node-rotation-controllerGraceful node rotation for Karpenter

Karpenter's expireAfter drains nodes at unpredictable times — this controller rotates them gracefully, inside your maintenance window, before that happens.

Get Started

Specification

Runbook

Zero-downtime surge

A replacement node is ready before the old one is drained. The controller induces NodePool-owned capacity via a temporary placeholder Pod — Karpenter provisions the node, PDBs govern the drain.

Window-bounded rotation

Rotation starts only inside your configured maintenance window. An automatically derived age threshold ensures every node gets multiple chances to rotate before its expireAfter deadline.

Safe by default

expireAfter is kept as a backstop, never removed. If the controller is absent or a rotation fails, nodes still expire exactly as they would without it — never worse than the status quo.

Validated on EKS Auto Mode

End-to-end tested on real EKS Auto Mode clusters — including a 12-hour unattended soak, zonal-PV rebind, leader failover, and the graceful-to-forceful-fallback boundary.