Built this at SAP to solve a real problem: clusters waste resources even when autoscaler thinks everything’s fine. If a node hits 50% CPU but only uses 10% RAM, the autoscaler keeps it around — but you’re paying for that unused RAM.

Node Refiner calculates excess resources across all nodes and figures out which ones can be safely drained and removed. It’s a Kubernetes Operator that runs alongside your cluster autoscaler, not a replacement for it.

Open sourced under SAP. Written in Go.