mirror of
https://github.com/openai/codex.git
synced 2026-11-15 21:50:46 +00:00
Interrupt and drain canceled classifier responses before socket reuse (#53698)
## Why Canceled classifier requests need bounded cleanup that preserves received token usage and returns only safe WebSocket connections to the pool. ## What changed - Own classifier response streams and connection leases together, interrupting dropped responses with a one-second cleanup deadline. - Require a matching interruption acknowledgement and response completion before reusing a canceled response's socket. Retire sockets on rejection, invalid acknowledgements, or timeout while preserving any received completion and token usage. - Skip requests canceled before dispatch and return their unused sockets to the pool. - Distinguish cancellation from ordinary steering, which continues draining without requiring an interruption acknowledgement. ## Testing Add regression coverage for cancellation, acknowledgement ordering and failures, cleanup timeouts, socket reuse, and token accounting under backpressure. Extend steering tests to cover rejected and missing acknowledgements. GitOrigin-RevId: c4e7aac49d36cea2c2d0e9f2ba5f6147f67471dd
This commit is contained in:
@@ -117,7 +117,7 @@ fn normalize_image_detail(detail: &mut Option<ImageDetail>, model_info: &ModelIn
|
||||
|
||||
pub struct ResponseStream {
|
||||
pub(crate) rx_event: mpsc::Receiver<Result<ResponseEvent>>,
|
||||
pub(crate) interrupt: Option<oneshot::Sender<()>>,
|
||||
pub(crate) interrupt: Option<oneshot::Sender<codex_api::ResponseInterrupt>>,
|
||||
/// Signals the mapper task that the consumer stopped polling before the
|
||||
/// provider stream reached its own terminal event.
|
||||
pub(crate) consumer_dropped: CancellationToken,
|
||||
|
||||
@@ -2689,7 +2689,7 @@ async fn try_run_sampling_request(
|
||||
Ok(Err(_)) => {
|
||||
if let Some(interrupt) = stream.interrupt.take() {
|
||||
if step_context.settings.model_info.use_responses_lite {
|
||||
let _ = interrupt.send(());
|
||||
let _ = interrupt.send(codex_api::ResponseInterrupt::Drain);
|
||||
}
|
||||
// Drain the response normally before reusing its connection and history.
|
||||
preempt = CancellationToken::new();
|
||||
|
||||
@@ -1239,16 +1239,23 @@ async fn injected_response_item_reopens_turn_after_final_answer() {
|
||||
server.shutdown().await;
|
||||
}
|
||||
|
||||
#[test_case(false; "completed_reasoning")]
|
||||
#[test_case(true; "discarded_reasoning")]
|
||||
#[test_case(false, Some("response.interrupt.accepted"); "completed_reasoning")]
|
||||
#[test_case(true, Some("response.interrupt.accepted"); "discarded_reasoning")]
|
||||
#[test_case(false, Some("response.interrupt.failed"); "rejected_interrupt")]
|
||||
#[test_case(false, None; "missing_acknowledgement")]
|
||||
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
|
||||
async fn steer_interrupts_and_drains_websocket(discard_partial: bool) -> anyhow::Result<()> {
|
||||
async fn steer_interrupts_and_drains_websocket(
|
||||
discard_partial: bool,
|
||||
acknowledgement: Option<&str>,
|
||||
) -> anyhow::Result<()> {
|
||||
core_test_support::skip_if_no_network!(Ok(()));
|
||||
|
||||
let mut interrupted = ev_completed_with_tokens("resp-interrupted", /*total_tokens*/ 42);
|
||||
interrupted["type"] = json!("response.incomplete");
|
||||
interrupted["response"]["status"] = json!("incomplete");
|
||||
interrupted["response"]["incomplete_details"] = json!({"reason": "interrupted"});
|
||||
if acknowledgement == Some("response.interrupt.accepted") {
|
||||
interrupted["type"] = json!("response.incomplete");
|
||||
interrupted["response"]["status"] = json!("incomplete");
|
||||
interrupted["response"]["incomplete_details"] = json!({"reason": "interrupted"});
|
||||
}
|
||||
let server = responses::start_websocket_server(vec![vec![
|
||||
vec![ev_response_created("warm-1"), ev_completed("warm-1")],
|
||||
vec![
|
||||
@@ -1256,13 +1263,15 @@ async fn steer_interrupts_and_drains_websocket(discard_partial: bool) -> anyhow:
|
||||
ev_reasoning_item_added("reason-1", &["thinking"]),
|
||||
],
|
||||
// Finish the active response only after receiving response.interrupt.
|
||||
vec![
|
||||
json!({
|
||||
"type": "response.interrupt.accepted",
|
||||
"response_id": "resp-interrupted",
|
||||
"sequence_number": 3,
|
||||
[
|
||||
acknowledgement.map(|kind| {
|
||||
json!({
|
||||
"type": kind,
|
||||
"response_id": "resp-interrupted",
|
||||
"sequence_number": 3,
|
||||
})
|
||||
}),
|
||||
if discard_partial {
|
||||
Some(if discard_partial {
|
||||
json!({
|
||||
"type": "response.output_item.interrupted",
|
||||
"response_id": "resp-interrupted",
|
||||
@@ -1272,9 +1281,12 @@ async fn steer_interrupts_and_drains_websocket(discard_partial: bool) -> anyhow:
|
||||
})
|
||||
} else {
|
||||
ev_reasoning_item("reason-1", &["thinking"], &[])
|
||||
},
|
||||
interrupted,
|
||||
],
|
||||
}),
|
||||
Some(interrupted),
|
||||
]
|
||||
.into_iter()
|
||||
.flatten()
|
||||
.collect(),
|
||||
vec![
|
||||
ev_response_created("resp-follow-up"),
|
||||
ev_completed("resp-follow-up"),
|
||||
|
||||
Reference in New Issue
Block a user