Overview¶

At this point the CSI driver and CSP should be installed and configured.

PVC access modes¶

The HPE CSI Driver for Kubernetes is primarily a ReadWriteOnce (RWO) CSI implementation for block based storage. The CSI driver also supports ReadWriteMany (RWX) and ReadOnlyMany (ROX) using a NFS Server Provisioner. It's enabled by transparently deploying a NFS server for each Persistent Volume Claim (PVC) against a StorageClass where it's enabled, that in turn is backed by a traditional RWO claim. Most of the examples featured on SCOD are illustrated as RWO using block based storage, but many of the examples apply in the majority of use cases.

The NFS Server Provisioner is not enabled by the default StorageClass and needs a custom StorageClass. The following sections are tailored to help understand the NFS Server Provisioner capabilities.

• Using the NFS Server Provisioner
• NFS Server Provisioner StorageClass parameters
• Diagnosing the NFS Server Provisioner issues

Enabling CSI snapshots¶

Support for VolumeSnapshotClasses and VolumeSnapshots is available from Kubernetes 1.17+. The snapshot beta CRDs and the common snapshot controller needs to be installed manually. As per Kubernetes SIG Storage, these should not be installed as part of a CSI driver and should be deployed by the Kubernetes cluster vendor or user.

Install snapshot beta CRDs and common snapshot controller (once per Kubernetes cluster, independent of any CSI drivers).

git clone https://github.com/kubernetes-csi/external-snapshotter
cd external-snapshotter
git checkout release-2.0
kubectl apply -f config/crd -f deploy/kubernetes/snapshot-controller

git clone https://github.com/kubernetes-csi/external-snapshotter
cd external-snapshotter
git checkout release-3.0
kubectl apply -f client/config/crd -f deploy/kubernetes/snapshot-controller

Base StorageClass parameters¶

Each CSP has its own set of unique parameters to control the provisioning behavior. These examples serve as a base StorageClass example for each version of Kubernetes. See the respective CSP for more elaborate examples.

# Renamed csi.storage.k8s.io/resizer-secret-name to
# csi.storage.k8s.io/controller-expand-secret-name
#
# Alpha features: PVC cloning and Pod inline ephemeral volumes
#
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  annotations:
    storageclass.kubernetes.io/is-default-class: "true"
  name: hpe-standard
provisioner: csi.hpe.com
parameters:
  csi.storage.k8s.io/fstype: xfs
  csi.storage.k8s.io/controller-expand-secret-name: hpe-backend
  csi.storage.k8s.io/controller-expand-secret-namespace: kube-system
  csi.storage.k8s.io/controller-publish-secret-name: hpe-backend
  csi.storage.k8s.io/controller-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-publish-secret-name: hpe-backend
  csi.storage.k8s.io/node-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-stage-secret-name: hpe-backend
  csi.storage.k8s.io/node-stage-secret-namespace: kube-system
  csi.storage.k8s.io/provisioner-secret-name: hpe-backend
  csi.storage.k8s.io/provisioner-secret-namespace: kube-system
  description: "Volume created by the HPE CSI Driver for Kubernetes"
reclaimPolicy: Delete
allowVolumeExpansion: true

# Alpha feature: volume expansion
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  annotations:
    storageclass.kubernetes.io/is-default-class: "true"
  name: hpe-standard
provisioner: csi.hpe.com
parameters:
  csi.storage.k8s.io/fstype: xfs
  csi.storage.k8s.io/resizer-secret-name: hpe-backend
  csi.storage.k8s.io/resizer-secret-namespace: kube-system
  csi.storage.k8s.io/controller-publish-secret-name: hpe-backend
  csi.storage.k8s.io/controller-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-publish-secret-name: hpe-backend
  csi.storage.k8s.io/node-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-stage-secret-name: hpe-backend
  csi.storage.k8s.io/node-stage-secret-namespace: kube-system
  csi.storage.k8s.io/provisioner-secret-name: hpe-backend
  csi.storage.k8s.io/provisioner-secret-namespace: kube-system
  description: "Volume created by the HPE CSI Driver for Kubernetes"
reclaimPolicy: Delete
# Required to allow volume expansion
allowVolumeExpansion: true

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  annotations:
    storageclass.kubernetes.io/is-default-class: "true"
  name: hpe-standard
provisioner: csi.hpe.com
parameters:
  csi.storage.k8s.io/fstype: xfs
  csi.storage.k8s.io/controller-publish-secret-name: hpe-backend
  csi.storage.k8s.io/controller-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-publish-secret-name: hpe-backend
  csi.storage.k8s.io/node-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-stage-secret-name: hpe-backend
  csi.storage.k8s.io/node-stage-secret-namespace: kube-system
  csi.storage.k8s.io/provisioner-secret-name: hpe-backend
  csi.storage.k8s.io/provisioner-secret-namespace: kube-system
  description: "Volume created by the HPE CSI Driver for Kubernetes"
reclaimPolicy: Delete

Common HPE CSI Driver StorageClass parameters across CSPs.

¹ = Parameter is mutable using the CSI Volume Mutator.

Provisioning concepts¶

These instructions are provided as an example on how to use the HPE CSI Driver with a CSP supported by HPE.

• Create a PersistentVolumeClaim from a StorageClass
• Ephemeral inline volume
• Raw block volume
• Using CSI snapshots
• Expanding PVCs
• Using PVC overrides
• Using volume mutations
• Using the NFS Server Provisioner

Create a PersistentVolumeClaim from a StorageClass¶

The below YAML declarations are meant to be created with kubectl create. Either copy the content to a file on the host where kubectl is being executed, or copy & paste into the terminal, like this:

kubectl create -f-
< paste the YAML >
^D (CTRL + D)

To get started, create a StorageClass API object referencing the CSI driver Secret relevant to the backend.

These examples are for Kubernetes 1.15+

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: hpe-scod
provisioner: csi.hpe.com
parameters:
  csi.storage.k8s.io/fstype: xfs
  csi.storage.k8s.io/controller-expand-secret-name: hpe-backend
  csi.storage.k8s.io/controller-expand-secret-namespace: kube-system
  csi.storage.k8s.io/controller-publish-secret-name: hpe-backend
  csi.storage.k8s.io/controller-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-publish-secret-name: hpe-backend
  csi.storage.k8s.io/node-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-stage-secret-name: hpe-backend
  csi.storage.k8s.io/node-stage-secret-namespace: kube-system
  csi.storage.k8s.io/provisioner-secret-name: hpe-backend
  csi.storage.k8s.io/provisioner-secret-namespace: kube-system
  description: "Volume created by the HPE CSI Driver for Kubernetes"
  accessProtocol: iscsi
reclaimPolicy: Delete
allowVolumeExpansion: true

Create a PersistentVolumeClaim. This object declaration ensures a PersistentVolume is created and provisioned on your behalf, make sure to reference the correct .spec.storageClassName:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-first-pvc
spec:
  accessModes:
  - ReadWriteOnce
  resources:
    requests:
      storage: 32Gi
  storageClassName: hpe-scod

After the PersistentVolumeClaim has been declared, check that a new PersistentVolume is created based on your claim:

kubectl get pv
NAME              CAPACITY ACCESS MODES RECLAIM POLICY STATUS CLAIM                STORAGECLASS AGE
pvc-13336da3-7... 32Gi     RWO          Delete         Bound  default/my-first-pvc hpe-scod     3s

The above output means that the HPE CSI Driver successfully provisioned a new volume. The volume is not attached to any node yet. It will only be attached to a node if a scheduled workload requests the PersistentVolumeClaim. Now, let us create a Pod that refers to the above volume. When the Pod is created, the volume will be attached, formatted and mounted according to the specification.

kind: Pod
apiVersion: v1
metadata:
  name: my-pod
spec:
  containers:
    - name: pod-datelog-1
      image: nginx
      command: ["bin/sh"]
      args: ["-c", "while true; do date >> /data/mydata.txt; sleep 1; done"]
      volumeMounts:
        - name: export1
          mountPath: /data
    - name: pod-datelog-2
      image: debian
      command: ["bin/sh"]
      args: ["-c", "while true; do date >> /data/mydata.txt; sleep 1; done"]
      volumeMounts:
        - name: export1
          mountPath: /data
  volumes:
    - name: export1
      persistentVolumeClaim:
        claimName: my-first-pvc

Check if the Pod is running successfully.

kubectl get pod my-pod
NAME        READY   STATUS    RESTARTS   AGE
my-pod      2/2     Running   0          2m29s

Ephemeral inline volume¶

It's possible to declare a volume "inline" a Pod specification. The volume is ephemeral and only persists as long as the Pod is running. If the Pod gets rescheduled, deleted or upgraded, the volume is deleted and a new volume gets provisioned if it gets restarted.

Ephemeral inline volumes are not associated with a StorageClass, hence a Secret needs to be provided inline with the volume.

There are two ways to declare the Secret with ephemeral inline volumes, either the Secret is in the same Namespace as the workload being declared or it resides in a foreign Namespace.

Local Secret:

apiVersion: v1
kind: Pod
metadata:
  name: my-pod-inline-mount-1
spec:
  containers:
    - name: pod-datelog-1
      image: nginx
      command: ["bin/sh"]
      args: ["-c", "while true; do date >> /data/mydata.txt; sleep 1; done"]
      volumeMounts:
        - name: my-volume-1
          mountPath: /data
  volumes:
    - name: my-volume-1
      csi:
       driver: csi.hpe.com
       nodePublishSecretRef:
         name: hpe-backend
       fsType: ext3
       volumeAttributes:
         csi.storage.k8s.io/ephemeral: "true"
         accessProtocol: "iscsi"
         size: "5Gi"

Foreign Secret:

apiVersion: v1
kind: Pod
metadata:
  name: my-pod-inline-mount-2
spec:
  containers:
    - name: pod-datelog-1
      image: nginx
      command: ["bin/sh"]
      args: ["-c", "while true; do date >> /data/mydata.txt; sleep 1; done"]
      volumeMounts:
        - name: my-volume-1
          mountPath: /data
  volumes:
    - name: my-volume-1
      csi:
       driver: csi.hpe.com
       fsType: ext3
       volumeAttributes:
         csi.storage.k8s.io/ephemeral: "true"
         inline-volume-secret-name: hpe-backend
         inline-volume-secret-namespace: kube-system
         accessProtocol: "iscsi"
         size: "7Gi"

The parameters used in the examples are the bare minimum required parameters. Any parameters supported by the HPE CSI Driver and backend CSP may be used for ephemeral inline volumes. See the base StorageClass parameters or the respective CSP being used.

Raw block volume¶

The default volumeMode for a PersistentVolumeClaim is Filesystem. If a raw block volume is desired, volumeMode needs to be set to Block. No filesystem will be created. Example:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-pvc-block
spec:
  accessModes:
  - ReadWriteOnce
  resources:
    requests:
      storage: 32Gi
  storageClassName: hpe-scod
  volumeMode: Block

Mapping the device in a Pod specification is slightly different than using regular filesystems as a volumeDevices section is added instead of a volumeMounts stanza:

apiVersion: v1
kind: Pod
metadata:
  name: my-pod-block
spec:
  containers:
    - name: my-null-pod
      image: fedora:31
      command: ["/bin/sh", "-c"]
      args: [ "tail -f /dev/null" ]
      volumeDevices:
        - name: data
          devicePath: /dev/xvda
  volumes:
    - name: data
      persistentVolumeClaim:
        claimName: my-pvc-block

Using CSI snapshots¶

CSI introduces snapshots as native objects in Kubernetes that allows end-users to provision VolumeSnapshot objects from an existing PersistentVolumeClaim. New PVCs may then be created using the snapshot as a source.

Start by creating a VolumeSnapshotClass referencing the Secret and defining additional snapshot parameters.

Kubernetes 1.17+ (CSI snapshots in beta)

apiVersion: snapshot.storage.k8s.io/v1beta1
kind: VolumeSnapshotClass
metadata:
  name: hpe-snapshot
  annotations:
    snapshot.storage.kubernetes.io/is-default-class: "true"
driver: csi.hpe.com
deletionPolicy: Delete
parameters:
  description: "Snapshot created by the HPE CSI Driver"
  csi.storage.k8s.io/snapshotter-secret-name: hpe-backend
  csi.storage.k8s.io/snapshotter-secret-namespace: kube-system

apiVersion: snapshot.storage.k8s.io/v1beta1
kind: VolumeSnapshotClass
metadata:
  name: hpe-snapshot
  annotations:
    snapshot.storage.kubernetes.io/is-default-class: "true"
driver: csi.hpe.com
deletionPolicy: Delete
parameters:
  description: "Snapshot created by the HPE CSI Driver"
  csi.storage.k8s.io/snapshotter-secret-name: hpe-backend
  csi.storage.k8s.io/snapshotter-secret-namespace: kube-system
  csi.storage.k8s.io/snapshotter-list-secret-name: hpe-backend
  csi.storage.k8s.io/snapshotter-list-secret-namespace: kube-system

Create a VolumeSnapshot. This will create a new snapshot of the volume.

apiVersion: snapshot.storage.k8s.io/v1beta1
kind: VolumeSnapshot
metadata:
  name: my-snapshot
spec:
  source:
    persistentVolumeClaimName: my-pvc

Check that a new VolumeSnapshot is created based on your claim:

kubectl describe volumesnapshot my-snapshot
Name:         my-snapshot
Namespace:    default
...
Status:
  Creation Time:  2019-05-22T15:51:28Z
  Ready:          true
  Restore Size:   32Gi

It's now possible to create a new PersistentVolumeClaim from the VolumeSnapshot.

---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-pvc-from-snapshot
spec:
  dataSource:
    name: my-snapshot
    kind: VolumeSnapshot
    apiGroup: snapshot.storage.k8s.io
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 32Gi

The .data.dataSource attribute may also clone PersistentVolumeClaim directly, without creating a VolumeSnapshot.

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-pvc-from-pvc
spec:
  dataSource:
    name: my-pvc
    kind: PersistentVolumeClaim
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 32Gi

Again, the size in .spec.resources.requests.storage must match the source PersistentVolumeClaim. This can get sticky from an automation perspective as volume expansion is being used on the source volume. It's recommended to inspect the source PersistentVolumeClaim or VolumeSnapshot size prior to creating a clone.

Expanding PVCs¶

To perform expansion operations on Kubernetes 1.14+, you must enhance your StorageClass with some additional attributes. Please see base StorageClass parameters.

Then, a volume provisioned by a StorageClass with expansion attributes may have its PersistentVolumeClaim expanded by altering the .spec.resources.requests.storage key of the PersistentVolumeClaim.

This may be done by the kubectl patch command.

kubectl patch pvc/my-pvc --patch '{"spec": {"resources": {"requests": {"storage": "64Gi"}}}}'
persistentvolumeclaim/my-pvc patched

The new PersistentVolumeClaim size may be observed with kubectl get pvc/my-pvc after a few moments.

Using PVC Overrides¶

The HPE CSI Driver allows the PersistentVolumeClaim to override the StorageClass parameters by annotating the PersistentVolumeClaim. Define the parameters allowed to be overridden in the StorageClass by setting the allowOverrides parameter:

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: hpe-scod-override
provisioner: csi.hpe.com
parameters:
  csi.storage.k8s.io/fstype: xfs
  csi.storage.k8s.io/provisioner-secret-name: hpe-backend
  csi.storage.k8s.io/provisioner-secret-namespace: kube-system
  csi.storage.k8s.io/controller-publish-secret-name: hpe-backend
  csi.storage.k8s.io/controller-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-stage-secret-name: hpe-backend
  csi.storage.k8s.io/node-stage-secret-namespace: kube-system
  csi.storage.k8s.io/node-publish-secret-name: hpe-backend
  csi.storage.k8s.io/node-publish-secret-namespace: kube-system
  description: "Volume provisioned by the HPE CSI Driver"
  accessProtocol: iscsi
  allowOverrides: description,accessProtocol

The end-user may now control those parameters (the StorageClass provides the default values).

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-pvc-override
  annotations:
    csi.hpe.com/description: "This is my custom description"
    csi.hpe.com/accessProtocol: fc
spec:
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 100Gi
  storageClassName: hpe-scod-override

Using volume mutations¶

The HPE CSI Driver (version 1.3.0 and later) allows the CSP backend volume to be mutated by annotating the PersistentVolumeClaim. Define the parameters allowed to be mutated in the StorageClass by setting the allowMutations parameter.

In the StorageClass, also make sure that the csi.storage.k8s.io/controller-provisioner-secret-name and csi.storage.k8s.io/controller-provisioner-secret-namespace are set, as those are used by the csi-extensions and csi-volume-mutator sidecars.

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: hpe-scod-mutation
provisioner: csi.hpe.com
parameters:
  csi.storage.k8s.io/fstype: xfs
  csi.storage.k8s.io/provisioner-secret-name: hpe-backend
  csi.storage.k8s.io/provisioner-secret-namespace: kube-system
  csi.storage.k8s.io/controller-publish-secret-name: hpe-backend
  csi.storage.k8s.io/controller-publish-secret-namespace: kube-system
  csi.storage.k8s.io/node-stage-secret-name: hpe-backend
  csi.storage.k8s.io/node-stage-secret-namespace: kube-system
  csi.storage.k8s.io/node-publish-secret-name: hpe-backend
  csi.storage.k8s.io/node-publish-secret-namespace: kube-system
  description: "Volume provisioned by the HPE CSI Driver"
  allowMutations: description

The end-user may now control those parameters by editing or patching the PersistentVolumeClaim.

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-pvc-mutation
  annotations:
    csi.hpe.com/description: "My description needs to change"
spec:
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 100Gi
  storageClassName: hpe-scod-mutation

Using the NFS Server Provisioner¶

Enabling the NFS Server Provisioner to allow RWX and ROX access mode for a PVC is straightforward. Create a new StorageClass and set .parameters.nfsResources to "true". Any subsequent claim to the StorageClass will create a NFS server Deployment on the cluster with the associated objects running on top of a RWO PVC.

Any RWO claim made against the StorageClass will also create a NFS server Deployment. This allows diverse connectivity options among the Kubernetes worker nodes as the HPE CSI Driver will look for nodes labelled csi.hpe.com/hpe-nfs=true before submitting the workload for scheduling. This allows dedicated NFS worker nodes without user workloads using taints and tolerations.

By default, the NFS Server Provisioner deploy resources in the "hpe-nfs" Namespace. This makes it easy to manage and diagnose. However, to use CSI data management capabilities on the PVCs, the NFS resources need to be deployed in the same Namespace as the RWX/ROX requesting PVC. This is controlled by the nfsNamespace StorageClass parameter. See base StorageClass parameters for more information.

Example use of accessModes:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-rwo-pvc
spec:
  accessModes:
  - ReadWriteOnce
  resources:
    requests:
      storage: 32Gi
  storageClassName: hpe-nfs

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-rwx-pvc
spec:
  accessModes:
  - ReadWriteMany
  resources:
    requests:
      storage: 32Gi
  storageClassName: hpe-nfs

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: my-rox-pvc
spec:
  accessModes:
  - ReadOnlyMany
  resources:
    requests:
      storage: 32Gi
  storageClassName: hpe-nfs

In the case of declaring a ROX PVC, the requesting Pod specification needs to request the PVC as read-only. Example:

apiVersion: v1
kind: Pod
metadata:
  name: pod-rox
spec:
  containers:
  - image: busybox
    name: busybox
    command:
      - "sleep"
      - "300"
    volumeMounts:
    - mountPath: /data
      name: my-vol
      readOnly: true
  volumes:
  - name: my-vol
    persistentVolumeClaim:
      claimName: my-rox-pvc
      readOnly: true

Requesting an empty read-only volume might not seem practical. The primary use case is to source existing datasets into immutable applications, using either a backend CSP cloning capability or CSI data management feature such as snapshots or existing PVCs.

Limitations and considerations for the NFS Server Provisioner¶

The current hardcoded limit for the NFS Server Provisioner is 20 NFS servers per Kubernetes worker node. The NFS server Deployment is currently setup in a completely unfettered resource mode where it will consume as much memory and CPU as it requests.

The two StorageClass parameters nfsResourceLimitsCpuM and nfsResourceLimitsMemoryMi control how much CPU and memory it may consume. Tests show that the NFS server consume about 150MiB at instantiation. These parameters will have defaults ready for GA.

The HPE CSI Driver now also incorporates a Pod Monitor to delete Pods that have become unavailable due to the Pod status changing to NodeLost or a node becoming unreachable that the Pod runs on. Be default the Pod Monitor only watches the NFS Server Provisioner Deployments. It may be used for any Deployment. See Pod Monitor on how to use it.

See diagnosing NFS Server Provisioner issues for further details.

Further reading¶

The official Kubernetes documentation contains comprehensive documentation on how to markup PersistentVolumeClaim and StorageClass API objects to tweak certain behaviors.

Each CSP has a set of unique StorageClass parameters that may be tweaked to accommodate a wide variety of use cases. Please see the documentation of the respective CSP for more details.