CSI driver reference

The following manifests describe supported CSI driver fields in the Storage Class and Persistent Volume resources.

Storage Class

The following manifest includes all of the supported fields in the Storage Class.

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: parallelstore-csi-sc
provisioner: parallelstore.csi.storage.gke.io
parameters:
  network: VPC_NETWORK_NAME # optional
  labels: KEY: VALUE # optional
  description: DESCRIPTION # optional
  reservedIpRange: IP_RANGE_NAME # optional
volumeBindingMode: VOLUME_BINDING_MODE # default is "immediate"; or "waitForFirstConsumer"
allowedTopologies: # optional
- matchLabelExpressions:
  - key: topology.gke.io/zone
    values:
    - ZONE
mountOptions:
  -  DFUSE_MOUNT_OPTIONS # enable-caching, enable-wb-cache, thread-count, eq-count

All of the following fields are optional:

  • network: The VPC network in which to create the Parallelstore instance. If this property is not specified, the network of the GKE cluster is used.

    To create a Parallelstore instance in a shared VPC network, the full name of the network must be provided. For example: projects/PROJECT_ID/global/networks/VPC_NETWORK_NAME

  • labels: User-specified key-value pairs to attach to this instance.

  • description: A description of the instance. Must be 2048 characters or less.

  • reservedIpRange: The name of an allocated IP address range. If unspecified, all ranges are considered.

  • allowedTopologies: Specifies the zone in which to create the Parallelstore instance. If this field is not specified, the instance is created in the same zone as the CSI driver deployment, with immediate volume binding mode.

  • volumeBindingMode: Specifies the topology selection mode. The immediate mode uses the value specified in allowedTopologies. The waitForFirstConsumer mode assigns the topology to the zone in which the GKE pod is scheduled. This can cause pods to be blocked while waiting for a volume to be provisioned.

  • mountOptions: Specifies dfuse mount options. Parallelstore instance-backed volumes inherit these values in the .spec.mountOptionsfield of the persistent volume.

    Options are:

    • disable-caching: Disables all caching. By default, full caching (except for write-back cache) is enabled when mounting a Parallelstore instance in GKE.
    • enable-wb-cache: Use write-back cache rather than write-through.
    • thread-count: Number of threads to use. This value is the sum of the number of FUSE threads and the number of event queues.
    • eq-count: Number of event queues to use.

    For information about these options, see Performance considerations.

Persistent Volume

The following manifest includes all of the supported fields in the Persistent Volume.

apiVersion: v1
kind: PersistentVolume
metadata:
  name: parallelstore-pv
  annotations:
    pv.kubernetes.io/provisioned-by: parallelstore.csi.storage.gke.io
spec:
  storageClassName: parallelstore-csi-sc
  capacity:
    storage: INSTANCE_SIZE # 12Ti
  accessModes:
    - ReadWriteMany
  persistentVolumeReclaimPolicy: Retain # or "Delete"
  volumeMode: Filesystem
  mountOptions:
    - DFUSE_MOUNT_OPTIONS # enable-caching, enable-wb-cache, thread-count, eq-count
  csi:
    driver: parallelstore.csi.storage.gke.io
    volumeHandle: PROJECT_ID/LOCATION/INSTANCE_NAME/default-pool/default-container
    volumeAttributes:
      accessPoints: ACCESS_POINTS # comma-separated list of IP addresses
      network: VPC_NETWORK # optional
  • capacity: The capacity of the Parallelstore instance. Must be one of: 12Ti, 16Ti, or 20Ti.
  • mountOptions: Inherit from the mountOptions field in StorageClass.
  • csi.volumeHandle: An exclusive identifier for a Parallelstore instance. Must use the format: PROJECT_ID/LOCATION/INSTANCE_NAME/default-pool/default-container
  • csi.volumeAttributes.accessPoints: A list of IP addresses for Parallelstore server nodes, separated by commas.
  • csi.volumeAttributes.network: The VPC network of the Parallelstore instance, which must align with the network of the GKE cluster. This field is used to verify that the provided Parallelstore instance is in a valid network before the CSI Driver mounts it.