Stop guardian reviews when parent turns are interrupted

This commit is contained in:
Ray Morgan
2026-06-08 13:52:35 -07:00
parent 52b359b249
commit 3d437e3f57
3 changed files with 143 additions and 0 deletions

View File

@@ -303,6 +303,32 @@ impl GuardianReviewSessionManager {
}
}
pub(crate) async fn interrupt_active_reviews(&self) {
let review_sessions = {
let state = self.state.lock().await;
state
.trunk
.iter()
.chain(&state.ephemeral_reviews)
.cloned()
.collect::<Vec<_>>()
};
let active_review_sessions = review_sessions
.into_iter()
.filter(|review_session| review_session.review_lock.try_acquire().is_err())
.collect::<Vec<_>>();
for review_session in &active_review_sessions {
let _ = review_session.codex.submit(Op::Interrupt).await;
}
for review_session in active_review_sessions {
let _ = tokio::time::timeout(
GUARDIAN_INTERRUPT_DRAIN_TIMEOUT,
review_session.review_lock.acquire(),
)
.await;
}
}
#[expect(
clippy::await_holding_invalid_type,
reason = "review session selection and trunk spawning must stay serialized"
@@ -561,6 +587,17 @@ impl GuardianReviewSessionManager {
}
Err(outcome) => return (outcome, GuardianReviewAnalyticsResult::without_session()),
};
let review_guard = match review_session.review_lock.acquire().await {
Ok(review_guard) => review_guard,
Err(err) => {
return (
GuardianReviewSessionOutcome::SessionFailed(anyhow!(
"guardian review lock closed: {err}"
)),
GuardianReviewAnalyticsResult::without_session(),
);
}
};
self.register_active_ephemeral(Arc::clone(&review_session))
.await;
let mut cleanup =
@@ -573,6 +610,7 @@ impl GuardianReviewSessionManager {
deadline,
))
.await;
drop(review_guard);
if let Some(review_session) = self.take_active_ephemeral(&review_session).await {
cleanup.disarm();
review_session.shutdown_in_background();

View File

@@ -3298,6 +3298,9 @@ impl Session {
pub async fn interrupt_task(self: &Arc<Self>) {
info!("interrupt received: abort current task, if any");
let had_active_turn = self.active_turn.lock().await.is_some();
self.guardian_review_session
.interrupt_active_reviews()
.await;
self.abort_all_tasks(TurnAbortReason::Interrupted).await;
if !had_active_turn {
self.cancel_mcp_startup().await;

View File

@@ -6,6 +6,7 @@ use codex_core::sandboxing::SandboxPermissions;
use codex_protocol::config_types::ApprovalsReviewer;
use codex_protocol::protocol::AskForApproval;
use codex_protocol::protocol::EventMsg;
use codex_protocol::protocol::GuardianAssessmentStatus;
use codex_protocol::protocol::Op;
use codex_protocol::protocol::SandboxPolicy;
use codex_protocol::user_input::UserInput;
@@ -14,8 +15,10 @@ use core_test_support::responses::ev_assistant_message;
use core_test_support::responses::ev_completed;
use core_test_support::responses::ev_function_call;
use core_test_support::responses::ev_response_created;
use core_test_support::responses::mount_response_sequence;
use core_test_support::responses::mount_sse_sequence;
use core_test_support::responses::sse;
use core_test_support::responses::sse_response;
use core_test_support::responses::start_mock_server;
use core_test_support::skip_if_no_network;
use core_test_support::skip_if_sandbox;
@@ -29,6 +32,105 @@ use std::os::unix::fs::PermissionsExt;
use std::time::Duration;
use tempfile::TempDir;
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn interrupting_parent_turn_shuts_down_active_guardian_review() -> Result<()> {
skip_if_no_network!(Ok(()));
skip_if_sandbox!(Ok(()));
let server = start_mock_server().await;
let approval_policy = AskForApproval::OnRequest;
let sandbox_policy = SandboxPolicy::WorkspaceWrite {
writable_roots: vec![],
network_access: false,
exclude_tmpdir_env_var: true,
exclude_slash_tmp: true,
};
let sandbox_policy_for_config = sandbox_policy.clone();
let mut builder = test_codex().with_config(move |config| {
config.permissions.approval_policy = Constrained::allow_any(approval_policy);
config
.set_legacy_sandbox_policy(sandbox_policy_for_config)
.expect("set sandbox policy");
});
let test = builder.build(&server).await?;
let tool_args = json!({
"cmd": "echo guardian-review",
"yield_time_ms": 1_000_u64,
"sandbox_permissions": SandboxPermissions::RequireEscalated,
"justification": "Exercise Guardian cancellation.",
});
let parent_response = sse_response(sse(vec![
ev_response_created("resp-parent-tool"),
ev_function_call(
"exec-call",
"exec_command",
&serde_json::to_string(&tool_args)?,
),
ev_completed("resp-parent-tool"),
]));
let guardian_response = sse_response(sse(vec![ev_response_created("resp-guardian-delayed")]))
.set_delay(Duration::from_secs(60));
let responses =
mount_response_sequence(&server, vec![parent_response, guardian_response]).await;
test.codex
.submit(Op::UserInput {
environments: None,
items: vec![UserInput::Text {
text: "run a command that requires Guardian review".into(),
text_elements: Vec::new(),
}],
final_output_json_schema: None,
responsesapi_client_metadata: None,
additional_context: Default::default(),
thread_settings: codex_protocol::protocol::ThreadSettingsOverrides {
cwd: Some(test.cwd.path().to_path_buf()),
approval_policy: Some(approval_policy),
approvals_reviewer: Some(ApprovalsReviewer::AutoReview),
sandbox_policy: Some(sandbox_policy),
..Default::default()
},
})
.await?;
let guardian_started = wait_for_event(&test.codex, |event| {
matches!(event, EventMsg::GuardianAssessment(_))
})
.await;
let EventMsg::GuardianAssessment(guardian_started) = guardian_started else {
unreachable!("wait predicate only accepts guardian assessments");
};
tokio::time::timeout(Duration::from_secs(5), async {
while responses.requests().len() < 2 {
tokio::time::sleep(Duration::from_millis(10)).await;
}
})
.await
.expect("guardian review request should be in flight before interrupt");
test.codex.submit(Op::Interrupt).await?;
let guardian_statuses = tokio::time::timeout(Duration::from_secs(5), async {
let mut statuses = vec![guardian_started.status];
loop {
match test.codex.next_event().await?.msg {
EventMsg::GuardianAssessment(event) => statuses.push(event.status),
EventMsg::TurnAborted(_) => return Ok::<_, anyhow::Error>(statuses),
_ => {}
}
}
})
.await??;
assert_eq!(
guardian_statuses,
vec![
GuardianAssessmentStatus::InProgress,
GuardianAssessmentStatus::Aborted,
]
);
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn guardian_review_session_does_not_inherit_legacy_notify() -> Result<()> {
skip_if_no_network!(Ok(()));