🧱 MVP: Build SageMaker in Development

[Gary-H9]

User Story

As a member of the team taking part in the SageMaker spike.
I want to build a MVP of the SageMaker product.
So that we can properly assess the product.

Value / Purpose

Informs the spike on this piece of work.

Useful Contacts

Gary + Yikang

Hypothesis

If we build the simplest version of SageMaker possible...
Then... we can better understand the project.

Proposal

Build the simplest, probably most open version of SageMaker in the analytical-platform-development AWS account. Do this in the data-platform repository.

Out of Scope

• Endpoints.
• RStudio

Definition of Done

• Follow this setup guide. Semi-irrelevant as some of the setup is covered by the terraform module.
• Setup a SageMaker domain.
• Setup a Share
• POC for integration between Control panel and Sagemaker

[Gary-H9]

I followed the aforementioned setup guide and there were AWS components mentioned, EFS, S3, etc. These are handled by the module and do not need to be declared in the code.

Initially I wanted to setup the solution using SSO. However, if this solution is adopted it will liaise with control panel to manipulate user permissions. As such the auth_mode we will use will be IAM.

I've also updated Yikang's Google doc to include an explainer on what an AWS SageMaker Domain is, this can be pulled into more formal documentation later.

[Gary-H9]

Enabled user ability to spawn spaces with the following in sagemaker.tf:

  default_space_settings {
    execution_role = aws_iam_role.sagemaker_studio_execution_role.arn
  }

[Gary-H9]

RStudio is now out of scope of the MVP as "RStudio on Amazon SageMaker is a paid product and requires that each user is appropriately licensed." (source)

[Gary-H9]

Setup a Share manually (shared-space-1), these should in theory be setup by users so I didn't want to Terraform this.

[Gary-H9]

Further details regarding price are included in the #data-platform-sagemaker Slack channel.

[ymao2]

linked the my for integration part to this ticket

[Gary-H9]

Completed and branch merged.