Repository navigation
🐛 fix(fleetreport): treat the hive-held stage dwell warning as benign - #15017
Conversation
The stage labeller warns once per PR per quiet period when a PR waiting on the hive sits past its stage dwell. The repeated-error feed keys on the message alone, so a busy hive with a backed-up reviewer lane had every held PR warn once and the burst across unrelated PRs was filed as an unclassified repeated runtime error. The condition stays in the log and the dashboard's hive-held count. Hive-Run: hivecommons#15009 Hive-Plan: selfreport benign-message exclusion Hive-Spec: fleetreport#benign-messages Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01HRH6kiRiadEpX2Et372Lop Signed-off-by: Karibhou <Karibhou@users.noreply.github.com>
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of No conflicting open PRs or same-file overlaps were found. |
|
Another agent tried to open a PR from |
There was a problem hiding this comment.
Read against v5 8e26835. Looks correct to me.
src/pkg/fleetreport/fleetreport.go:786-787lowercases the message before thebenignSubstringsloop at:821-825, so the all-lowercase entry"stage labels: pr waits on the hive past the stage dwell"matches the live warning text atsrc/pkg/github/automerge/automerge_stage_labels.go:967("stage labels: PR waits on the hive past the stage dwell — reviewer lane needs attention"). The test's negative case ("stage labels: provisioning failed") does not share the substring, so otherstage labels:failures stay reportable.- The entry is a substring, not an exact match, following the
forward-merge guardentry directly above it — consistent with the file's pattern.
Relation to #15020: that PR fixes one source of the warning (stageActivity flagging a no-linked-issue contributor PR as waiting on the hive). It does not replace this one — the warning still fires legitimately for held 050 PRs on a backed-up reviewer lane, and this PR is what keeps that burst out of unclassified_error_repeated. Different files, no conflict; both can land.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
Looks correct to me.
Verified against v5: IsBenignMessage lower-cases the input before matching (src/pkg/fleetreport/fleetreport.go:787), so the lower-case "stage labels: pr waits on the hive past the stage dwell" substring matches the mixed-case warning emitted at src/pkg/github/automerge/automerge_stage_labels.go:967. The warning is latched per PR by HiveHeldWarned (same file, line 968-969), so the "15 in 10m" burst in #15009 is indeed many PRs warning once each, as the body says.
Not a duplicate of #15020. #15020 fixes the specific misdiagnosis in the #15009 diagnostic (reason=contributor-lane:no-linked-issue should never count as waiting on the hive) at stageActivity; this PR stops the fleet feed from filing any instance of the warning as a repeated runtime error. The two are complementary and both can merge. Note the trade-off this one makes: after it lands, a genuinely backed-up reviewer lane is visible only in that hive's log and dashboard hive-held count, never fleet-wide — consistent with the #14727/#14430 precedent, but worth a maintainer knowing.
— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Ready for a human merge: verdict approved @2803565, 57/57 checks green, DCO yes, soak 30m0s elapsed. Blocked only by The hive never approves on your behalf: |
…issues, show and release hive-set holds (#15080) The #9665 reporter-trust merge hold judged #15020, a hive-authored PR, via a bare '#15017' mention that resolved to a sibling PR, attributed that PR's author as an untrusted issue reporter and held the PR two seconds after it opened; the tracker then showed only 'hold +1 · Needs human decision'. - Gate rationale is closing-keyword and declared issue links only; a link that resolves to a PR is skipped; hive self-reports are trusted; the gate is skipped for hive-authored PRs (kept for contributor and shared-PAT PRs). - Stage-label pass records hold provenance (by/source/reason/since) from the reporter-trust notice and the newest hold label event; tracker pill reads 'Held by hive · reporter trust (issue #N by @x, contributor)' or 'Held by @login'; the decision popover asks release/keep with Release leading; the 085 comment names who set the hold and why. - The same pass releases a hive-set reporter-trust hold on a hive-authored PR and logs 'stage labels: releasing self-imposed reporter-trust hold'; operator holds are left alone. Fixes #15075 Signed-off-by: Andrew Anderson <andy@clubanderson.com>
hanthor
left a comment
There was a problem hiding this comment.
Reviewed at 2803565. No blocking findings. IsBenignMessage lowercases the producer message before substring matching, so the new entry matches the live mixed-case warning. The change filters only its generic repeated-runtime-error observation while preserving the original warning and stage/dashboard hive-held signal. The negative regression leaves unrelated stage-label failures reportable. This complements #15020 rather than duplicating it. Current full check rollup passed apart from normal missing-approval Tide gating; retain all normal merge gates.
|
/lgtm Human review completed for 2803565. Accepting the explicit classification trade-off: the warning/dashboard signal remains, but this operational dwell message no longer creates a generic repeated-runtime-error fleet report. Normal CI and merge gates remain required. |
|
LGTM label has been added. DetailsGit tree hash: b8cd4480a96e54e4e30c6ec2f54649adfbdb069a |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: hanthor The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
Summary
stage labels: PR waits on the hive past the stage dwell — reviewer lane needs attention(src/pkg/github/automerge/automerge_stage_labels.go) is already one-shot per PR:HiveHeldWarnedlatches it until the PR sees activity, and the same condition is surfaced as the dashboard's hive-held count (#12534).The generic repeated-error feed (
slogfeed.Key) keys on the normalised message only — thepr=attribute is not part of the key — so on a busy hive whose reviewer lane is backed up, each held PR warning once adds up to "15 in 10m" of one condition and is filed asunclassified_error_repeated. That is a per-hive operational signal, not a runtime defect.This follows the established pattern (#14727, #14430, #14452, …): add the message to
benignSubstringsinsrc/pkg/fleetreport/fleetreport.go, with a regression test inruntime_test.go(benign match + a negative case so stage-label failures stay reportable) and achangelog.d/fixed-15009.mdfragment.Tests were not run locally per AGENTS.md; CI is the verdict.
Fixes #15009
🤖 Generated with Claude Code
https://claude.ai/code/session_01HRH6kiRiadEpX2Et372Lop
— hive: backend=claude model=claude-opus-5-5
🐝 Hive Agent:
contributor| SHA:28035652a