ClusterAutoscaler does not scale up nodes evenly across AZ even with with balanceSimilarNodeGroups: true

ClusterAutoscaler does not scale the nodes evenly in AzurePublicCloud OCP IPI

4.10.30

100%

1. Install an OCP cluster on AzurePublicCloud.
2. Create 3 machinesets each in single AZ
3. Use the below ClusterAutoscaler definition:

apiVersion: autoscaling.openshift.io/v1
kind: ClusterAutoscaler
metadata:
  name: default
spec:
  podPriorityThreshold: -10
  ignoreDaemonsetsUtilization: true
  balanceSimilarNodeGroups: true
  scaleDown:
    enabled: true
    delayAfterAdd: 30m
    delayAfterDelete: 5m
    delayAfterFailure: 3m
    unneededTime: 60m

4. Deploy application workload to trigger node scaleup

Uneven node scale up happens, for example like below:

# oc get machinesets -A
NAMESPACE               NAME                              DESIRED   CURRENT   READY   AVAILABLE   AGE
openshift-machine-api   useast2-2-frqmg-worker-eastus21   7         7         7       7           4d6h
openshift-machine-api   useast2-2-frqmg-worker-eastus22   3         3         3       3           4d6h
openshift-machine-api   useast2-2-frqmg-worker-eastus23   19        19        19      19          4d6h
```

The nodes should be evenly spread across all the availability zones.