Skip to content

Launch

End users (e.g. data scientists and ML researchers) will use a self-service workflow in the Developer Hub to deploy Ray as a Service into a vCluster based compute instance. A successful deployment of Ray to a vcluster based compute instance can take ~5 minutes. Once deployed, a dedicated Ray as Service tenant is created on the Host cluster for the end user. The diagram below describes the sequence of steps that are automatically performed to set this up.

Onboarding Experience in Developer Hub

Note

For testing and evaluation purposes, by default, Rafay automatically programs DNS for the Ray Endpoint URL and configures it for secure access using a trusted digital certificate. For production deployments, administrators should specify both DNS and Certificate details.

Add User to Developer Hub

In order for the end user to access the Developer Hub, Org Admins need to update the user's role with the required privileges

  • Navigate to System -> Users
  • Click on the User and click on the Projects tab
  • Click on "Assign", select the "project" that you would like this user to use
  • Select one of the following roles: "Developer", "Data Scientist", "MLOps Engineer" or "GenAI Developer".
  • Save

Note that all these roles are identical. So, select the name that suits the user's role the best.

Info

Instead of managing user roles one at a time, it is more practical and efficient to assign users to a custom group which has the correct role assigned to it.

End User Self Service

  • Access the Rafay Console and login using your credentials
  • For users with the correct role, they will now have access to the Developer Hub. This portal will present all the self service SKUs published by the administrator.

Step 1: Compute Instance

The Ray Service will be deployed on an already provisioned "compute instance". Skip this step if you already have a well resourced compute instance.

  • Click on Compute Instances
  • Select a suitable SKU and deploy

It can take ~5 minutes for all the compute instance to be successfully deployed and operational.

Step 2: Ray Service Instance

  • Click on Custom Services
  • Select the Ray SKU and specify the compute instance from the step above as the target

Ray SKU

  • Now, click on Deploy

This step can take a few minutes to complete. Once successfully deployed, the end user can access their web-based Ray Dashboard by clicking on the "Ray Endpoint URL". They need to provide "Access Credentials before they can access the dashboard.

Ray Tenant Endpoint