Repository navigation
ci: turn on the Triage and Retry System by default (no retry-ci label needed) - #44956
Conversation
|
CLA Signature Action: All authors have signed the CLA. You may need to manually re-run the blocking PR check if it doesn't pass in a few minutes. |
Builds ready [15b146e] [reused from 7b6dde7]
⚡ Performance Benchmarks (Total: 🟢 2 pass · 🟡 2 warn · 🔴 4 fail)
Bundle size diffs
|
Builds ready [edc526b] [reused from a738050]
⚡ Performance Benchmarks (Total: 🟢 0 pass · 🟡 0 warn · 🔴 1 fail)
Bundle size diffs
|
Builds ready [d8ad6db] [reused from a738050]
⚡ Performance Benchmarks (Total: 🟢 12 pass · 🟡 6 warn · 🔴 4 fail)
Bundle size diffs
|
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit d8ad6db. Configure here.
Builds ready [cf48d08] [reused from a738050]
⚡ Performance Benchmarks (Total: 🟢 12 pass · 🟡 6 warn · 🔴 4 fail)
Bundle size diffs
|
| // deferred status would never post — leaving the queue stuck. Instead, post on | ||
| // any merge_group where we won't retry. If the gate didn't actually defer, this | ||
| // posts a redundant failure status — harmless, since ci-status-gate already | ||
| // posted one. |
There was a problem hiding this comment.
Pull-request retries bypass the new freshness guard. If commit A fails and commit B is pushed before triage reruns commit A, GitHub retains commit A’s original GITHUB_REF; main.yml groups PR runs by that ref with cancel-in-progress, so the stale rerun can cancel commit B’s current CI. Commit B can then remain without a terminal All jobs pass status. Add an open/current-head check against HEAD_SHA, which should be analogous to the merge-queue verification.
with Love,
Codex 5.6 Sol Extra-High
Builds ready [4ca1e60] [reused from 7dce891]
⚡ Performance Benchmarks (Total: 🟢 6 pass · 🟡 5 warn · 🔴 4 fail)
Bundle size diffs
|
davidmurdoch
left a comment
There was a problem hiding this comment.
my robot says its good.

Description
In March, I started the Triage and Retry System, and it's been fully operational since April. But it's kind of been in stealth mode, where someone has to apply the
retry-cilabel for it to work.While it's been in stealth mode, I've been monitoring the decisions it would make with a custom Sentry dashboard https://metamask.sentry.io/dashboard/521041/?project=4510302346608640, and adjusting .github/rules/retry-config.jsonc so it makes better decisions about when to retry.
Now it has almost 4 months of data, and I'm confident it will make intelligent decisions.
The new rules (applies for both pull requests and Merge Queue runs)
retry-cilabel allows attempt 3 (which will remove the label), and adding it again allows attempt 4It also handles the edge cases of retry failure, cancellation, manual dequeue, and replaced merge-group entries without rerunning stale CI or leaving merge-queue status pending.
Testing
Changelog
CHANGELOG entry: null
Related issues
Closes: MetaMask/MetaMask-planning#7454
Note
Medium Risk
Changes default CI retry behavior and merge-queue status deferral for all PRs; incorrect classification or budget sync could retry real failures or leave the queue stuck, though policy is centralized and heavily tested.
Overview
Turns on the Triage and Retry System by default for PR and merge-queue runs: retryable (transient) failures on attempt 1 rerun automatically to attempt 2 without the
retry-cilabel. The label still funds extra retries (attempts 2→3 and 3→4) and is removed only on label-funded reruns, via a sharedretry-budgetpolicy used byclassify-failures.mtsand mirrored inci-status-gate.ymldeferral logic.Merge queue and stale-run safety: before rerunning failed merge-group jobs, triage verifies the queue entry and temporary ref are still current; stale or unverified entries get a terminal All jobs pass failure instead of a wasted rerun. For
pull_request, retries are skipped if the PR head SHA has moved past the failed run.Classification tuning:
retry-config.jsoncadds transient patterns (OIDC expiry, Sonar/GitHub flakes, etc.) and deterministic patterns (ESLint/node-test summaries, Sass, baselines, publish/Storybook denials) so auto-retries target infra flakes, not real breakages.New unit tests cover the retry-budget matrix and merge-queue verification helpers.
Reviewed by Cursor Bugbot for commit 4ca1e60. Bugbot is set up for automated code reviews on this repo. Configure here.