Hi All,
I'm using k8s-v1.19.13. I found that the pod will be stuck on terminating status after the node is dead. According to the CSI spec, the CO should unpublish the volume after the pod is finished. So if one pod(created by Deployment with 1 replica) has mounted a RWO pvc(e.g. iscsi-volume) and the node is out of power by accident, then the pod will never reborn again on other nodes. Because the dead pod got stuck on the offline node and could not finish finalizing job, the new pod keeps being on 'ContainerCreating' status and says the pvc was already used in another pod.
The reason for using 1-replica deployment mounting RWO pvc is that the application did't provide any HA solutions, and I can't figure out if I can use more than 1 replica mouting same RWX pvc for this application correctly.
Is there any way to release the volume gracefully? Or how to make the RWO pvc works properly in use cases like this?
Thanks!