daily-fix: poller CPU-phase stall predicate + per-instance idle counter
kind: infra#wf-fix#wf-fix-fp:07c2abee2b55#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 2026-07-03 (backfill run 2026-07-04) from the day's transcript problem sweep.
Goal
Stall predicate consults pod/instance-side process-CPU%% and output-file mtime before declaring a CPU-analysis phase stalled; the idle-GPU advisory keys its minutes counter to the current instance's creation timestamp (reset on relaunch). Verify first what #873's landed tripwire already covers.
Workflow gap
- Bug observed: 2026-07-03: the poller false-stalled 'constantly' on #813's CPU-bound analysis tail (~6h) — dismissed each time by checking process CPU + output mtimes manually; and the gpu-idle-advisory printed a stale CUMULATIVE minutes counter (486 min / 543 min) for instances only ~17 min old — two false 'burning' reads in one day.
- Why it matters: False alarms train operators to ignore the monitor — the one channel that catches real spend leaks (#664 class).
- Confidence (emitter): medium
Scope / surfaces
- Primary target:
scripts/poll_pipeline.py, src/explore_persona_space/backends/ - Grep the workflow surface for the pattern before editing and update every hit; list them in the plan.
Constraints / invariants
scripts/workflow_lint.pydefault run passes; ruff on touched files passes.- This task was auto-filed by the /daily three-route classifier (route 2 — behavior/logic change, independent review).
Provenance
- workflow_fix_target: scripts/poll_pipeline.py, src/explore_persona_space/backends/
- fingerprint: 07c2abee2b55
- source: /daily 2026-07-03 problem sweep (transcripts of 2026-07-03 UTC)
Activity