Which component are you using?:
/area cluster-autoscaler

What version of the component are you using?:

Component version: build from latest main

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

$ kubectl version
Server Version: v1.32.3+rke2r1

What happened instead?:

When using the ClusterAPI provider, lowering the cluster.x-k8s.io/cluster-api-autoscaler-node-group-max-size annotation below the current replica count creates a deadlock where the autoscaler cannot scale down.

DeleteNodes calls SetSize(replicas - 1) for each node being removed. SetSize unconditionally rejects any value above maxSize:

case nreplicas > r.maxSize:
   return fmt.Errorf("size increase too large - desired:%d max:%d", nreplicas, r.maxSize)

For example, with 4 running replicas and max annotation changed from 4 to 2, the autoscaler attempts SetSize(3) which fails because 3 > 2. The group can never converge to the desired size.

E0320 12:03:47.710228  1 delete_in_batch.go:206] Scale-down: couldn't delete empty node "node-007", status error: failed to delete nodes from group MachineDeployment/.../cpu-worker: size increase too large - desired:3 max:2

The maxSize check in SetSize should only apply when the replica count is actually increasing, not when it's decreasing toward the valid range.