Repository navigation
Description
Target: v6.
Target branch:
v6only. Open the PR with--base v6.hive-open-prdefaults to the repo's default branch (v5) when--baseis omitted.
Citations:path:lineagainstv6@26e5fc227, paths relative tosrc/.
An adopted PR's 074 stall clock runs while the PR is still waiting, unattempted, in the fixer's queue, so a busy single fixer turns queued rebases into 085 parks.
Owner decision needed before implementing: should the 074 dwell pause while the fixer has not reached the PR, or should the fixer's throughput change instead (for example more than one adopted rebase per kick, or a dedicated adoption fixer)? This issue describes the first option. The second would close it too.
What happened
projectbluefin/utah-packages on 2026-10-28. Four contributor fork PRs, #761, #772, #773 and #776, were adopted in place (072 → stage/074-adopted-by-hive) between 11:32 and 11:36Z. They were conflicting, so each needed a rebase.
- They were routed correctly. At 11:43:04Z one cycle routed 11 adopted rows to the single
scanner(9 rebase, 2 fix-ci), logged asadopted PR routed to fixer work=rebase(cmd/hive/merge_eligibility.go:1220). The rows were re-routed every cycle after that. - The scanner worked one PR per kick. It rebased feat: K8s scaffold — Dockerfile, kustomize, secrets #770 at 13:46Z, but did not reach the other four within the dwell.
- All four parked at once. At 14:31–14:35Z, exactly at the 3h 074 dwell, they parked at
stage/085-needs-decisionasadoption-stalled:conflicting. - The park killed the queued fix. At about 15:00Z the scanner reached them and refused: "not resolved by hive because PR lacks stage/074-adopted-by-hive".
So the stall was the queue, not anything the hive was blocked on. The park then sent the work to a maintainer.
Code
- The stall clock.
adoptionStale(pkg/github/automerge/automerge_contributor_adopt.go:1486) measures fromRecord.AdoptedHeadSince()(pkg/stagelabels/store.go:913), the adoption or the hive's last push. - Its consumers.
inPlaceRehomeReason(:1523, which givesrebase-rehomefor idle rebase work) andadoptionStalledReason(:1545) act on it inescapeStalledAdoption(:1572). - No "attempted" record exists. Nothing records whether the fixer has attempted a queued conflict row for a head. The escalation store's re-engagement ledger (
StaleRed,ReEngageEligible) covers red-CI PRs only.
Change (if the owner picks the dwell pause)
- Record when the fixer first attempted an adopted row's head. The attempt is the first pushable row of a scanner kick that the agent accepted. Persist it per (repo, PR, head) next to the stage record, or in the fixer-routing state that writes
ci-failing.json. Also expose when the row was first queued. - Pause the clock. In
adoptionStale, while the PR is queued and not yet attempted on this head, count the dwell frommax(AdoptedHeadSince, attemptedAt)instead ofAdoptedHeadSince. - Hard cap. Add a cap, for example 4× the 074 dwell from
AdoptedHeadSince, so a dead or paused fixer still escalates. Do not pause at all when the fixer is unavailable (agentAvailability, used atcmd/hive/main.go:7794). - Unchanged: the escalation note and its reasons.
Acceptance criteria
- An adopted PR that has waited in the fixer queue, unattempted, for longer than the 074 dwell is not parked at 085 for that reason.
- Once the fixer has attempted the head, the normal dwell applies from the attempt.
- With the fixer unavailable or paused, or past the hard cap, the PR parks as today.
- A same-repo or in-place adoption with no queued fixer row behaves as today.
Test plan
automerge_adoption_escape_test.go/automerge_adoption_rebase_test.go, each with a fake fixer-queue source:- queued and unattempted past the dwell: no stall and no rehome;
- attempted 1h ago on a 3h dwell: not stale;
- attempted 4h ago: stale;
- fixer unavailable: stale at the dwell;
- past the hard cap while queued: stale.
- A unit test for the attempt record: a new head resets it.
- Docs: the 074 dwell description in
src/docs/labels-and-control-signals.md(thestage/074-adopted-by-hiverow) andhive-merge.md.
Out of scope
- How many PRs the scanner handles per kick.
- The fork re-home refusal (🔒 fix(adoption): never move a fork PR's code into a base-repo branch (re-home ran fork code as trusted) #14306, 🐛 fix(adoption): keep a fork PR the hive cannot push to with its author until the 072 dwell #15018).
- The merge-queue base refresh and review-verdict reset, which are separate PRs.
Activity
hivecommons-hive commentedon Oct 28, 2026
🔒 Claimed by @thatguywob (contributor on hive hosted-available-oke-11-placeholder-r05x). The claim lasts 30m0s without activity and renews while the work continues. Reply
/unclaimto release, or/claimto take it over if you outrank the holder.thatguywob commentedon Oct 28, 2026
The issue explicitly leaves an owner decision open before implementation: pause the stage/074 dwell while an available fixer has queued but not yet attempted the adopted head (with the proposed hard cap), or change fixer throughput instead. Please select the intended approach; the proposed implementation and acceptance tests are conditional on choosing the dwell pause. No owner choice is recorded in the current issue comments.
Verified that
upstream/v6exists atc7edf3af5; this is a design choice in this issue, not a missing base branch or companion checkout. No implementation PR was opened.— hive: backend=codex
🐝 Hive Agent:
contributor| SHA:unknownhivecommons-hive commentedon Oct 28, 2026
🔓 Claim by @thatguywob released (lease revoked).
hivecommons-hive commentedon Oct 28, 2026
What to reply
/hive approve: go with the recommendation/hive decision <your instructions>: go ahead, following your instructions/hive help: show these againIf what this issue waited on has been done (a merged fix, or an action taken outside this repository), a maintainer can reply
/fixedto close it as completed; Hive clears its park labels.Only a maintainer with write access can un-park; everything else is ignored. Full command reference: https://github.com/hivecommons/hive/blob/v5/src/docs/maintainer-commands.md.
hivecommons-hive commentedon Oct 28, 2026
/hive decision go with the recommendation; where options remain, take the one that keeps work moving and is reversible
Rationale: no owner rule singles this out; the standing default is the hive's own recommendation.
Decided by the hive at @clubanderson's request as the owner's delegate under the owner-delegate policy (rule
default-approve).hivecommons-hive commentedon Oct 28, 2026
Un-parked by @clubanderson. Decision: go with the recommendation; where options remain, take the one that keeps work moving and is reversible
stage/045-needs-decisionis cleared when present andapproved-directionis applied; anyholdlabel is left exactly as it was.Recorded from the hive dashboard.
hivecommons-hive commentedon Oct 28, 2026
🔒 Claimed by agent
scanner(agent on hive hosted-available-oke-11-placeholder-r05x). The claim lasts 2h0m0s without activity and renews while the work continues. Reply/unclaimto release, or/claimto take it over if you outrank the holder.hivecommons-hive commentedon Oct 29, 2026
Fixed by #15131 (merged 9f58413). Closing. Reply /reopen if the problem persists.
github-actions commentedon Oct 29, 2026
@Danathar this issue was closed by @hivecommons-hive[bot] as completed.
If this was closed by a merged PR, the fix only takes effect once it ships in a tagged release and your deployment updates to it — check your running version against the fix before concluding the fix failed.
If you are still seeing the problem, comment
/reopenand it will be reopened for you — you do not need write access. Please add what you are still observing (running version, logs, steps) so it is not closed the same way again.