What is the use case for Pod Autoscaler given that costs are in the Nodes?

HPA will add or remove more Pods as needed. But this requires available Nodes in the Cluster, so these are not autoscaled and the maximum potentially needed number of Nodes must be kept running.

Most expense is in the VMs/machines underlying the Nodes. Without cluster autoscaling, I could just keep the maximum number of Pods, never scaling down, and it would make no financial difference.

So how does HPA help?

Higher priority pods can preempt lower-priority ones.

Some clusters also have cluster autoscaling enabled when pods are pending formlack of resources.

1 Like