Repository navigation
fix(storage): separate admission proof build and execution budgets - #5799
Conversation
Part of #5665. Hosted run 41551806702 exhausted the old four-minute wrapper during cold Kubernetes dependency compilation. Build the unchanged native proof separately within a bounded setup phase, retain its two-minute test timeout, and cap both phases below the existing twenty-minute package deadline.
Hosted CI run 41551806702, job 125645226990, killed the wrapper after 240.14 seconds while the child Go command was still downloading and compiling the Kubernetes admission dependencies. The native proof had not produced a verdict. This is a source fix rather than a rerun of that failed attempt. At 9d427c3, rendering and compilation share a fifteen-minute setup bound. The compiled test binary retains the two-minute native test timeout and four-minute outer execution bound; pipe cleanup is also bounded. The phases remain below the existing twenty-minute package limit. Every native admission case still executes, with the original working directory and rendered production profile. The actual complete wrapper passed locally after the final change in 9.16 seconds. The initial revised wrapper also passed in 16.31 seconds. Lint, formatting and whitespace checks pass. Correctness and simplification self-review found no remaining issue. No production or deletion-policy behavior changes. Hosted cold-run verification remains pending in this PR's CI. |
At current head 9d427c3, CodeRabbit's repository-scoped quota refusal was re-read directly from #5792 (comment). The authenticated provider event at 01:10:12 UTC says the next included review is available in 54 minutes, until 02:04:12 UTC. Its explicit scope is this public repository. No redundant request is being sent during that window; continuing to Codex. A quota refusal is not a review pass. |
@codex review |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
|
Codex Review: Didn't find any major issues. Bravo. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Evaluated current head 9d427c3: all 49 checks are settled without failures. Hosted CI41557747453 passed the full Go regression suite; the actual native admission proof completed in 183.20 seconds. The proof still runs every case with its two-minute execution timeout. The cold setup phase and outer execution are separately bounded below the unchanged twenty-minute package ceiling. Local final proof, lint, formatting and whitespace checks pass. This changes only the test wrapper; there is no production runtime or policy mutation. Codex completed a clean exact-head review: summary #issuecomment-6592793394 and receipt #issuecomment-6592804097. CodeRabbit was skipped under freshly verified repository-scoped quota evidence, not treated as green. Complete conversation, review, inline and thread inventories contain no unresolved findings or maintainer instructions. The published body passes its readback gate. Fresh platform health passes and GitHub reports CLEAN. All readable required checks/workflows pass; the configured code-quality rule remains unreadable and is left to the protected merge API. The failure at the earlier head remains recorded and was not rerun. Hosted execution above is the behavior-level acceptance for this code-only change. |
Why
The benchmark deletion safeguard's native test can exhaust its setup budget while compiling dependencies on a cold runner, blocking reviewed platform changes before the safeguard is exercised.
What
Give compilation and execution separate bounded budgets while preserving the full safeguard proof and its existing execution timeout.
Part of #5665