Eve library

ISIS v2 WS09-004: RunPod Warm/Cold Policy Optimization

Guide

Revision 1

On this page

Scope#

Implemented workload-class + SLA-tier warm/cold policy optimization for RunPod endpoint desired-state management.

Implemented Artifacts#

  • Policy contract:
    • infra/runpod/endpoints/warm-cold-policy.json
  • Policy apply/check script:
    • scripts/isis/apply_runpod_warm_cold_policy.mjs
  • Desired-state optimized by policy:
    • infra/runpod/endpoints/desired-state.json
  • CI/CD preflight gates enforcing policy alignment:
    • .github/workflows/runpod-endpoint-lifecycle.yml
    • .github/workflows/runpod-endpoint-drift-detection.yml
    • .github/workflows/runpod-endpoint-reconcile.yml
    • .github/workflows/runpod-endpoint-image-build.yml

Optimization Model#

Policy dimensions:

  • Workload class: interactive-short, interactive-medium, batch-long, sequence-long
  • SLA tier: critical, standard, economy
  • Warmness profile: warm, balanced, cold

Policy outputs per endpoint/environment:

  • workersMin
  • workersMax
  • idleTimeout

Staging values are automatically scaled down from production policy targets.

Verification#

bash
node scripts/isis/apply_runpod_warm_cold_policy.mjs --environment all --check
python3 scripts/isis/validate_runpod_desired_state.py --file infra/runpod/endpoints/desired-state.json

Outcome#

WS09-004 is now closed with deterministic warm/cold endpoint sizing rules, SLA-aware policy tiers, and enforced CI preflight validation to prevent configuration drift.