Skip to content

Part 4: Workload

What Will You Do

In this part of the self-paced exercise, you will deploy a "GPU workload" to your AKS cluster that has a GPU node.

Step 1: Namespace

In a typical production environment, administrators will have already created a "Kubernetes Namespace" for your workload. In this exercise, let us go ahead and create a namespace.

  • Open Terminal (on macOS/Linux) or Command Prompt (Windows) and navigate to the folder where you forked the Git repository
  • Navigate to the folder "/getstarted/gpuaks/workload"
  • Type the command
rctl create ns -f namespace.yaml

This step creates a namespace in your project. The controller can create a namespace on "multiple clusters" based on its placement policy. 

rctl publish ns gputest

Verify

To verify that then namespace was successfully created on your AKS cluster

rctl get ns gputest

You should see results like the following. Note that the namespace was successfully created on your AKS cluster. 

+---------+-------------+------------------------------+------------------+---------+--------+-------------+
| NAME    | TYPE        | CREATEDAT                    | DEPLOYEDCLUSTERS | ID      | LABELS | ANNOTATIONS |
+---------+-------------+------------------------------+------------------+---------+--------+-------------+
| gputest | RafayWizard | Fri Jul 14 17:26:19 UTC 2023 | demo-gpu-aks     | 27pd4p2 | []     | []          |
+---------+-------------+------------------------------+------------------+---------+--------+-------------+

Step 2: Deploy Workload

The "gputest.yaml" file contains the declarative specification for our GPU Workload. Let us review it. 

name: gputest
namespace: gputest
project: default
type: NativeYaml
clusters: demo-gpu-aks
payload: ./gpu-job.yaml

Note that the workload's name is "gputest" and it is of type "k8s YAML". The actual k8s YAML file is in the payload "gpu-job.yaml". 

rctl create workload gputest.yaml

If there were no errors, you should see a message like below

Workload created successfully

Now, let us publish the newly created workload to the downstream clusters. The workload can be deployed to multiple clusters as per the configured "placement policy". In this case, you are deploying to a single AKS cluster with the name "demo-gpu-aks". 

rctl publish workload gputest

Step 4: Verify

In the web console, click on Applications -> Workloads. You should see something like the following.

Published Workload

Step 5: GPU Dashboard

The GPU workload you deployed will consume the GPU attached to the AKS cluster. Administrators that wish to view GPU metrics have access to an "integrated GPU dashboard"

  • In your project, navigate to Infrastructure -> Clusters
  • Click on GPU in the cluster card to open the GPU selector

GPU Selector

  • Click Go to GPU to view its dashboard. You should see something like the following

GPU Dashboard

Recap

Congratulations! At this point, you have successfuly

  • Deployed a "GPU Workload" to the AKS Cluster and reviewed the integrated "GPU Dashboards"