A step-by-step migration guide from Kubernetes Cluster Autoscaler to Karpenter on EKS. Covers the motivation (8% average CPU utilization in autoscaled clusters, slow provisioning), prerequisites (IAM roles, subnet/security group tagging), parallel-run isolation strategies using taints or nodeSelectors, node group to NodePool/EC2NodeClass mapping with full YAML examples, incremental workload cutover, rollback procedures at every stage, and common pitfalls like missing discovery tags, over-restrictive instance requirements, missing PodDisruptionBudgets, and aggressive consolidation on stateful workloads.
Table of contents
Key takeawaysWhy teams migrate from Cluster Autoscaler to KarpenterPrerequisitesStep by step: migrating from Cluster Autoscaler to KarpenterRollback and safetyCommon migration pitfallsAfter migration: closing the CPU waste gapFrequently Asked Questions89 Impressions