vSphere Cloud provider and VMDK dynamic provisioning

Red Hat Logo

The release of OpenShift Container Platform 3.6 brings support for vsphere cloud provider. This provides vsphere VMDK dynamic provisioning for persistent volumes for container workloads. The storage presented to vsphere virtual machines as a VMDK has ReadWriteOnce access mode.

In the OCP 3.6 on vSphere reference architecture, much of this process is automated and can be implemented easily.

Virtual Machine Disks or VMDKs exists in virtual machines. Configuring the OCP cluster for vsphere cloud provider support requires:

  • Master Node Configuration
  • Application Node Configuration
  • vCenter or vSphere Requirements

The kubernetes docs do a great job of highlighting the requirements for vSphere.

First, both master and nodes need the following parameters set and services restarted:

  • cloud-provider=vsphere
  • cloud-config=path_to_config.conf

Master configuration

$ vi /etc/origin/master/master-config.yaml

kubernetesMasterConfig:
  apiServerArguments: 
    cloud-config:
    - /etc/vsphere/vsphere.conf
    cloud-provider:
    - vsphere
  controllerArguments: 
    cloud-config:
    - /etc/vsphere/vsphere.conf
    cloud-provider:
    - vsphere

$ sudo systemctl restart atomic-openshift-master.service

Node configuration

$ vi /etc/origin/node/node-config.yaml

kubeletArguments:
  cloud-config:
  - /etc/vsphere/vsphere.conf
  cloud-provider:
  - vsphere

$ sudo systemctl restart atomic-openshift-node.service

Next, the vsphere.conf file should loosely resemble this:

$ cat /etc/vsphere/vsphere.conf 
[Global]
user = "administrator@vsphere.local"
password = "vcenter_password"
server = "10.*.*.25" 
port = 443
insecure-flag = 1
datacenter = Boston
datastore = ose3-vmware-prod
working-dir = /Boston/vm/ocp36/
[Disk]
scsicontrollertype = pvscsi

The variables are discussed more in the Kubernetes document. The working-dir variable is the folder that houses the OpenShift guest machines. In vSphere, the folder syntax will be /Datacenter/vm/foldername.

Using GOVC

The tool govc is a GO-based application for interacting with vSphere and vCenter.

The documentation for govc is located in github:

https://github.com/vmware/govmomi/tree/master/govc

First, download govc then export the vars that govc needs to query vCenter:

curl -LO https://github.com/vmware/govmomi/releases/download/v0.15.0/govc_linux_amd64.gz
gunzip govc_linux_amd64.gz
chmod +x govc_linux_amd64
cp govc_linux_amd64 /usr/bin/

export GOVC_URL='vCenter IP OR FQDN'
export GOVC_USERNAME='administrator@vsphere.local'
export GOVC_PASSWORD='vCenter Password'
export GOVC_INSECURE=1

$ govc ls
/Boston/vm
/Boston/network
/Boston/host
/Boston/datastore

$ govc ls /Boston/vm
/Boston/vm/ocp36

$ govc ls /Boston/vm/ocp36
/Boston/vm/ocp36/haproxy-0
/Boston/vm/ocp36/app-0
/Boston/vm/ocp36/infra-0
/Boston/vm/ocp36/haproxy-1
/Boston/vm/ocp36/master-0
/Boston/vm/ocp36/nfs-0

vSphere Pre-requisites

  • disk.enableUUID

Next, the enableUUID parameter needs to be set on all virtual machines in the OCP cluster. The option is necessary so that the VMDK always presents a consistent UUID to the VM, this allows the new disk to be mounted properly.

enableUUID

The enableUUID option can be set at the template or VM in the vCenter client. Next, the template would then be used to deploy the OpenShift VMs.

Additionally, govc can be used to set this as well:

for each VM in `govc ls /Boston/vm/ocp36/`;do govc vm.change -e="disk.enableUUID=1" -vm="$VM"

Lastly, with the cloud provider configuration in place a new storage class to deploy VMDKs can be created.

Storage Classes and VMDK Provisioning

The datastore storage class can be created with the following yaml file:

$ vi cloud-provider-storage-class.yaml
kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
  name: "ose3-vmware-prod"
provisioner: kubernetes.io/vsphere-volume
parameters:
    diskformat: zeroedthick
    datastore: "ose3-vmware-prod"

oc create -f cloud-provider-storage-class.yaml

Now that the StorageClass object is created. The oc command can be used to verify the StorageClass exists.

$ oc get sc
NAME               TYPE
ose3-vmware-prod   kubernetes.io/vsphere-volume

$ oc describe sc ose3-vmware-prod
Name:           ose3-vmware-prod
IsDefaultClass: No
Annotations:    <none>
Provisioner:    kubernetes.io/vsphere-volume
Parameters:     datastore=ose3-vmware-prod,diskformat=zeroedthick
Events:         <none>

OpenShift can now dynamically provision VMDKs for persistent container storage within the OpenShift environment.

$ vi storage-class-vmware-claim.yaml

kind: PersistentVolumeClaim
apiVersion: v1
metadata:
  name: ose3-vmware-prod
  annotations:
    volume.beta.kubernetes.io/storage-class: ose3-vmware-prod
spec:
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 2Gi

$ oc create -f storage-class-vmware-claim.yaml

$ oc describe pvc ose3-vmware-prod

Name:           ose3-vmware-prod
Namespace:      default
StorageClass:   ose3-vmware-prod
Status:         Bound
Volume:         pvc-cc8a9970-7c76-11e7-ae86-005056a571ee
Labels:         <none>
Annotations:    pv.kubernetes.io/bind-completed=yes
                pv.kubernetes.io/bound-by-controller=yes
                volume.beta.kubernetes.io/storage-class=vmware-datastore-ssd
                volume.beta.kubernetes.io/storage-provisioner=kubernetes.io/vsphere-volume
Capacity:       2Gi
Access Modes:   RWO
Events:
  FirstSeen LastSeen    Count   From                SubObjectPath   Type        Reason          Message
  --------- --------    -----   ----                -------------   --------    ------          -------
  19s       19s     1   persistentvolume-controller         Normal      ProvisioningSucceeded   Successfully provisioned volume pvc-cc8a9970-7c76-11e7-ae86-005056a571ee using kubernetes.io/vsphere-volume

Now, in vCenter, a couple of changes are initiated:

First, the new disk is created.

ocp-new-disk

Secondly, the disk is ready to be consumed by a VM to be attached to a POD.

Lastly, while datastores are generally accessible via shared storage across a vCenter cluster, the VMDKs are tied to a specific machine. This explains the ‘ReadWriteOnce’ limitation of the persistent storage.


 

Share