Quickstart: Create a SageMaker sandbox domain to launch Amazon EMR clusters in Studio - Amazon SageMaker
Step 1: Create your test environmentStep 2: Launch a new Amazon EMR clusterStep 3: Connect a notebook to your clusterStep 4: Clean up your resources

Quickstart: Create a SageMaker sandbox domain to launch Amazon EMR clusters in Studio

This section walks you through the quick set up of a complete test environment in Amazon SageMaker Studio. You will be creating a new Studio domain that lets users launch new Amazon EMR clusters directly from Studio. The steps provide an example notebook that you can connect to an Amazon EMR cluster to start running Spark workloads. Using this notebook, you will build a Retrieval Augmented Generation System (RAG) using Amazon EMR Spark distributed processing and OpenSearch vector database.

Note

To get started, sign in to the AWS Management Console using an AWS Identity and Access Management (IAM) user account with admin permissions. For information on how to sign up for an AWS account and create a user with administrative access, see Complete Amazon SageMaker prerequisites.

Step 1: Create a SageMaker domain for launching Amazon EMR clusters in Studio

In the following steps, you apply a AWS CloudFormation stack to automatically create a new SageMaker domain. The stack also creates a user profile and configures the needed environment and permissions. The SageMaker domain is configured to let you directly launch Amazon EMR clusters from Studio. For this example, the Amazon EMR clusters are created in the same AWS account as SageMaker without authentication. You can find additional AWS CloudFormation stacks supporting various authentication methods like Kerberos in the getting_started GitHub repository.

Note

SageMaker allows 5 Studio domains per AWS account and AWS Region by default. Ensure your account has no more than 4 domains in your region before you create your stack.

Follow these steps to set up a SageMaker domain for launching Amazon EMR clusters from Studio.

  1. Download the raw file of this AWS CloudFormation template from the sagemaker-studio-emr GitHub repository.

  2. Go to the AWS CloudFormation console: https://console.aws.amazon.com/cloudformation

  3. Choose Create stack and select With new resources (standard) from the drop down menu.

  4. In Step 1:

    1. In the Prepare template section, select Choose an existing template.

    2. In the Specify template section, choose Upload a template file.

    3. Upload the downloaded AWS CloudFormation template and choose Next.

  5. In Step 2, enter a Stack name and a SageMakerDomainName then choose Next.

  6. In Step 3, keep all default values and choose Next.

  7. In Step 4, check the box to acknowledge resource creation and choose Create stack. This creates a Studio domain in your account and region.

Step 2: Launch a new Amazon EMR cluster from Studio UI

In the following steps, you create a new Amazon EMR cluster from the Studio UI.

  1. Go to the SageMaker console at https://console.aws.amazon.com/sagemaker/ and choose Domains in the left menu.

  2. Click on your domain name GenerativeAIDomain to open the Domain details page.

  3. Launch Studio from the user profile genai-user.

  4. In the left navigation pane, go to Data then Amazon EMR Clusters.

  5. On the Amazon EMR clusters page, choose Create. Select the template SageMaker Studio Domain No Auth EMR created by the AWS CloudFormation stack and then choose Next.

  6. Enter a name for the new Amazon EMR cluster. Optionally update other parameters such as the instance type of core and master nodes, idle timeout, or number of core nodes.

  7. Choose Create resource to launch the new Amazon EMR cluster.

    After creating the Amazon EMR cluster, follow the status on the EMR Clusters page. When the status changes to Running/Waiting, your Amazon EMR cluster is ready to use in Studio.

Step 3: Connect a JupyterLab notebook to the Amazon EMR cluster

In the following steps, you connect a notebook in JupyterLab to your running Amazon EMR cluster. For this example, you import a notebook allowing you to build a Retrieval Augmented Generation (RAG) system using Amazon EMR Spark distributed processing and OpenSearch vector database.

  1. Launch JupyterLab

    From Studio, launch the JupyterLab application.

  2. Create a private space

    If you have not created a space for your JupyterLab application, choose Create a JupyterLab space. Enter a name for the space, and keep the space as Private. Leave all other settings at their default values, and then choose Create space.

    Otherwise, run your JupyterLab space to launch a JupyterLab application.

  3. Deploy your LLM and embedding models for inference

    • From the top menu, choose File, New, and then Terminal.

    • In the terminal, run the following command.

      wget --no-check-certificate https://raw.githubusercontent.com/aws-samples/sagemaker-studio-foundation-models/main/lab-00-setup/Lab_0_Warm_Up_Deploy_EmbeddingModel_Llama2_on_Nvidia.ipynb
mkdir AWSGuides
cd AWSGuides
wget --no-check-certificate https://raw.githubusercontent.com/aws-samples/sagemaker-studio-foundation-models/main/lab-03-rag/AWSGuides/AmazonSageMakerDeveloperGuide.pdf
wget --no-check-certificate https://raw.githubusercontent.com/aws-samples/sagemaker-studio-foundation-models/main/lab-03-rag/AWSGuides/EC2DeveloperGuide.pdf
wget --no-check-certificate https://raw.githubusercontent.com/aws-samples/sagemaker-studio-foundation-models/main/lab-03-rag/AWSGuides/S3DeveloperGuide.pdf

      This retrieves the the Lab_0_Warm_Up_Deploy_EmbeddingModel_Llama2_on_Nvidia.ipynb notebook to your local directory and downloads three PDF files into a local AWSGuides folder.

    • Open lab-00-setup/Lab_0_Warm_Up_Deploy_EmbeddingModel_Llama2_on_Nvidia.ipynb, keep the Python 3 (ipykernel) kernel, and run each cell.

      Warning

      In the Llama 2 License Agreement section, ensure to accept the Llama2 EULA before you continue.

      The notebook deploys two models, Llama 2 and all-MiniLM-L6-v2 Models, on ml.g5.2xlarge for inference.

      The deployment of the models and the creation of the endpoints may take some time.

  4. Open your main notebook

    In JupyterLab, open your terminal and run the following command.

    cd ..
wget --no-check-certificate https://raw.githubusercontent.com/aws-samples/sagemaker-studio-foundation-models/main/lab-03-rag/Lab_3_RAG_on_SageMaker_Studio_using_EMR.ipynb

    You should see the additional Lab_3_RAG_on_SageMaker_Studio_using_EMR.ipynb notebook in the left panel of JupyterLab.

  5. Choose a PySpark kernel

    Open your Lab_3_RAG_on_SageMaker_Studio_using_EMR.ipynb notebook and ensure that you are using the SparkMagic PySpark kernel. You can switch kernel at the top right of your notebook. Choose the current kernel name to open up a kernel selection modal, and then choose SparkMagic PySpark.

  6. Connect your notebook to the cluster

    1. At the top right of your notebook, choose Cluster. This action opens a modal window that lists all of the running clusters that you have permission to access.

    2. Select your cluster then choose Connect. A new credential type selection modal window opens up.

    3. Choose No credential and then Connect.

      Modal showing selection of Amazon EMR credentials for JupyterLab notebooks.

    4. A notebook cell automatically populates and runs. The notebook cell loads the sagemaker_studio_analytics_extension.magics extension, which provides functionality to connect to the Amazon EMR cluster. It then uses the %sm_analytics magic command to initiate the connection to your Amazon EMR cluster and the Spark application.

      Note

      Ensure that the connection string to your Amazon EMR cluster has an authentication type set to None. This is illustrated by the value --auth-type None in the following example. You can modify the field if necessary.

      %load_ext sagemaker_studio_analytics_extension.magics
%sm_analytics emr connect --verify-certificate False --cluster-id your-cluster-id --auth-type None --language python

    5. Once you successfully establish the connection, your connection cell output message should display your SparkSession details including your cluster ID, YARN application ID, and a link to the Spark UI to monitor your Spark jobs.

You are ready to use the Lab_3_RAG_on_SageMaker_Studio_using_EMR.ipynb notebook. This example notebook runs distributed PySpark workloads for building a RAG system using LangChain and OpenSearch.

Step 4: Clean up your AWS CloudFormation stack

After you are finished, make sure to terminate your two endpoints and delete your AWS CloudFormation stack to prevent continued charges. Deleting the stack cleans up all the resources that were provisioned by the stack.

To delete your AWS CloudFormation stack when you are done with it

  1. Go to the AWS CloudFormation console: https://console.aws.amazon.com/cloudformation

  2. Select the stack you want to delete. You can search for it by name or find it in the list of stacks.

  3. Click the Delete button to finalize deleting the stack and then Delete again to acknowledge that this will delete all resources created by the stack.

    Wait for the stack deletion to complete. This can take a few minutes. AWS CloudFormation automatically cleans up all resources defined in the stack template.

  4. Verify that all resources created by the stack have been deleted. For example, check for any leftover Amazon EMR cluster.

To remove the API endpoints for a model

  1. Go to the SageMaker console: https://console.aws.amazon.com/sagemaker/.

  2. In the left navigation pane, choose Inference and then Endpoints.

  3. Select the endpoint hf-allminil6v2-embedding-ep and then choose Delete in the Actions drop down list. Repeat the step for the endpoint meta-llama2-7b-chat-tg-ep.