Create and connect to a Parallelstore instance from Google Kubernetes Engine

Parallelstore is available by invitation only. If you'd like to request access to Parallelstore in your Google Cloud project, contact your sales representative.

This guide describes how you can create a new Kubernetes volume backed by a Parallelstore CSI driver in GKE with dynamic provisioning. This lets you create storage backed by fully managed Parallelstore instances on-demand, and access them as volumes for your stateful workloads.

Before you begin

Before you start, make sure you have performed the following tasks:

• Enable the Parallelstore API and the Google Kubernetes Engine API.
Enable APIs
• If you want to use the Google Cloud CLI for this task, install and then initialize the gcloud CLI. If you previously installed the gcloud CLI, get the latest version by running gcloud components update.
  Note: For existing gcloud CLI installations, make sure to set the compute/region and compute/zone properties. By setting default locations, you can avoid errors in gcloud CLI like the following: One of [--zone, --region] must be supplied: Please specify location.

• See the CSI driver overview for limitations and requirements.
• Configure a VPC network.
• If you want to use a GKE Standard cluster, make sure to enable the CSI driver.

Create a new volume using the Parallelstore CSI driver

The following sections describe the typical process for creating a Kubernetes volume backed by a Parallelstore CSI driver in GKE:

1. (Optional) Create a StorageClass.
2. Use a PersistentVolumeClaim to access the volume.
3. (Optional) Configure resources for the sidecar container.
4. Create a workload that consumes the volume.

(Optional) Create a storage class

When the Parallelstore CSI driver is enabled, GKE automatically creates a StorageClass named parallelstore-rwx for provisioning Parallelstore instances. This StorageClass directs the CSI driver to provision Parallelstore instances in the same region as your GKE cluster to ensure optimal I/O performance.

Optionally, you can create a custom StorageClass with a specific topology. To do so, follow these steps:

1. Save the following StorageClass manifest in a file named parallelstore-class.yaml:

     apiVersion: storage.k8s.io/v1
     kind: StorageClass
     metadata:
       name: parallelstore-class
     provisioner: parallelstore.csi.storage.gke.io
     volumeBindingMode: Immediate
     reclaimPolicy: Delete
     allowedTopologies:
     - matchLabelExpressions:
       - key: topology.gke.io/zone
         values:
           LOCATION
   

   Replace the following:

   • LOCATION: the Compute Engine zone containing the cluster. You must specify a supported zone for the Parallelstore CSI driver.

   For the full list of fields supported in the StorageClass, refer to the Parallelstore CSI reference documentation.

2. Create the StorageClass by running this command:

   kubectl create -f parallelstore-class.yaml
   

Use a PersistentVolumeClaim to access the volume

You can create a PersistentVolumeClaim resource that references the Parallelstore CSI driver's StorageClass.

The following manifest file shows an example of how to create a PersistentVolumeClaim in ReadWriteMany access mode that references the StorageClass you created earlier.

1. Save the following manifest in a file named parallelstore-pvc.yaml:

     apiVersion: v1
     kind: PersistentVolumeClaim
     metadata:
       name: parallelstore-pvc
     spec:
       accessModes:
       - ReadWriteMany
       resources:
         requests:
           storage: STORAGE_SIZE
       storageClassName: parallelstore-class
   

   Replace STORAGE_SIZE with the storage size; for example, 12000Gi. The value must be in the range from 12,000 GiB to 100,000 GiB (in multiples of 4,000 GiB).

2. Create the PersistentVolumeClaim by running this command:

     kubectl create -f parallelstore-pvc.yaml
   

(Optional) Configure resources for the sidecar container

When you create a workload Pod that uses Parallelstore-backed volumes, the CSI driver determines whether your volume is based on Parallelstore instances.

If the driver detects that your volume is Parallelstore-based, or if you specify the annotation gke-parallelstore/volumes: "true", the CSI driver automatically injects a sidecar container named gke-parallelstore-sidecar into your Pod. This sidecar container mounts the Parallelstore instance to your workload.

Caution: Before deploying a Pod that consumes the PersistentVolumeClaim, ensure that either the parameter gke-parallelstore/volumes: "true" is specified in the Pod annotation, or use kubectl get pvc to verify that GKE created the PersistentVolume and PersistentVolumeClaim. Failure to do so may result in sidecar injection failures.

By default, GKE configures the sidecar container with the following resource requests, with resource limits unset:

• 250 m CPU
• 512 MiB memory
• 10 MiB ephemeral storage

To overwrite these values, you can optionally specify the annotation gke-parallelstore/[cpu-request|memory-request|cpu-limit|memory-limit|ephemeral-storage-request] as shown in the following example:

apiVersion: v1
kind: Pod
metadata:
  annotations:
    gke-parallelstore/volumes: "true"
    gke-parallelstore/cpu-request: 500m
    gke-parallelstore/memory-request: 1Gi
    gke-parallelstore/ephemeral-storage-request: 500Mi
    gke-parallelstore/cpu-limit: 1000m
    gke-parallelstore/memory-limit: 2Gi
    gke-parallelstore/ephemeral-storage-limit: 1Gi

Use the following considerations when deciding the amount of resources to allocate:

• If one of the request or limit values is set and another is unset, they will both be set to the same, specified value.
• Allocate more CPU to the sidecar container if your workloads need higher throughput. Insufficient CPU will cause I/O throttling.
• You can use value "0" to unset any resource limits on Standard clusters; for example, gke-parallelstore/memory-limit: "0"removes the memory limit for the sidecar container. This is useful when you cannot decide on the amount of resources gke-parallelstore-sidecar needs for your workloads, and want to let the sidecar consume all the available resources on a node.

Create a workload that consumes the volume

This section shows an example of how to create a Pod that consumes the PersistentVolumeClaim resource you created earlier.

Multiple Pods can share the same PersistentVolumeClaim resource.

1. Save the following manifest in a file named my-pod.yaml.

     apiVersion: v1
     kind: Pod
     metadata:
       name: my-pod
     spec:
       containers:
       - name: nginx
         image: nginx
         volumeMounts:
           - name: parallelstore-volume
             mountPath: /data
       volumes:
       - name: parallelstore-volume
         persistentVolumeClaim:
           claimName: parallelstore-pvc
   

2. Run the following command to apply the manifest to the cluster.

     kubectl apply -f my-pod.yaml
   

   The Pod will wait until the PersistentVolumeClaim is provisioned before it starts running. This operation might take several minutes to complete.

Manage the Parallelstore CSI driver

This section covers how you can enable and disable the Parallelstore CSI driver, if needed.

Enable the Parallelstore CSI driver on a new cluster

Note: The Parallelstore CSI driver is enabled by default for Autopilot clusters. You don't need to enable the driver when you create a new Autopilot cluster.

To enable the Parallelstore CSI driver when creating a new Standard cluster, run the following command with the Google Cloud CLI:

gcloud container clusters create CLUSTER_NAME \
    --location=LOCATION \
    --network=NETWORK_NAME \
    --addons=ParallelstoreCsiDriver \
    --cluster-version=VERSION

Replace the following:

• CLUSTER_NAME: the name of your cluster.
• LOCATION: the Compute Engine zone containing the cluster. You must specify a supported zone for the Parallelstore CSI driver.
• NETWORK_NAME: name of the VPC network you created in Configure a VPC network.
• VERSION: the GKE version number. You must specify a supported version number to use this feature, such as GKE version 1.29 or later. Alternatively, you can use the --release-channel flag and specify a release channel.

Enable the Parallelstore CSI driver on an existing cluster

To enable the driver on an existing GKE Standard cluster, run the following command with the Google Cloud CLI:

gcloud container clusters update CLUSTER_NAME \
  --location=LOCATION \
  --update-addons=ParallelstoreCsiDriver=ENABLED

Replace the following:

• CLUSTER_NAME : the name of your cluster.
• LOCATION: the Compute Engine zone containing the cluster. You must specify a supported zone for the Parallelstore CSI driver.

Make sure that your GKE cluster is running in the same VPC network that you set up in Configure a VPC network. To verify the VPC network for a GKE cluster, you can check in the Google Cloud console, or through the command gcloud container clusters describe $(CLUSTER) --format="value(networkConfig.network)" --location=$(LOCATION).

Disable the Parallelstore CSI driver

You can disable the Parallelstore CSI driver on an existing Autopilot or Standard cluster by using the Google Cloud CLI.

Warning: We strongly recommend not to manually disable the Parallelstore CSI driver on Autopilot clusters. Doing so causes any Pods using PersistentVolumes owned by the driver to fail to terminate. New Pods attempting to use those PersistentVolumes will also fail to start.

gcloud container clusters update CLUSTER_NAME \
    --location=LOCATION \
    --update-addons=ParallelstoreCsiDriver=DISABLED

Replace the following:

• CLUSTER_NAME : the name of your cluster.
• LOCATION: the Compute Engine zone containing the cluster. You must specify a supported zone for the Parallelstore CSI driver.

Use fsGroup with Parallelstore volumes

The Parallelstore CSI driver supports changing the group ownership of the root level directory of the mounted file system to match a user-requested fsGroup specified in the Pod's SecurityContext. This feature is only supported in GKE clusters version 1.29.5 or later, or version 1.30.1 or later.

Troubleshooting

For troubleshooting guidance, refer to the Troubleshooting page in the Parallelstore documentation.

What's next

• Explore the Parallelstore CSI reference documentation.
• Learn how to use the Parallelstore interception library to improve workload performance.
• Learn how to transfer data to Parallelstore from Cloud Storage.
• Try the tutorial to train a TensorFlow model with Keras on GKE.