));
this.restartBackoffMs = Math.min(this.restartBackoffMs * 2, MAX_RESTART_BACKOFF_MS);
this.restartAttempts++;
try {
const spawned = await this.bridge.spawn();
if (!spawned) {
console.error('[RemoteIngressEdge] Failed to respawn binary');
return;
}
this.bridge.on('exit', this.handleCrashRecovery);
await this.bridge.sendCommand('startEdge', {
hubHost: this.savedConfig.hubHost,
hubPort: this.savedConfig.hubPort ?? 8443,
edgeId: this.savedConfig.edgeId,
secret: this.savedConfig.secret,
});
this.started = true;
this.restartAttempts = 0;
this.restartBackoffMs = 1000;
console.log('[RemoteIngressEdge] Successfully recovered from crash');
this.emit('crashRecovered');
} catch (err) {
console.error(`[RemoteIngressEdge] Crash recovery failed: ${err}`);
}
};
}

View File

@@ -11,6 +11,8 @@ type THubCommands = {
params: {
tunnelPort: number;
targetHost?: string;
tlsCertPem?: string;
tlsKeyPem?: string;
};
result: { started: boolean };
};
@@ -33,6 +35,7 @@ type THubCommands = {
edgeId: string;
connectedAt: number;
activeStreams: number;
peerAddr: string;
}>;
};
};
@@ -41,11 +44,25 @@ type THubCommands = {
export interface IHubConfig {
tunnelPort?: number;
targetHost?: string;
tls?: {
certPem?: string;
keyPem?: string;
};
}
type TAllowedEdge = { id: string; secret: string; listenPorts?: number[]; stunIntervalSecs?: number };
const MAX_RESTART_ATTEMPTS = 10;
const MAX_RESTART_BACKOFF_MS = 30_000;
export class RemoteIngressHub extends EventEmitter {
private bridge: InstanceType<typeof plugins.smartrust.RustBridge<THubCommands>>;
private started = false;
private stopping = false;
private savedConfig: IHubConfig | null = null;
private savedEdges: TAllowedEdge[] = [];
private restartBackoffMs = 1000;
private restartAttempts = 0;
constructor() {
super();
@@ -73,7 +90,7 @@ export class RemoteIngressHub extends EventEmitter {
});
// Forward events from Rust binary
this.bridge.on('management:edgeConnected', (data: { edgeId: string }) => {
this.bridge.on('management:edgeConnected', (data: { edgeId: string; peerAddr: string }) => {
this.emit('edgeConnected', data);
});
this.bridge.on('management:edgeDisconnected', (data: { edgeId: string }) => {
@@ -91,29 +108,42 @@ export class RemoteIngressHub extends EventEmitter {
* Start the hub — spawns the Rust binary and starts the tunnel server.
*/
public async start(config: IHubConfig = {}): Promise<void> {
this.savedConfig = config;
this.stopping = false;
const spawned = await this.bridge.spawn();
if (!spawned) {
throw new Error('Failed to spawn remoteingress-bin');
}
// Register crash recovery handler
this.bridge.on('exit', this.handleCrashRecovery);
await this.bridge.sendCommand('startHub', {
tunnelPort: config.tunnelPort ?? 8443,
targetHost: config.targetHost ?? '127.0.0.1',
...(config.tls?.certPem && config.tls?.keyPem
? { tlsCertPem: config.tls.certPem, tlsKeyPem: config.tls.keyPem }
: {}),
});
this.started = true;
this.restartAttempts = 0;
this.restartBackoffMs = 1000;
}
/**
* Stop the hub and kill the Rust process.
*/
public async stop(): Promise<void> {
this.stopping = true;
if (this.started) {
try {
await this.bridge.sendCommand('stopHub', {} as Record<string, never>);
} catch {
// Process may already be dead
}
this.bridge.removeListener('exit', this.handleCrashRecovery);
this.bridge.kill();
this.started = false;
}
@@ -122,7 +152,8 @@ export class RemoteIngressHub extends EventEmitter {
/**
* Update the list of allowed edges that can connect to this hub.
*/
public async updateAllowedEdges(edges: Array<{ id: string; secret: string; listenPorts?: number[]; stunIntervalSecs?: number }>): Promise<void> {
public async updateAllowedEdges(edges: TAllowedEdge[]): Promise<void> {
this.savedEdges = edges;
await this.bridge.sendCommand('updateAllowedEdges', { edges });
}
@@ -139,4 +170,62 @@ export class RemoteIngressHub extends EventEmitter {
public get running(): boolean {
return this.bridge.running;
}
/**
* Handle unexpected Rust binary crash — auto-restart with backoff.
*/
private handleCrashRecovery = async (code: number | null, signal: string | null) => {
if (this.stopping || !this.started || !this.savedConfig) {
return;
}
console.error(
`[RemoteIngressHub] Rust binary crashed (code=${code}, signal=${signal}), ` +
`attempt ${this.restartAttempts + 1}/${MAX_RESTART_ATTEMPTS}`
);
this.started = false;
if (this.restartAttempts >= MAX_RESTART_ATTEMPTS) {
console.error('[RemoteIngressHub] Max restart attempts reached, giving up');
this.emit('crashRecoveryFailed');
return;
}
await new Promise(resolve => setTimeout(resolve, this.restartBackoffMs));
this.restartBackoffMs = Math.min(this.restartBackoffMs * 2, MAX_RESTART_BACKOFF_MS);
this.restartAttempts++;
try {
const spawned = await this.bridge.spawn();
if (!spawned) {
console.error('[RemoteIngressHub] Failed to respawn binary');
return;
}
this.bridge.on('exit', this.handleCrashRecovery);
const config = this.savedConfig;
await this.bridge.sendCommand('startHub', {
tunnelPort: config.tunnelPort ?? 8443,
targetHost: config.targetHost ?? '127.0.0.1',
...(config.tls?.certPem && config.tls?.keyPem
? { tlsCertPem: config.tls.certPem, tlsKeyPem: config.tls.keyPem }
: {}),
});
// Restore allowed edges
if (this.savedEdges.length > 0) {
await this.bridge.sendCommand('updateAllowedEdges', { edges: this.savedEdges });
}
this.started = true;
this.restartAttempts = 0;
this.restartBackoffMs = 1000;
console.log('[RemoteIngressHub] Successfully recovered from crash');
this.emit('crashRecovered');
} catch (err) {
console.error(`[RemoteIngressHub] Crash recovery failed: ${err}`);
}
};
}