daily-fix: GCP finalize false-failed when deliverables compl
kind: infra#wf-fix#wf-fix-fp:015c846696b4#daily-auto-filed
track:Highlight text on any card (body, plan, or an event) to anchor a comment, or leave a whole-task comment here. Mention @claude to summon a reply.
No comments yet.
Overview / Motivation
Auto-filed by /daily 2026-07-04 (route 2: behavior/logic change -> independent review), from the nightly transcript problem sweep.
Goal
When the finalize/tail step exits non-zero AFTER the workload's declared deliverables verify uploaded, publish a distinct phase (e.g. finalize_failed_artifacts_ok) instead of the terminal failed that triggers crash-response machinery.
Workflow gap
- Bug observed: Twice on 2026-07-04 a GCE workload ended with eps/phase=failed from a tail/finalize non-zero exit even though every declared deliverable was verified complete on HF: #811's A100 at ~7.75h ('finalize hiccup, all deliverables complete') and #811's instance-stage tail crash after the 10.86 GB store landed. Each cost a manual diagnosis cycle to establish nothing was lost.
- Why it is a workflow gap: the failure originates in the workflow surface / helper named below, not in any one experiment.
- Confidence (emitter): medium
Scope / surfaces
- Primary target:
src/explore_persona_space/backends/gcp.py - Sessions: f2ef929f (PM loop, 00:27-01:28 UTC), 86923fc5 (#811, ~00:30-01:30 UTC). Related: #1004 (rc-masking false phase=done) is the OPPOSITE defect (done on crash); keep the two coherent.
Constraints / invariants
- Workflow-surface only;
scripts/workflow_lint.py --check-asks+--check-referencesstay green; ruff clean on touched files.
Provenance
- workflow_fix_target: src/explore_persona_space/backends/gcp.py
- source: /daily 2026-07-04 problem sweep (transcript-mined)
Activity