We need to upgrade virtlet, which means the old virtlet daemonset pods need to be deleted and new pods will be created. I found during the upgrade, the running VMs are recreated. It is an issue for production cluster.