The concept of platform-based cloud providers in Kubernetes are being migrated to a more agnostic approach. An important part of any provider is the container persistent storage that the cloud provides. Kubernetes Container Storage Interface, or CSI, provides an interface for exposing arbitrary block and file storage systems to containerized workloads in Kubernetes. 

While the framework exists in upstream Kubernetes, the storage provider is in charge of their own drivers. This means timely updates can be included in the platform itself and not in Kubernetes. This article will explain the process of installing and configuring CSI for vSphere and will discuss some of the benefits of migrating to CSI vs. the in-tree vSphere Storage for Kubernetes

This installation can be in tandem with the default vSphere storage class deployed and managed by OpenShift with the pre-existing or fully-automated installation experience. In the event that the new storage class is preferred, it can be made the default replacing the vSphere Storage for Kubernetes StorageClass.

Platform Requirements

The required platform for vSphere CSI is at least vSphere 6.7 U3. This particular update includes vSphere’s Cloud Native Storage, which provides ease of use in the vCenter console. 

Additionally, cluster VMs will need “disk.enableUUID” and VM hardware version 15 or higher.

# govc find / -type m -runtime.powerState poweredOn -name 'ocp-*' | xargs -L 1 govc vm.power -off $1

# govc find / -type m -runtime.powerState poweredOff -name 'ocp-*' | xargs -L 1 govc vm.change -e="disk.enableUUID=1" -vm $1

# govc find / -type m -runtime.powerState poweredOff -name 'ocp-*' | xargs -L 1 govc vm.upgrade -version=15 -vm $1

# govc find / -type m -runtime.powerState poweredOff -name 'demo-*' | grep -v rhcos | xargs -L 1 govc vm.power -on $1

Installing the vSphere Cloud Provider Interface

The next step for installing CSI on OpenShift is to create a secret, which will be used for CSI’s configuration and access to the vCenter API.

It is important to note that the cluster-id in the below must be different per cluster, or volumes will end up being mounted into the wrong OCP cluster.

# vim csi-vsphere.conf

cluster-id = "csi-vsphere-cluster"
[VirtualCenter ""]
insecure-flag = "true"
user = "Administrator@vsphere.local"
password = "SuperPassword"
port = "443"
datacenters = "RDU"

# oc create secret generic vsphere-config-secret --from-file=csi-vsphere.conf --namespace=kube-system
# oc get secret vsphere-config-secret --namespace=kube-system
NAME                TYPE     DATA   AGE
vsphere-config-secret   Opaque   1      43s

If you have some experiences with the in-tree vSphere Cloud Provider (VCP), the above format should be familiar to you. The following VMware article discusses more about this method of configuration and installation on a vanilla K8s platform.

Storage Tags for Storage Policy Names

One of the biggest limitations of the old vSphere cloud provider was the relationship between a storage class and a datastore. This one-to-one relationship made multiple storage classes a requirement and imposed inherent limitations on datastore clustering as well. When using the CSI driver, a storage policy name can be used to distribute across multiple datastores. 

The following tag-based placement rule was applied to the aos-vsphere datastore in question. Multiple datastores could have been added to this policy by assigning the same tag.  

Now that the vSphere infrastructure has been prepared, the CSI drivers can be installed. 

Installing the vSphere CSI Driver

Using the upstream CSI manifests is compatible by default with OCP, first ensure the vSphere CPI has provided each node with a ProviderID, without this the CSI will not work:

# oc describe nodes | grep "ProviderID"

NOTE: The version of manifest files is dependent on the version of vSphere. The example below uses vSphere 7U1. All of the versions are listed here:

Install the vSphere CSI driver based on what version of OCP/K8s you have - For K8s 1.17+ or OCP 4.4 and above:

# oc apply -f
# oc apply -f
# oc apply -f

For K8s <1.17 or OCP 4.3

# oc apply -f
# oc apply -f
# oc apply -f

After the installation, verify success by querying one of the new custom resource definitions:

# oc get  CSINode
NAME              CREATED AT
control-plane-0   2020-03-02T18:21:44Z

Storage Class Creation and PVC Deployment

Now, the storage class should be deployed and tested. In the event that the vSphere cloud provider storage class was also present, the additional storage class can be created then used for migrations and workload use in tandem. 

# oc get sc
NAME             PROVISIONER                    AGE
csi-sc          6m44s
thin (default)   3h29m

# vi csi-sc.yaml
kind: StorageClass
 name: csi-sc
 annotations: "false"
 StoragePolicyName: "GoldVM"

# oc create -f csi-sc.yaml

# Test the new storage class
# vi csi-pvc.yaml
kind: PersistentVolumeClaim
apiVersion: v1
 name: csi-pvc2
 annotations: csi-sc
   - ReadWriteOnce
     storage: 30Gi

# oc create -f csi-pvc.yaml

# oc get pvc
NAME       STATUS   VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS   AGE
csi-pvc2   Bound    pvc-8892e718-d89a-4267-9826-e2beb362e723   30Gi       RWO            csi-sc         9m27s

The following information should be in the events log on vCenter: 

In the event that the new CSI storage class is the preferred one, patch both of the classes:

# oc patch storageclass thin -p '{"metadata": {"annotations": \
   {"": "false"}}}'
# oc patch storageclass csi-sc -p '{"metadata": {"annotations": \
   {"": "true"}}}'


In closing, this article has laid out directions for deploying and using CSI driver support on vSphere using OpenShift. CSI driver support and container native storage greatly simplify deployment and management of persistent volume container workloads. 

Additionally, the disks  are no longer indexed with virtual machines and are considered first-class disks. 

For more information on CSI and OpenShift, check out this solution brief I worked on with VMware.


How-tos, OpenShift 4.3, vSphere, Installation

< Back to the blog