Kubernetes cost-efficiency via scale-down of underutilized nodes with Persistent Volume Claim

We are excited to deliver additional cost savings for Kubernetes users running their workloads on Ocean by Spot. We now automatically scale down any underutilized nodes with Persistent Volume Claims (PVCs).

Previously, our Autoscaler Scale Down process, operating under an abundance of caution, would not consider pods with stateful workloads (utilizing PVCs) on underutilized nodes for migration. We have made enhancements to our bin packing algorithm and can reliably move pods with PVCs to scale down underutilized nodes. The cost savings achieved by this add a new layer of cost-efficiency to our core cost optimization solution for AWS spot instances, reservations and savings plans.

With this new feature, the Ocean Autoscaler would relocate Kubernetes Pod/s (from underutilized nodes) with a PVC to other nodes (when possible) within the infrastructure. Currently, we support scale down on AWS, for PVCs delivered via AWS Elastic Block Store (EBS), Elastic File Store (EFS) and any 3rd party Cloud Native Storage (block/file) providers.

For workloads that do not handle disruption very well, such as monolithic containerized apps or cron jobs, we advise to label them with spotinst.io/restrict-scale-down set to true. Any node running a pod with this label will not be scaled down by Ocean’s Autoscaler.       

As an Ocean by Spot user, there’s no configuration needed to take advantage of this capability. You should see additional cost savings accumulate in your account over time.