daily-fix: plan-time RAM/RSS routing gate + p90 fence sizing
kind: infra#wf-fix#wf-fix-fp:e001f5b88f11#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
planner section 9 sizes each CPU phase's projected RSS and routes >=~16 GB (or >2 concurrent multi-GB phases) off the shared VM to cpu-mid/cpu-bigmem — the RAM sibling of the 50 GB disk carve-out; and sizes --max-run-duration off the p90 per-cell estimate, not the mean.
Workflow gap
- Bug observed: 2026-07-03: #778's 22-GiB-RSS null battery was earlyoom-killed 3x on the starved shared VM (125 GB, 95G used) before a cpu-bigmem pivot; #833's chain fits lost 5 cells to earlyoom (two ~13-15 GB phases concurrently resident) and then overran its 36h GCP fence because per-cell wall-time ran ~2x the plan mean.
- Why it matters: The existing section-9 rules gate on disk footprint and wall-time only; RSS on the shared VM is the recurring killer.
- Confidence (emitter): medium
Scope / surfaces
- Primary target:
.claude/rules/plan-compute-sizing.md, .claude/agents/planner.md - 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: .claude/rules/plan-compute-sizing.md, .claude/agents/planner.md
- fingerprint: e001f5b88f11
- source: /daily 2026-07-03 problem sweep (transcripts of 2026-07-03 UTC)
Activity