v4.8.4

changelog.md

@@ -1,5 +1,179 @@
 # 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
+
+- Enable TCP keepalive on edge and hub connections to detect silent network failures sooner
+- Trigger immediate reconnect or disconnect when tunnel writer tasks fail instead of waiting for liveness timeouts
+- Prevent active stream counter underflow during concurrent connection cleanup
+
+## 2026-03-16 - 4.7.0 - feat(edge,protocol,test)
+add configurable edge bind address and expand flow-control test coverage
+
+- adds an optional bindAddress configuration for edge TCP listeners, defaulting to 0.0.0.0 when not provided
+- passes bindAddress through the TypeScript edge client and Rust edge runtime so local test setups can bind to localhost
+- adds protocol unit tests for adaptive stream window sizing and window update frame encoding/decoding
+- introduces end-to-end flow-control tests and updates the test script to build before running tests
+
+## 2026-03-16 - 4.6.1 - fix(remoteingress-core)
+avoid spurious tunnel disconnect events and increase control channel capacity
+
+- Emit TunnelDisconnected only after an established connection is actually lost, preventing false disconnect events during failed reconnect attempts.
+- Increase edge and hub control-channel buffer sizes from 64 to 256 to better prioritize control frames under load.
+
+## 2026-03-16 - 4.6.0 - feat(remoteingress-core)
+add adaptive per-stream flow control based on active stream counts
+
+- Track active stream counts on edge and hub connections to size per-stream flow control windows dynamically.
+- Cap WINDOW_UPDATE increments and read sizes to the adaptive window so bandwidth is shared more evenly across concurrent streams.
+- Apply the adaptive logic to both upload and download paths on edge and hub stream handlers.
+
+## 2026-03-16 - 4.5.12 - fix(remoteingress-core)
+improve tunnel liveness handling and enable TCP keepalive for accepted client sockets
+
+- Avoid disconnecting edges when PING or PONG frames cannot be queued because the control channel is temporarily full.
+- Enable TCP_NODELAY and TCP keepalive on accepted client connections to help detect stale or dropped clients.
+
+## 2026-03-16 - 4.5.11 - fix(repo)
+no changes to commit
+
+
+## 2026-03-16 - 4.5.10 - fix(remoteingress-core)
+guard zero-window reads to avoid false EOF handling on stalled streams
+
+- Prevent upload and download loops from calling read on an empty buffer when flow-control window remains at 0 after stall timeout
+- Log a warning and close the affected stream instead of misinterpreting Ok(0) as end-of-file
+
+## 2026-03-16 - 4.5.9 - fix(remoteingress-core)
+delay stream close until downstream response draining finishes to prevent truncated transfers
+
+- Waits for the hub-to-client download task to finish before sending the stream CLOSE frame
+- Prevents upstream reads from being cancelled mid-response during asymmetric transfers such as git fetch
+- Retains the existing timeout so stalled downloads still clean up safely
+
+## 2026-03-16 - 4.5.8 - fix(remoteingress-core)
+ensure upstream writes cancel promptly and reliably deliver CLOSE_BACK frames
+
+- listen for stream cancellation while waiting on upstream write timeouts so FRAME_CLOSE does not block for up to 60 seconds
+- replace try_send with send().await when emitting CLOSE_BACK frames to avoid silently dropping close notifications when the data channel is full
+
+## 2026-03-16 - 4.5.7 - fix(remoteingress-core)
+improve tunnel reconnect and frame write efficiency
+
+- Reuse the TLS connector across edge reconnections to preserve session resumption state and reduce reconnect latency.
+- Buffer hub and edge frame writes to coalesce small control and data frames into fewer TLS records and syscalls while still flushing each frame promptly.
+
+## 2026-03-16 - 4.5.6 - fix(remoteingress-core)
+disable Nagle's algorithm on edge, hub, and upstream TCP sockets to reduce control-frame latency
+
+- Enable TCP_NODELAY on the edge connection to the hub for faster PING/PONG and WINDOW_UPDATE delivery
+- Apply TCP_NODELAY on accepted hub streams before TLS handling
+- Enable TCP_NODELAY on SmartProxy upstream connections before sending the PROXY header
+
+## 2026-03-16 - 4.5.5 - fix(remoteingress-core)
+wait for hub-to-client draining before cleanup and reliably send close frames
+
+- switch CLOSE frame delivery on the data channel from try_send to send().await to avoid dropping it when the channel is full
+- delay stream cleanup until the hub-to-client task finishes or times out so large downstream responses continue after upload EOF
+- add a bounded 5-minute wait for download draining to prevent premature termination of asymmetric transfers such as git fetch
+
+## 2026-03-15 - 4.5.4 - fix(remoteingress-core)
+preserve stream close ordering and add flow-control stall timeouts
+
+- Send CLOSE and CLOSE_BACK frames on the data channel so they arrive after the final stream data frames.
+- Log and abort stalled upload and download paths when flow-control windows stay empty for 120 seconds.
+- Apply a 60-second timeout when writing buffered stream data to the upstream connection to prevent hung streams.
+
+## 2026-03-15 - 4.5.3 - fix(remoteingress-core)
+prioritize control frames over data in edge and hub tunnel writers
+
+- Split tunnel/frame writers into separate control and data channels in edge and hub
+- Use biased select loops so PING, PONG, WINDOW_UPDATE, OPEN, and CLOSE frames are sent before data frames
+- Route stream data through dedicated data channels while keeping OPEN, CLOSE, and flow-control updates on control channels to prevent keepalive starvation under load
+
+## 2026-03-15 - 4.5.2 - fix(remoteingress-core)
+improve stream flow control retries and increase channel buffer capacity
+
+- increase per-stream mpsc channel capacity from 128 to 256 on both edge and hub paths
+- only reset accumulated window update bytes after a successful try_send to avoid dropping flow-control credits when the update channel is busy
+
+## 2026-03-15 - 4.5.1 - fix(protocol)
+increase per-stream flow control window and channel buffers to improve high-RTT throughput
+
+- raise the initial stream window from 256 KB to 4 MB to allow more in-flight data per stream
+- increase edge and hub mpsc channel capacities from 16 to 128 to better absorb throughput under flow control
+
+## 2026-03-15 - 4.5.0 - feat(remoteingress-core)
+add per-stream flow control for edge and hub tunnel data transfer
+
+- introduce WINDOW_UPDATE frame types and protocol helpers for per-stream flow control
+- track per-stream send windows on both edge and hub to limit reads based on available capacity
+- send window updates after downstream writes to reduce channel pressure during large transfers
+
+## 2026-03-15 - 4.4.1 - fix(remoteingress-core)
+prevent stream data loss by applying backpressure and closing saturated channels
+
+- replace non-blocking frame writes with awaited sends in per-stream tasks so large transfers respect backpressure instead of dropping data
+- close and remove streams when back-channel or data channels fill up to avoid TCP stream corruption from silently dropped frames
+
+## 2026-03-03 - 4.4.0 - feat(remoteingress)
+add heartbeat PING/PONG and liveness timeouts; implement fast-reconnect/backoff reset and JS crash-recovery auto-restart
+
+- protocol: add FRAME_PING and FRAME_PONG and unit tests for ping/pong frames
+- edge (Rust): reset backoff after successful connection, respond to PING with PONG, track liveness via deadline and reconnect on timeout, use Duration/Instant helpers
+- hub (Rust): send periodic PING to edges, handle PONGs, enforce liveness timeout and disconnect inactive edges, use tokio interval and time utilities
+- ts: RemoteIngressEdge and RemoteIngressHub: add crash-recovery auto-restart with exponential backoff and max attempts, save/restore config and allowed edges, register/remove exit handlers, ensure stop() marks stopping and cleans up listeners
+- minor API/typing: introduce TAllowedEdge alias and persist allowed edges for restart recovery
+
+## 2026-02-26 - 4.3.0 - feat(hub)
+add optional TLS certificate/key support to hub start config and bridge
+
+- TypeScript: add tls.certPem and tls.keyPem to IHubConfig and include tlsCertPem/tlsKeyPem in startHub bridge command when both are provided
+- TypeScript: extend startHub params with tlsCertPem and tlsKeyPem and conditionally send them
+- Rust: change HubConfig serde attributes for tls_cert_pem and tls_key_pem from skip to default so absent PEM fields deserialize as None
+- Enables optional provisioning of TLS certificate and key to the hub when provided from the JS side
+
 ## 2026-02-26 - 4.2.0 - feat(core)
 expose edge peer address in hub events and migrate writers to channel-based, non-blocking framing with stream limits and timeouts
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@serve.zone/remoteingress",
-  "version": "4.2.0",
+  "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",
@@ -9,7 +9,7 @@
   "author": "Task Venture Capital GmbH",
   "license": "MIT",
   "scripts": {
-    "test": "(tstest test/ --verbose --logfile --timeout 60)",
+    "test": "(pnpm run build && tstest test/ --verbose --logfile --timeout 60)",
     "build": "(tsbuild tsfolders --allowimplicitany && tsrust)",
     "buildDocs": "(tsdoc)"
   },

rust/Cargo.lock

@@ -558,6 +558,7 @@ dependencies = [
  "rustls-pemfile",
  "serde",
  "serde_json",
+ "socket2 0.5.10",
  "tokio",
  "tokio-rustls",
  "tokio-util",
@@ -567,7 +568,9 @@ dependencies = [
 name = "remoteingress-protocol"
 version = "2.0.0"
 dependencies = [
+ "log",
  "tokio",
+ "tokio-util",
 ]
 
 [[package]]
@@ -701,6 +704,16 @@ version = "1.15.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "67b1b7a3b5fe4f1376887184045fcf45c69e92af734b7aaddc05fb777b6fbd03"
 
+[[package]]
+name = "socket2"
+version = "0.5.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e22376abed350d73dd1cd119b57ffccad95b4e585a7cda43e286245ce23c0678"
+dependencies = [
+ "libc",
+ "windows-sys 0.52.0",
+]
+
 [[package]]
 name = "socket2"
 version = "0.6.2"
@@ -765,7 +778,7 @@ dependencies = [
  "parking_lot",
  "pin-project-lite",
  "signal-hook-registry",
- "socket2",
+ "socket2 0.6.2",
  "tokio-macros",
  "windows-sys 0.61.2",
 ]

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/Cargo.toml

@@ -14,3 +14,4 @@ serde_json = "1"
 log = "0.4"
 rustls-pemfile = "2"
 tokio-util = "0.7"
+socket2 = "0.5"

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

@@ -1,16 +1,39 @@
 use std::collections::HashMap;
 use std::sync::atomic::{AtomicU32, Ordering};
 use std::sync::Arc;
-use tokio::io::{AsyncBufReadExt, AsyncReadExt, AsyncWriteExt, BufReader};
+use std::time::Duration;
+use tokio::io::{AsyncReadExt, AsyncWriteExt};
 use tokio::net::{TcpListener, TcpStream};
-use tokio::sync::{mpsc, Mutex, RwLock};
+use tokio::sync::{mpsc, Mutex, Notify, RwLock};
 use tokio::task::JoinHandle;
+use tokio::time::{Instant, sleep_until};
 use tokio_rustls::TlsConnector;
 use tokio_util::sync::CancellationToken;
 use serde::{Deserialize, Serialize};
 
 use remoteingress_protocol::*;
 
+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 {
+    /// 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>,
+    /// Notifier to wake the client reader when the window opens.
+    window_notify: Arc<Notify>,
+}
+
 /// Edge configuration (hub-host + credentials only; ports come from hub).
 #[derive(Debug, Clone, Deserialize, Serialize)]
 #[serde(rename_all = "camelCase")]
@@ -19,6 +42,10 @@ pub struct EdgeConfig {
     pub hub_port: u16,
     pub edge_id: String,
     pub secret: String,
+    /// Optional bind address for TCP listeners (defaults to "0.0.0.0").
+    /// Useful for testing on localhost where edge and upstream share the same machine.
+    #[serde(default)]
+    pub bind_address: Option<String>,
 }
 
 /// Handshake config received from hub after authentication.
@@ -47,7 +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")]
@@ -181,6 +209,14 @@ async fn edge_main_loop(
     let mut backoff_ms: u64 = 1000;
     let max_backoff_ms: u64 = 30000;
 
+    // Build TLS config ONCE outside the reconnect loop — preserves session
+    // cache across reconnections for TLS session resumption (saves 1 RTT).
+    let tls_config = rustls::ClientConfig::builder()
+        .dangerous()
+        .with_custom_certificate_verifier(Arc::new(NoCertVerifier))
+        .with_no_client_auth();
+    let connector = TlsConnector::from(Arc::new(tls_config));
+
     loop {
         // Create a per-connection child token
         let connection_token = cancel_token.child_token();
@@ -196,14 +232,31 @@ async fn edge_main_loop(
             &listen_ports,
             &mut shutdown_rx,
             &connection_token,
+            &connector,
         )
         .await;
 
         // Cancel connection token to kill all orphaned tasks from this cycle
         connection_token.cancel();
 
+        // Reset backoff after a successful connection for fast reconnect
+        let was_connected = *connected.read().await;
+        if was_connected {
+            backoff_ms = 1000;
+            log::info!("Was connected; resetting backoff to {}ms for fast reconnect", backoff_ms);
+        }
+
         *connected.write().await = false;
-        let _ = event_tx.try_send(EdgeEvent::TunnelDisconnected);
+        // 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 { reason: reason.clone() });
+        }
         active_streams.store(0, Ordering::Relaxed);
         // Reset stream ID counter for next connection cycle
         next_stream_id.store(1, Ordering::Relaxed);
@@ -211,10 +264,10 @@ 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(std::time::Duration::from_millis(backoff_ms)) => {}
+                    _ = tokio::time::sleep(Duration::from_millis(backoff_ms)) => {}
                     _ = cancel_token.cancelled() => break,
                     _ = shutdown_rx.recv() => break,
                 }
@@ -226,7 +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(
@@ -239,63 +372,78 @@ async fn connect_to_hub_and_run(
     listen_ports: &Arc<RwLock<Vec<u16>>>,
     shutdown_rx: &mut mpsc::Receiver<()>,
     connection_token: &CancellationToken,
+    connector: &TlsConnector,
 ) -> EdgeLoopResult {
-    // Build TLS connector that skips cert verification (auth is via secret)
-    let tls_config = rustls::ClientConfig::builder()
-        .dangerous()
-        .with_custom_certificate_verifier(Arc::new(NoCertVerifier))
-        .with_no_client_auth();
-
-    let connector = TlsConnector::from(Arc::new(tls_config));
 
     let addr = format!("{}:{}", config.hub_host, config.hub_port);
     let tcp = match TcpStream::connect(&addr).await {
-        Ok(s) => s,
+        Ok(s) => {
+            // Disable Nagle's algorithm for low-latency control frames (PING/PONG, WINDOW_UPDATE)
+            let _ = s.set_nodelay(true);
+            // TCP keepalive detects silent network failures (NAT timeout, path change)
+            // faster than the 45s application-level liveness timeout.
+            let ka = socket2::TcpKeepalive::new()
+                .with_time(Duration::from_secs(30));
+            #[cfg(target_os = "linux")]
+            let ka = ka.with_interval(Duration::from_secs(10));
+            let _ = socket2::SockRef::from(&s).set_tcp_keepalive(&ka);
+            s
+        }
         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));
         }
     };
 
@@ -336,119 +484,112 @@ async fn connect_to_hub_and_run(
                 _ = stun_token.cancelled() => break,
             }
             tokio::select! {
-                _ = tokio::time::sleep(std::time::Duration::from_secs(stun_interval)) => {}
+                _ = tokio::time::sleep(Duration::from_secs(stun_interval)) => {}
                 _ = stun_token.cancelled() => break,
             }
         }
     });
 
-    // Client socket map: stream_id -> sender for writing data back to client
-    let client_writers: Arc<Mutex<HashMap<u32, mpsc::Sender<Vec<u8>>>>> =
+    // Client socket map: stream_id -> per-stream state (back channel + flow control)
+    let client_writers: Arc<Mutex<HashMap<u32, EdgeStreamState>>> =
         Arc::new(Mutex::new(HashMap::new()));
 
-    // A5: Channel-based tunnel writer replaces Arc<Mutex<WriteHalf>>
-    let (tunnel_writer_tx, mut tunnel_writer_rx) = mpsc::channel::<Vec<u8>>(4096);
-    let tw_token = connection_token.clone();
-    let tunnel_writer_handle = tokio::spawn(async move {
-        loop {
-            tokio::select! {
-                data = tunnel_writer_rx.recv() => {
-                    match data {
-                        Some(frame_data) => {
-                            if write_half.write_all(&frame_data).await.is_err() {
-                                break;
-                            }
-                        }
-                        None => break,
-                    }
-                }
-                _ = tw_token.cancelled() => break,
-            }
-        }
-    });
+    // 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);
+    let tunnel_writer_tx = tunnel_ctrl_tx.clone();
 
-    // 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(
         &handshake.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,
     );
 
-    // 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)) => {
-                        match frame.frame_type {
-                            FRAME_DATA_BACK => {
-                                // A1: Non-blocking send to prevent head-of-line blocking
-                                let writers = client_writers.lock().await;
-                                if let Some(tx) = writers.get(&frame.stream_id) {
-                                    if tx.try_send(frame.payload).is_err() {
-                                        log::warn!("Stream {} back-channel full, dropping frame", frame.stream_id);
-                                    }
-                                }
-                            }
-                            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,
-                                        &client_writers,
-                                        active_streams,
-                                        next_stream_id,
-                                        &config.edge_id,
-                                        connection_token,
-                                    );
-                                }
-                            }
-                            _ => {
-                                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;
-                    }
+    // 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));
+
+    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);
+            }
+        }
+
+        // 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);
                 }
             }
-            _ = connection_token.cancelled() => {
-                log::info!("Connection cancelled");
-                break EdgeLoopResult::Shutdown;
+            remoteingress_protocol::TunnelEvent::Eof => {
+                log::info!("Hub disconnected (EOF)");
+                break EdgeLoopResult::Reconnect("hub_eof".to_string());
             }
-            _ = shutdown_rx.recv() => {
+            remoteingress_protocol::TunnelEvent::ReadError(e) => {
+                log::error!("Hub frame read error: {}", e);
+                break EdgeLoopResult::Reconnect(format!("hub_frame_error: {}", e));
+            }
+            remoteingress_protocol::TunnelEvent::WriteError(e) => {
+                log::error!("Tunnel write error: {}", e);
+                break EdgeLoopResult::Reconnect(format!("tunnel_write_error: {}", e));
+            }
+            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();
     }
@@ -460,12 +601,14 @@ async fn connect_to_hub_and_run(
 fn apply_port_config(
     new_ports: &[u16],
     port_listeners: &mut HashMap<u16, JoinHandle<()>>,
-    tunnel_writer_tx: &mpsc::Sender<Vec<u8>>,
-    client_writers: &Arc<Mutex<HashMap<u32, mpsc::Sender<Vec<u8>>>>>,
+    tunnel_ctrl_tx: &mpsc::Sender<Vec<u8>>,
+    tunnel_data_tx: &mpsc::Sender<Vec<u8>>,
+    client_writers: &Arc<Mutex<HashMap<u32, EdgeStreamState>>>,
     active_streams: &Arc<AtomicU32>,
     next_stream_id: &Arc<AtomicU32>,
     edge_id: &str,
     connection_token: &CancellationToken,
+    bind_address: &str,
 ) {
     let new_set: std::collections::HashSet<u16> = new_ports.iter().copied().collect();
     let old_set: std::collections::HashSet<u16> = port_listeners.keys().copied().collect();
@@ -480,15 +623,17 @@ fn apply_port_config(
 
     // Add new ports
     for &port in new_set.difference(&old_set) {
-        let tunnel_writer_tx = tunnel_writer_tx.clone();
+        let tunnel_ctrl_tx = tunnel_ctrl_tx.clone();
+        let tunnel_data_tx = tunnel_data_tx.clone();
         let client_writers = client_writers.clone();
         let active_streams = active_streams.clone();
         let next_stream_id = next_stream_id.clone();
         let edge_id = edge_id.to_string();
         let port_token = connection_token.child_token();
 
+        let bind_addr = bind_address.to_string();
         let handle = tokio::spawn(async move {
-            let listener = match TcpListener::bind(("0.0.0.0", port)).await {
+            let listener = match TcpListener::bind((bind_addr.as_str(), port)).await {
                 Ok(l) => l,
                 Err(e) => {
                     log::error!("Failed to bind port {}: {}", port, e);
@@ -502,8 +647,18 @@ fn apply_port_config(
                     accept_result = listener.accept() => {
                         match accept_result {
                             Ok((client_stream, client_addr)) => {
+                                // TCP keepalive detects dead clients that disappear without FIN.
+                                // Without this, zombie streams accumulate and never get cleaned up.
+                                let _ = client_stream.set_nodelay(true);
+                                let ka = socket2::TcpKeepalive::new()
+                                    .with_time(Duration::from_secs(60));
+                                #[cfg(target_os = "linux")]
+                                let ka = ka.with_interval(Duration::from_secs(60));
+                                let _ = socket2::SockRef::from(&client_stream).set_tcp_keepalive(&ka);
+
                                 let stream_id = next_stream_id.fetch_add(1, Ordering::Relaxed);
-                                let tunnel_writer_tx = tunnel_writer_tx.clone();
+                                let tunnel_ctrl_tx = tunnel_ctrl_tx.clone();
+                                let tunnel_data_tx = tunnel_data_tx.clone();
                                 let client_writers = client_writers.clone();
                                 let active_streams = active_streams.clone();
                                 let edge_id = edge_id.clone();
@@ -518,12 +673,25 @@ fn apply_port_config(
                                         stream_id,
                                         port,
                                         &edge_id,
-                                        tunnel_writer_tx,
+                                        tunnel_ctrl_tx,
+                                        tunnel_data_tx,
                                         client_writers,
                                         client_token,
+                                        Arc::clone(&active_streams),
                                     )
                                     .await;
-                                    active_streams.fetch_sub(1, Ordering::Relaxed);
+                                    // Saturating decrement: prevent underflow when
+                                    // edge_main_loop's store(0) races with task cleanup.
+                                    loop {
+                                        let current = active_streams.load(Ordering::Relaxed);
+                                        if current == 0 { break; }
+                                        if active_streams.compare_exchange_weak(
+                                            current, current - 1,
+                                            Ordering::Relaxed, Ordering::Relaxed,
+                                        ).is_ok() {
+                                            break;
+                                        }
+                                    }
                                 });
                             }
                             Err(e) => {
@@ -548,9 +716,11 @@ async fn handle_client_connection(
     stream_id: u32,
     dest_port: u16,
     edge_id: &str,
-    tunnel_writer_tx: mpsc::Sender<Vec<u8>>,
-    client_writers: Arc<Mutex<HashMap<u32, mpsc::Sender<Vec<u8>>>>>,
+    tunnel_ctrl_tx: mpsc::Sender<Vec<u8>>,
+    tunnel_data_tx: mpsc::Sender<Vec<u8>>,
+    client_writers: Arc<Mutex<HashMap<u32, EdgeStreamState>>>,
     client_token: CancellationToken,
+    active_streams: Arc<AtomicU32>,
 ) {
     let client_ip = client_addr.ip().to_string();
     let client_port = client_addr.port();
@@ -558,33 +728,74 @@ async fn handle_client_connection(
     // Determine edge IP (use 0.0.0.0 as placeholder — hub doesn't use it for routing)
     let edge_ip = "0.0.0.0";
 
-    // Send OPEN frame with PROXY v1 header via writer channel
+    // Send OPEN frame with PROXY v1 header via control channel
     let proxy_header = build_proxy_v1_header(&client_ip, edge_ip, client_port, dest_port);
     let open_frame = encode_frame(stream_id, FRAME_OPEN, proxy_header.as_bytes());
-    if tunnel_writer_tx.send(open_frame).await.is_err() {
+    if tunnel_ctrl_tx.send(open_frame).await.is_err() {
         return;
     }
 
-    // Set up channel for data coming back from hub
-    let (back_tx, mut back_rx) = mpsc::channel::<Vec<u8>>(256);
+    // 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;
-        writers.insert(stream_id, back_tx);
+        writers.insert(stream_id, EdgeStreamState {
+            back_tx,
+            send_window: Arc::clone(&send_window),
+            window_notify: Arc::clone(&window_notify),
+        });
     }
 
     let (mut client_read, mut client_write) = client_stream.into_split();
 
-    // Task: hub -> client
+    // Task: hub -> client (download direction)
+    // After writing to client TCP, send WINDOW_UPDATE to hub so it can send more
     let hub_to_client_token = client_token.clone();
-    let hub_to_client = tokio::spawn(async move {
+    let wu_tx = tunnel_ctrl_tx.clone();
+    let active_streams_h2c = Arc::clone(&active_streams);
+    let mut hub_to_client = tokio::spawn(async move {
+        let mut consumed_since_update: u32 = 0;
         loop {
             tokio::select! {
                 data = back_rx.recv() => {
                     match data {
                         Some(data) => {
+                            let len = data.len() as u32;
                             if client_write.write_all(&data).await.is_err() {
                                 break;
                             }
+                            // Track consumption for adaptive flow control.
+                            // The increment is capped to the adaptive window so the sender's
+                            // effective window shrinks to match current demand (fewer streams
+                            // = larger window, more streams = smaller window per stream).
+                            consumed_since_update += len;
+                            let adaptive_window = remoteingress_protocol::compute_window_for_stream_count(
+                                active_streams_h2c.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, 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,
+                                }
+                            }
                         }
                         None => break,
                     }
@@ -592,21 +803,63 @@ async fn handle_client_connection(
                 _ = hub_to_client_token.cancelled() => break,
             }
         }
+        // Send final window update for any remaining consumed bytes
+        if consumed_since_update > 0 {
+            let frame = encode_window_update(stream_id, FRAME_WINDOW_UPDATE, consumed_since_update);
+            let _ = wu_tx.send(frame).await;
+        }
         let _ = client_write.shutdown().await;
     });
 
-    // Task: client -> hub (via writer channel)
-    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).
+        // 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,
+                _ = client_token.cancelled() => break,
+                _ = tokio::time::sleep(Duration::from_secs(120)) => {
+                    log::warn!("Stream {} upload stalled (window empty for 120s)", stream_id);
+                    break;
+                }
+            }
+        }
+        if client_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 {} upload: 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(
+            active_streams.load(Ordering::Relaxed),
+        ) as usize;
+        let max_read = w.min(32768).min(adaptive_cap);
+
         tokio::select! {
-            read_result = client_read.read(&mut buf) => {
+            read_result = client_read.read(&mut buf[FRAME_HEADER_SIZE..FRAME_HEADER_SIZE + max_read]) => {
                 match read_result {
                     Ok(0) => break,
                     Ok(n) => {
-                        let data_frame = encode_frame(stream_id, FRAME_DATA, &buf[..n]);
-                        // A5: Use try_send to avoid blocking if writer channel is full
-                        if tunnel_writer_tx.try_send(data_frame).is_err() {
-                            log::warn!("Stream {} tunnel writer full, closing", stream_id);
+                        send_window.fetch_sub(n as u32, Ordering::Release);
+                        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;
                         }
                     }
@@ -617,18 +870,29 @@ async fn handle_client_connection(
         }
     }
 
-    // Send CLOSE frame (only if not cancelled)
+    // Wait for the download task (hub → client) to finish BEFORE sending CLOSE.
+    // Upload EOF (client done sending) does NOT mean the response is done.
+    // For asymmetric transfers like git fetch (small request, large response),
+    // the response is still streaming when the upload finishes.
+    // Sending CLOSE before the response finishes would cause the hub to cancel
+    // the upstream reader mid-response, truncating the data.
+    let _ = tokio::time::timeout(
+        Duration::from_secs(300),  // 5 min max wait for download to finish
+        &mut hub_to_client,
+    ).await;
+
+    // NOW send CLOSE — the response has been fully delivered (or timed out).
     if !client_token.is_cancelled() {
         let close_frame = encode_frame(stream_id, FRAME_CLOSE, &[]);
-        let _ = tunnel_writer_tx.try_send(close_frame);
+        let _ = tunnel_data_tx.send(close_frame).await;
     }
 
-    // Cleanup
+    // Clean up
     {
         let mut writers = client_writers.lock().await;
         writers.remove(&stream_id);
     }
-    hub_to_client.abort();
+    hub_to_client.abort(); // No-op if already finished; safety net if timeout fired
     let _ = edge_id; // used for logging context
 }
 
@@ -660,6 +924,7 @@ mod tests {
             hub_port: 9999,
             edge_id: "e1".to_string(),
             secret: "sec".to_string(),
+            bind_address: None,
         };
         let json = serde_json::to_string(&config).unwrap();
         let back: EdgeConfig = serde_json::from_str(&json).unwrap();
@@ -731,9 +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]
@@ -775,6 +1041,7 @@ mod tests {
             hub_port: 8443,
             edge_id: "test-edge".to_string(),
             secret: "test-secret".to_string(),
+            bind_address: None,
         });
         let status = edge.get_status().await;
         assert!(!status.running);
@@ -791,6 +1058,7 @@ mod tests {
             hub_port: 8443,
             edge_id: "e".to_string(),
             secret: "s".to_string(),
+            bind_address: None,
         });
         let rx1 = edge.take_event_rx().await;
         assert!(rx1.is_some());
@@ -805,6 +1073,7 @@ mod tests {
             hub_port: 8443,
             edge_id: "e".to_string(),
             secret: "s".to_string(),
+            bind_address: None,
         });
         edge.stop().await; // should not panic
         let status = edge.get_status().await;

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

@@ -1,23 +1,50 @@
 use std::collections::HashMap;
 use std::sync::Arc;
-use tokio::io::{AsyncBufReadExt, AsyncReadExt, AsyncWriteExt, BufReader};
+use std::sync::atomic::{AtomicU32, Ordering};
+use std::time::Duration;
+use tokio::io::{AsyncReadExt, AsyncWriteExt};
 use tokio::net::{TcpListener, TcpStream};
-use tokio::sync::{mpsc, Mutex, RwLock, Semaphore};
+use tokio::sync::{mpsc, Mutex, Notify, RwLock, Semaphore};
+use tokio::time::{interval, sleep_until, Instant};
 use tokio_rustls::TlsAcceptor;
 use tokio_util::sync::CancellationToken;
 use serde::{Deserialize, Serialize};
 
 use remoteingress_protocol::*;
 
+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 {
+    /// 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).
+    /// Decremented by the upstream reader, incremented by FRAME_WINDOW_UPDATE from edge.
+    send_window: Arc<AtomicU32>,
+    /// Notifier to wake the upstream reader when the window opens.
+    window_notify: Arc<Notify>,
+}
+
 /// Hub configuration.
 #[derive(Debug, Clone, Deserialize, Serialize)]
 #[serde(rename_all = "camelCase")]
 pub struct HubConfig {
     pub tunnel_port: u16,
     pub target_host: Option<String>,
-    #[serde(skip)]
+    #[serde(default)]
     pub tls_cert_pem: Option<String>,
-    #[serde(skip)]
+    #[serde(default)]
     pub tls_key_pem: Option<String>,
 }
 
@@ -76,7 +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")]
@@ -107,7 +134,7 @@ pub struct TunnelHub {
 struct ConnectedEdgeInfo {
     connected_at: u64,
     peer_addr: String,
-    active_streams: Arc<Mutex<HashMap<u32, (mpsc::Sender<Vec<u8>>, CancellationToken)>>>,
+    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,
@@ -173,11 +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(),
             });
         }
@@ -271,6 +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,
@@ -282,13 +592,33 @@ async fn handle_edge_connection(
     edge_token: CancellationToken,
     peer_addr: String,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-    let tls_stream = acceptor.accept(stream).await?;
-    let (read_half, mut write_half) = tokio::io::split(tls_stream);
-    let mut buf_reader = BufReader::new(read_half);
+    // Disable Nagle's algorithm for low-latency control frames (PING/PONG, WINDOW_UPDATE)
+    stream.set_nodelay(true)?;
+    // TCP keepalive detects silent network failures (NAT timeout, path change)
+    // faster than the 45s application-level liveness timeout.
+    let ka = socket2::TcpKeepalive::new()
+        .with_time(Duration::from_secs(30));
+    #[cfg(target_os = "linux")]
+    let ka = ka.with_interval(Duration::from_secs(10));
+    let _ = socket2::SockRef::from(&stream).set_tcp_keepalive(&ka);
+    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();
@@ -328,11 +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, (mpsc::Sender<Vec<u8>>, CancellationToken)>>> =
-        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()
@@ -348,37 +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(),
             },
         );
     }
 
-    // A5: Channel-based writer replaces Arc<Mutex<WriteHalf>>
-    // All frame writes go through this channel → dedicated writer task serializes them
-    let (frame_writer_tx, mut frame_writer_rx) = mpsc::channel::<Vec<u8>>(4096);
-    let writer_token = edge_token.clone();
-    let writer_handle = tokio::spawn(async move {
-        loop {
-            tokio::select! {
-                data = frame_writer_rx.recv() => {
-                    match data {
-                        Some(frame_data) => {
-                            if write_half.write_all(&frame_data).await.is_err() {
-                                break;
-                            }
-                        }
-                        None => break,
-                    }
-                }
-                _ = writer_token.cancelled() => break,
-            }
-        }
-    });
+    // 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);
 
     // 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 {
@@ -407,187 +724,99 @@ async fn handle_edge_connection(
     // A4: Semaphore to limit concurrent streams per edge
     let stream_semaphore = Arc::new(Semaphore::new(MAX_STREAMS_PER_EDGE));
 
-    // Frame reading loop
-    let mut frame_reader = FrameReader::new(buf_reader);
+    // Heartbeat: periodic PING and liveness timeout
+    let ping_interval_dur = Duration::from_secs(15);
+    let liveness_timeout_dur = Duration::from_secs(45);
+    let mut ping_ticker = interval(ping_interval_dur);
+    ping_ticker.tick().await; // consume the immediate first tick
+    let mut last_activity = Instant::now();
+    let mut liveness_deadline = Box::pin(sleep_until(last_activity + liveness_timeout_dur));
 
-    loop {
-        tokio::select! {
-            frame_result = frame_reader.next_frame() => {
-                match frame_result {
-                    Ok(Some(frame)) => {
-                        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;
-                                    }
-                                };
+    // Single-owner I/O engine — no tokio::io::split, no mutex
+    let mut tunnel_io = remoteingress_protocol::TunnelIo::new(tls_stream, Vec::new());
 
-                                // Payload is PROXY v1 header line
-                                let proxy_header = String::from_utf8_lossy(&frame.payload).to_string();
+    let mut disconnect_reason = "unknown".to_string();
 
-                                // Parse destination port from PROXY header
-                                let dest_port = parse_dest_port_from_proxy(&proxy_header).unwrap_or(443);
+    '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,
+                });
+            }
+        }
 
-                                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 = frame_writer_tx.clone();
-                                let target = target_host.clone();
-                                let stream_token = edge_token.child_token();
+        // 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;
+            }
+        }
 
-                                let _ = event_tx.try_send(HubEvent::StreamOpened {
-                                    edge_id: edge_id.clone(),
-                                    stream_id,
-                                });
+        // 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;
 
-                                // Create channel for data from edge to this stream
-                                let (data_tx, mut data_rx) = mpsc::channel::<Vec<u8>>(256);
-                                {
-                                    let mut s = streams.lock().await;
-                                    s.insert(stream_id, (data_tx, stream_token.clone()));
-                                }
-
-                                // Spawn task: connect to SmartProxy, send PROXY header, pipe data
-                                tokio::spawn(async move {
-                                    let _permit = permit; // hold semaphore permit until stream completes
-
-                                    let result = async {
-                                        // A2: Connect to SmartProxy with timeout
-                                        let mut upstream = tokio::time::timeout(
-                                            std::time::Duration::from_secs(10),
-                                            TcpStream::connect((target.as_str(), dest_port)),
-                                        )
-                                        .await
-                                        .map_err(|_| -> Box<dyn std::error::Error + Send + Sync> {
-                                            format!("connect to SmartProxy {}:{} timed out (10s)", target, dest_port).into()
-                                        })??;
-
-                                        upstream.write_all(proxy_header.as_bytes()).await?;
-
-                                        let (mut up_read, mut up_write) =
-                                            upstream.into_split();
-
-                                        // Forward data from edge (via channel) to SmartProxy
-                                        let writer_token = stream_token.clone();
-                                        let writer_for_edge_data = tokio::spawn(async move {
-                                            loop {
-                                                tokio::select! {
-                                                    data = data_rx.recv() => {
-                                                        match data {
-                                                            Some(data) => {
-                                                                if up_write.write_all(&data).await.is_err() {
-                                                                    break;
-                                                                }
-                                                            }
-                                                            None => break,
-                                                        }
-                                                    }
-                                                    _ = writer_token.cancelled() => break,
-                                                }
-                                            }
-                                            let _ = up_write.shutdown().await;
-                                        });
-
-                                        // Forward data from SmartProxy back to edge via writer channel
-                                        let mut buf = vec![0u8; 32768];
-                                        loop {
-                                            tokio::select! {
-                                                read_result = up_read.read(&mut buf) => {
-                                                    match read_result {
-                                                        Ok(0) => break,
-                                                        Ok(n) => {
-                                                            let frame =
-                                                                encode_frame(stream_id, FRAME_DATA_BACK, &buf[..n]);
-                                                            // A5: Use try_send to avoid blocking if writer channel is full
-                                                            if writer_tx.try_send(frame).is_err() {
-                                                                log::warn!("Stream {} writer channel full, closing", stream_id);
-                                                                break;
-                                                            }
-                                                        }
-                                                        Err(_) => break,
-                                                    }
-                                                }
-                                                _ = stream_token.cancelled() => break,
-                                            }
-                                        }
-
-                                        // Send CLOSE_BACK to edge (only if not cancelled)
-                                        if !stream_token.is_cancelled() {
-                                            let close_frame = encode_frame(stream_id, FRAME_CLOSE_BACK, &[]);
-                                            let _ = writer_tx.try_send(close_frame);
-                                        }
-
-                                        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 on error (only if not cancelled)
-                                        if !stream_token.is_cancelled() {
-                                            let close_frame = encode_frame(stream_id, FRAME_CLOSE_BACK, &[]);
-                                            let _ = writer_tx.try_send(close_frame);
-                                        }
-                                    }
-
-                                    // 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,
-                                        });
-                                    }
-                                });
-                            }
-                            FRAME_DATA => {
-                                // A1: Non-blocking send to prevent head-of-line blocking
-                                let s = streams.lock().await;
-                                if let Some((tx, _)) = s.get(&frame.stream_id) {
-                                    if tx.try_send(frame.payload).is_err() {
-                                        log::warn!("Stream {} data channel full, dropping frame", frame.stream_id);
-                                    }
-                                }
-                            }
-                            FRAME_CLOSE => {
-                                let mut s = streams.lock().await;
-                                if let Some((_, token)) = s.remove(&frame.stream_id) {
-                                    token.cancel();
-                                    let _ = event_tx.try_send(HubEvent::StreamClosed {
-                                        edge_id: edge_id.clone(),
-                                        stream_id: frame.stream_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;
                 }
             }
-            _ = edge_token.cancelled() => {
+            remoteingress_protocol::TunnelEvent::Eof => {
+                log::info!("Edge {} disconnected (EOF)", edge_id);
+                disconnect_reason = "edge_eof".to_string();
+                break;
+            }
+            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;
+            }
+            remoteingress_protocol::TunnelEvent::Cancelled => {
                 log::info!("Edge {} cancelled by hub", edge_id);
+                disconnect_reason = "cancelled_by_hub".to_string();
                 break;
             }
         }
@@ -596,13 +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(())
@@ -825,10 +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

@@ -4,4 +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;
@@ -7,6 +11,10 @@ pub const FRAME_CLOSE: u8 = 0x03;
 pub const FRAME_DATA_BACK: u8 = 0x04;
 pub const FRAME_CLOSE_BACK: u8 = 0x05;
 pub const FRAME_CONFIG: u8 = 0x06;     // Hub -> Edge: configuration update
+pub const FRAME_PING: u8 = 0x07;       // Hub -> Edge: heartbeat probe
+pub const FRAME_PONG: u8 = 0x08;       // Edge -> Hub: heartbeat response
+pub const FRAME_WINDOW_UPDATE: u8 = 0x09;      // Edge -> Hub: per-stream flow control
+pub const FRAME_WINDOW_UPDATE_BACK: u8 = 0x0A;  // Hub -> Edge: per-stream flow control
 
 // Frame header size: 4 (stream_id) + 1 (type) + 4 (length) = 9 bytes
 pub const FRAME_HEADER_SIZE: usize = 9;
@@ -14,6 +22,37 @@ pub const FRAME_HEADER_SIZE: usize = 9;
 // Maximum payload size (16 MB)
 pub const MAX_PAYLOAD_SIZE: u32 = 16 * 1024 * 1024;
 
+// Per-stream flow control constants
+/// Initial per-stream window size (4 MB). Sized for full throughput at high RTT:
+/// at 100ms RTT, this sustains ~40 MB/s per stream.
+pub const INITIAL_STREAM_WINDOW: u32 = 4 * 1024 * 1024;
+/// Send WINDOW_UPDATE after consuming this many bytes (half the initial window).
+pub const WINDOW_UPDATE_THRESHOLD: u32 = INITIAL_STREAM_WINDOW / 2;
+/// Maximum window size to prevent overflow.
+pub const MAX_WINDOW_SIZE: u32 = 16 * 1024 * 1024;
+
+/// Encode a WINDOW_UPDATE frame for a specific stream.
+pub fn encode_window_update(stream_id: u32, frame_type: u8, increment: u32) -> Vec<u8> {
+    encode_frame(stream_id, frame_type, &increment.to_be_bytes())
+}
+
+/// Compute the target per-stream window size based on the number of active streams.
+/// Total memory budget is ~32MB shared across all streams. As more streams are active,
+/// each gets a smaller window. This adapts to current demand — few streams get high
+/// throughput, many streams save memory and reduce control frame pressure.
+pub fn compute_window_for_stream_count(active: u32) -> u32 {
+    let per_stream = (32 * 1024 * 1024u64) / (active.max(1) as u64);
+    per_stream.clamp(64 * 1024, INITIAL_STREAM_WINDOW as u64) as u32
+}
+
+/// Decode a WINDOW_UPDATE payload into a byte increment. Returns None if payload is malformed.
+pub fn decode_window_update(payload: &[u8]) -> Option<u32> {
+    if payload.len() != 4 {
+        return None;
+    }
+    Some(u32::from_be_bytes([payload[0], payload[1], payload[2], payload[3]]))
+}
+
 /// A single multiplexed frame.
 #[derive(Debug, Clone)]
 pub struct Frame {
@@ -33,6 +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(
@@ -85,9 +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),
             ));
         }
 
@@ -109,10 +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";
@@ -261,6 +607,8 @@ mod tests {
             FRAME_DATA_BACK,
             FRAME_CLOSE_BACK,
             FRAME_CONFIG,
+            FRAME_PING,
+            FRAME_PONG,
         ];
 
         let mut data = Vec::new();
@@ -293,4 +641,149 @@ mod tests {
         assert_eq!(frame.frame_type, FRAME_CLOSE);
         assert!(frame.payload.is_empty());
     }
+
+    #[test]
+    fn test_encode_frame_ping_pong() {
+        // PING: stream_id=0, empty payload (control frame)
+        let ping = encode_frame(0, FRAME_PING, &[]);
+        assert_eq!(ping[4], FRAME_PING);
+        assert_eq!(&ping[0..4], &0u32.to_be_bytes());
+        assert_eq!(ping.len(), FRAME_HEADER_SIZE);
+
+        // PONG: stream_id=0, empty payload (control frame)
+        let pong = encode_frame(0, FRAME_PONG, &[]);
+        assert_eq!(pong[4], FRAME_PONG);
+        assert_eq!(&pong[0..4], &0u32.to_be_bytes());
+        assert_eq!(pong.len(), FRAME_HEADER_SIZE);
+    }
+
+    // --- compute_window_for_stream_count tests ---
+
+    #[test]
+    fn test_adaptive_window_zero_streams() {
+        // 0 streams treated as 1: 32MB/1 = 32MB → clamped to 4MB max
+        assert_eq!(compute_window_for_stream_count(0), INITIAL_STREAM_WINDOW);
+    }
+
+    #[test]
+    fn test_adaptive_window_one_stream() {
+        // 32MB/1 = 32MB → clamped to 4MB max
+        assert_eq!(compute_window_for_stream_count(1), INITIAL_STREAM_WINDOW);
+    }
+
+    #[test]
+    fn test_adaptive_window_at_max_boundary() {
+        // 32MB/8 = 4MB = exactly INITIAL_STREAM_WINDOW
+        assert_eq!(compute_window_for_stream_count(8), INITIAL_STREAM_WINDOW);
+    }
+
+    #[test]
+    fn test_adaptive_window_just_below_max() {
+        // 32MB/9 = 3,728,270 — first value below INITIAL_STREAM_WINDOW
+        let w = compute_window_for_stream_count(9);
+        assert!(w < INITIAL_STREAM_WINDOW);
+        assert_eq!(w, (32 * 1024 * 1024u64 / 9) as u32);
+    }
+
+    #[test]
+    fn test_adaptive_window_16_streams() {
+        // 32MB/16 = 2MB
+        assert_eq!(compute_window_for_stream_count(16), 2 * 1024 * 1024);
+    }
+
+    #[test]
+    fn test_adaptive_window_100_streams() {
+        // 32MB/100 = 335,544 bytes (~327KB)
+        let w = compute_window_for_stream_count(100);
+        assert_eq!(w, (32 * 1024 * 1024u64 / 100) as u32);
+        assert!(w > 64 * 1024); // above floor
+        assert!(w < INITIAL_STREAM_WINDOW as u32); // below ceiling
+    }
+
+    #[test]
+    fn test_adaptive_window_200_streams() {
+        // 32MB/200 = 167,772 bytes (~163KB), above 64KB floor
+        let w = compute_window_for_stream_count(200);
+        assert_eq!(w, (32 * 1024 * 1024u64 / 200) as u32);
+        assert!(w > 64 * 1024);
+    }
+
+    #[test]
+    fn test_adaptive_window_500_streams() {
+        // 32MB/500 = 67,108 bytes (~65.5KB), just above 64KB floor
+        let w = compute_window_for_stream_count(500);
+        assert_eq!(w, (32 * 1024 * 1024u64 / 500) as u32);
+        assert!(w > 64 * 1024);
+    }
+
+    #[test]
+    fn test_adaptive_window_at_min_boundary() {
+        // 32MB/512 = 65,536 = exactly 64KB floor
+        assert_eq!(compute_window_for_stream_count(512), 64 * 1024);
+    }
+
+    #[test]
+    fn test_adaptive_window_below_min_clamped() {
+        // 32MB/513 = 65,408 → clamped up to 64KB
+        assert_eq!(compute_window_for_stream_count(513), 64 * 1024);
+    }
+
+    #[test]
+    fn test_adaptive_window_1000_streams() {
+        // 32MB/1000 = 33,554 → clamped to 64KB
+        assert_eq!(compute_window_for_stream_count(1000), 64 * 1024);
+    }
+
+    #[test]
+    fn test_adaptive_window_max_u32() {
+        // Extreme: u32::MAX streams → tiny value → clamped to 64KB
+        assert_eq!(compute_window_for_stream_count(u32::MAX), 64 * 1024);
+    }
+
+    #[test]
+    fn test_adaptive_window_monotonically_decreasing() {
+        // Window should decrease (or stay same) as stream count increases
+        let mut prev = compute_window_for_stream_count(1);
+        for n in [2, 5, 10, 50, 100, 200, 500, 512, 1000] {
+            let w = compute_window_for_stream_count(n);
+            assert!(w <= prev, "window increased from {} to {} at n={}", prev, w, n);
+            prev = w;
+        }
+    }
+
+    #[test]
+    fn test_adaptive_window_total_budget_bounded() {
+        // active × per_stream_window should never exceed 32MB (+ clamp overhead for high N)
+        for n in [1, 10, 50, 100, 200, 500] {
+            let w = compute_window_for_stream_count(n);
+            let total = w as u64 * n as u64;
+            assert!(total <= 32 * 1024 * 1024, "total {}MB exceeds budget at n={}", total / (1024*1024), n);
+        }
+    }
+
+    // --- encode/decode window_update roundtrip ---
+
+    #[test]
+    fn test_window_update_roundtrip() {
+        for &increment in &[0u32, 1, 64 * 1024, INITIAL_STREAM_WINDOW, MAX_WINDOW_SIZE, u32::MAX] {
+            let frame = encode_window_update(42, FRAME_WINDOW_UPDATE, increment);
+            assert_eq!(frame[4], FRAME_WINDOW_UPDATE);
+            let decoded = decode_window_update(&frame[FRAME_HEADER_SIZE..]);
+            assert_eq!(decoded, Some(increment));
+        }
+    }
+
+    #[test]
+    fn test_window_update_back_roundtrip() {
+        let frame = encode_window_update(7, FRAME_WINDOW_UPDATE_BACK, 1234567);
+        assert_eq!(frame[4], FRAME_WINDOW_UPDATE_BACK);
+        assert_eq!(decode_window_update(&frame[FRAME_HEADER_SIZE..]), Some(1234567));
+    }
+
+    #[test]
+    fn test_decode_window_update_malformed() {
+        assert_eq!(decode_window_update(&[]), None);
+        assert_eq!(decode_window_update(&[0, 0, 0]), None);
+        assert_eq!(decode_window_update(&[0, 0, 0, 0, 0]), None);
+    }
 }

test/test.flowcontrol.node.ts

@@ -0,0 +1,475 @@
+import { expect, tap } from '@push.rocks/tapbundle';
+import * as net from 'net';
+import * as crypto from 'crypto';
+import { RemoteIngressHub, RemoteIngressEdge } from '../ts/index.js';
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+/** Find N free ports by binding to port 0 and collecting OS-assigned ports. */
+async function findFreePorts(count: number): Promise<number[]> {
+  const servers: net.Server[] = [];
+  const ports: number[] = [];
+  for (let i = 0; i < count; i++) {
+    const server = net.createServer();
+    await new Promise<void>((resolve) => server.listen(0, '127.0.0.1', resolve));
+    ports.push((server.address() as net.AddressInfo).port);
+    servers.push(server);
+  }
+  await Promise.all(servers.map((s) => new Promise<void>((resolve) => s.close(() => resolve()))));
+  return ports;
+}
+
+type TrackingServer = net.Server & { destroyAll: () => void };
+
+/** Start a TCP echo server that tracks connections for force-close. */
+function startEchoServer(port: number, host: string): Promise<TrackingServer> {
+  return new Promise((resolve, reject) => {
+    const connections = new Set<net.Socket>();
+    const server = net.createServer((socket) => {
+      connections.add(socket);
+      socket.on('close', () => connections.delete(socket));
+
+      // Skip PROXY protocol v1 header line before echoing
+      let proxyHeaderParsed = false;
+      let pendingBuf = Buffer.alloc(0);
+      socket.on('data', (data: Buffer) => {
+        if (!proxyHeaderParsed) {
+          pendingBuf = Buffer.concat([pendingBuf, data]);
+          const idx = pendingBuf.indexOf('\r\n');
+          if (idx !== -1) {
+            proxyHeaderParsed = true;
+            const remainder = pendingBuf.subarray(idx + 2);
+            if (remainder.length > 0) {
+              socket.write(remainder);
+            }
+          }
+          return;
+        }
+        socket.write(data);
+      });
+      socket.on('error', () => {});
+    }) as TrackingServer;
+
+    server.destroyAll = () => {
+      for (const conn of connections) conn.destroy();
+      connections.clear();
+    };
+
+    server.on('error', reject);
+    server.listen(port, host, () => resolve(server));
+  });
+}
+
+/**
+ * Start a server that sends a large response immediately on first data received.
+ * Does NOT wait for end (the tunnel protocol has no half-close).
+ * On receiving first data chunk after PROXY header, sends responseSize bytes then closes.
+ */
+function startLargeResponseServer(port: number, host: string, responseSize: number): Promise<TrackingServer> {
+  return new Promise((resolve, reject) => {
+    const connections = new Set<net.Socket>();
+    const server = net.createServer((socket) => {
+      connections.add(socket);
+      socket.on('close', () => connections.delete(socket));
+
+      let proxyHeaderParsed = false;
+      let pendingBuf = Buffer.alloc(0);
+      let responseSent = false;
+
+      socket.on('data', (data: Buffer) => {
+        if (!proxyHeaderParsed) {
+          pendingBuf = Buffer.concat([pendingBuf, data]);
+          const idx = pendingBuf.indexOf('\r\n');
+          if (idx !== -1) {
+            proxyHeaderParsed = true;
+            const remainder = pendingBuf.subarray(idx + 2);
+            if (remainder.length > 0 && !responseSent) {
+              responseSent = true;
+              sendLargeResponse(socket, responseSize);
+            }
+          }
+          return;
+        }
+        if (!responseSent) {
+          responseSent = true;
+          sendLargeResponse(socket, responseSize);
+        }
+      });
+      socket.on('error', () => {});
+    }) as TrackingServer;
+
+    server.destroyAll = () => {
+      for (const conn of connections) conn.destroy();
+      connections.clear();
+    };
+
+    server.on('error', reject);
+    server.listen(port, host, () => resolve(server));
+  });
+}
+
+function sendLargeResponse(socket: net.Socket, totalBytes: number) {
+  const chunkSize = 32 * 1024;
+  let sent = 0;
+  const writeChunk = () => {
+    while (sent < totalBytes) {
+      const toWrite = Math.min(chunkSize, totalBytes - sent);
+      // Use a deterministic pattern for verification
+      const chunk = Buffer.alloc(toWrite, (sent % 256) & 0xff);
+      const canContinue = socket.write(chunk);
+      sent += toWrite;
+      if (!canContinue) {
+        socket.once('drain', writeChunk);
+        return;
+      }
+    }
+    socket.end();
+  };
+  writeChunk();
+}
+
+/** Force-close a server: destroy all connections, then close. */
+async function forceCloseServer(server: TrackingServer): Promise<void> {
+  server.destroyAll();
+  await new Promise<void>((resolve) => server.close(() => resolve()));
+}
+
+interface TestTunnel {
+  hub: RemoteIngressHub;
+  edge: RemoteIngressEdge;
+  edgePort: number;
+  cleanup: () => Promise<void>;
+}
+
+/**
+ * Start a full hub + edge tunnel.
+ * Edge binds to 127.0.0.1, upstream server binds to 127.0.0.2.
+ * Hub targetHost = 127.0.0.2 so hub -> upstream doesn't loop back to edge.
+ */
+async function startTunnel(edgePort: number, hubPort: number): Promise<TestTunnel> {
+  const hub = new RemoteIngressHub();
+  const edge = new RemoteIngressEdge();
+
+  await hub.start({
+    tunnelPort: hubPort,
+    targetHost: '127.0.0.2',
+  });
+
+  await hub.updateAllowedEdges([
+    { id: 'test-edge', secret: 'test-secret', listenPorts: [edgePort] },
+  ]);
+
+  const connectedPromise = new Promise<void>((resolve, reject) => {
+    const timeout = setTimeout(() => reject(new Error('Edge did not connect within 10s')), 10000);
+    edge.once('tunnelConnected', () => {
+      clearTimeout(timeout);
+      resolve();
+    });
+  });
+
+  await edge.start({
+    hubHost: '127.0.0.1',
+    hubPort,
+    edgeId: 'test-edge',
+    secret: 'test-secret',
+    bindAddress: '127.0.0.1',
+  });
+
+  await connectedPromise;
+  await new Promise((resolve) => setTimeout(resolve, 500));
+
+  return {
+    hub,
+    edge,
+    edgePort,
+    cleanup: async () => {
+      await edge.stop();
+      await hub.stop();
+    },
+  };
+}
+
+/**
+ * Send data through the tunnel and collect the echoed response.
+ */
+function sendAndReceive(port: number, data: Buffer, timeoutMs = 30000): Promise<Buffer> {
+  return new Promise((resolve, reject) => {
+    const chunks: Buffer[] = [];
+    let totalReceived = 0;
+    const expectedLength = data.length;
+    let settled = false;
+
+    const client = net.createConnection({ host: '127.0.0.1', port }, () => {
+      client.write(data);
+      client.end();
+    });
+
+    const timer = setTimeout(() => {
+      if (!settled) {
+        settled = true;
+        client.destroy();
+        reject(new Error(`Timeout after ${timeoutMs}ms — received ${totalReceived}/${expectedLength} bytes`));
+      }
+    }, timeoutMs);
+
+    client.on('data', (chunk: Buffer) => {
+      chunks.push(chunk);
+      totalReceived += chunk.length;
+      if (totalReceived >= expectedLength && !settled) {
+        settled = true;
+        clearTimeout(timer);
+        client.destroy();
+        resolve(Buffer.concat(chunks));
+      }
+    });
+
+    client.on('end', () => {
+      if (!settled) {
+        settled = true;
+        clearTimeout(timer);
+        resolve(Buffer.concat(chunks));
+      }
+    });
+
+    client.on('error', (err) => {
+      if (!settled) {
+        settled = true;
+        clearTimeout(timer);
+        reject(err);
+      }
+    });
+  });
+}
+
+/**
+ * Connect to the tunnel, send a small request, and collect a large response.
+ * Does NOT call end() — the tunnel has no half-close.
+ * Instead, collects until expectedResponseSize bytes arrive.
+ */
+function sendAndReceiveLarge(
+  port: number,
+  data: Buffer,
+  expectedResponseSize: number,
+  timeoutMs = 60000,
+): Promise<Buffer> {
+  return new Promise((resolve, reject) => {
+    const chunks: Buffer[] = [];
+    let totalReceived = 0;
+    let settled = false;
+
+    const client = net.createConnection({ host: '127.0.0.1', port }, () => {
+      client.write(data);
+      // Do NOT call client.end() — the server will respond immediately
+      // and the tunnel CLOSE will happen when the download finishes
+    });
+
+    const timer = setTimeout(() => {
+      if (!settled) {
+        settled = true;
+        client.destroy();
+        reject(new Error(`Timeout after ${timeoutMs}ms — received ${totalReceived}/${expectedResponseSize} bytes`));
+      }
+    }, timeoutMs);
+
+    client.on('data', (chunk: Buffer) => {
+      chunks.push(chunk);
+      totalReceived += chunk.length;
+      if (totalReceived >= expectedResponseSize && !settled) {
+        settled = true;
+        clearTimeout(timer);
+        client.destroy();
+        resolve(Buffer.concat(chunks));
+      }
+    });
+
+    client.on('end', () => {
+      if (!settled) {
+        settled = true;
+        clearTimeout(timer);
+        resolve(Buffer.concat(chunks));
+      }
+    });
+
+    client.on('error', (err) => {
+      if (!settled) {
+        settled = true;
+        clearTimeout(timer);
+        reject(err);
+      }
+    });
+  });
+}
+
+function sha256(buf: Buffer): string {
+  return crypto.createHash('sha256').update(buf).digest('hex');
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+let tunnel: TestTunnel;
+let echoServer: TrackingServer;
+let hubPort: number;
+let edgePort: number;
+
+tap.test('setup: start echo server and tunnel', async () => {
+  [hubPort, edgePort] = await findFreePorts(2);
+
+  echoServer = await startEchoServer(edgePort, '127.0.0.2');
+  tunnel = await startTunnel(edgePort, hubPort);
+
+  expect(tunnel.hub.running).toBeTrue();
+});
+
+tap.test('single stream: 32MB transfer exceeding initial 4MB window', async () => {
+  const size = 32 * 1024 * 1024;
+  const data = crypto.randomBytes(size);
+  const expectedHash = sha256(data);
+
+  const received = await sendAndReceive(edgePort, data, 60000);
+
+  expect(received.length).toEqual(size);
+  expect(sha256(received)).toEqual(expectedHash);
+});
+
+tap.test('200 concurrent streams with 64KB each', async () => {
+  const streamCount = 200;
+  const payloadSize = 64 * 1024;
+
+  const promises = Array.from({ length: streamCount }, () => {
+    const data = crypto.randomBytes(payloadSize);
+    const hash = sha256(data);
+    return sendAndReceive(edgePort, data, 30000).then((received) => ({
+      sent: hash,
+      received: sha256(received),
+      sizeOk: received.length === payloadSize,
+    }));
+  });
+
+  const results = await Promise.all(promises);
+  const failures = results.filter((r) => !r.sizeOk || r.sent !== r.received);
+
+  expect(failures.length).toEqual(0);
+});
+
+tap.test('512 concurrent streams at minimum window boundary (16KB each)', async () => {
+  const streamCount = 512;
+  const payloadSize = 16 * 1024;
+
+  const promises = Array.from({ length: streamCount }, () => {
+    const data = crypto.randomBytes(payloadSize);
+    const hash = sha256(data);
+    return sendAndReceive(edgePort, data, 60000).then((received) => ({
+      sent: hash,
+      received: sha256(received),
+      sizeOk: received.length === payloadSize,
+    }));
+  });
+
+  const results = await Promise.all(promises);
+  const failures = results.filter((r) => !r.sizeOk || r.sent !== r.received);
+
+  expect(failures.length).toEqual(0);
+});
+
+tap.test('asymmetric transfer: 4KB request -> 4MB response', async () => {
+  // Swap to large-response server
+  await forceCloseServer(echoServer);
+  const responseSize = 4 * 1024 * 1024; // 4 MB
+  const largeServer = await startLargeResponseServer(edgePort, '127.0.0.2', responseSize);
+
+  try {
+    const requestData = crypto.randomBytes(4 * 1024); // 4 KB
+    const received = await sendAndReceiveLarge(edgePort, requestData, responseSize, 60000);
+    expect(received.length).toEqual(responseSize);
+  } finally {
+    // Always restore echo server even on failure
+    await forceCloseServer(largeServer);
+    echoServer = await startEchoServer(edgePort, '127.0.0.2');
+  }
+});
+
+tap.test('100 streams x 1MB each (100MB total exceeding 32MB budget)', async () => {
+  const streamCount = 100;
+  const payloadSize = 1 * 1024 * 1024;
+
+  const promises = Array.from({ length: streamCount }, () => {
+    const data = crypto.randomBytes(payloadSize);
+    const hash = sha256(data);
+    return sendAndReceive(edgePort, data, 120000).then((received) => ({
+      sent: hash,
+      received: sha256(received),
+      sizeOk: received.length === payloadSize,
+    }));
+  });
+
+  const results = await Promise.all(promises);
+  const failures = results.filter((r) => !r.sizeOk || r.sent !== r.received);
+
+  expect(failures.length).toEqual(0);
+});
+
+tap.test('active stream counter tracks concurrent connections', async () => {
+  const N = 50;
+
+  // Open N connections and keep them alive (send data but don't close)
+  const sockets: net.Socket[] = [];
+  const connectPromises = Array.from({ length: N }, () => {
+    return new Promise<net.Socket>((resolve, reject) => {
+      const sock = net.createConnection({ host: '127.0.0.1', port: edgePort }, () => {
+        resolve(sock);
+      });
+      sock.on('error', () => {});
+      setTimeout(() => reject(new Error('connect timeout')), 5000);
+    });
+  });
+
+  const connected = await Promise.all(connectPromises);
+  sockets.push(...connected);
+
+  // Brief delay for stream registration to propagate
+  await new Promise((resolve) => setTimeout(resolve, 500));
+
+  // Verify the edge reports >= N active streams.
+  // This counter is the input to compute_window_for_stream_count(),
+  // so its accuracy determines whether adaptive window sizing is correct.
+  const status = await tunnel.edge.getStatus();
+  expect(status.activeStreams).toBeGreaterThanOrEqual(N);
+
+  // Clean up: destroy all sockets (the tunnel's 300s stream timeout will handle cleanup)
+  for (const sock of sockets) {
+    sock.destroy();
+  }
+});
+
+tap.test('50 streams x 2MB each (forces multiple window refills per stream)', async () => {
+  // At 50 concurrent streams: adaptive window = 32MB/50 = 655KB per stream
+  // Each stream sends 2MB → needs ~3 WINDOW_UPDATE refill cycles per stream
+  const streamCount = 50;
+  const payloadSize = 2 * 1024 * 1024;
+
+  const promises = Array.from({ length: streamCount }, () => {
+    const data = crypto.randomBytes(payloadSize);
+    const hash = sha256(data);
+    return sendAndReceive(edgePort, data, 120000).then((received) => ({
+      sent: hash,
+      received: sha256(received),
+      sizeOk: received.length === payloadSize,
+    }));
+  });
+
+  const results = await Promise.all(promises);
+  const failures = results.filter((r) => !r.sizeOk || r.sent !== r.received);
+
+  expect(failures.length).toEqual(0);
+});
+
+tap.test('teardown: stop tunnel and echo server', async () => {
+  await tunnel.cleanup();
+  await forceCloseServer(echoServer);
+});
+
+export default tap.start();

ts/00_commitinfo_data.ts

@@ -3,6 +3,6 @@
  */
 export const commitinfo = {
   name: '@serve.zone/remoteingress',
-  version: '4.2.0',
+  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

@@ -14,6 +14,7 @@ type TEdgeCommands = {
       hubPort: number;
       edgeId: string;
       secret: string;
+      bindAddress?: string;
     };
     result: { started: boolean };
   };
@@ -38,11 +39,19 @@ export interface IEdgeConfig {
   hubPort?: number;
   edgeId: string;
   secret: string;
+  bindAddress?: string;
 }
 
+const MAX_RESTART_ATTEMPTS = 10;
+const MAX_RESTART_BACKOFF_MS = 30_000;
+
 export class RemoteIngressEdge extends EventEmitter {
   private bridge: InstanceType<typeof plugins.smartrust.RustBridge<TEdgeCommands>>;
   private started = false;
+  private stopping = false;
+  private savedConfig: IEdgeConfig | null = null;
+  private restartBackoffMs = 1000;
+  private restartAttempts = 0;
   private statusInterval: ReturnType<typeof setInterval> | undefined;
 
   constructor() {
@@ -74,8 +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);
@@ -109,19 +120,28 @@ export class RemoteIngressEdge extends EventEmitter {
       edgeConfig = config;
     }
 
+    this.savedConfig = edgeConfig;
+    this.stopping = false;
+
     const spawned = await this.bridge.spawn();
     if (!spawned) {
       throw new Error('Failed to spawn remoteingress-bin');
     }
 
+    // Register crash recovery handler
+    this.bridge.on('exit', this.handleCrashRecovery);
+
     await this.bridge.sendCommand('startEdge', {
       hubHost: edgeConfig.hubHost,
       hubPort: edgeConfig.hubPort ?? 8443,
       edgeId: edgeConfig.edgeId,
       secret: edgeConfig.secret,
+      ...(edgeConfig.bindAddress ? { bindAddress: edgeConfig.bindAddress } : {}),
     });
 
     this.started = true;
+    this.restartAttempts = 0;
+    this.restartBackoffMs = 1000;
 
     // Start periodic status logging
     this.statusInterval = setInterval(async () => {
@@ -142,6 +162,7 @@ export class RemoteIngressEdge extends EventEmitter {
    * Stop the edge and kill the Rust process.
    */
   public async stop(): Promise<void> {
+    this.stopping = true;
     if (this.statusInterval) {
       clearInterval(this.statusInterval);
       this.statusInterval = undefined;
@@ -152,6 +173,7 @@ export class RemoteIngressEdge extends EventEmitter {
       } catch {
         // Process may already be dead
       }
+      this.bridge.removeListener('exit', this.handleCrashRecovery);
       this.bridge.kill();
       this.started = false;
     }
@@ -170,4 +192,56 @@ export class RemoteIngressEdge extends EventEmitter {
   public get running(): boolean {
     return this.bridge.running;
   }
+
+  /**
+   * Handle unexpected Rust binary crash — auto-restart with backoff.
+   */
+  private handleCrashRecovery = async (code: number | null, signal: string | null) => {
+    if (this.stopping || !this.started || !this.savedConfig) {
+      return;
+    }
+
+    console.error(
+      `[RemoteIngressEdge] Rust binary crashed (code=${code}, signal=${signal}), ` +
+      `attempt ${this.restartAttempts + 1}/${MAX_RESTART_ATTEMPTS}`
+    );
+
+    this.started = false;
+
+    if (this.restartAttempts >= MAX_RESTART_ATTEMPTS) {
+      console.error('[RemoteIngressEdge] Max restart attempts reached, giving up');
+      this.emit('crashRecoveryFailed');
+      return;
+    }
+
+    await new Promise(resolve => setTimeout(resolve, this.restartBackoffMs));
+    this.restartBackoffMs = Math.min(this.restartBackoffMs * 2, MAX_RESTART_BACKOFF_MS);
+    this.restartAttempts++;
+
+    try {
+      const spawned = await this.bridge.spawn();
+      if (!spawned) {
+        console.error('[RemoteIngressEdge] Failed to respawn binary');
+        return;
+      }
+
+      this.bridge.on('exit', this.handleCrashRecovery);
+
+      await this.bridge.sendCommand('startEdge', {
+        hubHost: this.savedConfig.hubHost,
+        hubPort: this.savedConfig.hubPort ?? 8443,
+        edgeId: this.savedConfig.edgeId,
+        secret: this.savedConfig.secret,
+        ...(this.savedConfig.bindAddress ? { bindAddress: this.savedConfig.bindAddress } : {}),
+      });
+
+      this.started = true;
+      this.restartAttempts = 0;
+      this.restartBackoffMs = 1000;
+      console.log('[RemoteIngressEdge] Successfully recovered from crash');
+      this.emit('crashRecovered');
+    } catch (err) {
+      console.error(`[RemoteIngressEdge] Crash recovery failed: ${err}`);
+    }
+  };
 }

ts/classes.remoteingresshub.ts

@@ -11,6 +11,8 @@ type THubCommands = {
     params: {
       tunnelPort: number;
       targetHost?: string;
+      tlsCertPem?: string;
+      tlsKeyPem?: string;
     };
     result: { started: boolean };
   };
@@ -42,11 +44,25 @@ type THubCommands = {
 export interface IHubConfig {
   tunnelPort?: number;
   targetHost?: string;
+  tls?: {
+    certPem?: string;
+    keyPem?: string;
+  };
 }
 
+type TAllowedEdge = { id: string; secret: string; listenPorts?: number[]; stunIntervalSecs?: number };
+
+const MAX_RESTART_ATTEMPTS = 10;
+const MAX_RESTART_BACKOFF_MS = 30_000;
+
 export class RemoteIngressHub extends EventEmitter {
   private bridge: InstanceType<typeof plugins.smartrust.RustBridge<THubCommands>>;
   private started = false;
+  private stopping = false;
+  private savedConfig: IHubConfig | null = null;
+  private savedEdges: TAllowedEdge[] = [];
+  private restartBackoffMs = 1000;
+  private restartAttempts = 0;
 
   constructor() {
     super();
@@ -77,7 +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 }) => {
@@ -92,29 +110,42 @@ export class RemoteIngressHub extends EventEmitter {
    * Start the hub — spawns the Rust binary and starts the tunnel server.
    */
   public async start(config: IHubConfig = {}): Promise<void> {
+    this.savedConfig = config;
+    this.stopping = false;
+
     const spawned = await this.bridge.spawn();
     if (!spawned) {
       throw new Error('Failed to spawn remoteingress-bin');
     }
 
+    // Register crash recovery handler
+    this.bridge.on('exit', this.handleCrashRecovery);
+
     await this.bridge.sendCommand('startHub', {
       tunnelPort: config.tunnelPort ?? 8443,
       targetHost: config.targetHost ?? '127.0.0.1',
+      ...(config.tls?.certPem && config.tls?.keyPem
+        ? { tlsCertPem: config.tls.certPem, tlsKeyPem: config.tls.keyPem }
+        : {}),
     });
 
     this.started = true;
+    this.restartAttempts = 0;
+    this.restartBackoffMs = 1000;
   }
 
   /**
    * Stop the hub and kill the Rust process.
    */
   public async stop(): Promise<void> {
+    this.stopping = true;
     if (this.started) {
       try {
         await this.bridge.sendCommand('stopHub', {} as Record<string, never>);
       } catch {
         // Process may already be dead
       }
+      this.bridge.removeListener('exit', this.handleCrashRecovery);
       this.bridge.kill();
       this.started = false;
     }
@@ -123,7 +154,8 @@ export class RemoteIngressHub extends EventEmitter {
   /**
    * Update the list of allowed edges that can connect to this hub.
    */
-  public async updateAllowedEdges(edges: Array<{ id: string; secret: string; listenPorts?: number[]; stunIntervalSecs?: number }>): Promise<void> {
+  public async updateAllowedEdges(edges: TAllowedEdge[]): Promise<void> {
+    this.savedEdges = edges;
     await this.bridge.sendCommand('updateAllowedEdges', { edges });
   }
 
@@ -140,4 +172,62 @@ export class RemoteIngressHub extends EventEmitter {
   public get running(): boolean {
     return this.bridge.running;
   }
+
+  /**
+   * Handle unexpected Rust binary crash — auto-restart with backoff.
+   */
+  private handleCrashRecovery = async (code: number | null, signal: string | null) => {
+    if (this.stopping || !this.started || !this.savedConfig) {
+      return;
+    }
+
+    console.error(
+      `[RemoteIngressHub] Rust binary crashed (code=${code}, signal=${signal}), ` +
+      `attempt ${this.restartAttempts + 1}/${MAX_RESTART_ATTEMPTS}`
+    );
+
+    this.started = false;
+
+    if (this.restartAttempts >= MAX_RESTART_ATTEMPTS) {
+      console.error('[RemoteIngressHub] Max restart attempts reached, giving up');
+      this.emit('crashRecoveryFailed');
+      return;
+    }
+
+    await new Promise(resolve => setTimeout(resolve, this.restartBackoffMs));
+    this.restartBackoffMs = Math.min(this.restartBackoffMs * 2, MAX_RESTART_BACKOFF_MS);
+    this.restartAttempts++;
+
+    try {
+      const spawned = await this.bridge.spawn();
+      if (!spawned) {
+        console.error('[RemoteIngressHub] Failed to respawn binary');
+        return;
+      }
+
+      this.bridge.on('exit', this.handleCrashRecovery);
+
+      const config = this.savedConfig;
+      await this.bridge.sendCommand('startHub', {
+        tunnelPort: config.tunnelPort ?? 8443,
+        targetHost: config.targetHost ?? '127.0.0.1',
+        ...(config.tls?.certPem && config.tls?.keyPem
+          ? { tlsCertPem: config.tls.certPem, tlsKeyPem: config.tls.keyPem }
+          : {}),
+      });
+
+      // Restore allowed edges
+      if (this.savedEdges.length > 0) {
+        await this.bridge.sendCommand('updateAllowedEdges', { edges: this.savedEdges });
+      }
+
+      this.started = true;
+      this.restartAttempts = 0;
+      this.restartBackoffMs = 1000;
+      console.log('[RemoteIngressHub] Successfully recovered from crash');
+      this.emit('crashRecovered');
+    } catch (err) {
+      console.error(`[RemoteIngressHub] Crash recovery failed: ${err}`);
+    }
+  };
 }