Cluster Autoscaler incorrectly skips nodes with unschedulable field in scheduling simulation

[macsko]

Which component are you using?:

/area cluster-autoscaler

What version of the component are you using?:

I found it in the code and it seems to be there for a long time.

Component version:

What k8s version are you using (kubectl version)?:

kubectl version Output

$ kubectl version

What environment is this in?:

What did you expect to happen?:

If there is a pod that tolerates the node.kubernetes.io/unschedulable taint, then it could be schedulable on a node with unschedulable: true.

tolerations:
- key: "node.kubernetes.io/unschedulable"
  operator: "Exists"
  effect: "NoSchedule"

It should be honored by CA's plugin runner.

What happened instead?:

Plugin runner has a shortcut, that checks only the unschedulable field:

autoscaler/cluster-autoscaler/simulator/clustersnapshot/predicate/plugin_runner.go

Lines 77 to 81 in 955d0ca

	// Nodes with the Unschedulable bit set will be rejected by one of the plugins during the Filter phase below. We can check that quickly here
	// and short-circuit to avoid running the expensive Filter phase at all in this case.
	if nodeInfo.Node().Spec.Unschedulable {
	continue
	}

Then, if there is such pod with toleration specified above, it won't be taken into consideration by CA.

How to reproduce it (as minimally and precisely as possible):

Anything else we need to know?: