08/31/2022, 9:06 PM
Has anybody here used Karpenter instead of ClusterAutoscaler(CA) for EKS cluster node scaling? Currently we use CA with an on-demand managed node group. We are thinking about add new type node instance. With CA, we will need to add additional managed node group. Looks like this is completely unnecessary with Karpenter. So, how do we migrate from CA to Karpenter with a smooth transition? Could we have both CA & Karpenter deployed? Here is my plan, 1) deploy Karpenter; 2) scale current managed node group down to 0 instance which will force all existing pod to be moved to nodes that are managed by Karpenter provisioner I assume. 3) Finally we will decommission the managed group & undeploy CA. Will the above plan work? Is there better ways to do this? Thanks in advance for any tips.