[codex] Tighten transient retry policy suppression - #5025
Conversation
|
Quick follow-up: the Summary already covers the why / user-visible effect pretty well, but could you add the Surface area checklist and tick the right box (or Also marking this for QA early since a retry-policy regression here would affect a live run path; nothing needed from you yet — we'll use the label as the reminder to route QA once the PR is merge-ready. |
mrcfps
left a comment
There was a problem hiding this comment.
@PerishCode I reviewed the retry-policy changes across apps/daemon/src/run-retry-policy.ts, the focused daemon retry-policy tests, and the analytics suppressed-reason contract. The new allowlist is consistent with the classifier detail values I checked, and the added cases cover the main suppression paths introduced here. Nice work tightening this behavior while keeping the telemetry reason codes explicit.
Note: I attempted to run pnpm --filter @open-design/daemon test -- run-retry-policy.test.ts, but this reviewer worktree does not have dependencies installed (vitest was not found).
|
One PR-body follow-up before the final pass: could you add the Surface area checklist and a short Bug fix verification note describing the repro / regression seam this change covers? The Summary section already does the Why / user-visible-effect part, so no heading rename needed. |
AmyShang-alt
left a comment
There was a problem hiding this comment.
QA validation update: I re-checked the retry/billing safety boundaries locally on head db05ef4.
Validated commands:
- pnpm --dir apps/daemon exec vitest run tests/run-retry-policy.test.ts -> passed
- pnpm --dir apps/daemon typecheck -> passed
I also added temporary local boundary cases to exercise the concern directly:
- rate_limit + hard_quota + retryable=true still suppresses retry with hard_quota
- newly-allowed process_exit/fatal_rpc_error + retryable=true is still suppressed by userVisibleOutputSeen, toolCallSeen, artifactWriteSeen, and liveArtifactSeen
Those expanded local tests passed as 21/21. My previous request-changes review was too strong for this result; the existing policy gates do cover the retry/cost edge I was worried about. Approving based on this validation.
Co-authored-by: Siri-Ray <2667192167@qq.com>
Co-authored-by: Siri-Ray <2667192167@qq.com>
Summary
Validation
mainbefore opening the PR.