node-rotation-controller — Specification
Functional specification for a Kubernetes controller that proactively rotates Karpenter-managed nodes in a make-before-break (surge) fashion within a configurable maintenance window, before Karpenter's forceful expireAfter fires.
Japanese translation: docs/ja/specification/
Contents
- Overview — Background · Goals · Non-Goals · Terminology · Ecosystem Position
- Scope — Compatibility · Composition with Existing Mechanisms
- Design — Maintenance Window · Candidate Selection · Surge Sequence · Mid-surge Protection · Pod-level Behavior · Forceful Fallback · Zonal Workloads · Rollback · Backstop
- Operations — Capacity / Availability · Observability · RBAC · Cost
- Implementation — Architecture · Reconcile Loop · State Model · Configuration Schema
- Release — Versioning · Roadmap
- Risks & Status — Risks · Validated Assumptions · Open Questions
References
- Karpenter Disruption (official docs)
- Karpenter forceful-expiration design — establishes "user-side implementation" as a valid path
- Karpenter Discussion #1079 — Schedule for disruption — whitelist limitation of Disruption Budgets
- EKS Auto Mode docs
- EKS Auto Mode and maintenance window for "Drifted" nodes (AWS re:Post)