Keep Guardian v2 WebSocket pools warm with concurrent replenishment (#51378)

## Why

Serial connection opens let a stalled handshake delay healthy sockets. Waiting until the idle pool is empty to refill also leaves classification bursts using HTTP fallback.

## What changed

- Open up to four WebSockets concurrently and start replenishing when the idle pool falls to half capacity. Preserve successful in-flight opens when another open fails or times out.
- Make explicit prewarming wait for an existing refill before checking readiness.
- Add `ResponsesWebsocketClient::connect_with_connector` and reuse one connector per sampler, initializing TLS configuration on a blocking task while retaining per-connection authentication and destination policy checks.
- Record connection lease durations for admission, setup, and total time, tagged by transport.

## Testing

Add regression tests for a stalled handshake alongside a healthy socket and replenishment before the last idle socket is leased. Update cooldown, authentication, and transport retry tests to account for concurrent handshake completion.

GitOrigin-RevId: f4a8258977c1f2a58e00c6badc7b599280ef22ad
This commit is contained in:
jif
2026-10-06 14:42:57 +00:00
committed by copyberry
parent c0c230e673
commit 73178e7ca6
9 changed files with 343 additions and 74 deletions
@@ -349,6 +349,8 @@ impl ResponsesWebsocketConnection {
}
}
mod connector;
/// Client for connecting to the Responses WebSocket endpoint for one provider.
pub struct ResponsesWebsocketClient {
provider: Provider,
@@ -385,12 +387,6 @@ impl ResponsesWebsocketClient {
Self { provider, auth }
}
#[instrument(
name = "responses_websocket.connect",
level = "info",
skip_all,
fields(transport = "responses_websocket", api.path = "/responses")
)]
pub async fn connect(
&self,
http_client_factory: &HttpClientFactory,
@@ -399,24 +395,16 @@ impl ResponsesWebsocketClient {
turn_state: Option<Arc<OnceLock<String>>>,
telemetry: Option<Arc<dyn WebsocketTelemetry>>,
) -> Result<ResponsesWebsocketConnection, ApiError> {
let ws_url = self
.provider
.websocket_url_for_path("/responses")
.map_err(|err| ApiError::Stream(format!("failed to build websocket URL: {err}")))?;
let mut headers =
merge_request_headers(&self.provider.headers, extra_headers, default_headers);
self.auth.add_auth_headers(&mut headers);
let (stream, _status, server_reasoning_included, server_model) =
connect_websocket(ws_url, headers, http_client_factory, turn_state.clone()).await?;
Ok(ResponsesWebsocketConnection::new(
stream,
self.provider.stream_idle_timeout,
server_reasoning_included,
server_model,
let connector = WebSocketConnector::new(http_client_factory)
.map_err(|err| ApiError::Stream(format!("failed to configure websocket TLS: {err}")))?;
self.connect_with_connector(
&connector,
extra_headers,
default_headers,
turn_state,
telemetry,
))
)
.await
}
/// Opens a WebSocket connection long enough to validate the upgrade response.
@@ -442,10 +430,12 @@ impl ResponsesWebsocketClient {
merge_request_headers(&self.provider.headers, extra_headers, default_headers);
self.auth.add_auth_headers(&mut headers);
let connector = WebSocketConnector::new(http_client_factory)
.map_err(|err| ApiError::Stream(format!("failed to configure websocket TLS: {err}")))?;
let (mut stream, status, reasoning_included, server_model) = connect_websocket(
ws_url.clone(),
headers,
http_client_factory,
&connector,
/*turn_state*/ None,
)
.await?;
@@ -499,7 +489,7 @@ fn merge_request_headers(
async fn connect_websocket(
url: Url,
headers: HeaderMap,
http_client_factory: &HttpClientFactory,
connector: &WebSocketConnector,
turn_state: Option<Arc<OnceLock<String>>>,
) -> Result<(WsStream, StatusCode, bool, Option<String>), ApiError> {
info!("connecting to websocket: {url}");
@@ -510,8 +500,6 @@ async fn connect_websocket(
.map_err(|err| ApiError::Stream(format!("failed to build websocket request: {err}")))?;
request.headers_mut().extend(headers);
let connector = WebSocketConnector::new(http_client_factory)
.map_err(|err| ApiError::Stream(format!("failed to configure websocket TLS: {err}")))?;
let response = connector.connect(request, websocket_config()).await;
let (stream, response) = match response {
@@ -0,0 +1,49 @@
//! Establishes Responses sockets with caller-owned TLS and proxy configuration.
//! Connector reuse never caches authentication or bypasses per-connection destination policy.
use super::ResponsesWebsocketClient;
use super::ResponsesWebsocketConnection;
use super::connect_websocket;
use super::merge_request_headers;
use crate::ApiError;
use crate::WebsocketTelemetry;
use codex_websocket_client::WebSocketConnector;
use http::HeaderMap;
use std::sync::Arc;
use std::sync::OnceLock;
impl ResponsesWebsocketClient {
/// Opens a socket using the caller's reusable connector and this client's current auth.
/// The caller owns the lifetime of the connector's TLS configuration snapshot.
#[tracing::instrument(
name = "responses_websocket.connect",
level = "info",
skip_all,
fields(transport = "responses_websocket", api.path = "/responses")
)]
pub async fn connect_with_connector(
&self,
connector: &WebSocketConnector,
extra_headers: HeaderMap,
default_headers: HeaderMap,
turn_state: Option<Arc<OnceLock<String>>>,
telemetry: Option<Arc<dyn WebsocketTelemetry>>,
) -> Result<ResponsesWebsocketConnection, ApiError> {
let ws_url = self
.provider
.websocket_url_for_path("/responses")
.map_err(|err| ApiError::Stream(format!("failed to build websocket URL: {err}")))?;
let mut headers =
merge_request_headers(&self.provider.headers, extra_headers, default_headers);
self.auth.add_auth_headers(&mut headers);
let (stream, _status, server_reasoning_included, server_model) =
connect_websocket(ws_url, headers, connector, turn_state).await?;
Ok(ResponsesWebsocketConnection::new(
stream,
self.provider.stream_idle_timeout,
server_reasoning_included,
server_model,
telemetry,
))
}
}