Compare commits

...

1 Commits

Author SHA1 Message Date
Sayan Sisodiya
95b16816f5 wip; make compact request async submit + poll to avoid timeouts 2026-02-09 17:52:03 -08:00
6 changed files with 452 additions and 23 deletions

View File

@@ -28,6 +28,11 @@ The public interface of this crate is intentionally small and uniform:
- `instructions: &str` fully-resolved compaction instructions.
- Output: `Vec<ResponseItem>`.
- `CompactClient::compact_input(&CompactionInput, extra_headers)` wraps the JSON encoding and retry/telemetry wiring.
- Async operation mode:
- `CompactClient::submit_compact_operation(body, extra_headers)` returns `operation_id` from `responses/compact/operations`.
- `CompactClient::poll_compact_operation(operation_id, extra_headers)` returns either:
- `CompactOperationPollResult::Pending { poll_after_ms }`, or
- `CompactOperationPollResult::Completed(Vec<ResponseItem>)`.
- **Memory trace summarize endpoint**
- Input: `MemoryTraceSummarizeInput` (re-exported as `codex_api::MemoryTraceSummarizeInput`):

View File

@@ -5,9 +5,11 @@ use crate::error::ApiError;
use crate::provider::Provider;
use codex_client::HttpTransport;
use codex_client::RequestTelemetry;
use codex_client::Response as HttpResponse;
use codex_protocol::models::ResponseItem;
use http::HeaderMap;
use http::Method;
use http::StatusCode;
use serde::Deserialize;
use serde_json::to_value;
use std::sync::Arc;
@@ -16,6 +18,12 @@ pub struct CompactClient<T: HttpTransport, A: AuthProvider> {
session: EndpointSession<T, A>,
}
#[derive(Debug)]
pub enum CompactOperationPollResult {
Pending { poll_after_ms: Option<u64> },
Completed(Vec<ResponseItem>),
}
impl<T: HttpTransport, A: AuthProvider> CompactClient<T, A> {
pub fn new(transport: T, provider: Provider, auth: A) -> Self {
Self {
@@ -33,6 +41,14 @@ impl<T: HttpTransport, A: AuthProvider> CompactClient<T, A> {
"responses/compact"
}
fn operations_path() -> &'static str {
"responses/compact/operations"
}
fn operation_path(operation_id: &str) -> String {
format!("responses/compact/operations/{operation_id}")
}
pub async fn compact(
&self,
body: serde_json::Value,
@@ -56,6 +72,98 @@ impl<T: HttpTransport, A: AuthProvider> CompactClient<T, A> {
.map_err(|e| ApiError::Stream(format!("failed to encode compaction input: {e}")))?;
self.compact(body, extra_headers).await
}
pub async fn submit_compact_operation(
&self,
body: serde_json::Value,
extra_headers: HeaderMap,
) -> Result<String, ApiError> {
let resp = self
.session
.execute(
Method::POST,
Self::operations_path(),
extra_headers,
Some(body),
)
.await?;
if resp.status != StatusCode::ACCEPTED {
return Err(response_to_api_error(
&resp,
"compact operation submit failed",
));
}
let parsed: CompactOperationPendingResponse =
serde_json::from_slice(&resp.body).map_err(|e| ApiError::Stream(e.to_string()))?;
if parsed.operation_id.is_empty() {
return Err(ApiError::Stream(
"compact operation submit returned empty operation_id".to_string(),
));
}
if parsed.status != "pending" {
let status = parsed.status;
return Err(ApiError::Stream(format!(
"unexpected compact operation status: {status}"
)));
}
Ok(parsed.operation_id)
}
pub async fn poll_compact_operation(
&self,
operation_id: &str,
extra_headers: HeaderMap,
) -> Result<CompactOperationPollResult, ApiError> {
let resp = self
.session
.execute(
Method::GET,
&Self::operation_path(operation_id),
extra_headers,
None,
)
.await?;
if resp.status == StatusCode::ACCEPTED {
let parsed: CompactOperationPendingResponse =
serde_json::from_slice(&resp.body).map_err(|e| ApiError::Stream(e.to_string()))?;
if parsed.status == "pending" {
return Ok(CompactOperationPollResult::Pending {
poll_after_ms: parsed.poll_after_ms,
});
}
let status = parsed.status;
return Err(ApiError::Stream(format!(
"unexpected compact operation status: {status}"
)));
}
if resp.status != StatusCode::OK {
return Err(response_to_api_error(
&resp,
"compact operation poll failed",
));
}
let parsed: CompactHistoryResponse =
serde_json::from_slice(&resp.body).map_err(|e| ApiError::Stream(e.to_string()))?;
Ok(CompactOperationPollResult::Completed(parsed.output))
}
}
fn response_to_api_error(resp: &HttpResponse, fallback_message: &str) -> ApiError {
let message = String::from_utf8_lossy(&resp.body).trim().to_string();
ApiError::Api {
status: resp.status,
message: if message.is_empty() {
fallback_message.to_string()
} else {
message
},
}
}
#[derive(Debug, Deserialize)]
@@ -63,28 +171,25 @@ struct CompactHistoryResponse {
output: Vec<ResponseItem>,
}
#[derive(Debug, Deserialize)]
struct CompactOperationPendingResponse {
operation_id: String,
status: String,
poll_after_ms: Option<u64>,
}
#[cfg(test)]
mod tests {
use super::*;
use crate::provider::RetryConfig;
use async_trait::async_trait;
use codex_client::Request;
use codex_client::Response;
use codex_client::StreamResponse;
use codex_client::TransportError;
#[derive(Clone, Default)]
struct DummyTransport;
#[async_trait]
impl HttpTransport for DummyTransport {
async fn execute(&self, _req: Request) -> Result<Response, TransportError> {
Err(TransportError::Build("execute should not run".to_string()))
}
async fn stream(&self, _req: Request) -> Result<StreamResponse, TransportError> {
Err(TransportError::Build("stream should not run".to_string()))
}
}
use http::HeaderMap;
use pretty_assertions::assert_eq;
use std::time::Duration;
#[derive(Clone, Default)]
struct DummyAuth;
@@ -95,11 +200,67 @@ mod tests {
}
}
#[test]
fn path_is_responses_compact() {
assert_eq!(
CompactClient::<DummyTransport, DummyAuth>::path(),
"responses/compact"
fn provider(base_url: &str) -> Provider {
Provider {
name: "test".to_string(),
base_url: base_url.to_string(),
query_params: None,
headers: HeaderMap::new(),
retry: RetryConfig {
max_attempts: 1,
base_delay: Duration::from_millis(1),
retry_429: false,
retry_5xx: true,
retry_transport: true,
},
stream_idle_timeout: Duration::from_secs(1),
}
}
#[derive(Clone)]
struct StaticResponseTransport {
response: Response,
}
#[async_trait]
impl HttpTransport for StaticResponseTransport {
async fn execute(&self, _req: Request) -> Result<Response, TransportError> {
Ok(self.response.clone())
}
async fn stream(&self, _req: Request) -> Result<StreamResponse, TransportError> {
Err(TransportError::Build("stream should not run".to_string()))
}
}
#[tokio::test]
async fn submit_compact_operation_returns_pending_status() {
let response_body = compact_operation_pending_payload("op_123");
let client = CompactClient::new(
StaticResponseTransport {
response: Response {
status: StatusCode::ACCEPTED,
headers: HeaderMap::new(),
body: bytes::Bytes::from(response_body.to_string()),
},
},
provider("https://example.com/api/codex"),
DummyAuth,
);
let submit_result = client
.submit_compact_operation(serde_json::json!({}), HeaderMap::new())
.await
.expect("submit should succeed");
assert_eq!(submit_result, "op_123");
}
fn compact_operation_pending_payload(operation_id: &str) -> serde_json::Value {
serde_json::json!({
"operation_id": operation_id,
"status": "pending",
"poll_after_ms": 1000,
})
}
}

View File

@@ -28,6 +28,7 @@ pub use crate::common::ResponsesApiRequest;
pub use crate::common::create_text_param_for_request;
pub use crate::endpoint::aggregate::AggregateStreamExt;
pub use crate::endpoint::compact::CompactClient;
pub use crate::endpoint::compact::CompactOperationPollResult;
pub use crate::endpoint::memories::MemoriesClient;
pub use crate::endpoint::models::ModelsClient;
pub use crate::endpoint::responses::ResponsesClient;

View File

@@ -36,6 +36,7 @@ use crate::api_bridge::auth_provider_from_auth;
use crate::api_bridge::map_api_error;
use crate::auth::UnauthorizedRecovery;
use codex_api::CompactClient as ApiCompactClient;
use codex_api::CompactOperationPollResult;
use codex_api::CompactionInput as ApiCompactionInput;
use codex_api::MemoriesClient as ApiMemoriesClient;
use codex_api::MemoryTrace as ApiMemoryTrace;
@@ -83,6 +84,7 @@ use tokio::sync::oneshot::error::TryRecvError;
use tokio_tungstenite::tungstenite::Error;
use tokio_tungstenite::tungstenite::Message;
use tracing::warn;
use uuid::Uuid;
use crate::AuthManager;
use crate::auth::CodexAuth;
@@ -105,6 +107,11 @@ pub const X_CODEX_TURN_METADATA_HEADER: &str = "x-codex-turn-metadata";
pub const X_RESPONSESAPI_INCLUDE_TIMING_METRICS_HEADER: &str =
"x-responsesapi-include-timing-metrics";
const RESPONSES_WEBSOCKETS_V2_BETA_HEADER_VALUE: &str = "responses_websockets=2026-02-06";
const X_CODEX_IDEMPOTENCY_KEY_HEADER: &str = "x-codex-idempotency-key";
const COMPACT_OPERATION_INITIAL_POLL_INTERVAL: Duration = Duration::from_millis(250);
const COMPACT_OPERATION_MIN_POLL_INTERVAL: Duration = Duration::from_millis(100);
const COMPACT_OPERATION_MAX_POLL_INTERVAL: Duration = Duration::from_secs(2);
const COMPACT_OPERATION_POLL_DEADLINE: Duration = Duration::from_secs(15 * 60);
/// Session-scoped state shared by all [`ModelClient`] clones.
///
/// This is intentionally kept minimal so `ModelClient` does not need to hold a full `Config`. Most
@@ -268,11 +275,83 @@ impl ModelClient {
instructions: &instructions,
};
let body = serde_json::to_value(&payload).map_err(|err| {
map_api_error(ApiError::Stream(format!(
"failed to encode compaction input: {err}"
)))
})?;
let extra_headers = self.build_subagent_headers();
client
.compact_input(&payload, extra_headers)
.await
.map_err(map_api_error)
let mut submit_headers = extra_headers.clone();
let compact_id = Uuid::new_v4();
let idempotency_key = format!("compact-{compact_id}");
let idempotency_header_value = HeaderValue::from_str(&idempotency_key).map_err(|err| {
map_api_error(ApiError::Stream(format!(
"failed to encode compact idempotency key header: {err}"
)))
})?;
submit_headers.insert(X_CODEX_IDEMPOTENCY_KEY_HEADER, idempotency_header_value);
let operation_id = match client.submit_compact_operation(body, submit_headers).await {
Ok(operation_id) => operation_id,
Err(err) if Self::is_compact_operations_endpoint_unsupported(&err) => {
return client
.compact_input(&payload, extra_headers)
.await
.map_err(map_api_error);
}
Err(err) => return Err(map_api_error(err)),
};
let mut poll_interval = COMPACT_OPERATION_INITIAL_POLL_INTERVAL;
let deadline = tokio::time::Instant::now() + COMPACT_OPERATION_POLL_DEADLINE;
loop {
if tokio::time::Instant::now() >= deadline {
return Err(map_api_error(ApiError::Stream(
"timeout waiting for compact operation".to_string(),
)));
}
let poll_result = client
.poll_compact_operation(&operation_id, extra_headers.clone())
.await
.map_err(map_api_error)?;
match poll_result {
CompactOperationPollResult::Completed(output) => return Ok(output),
CompactOperationPollResult::Pending { poll_after_ms } => {
let poll_after_duration = poll_after_ms
.map(Duration::from_millis)
.map(Self::clamp_compact_operation_poll_interval);
let sleep_duration = poll_after_duration.unwrap_or(poll_interval);
tokio::time::sleep(sleep_duration).await;
poll_interval = poll_after_duration.unwrap_or_else(|| {
Self::clamp_compact_operation_poll_interval(poll_interval.saturating_mul(2))
});
}
}
}
}
fn is_compact_operations_endpoint_unsupported(err: &ApiError) -> bool {
match err {
ApiError::Api { status, .. } => {
*status == HttpStatusCode::NOT_FOUND
|| *status == HttpStatusCode::METHOD_NOT_ALLOWED
}
ApiError::Transport(TransportError::Http { status, .. }) => {
*status == HttpStatusCode::NOT_FOUND
|| *status == HttpStatusCode::METHOD_NOT_ALLOWED
}
_ => false,
}
}
fn clamp_compact_operation_poll_interval(interval: Duration) -> Duration {
interval.clamp(
COMPACT_OPERATION_MIN_POLL_INTERVAL,
COMPACT_OPERATION_MAX_POLL_INTERVAL,
)
}
/// Builds memory summaries for each provided normalized trace.

View File

@@ -756,6 +756,31 @@ fn compact_mock() -> (MockBuilder, ResponseMock) {
(mock, response_mock)
}
fn compact_operation_submit_mock() -> (MockBuilder, ResponseMock) {
let response_mock = ResponseMock::new();
let mock = Mock::given(method("POST"))
.and(path_regex(".*/responses/compact/operations$"))
.and(response_mock.clone());
(mock, response_mock)
}
fn compact_operation_poll_mock(operation_id: &str) -> (MockBuilder, ResponseMock) {
let response_mock = ResponseMock::new();
let path = format!(".*/responses/compact/operations/{operation_id}$");
let mock = Mock::given(method("GET"))
.and(path_regex(&path))
.and(response_mock.clone());
(mock, response_mock)
}
fn compact_operation_pending_payload(operation_id: &str) -> serde_json::Value {
serde_json::json!({
"operation_id": operation_id,
"status": "pending",
"poll_after_ms": 1000,
})
}
fn models_mock() -> (MockBuilder, ModelsMock) {
let models_mock = ModelsMock::new();
let mock = Mock::given(method("GET"))
@@ -820,6 +845,55 @@ pub async fn mount_compact_json_once(server: &MockServer, body: serde_json::Valu
response_mock
}
pub async fn mount_compact_operation_submit_once(
server: &MockServer,
operation_id: &str,
) -> ResponseMock {
let (mock, response_mock) = compact_operation_submit_mock();
mock.respond_with(
ResponseTemplate::new(202)
.insert_header("content-type", "application/json")
.set_body_json(compact_operation_pending_payload(operation_id)),
)
.up_to_n_times(1)
.mount(server)
.await;
response_mock
}
pub async fn mount_compact_operation_poll_pending_once(
server: &MockServer,
operation_id: &str,
) -> ResponseMock {
let (mock, response_mock) = compact_operation_poll_mock(operation_id);
mock.respond_with(
ResponseTemplate::new(202)
.insert_header("content-type", "application/json")
.set_body_json(compact_operation_pending_payload(operation_id)),
)
.up_to_n_times(1)
.mount(server)
.await;
response_mock
}
pub async fn mount_compact_operation_poll_done_once(
server: &MockServer,
operation_id: &str,
body: serde_json::Value,
) -> ResponseMock {
let (mock, response_mock) = compact_operation_poll_mock(operation_id);
mock.respond_with(
ResponseTemplate::new(200)
.insert_header("content-type", "application/json")
.set_body_json(body.clone()),
)
.up_to_n_times(1)
.mount(server)
.await;
response_mock
}
pub async fn mount_models_once(server: &MockServer, body: ModelsResponse) -> ModelsMock {
let (mock, models_mock) = models_mock();
mock.respond_with(

View File

@@ -156,6 +156,115 @@ async fn remote_compact_replaces_history_for_followups() -> Result<()> {
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn remote_compact_operations_happy_path_replaces_history_for_followups() -> Result<()> {
skip_if_no_network!(Ok(()));
let harness = TestCodexHarness::with_builder(
test_codex().with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing()),
)
.await?;
let codex = harness.test().codex.clone();
let responses_mock = responses::mount_sse_sequence(
harness.server(),
vec![
responses::sse(vec![
responses::ev_assistant_message("m1", "FIRST_REMOTE_REPLY"),
responses::ev_completed("resp-1"),
]),
responses::sse(vec![
responses::ev_assistant_message("m2", "AFTER_COMPACT_REPLY"),
responses::ev_completed("resp-2"),
]),
],
)
.await;
let compacted_history = vec![
ResponseItem::Message {
id: None,
role: "user".to_string(),
content: vec![ContentItem::InputText {
text: "REMOTE_COMPACTED_SUMMARY".to_string(),
}],
end_turn: None,
phase: None,
},
ResponseItem::Compaction {
encrypted_content: "ENCRYPTED_COMPACTION_SUMMARY".to_string(),
},
];
let operation_id = "op_compact_happy_1";
let submit_mock =
responses::mount_compact_operation_submit_once(harness.server(), operation_id).await;
let poll_pending_mock =
responses::mount_compact_operation_poll_pending_once(harness.server(), operation_id).await;
let poll_done_mock = responses::mount_compact_operation_poll_done_once(
harness.server(),
operation_id,
serde_json::json!({ "output": compacted_history.clone() }),
)
.await;
let legacy_compact_mock = responses::mount_compact_json_once(
harness.server(),
serde_json::json!({ "output": compacted_history.clone() }),
)
.await;
codex
.submit(Op::UserInput {
items: vec![UserInput::Text {
text: "hello remote compact".into(),
text_elements: Vec::new(),
}],
final_output_json_schema: None,
})
.await?;
wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnComplete(_))).await;
codex.submit(Op::Compact).await?;
wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnComplete(_))).await;
codex
.submit(Op::UserInput {
items: vec![UserInput::Text {
text: "after compact".into(),
text_elements: Vec::new(),
}],
final_output_json_schema: None,
})
.await?;
wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnComplete(_))).await;
let submit_request = submit_mock.single_request();
assert_eq!(submit_request.path(), "/v1/responses/compact/operations");
assert_eq!(poll_pending_mock.requests().len(), 1);
assert_eq!(poll_done_mock.requests().len(), 1);
assert_eq!(legacy_compact_mock.requests().len(), 0);
let follow_up_body = responses_mock
.requests()
.last()
.expect("follow-up request missing")
.body_json()
.to_string();
assert!(
follow_up_body.contains("REMOTE_COMPACTED_SUMMARY"),
"expected follow-up request to use compacted history"
);
assert!(
follow_up_body.contains("ENCRYPTED_COMPACTION_SUMMARY"),
"expected follow-up request to include compaction summary item"
);
assert!(
!follow_up_body.contains("FIRST_REMOTE_REPLY"),
"expected follow-up request to drop pre-compaction assistant messages"
);
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn remote_compact_runs_automatically() -> Result<()> {
skip_if_no_network!(Ok(()));