Repository navigation
π fix(adoption): an adopted red PR is routed to the fixer once per red episode, and not at all once parked at 085 - #17081
Conversation
An adopted contributor PR red at an unchanged head was logged "adopted PR routed to fixer" on every eligibility pass (61 times each for projectbluefin/hplip-printer-app#199 and ghostscript-printer-app#190 in eight hours) and stayed in the scanner's FIX-BEFORE-NEW list even after the hive parked it at stage/085 with nothing to push, so the scanner re-triaged an upstream fsdk-containers bug it had already diagnosed. - Route and announce fix-ci once per (head, failing-check set); the row stays listed and queued for the 074 dwell between announcements. - An adopted PR whose current head is parked at 085 (AdoptionEscalated) leaves ci-failing.json until the head moves or the escalation is released, with one log line naming the reason. - The sentinel alert alone still does not block repair (hivecommons#13280). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com>
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
|
@Danathar this PR is back in your court ( Failing checks:
Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path. Queued in your relay fix queue (2 waiting): fixes are handed out before new work as soon as you have capacity. |
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
changelog-fragment-guard failed because the fragment sat under src/changelog.d/, which is never compiled into CHANGELOG.md. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1 Signed-off-by: Danathar <Danathar@users.noreply.github.com>
There was a problem hiding this comment.
Conclusion: looks correct to me; the diff does what the body says and I found no defect.
What I read on origin/v6:
adoptedFixCIParkedrelies onstagelabels.Record.AdoptionEscalated(head)(src/pkg/stagelabels/store.go:1635), which is head-pinned, so a new push on a parked PR correctly falls through to the fixer path β the test atadopted_fix_ci_routes_test.go:112-121covers exactly that.- The release paths the park depends on exist and don't need the fixer to have run:
ClearAdoptionEscalationis called from the #14500 recovery (automerge_contributor_adopt.go:2405), from stall-reason clearing (:2435), and fromautomerge_adoption_intent.go:186. Removing the row fromci-failing.jsontherefore cannot strand a parked PR. retain(openPRs)is built fromcandidates, which isactionable.PRs.Items + Held(merge_eligibility.go:1316-1321) β the full open set β so a PR temporarily held does not lose its episode and get re-announced. Thelen(openPRs) > 0guard against an empty read is a sensible defensive choice.fixCIEpisodesorts and compacts check names, so check-order churn in the GitHub payload is not a new episode (covered atadopted_fix_ci_routes_test.go:71-77).
One observation, not a defect: a CIStatus: failure row with an empty FailingChecks (checks not yet loaded) forms the episode fix-ci\0<head>\0, and once check names arrive it is announced a second time. That seems like the right behaviour β the second line carries information the first did not β just noting it so nobody is surprised by two lines at one head.
Scope note for the author: as the body says, until the 085 park lands the row still appears in ci-failing.json on every sweep; only the log line is deduplicated. So the "re-triaged by a scanner kick" cost in the Summary is addressed by the park check, not the episode ledger.
β hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Hive status β updated in place; edits do not notify. Full history is in the Activity Monitor. Review findings from Read the diff against
One wording note (info): the changelog fragment and |
Summary
An adopted contributor PR that stays red at the same head was "routed to fixer" on every eligibility pass, and it stayed in the scanner's FIX-BEFORE-NEW list even after the hive had parked it at
stage/085saying it had nothing to push. Two projectbluefin printer-app PRs were routed 61 times each in eight hours. One scanner kick spent two background sub-agents re-triaging them, and came to the same "upstream bug, nothing to push" answer it had already posted. With this change, each red episode (the head plus its failing checks) is routed once. A PR the hive has parked at 085 on its current head leaves the fixer list until the head moves or the park is released.What went wrong
2b9a441, unchanged since 11-06 13:11Z) and Reduce the printer application downloadΒ projectbluefin/ghostscript-printer-app#190 (head42b7364) loggedadopted PR routed to fixer(work=fix-ci, fixer=scanner) about every seven minutes, 61 times each between 11:08Z and 19:15Z on 11-07. Make the printer application download smallerΒ projectbluefin/gutenprint-printer-app#196 and fix(parity): resolve factory outputs libfdk-aac through libfdk-aac-freeΒ projectbluefin/utah-packages#917 did the same until new pushes. It is still happening on 7b9eeb8: fix: eliminate GraphQL from api-collector, read actionable from REST cacheΒ #199 and add send button for custom agent promptsΒ #190 were routed at 18:27Z, 19:08Z and 19:16Z.include/slim-printing.ymlacross the junction. That bug is tracked upstream (Refs [scanner] fix(printing): make the shared slim recipe usable from the printer applicationsΒ projectbluefin/fsdk-containers#509, triage/accepted 18:17Z). The scanner already said so on fix: eliminate GraphQL from api-collector, read actionable from REST cacheΒ #199 and add send button for custom agent promptsΒ #190 at 03:07Z ("No code was pushed; the upstream fix is tracked in fsdk-containers#509").stage/085withadoption-stalled:...("the hive has nothing it can push"). The merge-eligibility pass kept listing them anyway. The scanner's 17:30Z governor kick carried them in FIX-BEFORE-NEW and started a "Repair hplip 199 + ghostscript 190" sub-agent, which again pushed nothing and blamed fsdk-containers.What changes
writeMergeEligibleremembers the last red episode it routed for each adopted PR. When the head and the set of failing checks are unchanged, the row stays listed for the fixer's kick and queued for the 074 dwell, but it is not routed or logged again. A new head or a newly failing check routes once. This follows how adopted rebases were already announced once per attempt (π fix(automerge): time-box adopted-PR rebase attempts and re-home when exhaustedΒ #16028).stagelabels.Record.AdoptionEscalated) is not fixer work. It is left out ofci-failing.json, and oneadopted PR parked at needs-decision; fix-ci not routedline names the reason. A new head, or a release of the escalation (blocker cleared, a Adopted PRs red only from a since-fixed shared-CI flake are escalated instead of updated/re-runΒ #14500 recovery attempt, operator intent), puts it back.TestClassifyMergeEligibility_AdoptedPRReachesFixerPastSentinel), so this PR does not change it.Files
src/cmd/hive/adopted_fix_ci_routes.go: the per-PR episode ledger (in memory, pruned to open PRs) and the 085 park check.src/cmd/hive/merge_eligibility.go: the fix-ci branch uses both.src/docs/labels-and-control-signals/stage-labels.md: one sentence on the new behavior.src/changelog.d/fixed-adopted-fix-ci-route-once.mdTests
TestWriteMergeEligible_AdoptedFixCIRoutedOncePerHeadAndCheckSet: same head and same checks (reordered) on the second sweep: listed, routed 0 times. A new failing check routes once. A new head routes once, then goes quiet.TestWriteMergeEligible_AdoptedPRParkedAt085IsNotFixCIWork: parked at this head: not listed and not routed on two sweeps, with one park line. Released: routed once. A new head after a park on the old head: routed once.2161745f3): the second sweep re-routes, and the parked PR is listed and routed. Both pass with the fix.go test ./cmd/hive/passes in full.go vet, and the docs link and citation checks, are clean.Follow-up, not in this PR: the fixer has no machine-readable way to say "red because of an open issue in another repo".
hive-ci-externalcovers only a fork parent that is red too. The scanner's "tracked in fsdk-containers#509" note was prose, so until the 085 park the PR stays on the fixer's list.π€ Generated with Claude Code
https://claude.ai/code/session_01BQYnPGFJx81c8JXR3o1sg1