Repository navigation
🐛 fix(fleetreport): dampener suppressions are not unclassified runtime errors - #17109
Conversation
…errors Signed-off-by: scanner <280983584+hivecommons-hive[bot]@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 |
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. |
|
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor. hive/awaiting-ci: CI pending — sweep will merge when green. 🐝 Hive Agent: — hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88 Hive filed review findings that do not block this PR as backlog items:
Cap: 10 backlog item(s) per PR per day. Reviewed by Reviewer notesRead the diff against the v5 tree and the emitting site. Looks correct to me.
No findings. Review findings from Read the emitter and the handler, not just the diff. Looks correct to me.
Not approving (that is a human's call), but I found nothing to block on. Evidence bundle `#171098dba1c0` sha256:51087aad0a3f411fda9fb9455fb9cf28d88977879e9841ddadc7553e905ac4c9 — download from the dashboard or `hivectl review evidence #17109`. |
There was a problem hiding this comment.
Read the filter against the emitter: src/pkg/stagelabels/store.go:781 logs exactly "stage transition suppressed by dampener", so the exact-match in stageDampenerSuppression (slogfeed.go:235) catches it, and the early return at slogfeed.go:117 runs before Key()/LogFedClass, so no observation can leak through another class. Same shape as sentinelPacingSkip; test mirrors TestReviewRetryServerErrorLogsWithoutFleetReport. Looks correct to me.
One note, not a defect: #17103 showed 12 suppressions in 10 min on a single PR (#17073 flapping 071-needs-author ↔ 050-review-pending). Dropping the selfreport is right, but the underlying flap stays invisible to the fleet now — worth a separate look if it recurs.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
…slogfeed skip list Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com>
There was a problem hiding this comment.
Read the diff against the tree at 7e05404. Looks correct to me.
- The filter string in
src/pkg/fleetreport/slogfeed/slogfeed.go:236matches the only emitter byte-for-byte:src/pkg/stagelabels/store.go:781(slog.Warn("stage transition suppressed by dampener", …), inside the #16974 dampener branch that returnstr.Suppressed = truewithout touching the label). No other call site logs that message, so nothing else is silenced. stageDampenerSuppressionfollows the same shape as the neighbouring message-match helpers (automergeMergeGuardSkipatslogfeed.go:244,reviewRetryServerErrorat:298) and is wired into the same early-return chain infeed(:117), so the record still reaches the wrapped handler — the new test asserts both (12 records pass through, 0 observations).src/docs/fleet-report.mdandchangelog.d/fixed-17103.mddescribe the behaviour as implemented.
CI is green apart from tide pending.
— 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 at the merge-base:
- The only emitter of this message is
src/pkg/stagelabels/store.go:781(slog.Warn("stage transition suppressed by dampener", …)), so the exact-match predicatestageDampenerSuppressioninsrc/pkg/fleetreport/slogfeed/slogfeed.gocatches every instance and nothing else. - The predicate is wired into the early-return chain in
feed()ahead ofKey()/LogFedClass, the same shape asreviewRetryServerErrorandautomergeMergeGuardSkip(slogfeed.go:236-291), so the WARN still reaches the wrapped handler viaHandle()(slogfeed.go:80-86) and only the fleet observation is skipped — which is what the test asserts (12 records through, 0 observations).
One info-level note, not blocking: the new paragraph in src/docs/fleet-report.md cites pkg/fleetreport/slogfeed/slogfeed.go while the neighbouring paragraphs cite src/…-prefixed paths (src/cmd/hive/main_sweeps.go, src/crux/agent/manager_kick.go). The citation guard passed, so this is cosmetic.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
Read the change against the v5 tree; looks correct to me.
- The emitter is
src/pkg/stagelabels/store.go:781—slog.Warn("stage transition suppressed by dampener", ...), fired only fromisDampenedLockedinside the #16974 flap guard, so the exact-string match instageDampenerSuppression(slogfeed.go:235-237) catches precisely that record and nothing else. - The skip sits in the early-return chain at
slogfeed.go:117, ahead ofKey()/LogFedClass, same shape asexpectedKickHold(slogfeed.go:138). The record still reaches the wrapped handler, which the test asserts (12 forwarded, 0 observations). src/docs/fleet-report.mdgains the matching paragraph, soCloses #17111(the docs-currency follow-up) is accurate, not just #17103.
Nothing to request. Merge state is only waiting on tide/overlayfs-exec-guard.
— hive: agent=reviewer-2 backend=copilot model=claude-fable-5.1 copilot=1.0.88
There was a problem hiding this comment.
Looks correct to me. Read the emitter and the only ingestion path:
src/pkg/stagelabels/store.go:781emits exactly"stage transition suppressed by dampener"(withkey/head/from/to/reasonattrs), so thestrings.TrimSpace(r.Message) ==match in the newstageDampenerSuppression(slogfeed.go:233) is exact, not a prefix guess.slogfeed.feedis the only producer ofClassUnclassifiedRepeatedobservations (fleetreport.IsBenignMessagehas no other callers outside slogfeed), so short-circuiting infeedfully stops the selfreport; nothing else re-derives the class from logs.- The test fails without the fix:
newFeedobserves every WARN+ record, so 12 WARNs produce 12 observations andlen(*seen) != 0trips; thestrings.Count(out, "\n") == 12check confirms the records still reach the wrapped handler (local logs keep them), matching the stated intent.
One non-blocking idiom note: the three existing exact-message, attr-independent benign cases (fleetreport.go:895, :901, :905) live in IsBenignMessage, while the slogfeed predicates are used for matches that also need to inspect attrs. This one needs no attr inspection, so it could have gone in IsBenignMessage; either placement is functionally equivalent here.
Docs paragraph in src/docs/fleet-report.md and the changelog.d/fixed-17103.md fragment match the behaviour change.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Thank you for your contribution! Your PR has been merged. We'd love to hear how your experience was: share feedback |
…time error (#17147) The stage engine's same-head reversal guard (suppressStageFlap) logs "stage flap suppressed" at WARN each time it refuses to undo its last label write. That is the guard working as designed, but the slogfeed skip added in #17109 only matched the stage-label store's dampener message, so repeated suppressions still filed unclassified_error_repeated selfreports. Skip both messages in stageDampenerSuppression. Fixes #17142 Hive-Run: #17142 Hive-Plan: selfreport-triage Hive-Spec: fleet-report#slogfeed-skip-list Signed-off-by: Karibhou <Karibhou@users.noreply.github.com> Co-authored-by: Karibhou <Karibhou@users.noreply.github.com>
The dampener WARN (
stage transition suppressed by dampener) is the stage engine working as designed; counting it as a repeated unclassified error files fleet selfreports for benign flap suppression. The slogfeed handler now skips it (stays in local logs), with a test. Closes #17103— hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88
Closes #17111