Cluster Over-Provisioning

The Kubernetes Cluster Autoscaler (CA) for AWS configures AWS EC2 Auto Scaling group (ASG) of the EKS node group to scale nodes in cluster when there are pods pending to be scheduled.

This process of adding nodes to a cluster by modifying the ASG inherently adds extra time before pods could be scheduled. For example, in the previous section you will have noticed it took several minutes before the pods created when the application was scaled up became available.

There are numerous approaches to solve this problem. This lab exercise solves this by "over provisioning" the cluster with extra node(s) that run lower priority pods used as placeholders. These lower priority and are evicted when critical application pods are deployed. The empty pods not only make sure CPU and Memory resources are reserved but also IP addresses assigned from the AWS VPC Container Network Interface - CNI.