v4.8.4

changelog.md

@@ -1,5 +1,46 @@
 # Changelog
 
+## 2026-03-17 - 4.8.4 - fix(remoteingress-core)
+prevent stream stalls by guaranteeing flow-control updates and avoiding bounded per-stream channel overflows
+
+- Replace bounded per-stream data channels with unbounded channels on edge and hub, relying on existing WINDOW_UPDATE flow control to limit bytes in flight
+- Use awaited sends for FRAME_WINDOW_UPDATE and FRAME_WINDOW_UPDATE_BACK so updates are not dropped and streams do not deadlock under backpressure
+- Clean up stream state when channel receivers have already exited instead of closing active streams because a bounded queue filled
+
+## 2026-03-17 - 4.8.3 - fix(protocol,edge)
+optimize tunnel frame handling and zero-copy uploads in edge I/O
+
+- extract hub frame processing into a shared edge handler to remove duplicated tunnel logic
+- add zero-copy frame header encoding and read payloads directly into framed buffers for client-to-hub uploads
+- refactor TunnelIo read/write state to avoid unsafe queue access and reduce buffer churn with incremental parsing
+
+## 2026-03-17 - 4.8.2 - fix(rust-edge)
+refactor tunnel I/O to preserve TLS state and prioritize control frames
+
+- replace split TLS handling with a single-owner TunnelIo to avoid handshake and buffered read corruption
+- prioritize control frames over data frames to prevent WINDOW_UPDATE starvation and flow-control deadlocks
+- improve tunnel reliability with incremental frame parsing, liveness/error events, and corrupt frame header logging
+
+## 2026-03-17 - 4.8.1 - fix(remoteingress-core)
+remove tunnel writer timeouts from edge and hub buffered writes
+
+- Drops the 30 second timeout wrapper around writer.write_all and writer.flush in both edge and hub tunnel writers.
+- Updates error logging to report write failures without referring to stalled writes.
+
+## 2026-03-17 - 4.8.0 - feat(events)
+include disconnect reasons in edge and hub management events
+
+- Add reason fields to tunnelDisconnected and edgeDisconnected events emitted from the Rust core and binary bridge
+- Propagate specific disconnect causes such as EOF, liveness timeout, writer failure, handshake failure, and hub cancellation
+- Update TypeScript edge and hub classes to log and forward disconnect reason data
+- Extend serialization tests to cover the new reason fields
+
+## 2026-03-17 - 4.7.2 - fix(remoteingress-core)
+add tunnel write timeouts and scale initial stream windows by active stream count
+
+- Wrap tunnel frame writes and flushes in a 30-second timeout on both edge and hub to detect stalled writers and trigger faster reconnect or cleanup.
+- Compute each stream's initial send window from the current active stream count instead of using a fixed window to keep total in-flight data within the 32MB budget.
+
 ## 2026-03-17 - 4.7.1 - fix(remoteingress-core)
 improve tunnel failure detection and reconnect handling
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@serve.zone/remoteingress",
-  "version": "4.7.1",
+  "version": "4.8.4",
   "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",

rust/Cargo.lock

@@ -568,7 +568,9 @@ dependencies = [
 name = "remoteingress-protocol"
 version = "2.0.0"
 dependencies = [
+ "log",
  "tokio",
+ "tokio-util",
 ]
 
 [[package]]

rust/crates/remoteingress-bin/src/main.rs

@@ -173,10 +173,10 @@ async fn handle_request(
                                     serde_json::json!({ "edgeId": edge_id, "peerAddr": peer_addr }),
                                 );
                             }
-                            HubEvent::EdgeDisconnected { edge_id } => {
+                            HubEvent::EdgeDisconnected { edge_id, reason } => {
                                 send_event(
                                     "edgeDisconnected",
-                                    serde_json::json!({ "edgeId": edge_id }),
+                                    serde_json::json!({ "edgeId": edge_id, "reason": reason }),
                                 );
                             }
                             HubEvent::StreamOpened {
@@ -295,8 +295,8 @@ async fn handle_request(
                             EdgeEvent::TunnelConnected => {
                                 send_event("tunnelConnected", serde_json::json!({}));
                             }
-                            EdgeEvent::TunnelDisconnected => {
-                                send_event("tunnelDisconnected", serde_json::json!({}));
+                            EdgeEvent::TunnelDisconnected { reason } => {
+                                send_event("tunnelDisconnected", serde_json::json!({ "reason": reason }));
                             }
                             EdgeEvent::PublicIpDiscovered { ip } => {
                                 send_event(

rust/crates/remoteingress-core/src/edge.rs

@@ -2,7 +2,7 @@ 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::io::{AsyncReadExt, AsyncWriteExt};
 use tokio::net::{TcpListener, TcpStream};
 use tokio::sync::{mpsc, Mutex, Notify, RwLock};
 use tokio::task::JoinHandle;
@@ -13,10 +13,20 @@ use serde::{Deserialize, Serialize};
 
 use remoteingress_protocol::*;
 
+type EdgeTlsStream = tokio_rustls::client::TlsStream<TcpStream>;
+
+/// Result of processing a frame (shared with hub.rs pattern).
+#[allow(dead_code)]
+enum EdgeFrameAction {
+    Continue,
+    Disconnect(String),
+}
+
 /// 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>>,
+    /// Unbounded channel to deliver FRAME_DATA_BACK payloads to the hub_to_client task.
+    /// Unbounded because flow control (WINDOW_UPDATE) already limits bytes-in-flight.
+    back_tx: mpsc::UnboundedSender<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>,
@@ -64,7 +74,8 @@ struct ConfigUpdate {
 #[serde(tag = "type")]
 pub enum EdgeEvent {
     TunnelConnected,
-    TunnelDisconnected,
+    #[serde(rename_all = "camelCase")]
+    TunnelDisconnected { reason: String },
     #[serde(rename_all = "camelCase")]
     PublicIpDiscovered { ip: String },
     #[serde(rename_all = "camelCase")]
@@ -236,10 +247,15 @@ async fn edge_main_loop(
         }
 
         *connected.write().await = false;
+        // Extract reason for disconnect event
+        let reason = match &result {
+            EdgeLoopResult::Reconnect(r) => r.clone(),
+            EdgeLoopResult::Shutdown => "shutdown".to_string(),
+        };
         // Only emit disconnect event on actual disconnection, not on failed reconnects.
         // Failed reconnects never reach line 335 (handshake success), so was_connected is false.
         if was_connected {
-            let _ = event_tx.try_send(EdgeEvent::TunnelDisconnected);
+            let _ = event_tx.try_send(EdgeEvent::TunnelDisconnected { reason: reason.clone() });
         }
         active_streams.store(0, Ordering::Relaxed);
         // Reset stream ID counter for next connection cycle
@@ -248,7 +264,7 @@ async fn edge_main_loop(
 
         match result {
             EdgeLoopResult::Shutdown => break,
-            EdgeLoopResult::Reconnect => {
+            EdgeLoopResult::Reconnect(_) => {
                 log::info!("Reconnecting in {}ms...", backoff_ms);
                 tokio::select! {
                     _ = tokio::time::sleep(Duration::from_millis(backoff_ms)) => {}
@@ -263,7 +279,87 @@ async fn edge_main_loop(
 
 enum EdgeLoopResult {
     Shutdown,
-    Reconnect,
+    Reconnect(String),  // reason for disconnection
+}
+
+/// Process a single frame received from the hub side of the tunnel.
+/// Handles FRAME_DATA_BACK, FRAME_WINDOW_UPDATE_BACK, FRAME_CLOSE_BACK, FRAME_CONFIG, FRAME_PING.
+async fn handle_edge_frame(
+    frame: Frame,
+    tunnel_io: &mut remoteingress_protocol::TunnelIo<EdgeTlsStream>,
+    client_writers: &Arc<Mutex<HashMap<u32, EdgeStreamState>>>,
+    listen_ports: &Arc<RwLock<Vec<u16>>>,
+    event_tx: &mpsc::Sender<EdgeEvent>,
+    tunnel_writer_tx: &mpsc::Sender<Vec<u8>>,
+    tunnel_data_tx: &mpsc::Sender<Vec<u8>>,
+    port_listeners: &mut HashMap<u16, JoinHandle<()>>,
+    active_streams: &Arc<AtomicU32>,
+    next_stream_id: &Arc<AtomicU32>,
+    edge_id: &str,
+    connection_token: &CancellationToken,
+    bind_address: &str,
+) -> EdgeFrameAction {
+    match frame.frame_type {
+        FRAME_DATA_BACK => {
+            // Dispatch to per-stream unbounded channel. Flow control (WINDOW_UPDATE)
+            // limits bytes-in-flight, so the channel won't grow unbounded. send() only
+            // fails if the receiver is dropped (hub_to_client task already exited).
+            let mut writers = client_writers.lock().await;
+            if let Some(state) = writers.get(&frame.stream_id) {
+                if state.back_tx.send(frame.payload).is_err() {
+                    // Receiver dropped — hub_to_client task already exited, clean up
+                    writers.remove(&frame.stream_id);
+                }
+            }
+        }
+        FRAME_WINDOW_UPDATE_BACK => {
+            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 => {
+            let mut writers = client_writers.lock().await;
+            writers.remove(&frame.stream_id);
+        }
+        FRAME_CONFIG => {
+            if let Ok(update) = serde_json::from_slice::<ConfigUpdate>(&frame.payload) {
+                log::info!("Config update from hub: ports {:?}", update.listen_ports);
+                *listen_ports.write().await = update.listen_ports.clone();
+                let _ = event_tx.try_send(EdgeEvent::PortsUpdated {
+                    listen_ports: update.listen_ports.clone(),
+                });
+                apply_port_config(
+                    &update.listen_ports,
+                    port_listeners,
+                    tunnel_writer_tx,
+                    tunnel_data_tx,
+                    client_writers,
+                    active_streams,
+                    next_stream_id,
+                    edge_id,
+                    connection_token,
+                    bind_address,
+                );
+            }
+        }
+        FRAME_PING => {
+            // Queue PONG directly — no channel round-trip, guaranteed delivery
+            tunnel_io.queue_ctrl(encode_frame(0, FRAME_PONG, &[]));
+        }
+        _ => {
+            log::warn!("Unexpected frame type {} from hub", frame.frame_type);
+        }
+    }
+    EdgeFrameAction::Continue
 }
 
 async fn connect_to_hub_and_run(
@@ -295,49 +391,59 @@ async fn connect_to_hub_and_run(
         }
         Err(e) => {
             log::error!("Failed to connect to hub at {}: {}", addr, e);
-            return EdgeLoopResult::Reconnect;
+            return EdgeLoopResult::Reconnect(format!("tcp_connect_failed: {}", e));
         }
     };
 
     let server_name = rustls::pki_types::ServerName::try_from(config.hub_host.clone())
         .unwrap_or_else(|_| rustls::pki_types::ServerName::try_from("remoteingress-hub".to_string()).unwrap());
 
-    let tls_stream = match connector.connect(server_name, tcp).await {
+    let mut tls_stream = match connector.connect(server_name, tcp).await {
         Ok(s) => s,
         Err(e) => {
             log::error!("TLS handshake failed: {}", e);
-            return EdgeLoopResult::Reconnect;
+            return EdgeLoopResult::Reconnect(format!("tls_handshake_failed: {}", e));
         }
     };
 
-    let (read_half, mut write_half) = tokio::io::split(tls_stream);
-
-    // Send auth line
+    // Send auth line (we own the whole stream — no split)
     let auth_line = format!("EDGE {} {}\n", config.edge_id, config.secret);
-    if write_half.write_all(auth_line.as_bytes()).await.is_err() {
-        return EdgeLoopResult::Reconnect;
+    if tls_stream.write_all(auth_line.as_bytes()).await.is_err() {
+        return EdgeLoopResult::Reconnect("auth_write_failed".to_string());
+    }
+    if tls_stream.flush().await.is_err() {
+        return EdgeLoopResult::Reconnect("auth_flush_failed".to_string());
     }
 
-    // Read handshake response line from hub (JSON with initial config)
-    let mut buf_reader = BufReader::new(read_half);
-    let mut handshake_line = String::new();
-    match buf_reader.read_line(&mut handshake_line).await {
-        Ok(0) => {
-            log::error!("Hub rejected connection (EOF before handshake)");
-            return EdgeLoopResult::Reconnect;
-        }
-        Ok(_) => {}
-        Err(e) => {
-            log::error!("Failed to read handshake response: {}", e);
-            return EdgeLoopResult::Reconnect;
+    // Read handshake line byte-by-byte (no BufReader — into_inner corrupts TLS state)
+    let mut handshake_bytes = Vec::with_capacity(512);
+    let mut byte = [0u8; 1];
+    loop {
+        match tls_stream.read_exact(&mut byte).await {
+            Ok(_) => {
+                handshake_bytes.push(byte[0]);
+                if byte[0] == b'\n' { break; }
+                if handshake_bytes.len() > 8192 {
+                    return EdgeLoopResult::Reconnect("handshake_too_long".to_string());
+                }
+            }
+            Err(e) if e.kind() == std::io::ErrorKind::UnexpectedEof => {
+                log::error!("Hub rejected connection (EOF before handshake)");
+                return EdgeLoopResult::Reconnect("hub_rejected_eof".to_string());
+            }
+            Err(e) => {
+                log::error!("Failed to read handshake response: {}", e);
+                return EdgeLoopResult::Reconnect(format!("handshake_read_failed: {}", e));
+            }
         }
     }
+    let handshake_line = String::from_utf8_lossy(&handshake_bytes);
 
     let handshake: HandshakeConfig = match serde_json::from_str(handshake_line.trim()) {
         Ok(h) => h,
         Err(e) => {
             log::error!("Invalid handshake response: {}", e);
-            return EdgeLoopResult::Reconnect;
+            return EdgeLoopResult::Reconnect(format!("handshake_invalid: {}", e));
         }
     };
 
@@ -388,52 +494,13 @@ async fn connect_to_hub_and_run(
     let client_writers: Arc<Mutex<HashMap<u32, EdgeStreamState>>> =
         Arc::new(Mutex::new(HashMap::new()));
 
-    // QoS dual-channel tunnel writer: control frames (PONG/WINDOW_UPDATE/CLOSE/OPEN)
-    // have priority over data frames (DATA). Prevents PING starvation under load.
+    // QoS dual-channel: ctrl frames have priority over data frames.
+    // Stream handlers send through these channels → TunnelIo drains them.
     let (tunnel_ctrl_tx, mut tunnel_ctrl_rx) = mpsc::channel::<Vec<u8>>(256);
     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();
-    // Oneshot to signal the reader loop when the writer dies from a write error.
-    // This avoids the 45s liveness timeout delay when the tunnel is already dead.
-    let (writer_dead_tx, mut writer_dead_rx) = tokio::sync::oneshot::channel::<()>();
-    let tunnel_writer_handle = tokio::spawn(async move {
-        // BufWriter coalesces small writes (frame headers, control frames) into fewer
-        // TLS records and syscalls. Flushed after each frame to avoid holding data.
-        let mut writer = tokio::io::BufWriter::with_capacity(65536, write_half);
-        let mut write_error = false;
-        loop {
-            tokio::select! {
-                biased; // control frames always take priority over data
-                ctrl = tunnel_ctrl_rx.recv() => {
-                    match ctrl {
-                        Some(frame_data) => {
-                            if writer.write_all(&frame_data).await.is_err() { write_error = true; break; }
-                            if writer.flush().await.is_err() { write_error = true; break; }
-                        }
-                        None => break,
-                    }
-                }
-                data = tunnel_data_rx.recv() => {
-                    match data {
-                        Some(frame_data) => {
-                            if writer.write_all(&frame_data).await.is_err() { write_error = true; break; }
-                            if writer.flush().await.is_err() { write_error = true; break; }
-                        }
-                        None => break,
-                    }
-                }
-                _ = tw_token.cancelled() => break,
-            }
-        }
-        if write_error {
-            log::error!("Tunnel writer failed, signalling reader for fast reconnect");
-            let _ = writer_dead_tx.send(());
-        }
-    });
 
-    // Start TCP listeners for initial ports (hot-reloadable)
+    // Start TCP listeners for initial ports
     let mut port_listeners: HashMap<u16, JoinHandle<()>> = HashMap::new();
     let bind_address = config.bind_address.as_deref().unwrap_or("0.0.0.0");
     apply_port_config(
@@ -449,122 +516,80 @@ async fn connect_to_hub_and_run(
         bind_address,
     );
 
-    // Heartbeat: liveness timeout detects silent hub failures
+    // Single-owner I/O engine — no tokio::io::split, no mutex
+    let mut tunnel_io = remoteingress_protocol::TunnelIo::new(tls_stream, Vec::new());
+
     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 {
-        tokio::select! {
-            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);
+    let result = 'io_loop: loop {
+        // Drain any buffered frames
+        loop {
+            let frame = match tunnel_io.try_parse_frame() {
+                Some(Ok(f)) => f,
+                Some(Err(e)) => {
+                    log::error!("Hub frame error: {}", e);
+                    break 'io_loop EdgeLoopResult::Reconnect(format!("hub_frame_error: {}", e));
+                }
+                None => break,
+            };
+            last_activity = Instant::now();
+            liveness_deadline.as_mut().reset(last_activity + liveness_timeout_dur);
+            if let EdgeFrameAction::Disconnect(reason) = handle_edge_frame(
+                frame, &mut tunnel_io, &client_writers, listen_ports, event_tx,
+                &tunnel_writer_tx, &tunnel_data_tx, &mut port_listeners,
+                active_streams, next_stream_id, &config.edge_id, connection_token, bind_address,
+            ).await {
+                break 'io_loop EdgeLoopResult::Reconnect(reason);
+            }
+        }
 
-                        match frame.frame_type {
-                            FRAME_DATA_BACK => {
-                                // 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 => {
-                                let mut writers = client_writers.lock().await;
-                                writers.remove(&frame.stream_id);
-                            }
-                            FRAME_CONFIG => {
-                                if let Ok(update) = serde_json::from_slice::<ConfigUpdate>(&frame.payload) {
-                                    log::info!("Config update from hub: ports {:?}", update.listen_ports);
-                                    *listen_ports.write().await = update.listen_ports.clone();
-                                    let _ = event_tx.try_send(EdgeEvent::PortsUpdated {
-                                        listen_ports: update.listen_ports.clone(),
-                                    });
-                                    apply_port_config(
-                                        &update.listen_ports,
-                                        &mut port_listeners,
-                                        &tunnel_writer_tx,
-                                        &tunnel_data_tx,
-                                        &client_writers,
-                                        active_streams,
-                                        next_stream_id,
-                                        &config.edge_id,
-                                        connection_token,
-                                        bind_address,
-                                    );
-                                }
-                            }
-                            FRAME_PING => {
-                                let pong_frame = encode_frame(0, FRAME_PONG, &[]);
-                                if tunnel_writer_tx.try_send(pong_frame).is_err() {
-                                    // Control channel full (WINDOW_UPDATE burst from many streams).
-                                    // DON'T disconnect — the 45s liveness timeout gives margin
-                                    // for the channel to drain and the next PONG to succeed.
-                                    log::warn!("PONG send failed, control channel full — skipping this cycle");
-                                }
-                                log::trace!("Received PING from hub, sent PONG");
-                            }
-                            _ => {
-                                log::warn!("Unexpected frame type {} from hub", frame.frame_type);
-                            }
-                        }
-                    }
-                    Ok(None) => {
-                        log::info!("Hub disconnected (EOF)");
-                        break EdgeLoopResult::Reconnect;
-                    }
-                    Err(e) => {
-                        log::error!("Hub frame error: {}", e);
-                        break EdgeLoopResult::Reconnect;
-                    }
+        // Poll I/O: write(ctrl→data), flush, read, channels, timers
+        let event = std::future::poll_fn(|cx| {
+            tunnel_io.poll_step(cx, &mut tunnel_ctrl_rx, &mut tunnel_data_rx, &mut liveness_deadline, connection_token)
+        }).await;
+
+        match event {
+            remoteingress_protocol::TunnelEvent::Frame(frame) => {
+                last_activity = Instant::now();
+                liveness_deadline.as_mut().reset(last_activity + liveness_timeout_dur);
+                if let EdgeFrameAction::Disconnect(reason) = handle_edge_frame(
+                    frame, &mut tunnel_io, &client_writers, listen_ports, event_tx,
+                    &tunnel_writer_tx, &tunnel_data_tx, &mut port_listeners,
+                    active_streams, next_stream_id, &config.edge_id, connection_token, bind_address,
+                ).await {
+                    break EdgeLoopResult::Reconnect(reason);
                 }
             }
-            _ = &mut liveness_deadline => {
-                log::warn!("Hub liveness timeout (no frames for {}s), reconnecting",
-                    liveness_timeout_dur.as_secs());
-                break EdgeLoopResult::Reconnect;
+            remoteingress_protocol::TunnelEvent::Eof => {
+                log::info!("Hub disconnected (EOF)");
+                break EdgeLoopResult::Reconnect("hub_eof".to_string());
             }
-            _ = &mut writer_dead_rx => {
-                log::error!("Tunnel writer died, reconnecting immediately");
-                break EdgeLoopResult::Reconnect;
+            remoteingress_protocol::TunnelEvent::ReadError(e) => {
+                log::error!("Hub frame read error: {}", e);
+                break EdgeLoopResult::Reconnect(format!("hub_frame_error: {}", e));
             }
-            _ = connection_token.cancelled() => {
-                log::info!("Connection cancelled");
-                break EdgeLoopResult::Shutdown;
+            remoteingress_protocol::TunnelEvent::WriteError(e) => {
+                log::error!("Tunnel write error: {}", e);
+                break EdgeLoopResult::Reconnect(format!("tunnel_write_error: {}", e));
             }
-            _ = shutdown_rx.recv() => {
+            remoteingress_protocol::TunnelEvent::LivenessTimeout => {
+                log::warn!("Hub liveness timeout (no frames for {}s), reconnecting", liveness_timeout_dur.as_secs());
+                break EdgeLoopResult::Reconnect("liveness_timeout".to_string());
+            }
+            remoteingress_protocol::TunnelEvent::Cancelled => {
+                if shutdown_rx.try_recv().is_ok() {
+                    break EdgeLoopResult::Shutdown;
+                }
                 break EdgeLoopResult::Shutdown;
             }
         }
     };
 
-    // Cancel connection token to propagate to all child tasks BEFORE aborting
+    // Cleanup
     connection_token.cancel();
     stun_handle.abort();
-    tunnel_writer_handle.abort();
     for (_, h) in port_listeners.drain() {
         h.abort();
     }
@@ -710,9 +735,16 @@ async fn handle_client_connection(
         return;
     }
 
-    // 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));
+    // Per-stream unbounded back-channel. Flow control (WINDOW_UPDATE) limits
+    // bytes-in-flight, so this won't grow unbounded. Unbounded avoids killing
+    // streams due to channel overflow — backpressure slows streams, never kills them.
+    let (back_tx, mut back_rx) = mpsc::unbounded_channel::<Vec<u8>>();
+    // Adaptive initial window: scale with current stream count to keep total in-flight
+    // data within the 32MB budget. Prevents burst flooding when many streams open.
+    let initial_window = remoteingress_protocol::compute_window_for_stream_count(
+        active_streams.load(Ordering::Relaxed),
+    );
+    let send_window = Arc::new(AtomicU32::new(initial_window));
     let window_notify = Arc::new(Notify::new());
     {
         let mut writers = client_writers.lock().await;
@@ -753,10 +785,16 @@ async fn handle_client_connection(
                             if consumed_since_update >= threshold {
                                 let increment = consumed_since_update.min(adaptive_window);
                                 let frame = encode_window_update(stream_id, FRAME_WINDOW_UPDATE, increment);
-                                if wu_tx.try_send(frame).is_ok() {
-                                    consumed_since_update -= increment;
+                                // Use send().await for guaranteed delivery — dropping WINDOW_UPDATEs
+                                // causes permanent flow stalls. Safe: runs in per-stream task, not main loop.
+                                tokio::select! {
+                                    result = wu_tx.send(frame) => {
+                                        if result.is_ok() {
+                                            consumed_since_update -= increment;
+                                        }
+                                    }
+                                    _ = hub_to_client_token.cancelled() => break,
                                 }
-                                // If try_send fails, keep accumulating — retry on next threshold
                             }
                         }
                         None => break,
@@ -768,20 +806,26 @@ async fn handle_client_connection(
         // 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 _ = wu_tx.send(frame).await;
         }
         let _ = client_write.shutdown().await;
     });
 
-    // Task: client -> hub (upload direction) with per-stream flow control
-    let mut buf = vec![0u8; 32768];
+    // Task: client -> hub (upload direction) with per-stream flow control.
+    // Zero-copy: read payload directly after the header, then prepend header.
+    let mut buf = vec![0u8; FRAME_HEADER_SIZE + 32768];
     loop {
-        // Wait for send window to have capacity (with stall timeout)
+        // Wait for send window to have capacity (with stall timeout).
+        // Safe pattern: register notified BEFORE checking the condition
+        // to avoid missing a notify_one that fires between load and select.
         loop {
+            let notified = window_notify.notified();
+            tokio::pin!(notified);
+            notified.as_mut().enable();
             let w = send_window.load(Ordering::Acquire);
             if w > 0 { break; }
             tokio::select! {
-                _ = window_notify.notified() => continue,
+                _ = notified => continue,
                 _ = client_token.cancelled() => break,
                 _ = tokio::time::sleep(Duration::from_secs(120)) => {
                     log::warn!("Stream {} upload stalled (window empty for 120s)", stream_id);
@@ -804,15 +848,16 @@ async fn handle_client_connection(
         let adaptive_cap = remoteingress_protocol::compute_window_for_stream_count(
             active_streams.load(Ordering::Relaxed),
         ) as usize;
-        let max_read = w.min(buf.len()).min(adaptive_cap);
+        let max_read = w.min(32768).min(adaptive_cap);
 
         tokio::select! {
-            read_result = client_read.read(&mut buf[..max_read]) => {
+            read_result = client_read.read(&mut buf[FRAME_HEADER_SIZE..FRAME_HEADER_SIZE + 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]);
+                        encode_frame_header(&mut buf, stream_id, FRAME_DATA, n);
+                        let data_frame = buf[..FRAME_HEADER_SIZE + n].to_vec();
                         if tunnel_data_tx.send(data_frame).await.is_err() {
                             log::warn!("Stream {} data channel closed, closing", stream_id);
                             break;
@@ -951,9 +996,10 @@ mod tests {
 
     #[test]
     fn test_edge_event_tunnel_disconnected() {
-        let event = EdgeEvent::TunnelDisconnected;
+        let event = EdgeEvent::TunnelDisconnected { reason: "hub_eof".to_string() };
         let json = serde_json::to_value(&event).unwrap();
         assert_eq!(json["type"], "tunnelDisconnected");
+        assert_eq!(json["reason"], "hub_eof");
     }
 
     #[test]

rust/crates/remoteingress-core/src/hub.rs

@@ -2,7 +2,7 @@ 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::io::{AsyncReadExt, AsyncWriteExt};
 use tokio::net::{TcpListener, TcpStream};
 use tokio::sync::{mpsc, Mutex, Notify, RwLock, Semaphore};
 use tokio::time::{interval, sleep_until, Instant};
@@ -12,10 +12,21 @@ use serde::{Deserialize, Serialize};
 
 use remoteingress_protocol::*;
 
+type HubTlsStream = tokio_rustls::server::TlsStream<TcpStream>;
+
+/// Result of processing a frame.
+#[allow(dead_code)]
+enum FrameAction {
+    Continue,
+    Disconnect(String),
+}
+
 /// 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>>,
+    /// Unbounded channel to deliver FRAME_DATA payloads to the upstream writer task.
+    /// Unbounded because flow control (WINDOW_UPDATE) already limits bytes-in-flight.
+    /// A bounded channel would kill streams instead of applying backpressure.
+    data_tx: mpsc::UnboundedSender<Vec<u8>>,
     /// Cancellation token for this stream.
     cancel_token: CancellationToken,
     /// Send window for FRAME_DATA_BACK (download direction).
@@ -92,7 +103,7 @@ pub enum HubEvent {
     #[serde(rename_all = "camelCase")]
     EdgeConnected { edge_id: String, peer_addr: String },
     #[serde(rename_all = "camelCase")]
-    EdgeDisconnected { edge_id: String },
+    EdgeDisconnected { edge_id: String, reason: String },
     #[serde(rename_all = "camelCase")]
     StreamOpened { edge_id: String, stream_id: u32 },
     #[serde(rename_all = "camelCase")]
@@ -123,7 +134,7 @@ pub struct TunnelHub {
 struct ConnectedEdgeInfo {
     connected_at: u64,
     peer_addr: String,
-    active_streams: Arc<Mutex<HashMap<u32, HubStreamState>>>,
+    edge_stream_count: Arc<AtomicU32>,
     config_tx: mpsc::Sender<EdgeConfigUpdate>,
     #[allow(dead_code)] // kept alive for Drop — cancels child tokens when edge is removed
     cancel_token: CancellationToken,
@@ -189,11 +200,10 @@ impl TunnelHub {
 
         let mut connected = Vec::new();
         for (id, info) in edges.iter() {
-            let streams = info.active_streams.lock().await;
             connected.push(ConnectedEdgeStatus {
                 edge_id: id.clone(),
                 connected_at: info.connected_at,
-                active_streams: streams.len(),
+                active_streams: info.edge_stream_count.load(Ordering::Relaxed) as usize,
                 peer_addr: info.peer_addr.clone(),
             });
         }
@@ -287,6 +297,290 @@ impl Drop for TunnelHub {
 /// Maximum concurrent streams per edge connection.
 const MAX_STREAMS_PER_EDGE: usize = 1024;
 
+/// Process a single frame received from the edge side of the tunnel.
+/// Handles FRAME_OPEN, FRAME_DATA, FRAME_WINDOW_UPDATE, FRAME_CLOSE, and FRAME_PONG.
+async fn handle_hub_frame(
+    frame: Frame,
+    tunnel_io: &mut remoteingress_protocol::TunnelIo<HubTlsStream>,
+    streams: &mut HashMap<u32, HubStreamState>,
+    stream_semaphore: &Arc<Semaphore>,
+    edge_stream_count: &Arc<AtomicU32>,
+    edge_id: &str,
+    event_tx: &mpsc::Sender<HubEvent>,
+    ctrl_tx: &mpsc::Sender<Vec<u8>>,
+    data_tx: &mpsc::Sender<Vec<u8>>,
+    target_host: &str,
+    edge_token: &CancellationToken,
+    cleanup_tx: &mpsc::Sender<u32>,
+) -> FrameAction {
+    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, &[]);
+                    tunnel_io.queue_ctrl(close_frame);
+                    return FrameAction::Continue;
+                }
+            };
+
+            // Payload is PROXY v1 header line
+            let proxy_header = String::from_utf8_lossy(&frame.payload).to_string();
+
+            // Parse destination port from PROXY header
+            let dest_port = parse_dest_port_from_proxy(&proxy_header).unwrap_or(443);
+
+            let stream_id = frame.stream_id;
+            let cleanup = cleanup_tx.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.to_string();
+            let stream_token = edge_token.child_token();
+
+            let _ = event_tx.try_send(HubEvent::StreamOpened {
+                edge_id: edge_id.to_string(),
+                stream_id,
+            });
+
+            // Create channel for data from edge to this stream
+            let (stream_data_tx, mut stream_data_rx) = mpsc::unbounded_channel::<Vec<u8>>();
+            // Adaptive initial window: scale with current stream count
+            // to keep total in-flight data within the 32MB budget.
+            let initial_window = compute_window_for_stream_count(
+                edge_stream_count.load(Ordering::Relaxed),
+            );
+            let send_window = Arc::new(AtomicU32::new(initial_window));
+            let window_notify = Arc::new(Notify::new());
+            streams.insert(stream_id, HubStreamState {
+                data_tx: stream_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
+            let stream_counter = Arc::clone(edge_stream_count);
+            tokio::spawn(async move {
+                let _permit = permit; // hold semaphore permit until stream completes
+                stream_counter.fetch_add(1, Ordering::Relaxed);
+
+                let result = async {
+                    // 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.set_nodelay(true)?;
+                    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 stream_counter_w = Arc::clone(&stream_counter);
+                    let writer_for_edge_data = tokio::spawn(async move {
+                        let mut consumed_since_update: u32 = 0;
+                        loop {
+                            tokio::select! {
+                                data = stream_data_rx.recv() => {
+                                    match data {
+                                        Some(data) => {
+                                            let len = data.len() as u32;
+                                            // Check cancellation alongside the write so we respond
+                                            // promptly to FRAME_CLOSE instead of blocking up to 60s.
+                                            let write_result = tokio::select! {
+                                                r = tokio::time::timeout(
+                                                    Duration::from_secs(60),
+                                                    up_write.write_all(&data),
+                                                ) => r,
+                                                _ = writer_token.cancelled() => break,
+                                            };
+                                            match write_result {
+                                                Ok(Ok(())) => {}
+                                                Ok(Err(_)) => break,
+                                                Err(_) => {
+                                                    log::warn!("Stream {} write to upstream timed out (60s)", stream_id);
+                                                    break;
+                                                }
+                                            }
+                                            // Track consumption for adaptive flow control.
+                                            // Increment capped to adaptive window to limit per-stream in-flight data.
+                                            consumed_since_update += len;
+                                            let adaptive_window = remoteingress_protocol::compute_window_for_stream_count(
+                                                stream_counter_w.load(Ordering::Relaxed),
+                                            );
+                                            let threshold = adaptive_window / 2;
+                                            if consumed_since_update >= threshold {
+                                                let increment = consumed_since_update.min(adaptive_window);
+                                                let frame = encode_window_update(stream_id, FRAME_WINDOW_UPDATE_BACK, increment);
+                                                // Use send().await for guaranteed delivery — dropping WINDOW_UPDATEs
+                                                // causes permanent flow stalls. Safe: runs in per-stream task, not main loop.
+                                                tokio::select! {
+                                                    result = wub_tx.send(frame) => {
+                                                        if result.is_ok() {
+                                                            consumed_since_update -= increment;
+                                                        }
+                                                    }
+                                                    _ = writer_token.cancelled() => break,
+                                                }
+                                            }
+                                        }
+                                        None => break,
+                                    }
+                                }
+                                _ = 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.send(frame).await;
+                        }
+                        let _ = up_write.shutdown().await;
+                    });
+
+                    // Forward data from SmartProxy back to edge via writer channel
+                    // with per-stream flow control (check send_window before reading).
+                    // Zero-copy: read payload directly after the header, then prepend header.
+                    let mut buf = vec![0u8; FRAME_HEADER_SIZE + 32768];
+                    loop {
+                        // Wait for send window to have capacity (with stall timeout).
+                        // Safe pattern: register notified BEFORE checking the condition
+                        // to avoid missing a notify_one that fires between load and select.
+                        loop {
+                            let notified = window_notify.notified();
+                            tokio::pin!(notified);
+                            notified.as_mut().enable();
+                            let w = send_window.load(Ordering::Acquire);
+                            if w > 0 { break; }
+                            tokio::select! {
+                                _ = notified => continue,
+                                _ = stream_token.cancelled() => break,
+                                _ = tokio::time::sleep(Duration::from_secs(120)) => {
+                                    log::warn!("Stream {} download stalled (window empty for 120s)", stream_id);
+                                    break;
+                                }
+                            }
+                        }
+                        if stream_token.is_cancelled() { break; }
+
+                        // Limit read size to available window.
+                        // IMPORTANT: if window is 0 (stall timeout fired), we must NOT
+                        // read into an empty buffer — read(&mut buf[..0]) returns Ok(0)
+                        // which would be falsely interpreted as EOF.
+                        let w = send_window.load(Ordering::Acquire) as usize;
+                        if w == 0 {
+                            log::warn!("Stream {} download: window still 0 after stall timeout, closing", stream_id);
+                            break;
+                        }
+                        // Adaptive: cap read to current per-stream target window
+                        let adaptive_cap = remoteingress_protocol::compute_window_for_stream_count(
+                            stream_counter.load(Ordering::Relaxed),
+                        ) as usize;
+                        let max_read = w.min(32768).min(adaptive_cap);
+
+                        tokio::select! {
+                            read_result = up_read.read(&mut buf[FRAME_HEADER_SIZE..FRAME_HEADER_SIZE + max_read]) => {
+                                match read_result {
+                                    Ok(0) => break,
+                                    Ok(n) => {
+                                        send_window.fetch_sub(n as u32, Ordering::Release);
+                                        encode_frame_header(&mut buf, stream_id, FRAME_DATA_BACK, n);
+                                        let frame = buf[..FRAME_HEADER_SIZE + n].to_vec();
+                                        if data_writer_tx.send(frame).await.is_err() {
+                                            log::warn!("Stream {} data channel closed, closing", stream_id);
+                                            break;
+                                        }
+                                    }
+                                    Err(_) => break,
+                                }
+                            }
+                            _ = stream_token.cancelled() => break,
+                        }
+                    }
+
+                    // Send CLOSE_BACK via DATA channel (must arrive AFTER last DATA_BACK).
+                    // Use send().await to guarantee delivery (try_send silently drops if full).
+                    if !stream_token.is_cancelled() {
+                        let close_frame = encode_frame(stream_id, FRAME_CLOSE_BACK, &[]);
+                        let _ = data_writer_tx.send(close_frame).await;
+                    }
+
+                    writer_for_edge_data.abort();
+                    Ok::<(), Box<dyn std::error::Error + Send + Sync>>(())
+                }
+                .await;
+
+                if let Err(e) = result {
+                    log::error!("Stream {} error: {}", stream_id, e);
+                    // Send CLOSE_BACK via DATA channel on error (must arrive after any DATA_BACK).
+                    // Use send().await to guarantee delivery.
+                    if !stream_token.is_cancelled() {
+                        let close_frame = encode_frame(stream_id, FRAME_CLOSE_BACK, &[]);
+                        let _ = data_writer_tx.send(close_frame).await;
+                    }
+                }
+
+                // Signal main loop to remove stream from the map
+                let _ = cleanup.send(stream_id).await;
+                stream_counter.fetch_sub(1, Ordering::Relaxed);
+            });
+        }
+        FRAME_DATA => {
+            // Dispatch to per-stream unbounded channel. Flow control (WINDOW_UPDATE)
+            // limits bytes-in-flight, so the channel won't grow unbounded. send() only
+            // fails if the receiver is dropped (stream handler already exited).
+            if let Some(state) = streams.get(&frame.stream_id) {
+                if state.data_tx.send(frame.payload).is_err() {
+                    // Receiver dropped — stream handler already exited, clean up
+                    streams.remove(&frame.stream_id);
+                }
+            }
+        }
+        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 {
+                    if let Some(state) = streams.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 => {
+            if let Some(state) = streams.remove(&frame.stream_id) {
+                state.cancel_token.cancel();
+                let _ = event_tx.try_send(HubEvent::StreamClosed {
+                    edge_id: edge_id.to_string(),
+                    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);
+        }
+    }
+    FrameAction::Continue
+}
+
 /// Handle a single edge connection: authenticate, then enter frame loop.
 async fn handle_edge_connection(
     stream: TcpStream,
@@ -307,13 +601,24 @@ async fn handle_edge_connection(
     #[cfg(target_os = "linux")]
     let ka = ka.with_interval(Duration::from_secs(10));
     let _ = socket2::SockRef::from(&stream).set_tcp_keepalive(&ka);
-    let tls_stream = acceptor.accept(stream).await?;
-    let (read_half, mut write_half) = tokio::io::split(tls_stream);
-    let mut buf_reader = BufReader::new(read_half);
+    let mut tls_stream = acceptor.accept(stream).await?;
 
-    // Read auth line: "EDGE <edgeId> <secret>\n"
-    let mut auth_line = String::new();
-    buf_reader.read_line(&mut auth_line).await?;
+    // Byte-by-byte auth line reading (no BufReader).
+    // Auth line: "EDGE <edgeId> <secret>\n"
+    let mut auth_buf = Vec::with_capacity(512);
+    loop {
+        let mut byte = [0u8; 1];
+        tls_stream.read_exact(&mut byte).await?;
+        if byte[0] == b'\n' {
+            break;
+        }
+        auth_buf.push(byte[0]);
+        if auth_buf.len() > 4096 {
+            return Err("auth line too long".into());
+        }
+    }
+    let auth_line = String::from_utf8(auth_buf)
+        .map_err(|_| "auth line not valid UTF-8")?;
     let auth_line = auth_line.trim();
 
     let parts: Vec<&str> = auth_line.splitn(3, ' ').collect();
@@ -353,11 +658,15 @@ async fn handle_edge_connection(
     };
     let mut handshake_json = serde_json::to_string(&handshake)?;
     handshake_json.push('\n');
-    write_half.write_all(handshake_json.as_bytes()).await?;
+    tls_stream.write_all(handshake_json.as_bytes()).await?;
+    tls_stream.flush().await?;
 
     // Track this edge
-    let streams: Arc<Mutex<HashMap<u32, HubStreamState>>> =
-        Arc::new(Mutex::new(HashMap::new()));
+    let mut streams: HashMap<u32, HubStreamState> = HashMap::new();
+    // Per-edge active stream counter for adaptive flow control
+    let edge_stream_count = Arc::new(AtomicU32::new(0));
+    // Cleanup channel: spawned stream tasks send stream_id here when done
+    let (cleanup_tx, mut cleanup_rx) = mpsc::channel::<u32>(256);
     let now = std::time::SystemTime::now()
         .duration_since(std::time::UNIX_EPOCH)
         .unwrap_or_default()
@@ -373,61 +682,20 @@ async fn handle_edge_connection(
             ConnectedEdgeInfo {
                 connected_at: now,
                 peer_addr,
-                active_streams: streams.clone(),
+                edge_stream_count: edge_stream_count.clone(),
                 config_tx,
                 cancel_token: edge_token.clone(),
             },
         );
     }
 
-    // Per-edge active stream counter for adaptive flow control
-    let edge_stream_count = Arc::new(AtomicU32::new(0));
-
-    // 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.
+    // QoS dual-channel: ctrl frames have priority over data frames.
+    // Stream handlers send through these channels -> TunnelIo drains them.
     let (ctrl_tx, mut ctrl_rx) = mpsc::channel::<Vec<u8>>(256);
     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_dead_tx, mut writer_dead_rx) = tokio::sync::oneshot::channel::<()>();
-    let writer_handle = tokio::spawn(async move {
-        // BufWriter coalesces small writes (frame headers, control frames) into fewer
-        // TLS records and syscalls. Flushed after each frame to avoid holding data.
-        let mut writer = tokio::io::BufWriter::with_capacity(65536, write_half);
-        let mut write_error = false;
-        loop {
-            tokio::select! {
-                biased; // control frames always take priority over data
-                ctrl = ctrl_rx.recv() => {
-                    match ctrl {
-                        Some(frame_data) => {
-                            if writer.write_all(&frame_data).await.is_err() { write_error = true; break; }
-                            if writer.flush().await.is_err() { write_error = true; break; }
-                        }
-                        None => break,
-                    }
-                }
-                data = data_rx.recv() => {
-                    match data {
-                        Some(frame_data) => {
-                            if writer.write_all(&frame_data).await.is_err() { write_error = true; break; }
-                            if writer.flush().await.is_err() { write_error = true; break; }
-                        }
-                        None => break,
-                    }
-                }
-                _ = writer_token.cancelled() => break,
-            }
-        }
-        if write_error {
-            log::error!("Tunnel writer to edge failed, signalling reader for fast cleanup");
-            let _ = writer_dead_tx.send(());
-        }
-    });
 
     // Spawn task to forward config updates as FRAME_CONFIG frames
-    let config_writer_tx = frame_writer_tx.clone();
+    let config_writer_tx = ctrl_tx.clone();
     let config_edge_id = edge_id.clone();
     let config_token = edge_token.clone();
     let config_handle = tokio::spawn(async move {
@@ -464,315 +732,91 @@ async fn handle_edge_connection(
     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);
+    // Single-owner I/O engine — no tokio::io::split, no mutex
+    let mut tunnel_io = remoteingress_protocol::TunnelIo::new(tls_stream, Vec::new());
 
-    loop {
-        tokio::select! {
-            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);
+    let mut disconnect_reason = "unknown".to_string();
 
-                        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;
-                                    }
-                                };
+    'hub_loop: loop {
+        // Drain completed stream cleanups from spawned tasks
+        while let Ok(stream_id) = cleanup_rx.try_recv() {
+            if streams.remove(&stream_id).is_some() {
+                let _ = event_tx.try_send(HubEvent::StreamClosed {
+                    edge_id: edge_id.clone(),
+                    stream_id,
+                });
+            }
+        }
 
-                                // Payload is PROXY v1 header line
-                                let proxy_header = String::from_utf8_lossy(&frame.payload).to_string();
+        // Drain any buffered frames
+        loop {
+            let frame = match tunnel_io.try_parse_frame() {
+                Some(Ok(f)) => f,
+                Some(Err(e)) => {
+                    log::error!("Edge {} frame error: {}", edge_id, e);
+                    disconnect_reason = format!("edge_frame_error: {}", e);
+                    break 'hub_loop;
+                }
+                None => break,
+            };
+            last_activity = Instant::now();
+            liveness_deadline.as_mut().reset(last_activity + liveness_timeout_dur);
+            if let FrameAction::Disconnect(reason) = handle_hub_frame(
+                frame, &mut tunnel_io, &mut streams, &stream_semaphore, &edge_stream_count,
+                &edge_id, &event_tx, &ctrl_tx, &data_tx, &target_host, &edge_token,
+                &cleanup_tx,
+            ).await {
+                disconnect_reason = reason;
+                break 'hub_loop;
+            }
+        }
 
-                                // Parse destination port from PROXY header
-                                let dest_port = parse_dest_port_from_proxy(&proxy_header).unwrap_or(443);
+        // Poll I/O: write(ctrl->data), flush, read, channels, timers
+        let event = std::future::poll_fn(|cx| {
+            // Queue PING if ticker fires
+            if ping_ticker.poll_tick(cx).is_ready() {
+                tunnel_io.queue_ctrl(encode_frame(0, FRAME_PING, &[]));
+            }
+            tunnel_io.poll_step(cx, &mut ctrl_rx, &mut data_rx, &mut liveness_deadline, &edge_token)
+        }).await;
 
-                                let stream_id = frame.stream_id;
-                                let edge_id_clone = edge_id.clone();
-                                let event_tx_clone = event_tx.clone();
-                                let streams_clone = streams.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();
-
-                                let _ = event_tx.try_send(HubEvent::StreamOpened {
-                                    edge_id: edge_id.clone(),
-                                    stream_id,
-                                });
-
-                                // 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, 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
-                                let stream_counter = Arc::clone(&edge_stream_count);
-                                tokio::spawn(async move {
-                                    let _permit = permit; // hold semaphore permit until stream completes
-                                    stream_counter.fetch_add(1, Ordering::Relaxed);
-
-                                    let result = async {
-                                        // 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.set_nodelay(true)?;
-                                        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 stream_counter_w = Arc::clone(&stream_counter);
-                                        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;
-                                                                // Check cancellation alongside the write so we respond
-                                                                // promptly to FRAME_CLOSE instead of blocking up to 60s.
-                                                                let write_result = tokio::select! {
-                                                                    r = tokio::time::timeout(
-                                                                        Duration::from_secs(60),
-                                                                        up_write.write_all(&data),
-                                                                    ) => r,
-                                                                    _ = writer_token.cancelled() => break,
-                                                                };
-                                                                match write_result {
-                                                                    Ok(Ok(())) => {}
-                                                                    Ok(Err(_)) => break,
-                                                                    Err(_) => {
-                                                                        log::warn!("Stream {} write to upstream timed out (60s)", stream_id);
-                                                                        break;
-                                                                    }
-                                                                }
-                                                                // Track consumption for adaptive flow control.
-                                                                // Increment capped to adaptive window to limit per-stream in-flight data.
-                                                                consumed_since_update += len;
-                                                                let adaptive_window = remoteingress_protocol::compute_window_for_stream_count(
-                                                                    stream_counter_w.load(Ordering::Relaxed),
-                                                                );
-                                                                let threshold = adaptive_window / 2;
-                                                                if consumed_since_update >= threshold {
-                                                                    let increment = consumed_since_update.min(adaptive_window);
-                                                                    let frame = encode_window_update(stream_id, FRAME_WINDOW_UPDATE_BACK, increment);
-                                                                    if wub_tx.try_send(frame).is_ok() {
-                                                                        consumed_since_update -= increment;
-                                                                    }
-                                                                    // If try_send fails, keep accumulating — retry on next threshold
-                                                                }
-                                                            }
-                                                            None => break,
-                                                        }
-                                                    }
-                                                    _ = 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 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 (with stall timeout)
-                                            loop {
-                                                let w = send_window.load(Ordering::Acquire);
-                                                if w > 0 { break; }
-                                                tokio::select! {
-                                                    _ = window_notify.notified() => continue,
-                                                    _ = stream_token.cancelled() => break,
-                                                    _ = tokio::time::sleep(Duration::from_secs(120)) => {
-                                                        log::warn!("Stream {} download stalled (window empty for 120s)", stream_id);
-                                                        break;
-                                                    }
-                                                }
-                                            }
-                                            if stream_token.is_cancelled() { break; }
-
-                                            // Limit read size to available window.
-                                            // IMPORTANT: if window is 0 (stall timeout fired), we must NOT
-                                            // read into an empty buffer — read(&mut buf[..0]) returns Ok(0)
-                                            // which would be falsely interpreted as EOF.
-                                            let w = send_window.load(Ordering::Acquire) as usize;
-                                            if w == 0 {
-                                                log::warn!("Stream {} download: window still 0 after stall timeout, closing", stream_id);
-                                                break;
-                                            }
-                                            // Adaptive: cap read to current per-stream target window
-                                            let adaptive_cap = remoteingress_protocol::compute_window_for_stream_count(
-                                                stream_counter.load(Ordering::Relaxed),
-                                            ) as usize;
-                                            let max_read = w.min(buf.len()).min(adaptive_cap);
-
-                                            tokio::select! {
-                                                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]);
-                                                            if data_writer_tx.send(frame).await.is_err() {
-                                                                log::warn!("Stream {} data channel closed, closing", stream_id);
-                                                                break;
-                                                            }
-                                                        }
-                                                        Err(_) => break,
-                                                    }
-                                                }
-                                                _ = stream_token.cancelled() => break,
-                                            }
-                                        }
-
-                                        // Send CLOSE_BACK via DATA channel (must arrive AFTER last DATA_BACK).
-                                        // Use send().await to guarantee delivery (try_send silently drops if full).
-                                        if !stream_token.is_cancelled() {
-                                            let close_frame = encode_frame(stream_id, FRAME_CLOSE_BACK, &[]);
-                                            let _ = data_writer_tx.send(close_frame).await;
-                                        }
-
-                                        writer_for_edge_data.abort();
-                                        Ok::<(), Box<dyn std::error::Error + Send + Sync>>(())
-                                    }
-                                    .await;
-
-                                    if let Err(e) = result {
-                                        log::error!("Stream {} error: {}", stream_id, e);
-                                        // Send CLOSE_BACK via DATA channel on error (must arrive after any DATA_BACK).
-                                        // Use send().await to guarantee delivery.
-                                        if !stream_token.is_cancelled() {
-                                            let close_frame = encode_frame(stream_id, FRAME_CLOSE_BACK, &[]);
-                                            let _ = data_writer_tx.send(close_frame).await;
-                                        }
-                                    }
-
-                                    // Clean up stream (guard against duplicate if FRAME_CLOSE already removed it)
-                                    let was_present = {
-                                        let mut s = streams_clone.lock().await;
-                                        s.remove(&stream_id).is_some()
-                                    };
-                                    if was_present {
-                                        let _ = event_tx_clone.try_send(HubEvent::StreamClosed {
-                                            edge_id: edge_id_clone,
-                                            stream_id,
-                                        });
-                                    }
-                                    stream_counter.fetch_sub(1, Ordering::Relaxed);
-                                });
-                            }
-                            FRAME_DATA => {
-                                // 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(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);
-                            }
-                        }
-                    }
-                    Ok(None) => {
-                        log::info!("Edge {} disconnected (EOF)", edge_id);
-                        break;
-                    }
-                    Err(e) => {
-                        log::error!("Edge {} frame error: {}", edge_id, e);
-                        break;
-                    }
+        match event {
+            remoteingress_protocol::TunnelEvent::Frame(frame) => {
+                last_activity = Instant::now();
+                liveness_deadline.as_mut().reset(last_activity + liveness_timeout_dur);
+                if let FrameAction::Disconnect(reason) = handle_hub_frame(
+                    frame, &mut tunnel_io, &mut streams, &stream_semaphore, &edge_stream_count,
+                    &edge_id, &event_tx, &ctrl_tx, &data_tx, &target_host, &edge_token,
+                    &cleanup_tx,
+                ).await {
+                    disconnect_reason = reason;
+                    break;
                 }
             }
-            _ = ping_ticker.tick() => {
-                let ping_frame = encode_frame(0, FRAME_PING, &[]);
-                if frame_writer_tx.try_send(ping_frame).is_err() {
-                    // Control channel full — skip this PING cycle.
-                    // The 45s liveness timeout gives margin for the channel to drain.
-                    log::warn!("PING send to edge {} failed, control channel full — skipping", edge_id);
-                }
-                log::trace!("Sent PING to edge {}", edge_id);
+            remoteingress_protocol::TunnelEvent::Eof => {
+                log::info!("Edge {} disconnected (EOF)", edge_id);
+                disconnect_reason = "edge_eof".to_string();
+                break;
             }
-            _ = &mut liveness_deadline => {
+            remoteingress_protocol::TunnelEvent::ReadError(e) => {
+                log::error!("Edge {} frame error: {}", edge_id, e);
+                disconnect_reason = format!("edge_frame_error: {}", e);
+                break;
+            }
+            remoteingress_protocol::TunnelEvent::WriteError(e) => {
+                log::error!("Tunnel write error to edge {}: {}", edge_id, e);
+                disconnect_reason = format!("tunnel_write_error: {}", e);
+                break;
+            }
+            remoteingress_protocol::TunnelEvent::LivenessTimeout => {
                 log::warn!("Edge {} liveness timeout (no frames for {}s), disconnecting",
                     edge_id, liveness_timeout_dur.as_secs());
+                disconnect_reason = "liveness_timeout".to_string();
                 break;
             }
-            _ = &mut writer_dead_rx => {
-                log::error!("Tunnel writer to edge {} died, disconnecting immediately", edge_id);
-                break;
-            }
-            _ = edge_token.cancelled() => {
+            remoteingress_protocol::TunnelEvent::Cancelled => {
                 log::info!("Edge {} cancelled by hub", edge_id);
+                disconnect_reason = "cancelled_by_hub".to_string();
                 break;
             }
         }
@@ -781,13 +825,13 @@ 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);
     }
     let _ = event_tx.try_send(HubEvent::EdgeDisconnected {
         edge_id: edge_id.clone(),
+        reason: disconnect_reason,
     });
 
     Ok(())
@@ -1010,10 +1054,12 @@ mod tests {
     fn test_hub_event_edge_disconnected_serialize() {
         let event = HubEvent::EdgeDisconnected {
             edge_id: "edge-2".to_string(),
+            reason: "liveness_timeout".to_string(),
         };
         let json = serde_json::to_value(&event).unwrap();
         assert_eq!(json["type"], "edgeDisconnected");
         assert_eq!(json["edgeId"], "edge-2");
+        assert_eq!(json["reason"], "liveness_timeout");
     }
 
     #[test]

rust/crates/remoteingress-protocol/Cargo.toml

@@ -4,7 +4,9 @@ version = "2.0.0"
 edition = "2021"
 
 [dependencies]
-tokio = { version = "1", features = ["io-util"] }
+tokio = { version = "1", features = ["io-util", "sync", "time"] }
+tokio-util = "0.7"
+log = "0.4"
 
 [dev-dependencies]
 tokio = { version = "1", features = ["io-util", "macros", "rt"] }

rust/crates/remoteingress-protocol/src/lib.rs

@@ -1,4 +1,8 @@
-use tokio::io::{AsyncRead, AsyncReadExt};
+use std::collections::VecDeque;
+use std::future::Future;
+use std::pin::Pin;
+use std::task::{Context, Poll};
+use tokio::io::{AsyncRead, AsyncReadExt, AsyncWrite, ReadBuf};
 
 // Frame type constants
 pub const FRAME_OPEN: u8 = 0x01;
@@ -68,6 +72,16 @@ pub fn encode_frame(stream_id: u32, frame_type: u8, payload: &[u8]) -> Vec<u8> {
     buf
 }
 
+/// Write a frame header into `buf[0..FRAME_HEADER_SIZE]`.
+/// The caller must ensure payload is already at `buf[FRAME_HEADER_SIZE..FRAME_HEADER_SIZE + payload_len]`.
+/// This enables zero-copy encoding: read directly into `buf[FRAME_HEADER_SIZE..]`, then
+/// prepend the header without copying the payload.
+pub fn encode_frame_header(buf: &mut [u8], stream_id: u32, frame_type: u8, payload_len: usize) {
+    buf[0..4].copy_from_slice(&stream_id.to_be_bytes());
+    buf[4] = frame_type;
+    buf[5..9].copy_from_slice(&(payload_len as u32).to_be_bytes());
+}
+
 /// Build a PROXY protocol v1 header line.
 /// Format: `PROXY TCP4 <client_ip> <edge_ip> <client_port> <dest_port>\r\n`
 pub fn build_proxy_v1_header(
@@ -120,9 +134,13 @@ impl<R: AsyncRead + Unpin> FrameReader<R> {
         ]);
 
         if length > MAX_PAYLOAD_SIZE {
+            log::error!(
+                "CORRUPT FRAME HEADER: raw={:02x?} stream_id={} type=0x{:02x} length={}",
+                self.header_buf, stream_id, frame_type, length
+            );
             return Err(std::io::Error::new(
                 std::io::ErrorKind::InvalidData,
-                format!("frame payload too large: {} bytes", length),
+                format!("frame payload too large: {} bytes (header={:02x?})", length, self.header_buf),
             ));
         }
 
@@ -144,10 +162,303 @@ impl<R: AsyncRead + Unpin> FrameReader<R> {
     }
 }
 
+// ---------------------------------------------------------------------------
+// TunnelIo: single-owner I/O multiplexer for the TLS tunnel connection
+// ---------------------------------------------------------------------------
+
+/// Events produced by the TunnelIo event loop.
+#[derive(Debug)]
+pub enum TunnelEvent {
+    /// A complete frame was read from the remote side.
+    Frame(Frame),
+    /// The remote side closed the connection (EOF).
+    Eof,
+    /// A read error occurred.
+    ReadError(std::io::Error),
+    /// A write error occurred.
+    WriteError(std::io::Error),
+    /// No frames received for the liveness timeout duration.
+    LivenessTimeout,
+    /// The cancellation token was triggered.
+    Cancelled,
+}
+
+/// Write state extracted into a sub-struct so the borrow checker can see
+/// disjoint field access between `self.write` and `self.stream`.
+struct WriteState {
+    ctrl_queue: VecDeque<Vec<u8>>,   // PONG, WINDOW_UPDATE, CLOSE, OPEN — always first
+    data_queue: VecDeque<Vec<u8>>,   // DATA, DATA_BACK — only when ctrl is empty
+    offset: usize,                   // progress within current frame being written
+    flush_needed: bool,
+}
+
+impl WriteState {
+    fn has_work(&self) -> bool {
+        !self.ctrl_queue.is_empty() || !self.data_queue.is_empty()
+    }
+}
+
+/// Single-owner I/O engine for the tunnel TLS connection.
+///
+/// Owns the TLS stream directly — no `tokio::io::split()`, no mutex.
+/// Uses two priority write queues: ctrl frames (PONG, WINDOW_UPDATE, CLOSE, OPEN)
+/// are ALWAYS written before data frames (DATA, DATA_BACK). This prevents
+/// WINDOW_UPDATE starvation that causes flow control deadlocks.
+pub struct TunnelIo<S> {
+    stream: S,
+    // Read state: accumulate bytes, parse frames incrementally
+    read_buf: Vec<u8>,
+    read_pos: usize,
+    parse_pos: usize,
+    // Write state: extracted sub-struct for safe disjoint borrows
+    write: WriteState,
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin> TunnelIo<S> {
+    pub fn new(stream: S, initial_data: Vec<u8>) -> Self {
+        let read_pos = initial_data.len();
+        let mut read_buf = initial_data;
+        if read_buf.capacity() < 65536 {
+            read_buf.reserve(65536 - read_buf.len());
+        }
+        Self {
+            stream,
+            read_buf,
+            read_pos,
+            parse_pos: 0,
+            write: WriteState {
+                ctrl_queue: VecDeque::new(),
+                data_queue: VecDeque::new(),
+                offset: 0,
+                flush_needed: false,
+            },
+        }
+    }
+
+    /// Queue a high-priority control frame (PONG, WINDOW_UPDATE, CLOSE, OPEN).
+    pub fn queue_ctrl(&mut self, frame: Vec<u8>) {
+        self.write.ctrl_queue.push_back(frame);
+    }
+
+    /// Queue a lower-priority data frame (DATA, DATA_BACK).
+    pub fn queue_data(&mut self, frame: Vec<u8>) {
+        self.write.data_queue.push_back(frame);
+    }
+
+    /// Try to parse a complete frame from the read buffer.
+    /// Uses a parse_pos cursor to avoid drain() on every frame.
+    pub fn try_parse_frame(&mut self) -> Option<Result<Frame, std::io::Error>> {
+        let available = self.read_pos - self.parse_pos;
+        if available < FRAME_HEADER_SIZE {
+            return None;
+        }
+
+        let base = self.parse_pos;
+        let stream_id = u32::from_be_bytes([
+            self.read_buf[base], self.read_buf[base + 1],
+            self.read_buf[base + 2], self.read_buf[base + 3],
+        ]);
+        let frame_type = self.read_buf[base + 4];
+        let length = u32::from_be_bytes([
+            self.read_buf[base + 5], self.read_buf[base + 6],
+            self.read_buf[base + 7], self.read_buf[base + 8],
+        ]);
+
+        if length > MAX_PAYLOAD_SIZE {
+            let header = [
+                self.read_buf[base], self.read_buf[base + 1],
+                self.read_buf[base + 2], self.read_buf[base + 3],
+                self.read_buf[base + 4], self.read_buf[base + 5],
+                self.read_buf[base + 6], self.read_buf[base + 7],
+                self.read_buf[base + 8],
+            ];
+            log::error!(
+                "CORRUPT FRAME HEADER: raw={:02x?} stream_id={} type=0x{:02x} length={}",
+                header, stream_id, frame_type, length
+            );
+            return Some(Err(std::io::Error::new(
+                std::io::ErrorKind::InvalidData,
+                format!("frame payload too large: {} bytes (header={:02x?})", length, header),
+            )));
+        }
+
+        let total_frame_size = FRAME_HEADER_SIZE + length as usize;
+        if available < total_frame_size {
+            return None;
+        }
+
+        let payload = self.read_buf[base + FRAME_HEADER_SIZE..base + total_frame_size].to_vec();
+        self.parse_pos += total_frame_size;
+
+        // Compact when parse_pos > half the data to reclaim memory
+        if self.parse_pos > self.read_pos / 2 && self.parse_pos > 0 {
+            self.read_buf.drain(..self.parse_pos);
+            self.read_pos -= self.parse_pos;
+            self.parse_pos = 0;
+        }
+
+        Some(Ok(Frame { stream_id, frame_type, payload }))
+    }
+
+    /// Poll-based I/O step. Returns Ready on events, Pending when idle.
+    ///
+    /// Order: write(ctrl→data) → flush → read → channels → timers
+    pub fn poll_step(
+        &mut self,
+        cx: &mut Context<'_>,
+        ctrl_rx: &mut tokio::sync::mpsc::Receiver<Vec<u8>>,
+        data_rx: &mut tokio::sync::mpsc::Receiver<Vec<u8>>,
+        liveness_deadline: &mut Pin<Box<tokio::time::Sleep>>,
+        cancel_token: &tokio_util::sync::CancellationToken,
+    ) -> Poll<TunnelEvent> {
+        // 1. WRITE: drain ctrl queue first, then data queue.
+        //    TLS poll_write writes plaintext to session buffer (always Ready).
+        //    Batch up to 16 frames per poll cycle.
+        //    Safe: `self.write` and `self.stream` are disjoint fields.
+        let mut writes = 0;
+        while self.write.has_work() && writes < 16 {
+            let from_ctrl = !self.write.ctrl_queue.is_empty();
+            let frame = if from_ctrl {
+                self.write.ctrl_queue.front().unwrap()
+            } else {
+                self.write.data_queue.front().unwrap()
+            };
+            let remaining = &frame[self.write.offset..];
+
+            match Pin::new(&mut self.stream).poll_write(cx, remaining) {
+                Poll::Ready(Ok(0)) => {
+                    return Poll::Ready(TunnelEvent::WriteError(
+                        std::io::Error::new(std::io::ErrorKind::WriteZero, "write zero"),
+                    ));
+                }
+                Poll::Ready(Ok(n)) => {
+                    self.write.offset += n;
+                    self.write.flush_needed = true;
+                    if self.write.offset >= frame.len() {
+                        if from_ctrl { self.write.ctrl_queue.pop_front(); }
+                        else { self.write.data_queue.pop_front(); }
+                        self.write.offset = 0;
+                        writes += 1;
+                    }
+                }
+                Poll::Ready(Err(e)) => return Poll::Ready(TunnelEvent::WriteError(e)),
+                Poll::Pending => break,
+            }
+        }
+
+        // 2. FLUSH: push encrypted data from TLS session to TCP.
+        if self.write.flush_needed {
+            match Pin::new(&mut self.stream).poll_flush(cx) {
+                Poll::Ready(Ok(())) => self.write.flush_needed = false,
+                Poll::Ready(Err(e)) => return Poll::Ready(TunnelEvent::WriteError(e)),
+                Poll::Pending => {} // TCP waker will notify us
+            }
+        }
+
+        // 3. READ: drain stream until Pending to ensure the TCP waker is always registered.
+        //    Without this loop, a Ready return with partial frame data would consume
+        //    the waker without re-registering it, causing the task to sleep until a
+        //    timer or channel wakes it (potentially 15+ seconds of lost reads).
+        loop {
+            // Compact if needed to make room for reads
+            if self.parse_pos > 0 && self.read_buf.len() - self.read_pos < 32768 {
+                self.read_buf.drain(..self.parse_pos);
+                self.read_pos -= self.parse_pos;
+                self.parse_pos = 0;
+            }
+            if self.read_buf.len() < self.read_pos + 32768 {
+                self.read_buf.resize(self.read_pos + 32768, 0);
+            }
+            let mut rbuf = ReadBuf::new(&mut self.read_buf[self.read_pos..]);
+            match Pin::new(&mut self.stream).poll_read(cx, &mut rbuf) {
+                Poll::Ready(Ok(())) => {
+                    let n = rbuf.filled().len();
+                    if n == 0 {
+                        return Poll::Ready(TunnelEvent::Eof);
+                    }
+                    self.read_pos += n;
+                    if let Some(result) = self.try_parse_frame() {
+                        return match result {
+                            Ok(frame) => Poll::Ready(TunnelEvent::Frame(frame)),
+                            Err(e) => Poll::Ready(TunnelEvent::ReadError(e)),
+                        };
+                    }
+                    // Partial data — loop to call poll_read again so the TCP
+                    // waker is re-registered when it finally returns Pending.
+                }
+                Poll::Ready(Err(e)) => return Poll::Ready(TunnelEvent::ReadError(e)),
+                Poll::Pending => break,
+            }
+        }
+
+        // 4. CHANNELS: drain ctrl into ctrl_queue, data into data_queue.
+        let mut got_new = false;
+        loop {
+            match ctrl_rx.poll_recv(cx) {
+                Poll::Ready(Some(frame)) => { self.write.ctrl_queue.push_back(frame); got_new = true; }
+                Poll::Ready(None) => {
+                    return Poll::Ready(TunnelEvent::WriteError(
+                        std::io::Error::new(std::io::ErrorKind::BrokenPipe, "ctrl channel closed"),
+                    ));
+                }
+                Poll::Pending => break,
+            }
+        }
+        loop {
+            match data_rx.poll_recv(cx) {
+                Poll::Ready(Some(frame)) => { self.write.data_queue.push_back(frame); got_new = true; }
+                Poll::Ready(None) => {
+                    return Poll::Ready(TunnelEvent::WriteError(
+                        std::io::Error::new(std::io::ErrorKind::BrokenPipe, "data channel closed"),
+                    ));
+                }
+                Poll::Pending => break,
+            }
+        }
+
+        // 5. TIMERS
+        if liveness_deadline.as_mut().poll(cx).is_ready() {
+            return Poll::Ready(TunnelEvent::LivenessTimeout);
+        }
+        if cancel_token.is_cancelled() {
+            return Poll::Ready(TunnelEvent::Cancelled);
+        }
+
+        // 6. SELF-WAKE: only when we have frames AND flush is done.
+        //    If flush is pending, the TCP write-readiness waker will notify us.
+        //    If we got new channel frames, wake to write them.
+        if got_new || (!self.write.flush_needed && self.write.has_work()) {
+            cx.waker().wake_by_ref();
+        }
+
+        Poll::Pending
+    }
+
+    pub fn into_inner(self) -> S {
+        self.stream
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
 
+    #[test]
+    fn test_encode_frame_header() {
+        let payload = b"hello";
+        let mut buf = vec![0u8; FRAME_HEADER_SIZE + payload.len()];
+        buf[FRAME_HEADER_SIZE..].copy_from_slice(payload);
+        encode_frame_header(&mut buf, 42, FRAME_DATA, payload.len());
+        assert_eq!(buf, encode_frame(42, FRAME_DATA, payload));
+    }
+
+    #[test]
+    fn test_encode_frame_header_empty_payload() {
+        let mut buf = vec![0u8; FRAME_HEADER_SIZE];
+        encode_frame_header(&mut buf, 99, FRAME_CLOSE, 0);
+        assert_eq!(buf, encode_frame(99, FRAME_CLOSE, &[]));
+    }
+
     #[test]
     fn test_encode_frame() {
         let data = b"hello";

ts/00_commitinfo_data.ts

@@ -3,6 +3,6 @@
  */
 export const commitinfo = {
   name: '@serve.zone/remoteingress',
-  version: '4.7.1',
+  version: '4.8.4',
   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.'
 }

ts/classes.remoteingressedge.ts

@@ -83,8 +83,10 @@ export class RemoteIngressEdge extends EventEmitter {
     this.bridge.on('management:tunnelConnected', () => {
       this.emit('tunnelConnected');
     });
-    this.bridge.on('management:tunnelDisconnected', () => {
-      this.emit('tunnelDisconnected');
+    this.bridge.on('management:tunnelDisconnected', (data: { reason?: string }) => {
+      const reason = data?.reason ?? 'unknown';
+      console.log(`[RemoteIngressEdge] Tunnel disconnected: ${reason}`);
+      this.emit('tunnelDisconnected', data);
     });
     this.bridge.on('management:publicIpDiscovered', (data: { ip: string }) => {
       this.emit('publicIpDiscovered', data);

ts/classes.remoteingresshub.ts

@@ -93,7 +93,9 @@ export class RemoteIngressHub extends EventEmitter {
     this.bridge.on('management:edgeConnected', (data: { edgeId: string; peerAddr: string }) => {
       this.emit('edgeConnected', data);
     });
-    this.bridge.on('management:edgeDisconnected', (data: { edgeId: string }) => {
+    this.bridge.on('management:edgeDisconnected', (data: { edgeId: string; reason?: string }) => {
+      const reason = data?.reason ?? 'unknown';
+      console.log(`[RemoteIngressHub] Edge ${data.edgeId} disconnected: ${reason}`);
       this.emit('edgeDisconnected', data);
     });
     this.bridge.on('management:streamOpened', (data: { edgeId: string; streamId: number }) => {