v25.11.23

changelog.md

@@ -1,5 +1,81 @@
 # Changelog
 
+## 2026-03-17 - 25.11.23 - fix(rustproxy-http,rustproxy-metrics)
+reduce per-frame metrics overhead by batching body byte accounting
+
+- Buffer HTTP body byte counts and flush them every 64 KB, at end of stream, and on drop to keep totals accurate while preserving throughput sampling.
+- Skip zero-value counter updates in metrics collection to avoid unnecessary atomic and DashMap operations for the unused direction.
+
+## 2026-03-17 - 25.11.22 - fix(rustproxy-http)
+reuse healthy HTTP/2 upstream connections after requests with bodies
+
+- Registers successful HTTP/2 connections in the pool regardless of whether the proxied request included a body
+- Continues to avoid pooling upstream connections that returned 502 Bad Gateway responses
+
+## 2026-03-17 - 25.11.21 - fix(rustproxy-http)
+reuse pooled HTTP/2 connections for requests with and without bodies
+
+- remove the bodyless-request restriction from HTTP/2 pool checkout
+- always return successful HTTP/2 senders to the connection pool after requests
+
+## 2026-03-17 - 25.11.20 - fix(rustproxy-http)
+avoid downgrading cached backend protocol on H2 stream errors
+
+- Treat HTTP/2 stream-level failures as retryable request errors instead of evidence that the backend only supports HTTP/1.1
+- Keep protocol cache entries unchanged after successful H2 handshakes so future requests continue using HTTP/2
+- Lower log severity for this fallback path from warning to debug while still recording backend H2 failure metrics
+
+## 2026-03-16 - 25.11.19 - fix(rustproxy-http)
+avoid reusing pooled HTTP/2 connections for requests with bodies to prevent upload flow-control stalls
+
+- Limit HTTP/2 pool checkout to bodyless requests such as GET, HEAD, and DELETE
+- Skip re-registering HTTP/2 connections in the pool after requests that send a body
+- Prevent stalled uploads caused by depleted connection-level flow control windows on reused HTTP/2 connections
+
+## 2026-03-16 - 25.11.18 - fix(repo)
+no changes to commit
+
+
+## 2026-03-16 - 25.11.17 - fix(rustproxy-http)
+prevent stale HTTP/2 connection drivers from evicting newer pooled connections
+
+- add generation IDs to pooled HTTP/2 senders so pool removal only affects the matching connection
+- update HTTP/2 proxy and retry paths to register generation-tagged connections and skip eviction before registration completes
+
+## 2026-03-16 - 25.11.16 - fix(repo)
+no changes to commit
+
+
+## 2026-03-16 - 25.11.15 - fix(rustproxy-http)
+implement vectored write support for backend streams
+
+- Add poll_write_vectored forwarding for both plain and TLS backend stream variants
+- Expose is_write_vectored so the proxy can correctly report vectored write capability
+
+## 2026-03-16 - 25.11.14 - fix(rustproxy-http)
+forward vectored write support in ShutdownOnDrop AsyncWrite wrapper
+
+- Implements poll_write_vectored by delegating to the wrapped writer
+- Exposes is_write_vectored so the wrapper preserves underlying AsyncWrite capabilities
+
+## 2026-03-16 - 25.11.13 - fix(rustproxy-http)
+remove hot-path debug logging from HTTP/1 connection pool hits
+
+- Stops emitting debug logs when reusing HTTP/1 idle connections in the connection pool.
+- Keeps pool hit behavior unchanged while reducing overhead on a frequently executed path.
+
+## 2026-03-16 - 25.11.12 - fix(rustproxy-http)
+remove connection pool hit logging and keep logging limited to actual failures
+
+- Removes debug and warning logs for HTTP/2 connection pool hits and age checks.
+- Keeps pool behavior unchanged while reducing noisy per-request logging in the Rust HTTP proxy layer.
+
+## 2026-03-16 - 25.11.11 - fix(rustproxy-http)
+improve HTTP/2 proxy error logging with warning-level connection failures and debug error details
+
+- Adds debug-formatted error fields to HTTP/2 handshake, retry, fallback, and request failure logs
+- Promotes upstream HTTP/2 connection error logs from debug to warn to improve operational visibility
+
 ## 2026-03-16 - 25.11.10 - fix(rustproxy-http)
 validate pooled HTTP/2 connections asynchronously before reuse and evict stale senders
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@push.rocks/smartproxy",
-  "version": "25.11.10",
+  "version": "25.11.23",
   "private": false,
   "description": "A powerful proxy package with unified route-based configuration for high traffic management. Features include SSL/TLS support, flexible routing patterns, WebSocket handling, advanced security options, and automatic ACME certificate management.",
   "main": "dist_ts/index.js",

rust/crates/rustproxy-http/src/connection_pool.rs

@@ -4,13 +4,13 @@
 //! HTTP/2 connections are multiplexed (clone the sender for each request).
 
 use std::sync::Arc;
+use std::sync::atomic::{AtomicU64, Ordering};
 use std::time::{Duration, Instant};
 
 use bytes::Bytes;
 use dashmap::DashMap;
 use http_body_util::combinators::BoxBody;
 use hyper::client::conn::{http1, http2};
-use tracing::{debug, warn};
 
 /// Maximum idle connections per backend key.
 const MAX_IDLE_PER_KEY: usize = 16;
@@ -38,10 +38,13 @@ struct IdleH1 {
     idle_since: Instant,
 }
 
-/// A pooled HTTP/2 sender (multiplexed, Clone-able).
+/// A pooled HTTP/2 sender (multiplexed, Clone-able) with a generation tag.
 struct PooledH2 {
     sender: http2::SendRequest<BoxBody<Bytes, hyper::Error>>,
     created_at: Instant,
+    /// Unique generation ID. Connection drivers use this to only remove their OWN
+    /// entry, preventing phantom eviction when multiple connections share the same key.
+    generation: u64,
 }
 
 /// Backend connection pool.
@@ -50,6 +53,8 @@ pub struct ConnectionPool {
     h1_pool: Arc<DashMap<PoolKey, Vec<IdleH1>>>,
     /// HTTP/2 multiplexed connections indexed by backend key.
     h2_pool: Arc<DashMap<PoolKey, PooledH2>>,
+    /// Monotonic generation counter for H2 pool entries.
+    h2_generation: AtomicU64,
     /// Handle for the background eviction task.
     eviction_handle: Option<tokio::task::JoinHandle<()>>,
 }
@@ -69,6 +74,7 @@ impl ConnectionPool {
         Self {
             h1_pool,
             h2_pool,
+            h2_generation: AtomicU64::new(0),
             eviction_handle: Some(eviction_handle),
         }
     }
@@ -82,7 +88,7 @@ impl ConnectionPool {
         while let Some(idle) = idles.pop() {
             // Check if the connection is still alive and ready
             if idle.idle_since.elapsed() < IDLE_TIMEOUT && idle.sender.is_ready() && !idle.sender.is_closed() {
-                debug!("Pool hit (h1): {}:{}", key.host, key.port);
+                // H1 pool hit — no logging on hot path
                 return Some(idle.sender);
             }
             // Stale or closed — drop it
@@ -120,42 +126,51 @@ impl ConnectionPool {
         let pooled = entry.value();
         let age = pooled.created_at.elapsed();
 
-        // Check if the h2 connection is still alive and not too old
         if pooled.sender.is_closed() || age >= MAX_H2_AGE {
-            let reason = if pooled.sender.is_closed() { "closed" } else { "max_age" };
-            debug!("Pool evict (h2): {}:{} (reason={}, age={:.1}s)", key.host, key.port, reason, age.as_secs_f64());
             drop(entry);
             self.h2_pool.remove(key);
             return None;
         }
 
         if pooled.sender.is_ready() {
-            if age > Duration::from_secs(30) {
-                warn!("Pool hit (h2): {}:{} — connection age {:.1}s (>30s, may be stale)", key.host, key.port, age.as_secs_f64());
-            } else {
-                debug!("Pool hit (h2): {}:{} (age={:.1}s)", key.host, key.port, age.as_secs_f64());
-            }
             return Some((pooled.sender.clone(), age));
         }
         None
     }
 
-    /// Remove a dead HTTP/2 sender from the pool.
+    /// Remove a dead HTTP/2 sender from the pool (unconditional).
     /// Called when `send_request` fails to prevent subsequent requests from reusing the stale sender.
     pub fn remove_h2(&self, key: &PoolKey) {
         self.h2_pool.remove(key);
     }
 
-    /// Register an HTTP/2 sender in the pool. Since h2 is multiplexed,
-    /// only one sender per key is stored (it's Clone-able).
-    pub fn register_h2(&self, key: PoolKey, sender: http2::SendRequest<BoxBody<Bytes, hyper::Error>>) {
+    /// Remove an HTTP/2 sender ONLY if the current entry has the expected generation.
+    /// This prevents phantom eviction: when multiple connections share the same key,
+    /// an old connection's driver won't accidentally remove a newer connection's entry.
+    pub fn remove_h2_if_generation(&self, key: &PoolKey, expected_gen: u64) {
+        if let Some(entry) = self.h2_pool.get(key) {
+            if entry.value().generation == expected_gen {
+                drop(entry); // release DashMap ref before remove
+                self.h2_pool.remove(key);
+            }
+            // else: a newer connection replaced ours — don't touch it
+        }
+    }
+
+    /// Register an HTTP/2 sender in the pool. Returns the generation ID for this entry.
+    /// The caller should pass this generation to the connection driver so it can use
+    /// `remove_h2_if_generation` instead of `remove_h2` to avoid phantom eviction.
+    pub fn register_h2(&self, key: PoolKey, sender: http2::SendRequest<BoxBody<Bytes, hyper::Error>>) -> u64 {
+        let gen = self.h2_generation.fetch_add(1, Ordering::Relaxed);
         if sender.is_closed() {
-            return;
+            return gen;
         }
         self.h2_pool.insert(key, PooledH2 {
             sender,
             created_at: Instant::now(),
+            generation: gen,
         });
+        gen
     }
 
     /// Background eviction loop — runs every EVICTION_INTERVAL to remove stale connections.

rust/crates/rustproxy-http/src/counting_body.rs

@@ -9,10 +9,17 @@ use bytes::Bytes;
 use http_body::Frame;
 use rustproxy_metrics::MetricsCollector;
 
+/// Flush accumulated bytes to the metrics collector every 64 KB.
+/// This reduces per-frame DashMap shard-locked reads from ~15 to ~1 per 4 frames
+/// (assuming typical 16 KB upload frames).  The 1 Hz throughput sampler still sees
+/// data within one sampling period even at low transfer rates.
+const BYTE_FLUSH_THRESHOLD: u64 = 65_536;
+
 /// Wraps any `http_body::Body` and counts data bytes passing through.
 ///
-/// Each chunk is reported to the `MetricsCollector` immediately so that
-/// the throughput tracker (sampled at 1 Hz) reflects real-time data flow.
+/// Bytes are accumulated and flushed to the `MetricsCollector` every
+/// [`BYTE_FLUSH_THRESHOLD`] bytes (and on Drop) so the throughput tracker
+/// (sampled at 1 Hz) reflects real-time data flow without per-frame overhead.
 ///
 /// The inner body is pinned on the heap to support `!Unpin` types like `hyper::body::Incoming`.
 pub struct CountingBody<B> {
@@ -22,6 +29,8 @@ pub struct CountingBody<B> {
     source_ip: Option<String>,
     /// Whether we count bytes as "in" (request body) or "out" (response body).
     direction: Direction,
+    /// Accumulated bytes not yet flushed to the metrics collector.
+    pending_bytes: u64,
     /// Optional connection-level activity tracker. When set, poll_frame updates this
     /// to keep the idle watchdog alive during active body streaming (uploads/downloads).
     connection_activity: Option<Arc<AtomicU64>>,
@@ -57,6 +66,7 @@ impl<B> CountingBody<B> {
             route_id,
             source_ip,
             direction,
+            pending_bytes: 0,
             connection_activity: None,
             activity_start: None,
             active_requests: None,
@@ -81,14 +91,19 @@ impl<B> CountingBody<B> {
         self
     }
 
-    /// Report a chunk of bytes immediately to the metrics collector.
+    /// Flush accumulated bytes to the metrics collector.
     #[inline]
-    fn report_chunk(&self, len: u64) {
+    fn flush_pending(&mut self) {
+        if self.pending_bytes == 0 {
+            return;
+        }
+        let bytes = self.pending_bytes;
+        self.pending_bytes = 0;
         let route_id = self.route_id.as_deref();
         let source_ip = self.source_ip.as_deref();
         match self.direction {
-            Direction::In => self.metrics.record_bytes(len, 0, route_id, source_ip),
-            Direction::Out => self.metrics.record_bytes(0, len, route_id, source_ip),
+            Direction::In => self.metrics.record_bytes(bytes, 0, route_id, source_ip),
+            Direction::Out => self.metrics.record_bytes(0, bytes, route_id, source_ip),
         }
     }
 }
@@ -113,9 +128,12 @@ where
             Poll::Ready(Some(Ok(frame))) => {
                 if let Some(data) = frame.data_ref() {
                     let len = data.len() as u64;
-                    // Report bytes immediately so the 1 Hz throughput sampler sees them
-                    this.report_chunk(len);
-                    // Keep the connection-level idle watchdog alive during body streaming
+                    this.pending_bytes += len;
+                    if this.pending_bytes >= BYTE_FLUSH_THRESHOLD {
+                        this.flush_pending();
+                    }
+                    // Keep the connection-level idle watchdog alive on every frame
+                    // (this is just one atomic store — cheap enough per-frame)
                     if let (Some(activity), Some(start)) = (&this.connection_activity, &this.activity_start) {
                         activity.store(start.elapsed().as_millis() as u64, Ordering::Relaxed);
                     }
@@ -123,7 +141,11 @@ where
                 Poll::Ready(Some(Ok(frame)))
             }
             Poll::Ready(Some(Err(e))) => Poll::Ready(Some(Err(e))),
-            Poll::Ready(None) => Poll::Ready(None),
+            Poll::Ready(None) => {
+                // End of stream — flush any remaining bytes
+                this.flush_pending();
+                Poll::Ready(None)
+            }
             Poll::Pending => Poll::Pending,
         }
     }
@@ -139,6 +161,8 @@ where
 
 impl<B> Drop for CountingBody<B> {
     fn drop(&mut self) {
+        // Flush any remaining accumulated bytes so totals stay accurate
+        self.flush_pending();
         // Decrement the active-request counter so the HTTP idle watchdog
         // knows this response body is no longer streaming.
         if let Some(ref counter) = self.active_requests {

rust/crates/rustproxy-http/src/proxy_service.rs

@@ -109,6 +109,24 @@ impl tokio::io::AsyncWrite for BackendStream {
         }
     }
 
+    fn poll_write_vectored(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        bufs: &[std::io::IoSlice<'_>],
+    ) -> Poll<std::io::Result<usize>> {
+        match self.get_mut() {
+            BackendStream::Plain(s) => Pin::new(s).poll_write_vectored(cx, bufs),
+            BackendStream::Tls(s) => Pin::new(s).poll_write_vectored(cx, bufs),
+        }
+    }
+
+    fn is_write_vectored(&self) -> bool {
+        match self {
+            BackendStream::Plain(s) => s.is_write_vectored(),
+            BackendStream::Tls(s) => s.is_write_vectored(),
+        }
+    }
+
     fn poll_flush(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<std::io::Result<()>> {
         match self.get_mut() {
             BackendStream::Plain(s) => Pin::new(s).poll_flush(cx),
@@ -659,11 +677,9 @@ impl HttpProxyService {
                 h2: use_h2,
             };
 
-            // H2 pool checkout with async readiness validation.
-            // checkout_h2 does synchronous is_closed()/is_ready() checks, but these
-            // reflect cached state — the H2 connection driver (a separate tokio task)
-            // may not have processed a pending GOAWAY/RST yet. The ready().await
-            // forces the runtime to yield, giving the driver a chance to detect failures.
+            // H2 pool checkout — reuse pooled connections for all requests.
+            // The h2 crate properly replenishes connection-level flow control
+            // windows via release_capacity() as data is consumed.
             if use_h2 {
                 if let Some((mut sender, age)) = self.connection_pool.checkout_h2(&pool_key) {
                     match tokio::time::timeout(
@@ -990,7 +1006,7 @@ impl HttpProxyService {
         ) = match tokio::time::timeout(self.connect_timeout, h2_builder.handshake(io)).await {
             Ok(Ok(h)) => h,
             Ok(Err(e)) => {
-                error!(backend = %backend_key, domain = %domain, error = %e, "Backend H2 handshake failed");
+                error!(backend = %backend_key, domain = %domain, error = %e, error_debug = ?e, "Backend H2 handshake failed");
                 self.metrics.backend_handshake_error(&backend_key);
                 return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend H2 handshake failed"));
             }
@@ -1001,24 +1017,32 @@ impl HttpProxyService {
             }
         };
 
-        // Spawn the H2 connection driver; proactively evict from pool on exit
-        // so the next request gets a fresh connection instead of a dead sender.
+        // Shared generation ID: driver reads it after registration sets it.
+        // Uses u64::MAX as sentinel for "not yet registered" (driver waits/skips eviction).
+        let gen_holder = Arc::new(std::sync::atomic::AtomicU64::new(u64::MAX));
+
+        // Spawn the H2 connection driver; evict from pool on exit using generation-tagged
+        // removal to prevent phantom eviction when multiple connections share the same key.
         {
             let pool = Arc::clone(&self.connection_pool);
             let key = pool_key.clone();
+            let gen = Arc::clone(&gen_holder);
             tokio::spawn(async move {
                 if let Err(e) = conn.await {
-                    debug!("HTTP/2 upstream connection error: {}", e);
+                    warn!("HTTP/2 upstream connection error: {} ({:?})", e, e);
+                }
+                let g = gen.load(std::sync::atomic::Ordering::Relaxed);
+                if g != u64::MAX {
+                    pool.remove_h2_if_generation(&key, g);
                 }
-                pool.remove_h2(&key);
             });
         }
 
-        // Clone sender for potential pool registration; register only after first request succeeds
         let sender_for_pool = sender.clone();
         let result = self.forward_h2_with_sender(sender, parts, body, upstream_headers, upstream_path, route, route_id, source_ip, Some(pool_key), domain, conn_activity).await;
         if matches!(&result, Ok(ref resp) if resp.status() != StatusCode::BAD_GATEWAY) {
-            self.connection_pool.register_h2(pool_key.clone(), sender_for_pool);
+            let g = self.connection_pool.register_h2(pool_key.clone(), sender_for_pool);
+            gen_holder.store(g, std::sync::atomic::Ordering::Relaxed);
         }
         result
     }
@@ -1140,7 +1164,7 @@ impl HttpProxyService {
         ) = match tokio::time::timeout(self.connect_timeout, h2_builder.handshake(io)).await {
             Ok(Ok(h)) => h,
             Ok(Err(e)) => {
-                error!(backend = %backend_key, domain = %domain, error = %e, "H2 retry: handshake failed");
+                error!(backend = %backend_key, domain = %domain, error = %e, error_debug = ?e, "H2 retry: handshake failed");
                 self.metrics.backend_handshake_error(&backend_key);
                 self.metrics.backend_connection_closed(&backend_key);
                 return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend H2 retry handshake failed"));
@@ -1153,15 +1177,20 @@ impl HttpProxyService {
             }
         };
 
-        // Spawn the H2 connection driver; proactively evict from pool on exit.
+        // Spawn the H2 connection driver with generation-tagged eviction.
+        let gen_holder = Arc::new(std::sync::atomic::AtomicU64::new(u64::MAX));
         {
             let pool = Arc::clone(&self.connection_pool);
             let key = pool_key.clone();
+            let gen = Arc::clone(&gen_holder);
             tokio::spawn(async move {
                 if let Err(e) = conn.await {
-                    debug!("H2 retry: upstream connection error: {}", e);
+                    warn!("H2 retry: upstream connection error: {} ({:?})", e, e);
+                }
+                let g = gen.load(std::sync::atomic::Ordering::Relaxed);
+                if g != u64::MAX {
+                    pool.remove_h2_if_generation(&key, g);
                 }
-                pool.remove_h2(&key);
             });
         }
 
@@ -1189,7 +1218,8 @@ impl HttpProxyService {
         match sender.send_request(upstream_req).await {
             Ok(resp) => {
                 // Register in pool only after request succeeds
-                self.connection_pool.register_h2(pool_key.clone(), sender);
+                let g = self.connection_pool.register_h2(pool_key.clone(), sender);
+                gen_holder.store(g, std::sync::atomic::Ordering::Relaxed);
                 let result = self.build_streaming_response(resp, route, route_id, source_ip, conn_activity).await;
                 // Close the fresh backend connection (opened above)
                 self.metrics.backend_connection_closed(&backend_key);
@@ -1282,15 +1312,20 @@ impl HttpProxyService {
                 }
             }
             Ok(Ok((mut sender, conn))) => {
-                // Spawn the H2 connection driver; proactively evict from pool on exit.
+                // Spawn the H2 connection driver with generation-tagged eviction.
+                let gen_holder = Arc::new(std::sync::atomic::AtomicU64::new(u64::MAX));
                 {
                     let pool = Arc::clone(&self.connection_pool);
                     let key = pool_key.clone();
+                    let gen = Arc::clone(&gen_holder);
                     tokio::spawn(async move {
                         if let Err(e) = conn.await {
-                            debug!("HTTP/2 upstream connection error: {}", e);
+                            warn!("HTTP/2 upstream connection error: {} ({:?})", e, e);
+                        }
+                        let g = gen.load(std::sync::atomic::Ordering::Relaxed);
+                        if g != u64::MAX {
+                            pool.remove_h2_if_generation(&key, g);
                         }
-                        pool.remove_h2(&key);
                     });
                 }
 
@@ -1331,27 +1366,24 @@ impl HttpProxyService {
 
                 match sender.send_request(upstream_req).await {
                     Ok(upstream_response) => {
-                        // H2 works! Register sender in pool for multiplexed reuse
-                        self.connection_pool.register_h2(pool_key.clone(), sender);
+                        let g = self.connection_pool.register_h2(pool_key.clone(), sender);
+                        gen_holder.store(g, std::sync::atomic::Ordering::Relaxed);
                         self.build_streaming_response(upstream_response, route, route_id, source_ip, conn_activity).await
                     }
                     Err(e) => {
-                        // H2 request failed — backend advertises h2 via ALPN but doesn't
-                        // actually speak it. Update cache so future requests use H1.
+                        // H2 request failed on a stream level (e.g. RST_STREAM PROTOCOL_ERROR).
+                        // The H2 handshake succeeded, so the backend genuinely speaks H2 — don't
+                        // poison the protocol cache.  Only handshake-level failures (below) should
+                        // downgrade the cache to H1.
                         let bk = format!("{}:{}", upstream.host, upstream.port);
-                        warn!(
+                        debug!(
                             backend = %bk,
                             domain = %domain,
                             error = %e,
-                            "Auto-detect: H2 request failed, falling back to H1"
+                            error_debug = ?e,
+                            "H2 stream error, retrying this request as H1"
                         );
                         self.metrics.backend_h2_failure(&bk);
-                        let cache_key = crate::protocol_cache::ProtocolCacheKey {
-                            host: upstream.host.clone(),
-                            port: upstream.port,
-                            requested_host: requested_host.clone(),
-                        };
-                        self.protocol_cache.insert(cache_key, crate::protocol_cache::DetectedProtocol::H1);
 
                         // Retry as H1 for bodyless requests; return 502 for requests with bodies
                         if let Some((method, headers)) = retry_state {
@@ -1600,11 +1632,11 @@ impl HttpProxyService {
                 // Evict the dead sender so subsequent requests get fresh connections
                 if let Some(key) = pool_key {
                     let bk = format!("{}:{}", key.host, key.port);
-                    error!(backend = %bk, domain = %domain, error = %e, "Backend H2 request failed");
+                    error!(backend = %bk, domain = %domain, error = %e, error_debug = ?e, "Backend H2 request failed");
                     self.metrics.backend_request_error(&bk);
                     self.connection_pool.remove_h2(key);
                 } else {
-                    error!(domain = %domain, error = %e, "Backend H2 request failed");
+                    error!(domain = %domain, error = %e, error_debug = ?e, "Backend H2 request failed");
                 }
                 return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend H2 request failed"));
             }

rust/crates/rustproxy-http/src/shutdown_on_drop.rs

@@ -50,6 +50,18 @@ impl<S: AsyncRead + AsyncWrite + Unpin + Send + 'static> AsyncWrite for Shutdown
         Pin::new(self.get_mut().inner.as_mut().unwrap()).poll_write(cx, buf)
     }
 
+    fn poll_write_vectored(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        bufs: &[io::IoSlice<'_>],
+    ) -> Poll<io::Result<usize>> {
+        Pin::new(self.get_mut().inner.as_mut().unwrap()).poll_write_vectored(cx, bufs)
+    }
+
+    fn is_write_vectored(&self) -> bool {
+        self.inner.as_ref().unwrap().is_write_vectored()
+    }
+
     fn poll_flush(
         self: Pin<&mut Self>,
         cx: &mut Context<'_>,

rust/crates/rustproxy-metrics/src/collector.rs

@@ -259,40 +259,49 @@ impl MetricsCollector {
     /// Called per-chunk in the TCP copy loop. Only touches AtomicU64 counters —
     /// no Mutex is taken. The throughput trackers are fed during `sample_all()`.
     pub fn record_bytes(&self, bytes_in: u64, bytes_out: u64, route_id: Option<&str>, source_ip: Option<&str>) {
-        self.total_bytes_in.fetch_add(bytes_in, Ordering::Relaxed);
-        self.total_bytes_out.fetch_add(bytes_out, Ordering::Relaxed);
-
-        // Accumulate into lock-free pending throughput counters
-        self.global_pending_tp_in.fetch_add(bytes_in, Ordering::Relaxed);
-        self.global_pending_tp_out.fetch_add(bytes_out, Ordering::Relaxed);
+        // Short-circuit: only touch counters for the direction that has data.
+        // CountingBody always calls with one direction zero — skipping the zero
+        // direction avoids ~50% of DashMap shard-locked reads per call.
+        if bytes_in > 0 {
+            self.total_bytes_in.fetch_add(bytes_in, Ordering::Relaxed);
+            self.global_pending_tp_in.fetch_add(bytes_in, Ordering::Relaxed);
+        }
+        if bytes_out > 0 {
+            self.total_bytes_out.fetch_add(bytes_out, Ordering::Relaxed);
+            self.global_pending_tp_out.fetch_add(bytes_out, Ordering::Relaxed);
+        }
 
         // Per-route tracking: use get() first (zero-alloc fast path for existing entries),
         // fall back to entry() with to_string() only on the rare first-chunk miss.
         if let Some(route_id) = route_id {
-            if let Some(counter) = self.route_bytes_in.get(route_id) {
-                counter.fetch_add(bytes_in, Ordering::Relaxed);
-            } else {
-                self.route_bytes_in.entry(route_id.to_string())
-                    .or_insert_with(|| AtomicU64::new(0))
-                    .fetch_add(bytes_in, Ordering::Relaxed);
+            if bytes_in > 0 {
+                if let Some(counter) = self.route_bytes_in.get(route_id) {
+                    counter.fetch_add(bytes_in, Ordering::Relaxed);
+                } else {
+                    self.route_bytes_in.entry(route_id.to_string())
+                        .or_insert_with(|| AtomicU64::new(0))
+                        .fetch_add(bytes_in, Ordering::Relaxed);
+                }
             }
-            if let Some(counter) = self.route_bytes_out.get(route_id) {
-                counter.fetch_add(bytes_out, Ordering::Relaxed);
-            } else {
-                self.route_bytes_out.entry(route_id.to_string())
-                    .or_insert_with(|| AtomicU64::new(0))
-                    .fetch_add(bytes_out, Ordering::Relaxed);
+            if bytes_out > 0 {
+                if let Some(counter) = self.route_bytes_out.get(route_id) {
+                    counter.fetch_add(bytes_out, Ordering::Relaxed);
+                } else {
+                    self.route_bytes_out.entry(route_id.to_string())
+                        .or_insert_with(|| AtomicU64::new(0))
+                        .fetch_add(bytes_out, Ordering::Relaxed);
+                }
             }
 
             // Accumulate into per-route pending throughput counters (lock-free)
             if let Some(entry) = self.route_pending_tp.get(route_id) {
-                entry.0.fetch_add(bytes_in, Ordering::Relaxed);
-                entry.1.fetch_add(bytes_out, Ordering::Relaxed);
+                if bytes_in > 0 { entry.0.fetch_add(bytes_in, Ordering::Relaxed); }
+                if bytes_out > 0 { entry.1.fetch_add(bytes_out, Ordering::Relaxed); }
             } else {
                 let entry = self.route_pending_tp.entry(route_id.to_string())
                     .or_insert_with(|| (AtomicU64::new(0), AtomicU64::new(0)));
-                entry.0.fetch_add(bytes_in, Ordering::Relaxed);
-                entry.1.fetch_add(bytes_out, Ordering::Relaxed);
+                if bytes_in > 0 { entry.0.fetch_add(bytes_in, Ordering::Relaxed); }
+                if bytes_out > 0 { entry.1.fetch_add(bytes_out, Ordering::Relaxed); }
             }
         }
 
@@ -302,30 +311,34 @@ impl MetricsCollector {
             // This prevents orphaned entries when record_bytes races with
             // connection_closed (which evicts all per-IP data on last close).
             if self.ip_connections.contains_key(ip) {
-                if let Some(counter) = self.ip_bytes_in.get(ip) {
-                    counter.fetch_add(bytes_in, Ordering::Relaxed);
-                } else {
-                    self.ip_bytes_in.entry(ip.to_string())
-                        .or_insert_with(|| AtomicU64::new(0))
-                        .fetch_add(bytes_in, Ordering::Relaxed);
+                if bytes_in > 0 {
+                    if let Some(counter) = self.ip_bytes_in.get(ip) {
+                        counter.fetch_add(bytes_in, Ordering::Relaxed);
+                    } else {
+                        self.ip_bytes_in.entry(ip.to_string())
+                            .or_insert_with(|| AtomicU64::new(0))
+                            .fetch_add(bytes_in, Ordering::Relaxed);
+                    }
                 }
-                if let Some(counter) = self.ip_bytes_out.get(ip) {
-                    counter.fetch_add(bytes_out, Ordering::Relaxed);
-                } else {
-                    self.ip_bytes_out.entry(ip.to_string())
-                        .or_insert_with(|| AtomicU64::new(0))
-                        .fetch_add(bytes_out, Ordering::Relaxed);
+                if bytes_out > 0 {
+                    if let Some(counter) = self.ip_bytes_out.get(ip) {
+                        counter.fetch_add(bytes_out, Ordering::Relaxed);
+                    } else {
+                        self.ip_bytes_out.entry(ip.to_string())
+                            .or_insert_with(|| AtomicU64::new(0))
+                            .fetch_add(bytes_out, Ordering::Relaxed);
+                    }
                 }
 
                 // Accumulate into per-IP pending throughput counters (lock-free)
                 if let Some(entry) = self.ip_pending_tp.get(ip) {
-                    entry.0.fetch_add(bytes_in, Ordering::Relaxed);
-                    entry.1.fetch_add(bytes_out, Ordering::Relaxed);
+                    if bytes_in > 0 { entry.0.fetch_add(bytes_in, Ordering::Relaxed); }
+                    if bytes_out > 0 { entry.1.fetch_add(bytes_out, Ordering::Relaxed); }
                 } else {
                     let entry = self.ip_pending_tp.entry(ip.to_string())
                         .or_insert_with(|| (AtomicU64::new(0), AtomicU64::new(0)));
-                    entry.0.fetch_add(bytes_in, Ordering::Relaxed);
-                    entry.1.fetch_add(bytes_out, Ordering::Relaxed);
+                    if bytes_in > 0 { entry.0.fetch_add(bytes_in, Ordering::Relaxed); }
+                    if bytes_out > 0 { entry.1.fetch_add(bytes_out, Ordering::Relaxed); }
                 }
             }
         }

ts/00_commitinfo_data.ts

@@ -3,6 +3,6 @@
  */
 export const commitinfo = {
   name: '@push.rocks/smartproxy',
-  version: '25.11.10',
+  version: '25.11.23',
   description: 'A powerful proxy package with unified route-based configuration for high traffic management. Features include SSL/TLS support, flexible routing patterns, WebSocket handling, advanced security options, and automatic ACME certificate management.'
 }