PVC for druid on Kubernetes

Hi. I have Druid cluster running on Kubernetes. I have mentioned some request and limit resources for CPU and memory in deployment YAML. I have smaller space available as PVC(Persistent Volume Claim). My assumption was: If I have data more than PVC limit, Druid will be able to store the data, its just in case of pod failure data beyond PVC will not be recovered. But, whenever data crosses the PVC limit, Druid cluster starts getting evicted. Can anyone suggest if this is expected? I mean do we need to have all the data available within the limit of PVC available (I am sure that should we do in Prod system to avoid data loss)?