Real-world EKS
Module infographic
62 / 69

Revision poster: moving from Cluster Autoscaler to Karpenter.

Infographic: before and after of Cluster Autoscaler versus Karpenter and the five migration steps
Click the poster to enlarge it

Top: the before and after. Below: five steps, each reversible until the old group is deleted.

  1. 1Before: fixed node groups scaled by the Cluster Autoscaler.
  2. 2After: nodes launched just in time and sized to pending Pods.
  3. 3Install beside the old groups, define NodePool and EC2NodeClass, then scale the old group down gradually.
  4. 4Drain old nodes and verify the workloads.
  5. 5Remove the old autoscaler and its IAM permissions last.

Self-check: cover the poster and answer

  • Why keep a small managed node group for Karpenter itself?
  • What should you pin in the EC2NodeClass for production?

Move capacity gradually and keep the old group until the new one is proven.

BackNext