#CodeFlare SDK Tutorial

This tutorial demonstrates how to use the CodeFlare SDK to submit RayJobs against an existing RayCluster. You will learn how to spin up a Ray cluster, verify its status, submit a RayJob, and manage the cluster lifecycle.

#Prerequisites

• You have installed the Alauda Build of KubeRay Operator cluster plugin in your data science cluster, see Install Alauda Build of KubeRay Operator.
• You can access a namespace in Alauda AI, create a workbench, and the workbench is running a default workbench image that contains the CodeFlare SDK, for example, the Standard Data Science notebook. For information about creating workbenches, see Create Workbench.
• You have logged in to Alauda AI, started your workbench, and logged in to JupyterLab.

#Demo Notebook

Download the demo Jupyter notebook to follow along with this tutorial:

• Notebook: CodeFlare SDK RayJob Demo

Click the upward arrow button on the JupyterLab page to upload the downloaded notebook file.

#Procedure

Open the demo notebook in JupyterLab and follow the steps below. Each step corresponds to a section in the notebook.

#Step 1: Create the RayCluster

Run the first two code cells to import the CodeFlare SDK and create a Ray cluster with the ClusterConfiguration API. Before running, update the image parameter to a Ray cluster image that is compatible with your hardware architecture. If your cluster has no direct internet access, use an image from your internal registry.

The cluster.apply() call submits the cluster configuration and waits for it to be ready. You can adjust the timeout value as needed.

#Step 2: Verify the cluster status

Run the cluster.status() cell. If the cluster is not up immediately, run the cell a few more times until you see that it is in a Ready state.

#Step 3: Submit a RayJob

Run the RayJob cell to create and submit a job against the running cluster. Note the following parameters:

• job_name: A unique name for your RayJob.
• cluster_name: Must match the name of your existing RayCluster.
• entrypoint: The command to execute. In standard practice, this would point to a Python training script rather than the inline command used in the demo.

#Step 4: Monitor the RayJob

Run the rayjob.status() cell. This function outputs different tables based on the RayJob's current status. You can re-run the cell multiple times to observe the changes.

#Step 5: Clean up

Once the job has completed, run cluster.down() to shut down the cluster.