Karpenter
Last updated 2026-06-04
Karpenter is an open-source Kubernetes node provisioner, originally from AWS, that launches and removes nodes in real time to fit pending pods. Instead of relying on fixed, pre-defined node groups, it watches for unschedulable pods and provisions just-in-time capacity, picking instance types and sizes that match what those workloads actually request. For example, if a batch job needs more memory than the current nodes provide, Karpenter can launch a right-sized instance within seconds rather than waiting on a node group to scale. It also consolidates pods onto fewer nodes when utilization drops and removes the now-empty capacity, and it can prefer cheaper spot capacity for interruptible workloads, which can cut compute cost substantially. By matching the compute layer closely to real demand, it is a key tool for optimizing Kubernetes node spend. LevelFour reads Karpenter configuration to recommend and apply node-level optimizations alongside pod rightsizing.
Frequently asked questions
- How is Karpenter different from the Kubernetes Cluster Autoscaler?
- The Cluster Autoscaler scales pre-defined node groups up and down by a fixed instance type. Karpenter is groupless: it provisions individual nodes just-in-time, choosing the instance type and size that best fit pending pods, and it consolidates and removes underused nodes, giving it more flexibility to bin-pack and use diverse capacity.
- Can Karpenter use spot instances to reduce cost?
- Yes. Karpenter can provision spot capacity for interruptible workloads, often diversifying across instance types and zones to reduce the impact of reclamation. It handles spot interruption notices by draining and replacing affected nodes, making it a practical way to run fault-tolerant pods on discounted capacity while staying within scheduling constraints.
See also
LevelFour automates this across AWS, GCP, Azure, and Kubernetes with automated infrastructure-as-code pull requests.