Skip to main content

Provisioning GCP Lustre

Overview​

This guide explains how to provision a managed GCP Lustre instance.

Prerequisites​

  • Create a namespace for installation.
export NAMESPACE=llm-d-storage # or any other namespace (shorter names recommended)
kubectl create namespace ${NAMESPACE}

Cluster setup for provisioning a managed GCP Lustre instance (GKE cluster)​

To create a new GKE cluster, you need to first set up a separate VPC for provisioning a managed GCP Lustre instance. Follow the steps from Lustre guide to setup a separate VPC network.

If you have an existing cluster, you can update the same VPC network to enable provisioning a managed GCP Lustre instance. Ensure you initialize the variable NETWORK_NAME with your existing network name and skip the network creation command in the setup above.

Please update the $NETWORK_NAME and $LOCATION variables in storage_class.yaml to match your cluster configuration.

Ensure Lustre CSI driver is enabled on the cluster, otherwise it would fail to provision a managed GCP Lustre instance.

Provisioning​

GKE​

1. Create a StorageClass for Lustre:

kubectl apply -f guides/tiered-prefix-cache/storage/manifests/backends/lustre/storage_class.yaml  -n ${NAMESPACE}

2. Create a PVC:

Once the StorageClass is created, you can create a PVC and mount to your pod like so:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: lustre-pvc
spec:
  accessModes:
  - ReadWriteMany
  resources:
    requests:
      storage: 18000Gi # Range from 18000Gi to 954000Gi, must be in multiples of 9000 Gi
  storageClassName: lustre-class

Check the PVC​

kubectl get pvc -n ${NAMESPACE}

Output should show the PVC as Bound:

NAME         STATUS   VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS   VOLUMEATTRIBUTESCLASS   AGE
lustre-pvc   Bound    pvc-3c793698-XXXXXXX   36000Gi    RWX            lustre-class   <unset>                 6d

Cleanup​

kubectl delete -f ./storage_class.yaml -n ${NAMESPACE}