CNS-CSI 2.1 Improvements in vSphere with Kubernetes
There are a handful of improvements that have been made in this version of the CNS-CSI driver.
1. Volume Health
This feature will report the health of volumes that have been deployed on vSphere storage and can be queried from the PersistentVolumeClaim details. This feature is available in the vSphere with Kubernetes Supervisor Cluster and Tanzu Kubernetes Grid (TKG) “guest” clusters in vSphere 7.0U1.
chogan@chogan-a01 ~ % kubectl get pvc NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS AGE demo-pvc-vsan Bound pvc-d347c2ed-fcc6-4e3a-9fd2-a293fa568835 2Gi RWO vsan-default-storage-policy 36m chogan@chogan-a01 ~ % kubectl describe pvc | grep volumehealth volumehealth.storage.kubernetes.io/health: accessible
More details regarding Volume Health can be found here.
2. Volume Placement Observability
Another nice feature is that Physical Volume Placement details can now be seen directly from the Containers Volume view in the vSphere client.
3. Static PV support in both Supervisor and TKG cluster
This release also sees the ability to support Static Persistent Volumes in both the vSphere with Kubernetes Supervisor cluster, as well as the Tanzu Kubernetes Grid clusters. One of the main uses cases for Static PVs is to be able to re-use an existing PV, and schedule a new Pod or Pods to use it. This could be from a previously deployed application, or indeed a PV that has been restored from a backup. Further details about Static Persistent Volumes provisioning can be found here.
4. Offline Volume Grow
This was a feature that was introduced in the vSphere 7.0 CNS-CSI release for vanilla Kubernetes, although it was still a beta feature at the time. I talked about it in detail in this blog post. In this 7.0U1 release, we added the same functionality to Tanzu Kubernetes Grid clusters. It is still an offline operation, meaning that the Pod will need to be unscheduled, and rescheduled after the size of the PV has been increased. More details on how to do an offline volume grow operation can be found here.
CNS-CSI 2.1 Improvements in vanilla Kubernetes
There is one major improvement that we are making to the vanilla Kubernetes CSI driver.
1. CSI Migration (beta)
While we could offline migrate stateful applications between VCP and CSI using Velero, there is a requirement to enable this to be done seamlessly with a Live Migration feature. CSI Migration will be available in the Kubernetes 1.19 release as beta functionality. To be able to use migration of volumes from the earlier, in-tree VCP to CSI on vSphere, customers will also need vSphere 7.0U1 and CSI driver version 2.1. Persistent Volume (PV) operations, run on a Kubernetes cluster with the VCP, will be seamlessly redirected to the out-of-tree CSI driver.
Migration between the older, in-tree VCP and the newer vSphere CSI Driver is complicated by the fact that the VCP uses the path to a VMDK for the Persistent Volume lifecycle operations, whereas the CSI Driver uses the First Class Disk ID. The APIs needed for migration workflows (relating VMDKs to FCDs, and vice-versa) were added to vSphere 7.0U1.
More detailed information will follow, but this should certainly assist customers who are on distributions or have applications that currently use the older, in-tree, vSphere Cloud Provider (VCP), as at some point the in-tree VCP will be deprecated and removed from the Kubernetes distributions.