I have a node pool for one deployment with 200-1000 pods. They're set with a CPU based HPA. When the HPA scales down the deployment, it removes pods randomly, and eventually, I have an under-utilized node pool. The nodes aren't scaled down correctly because they still have at least one pod running.

I tried to find a solution and failed. Possible solutions, in my opinion:

1. HPA is aware of nodes utilization.
2. A PodDisruptionBudget for nodes?
3. Drain node if its CPU utilization is under a threshold.

Any help will be much appreciated.