Skip to content

fix(codex): retry terminal-only live completions #70

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
raine merged 4 commits into from
Jul 28, 2026
+475 −34
Merged

fix(codex): retry terminal-only live completions #70

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
fa8197b
fix(codex): retry empty live completions
HM-Silvey Jul 20, 2026
3f9bff9
test(codex): preserve signature-only live output
HM-Silvey Jul 20, 2026
bb42ead
test(codex): prove bounded 503 on repeated empty completions
HM-Silvey Jul 20, 2026
23d22ea
codex: capture only committed live events
raine Jul 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
127 changes: 98 additions & 29 deletions src/providers/codex/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@ use std::sync::Arc;

use crate::anthropic::error::json_error;
use crate::anthropic::schema::{CountTokensResponse, MessagesRequest};
use crate::anthropic::sse::parse_sse_events;
use crate::config;
use crate::logging::create_logger;
use crate::monitor::usage_from_anthropic_sse;
Expand Down Expand Up @@ -567,6 +568,9 @@ async fn live_stream_response_once(
) -> LiveStreamStart {
let mut translator = LiveStreamTranslator::new(message_id, model.to_string());
let mut upstream_sse_body = Vec::new();
// Keep protocol framing private until real output makes a transparent retry unsafe.
// Every branch that consumes pending_chunk returns, so it is never flushed twice.
let mut pending_chunk = Vec::new();
let mut generation_started = false;

while let Some(item) = upstream_events.recv().await {
Expand All @@ -592,7 +596,7 @@ async fn live_stream_response_once(
generation_started = true;
}
append_upstream_sse_payload(&mut upstream_sse_body, &payload);
let (chunk, terminal) = match translate_live_stream_payload(&mut translator, &payload, &ctx)
let (chunk, terminal) = match translate_live_stream_payload(&mut translator, &payload, None)
{
Ok(result) => result,
Err(message) => {
Expand Down Expand Up @@ -638,8 +642,18 @@ async fn live_stream_response_once(
return LiveStreamStart::Response(map_codex_failure_to_response(&message));
}
};
if !chunk.is_empty() {
record_live_stream_progress(&ctx, &chunk);
pending_chunk.extend_from_slice(&chunk);
if terminal
&& is_codex_success_terminal_event(&payload)
&& !translator.has_semantic_output()
{
return LiveStreamStart::Retry {
error: empty_live_completion_error(),
};
}
if translator.has_semantic_output() && !pending_chunk.is_empty() {
record_live_stream_downstream_capture(&ctx, &pending_chunk);
record_live_stream_progress(&ctx, &pending_chunk);
if terminal {
update_continuation_from_upstream(
ctx.session_id.as_deref(),
Expand All @@ -648,12 +662,12 @@ async fn live_stream_response_once(
&upstream_sse_body,
compact_boundary,
);
return LiveStreamStart::Response(single_live_stream_response(chunk));
return LiveStreamStart::Response(single_live_stream_response(pending_chunk));
}
return LiveStreamStart::Response(remaining_live_stream_response(
upstream_events,
translator,
chunk,
pending_chunk,
ctx,
turn_id,
request_body,
Expand All @@ -669,7 +683,12 @@ async fn live_stream_response_once(
&upstream_sse_body,
compact_boundary,
);
return LiveStreamStart::Response(empty_live_stream_response());
if pending_chunk.is_empty() {
return LiveStreamStart::Response(empty_live_stream_response());
}
record_live_stream_downstream_capture(&ctx, &pending_chunk);
record_live_stream_progress(&ctx, &pending_chunk);
return LiveStreamStart::Response(single_live_stream_response(pending_chunk));
}
}

Expand All @@ -684,6 +703,16 @@ async fn live_stream_response_once(
}
}

fn empty_live_completion_error() -> client::CodexError {
client::CodexError {
status: 503,
message: "Codex completed without producing output".to_string(),
detail: Some(EMPTY_CODEX_COMPLETION_DETAIL.to_string()),
retry_after: None,
origin: client::CodexErrorOrigin::WebSocket,
}
}

fn codex_generation_event(payload: &serde_json::Value) -> bool {
!matches!(
payload.get("type").and_then(|value| value.as_str()),
Expand All @@ -694,13 +723,31 @@ fn codex_generation_event(payload: &serde_json::Value) -> bool {
fn translate_live_stream_payload(
translator: &mut LiveStreamTranslator,
payload: &serde_json::Value,
ctx: &RequestContext,
traffic: Option<&crate::traffic::TrafficCapture>,
) -> Result<(Vec<u8>, bool), String> {
let chunk = translator.accept(payload, ctx.traffic.as_deref())?;
let chunk = translator.accept(payload, traffic)?;
let terminal = is_codex_terminal_event(payload) || translator.is_finished();
Ok((chunk, terminal))
}

fn record_live_stream_downstream_capture(ctx: &RequestContext, chunk: &[u8]) {
let Some(traffic) = ctx.traffic.as_ref() else {
return;
};
for event in parse_sse_events(chunk) {
let Ok(data) = serde_json::from_str::<serde_json::Value>(&event.data) else {
continue;
};
traffic.write_json_event(
"050-downstream-event",
&serde_json::json!({
"event": event.event.as_deref().unwrap_or("message"),
"data": data,
}),
);
}
}

fn record_live_stream_progress(ctx: &RequestContext, chunk: &[u8]) {
if let Some(monitor) = ctx.monitor.as_ref() {
let (input_tokens, output_tokens) = usage_from_anthropic_sse(chunk);
Expand Down Expand Up @@ -750,28 +797,31 @@ fn remaining_live_stream_response(
match item {
Ok(payload) => {
append_upstream_sse_payload(&mut upstream_sse_body, &payload);
let (chunk, terminal) =
match translate_live_stream_payload(&mut translator, &payload, &ctx) {
Ok(result) => result,
Err(message) => {
abort_request_state(
ctx.session_id.as_deref(),
turn_id,
compact_boundary,
&request_body,
);
let chunk = translator.error_chunk(
&message,
"api_error",
ctx.traffic.as_deref(),
);
if !chunk.is_empty() {
record_live_stream_progress(&ctx, &chunk);
let _ = tx.send(Ok(Bytes::from(chunk))).await;
}
return;
let (chunk, terminal) = match translate_live_stream_payload(
&mut translator,
&payload,
ctx.traffic.as_deref(),
) {
Ok(result) => result,
Err(message) => {
abort_request_state(
ctx.session_id.as_deref(),
turn_id,
compact_boundary,
&request_body,
);
let chunk = translator.error_chunk(
&message,
"api_error",
ctx.traffic.as_deref(),
);
if !chunk.is_empty() {
record_live_stream_progress(&ctx, &chunk);
let _ = tx.send(Ok(Bytes::from(chunk))).await;
}
};
return;
}
};
if !chunk.is_empty() {
record_live_stream_progress(&ctx, &chunk);
if tx.send(Ok(Bytes::from(chunk))).await.is_err() {
Expand Down Expand Up @@ -926,6 +976,13 @@ fn is_codex_terminal_event(payload: &serde_json::Value) -> bool {
)
}

fn is_codex_success_terminal_event(payload: &serde_json::Value) -> bool {
matches!(
payload.get("type").and_then(|v| v.as_str()),
Some("response.completed") | Some("response.done")
)
}

fn retryable_live_start_codex_error(err: &client::CodexError) -> bool {
if err.origin == client::CodexErrorOrigin::WebSocketHandshake {
return err.status == 0 || matches!(err.status, 429 | 500 | 502 | 503 | 504 | 529);
Expand Down Expand Up @@ -1481,6 +1538,18 @@ mod tests {
);
}

#[tokio::test]
async fn empty_live_completion_maps_to_explicit_service_unavailable() {
let err = empty_live_completion_error();

assert_eq!(err.status, 503);
assert_eq!(err.detail.as_deref(), Some(EMPTY_CODEX_COMPLETION_DETAIL));
assert_eq!(
map_codex_error_to_response(&err).status(),
StatusCode::SERVICE_UNAVAILABLE
);
}

#[test]
fn live_start_statusless_websocket_handshake_error_is_retryable() {
let err = client::CodexError {
Expand Down
Loading