@Vikash_Singh Back to your original issue. Can you please check what type are the executors?
kubectl get statefulsets
Does it return dremio-executor as one of them? If not you have an older version of helm charts where the executors are replicasets and yes, when the pod goes down for whatever reason, the uploads, reflections stored on the pod are lost. With the use of statefulsets, if the pod go down and come back up, you will not loose uploads, reflections as there is a persistent volume associated with each executor.
However, if you scale down your executors, there is a potential of loosing uploads and reflections. The reflections will be refreshed. For uploads, you can store them in ADLS. Check the latest helm charts where we make it easier to configure this.
If you have all the executors and their IP addresses have not changed - you should not get the error about “data with hard affinity”. Can you confirm your scenario?
Don’t try to manually configure this on AKS but use helm charts.