In September last year we announced the addition of Karpenter as experimental feature. Since then we have been making improvements to our implementation and gradually used some pilots internally and with some customers. We are happy to announce that Karpenter is now deployed by default (via AWS Fargate) on all our EKS clusters and we’ve migrated the system NodePool to use it instead of the standard Cluster Autoscaler.
As an immediate result we’ve noticed, as expected, much more optimized resource usage and immediate cost savings on most of our clusters. In the coming weeks, when thigs settle down, we’ll have a more concrete data for all our environments.
Next, in the coming months, our team will be evaluating together with our customers on migrating application workloads to run on Karpenter-managed NodePools too.