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  reserved-ip-range: IP_RANGE_NAME # optional  file-stripe-level: FILE_STRIPE_LEVEL # optional  directory-stripe-level: DIRECTORY_STRIPE_LEVEL # optional  mountLocality: MOUNT_LOCALITY # optional  dfuseCPURequest: DFUSE_CPU_REQUEST # optional  dfuseMemoryRequest: DFUSE_MEMORY_REQUEST # optional  dfuseCPULimit: DFUSE_CPU_LIMIT # optional  dfuseMemoryLimit: DFUSE_MEMORY_LIMIT # optional volumeBindingMode: VOLUME_BINDING_MODE # default is "immediate"; or "waitForFirstConsumer" allowedTopologies: # optional - matchLabelExpressions:  - key: topology.gke.io/zone  values:  - ZONE mountOptions:  - DFUSE_MOUNT_OPTIONS # disable-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.

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

  • file-stripe-level defines file striping settings. Allowed values are:

    • file-stripe-level-balanced
    • file-stripe-level-max
    • file-stripe-level-min

    See Performance considerations for details.

  • directory-stripe-level defines the striping level for directories. Allowed values are:

    • directory-stripe-level-balanced
    • directory-stripe-level-max
    • directory-stripe-level-min

    See Performance considerations for details.

  • mountLocality: Enables node-local mount mode for dynamic provisioning. Supported values are node or pod. The default is pod. If node is specified, you can also use the following options:

    • dfuseCPURequest: The CPU request for the dfuse process. The default is 250m.
    • dfuseMemoryRequest: The memory request for the dfuse process. The default is 512Mi.
    • dfuseCPULimit: The CPU limit for the dfuse process. The default is unset.
    • dfuseMemoryLimit: The memory limit for the dfuse process. The default is 10Gi.

  • 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 and kernel parameters, including read_ahead_kb and max_ratio. 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.
    • read_ahead_kb: Size of the read-ahead window in kilobytes.
    • max_ratio: Allows limiting a particular device to use not more than the given percentage of the write-back cache.

    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 # disable-caching, enable-wb-cache, thread-count, eq-count, read_ahead_kb, max_ratio  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  mountLocality: MOUNT_LOCALITY # optional  dfuseCPURequest: DFUSE_CPU_REQUEST # optional  dfuseMemoryRequest: DFUSE_MEMORY_REQUEST # optional  dfuseCPULimit: DFUSE_CPU_LIMIT # optional  dfuseMemoryLimit: DFUSE_MEMORY_LIMIT # 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.

  • csi.volumeAttributes.mountLocality: Enables node-local mount mode for dynamic provisioning. The supported values are node or pod. The default is pod. If node is specified, you can also use the following options:

    • csi.volumeAttributes.dfuseCPURequest: The CPU request for the dfuse process. The default is 250m.
    • csi.volumeAttributes.dfuseMemoryRequest: The memory request for the dfuse process. The default is 512Mi.
    • csi.volumeAttributes.dfuseCPULimit: The CPU limit for the dfuse process. The default is unset.
    • csi.volumeAttributes.dfuseMemoryLimit: The memory limit for the dfuse process. The default is 10Gi.