Repository navigation
fix(pr-merge): lane_room reads its post-merge answer past the 2-minute unverified age and disarms the lane #9577
Description
Problem
While any lane merge is unverified, lane_room (babysit_merge_policy.py) reports the lane room as unreadable, with reason post-merge answer unverified-stale. enable-auto-merge then disarms every armed lane pull request.
The shared post-merge answer may be at most SHARED_UNVERIFIED_MAX_AGE old while a merge is unverified (_shared_age_refusal). That limit is run_cache.MAX_AGE, 2 minutes. The post-merge-dispatch step takes longer than that on its own. In run 40060427179 the answer was stamped computedAt 03:58:09Z, the step finished at 03:59:45Z, and enable-auto-merge read the answer at 04:00:10Z, 2:01 later. So the answer is always stale by the time lane-room reads it, and every legitimately pending merge makes the room unreadable rather than counted.
The gate path handles the same case by recomputing the post-merge check and writing the answer back. lane_room does not recompute. in_flight already counts unverified lane merges toward merge.max-in-flight, so the room is held anyway while a merge is pending.
Fix
Have lane_room recompute the post-merge answer on unverified-stale, as the gate does. Do not ignore the age: a verification failure after computedAt must still refuse. A new test should show that a pending merge with an answer older than 2 minutes yields a counted room, not an unreadable one.
Evidence
- Runs 40059227921, 40059975039, 40060427179 and 40062210911 all log
lane room: {"refusal":null,"unreadable":"post-merge answer unverified-stale",...}. - The trigger was feat(performance): add an optional holdout harness to performance:climb, off by default #9554. It stayed pending on
read-back-unrecordedbecause its native arm enqueued at once with no AutoMergeEnabledEvent; that is fixed separately on fix/postmerge-native-immediate-enqueue.
Refs: #9508
Activity
github-actions commentedon Oct 24, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Small
Complexity: Medium
Uncertainty: Known
Flags: none
Understood: while a lane merge is unverified,
lane_roomreads a post-merge answer older than the 2-minute limit and reports the room unreadable. enable-auto-merge then disarms the armed lane PRs. The proposed fix is forlane_roomto recompute the answer onunverified-stale, as the gate path does. It keeps the age check, so a verification failure aftercomputedAtstill refuses. It also adds a test for a pending merge with an answer older than 2 minutes. The issue is clear and I need nothing more from you. I found no likely duplicates among the open issues. #9508 is already referenced.Automated triage, applied by the lanes App.
melodic-automation-lanes commentedon Oct 24, 2026
Not admitted to the issue lane: it has no
- [ ]task list to check. A recorded decision does not change these rules (.github/actions/resolve-config/admission-rules.json); a person who wants it worked addsagent-ready.kyle-sexton commentedon Oct 24, 2026
Scope addition (requested by the GitHub Support session, which monitors stuck PRs): a room hold logs only to the console today ('#N left: the lane's room could not be read ...', run 40064383267 at 05:00:38Z), so no PR or annotation shows why nothing arms. When lane_room refuses or is unreadable, enable-auto-merge emits one ::warning:: annotation and a step-summary line naming the reason and each PR left unarmed. Skip a per-PR comment: the room is re-read every scheduled tick.
fix(pr-merge): keep an unverified-stale answer nothing has moved sinc…
fix(pr-merge): carry lane records into the room and end a kept answer…
fix(pr-merge): report a room hold's pull requests by outcome and narr…
fix(pr-merge): drop tracker references and run citations from the roo…