If cGPU Basic Edition is installed in a Container Service for Kubernetes (ACK) dedicated cluster, cGPU cannot work as normal after you migrate the cluster workloads to an ACK Pro cluster. ACK Pro clusters support only cGPU Professional Edition. In this case, you must upgrade cGPU Basic Edition to cGPU Professional Edition in the ACK Pro cluster after the migration is completed. This topic describes how to upgrade cGPU Basic Edition to cGPU Professional Edition in an ACK Pro cluster.

Prerequisites

Workloads are migrated from an ACK dedicated cluster to an ACK Pro cluster. cGPU Basic Edition is installed in the ACK dedicated cluster before the migration. For more information, see Hot migration from ACK dedicated clusters to ACK Pro clusters.

Procedure

  1. Connect to ACK clusters by using kubectl.
  2. Download the Job YAML file that is used to change the node label and uninstall the original cGPU component. To download the YAML file, click gpushare-label-change.yaml.
  3. Run the following command to deploy the Job that runs the cGPU component:
    kubectl apply -f gpushare-label-change.yaml
  4. Run the following command to check whether the Job is deployed:
    kubectl get po -l app=change-gpushare-labels -n kube-system

    Expected output:

    NAME                             READY   STATUS      RESTARTS   AGE
    gpushare-label-migration-v****   0/1     Completed   0          89s
    The output shows that the Job is in the Completed state, which indicates that the Job is deployed.
  5. Install cGPU Professional Edition. For more information, see Install the cGPU component.
  6. Install a GPU memory inspection tool in the cluster. For more information, see Install and use the GPU memory inspection tool.

What to do next

For more information about how to test the GPU sharing, GPU scheduling, and GPU memory isolation features of cGPU Professional Edition, see Enable GPU sharing.