Compare commits
16 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
| 95adf56e52 | |||
| c96a493fb6 | |||
| b92587cc16 | |||
| b3dc0a6db2 | |||
| de3b8d3f58 | |||
| 75089ec975 | |||
| b106db932b | |||
| fb0c0dcc31 | |||
| 61b67b91a0 | |||
| fc64f5a95e | |||
| 90b83a9dbe | |||
| 508621e231 | |||
| 9ef21dcb41 | |||
| 0acd907431 | |||
| 80276a70e8 | |||
| 0d4399d7f1 |
50
changelog.md
50
changelog.md
@@ -1,5 +1,55 @@
|
||||
# Changelog
|
||||
|
||||
## 2026-03-12 - 25.10.7 - fix(rustproxy-http)
|
||||
remove Host header from HTTP/2 upstream requests while preserving it for HTTP/1 retries
|
||||
|
||||
- strips the Host header before sending HTTP/2 upstream requests so :authority from the URI is used instead
|
||||
- avoids 400 responses from nginx caused by sending both Host and :authority headers
|
||||
- keeps a cloned header set for bodyless request retries so HTTP/1 fallback still retains the Host header
|
||||
|
||||
## 2026-03-12 - 25.10.6 - fix(rustproxy-http)
|
||||
use the requested domain as HTTP/2 authority instead of the backend host and port
|
||||
|
||||
- build HTTP/2 absolute URIs from the client-facing domain so the :authority pseudo-header matches the Host header
|
||||
- remove backend port from generated HTTP/2 request URIs and fall back to the upstream host only when no domain is available
|
||||
- apply the authority handling consistently across pooled, inline, and generic upstream request paths
|
||||
|
||||
## 2026-03-12 - 25.10.5 - fix(rustproxy-http)
|
||||
configure HTTP/2 client builders with a Tokio timer for keep-alive handling
|
||||
|
||||
- Adds TokioTimer to all HTTP/2 client builder instances in proxy_service.
|
||||
- Ensures configured HTTP/2 keep-alive interval and timeout settings have the required timer runtime support.
|
||||
|
||||
## 2026-03-12 - 25.10.4 - fix(rustproxy-http)
|
||||
stabilize upstream HTTP/2 forwarding and fallback behavior
|
||||
|
||||
- Remove hop-by-hop headers before forwarding requests to HTTP/2 backends to comply with RFC 9113.
|
||||
- Use ALPN-enabled TLS configuration whenever HTTP/2 is possible, including explicit H2 connections and retries.
|
||||
- Add HTTP/2 handshake timeouts, tuned connection settings, and fallback to HTTP/1 when H2 negotiation times out or fails.
|
||||
- Register pooled HTTP/2 senders only after a successful first request to avoid reusing broken connections.
|
||||
- Build absolute URIs for HTTP/2 upstream requests so pseudo-headers such as scheme and authority are derived correctly.
|
||||
|
||||
## 2026-03-12 - 25.10.3 - fix(rustproxy-http)
|
||||
include request domain in backend proxy error and protocol detection logs
|
||||
|
||||
- Adds domain context to backend TCP/TLS connect, handshake, request failure, retry, and fallback log entries in the Rust HTTP proxy service.
|
||||
- Propagates the resolved host/domain through H1, H2, pooled, and fallback forwarding paths so backend-level diagnostics can be correlated with the original request domain.
|
||||
|
||||
## 2026-03-12 - 25.10.2 - fix(repo)
|
||||
no code changes to release
|
||||
|
||||
|
||||
## 2026-03-12 - 25.10.1 - fix(repo)
|
||||
no changes to commit
|
||||
|
||||
|
||||
## 2026-03-12 - 25.10.0 - feat(metrics)
|
||||
add per-backend connection, error, protocol, and pool metrics with stale backend pruning
|
||||
|
||||
- tracks backend connection lifecycle, connect timing, protocol detection, pool hit/miss rates, handshake/request errors, and h2 fallback failures in Rust metrics
|
||||
- exposes backend metrics through the TypeScript metrics adapter with backend listings, protocol lookup, and top error summaries
|
||||
- prunes backend metrics for backends no longer referenced by active routes, including preserved-port targets expanded across listening ports
|
||||
|
||||
## 2026-03-11 - 25.9.3 - fix(rustproxy-http)
|
||||
Evict stale HTTP/2 pooled senders and retry bodyless requests with fresh backend connections to avoid 502s
|
||||
|
||||
|
||||
@@ -1,6 +1,6 @@
|
||||
{
|
||||
"name": "@push.rocks/smartproxy",
|
||||
"version": "25.9.3",
|
||||
"version": "25.10.7",
|
||||
"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",
|
||||
|
||||
@@ -479,6 +479,7 @@ impl HttpProxyService {
|
||||
}
|
||||
|
||||
let upstream_key = format!("{}:{}", upstream.host, upstream.port);
|
||||
let domain_str = host.as_deref().unwrap_or("-");
|
||||
self.upstream_selector.connection_started(&upstream_key);
|
||||
|
||||
// Check for WebSocket upgrade
|
||||
@@ -538,6 +539,15 @@ impl HttpProxyService {
|
||||
}
|
||||
}
|
||||
|
||||
// Remove hop-by-hop headers (RFC 9113 §8.2.2 forbids connection-specific headers in H2)
|
||||
upstream_headers.remove("connection");
|
||||
upstream_headers.remove("keep-alive");
|
||||
upstream_headers.remove("proxy-connection");
|
||||
upstream_headers.remove("transfer-encoding");
|
||||
upstream_headers.remove("te");
|
||||
upstream_headers.remove("trailer");
|
||||
upstream_headers.remove("upgrade");
|
||||
|
||||
// Add standard reverse-proxy headers (X-Forwarded-*)
|
||||
{
|
||||
let original_host = host.as_deref().unwrap_or("");
|
||||
@@ -618,9 +628,11 @@ impl HttpProxyService {
|
||||
// H2 pool checkout (H2 senders are Clone and multiplexed)
|
||||
if use_h2 {
|
||||
if let Some(sender) = self.connection_pool.checkout_h2(&pool_key) {
|
||||
self.metrics.backend_pool_hit(&upstream_key);
|
||||
self.metrics.set_backend_protocol(&upstream_key, "h2");
|
||||
let result = self.forward_h2_pooled(
|
||||
sender, parts, body, upstream_headers, &upstream_path,
|
||||
route_match.route, route_id, &ip_str, &pool_key,
|
||||
route_match.route, route_id, &ip_str, &pool_key, domain_str,
|
||||
).await;
|
||||
self.upstream_selector.connection_ended(&upstream_key);
|
||||
return result;
|
||||
@@ -629,14 +641,17 @@ impl HttpProxyService {
|
||||
}
|
||||
|
||||
// --- Fresh connection path ---
|
||||
// Choose TLS config: use ALPN config for auto-detect probe, plain config otherwise
|
||||
let tls_config = if needs_alpn_probe {
|
||||
self.metrics.backend_pool_miss(&upstream_key);
|
||||
|
||||
// Choose TLS config: use ALPN config when H2 is possible (auto-detect probe OR explicit H2)
|
||||
let tls_config = if needs_alpn_probe || use_h2 {
|
||||
&self.backend_tls_config_alpn
|
||||
} else {
|
||||
&self.backend_tls_config
|
||||
};
|
||||
|
||||
// Establish backend connection
|
||||
let connect_start = std::time::Instant::now();
|
||||
let (backend, detected_h2) = if upstream.use_tls {
|
||||
match tokio::time::timeout(
|
||||
self.connect_timeout,
|
||||
@@ -661,25 +676,42 @@ impl HttpProxyService {
|
||||
};
|
||||
self.protocol_cache.insert(cache_key, detected);
|
||||
|
||||
debug!(
|
||||
"Auto-detected {} for backend {}:{}",
|
||||
if is_h2 { "HTTP/2" } else { "HTTP/1.1" },
|
||||
upstream.host, upstream.port
|
||||
info!(
|
||||
backend = %upstream_key,
|
||||
domain = %domain_str,
|
||||
protocol = if is_h2 { "h2" } else { "h1" },
|
||||
connect_time_ms = %connect_start.elapsed().as_millis(),
|
||||
"Backend protocol detected via ALPN"
|
||||
);
|
||||
|
||||
is_h2
|
||||
} else {
|
||||
use_h2
|
||||
};
|
||||
self.metrics.backend_connection_opened(&upstream_key, connect_start.elapsed());
|
||||
self.metrics.set_backend_protocol(&upstream_key, if final_h2 { "h2" } else { "h1" });
|
||||
(BackendStream::Tls(tls), final_h2)
|
||||
}
|
||||
Ok(Err(e)) => {
|
||||
error!("Failed TLS connect to upstream {}:{}: {}", upstream.host, upstream.port, e);
|
||||
error!(
|
||||
backend = %upstream_key,
|
||||
domain = %domain_str,
|
||||
connect_time_ms = %connect_start.elapsed().as_millis(),
|
||||
error = %e,
|
||||
"Backend TLS connect failed"
|
||||
);
|
||||
self.metrics.backend_connect_error(&upstream_key);
|
||||
self.upstream_selector.connection_ended(&upstream_key);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend TLS unavailable"));
|
||||
}
|
||||
Err(_) => {
|
||||
error!("Upstream TLS connect timeout for {}:{}", upstream.host, upstream.port);
|
||||
error!(
|
||||
backend = %upstream_key,
|
||||
domain = %domain_str,
|
||||
connect_time_ms = %connect_start.elapsed().as_millis(),
|
||||
"Backend TLS connect timeout"
|
||||
);
|
||||
self.metrics.backend_connect_error(&upstream_key);
|
||||
self.upstream_selector.connection_ended(&upstream_key);
|
||||
return Ok(error_response(StatusCode::GATEWAY_TIMEOUT, "Backend TLS connect timeout"));
|
||||
}
|
||||
@@ -694,15 +726,30 @@ impl HttpProxyService {
|
||||
let _ = socket2::SockRef::from(&s).set_tcp_keepalive(
|
||||
&socket2::TcpKeepalive::new().with_time(std::time::Duration::from_secs(60))
|
||||
);
|
||||
self.metrics.backend_connection_opened(&upstream_key, connect_start.elapsed());
|
||||
self.metrics.set_backend_protocol(&upstream_key, if use_h2 { "h2" } else { "h1" });
|
||||
(BackendStream::Plain(s), use_h2)
|
||||
}
|
||||
Ok(Err(e)) => {
|
||||
error!("Failed to connect to upstream {}:{}: {}", upstream.host, upstream.port, e);
|
||||
error!(
|
||||
backend = %upstream_key,
|
||||
domain = %domain_str,
|
||||
connect_time_ms = %connect_start.elapsed().as_millis(),
|
||||
error = %e,
|
||||
"Backend TCP connect failed"
|
||||
);
|
||||
self.metrics.backend_connect_error(&upstream_key);
|
||||
self.upstream_selector.connection_ended(&upstream_key);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend unavailable"));
|
||||
}
|
||||
Err(_) => {
|
||||
error!("Upstream connect timeout for {}:{}", upstream.host, upstream.port);
|
||||
error!(
|
||||
backend = %upstream_key,
|
||||
domain = %domain_str,
|
||||
connect_time_ms = %connect_start.elapsed().as_millis(),
|
||||
"Backend TCP connect timeout"
|
||||
);
|
||||
self.metrics.backend_connect_error(&upstream_key);
|
||||
self.upstream_selector.connection_ended(&upstream_key);
|
||||
return Ok(error_response(StatusCode::GATEWAY_TIMEOUT, "Backend connect timeout"));
|
||||
}
|
||||
@@ -724,22 +771,23 @@ impl HttpProxyService {
|
||||
self.forward_h2_with_fallback(
|
||||
io, parts, body, upstream_headers, &upstream_path,
|
||||
&upstream, route_match.route, route_id, &ip_str, &final_pool_key,
|
||||
host.clone(),
|
||||
host.clone(), domain_str,
|
||||
).await
|
||||
} else {
|
||||
// Explicit H2 mode: hard-fail on handshake error (preserved behavior)
|
||||
self.forward_h2(
|
||||
io, parts, body, upstream_headers, &upstream_path,
|
||||
&upstream, route_match.route, route_id, &ip_str, &final_pool_key,
|
||||
&upstream, route_match.route, route_id, &ip_str, &final_pool_key, domain_str,
|
||||
).await
|
||||
}
|
||||
} else {
|
||||
self.forward_h1(
|
||||
io, parts, body, upstream_headers, &upstream_path,
|
||||
&upstream, route_match.route, route_id, &ip_str, &final_pool_key,
|
||||
&upstream, route_match.route, route_id, &ip_str, &final_pool_key, domain_str,
|
||||
).await
|
||||
};
|
||||
self.upstream_selector.connection_ended(&upstream_key);
|
||||
self.metrics.backend_connection_closed(&upstream_key);
|
||||
result
|
||||
}
|
||||
|
||||
@@ -757,12 +805,16 @@ impl HttpProxyService {
|
||||
route_id: Option<&str>,
|
||||
source_ip: &str,
|
||||
pool_key: &crate::connection_pool::PoolKey,
|
||||
domain: &str,
|
||||
) -> Result<Response<BoxBody<Bytes, hyper::Error>>, hyper::Error> {
|
||||
let backend_key = format!("{}:{}", pool_key.host, pool_key.port);
|
||||
|
||||
// Try pooled H1 connection first — avoids TCP+TLS handshake
|
||||
if let Some(pooled_sender) = self.connection_pool.checkout_h1(pool_key) {
|
||||
self.metrics.backend_pool_hit(&backend_key);
|
||||
return self.forward_h1_with_sender(
|
||||
pooled_sender, parts, body, upstream_headers, upstream_path,
|
||||
route, route_id, source_ip, pool_key,
|
||||
route, route_id, source_ip, pool_key, domain,
|
||||
).await;
|
||||
}
|
||||
|
||||
@@ -773,7 +825,8 @@ impl HttpProxyService {
|
||||
) = match hyper::client::conn::http1::handshake(io).await {
|
||||
Ok(h) => h,
|
||||
Err(e) => {
|
||||
error!("Upstream handshake failed: {}", e);
|
||||
error!(backend = %backend_key, domain = %domain, error = %e, "Backend H1 handshake failed");
|
||||
self.metrics.backend_handshake_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend handshake failed"));
|
||||
}
|
||||
};
|
||||
@@ -784,7 +837,7 @@ impl HttpProxyService {
|
||||
}
|
||||
});
|
||||
|
||||
self.forward_h1_with_sender(sender, parts, body, upstream_headers, upstream_path, route, route_id, source_ip, pool_key).await
|
||||
self.forward_h1_with_sender(sender, parts, body, upstream_headers, upstream_path, route, route_id, source_ip, pool_key, domain).await
|
||||
}
|
||||
|
||||
/// Common H1 forwarding logic used by both fresh and pooled paths.
|
||||
@@ -799,6 +852,7 @@ impl HttpProxyService {
|
||||
route_id: Option<&str>,
|
||||
source_ip: &str,
|
||||
pool_key: &crate::connection_pool::PoolKey,
|
||||
domain: &str,
|
||||
) -> Result<Response<BoxBody<Bytes, hyper::Error>>, hyper::Error> {
|
||||
// Always use HTTP/1.1 for h1 backend connections (h2 incoming requests have version HTTP/2.0)
|
||||
let mut upstream_req = Request::builder()
|
||||
@@ -825,7 +879,9 @@ impl HttpProxyService {
|
||||
let upstream_response = match sender.send_request(upstream_req).await {
|
||||
Ok(resp) => resp,
|
||||
Err(e) => {
|
||||
error!("Upstream request failed: {}", e);
|
||||
let bk = format!("{}:{}", pool_key.host, pool_key.port);
|
||||
error!(backend = %bk, domain = %domain, error = %e, "Backend H1 request failed");
|
||||
self.metrics.backend_request_error(&bk);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend request failed"));
|
||||
}
|
||||
};
|
||||
@@ -850,18 +906,32 @@ impl HttpProxyService {
|
||||
route_id: Option<&str>,
|
||||
source_ip: &str,
|
||||
pool_key: &crate::connection_pool::PoolKey,
|
||||
domain: &str,
|
||||
) -> Result<Response<BoxBody<Bytes, hyper::Error>>, hyper::Error> {
|
||||
let backend_key = format!("{}:{}", pool_key.host, pool_key.port);
|
||||
let exec = hyper_util::rt::TokioExecutor::new();
|
||||
// Explicitly type the handshake with BoxBody for uniform pool type
|
||||
let mut h2_builder = hyper::client::conn::http2::Builder::new(exec);
|
||||
h2_builder
|
||||
.timer(hyper_util::rt::TokioTimer::new())
|
||||
.keep_alive_interval(std::time::Duration::from_secs(10))
|
||||
.keep_alive_timeout(std::time::Duration::from_secs(5))
|
||||
.adaptive_window(true)
|
||||
.initial_stream_window_size(2 * 1024 * 1024);
|
||||
let (sender, conn): (
|
||||
hyper::client::conn::http2::SendRequest<BoxBody<Bytes, hyper::Error>>,
|
||||
hyper::client::conn::http2::Connection<TokioIo<BackendStream>, BoxBody<Bytes, hyper::Error>, hyper_util::rt::TokioExecutor>,
|
||||
) = match hyper::client::conn::http2::handshake(exec, io).await {
|
||||
Ok(h) => h,
|
||||
Err(e) => {
|
||||
error!("HTTP/2 upstream handshake failed: {}", e);
|
||||
) = 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");
|
||||
self.metrics.backend_handshake_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend H2 handshake failed"));
|
||||
}
|
||||
Err(_) => {
|
||||
error!(backend = %backend_key, domain = %domain, "Backend H2 handshake timeout");
|
||||
self.metrics.backend_handshake_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::GATEWAY_TIMEOUT, "Backend H2 handshake timeout"));
|
||||
}
|
||||
};
|
||||
|
||||
tokio::spawn(async move {
|
||||
@@ -870,10 +940,13 @@ impl HttpProxyService {
|
||||
}
|
||||
});
|
||||
|
||||
// Register for multiplexed reuse
|
||||
self.connection_pool.register_h2(pool_key.clone(), sender.clone());
|
||||
|
||||
self.forward_h2_with_sender(sender, parts, body, upstream_headers, upstream_path, route, route_id, source_ip, Some(pool_key)).await
|
||||
// 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).await;
|
||||
if matches!(&result, Ok(ref resp) if resp.status() != StatusCode::BAD_GATEWAY) {
|
||||
self.connection_pool.register_h2(pool_key.clone(), sender_for_pool);
|
||||
}
|
||||
result
|
||||
}
|
||||
|
||||
/// Forward request using an existing (pooled) HTTP/2 sender.
|
||||
@@ -890,6 +963,7 @@ impl HttpProxyService {
|
||||
route_id: Option<&str>,
|
||||
source_ip: &str,
|
||||
pool_key: &crate::connection_pool::PoolKey,
|
||||
domain: &str,
|
||||
) -> Result<Response<BoxBody<Bytes, hyper::Error>>, hyper::Error> {
|
||||
// Save retry state for bodyless requests (cheap: Method is an enum, HeaderMap clones Arc-backed Bytes)
|
||||
let retry_state = if body.is_end_stream() {
|
||||
@@ -900,18 +974,18 @@ impl HttpProxyService {
|
||||
|
||||
let result = self.forward_h2_with_sender(
|
||||
sender, parts, body, upstream_headers, upstream_path,
|
||||
route, route_id, source_ip, Some(pool_key),
|
||||
route, route_id, source_ip, Some(pool_key), domain,
|
||||
).await;
|
||||
|
||||
// If the request failed (502) and we can retry with an empty body, do so
|
||||
let is_502 = matches!(&result, Ok(resp) if resp.status() == StatusCode::BAD_GATEWAY);
|
||||
if is_502 {
|
||||
if let Some((method, headers)) = retry_state {
|
||||
warn!("Stale pooled H2 sender for {}:{}, retrying with fresh connection",
|
||||
pool_key.host, pool_key.port);
|
||||
warn!(backend = %format!("{}:{}", pool_key.host, pool_key.port), domain = %domain,
|
||||
"Stale pooled H2 sender, retrying with fresh connection");
|
||||
return self.retry_h2_with_fresh_connection(
|
||||
method, headers, upstream_path,
|
||||
pool_key, route, route_id, source_ip,
|
||||
pool_key, route, route_id, source_ip, domain,
|
||||
).await;
|
||||
}
|
||||
}
|
||||
@@ -929,20 +1003,26 @@ impl HttpProxyService {
|
||||
route: &rustproxy_config::RouteConfig,
|
||||
route_id: Option<&str>,
|
||||
source_ip: &str,
|
||||
domain: &str,
|
||||
) -> Result<Response<BoxBody<Bytes, hyper::Error>>, hyper::Error> {
|
||||
let backend_key = format!("{}:{}", pool_key.host, pool_key.port);
|
||||
|
||||
// Establish fresh backend connection
|
||||
let retry_connect_start = std::time::Instant::now();
|
||||
let backend = if pool_key.use_tls {
|
||||
match tokio::time::timeout(
|
||||
self.connect_timeout,
|
||||
connect_tls_backend(&self.backend_tls_config, &pool_key.host, pool_key.port),
|
||||
connect_tls_backend(&self.backend_tls_config_alpn, &pool_key.host, pool_key.port),
|
||||
).await {
|
||||
Ok(Ok(tls)) => BackendStream::Tls(tls),
|
||||
Ok(Err(e)) => {
|
||||
error!("H2 retry: TLS connect failed for {}:{}: {}", pool_key.host, pool_key.port, e);
|
||||
error!(backend = %backend_key, domain = %domain, error = %e, "H2 retry: TLS connect failed");
|
||||
self.metrics.backend_connect_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend unavailable on H2 retry"));
|
||||
}
|
||||
Err(_) => {
|
||||
error!("H2 retry: TLS connect timeout for {}:{}", pool_key.host, pool_key.port);
|
||||
error!(backend = %backend_key, domain = %domain, "H2 retry: TLS connect timeout");
|
||||
self.metrics.backend_connect_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::GATEWAY_TIMEOUT, "Backend timeout on H2 retry"));
|
||||
}
|
||||
}
|
||||
@@ -956,27 +1036,45 @@ impl HttpProxyService {
|
||||
BackendStream::Plain(s)
|
||||
}
|
||||
Ok(Err(e)) => {
|
||||
error!("H2 retry: connect failed for {}:{}: {}", pool_key.host, pool_key.port, e);
|
||||
error!(backend = %backend_key, domain = %domain, error = %e, "H2 retry: TCP connect failed");
|
||||
self.metrics.backend_connect_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend unavailable on H2 retry"));
|
||||
}
|
||||
Err(_) => {
|
||||
error!("H2 retry: connect timeout for {}:{}", pool_key.host, pool_key.port);
|
||||
error!(backend = %backend_key, domain = %domain, "H2 retry: TCP connect timeout");
|
||||
self.metrics.backend_connect_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::GATEWAY_TIMEOUT, "Backend timeout on H2 retry"));
|
||||
}
|
||||
}
|
||||
};
|
||||
self.metrics.backend_connection_opened(&backend_key, retry_connect_start.elapsed());
|
||||
|
||||
let io = TokioIo::new(backend);
|
||||
let exec = hyper_util::rt::TokioExecutor::new();
|
||||
let mut h2_builder = hyper::client::conn::http2::Builder::new(exec);
|
||||
h2_builder
|
||||
.timer(hyper_util::rt::TokioTimer::new())
|
||||
.keep_alive_interval(std::time::Duration::from_secs(10))
|
||||
.keep_alive_timeout(std::time::Duration::from_secs(5))
|
||||
.adaptive_window(true)
|
||||
.initial_stream_window_size(2 * 1024 * 1024);
|
||||
let (mut sender, conn): (
|
||||
hyper::client::conn::http2::SendRequest<BoxBody<Bytes, hyper::Error>>,
|
||||
hyper::client::conn::http2::Connection<TokioIo<BackendStream>, BoxBody<Bytes, hyper::Error>, hyper_util::rt::TokioExecutor>,
|
||||
) = match hyper::client::conn::http2::handshake(exec, io).await {
|
||||
Ok(h) => h,
|
||||
Err(e) => {
|
||||
error!("H2 retry: handshake failed for {}:{}: {}", pool_key.host, pool_key.port, e);
|
||||
) = 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");
|
||||
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"));
|
||||
}
|
||||
Err(_) => {
|
||||
error!(backend = %backend_key, domain = %domain, "H2 retry: handshake timeout");
|
||||
self.metrics.backend_handshake_error(&backend_key);
|
||||
self.metrics.backend_connection_closed(&backend_key);
|
||||
return Ok(error_response(StatusCode::GATEWAY_TIMEOUT, "Backend H2 retry handshake timeout"));
|
||||
}
|
||||
};
|
||||
|
||||
tokio::spawn(async move {
|
||||
@@ -985,13 +1083,17 @@ impl HttpProxyService {
|
||||
}
|
||||
});
|
||||
|
||||
// Register fresh sender in pool for future requests
|
||||
self.connection_pool.register_h2(pool_key.clone(), sender.clone());
|
||||
|
||||
// Build request with empty body
|
||||
// Build request with empty body using absolute URI for H2 pseudo-headers
|
||||
let scheme = if pool_key.use_tls { "https" } else { "http" };
|
||||
let authority = if domain != "-" { domain } else { pool_key.host.as_str() };
|
||||
let h2_uri = format!("{}://{}{}", scheme, authority, upstream_path);
|
||||
let mut upstream_req = Request::builder()
|
||||
.method(method)
|
||||
.uri(upstream_path);
|
||||
.uri(&h2_uri);
|
||||
|
||||
// Remove Host header for H2 — :authority pseudo-header (from URI) is sufficient
|
||||
let mut upstream_headers = upstream_headers;
|
||||
upstream_headers.remove("host");
|
||||
|
||||
if let Some(headers) = upstream_req.headers_mut() {
|
||||
*headers = upstream_headers;
|
||||
@@ -1004,11 +1106,18 @@ impl HttpProxyService {
|
||||
|
||||
match sender.send_request(upstream_req).await {
|
||||
Ok(resp) => {
|
||||
self.build_streaming_response(resp, route, route_id, source_ip).await
|
||||
// Register in pool only after request succeeds
|
||||
self.connection_pool.register_h2(pool_key.clone(), sender);
|
||||
let result = self.build_streaming_response(resp, route, route_id, source_ip).await;
|
||||
// Close the fresh backend connection (opened above)
|
||||
self.metrics.backend_connection_closed(&backend_key);
|
||||
result
|
||||
}
|
||||
Err(e) => {
|
||||
error!("H2 retry: request failed for {}:{}: {}", pool_key.host, pool_key.port, e);
|
||||
self.connection_pool.remove_h2(pool_key);
|
||||
error!(backend = %backend_key, domain = %domain, error = %e, "H2 retry: request failed");
|
||||
self.metrics.backend_request_error(&backend_key);
|
||||
// Close the fresh backend connection (opened above)
|
||||
self.metrics.backend_connection_closed(&backend_key);
|
||||
Ok(error_response(StatusCode::BAD_GATEWAY, "Backend H2 request failed on retry"))
|
||||
}
|
||||
}
|
||||
@@ -1026,7 +1135,7 @@ impl HttpProxyService {
|
||||
io: TokioIo<BackendStream>,
|
||||
parts: hyper::http::request::Parts,
|
||||
body: Incoming,
|
||||
upstream_headers: hyper::HeaderMap,
|
||||
mut upstream_headers: hyper::HeaderMap,
|
||||
upstream_path: &str,
|
||||
upstream: &crate::upstream_selector::UpstreamSelection,
|
||||
route: &rustproxy_config::RouteConfig,
|
||||
@@ -1034,15 +1143,62 @@ impl HttpProxyService {
|
||||
source_ip: &str,
|
||||
pool_key: &crate::connection_pool::PoolKey,
|
||||
requested_host: Option<String>,
|
||||
domain: &str,
|
||||
) -> Result<Response<BoxBody<Bytes, hyper::Error>>, hyper::Error> {
|
||||
let exec = hyper_util::rt::TokioExecutor::new();
|
||||
let handshake_result: Result<(
|
||||
hyper::client::conn::http2::SendRequest<BoxBody<Bytes, hyper::Error>>,
|
||||
hyper::client::conn::http2::Connection<TokioIo<BackendStream>, BoxBody<Bytes, hyper::Error>, hyper_util::rt::TokioExecutor>,
|
||||
), hyper::Error> = hyper::client::conn::http2::handshake(exec, io).await;
|
||||
let mut h2_builder = hyper::client::conn::http2::Builder::new(exec);
|
||||
h2_builder
|
||||
.timer(hyper_util::rt::TokioTimer::new())
|
||||
.keep_alive_interval(std::time::Duration::from_secs(10))
|
||||
.keep_alive_timeout(std::time::Duration::from_secs(5))
|
||||
.adaptive_window(true)
|
||||
.initial_stream_window_size(2 * 1024 * 1024);
|
||||
let handshake_result = tokio::time::timeout(
|
||||
self.connect_timeout,
|
||||
h2_builder.handshake(io),
|
||||
).await;
|
||||
|
||||
match handshake_result {
|
||||
Ok((mut sender, conn)) => {
|
||||
Err(_) => {
|
||||
// H2 handshake timed out — fall back to H1
|
||||
let bk = format!("{}:{}", upstream.host, upstream.port);
|
||||
warn!(
|
||||
backend = %bk,
|
||||
domain = %domain,
|
||||
"H2 handshake timeout, falling back to H1"
|
||||
);
|
||||
self.metrics.backend_h2_failure(&bk);
|
||||
self.metrics.backend_handshake_error(&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);
|
||||
|
||||
match self.reconnect_backend(upstream, domain).await {
|
||||
Some(fallback_backend) => {
|
||||
let h1_pool_key = crate::connection_pool::PoolKey {
|
||||
host: upstream.host.clone(),
|
||||
port: upstream.port,
|
||||
use_tls: upstream.use_tls,
|
||||
h2: false,
|
||||
};
|
||||
let fallback_io = TokioIo::new(fallback_backend);
|
||||
let result = self.forward_h1(
|
||||
fallback_io, parts, body, upstream_headers, upstream_path,
|
||||
upstream, route, route_id, source_ip, &h1_pool_key, domain,
|
||||
).await;
|
||||
self.metrics.backend_connection_closed(&bk);
|
||||
result
|
||||
}
|
||||
None => {
|
||||
Ok(error_response(StatusCode::BAD_GATEWAY, "Backend unavailable after H2 timeout fallback"))
|
||||
}
|
||||
}
|
||||
}
|
||||
Ok(Ok((mut sender, conn))) => {
|
||||
tokio::spawn(async move {
|
||||
if let Err(e) = conn.await {
|
||||
debug!("HTTP/2 upstream connection error: {}", e);
|
||||
@@ -1050,18 +1206,25 @@ impl HttpProxyService {
|
||||
});
|
||||
|
||||
// Save retry state before consuming parts/body (for bodyless requests like GET)
|
||||
// Clone BEFORE removing Host — H1 fallback needs Host header
|
||||
let retry_state = if body.is_end_stream() {
|
||||
Some((parts.method.clone(), upstream_headers.clone()))
|
||||
} else {
|
||||
None
|
||||
};
|
||||
|
||||
// Remove Host header for H2 — :authority pseudo-header (from URI) is sufficient
|
||||
upstream_headers.remove("host");
|
||||
|
||||
// Build and send the h2 request inline (don't register in pool yet —
|
||||
// we need to verify the request actually succeeds first, because some
|
||||
// backends advertise h2 via ALPN but don't speak the h2 binary protocol).
|
||||
let scheme = if upstream.use_tls { "https" } else { "http" };
|
||||
let authority = if domain != "-" { domain } else { upstream.host.as_str() };
|
||||
let h2_uri = format!("{}://{}{}", scheme, authority, upstream_path);
|
||||
let mut upstream_req = Request::builder()
|
||||
.method(parts.method)
|
||||
.uri(upstream_path);
|
||||
.uri(&h2_uri);
|
||||
|
||||
if let Some(headers) = upstream_req.headers_mut() {
|
||||
*headers = upstream_headers;
|
||||
@@ -1086,10 +1249,14 @@ impl HttpProxyService {
|
||||
Err(e) => {
|
||||
// H2 request failed — backend advertises h2 via ALPN but doesn't
|
||||
// actually speak it. Update cache so future requests use H1.
|
||||
let bk = format!("{}:{}", upstream.host, upstream.port);
|
||||
warn!(
|
||||
"Auto-detect: H2 request failed for {}:{}, falling back to H1: {}",
|
||||
upstream.host, upstream.port, e
|
||||
backend = %bk,
|
||||
domain = %domain,
|
||||
error = %e,
|
||||
"Auto-detect: H2 request failed, falling back to H1"
|
||||
);
|
||||
self.metrics.backend_h2_failure(&bk);
|
||||
let cache_key = crate::protocol_cache::ProtocolCacheKey {
|
||||
host: upstream.host.clone(),
|
||||
port: upstream.port,
|
||||
@@ -1099,7 +1266,7 @@ impl HttpProxyService {
|
||||
|
||||
// Retry as H1 for bodyless requests; return 502 for requests with bodies
|
||||
if let Some((method, headers)) = retry_state {
|
||||
match self.reconnect_backend(upstream).await {
|
||||
match self.reconnect_backend(upstream, domain).await {
|
||||
Some(fallback_backend) => {
|
||||
let h1_pool_key = crate::connection_pool::PoolKey {
|
||||
host: upstream.host.clone(),
|
||||
@@ -1108,10 +1275,13 @@ impl HttpProxyService {
|
||||
h2: false,
|
||||
};
|
||||
let fallback_io = TokioIo::new(fallback_backend);
|
||||
self.forward_h1_empty_body(
|
||||
let result = self.forward_h1_empty_body(
|
||||
fallback_io, method, headers, upstream_path,
|
||||
route, route_id, source_ip, &h1_pool_key,
|
||||
).await
|
||||
route, route_id, source_ip, &h1_pool_key, domain,
|
||||
).await;
|
||||
// Close the reconnected backend connection (opened in reconnect_backend)
|
||||
self.metrics.backend_connection_closed(&bk);
|
||||
result
|
||||
}
|
||||
None => {
|
||||
Ok(error_response(StatusCode::BAD_GATEWAY, "Backend unavailable after H2 fallback"))
|
||||
@@ -1123,13 +1293,18 @@ impl HttpProxyService {
|
||||
}
|
||||
}
|
||||
}
|
||||
Err(e) => {
|
||||
Ok(Err(e)) => {
|
||||
// H2 handshake truly failed — fall back to H1
|
||||
// Body is NOT consumed yet, so we can retry the full request.
|
||||
let bk = format!("{}:{}", upstream.host, upstream.port);
|
||||
warn!(
|
||||
"H2 handshake failed for {}:{}, falling back to H1: {}",
|
||||
upstream.host, upstream.port, e
|
||||
backend = %bk,
|
||||
domain = %domain,
|
||||
error = %e,
|
||||
"H2 handshake failed, falling back to H1"
|
||||
);
|
||||
self.metrics.backend_h2_failure(&bk);
|
||||
self.metrics.backend_handshake_error(&bk);
|
||||
|
||||
// Update cache to H1 so subsequent requests skip H2
|
||||
let cache_key = crate::protocol_cache::ProtocolCacheKey {
|
||||
@@ -1140,7 +1315,7 @@ impl HttpProxyService {
|
||||
self.protocol_cache.insert(cache_key, crate::protocol_cache::DetectedProtocol::H1);
|
||||
|
||||
// Reconnect for H1 (the original io was consumed by the failed h2 handshake)
|
||||
match self.reconnect_backend(upstream).await {
|
||||
match self.reconnect_backend(upstream, domain).await {
|
||||
Some(fallback_backend) => {
|
||||
let h1_pool_key = crate::connection_pool::PoolKey {
|
||||
host: upstream.host.clone(),
|
||||
@@ -1149,10 +1324,13 @@ impl HttpProxyService {
|
||||
h2: false,
|
||||
};
|
||||
let fallback_io = TokioIo::new(fallback_backend);
|
||||
self.forward_h1(
|
||||
let result = self.forward_h1(
|
||||
fallback_io, parts, body, upstream_headers, upstream_path,
|
||||
upstream, route, route_id, source_ip, &h1_pool_key,
|
||||
).await
|
||||
upstream, route, route_id, source_ip, &h1_pool_key, domain,
|
||||
).await;
|
||||
// Close the reconnected backend connection (opened in reconnect_backend)
|
||||
self.metrics.backend_connection_closed(&bk);
|
||||
result
|
||||
}
|
||||
None => {
|
||||
Ok(error_response(StatusCode::BAD_GATEWAY, "Backend unavailable after H2 fallback"))
|
||||
@@ -1174,14 +1352,17 @@ impl HttpProxyService {
|
||||
route_id: Option<&str>,
|
||||
source_ip: &str,
|
||||
pool_key: &crate::connection_pool::PoolKey,
|
||||
domain: &str,
|
||||
) -> Result<Response<BoxBody<Bytes, hyper::Error>>, hyper::Error> {
|
||||
let backend_key = format!("{}:{}", pool_key.host, pool_key.port);
|
||||
let (mut sender, conn): (
|
||||
hyper::client::conn::http1::SendRequest<BoxBody<Bytes, hyper::Error>>,
|
||||
hyper::client::conn::http1::Connection<TokioIo<BackendStream>, BoxBody<Bytes, hyper::Error>>,
|
||||
) = match hyper::client::conn::http1::handshake(io).await {
|
||||
Ok(h) => h,
|
||||
Err(e) => {
|
||||
error!("H1 fallback: handshake failed: {}", e);
|
||||
error!(backend = %backend_key, domain = %domain, error = %e, "H1 fallback: handshake failed");
|
||||
self.metrics.backend_handshake_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend H1 fallback handshake failed"));
|
||||
}
|
||||
};
|
||||
@@ -1209,7 +1390,8 @@ impl HttpProxyService {
|
||||
let upstream_response = match sender.send_request(upstream_req).await {
|
||||
Ok(resp) => resp,
|
||||
Err(e) => {
|
||||
error!("H1 fallback: request failed: {}", e);
|
||||
error!(backend = %backend_key, domain = %domain, error = %e, "H1 fallback: request failed");
|
||||
self.metrics.backend_request_error(&backend_key);
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend H1 fallback request failed"));
|
||||
}
|
||||
};
|
||||
@@ -1224,19 +1406,27 @@ impl HttpProxyService {
|
||||
async fn reconnect_backend(
|
||||
&self,
|
||||
upstream: &crate::upstream_selector::UpstreamSelection,
|
||||
domain: &str,
|
||||
) -> Option<BackendStream> {
|
||||
let backend_key = format!("{}:{}", upstream.host, upstream.port);
|
||||
let reconnect_start = std::time::Instant::now();
|
||||
if upstream.use_tls {
|
||||
match tokio::time::timeout(
|
||||
self.connect_timeout,
|
||||
connect_tls_backend(&self.backend_tls_config, &upstream.host, upstream.port),
|
||||
).await {
|
||||
Ok(Ok(tls)) => Some(BackendStream::Tls(tls)),
|
||||
Ok(Ok(tls)) => {
|
||||
self.metrics.backend_connection_opened(&backend_key, reconnect_start.elapsed());
|
||||
Some(BackendStream::Tls(tls))
|
||||
}
|
||||
Ok(Err(e)) => {
|
||||
error!("H1 fallback: TLS reconnect failed for {}:{}: {}", upstream.host, upstream.port, e);
|
||||
error!(backend = %backend_key, domain = %domain, error = %e, "H1 fallback: TLS reconnect failed");
|
||||
self.metrics.backend_connect_error(&backend_key);
|
||||
None
|
||||
}
|
||||
Err(_) => {
|
||||
error!("H1 fallback: TLS reconnect timeout for {}:{}", upstream.host, upstream.port);
|
||||
error!(backend = %backend_key, domain = %domain, "H1 fallback: TLS reconnect timeout");
|
||||
self.metrics.backend_connect_error(&backend_key);
|
||||
None
|
||||
}
|
||||
}
|
||||
@@ -1250,14 +1440,17 @@ impl HttpProxyService {
|
||||
let _ = socket2::SockRef::from(&s).set_tcp_keepalive(
|
||||
&socket2::TcpKeepalive::new().with_time(std::time::Duration::from_secs(60))
|
||||
);
|
||||
self.metrics.backend_connection_opened(&backend_key, reconnect_start.elapsed());
|
||||
Some(BackendStream::Plain(s))
|
||||
}
|
||||
Ok(Err(e)) => {
|
||||
error!("H1 fallback: reconnect failed for {}:{}: {}", upstream.host, upstream.port, e);
|
||||
error!(backend = %backend_key, domain = %domain, error = %e, "H1 fallback: TCP reconnect failed");
|
||||
self.metrics.backend_connect_error(&backend_key);
|
||||
None
|
||||
}
|
||||
Err(_) => {
|
||||
error!("H1 fallback: reconnect timeout for {}:{}", upstream.host, upstream.port);
|
||||
error!(backend = %backend_key, domain = %domain, "H1 fallback: TCP reconnect timeout");
|
||||
self.metrics.backend_connect_error(&backend_key);
|
||||
None
|
||||
}
|
||||
}
|
||||
@@ -1276,10 +1469,23 @@ impl HttpProxyService {
|
||||
route_id: Option<&str>,
|
||||
source_ip: &str,
|
||||
pool_key: Option<&crate::connection_pool::PoolKey>,
|
||||
domain: &str,
|
||||
) -> Result<Response<BoxBody<Bytes, hyper::Error>>, hyper::Error> {
|
||||
// Build absolute URI for H2 pseudo-headers (:scheme, :authority)
|
||||
// Use the requested domain as authority (not backend address) so :authority matches Host header
|
||||
let scheme = if pool_key.map(|pk| pk.use_tls).unwrap_or(false) { "https" } else { "http" };
|
||||
let authority = if domain != "-" { domain } else {
|
||||
pool_key.map(|pk| pk.host.as_str()).unwrap_or("localhost")
|
||||
};
|
||||
let h2_uri = format!("{}://{}{}", scheme, authority, upstream_path);
|
||||
let mut upstream_req = Request::builder()
|
||||
.method(parts.method)
|
||||
.uri(upstream_path);
|
||||
.uri(&h2_uri);
|
||||
|
||||
// Remove Host header for H2 — :authority pseudo-header (from URI) is sufficient
|
||||
// Having both Host and :authority causes nginx to return 400
|
||||
let mut upstream_headers = upstream_headers;
|
||||
upstream_headers.remove("host");
|
||||
|
||||
if let Some(headers) = upstream_req.headers_mut() {
|
||||
*headers = upstream_headers;
|
||||
@@ -1300,10 +1506,14 @@ impl HttpProxyService {
|
||||
let upstream_response = match sender.send_request(upstream_req).await {
|
||||
Ok(resp) => resp,
|
||||
Err(e) => {
|
||||
error!("HTTP/2 upstream request failed: {}", e);
|
||||
// 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");
|
||||
self.metrics.backend_request_error(&bk);
|
||||
self.connection_pool.remove_h2(key);
|
||||
} else {
|
||||
error!(domain = %domain, error = %e, "Backend H2 request failed");
|
||||
}
|
||||
return Ok(error_response(StatusCode::BAD_GATEWAY, "Backend H2 request failed"));
|
||||
}
|
||||
|
||||
@@ -3,6 +3,7 @@ use serde::{Deserialize, Serialize};
|
||||
use std::collections::HashSet;
|
||||
use std::sync::atomic::{AtomicU64, Ordering};
|
||||
use std::sync::Mutex;
|
||||
use std::time::Duration;
|
||||
|
||||
use crate::throughput::{ThroughputSample, ThroughputTracker};
|
||||
|
||||
@@ -20,6 +21,7 @@ pub struct Metrics {
|
||||
pub throughput_recent_out_bytes_per_sec: u64,
|
||||
pub routes: std::collections::HashMap<String, RouteMetrics>,
|
||||
pub ips: std::collections::HashMap<String, IpMetrics>,
|
||||
pub backends: std::collections::HashMap<String, BackendMetrics>,
|
||||
pub throughput_history: Vec<ThroughputSample>,
|
||||
pub total_http_requests: u64,
|
||||
pub http_requests_per_sec: u64,
|
||||
@@ -52,6 +54,23 @@ pub struct IpMetrics {
|
||||
pub throughput_out_bytes_per_sec: u64,
|
||||
}
|
||||
|
||||
/// Per-backend metrics (keyed by "host:port").
|
||||
#[derive(Debug, Clone, Serialize, Deserialize)]
|
||||
#[serde(rename_all = "camelCase")]
|
||||
pub struct BackendMetrics {
|
||||
pub active_connections: u64,
|
||||
pub total_connections: u64,
|
||||
pub protocol: String,
|
||||
pub connect_errors: u64,
|
||||
pub handshake_errors: u64,
|
||||
pub request_errors: u64,
|
||||
pub total_connect_time_us: u64,
|
||||
pub connect_count: u64,
|
||||
pub pool_hits: u64,
|
||||
pub pool_misses: u64,
|
||||
pub h2_failures: u64,
|
||||
}
|
||||
|
||||
/// Statistics snapshot.
|
||||
#[derive(Debug, Clone, Serialize, Deserialize)]
|
||||
#[serde(rename_all = "camelCase")]
|
||||
@@ -69,6 +88,9 @@ const DEFAULT_RETENTION_SECONDS: usize = 3600;
|
||||
/// Maximum number of IPs to include in a snapshot (top by active connections).
|
||||
const MAX_IPS_IN_SNAPSHOT: usize = 100;
|
||||
|
||||
/// Maximum number of backends to include in a snapshot (top by total connections).
|
||||
const MAX_BACKENDS_IN_SNAPSHOT: usize = 100;
|
||||
|
||||
/// Metrics collector tracking connections and throughput.
|
||||
///
|
||||
/// Design: The hot path (`record_bytes`) is entirely lock-free — it only touches
|
||||
@@ -96,6 +118,19 @@ pub struct MetricsCollector {
|
||||
ip_pending_tp: DashMap<String, (AtomicU64, AtomicU64)>,
|
||||
ip_throughput: DashMap<String, Mutex<ThroughputTracker>>,
|
||||
|
||||
// ── Per-backend tracking (keyed by "host:port") ──
|
||||
backend_active: DashMap<String, AtomicU64>,
|
||||
backend_total: DashMap<String, AtomicU64>,
|
||||
backend_protocol: DashMap<String, String>,
|
||||
backend_connect_errors: DashMap<String, AtomicU64>,
|
||||
backend_handshake_errors: DashMap<String, AtomicU64>,
|
||||
backend_request_errors: DashMap<String, AtomicU64>,
|
||||
backend_connect_time_us: DashMap<String, AtomicU64>,
|
||||
backend_connect_count: DashMap<String, AtomicU64>,
|
||||
backend_pool_hits: DashMap<String, AtomicU64>,
|
||||
backend_pool_misses: DashMap<String, AtomicU64>,
|
||||
backend_h2_failures: DashMap<String, AtomicU64>,
|
||||
|
||||
// ── HTTP request tracking ──
|
||||
total_http_requests: AtomicU64,
|
||||
pending_http_requests: AtomicU64,
|
||||
@@ -134,6 +169,17 @@ impl MetricsCollector {
|
||||
ip_bytes_out: DashMap::new(),
|
||||
ip_pending_tp: DashMap::new(),
|
||||
ip_throughput: DashMap::new(),
|
||||
backend_active: DashMap::new(),
|
||||
backend_total: DashMap::new(),
|
||||
backend_protocol: DashMap::new(),
|
||||
backend_connect_errors: DashMap::new(),
|
||||
backend_handshake_errors: DashMap::new(),
|
||||
backend_request_errors: DashMap::new(),
|
||||
backend_connect_time_us: DashMap::new(),
|
||||
backend_connect_count: DashMap::new(),
|
||||
backend_pool_hits: DashMap::new(),
|
||||
backend_pool_misses: DashMap::new(),
|
||||
backend_h2_failures: DashMap::new(),
|
||||
total_http_requests: AtomicU64::new(0),
|
||||
pending_http_requests: AtomicU64::new(0),
|
||||
http_request_throughput: Mutex::new(ThroughputTracker::new(retention_seconds)),
|
||||
@@ -268,6 +314,113 @@ impl MetricsCollector {
|
||||
self.pending_http_requests.fetch_add(1, Ordering::Relaxed);
|
||||
}
|
||||
|
||||
// ── Per-backend recording methods ──
|
||||
|
||||
/// Record a successful backend connection with its connect duration.
|
||||
pub fn backend_connection_opened(&self, key: &str, connect_time: Duration) {
|
||||
self.backend_active
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
self.backend_total
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
self.backend_connect_time_us
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(connect_time.as_micros() as u64, Ordering::Relaxed);
|
||||
self.backend_connect_count
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
}
|
||||
|
||||
/// Record a backend connection closing.
|
||||
pub fn backend_connection_closed(&self, key: &str) {
|
||||
if let Some(counter) = self.backend_active.get(key) {
|
||||
let val = counter.load(Ordering::Relaxed);
|
||||
if val > 0 {
|
||||
counter.fetch_sub(1, Ordering::Relaxed);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Record a backend connect error (TCP or TLS connect failure/timeout).
|
||||
pub fn backend_connect_error(&self, key: &str) {
|
||||
self.backend_connect_errors
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
}
|
||||
|
||||
/// Record a backend handshake error (H1 or H2 handshake failure).
|
||||
pub fn backend_handshake_error(&self, key: &str) {
|
||||
self.backend_handshake_errors
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
}
|
||||
|
||||
/// Record a backend request error (send_request failure).
|
||||
pub fn backend_request_error(&self, key: &str) {
|
||||
self.backend_request_errors
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
}
|
||||
|
||||
/// Record a connection pool hit for a backend.
|
||||
pub fn backend_pool_hit(&self, key: &str) {
|
||||
self.backend_pool_hits
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
}
|
||||
|
||||
/// Record a connection pool miss for a backend.
|
||||
pub fn backend_pool_miss(&self, key: &str) {
|
||||
self.backend_pool_misses
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
}
|
||||
|
||||
/// Record an H2 failure (h2 attempted but fell back to h1).
|
||||
pub fn backend_h2_failure(&self, key: &str) {
|
||||
self.backend_h2_failures
|
||||
.entry(key.to_string())
|
||||
.or_insert_with(|| AtomicU64::new(0))
|
||||
.fetch_add(1, Ordering::Relaxed);
|
||||
}
|
||||
|
||||
/// Set the protocol in use for a backend ("h1" or "h2").
|
||||
pub fn set_backend_protocol(&self, key: &str, protocol: &str) {
|
||||
self.backend_protocol
|
||||
.entry(key.to_string())
|
||||
.and_modify(|v| {
|
||||
if v != protocol {
|
||||
*v = protocol.to_string();
|
||||
}
|
||||
})
|
||||
.or_insert_with(|| protocol.to_string());
|
||||
}
|
||||
|
||||
/// Remove per-backend metrics for backends no longer in any route target.
|
||||
pub fn retain_backends(&self, active_backends: &HashSet<String>) {
|
||||
self.backend_active.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_total.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_protocol.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_connect_errors.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_handshake_errors.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_request_errors.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_connect_time_us.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_connect_count.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_pool_hits.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_pool_misses.retain(|k, _| active_backends.contains(k));
|
||||
self.backend_h2_failures.retain(|k, _| active_backends.contains(k));
|
||||
}
|
||||
|
||||
/// Take a throughput sample on all trackers (cold path, call at 1Hz or configured interval).
|
||||
///
|
||||
/// Drains the lock-free pending counters and feeds the accumulated bytes
|
||||
@@ -488,6 +641,72 @@ impl MetricsCollector {
|
||||
});
|
||||
}
|
||||
|
||||
// Collect per-backend metrics, capped at top MAX_BACKENDS_IN_SNAPSHOT by total connections
|
||||
let mut backend_entries: Vec<(String, BackendMetrics)> = Vec::new();
|
||||
for entry in self.backend_total.iter() {
|
||||
let key = entry.key().clone();
|
||||
let total = entry.value().load(Ordering::Relaxed);
|
||||
let active = self.backend_active
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
let protocol = self.backend_protocol
|
||||
.get(&key)
|
||||
.map(|v| v.value().clone())
|
||||
.unwrap_or_else(|| "unknown".to_string());
|
||||
let connect_errors = self.backend_connect_errors
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
let handshake_errors = self.backend_handshake_errors
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
let request_errors = self.backend_request_errors
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
let total_connect_time_us = self.backend_connect_time_us
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
let connect_count = self.backend_connect_count
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
let pool_hits = self.backend_pool_hits
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
let pool_misses = self.backend_pool_misses
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
let h2_failures = self.backend_h2_failures
|
||||
.get(&key)
|
||||
.map(|c| c.load(Ordering::Relaxed))
|
||||
.unwrap_or(0);
|
||||
|
||||
backend_entries.push((key, BackendMetrics {
|
||||
active_connections: active,
|
||||
total_connections: total,
|
||||
protocol,
|
||||
connect_errors,
|
||||
handshake_errors,
|
||||
request_errors,
|
||||
total_connect_time_us,
|
||||
connect_count,
|
||||
pool_hits,
|
||||
pool_misses,
|
||||
h2_failures,
|
||||
}));
|
||||
}
|
||||
// Sort by total connections descending, then cap
|
||||
backend_entries.sort_by(|a, b| b.1.total_connections.cmp(&a.1.total_connections));
|
||||
backend_entries.truncate(MAX_BACKENDS_IN_SNAPSHOT);
|
||||
|
||||
let backends: std::collections::HashMap<String, BackendMetrics> = backend_entries.into_iter().collect();
|
||||
|
||||
// HTTP request rates
|
||||
let (http_rps, http_rps_recent) = self.http_request_throughput
|
||||
.lock()
|
||||
@@ -509,6 +728,7 @@ impl MetricsCollector {
|
||||
throughput_recent_out_bytes_per_sec: global_recent_out,
|
||||
routes,
|
||||
ips,
|
||||
backends,
|
||||
throughput_history,
|
||||
total_http_requests: self.total_http_requests.load(Ordering::Relaxed),
|
||||
http_requests_per_sec: http_rps,
|
||||
@@ -805,4 +1025,120 @@ mod tests {
|
||||
assert_eq!(snapshot.throughput_history[0].bytes_in, 100);
|
||||
assert_eq!(snapshot.throughput_history[4].bytes_in, 500);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_backend_metrics_basic() {
|
||||
let collector = MetricsCollector::new();
|
||||
let key = "backend1:8080";
|
||||
|
||||
// Open connections with timing
|
||||
collector.backend_connection_opened(key, Duration::from_millis(15));
|
||||
collector.backend_connection_opened(key, Duration::from_millis(25));
|
||||
|
||||
assert_eq!(collector.backend_active.get(key).unwrap().load(Ordering::Relaxed), 2);
|
||||
assert_eq!(collector.backend_total.get(key).unwrap().load(Ordering::Relaxed), 2);
|
||||
assert_eq!(collector.backend_connect_count.get(key).unwrap().load(Ordering::Relaxed), 2);
|
||||
// 15ms + 25ms = 40ms = 40_000us
|
||||
assert_eq!(collector.backend_connect_time_us.get(key).unwrap().load(Ordering::Relaxed), 40_000);
|
||||
|
||||
// Close one
|
||||
collector.backend_connection_closed(key);
|
||||
assert_eq!(collector.backend_active.get(key).unwrap().load(Ordering::Relaxed), 1);
|
||||
// total stays
|
||||
assert_eq!(collector.backend_total.get(key).unwrap().load(Ordering::Relaxed), 2);
|
||||
|
||||
// Record errors
|
||||
collector.backend_connect_error(key);
|
||||
collector.backend_handshake_error(key);
|
||||
collector.backend_request_error(key);
|
||||
collector.backend_h2_failure(key);
|
||||
collector.backend_pool_hit(key);
|
||||
collector.backend_pool_hit(key);
|
||||
collector.backend_pool_miss(key);
|
||||
|
||||
assert_eq!(collector.backend_connect_errors.get(key).unwrap().load(Ordering::Relaxed), 1);
|
||||
assert_eq!(collector.backend_handshake_errors.get(key).unwrap().load(Ordering::Relaxed), 1);
|
||||
assert_eq!(collector.backend_request_errors.get(key).unwrap().load(Ordering::Relaxed), 1);
|
||||
assert_eq!(collector.backend_h2_failures.get(key).unwrap().load(Ordering::Relaxed), 1);
|
||||
assert_eq!(collector.backend_pool_hits.get(key).unwrap().load(Ordering::Relaxed), 2);
|
||||
assert_eq!(collector.backend_pool_misses.get(key).unwrap().load(Ordering::Relaxed), 1);
|
||||
|
||||
// Protocol
|
||||
collector.set_backend_protocol(key, "h1");
|
||||
assert_eq!(collector.backend_protocol.get(key).unwrap().value(), "h1");
|
||||
collector.set_backend_protocol(key, "h2");
|
||||
assert_eq!(collector.backend_protocol.get(key).unwrap().value(), "h2");
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_backend_metrics_in_snapshot() {
|
||||
let collector = MetricsCollector::new();
|
||||
|
||||
collector.backend_connection_opened("b1:443", Duration::from_millis(10));
|
||||
collector.backend_connection_opened("b2:8080", Duration::from_millis(20));
|
||||
collector.set_backend_protocol("b1:443", "h2");
|
||||
collector.set_backend_protocol("b2:8080", "h1");
|
||||
collector.backend_connect_error("b1:443");
|
||||
|
||||
let snapshot = collector.snapshot();
|
||||
assert_eq!(snapshot.backends.len(), 2);
|
||||
|
||||
let b1 = snapshot.backends.get("b1:443").unwrap();
|
||||
assert_eq!(b1.active_connections, 1);
|
||||
assert_eq!(b1.total_connections, 1);
|
||||
assert_eq!(b1.protocol, "h2");
|
||||
assert_eq!(b1.connect_errors, 1);
|
||||
assert_eq!(b1.total_connect_time_us, 10_000);
|
||||
assert_eq!(b1.connect_count, 1);
|
||||
|
||||
let b2 = snapshot.backends.get("b2:8080").unwrap();
|
||||
assert_eq!(b2.protocol, "h1");
|
||||
assert_eq!(b2.connect_errors, 0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_retain_backends_prunes_stale() {
|
||||
let collector = MetricsCollector::new();
|
||||
|
||||
collector.backend_connection_opened("active:443", Duration::from_millis(5));
|
||||
collector.backend_connection_opened("stale:8080", Duration::from_millis(10));
|
||||
collector.set_backend_protocol("active:443", "h1");
|
||||
collector.set_backend_protocol("stale:8080", "h2");
|
||||
collector.backend_connect_error("stale:8080");
|
||||
|
||||
let active = HashSet::from(["active:443".to_string()]);
|
||||
collector.retain_backends(&active);
|
||||
|
||||
// active:443 should still exist
|
||||
assert!(collector.backend_total.get("active:443").is_some());
|
||||
assert!(collector.backend_protocol.get("active:443").is_some());
|
||||
|
||||
// stale:8080 should be fully removed
|
||||
assert!(collector.backend_active.get("stale:8080").is_none());
|
||||
assert!(collector.backend_total.get("stale:8080").is_none());
|
||||
assert!(collector.backend_protocol.get("stale:8080").is_none());
|
||||
assert!(collector.backend_connect_errors.get("stale:8080").is_none());
|
||||
assert!(collector.backend_connect_time_us.get("stale:8080").is_none());
|
||||
assert!(collector.backend_connect_count.get("stale:8080").is_none());
|
||||
assert!(collector.backend_pool_hits.get("stale:8080").is_none());
|
||||
assert!(collector.backend_pool_misses.get("stale:8080").is_none());
|
||||
assert!(collector.backend_h2_failures.get("stale:8080").is_none());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_backend_connection_closed_saturates() {
|
||||
let collector = MetricsCollector::new();
|
||||
let key = "b:80";
|
||||
|
||||
// Close without opening — should not underflow
|
||||
collector.backend_connection_closed(key);
|
||||
// No entry created
|
||||
assert!(collector.backend_active.get(key).is_none());
|
||||
|
||||
// Open one, close two — should saturate at 0
|
||||
collector.backend_connection_opened(key, Duration::from_millis(1));
|
||||
collector.backend_connection_closed(key);
|
||||
collector.backend_connection_closed(key);
|
||||
assert_eq!(collector.backend_active.get(key).unwrap().load(Ordering::Relaxed), 0);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -603,6 +603,31 @@ impl RustProxy {
|
||||
.collect();
|
||||
self.metrics.retain_routes(&active_route_ids);
|
||||
|
||||
// Prune per-backend metrics for backends no longer in any route target.
|
||||
// For PortSpec::Preserve routes, expand across all listening ports since
|
||||
// the actual runtime port depends on the incoming connection.
|
||||
let listening_ports = self.get_listening_ports();
|
||||
let active_backends: HashSet<String> = routes.iter()
|
||||
.filter_map(|r| r.action.targets.as_ref())
|
||||
.flat_map(|targets| targets.iter())
|
||||
.flat_map(|target| {
|
||||
let hosts: Vec<String> = target.host.to_vec().into_iter().map(|s| s.to_string()).collect();
|
||||
match &target.port {
|
||||
rustproxy_config::PortSpec::Fixed(p) => {
|
||||
hosts.into_iter().map(|h| format!("{}:{}", h, p)).collect::<Vec<_>>()
|
||||
}
|
||||
_ => {
|
||||
// Preserve/special: expand across all listening ports
|
||||
let lp = &listening_ports;
|
||||
hosts.into_iter()
|
||||
.flat_map(|h| lp.iter().map(move |p| format!("{}:{}", h, *p)))
|
||||
.collect::<Vec<_>>()
|
||||
}
|
||||
}
|
||||
})
|
||||
.collect();
|
||||
self.metrics.retain_backends(&active_backends);
|
||||
|
||||
// Atomically swap the route table
|
||||
let new_manager = Arc::new(new_manager);
|
||||
self.route_table.store(Arc::clone(&new_manager));
|
||||
|
||||
@@ -3,6 +3,6 @@
|
||||
*/
|
||||
export const commitinfo = {
|
||||
name: '@push.rocks/smartproxy',
|
||||
version: '25.9.3',
|
||||
version: '25.10.7',
|
||||
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.'
|
||||
}
|
||||
|
||||
@@ -67,6 +67,13 @@ export interface IMetrics {
|
||||
connections(): number;
|
||||
};
|
||||
|
||||
// Backend metrics
|
||||
backends: {
|
||||
byBackend(): Map<string, IBackendMetrics>;
|
||||
protocols(): Map<string, string>;
|
||||
topByErrors(limit?: number): Array<{ backend: string; errors: number }>;
|
||||
};
|
||||
|
||||
// Performance metrics
|
||||
percentiles: {
|
||||
connectionDuration(): { p50: number; p95: number; p99: number };
|
||||
@@ -98,6 +105,21 @@ export interface IMetricsConfig {
|
||||
prometheusPrefix: string; // Default: smartproxy_
|
||||
}
|
||||
|
||||
/**
|
||||
* Per-backend metrics
|
||||
*/
|
||||
export interface IBackendMetrics {
|
||||
protocol: string;
|
||||
activeConnections: number;
|
||||
totalConnections: number;
|
||||
connectErrors: number;
|
||||
handshakeErrors: number;
|
||||
requestErrors: number;
|
||||
avgConnectTimeMs: number;
|
||||
poolHitRate: number;
|
||||
h2Failures: number;
|
||||
}
|
||||
|
||||
/**
|
||||
* Internal interface for connection byte tracking
|
||||
*/
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
import type { IMetrics, IThroughputData, IThroughputHistoryPoint } from './models/metrics-types.js';
|
||||
import type { IMetrics, IBackendMetrics, IThroughputData, IThroughputHistoryPoint } from './models/metrics-types.js';
|
||||
import type { RustProxyBridge } from './rust-proxy-bridge.js';
|
||||
|
||||
/**
|
||||
@@ -169,6 +169,55 @@ export class RustMetricsAdapter implements IMetrics {
|
||||
},
|
||||
};
|
||||
|
||||
public backends = {
|
||||
byBackend: (): Map<string, IBackendMetrics> => {
|
||||
const result = new Map<string, IBackendMetrics>();
|
||||
if (this.cache?.backends) {
|
||||
for (const [key, bm] of Object.entries(this.cache.backends)) {
|
||||
const m = bm as any;
|
||||
const totalTimeUs = m.totalConnectTimeUs ?? 0;
|
||||
const count = m.connectCount ?? 0;
|
||||
const poolHits = m.poolHits ?? 0;
|
||||
const poolMisses = m.poolMisses ?? 0;
|
||||
const poolTotal = poolHits + poolMisses;
|
||||
result.set(key, {
|
||||
protocol: m.protocol ?? 'unknown',
|
||||
activeConnections: m.activeConnections ?? 0,
|
||||
totalConnections: m.totalConnections ?? 0,
|
||||
connectErrors: m.connectErrors ?? 0,
|
||||
handshakeErrors: m.handshakeErrors ?? 0,
|
||||
requestErrors: m.requestErrors ?? 0,
|
||||
avgConnectTimeMs: count > 0 ? (totalTimeUs / count) / 1000 : 0,
|
||||
poolHitRate: poolTotal > 0 ? poolHits / poolTotal : 0,
|
||||
h2Failures: m.h2Failures ?? 0,
|
||||
});
|
||||
}
|
||||
}
|
||||
return result;
|
||||
},
|
||||
protocols: (): Map<string, string> => {
|
||||
const result = new Map<string, string>();
|
||||
if (this.cache?.backends) {
|
||||
for (const [key, bm] of Object.entries(this.cache.backends)) {
|
||||
result.set(key, (bm as any).protocol ?? 'unknown');
|
||||
}
|
||||
}
|
||||
return result;
|
||||
},
|
||||
topByErrors: (limit: number = 10): Array<{ backend: string; errors: number }> => {
|
||||
const result: Array<{ backend: string; errors: number }> = [];
|
||||
if (this.cache?.backends) {
|
||||
for (const [key, bm] of Object.entries(this.cache.backends)) {
|
||||
const m = bm as any;
|
||||
const errors = (m.connectErrors ?? 0) + (m.handshakeErrors ?? 0) + (m.requestErrors ?? 0);
|
||||
if (errors > 0) result.push({ backend: key, errors });
|
||||
}
|
||||
}
|
||||
result.sort((a, b) => b.errors - a.errors);
|
||||
return result.slice(0, limit);
|
||||
},
|
||||
};
|
||||
|
||||
public percentiles = {
|
||||
connectionDuration: (): { p50: number; p95: number; p99: number } => {
|
||||
return { p50: 0, p95: 0, p99: 0 };
|
||||
|
||||
Reference in New Issue
Block a user