I'm running some Flink jobs on my EKS cluster in AWS. I am using spot instances to run these jobs, but am running into a problem with a PVC (EBS volume). I am fairly certain the issue is that the EBS volume only binds to one node in my cluster, and every time the spot instance essentially "restarts", the redeployed pod is not able to bind to the volume again, due to the spot instance having a new IP address.
I can clearly see this happening when the instance "restarts" due to the age of the node being a few minutes, and the Flink application cannot reschedule due to a volume node affinity conflict.