Repository navigation
ci(runners): send twelve default-runner jobs to the kirocrew-runners group - #25645
Conversation
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks. It reuses the fork-safe label form already used in 31 other workflows, the check names stay the same, and each site can be reverted with a one-line change. Notes
[DESIGN-REVIEWED] b4b5391 Description read[DESCRIPTION-READ] f5a1a10252b0cd261e779070455a2b5efc012d38c6add4e85a57bd933d4d3a3a That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS This only moves CI runners, and the PR is explicit that a fork falls back to Notes
What this change shipsInventory (13 items) — 12 justifiedIntent: keep this repository's maintenance and nightly CI lanes out of the shared, enterprise-wide standard runner queue. ADDITION (an operational change).
[FIRST-PRINCIPLES-REVIEWED] b4b5391 Description read[DESCRIPTION-READ] a75b3a917b2af037406d5da079d3c151b4bf7dcb7458337408e511cfe9283d7a That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
|
@tim-finnigan this branch now has merge conflicts with |
56f0d51 to
3a85dbf
Compare
|
HOLD at The change itself looks sound: twelve Blocker: Non-blocking AI notes, for the record: the Opus note that the doc says — crew-pr-reviewer (r135) |
3a85dbf to
3f9e9ea
Compare
3f9e9ea to
3c83cbb
Compare
|
self-added: no
|
|
/ai-review override fable 29831b2 deferred #26425: agent: owner-directed deferral. The finding is correct and is not rebutted. All five open routing PRs append to the same ci-and-reviews inventories, so a per-PR doc edit puts them back in conflict with each other. #26425 makes that edit once per landed wave. |
Human judgment recorded@tim-finnigan recorded a deferred release (class
This decision applies only to this commit. A new push requires a new judgment. |
|
Hold: one product question about Reviewed head: 29831b2 The change itself is sound. Twelve The open question: your sibling #25349 measured PR Readiness. Please confirm whether the cap blocker applies here. Either answer clears this: the cap has moved, a maintainer accepts landing it now, or the sweep is dropped from this PR.
AI findings judged non-blocking: Design (label literal repeated across workflows, rename cost) does not block because the parity test pins every site, so a missed rename fails CI. — crew-pr-reviewer (r137) |
…group Twelve jobs across ten workflows move off GitHub's shared standard pool onto the `kirocrew-runners` larger-runner group, named by its runner label with a fork fallback rather than by `group:`, because an organization group is unreachable from a fork and a job targeting one there waits forever for a runner. Nothing gets faster: the group is the same CPU and memory as a public-repository `ubuntu-latest`, so only queue position changes, and larger-runner minutes are billed where standard ones are free. Check names are unchanged, including nightly's `pod-scenarios`, whose `name:` and `os:` values stay exactly as they were with the decision moved into `runs-on` per leg.
29831b2 to
b4b5391
Compare
Problem / Motivation
Goal: Move twelve jobs across ten workflows off GitHub's shared standard
runner pool onto the
kirocrew-runnerslarger-runner group.Why it matters
The standard GitHub-hosted pool shares its concurrency enterprise-wide, so a
job on it waits behind work this repository does not control. The owner's call
is that these lanes should not sit there.
Not a goal
public-repository
ubuntu-latest.build-desktop.yml'sbuild-desktopjob orbuild-ffmpeg-decoder.yml'sbuildjob. Both pinubuntu-22.04andubuntu-22.04-armto match the fleet image's glibc 2.35 floor; that is acompatibility decision of its own.
What changed (motivation → approach → change)
Twelve jobs asked for a shared runner; now they ask for the group this
repository pays for. Nothing else about them changes.
main-red-watch.yml,main-red-pr-notice.yml,main-macos-failure-watch.ymlwatch,notice,pageissue-summary.yml,issue-triage.ymlsummarize,triagenightly.ymlversion,process-leak-invariant,pod-scenarios(Linux leg only)pr-readiness-sweep.ymlsweepgui-user-test.ymlgui-user-testbuild-desktop.ymlsmoke-linux-packagesbuild-ffmpeg-decoder.ymlpublishEleven were a hardcoded
runs-on: ubuntu-latestand become:The label, not
group:. A group resolves only inside the organization thatowns it, so a fork running its own copy of one of these workflows finds no
runner and the job waits forever with nothing to fall back to. The label form
carries its fallback in the
||leg. That costs the rename safety agroup:reference buys: rename the definition and these twelve lines break together.
The larger-runner totals in
docs/ci/ci-and-reviews.mdare left toPR #26303, which removes
those running totals, so this PR edits no doc.
nightly.yml'spod-scenariosis a three-leg matrix. Itsname:and itsos:values are untouched, so the rendered check names do not move; only theLinux leg's runner is decided, per leg, the way
ci.yml'se2e-boot-matrixdoes it.
Nothing gets faster, and that is the honest framing: same CPU and memory as
a public-repository
ubuntu-latest, so only queue position changes, andlarger-runner minutes are billed where standard ones are free. These lanes hold
no merge, so this is not a latency win being bought.
One exception is worth a reviewer's attention.
pr-readiness-sweepis how acompleted lane reaches the
PR Readinessverdict at all, becausepr-readiness.ymldoes not subscribe to thecompletedevent of the workflowsit monitors. A sweep that waits for a runner delays a green verdict reaching
the required status. It cannot make a verdict wrong, and cannot turn a red into
a merge, but for that one job queue position is the latency.
Backwards compatibility
Nothing is rejected, renamed or removed. Check names, triggers, permissions and
steps are identical, and a fork keeps
ubuntu-latestthrough the||leg.Rollback is a one-line revert per site.
Tests
test/test_ci_fleet_routing_expression_parity.pypins every off-fleet job tothe exact runner it is expected to use. Three of these jobs are already
_PERMANENT_EXCEPTIONSentries, so their_HOSTED_EXCEPTION_RUNNERSpin movesfrom the
ubuntu-latestdefault to the new expression. The pin stays an exactper-job equality, so a drift to a third runner still fails.
test/test_main_macos_failure_watch.pyasserts its watcher's runner byequality; that assertion now names the new expression. It is re-pinned, not
relaxed, for the same reason: a drift to a self-hosted runner, where "no
checkout" stops being the whole isolation story, must still fail.
No new test. The property both files already pin is exactly the one this diff
changes, and
a-ratchet-may-only-tightenwants it re-pinned rather thanwidened.
Manual verification
python3 scripts/local-gate.pybackend leg on the rebased head: 1265 passed,118 skipped across the 14 related files. Its frontend leg could not run on this
host, which failed
npm ciwith a registryE401before any check; nonode_modulesexists on this machine for any checkout, and this diff containsno
website/file.Both changed test files re-run alone and serially,
python -m pytest -n0 test/test_ci_fleet_routing_expression_parity.py test/test_main_macos_failure_watch.py: 34 passed, 26 skipped. Theflake-pattern ratchet, which the related-test runner does not select, is also
green:
python -m pytest -n0 test/test_flake_pattern_ratchet.py, 64 passed.Every one of the twelve
runs-onvalues was re-read back out of the parsedYAML and compared against the intended expression, together with
pod-scenarios's unchangedname:andos:list and the two deliberatelyuntouched matrix jobs.
Checked at the changed/unchanged boundary:
scripts/ci/runner_watchdog.pyactsonly on labels carrying its
codebuild-prefix, so it neither heals nornotices a
kirocrew-linux-4-corejob, and no job here gained a fleet label.runner.osconditions insidepod-scenariosstill readLinuxon the movedleg. No moved job declares a
container:orservices:block.Screenshots / video
Why no screenshot: CI runner routing only; no frontend file is touched and
no pixel changes.
Related Issues
Closes #25644
Checklist
npm cifailure that reproduces on every checkout on this machinedocs/ci/ci-and-reviews.mdare PR #26303's to remove