An HPA object watches the resource consumption of pods that are managed by a controller (Deployment, ReplicaSet, or StatefulSet) at a given interval and controls the replicas by comparing the desired target of certain metrics with their real usage. For instance, suppose that we have a Deployment controller with two pods initially, and they are currently using 1,000 m of CPU on average while we want the CPU percentage to be 200 m per pod. The associated HPA would calculate how many pods are needed for the desired target with 2*(1000 m/200 m) = 10, so it will adjust the replicas of the controller to 10 pods accordingly. Kubernetes will take care of the rest to schedule the eight new pods.
The evaluation interval is ...
