fix: address Bugbot round-5 findings on the release (doctor ready, task label, e2e gate)#383
Merged
Merged
Conversation
…acity probe checkNodeFit's StatusWarn covers two different situations: a can't-check (RESOURCE_REQUESTS unreadable, nodes unlistable) and the soft GPU fallback. The rollup only special-cased StatusFail, so a can't-check Warn fell through to ✔ Ready — the same false-green class already special-cased for Pod-health list failures. Can't-check now rolls up to the honest · can't-check line; the GPU-soft Warn stays Ready (CPU fallback). Bugbot round 5 on the release PR (#377). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
MAX(task) is a lexicographic string max: after --overwrite with a different --task, data list could keep grouping the dataset under the old task. Join on MAX(started_at) per table instead — started_at ships in the same DDL that creates the journal, so it exists whenever the table does; on a missing column the query errors and the listing degrades to inferred modality (existing best-effort path). Bugbot round 5 on the release PR (#377). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The suite gates used !cancelled() + the probe's outputs; when the changes job itself fails, outputs read empty and unlabeled PRs skipped both suites — a flaky detector silently dropped the pre-merge e2e gate this workflow restores. Run the suites on needs.changes.result == 'failure'. Bugbot round 5 on the release PR (#377). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Contributor
Author
|
bugbot run |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
✅ Bugbot reviewed your changes and found no new issues!
Comment @cursor review or bugbot run to trigger another review on this PR
Reviewed by Cursor Bugbot for commit 8e8bf8e. Configure here.
saadqbal
approved these changes
Jul 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The three Medium findings from Bugbot's re-scan of release PR #377 (head 685cf0b):
summarizeDoctoronly special-casedNode capacityFail, so a can't-check Warn (RESOURCE_REQUESTS unreadable / nodes unlistable) fell through to ✔Ready to run training— readiness asserted without a capacity probe. Can't-check now rolls up to the honest · can't-check line (mirroring the Pod-health list-failure case); the GPU-soft Warn intentionally stays Ready (CPU fallback).applyTaskLabelsusedMAX(task)— a lexicographic max, so--overwritewith a different--taskcould keep the old grouping. Now picks the task of the latest run (join onMAX(started_at)per table;started_atships in the journal's CREATE, and a missing column just degrades to the existing inferred-modality path).needs.changes.result == 'failure'(fail closed).Related
Ref #377 — Bugbot round 5. Threads answered there.
Type of change
Test plan
go build ./...,go test ./internal/cli/ ./internal/push/ -count=1— passbash scripts/check-style.sh— clean; workflow YAML parsesstarted_atexists since the table's introduction (fix(doctor): stop requests-proxy readiness false-green — honest neutral until the egress probe (cli#351) #369)Checklist
go build ./...,go vet, and the Lint job's checks pass locallybash scripts/check-style.shpasses🤖 Generated with Claude Code
Note
Low Risk
Targeted bug fixes in doctor messaging, a MySQL label query, and CI job conditions; no auth, secrets, or core training paths changed.
Overview
Addresses three Bugbot findings on release readiness: honest
doctorrollups, correctdata listtask labels, and a safer e2e pre-merge gate.tracebloc doctorno longer shows ✔ Ready to run training when Node capacity is only a Warn because capacity could not be verified (unlistable nodes or unreadableRESOURCE_REQUESTS). Those cases now roll up to a neutral couldn't check free compute line, matching the existing Pod-health list-failure behavior. GPU-soft capacity Warns still count as ready (CPU fallback).data listtask enrichment now uses the latest ingest run per table (MAX(started_at)join) instead ofMAX(task), so re-ingest with--overwriteand a different--taskis not stuck on the old lexicographic task string.Integration e2e (
integrationanddelete-teardown) now also run when the PRchangespaths job fails, so a broken probe does not skip both suites with empty outputs (fail-closed).Tests cover the new doctor rollups (can't-check → Unknown; GPU-soft → still Ready).
Reviewed by Cursor Bugbot for commit 8e8bf8e. Bugbot is set up for automated code reviews on this repo. Configure here.