To run this workshop, you need an AWS account, and a user identity with access to the following services:
You can use your own account, or an account provided through Event Engine as part of an AWS organized workshop. Using an account provided by Event Engine is the easier path, as you will have full access to all AWS services, and the account will terminate automatically when the event is over.
You should also have familiarity with using the AWS CLI, including configuring the CLI for a specific account and region profile. If not, please follow the CLI setup instructions. Make sure you have a default profile set up.
Finally, you'll need to have a kubernetes cluster set up with Kubeflow installed, and a local copy of the KFP SDK.
If you are running this workshop as part of an Event Engine lab, please log into the console using this link and enter the hash provided to you as part of the workshop.
If you are using your own AWS account, be sure you have access to create S3 buckets, run SageMaker training jobs, create SageMaker endpoints, and optionally create and manage EKS clusters.
After completing the workshop, remember to complete the cleanup section to remove any unnecessary AWS resources.
After you have your account identified, make sure to work in the
Also note your AWS account number. You find this in the console or by running
aws sts get-caller-identity on the CLI. We'll refer to this as
ACCOUNT going forward.
Whichever AWS account option you use, you may prefer to use a cloud-based environment rather than installing tools like
eksctl on your own machine. In that case, you can use a Cloud 9 IDE. See the getting started guide for how to set up your own Cloud 9 IDE.
If you use Cloud 9, you'll need to make a few changes to your IDE to let you run the labs from the environment.
AWS servicefor the entity and leave the service set to
AWS managed temporary credentials.
Note that this role grants a very broad set of permissions to your Cloud9 instance.
In your Cloud9 IDE, create the file
~/.aws/config with these lines, replacing
ACCOUNT with your own settings:
[default] region=us-east-1 account=ACCOUNT
You'll need a kubernetes (k8s) cluster with Kubeflow installed.
If you need to make a k8s cluster, you can create an EKS cluster using eksctl or by following one of the other Quick Start guides. For the purposes of this workshop, a small cluster with three
m5.xlarge nodes should be sufficient.
If you do not yet have Kubeflow installed, follow the instructions to install it on your k8s cluster.
Once your cluster is set up, follow the
IAM Permissions instructions to grant the Kubeflow Components permission to access AWS services on your behalf. Only follow the instructions under the
IAM Permissions section on this page; we will handle the other steps later in the workshop.
Install the KFP SDK.
If you want an easier way to deploy an EKS cluster with Kubeflow, you can use this CloudFormation template.