On the OpenShift cluster customer have, for example 2 "types" of machines set, one using SpotVM ("spotVMOptions: {}" in the machineset definition), one using the same VM type but without SpotVM (on-demand capacity). 
- Customer need these machines (whether it's Spot or not) to be reserved for the application.
- Hence they define Taints on each machineset and their equivalent   Tolerations in the namespace of the application.
- If for some reasons the HPA is scaling up the number of replicas of the application and there is no capacity available in one of the machineset (because there is no spot instance capacity left or because customer ran out of arm64 VM capacity in the region). 

- Would the Cluster Autoscaler be able to detect this (as the deployment of the machine will be failing) and try to increase the replicas of another machineset ? 
- Is there any way to define a deployment strategy (40% "Spot instances", 60% "on-demand") that would be compatible with the use of HPA/CA ?
 

Customer need this for their application specific requirement.(>3000 machines for this application in production at scale on 4 Azure regions).