mirror of
https://github.com/openai/codex.git
synced 2026-09-20 12:47:38 +00:00
docs: explain Guardian retry invariants
Co-authored-by: Codex noreply@openai.com
This commit is contained in:
@@ -668,6 +668,11 @@ async fn run_guardian_review(
|
||||
}
|
||||
}
|
||||
|
||||
/// Runs one Guardian review attempt and, for transient failures, one bounded retry.
|
||||
///
|
||||
/// Availability failures wait with full jitter before retrying with the normal
|
||||
/// deadline. Model timeouts retry immediately with a shorter deadline.
|
||||
/// Cancellation stops the retry without converting it into a denial.
|
||||
async fn run_guardian_review_session_with_retry(
|
||||
session: Arc<Session>,
|
||||
turn: Arc<TurnContext>,
|
||||
|
||||
@@ -911,6 +911,10 @@ fn event_matches_turn(event: &Event, expected_turn_id: &str) -> bool {
|
||||
}
|
||||
}
|
||||
|
||||
/// Builds the locked-down config used by Guardian review sessions.
|
||||
///
|
||||
/// Provider-level retries are disabled because the Guardian retry wrapper owns
|
||||
/// the complete retry budget and must issue at most one additional review.
|
||||
pub(crate) fn build_guardian_review_session_config(
|
||||
parent_config: &Config,
|
||||
live_network_config: Option<codex_network_proxy::NetworkProxyConfig>,
|
||||
|
||||
Reference in New Issue
Block a user