Compute SKU
During the infrastructure setup step, the Org Administrator would have navigated to the "Template Catalog" and cloned the compute and Jupyter notebook templates into this project. In this step, you will use the PaaS Studio to create and publish a compute profile (aka SKU) that will be made immediately available to end users for self-service consumption via the Developer Hub.
You will create a Computer Profiles called "small" that the end user can launch as a compute instance.
A very common approach used by organizations is to publish multiple flavors of compute SKUs. In our exercise, we will use the default resource sizes for three SKUs as shown below.
Note
It is possible to create SKUs without GPU for non-GPU use cases.
SKU | CPU | Memory | GPU |
---|---|---|---|
Small | 4 | 8Gb | 1 |
Medium | 8 | 16Gb | 2 |
Large | 16 | 32Gb | 4 |
As a convenience, the Rafay Platform provides three starter options for SKUs: small, medium and large. Admins can rename and customize these if necessary in the underlying Environment Template.
Info
The CPU and Memory resource specifications for the small SKU above is the minimum required for the Virtual Cluster based architecture. Resource allocations below this have been tested and will not work deterministically.
Progressive Templating¶
The PaaS Studio is the 2nd layer of progressive templating. In the first step as the Infra/Org Admin, when we cloned/created the template, we locked down some of the configuration. In this step, as a PaaS admin, you will progressively lock down a few more configurations to ensure that the end user has a simple, intuitive experience.
Small Compute Profile¶
In this step, let us start with the creation of a compute profile for a small SKU.
Navigate to the PaaS Studio with a PaaS Project Admin role (If you are logged in as an Org Admin with access to all projects, ensure you have selected the right project).
- Select Compute Profiles and click on New Compute Profile
General Section¶
In this section, you can provide general information about your Small SKU.
- Provide a unique name (e.g. small-gpu)
- Select the environment template and version for the compute instance that was shared to your project by the admin
- In the description field, enter something like "4 CPU, 8GB Memory, 1 GPU". This will be displayed to the end user when they select the SKU
- Click on Save
Once the profile has been created, we are ready to configure it with details that map to our table from above.
Design Profile Card¶
In this section, you can specify icons and a readme that will be displayed to the end user as additional information about the SKU. We will skip this section for our exercise.
Inputs & Constraints¶
The compute template from the catalog comes with a default set of commonly used input variables. The PaaS Studio administrator now has to make a decision on which of these should be made available as inputs for the end user.
- Navigate to the Input Settings configuration section.
- Disable the checkboxes for the "Allow Override"
Info
In our example, there is no input we wish to expose to the end user. Once they select the "Small" SKU, we wish to abstract away everything else from them
- For "Namespace Quota Size", ensure "small" is selected
- For "Host Cluster Name", enter the name of the underlying host cluster that has been provisioned in the prior step (e.g. mohan-lab-cluster)
- Click on Save.
You should now have a published compute profile called "small" like the image below.
Output Settings¶
This section allows the administrator to configure and specify what outputs should be presented to the user once the SKU is successfully fulfilled. In our example, we will use the default and not configure this section.
Verify Compute Profile¶
Info
This is a completely optional step.
Admins may wish to validate what the compute profile will look like for the end user. They can do this by following the steps below:
- click on Compute Instances
- Click on New Compute Instance
- You should now see the newly created "small" SKU as an option.
Note that this is similar to what the end users will be shown when they access the Self-Service Developer Hub Portal.