openai/codex
1
0
Fork 0
mirror of https://github.com/openai/codex.git synced 2026-10-24 18:18:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

Skip prefix compaction when snapshot is oversized

Browse Source
This commit is contained in:
Rohit Arunachalam
2026-04-16 11:45:22 -07:00
parent 9997539fa1
commit bdd4370c21
2 changed files with 139 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
codex-rs/core/src/compact_remote.rs
+9 -8
View File
@@ -81,17 +81,18 @@ pub(crate) async fn run_remote_prefix_compact_task(
let mut history = ContextManager::new();
history.replace(base_history);
let base_instructions = sess.get_base_instructions().await;
let deleted_items = trim_function_call_history_to_fit_context_window(
&mut history,
turn_context.as_ref(),
&base_instructions,
);
if deleted_items > 0 {
if let Some(context_window) = turn_context.model_context_window()
&& let Some(estimated_tokens) =
history.estimate_token_count_with_base_instructions(&base_instructions)
&& estimated_tokens > context_window
{
info!(
turn_id = %turn_context.sub_id,
deleted_items,
"trimmed history items before prefix compaction"
estimated_tokens,
context_window,
"prefix compaction snapshot exceeds context window"
);
return Err(CodexErr::ContextWindowExceeded);
}
let prompt_input = history.for_prompt(&turn_context.model_info.input_modalities);
codex-rs/core/tests/suite/compact_remote.rs
+130 -4
View File
@@ -156,6 +156,33 @@ impl Respond for PrefixRaceCompactResponder {
}
}
#[derive(Clone, Debug)]
struct ModeAwareCompactResponder {
requests: StdArc<StdMutex<Vec<serde_json::Value>>>,
}
impl Respond for ModeAwareCompactResponder {
fn respond(&self, request: &wiremock::Request) -> ResponseTemplate {
let body = request
.body_json::<serde_json::Value>()
.expect("compact request body should be JSON");
let prefix_mode = body.get("mode").and_then(serde_json::Value::as_str) == Some("prefix");
self.requests
.lock()
.expect("request lock poisoned")
.push(body);
let summary = if prefix_mode {
"PREFIX_SUMMARY"
} else {
"NORMAL_FALLBACK_SUMMARY"
};
ResponseTemplate::new(200)
.insert_header("content-type", "application/json")
.set_body_json(json!({ "output": compacted_summary_only_output(summary) }))
}
}
fn remote_realtime_test_codex_builder(
realtime_server: &responses::WebSocketTestServer,
) -> TestCodexBuilder {
@@ -1189,7 +1216,7 @@ async fn prefix_compact_uses_configured_threshold_percent() -> Result<()> {
test_codex()
.with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing())
.with_config(|config| {
config.model_context_window = Some(1_000);
config.model_context_window = Some(20_000);
config.model_auto_compact_token_limit = Some(800);
config.prefix_compaction.threshold_percent = Some(90);
enable_prefix_compaction(config);
@@ -1264,6 +1291,105 @@ async fn prefix_compact_uses_configured_threshold_percent() -> Result<()> {
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn oversized_prefix_snapshot_skips_prefix_compact_request() -> Result<()> {
skip_if_no_network!(Ok(()));
let large_reply = "LARGE_REMOTE_REPLY ".repeat(1_000);
let harness = TestCodexHarness::with_builder(
test_codex()
.with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing())
.with_config(|config| {
config.model_context_window = Some(1_000);
config.model_auto_compact_token_limit = Some(800);
enable_prefix_compaction(config);
}),
)
.await?;
let codex = harness.test().codex.clone();
let responses_mock = responses::mount_sse_sequence(
harness.server(),
vec![
sse(vec![
responses::ev_assistant_message("m1", &large_reply),
responses::ev_completed_with_tokens("resp-1", /*total_tokens*/ 650),
]),
sse(vec![
responses::ev_assistant_message("m2", "SECOND_REMOTE_REPLY"),
responses::ev_completed_with_tokens("resp-2", /*total_tokens*/ 850),
]),
sse(vec![
responses::ev_assistant_message("m3", "THIRD_REMOTE_REPLY"),
responses::ev_completed_with_tokens("resp-3", /*total_tokens*/ 200),
]),
],
)
.await;
let compact_requests = StdArc::new(StdMutex::new(Vec::new()));
Mock::given(method("POST"))
.and(path_regex(".*/responses/compact$"))
.respond_with(ModeAwareCompactResponder {
requests: StdArc::clone(&compact_requests),
})
.up_to_n_times(1)
.expect(1)
.mount(harness.server())
.await;
codex
.submit(Op::UserInput {
items: vec![UserInput::Text {
text: "FIRST_REMOTE_USER".into(),
text_elements: Vec::new(),
}],
final_output_json_schema: None,
responsesapi_client_metadata: None,
})
.await?;
wait_for_event(&codex, |event| matches!(event, EventMsg::TurnComplete(_))).await;
codex
.submit(Op::UserInput {
items: vec![UserInput::Text {
text: "SECOND_REMOTE_USER".into(),
text_elements: Vec::new(),
}],
final_output_json_schema: None,
responsesapi_client_metadata: None,
})
.await?;
wait_for_event(&codex, |event| matches!(event, EventMsg::TurnComplete(_))).await;
codex
.submit(Op::UserInput {
items: vec![UserInput::Text {
text: "THIRD_REMOTE_USER".into(),
text_elements: Vec::new(),
}],
final_output_json_schema: None,
responsesapi_client_metadata: None,
})
.await?;
wait_for_event(&codex, |event| matches!(event, EventMsg::TurnComplete(_))).await;
let compact_requests = compact_requests
.lock()
.expect("request lock poisoned")
.clone();
assert_eq!(compact_requests.len(), 1);
assert!(compact_requests[0].get("mode").is_none());
let requests = responses_mock.requests();
assert_eq!(requests.len(), 3);
let third_request_body = requests[2].body_json().to_string();
assert!(third_request_body.contains("NORMAL_FALLBACK_SUMMARY"));
assert!(!third_request_body.contains("PREFIX_SUMMARY"));
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn ready_prefix_compact_is_applied_by_pre_turn_auto_compact() -> Result<()> {
skip_if_no_network!(Ok(()));
@@ -1272,7 +1398,7 @@ async fn ready_prefix_compact_is_applied_by_pre_turn_auto_compact() -> Result<()
test_codex()
.with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing())
.with_config(|config| {
config.model_context_window = Some(1_000);
config.model_context_window = Some(20_000);
config.model_auto_compact_token_limit = Some(800);
enable_prefix_compaction(config);
}),
@@ -1393,7 +1519,7 @@ async fn running_prefix_compact_is_abandoned_when_auto_compact_fires() -> Result
test_codex()
.with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing())
.with_config(|config| {
config.model_context_window = Some(1_000);
config.model_context_window = Some(20_000);
config.model_auto_compact_token_limit = Some(800);
enable_prefix_compaction(config);
}),
@@ -1494,7 +1620,7 @@ async fn running_prefix_compact_started_on_follow_up_boundary_is_abandoned_when_
test_codex()
.with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing())
.with_config(|config| {
config.model_context_window = Some(1_000);
config.model_context_window = Some(20_000);
config.model_auto_compact_token_limit = Some(800);
enable_prefix_compaction(config);
}),