EMR Toolkit is a VS Code Extension to make it easier to develop Spark jobs on EMR.
Requirements
- A local AWS profile
- Access to the AWS API to list EMR and Glue resources
- Docker (if you want to use the devcontainer)
Features
Amazon EMR Explorer
The Amazon EMR Explorer allows you to browse job runs and steps across EMR on EC2, EMR on EKS, and EMR Serverless. To see the Explorer, choose the EMR icon in the Activity bar.
Note: If you do not have default AWS credentials or AWS_PROFILE
environment variable, use the EMR: Select AWS Profile
command to select your profile.
Glue Catalog Explorer
The Glue Catalog Explorer displays databases and tables in the Glue Data Catalog. By right-clicking on a table, you can select View Glue Table
that will show the table columns.
PySpark EMR Development Container
The toolkit provides an EMR: Create local Spark environment
command that creates a development container based off of an EMR on EKS image for the EMR version you choose. This container can be used to develop Spark and PySpark code locally that is fully compatible with your remote EMR environment.
You choose a region and EMR version you want to use, and the extension creates the relevant Dockerfile
and devcontainer.json
.
Once the container is created, follow the instructions in the emr-local.md
file to authenticate to ECR and use the Dev--Containers: Reopen in Container
command to build and open your local Spark environment.
You can choose to configure AWS authentication in the container in 1 of 3 ways:
- Use existing ~/.aws config - This mounts your ~/.aws directory to the container.
- Environment variables - If you already have AWS environment variables configured in your shell, the container will reference those variables.
- .env file - Creates a
.devcontainer/aws.env
file that you can populate with AWS credentials.
Spark Shell Support
The EMR Development Container is configured to run Spark in local mode. You can use it like any Spark-enabled environment. Inside the VS Code Terminal, you can use the pyspark
or spark-shell
commands to start a local Spark session.
Jupyter Notebook Support
By default, the EMR Development Container also supports Jupyter. Use the Create: New Jupyter Notebook command to create a new Jupyter notebook. The following code snippet shows how to initialize a Spark Session inside the notebook. By default, the Container environment is also configured to use the Glue Data Catalog so you can use spark.sql
commands against Glue tables.
from pyspark.sql import SparkSession
spark = (
SparkSession.builder.appName("EMRLocal")
.getOrCreate()
)
EMR Serverless Deployment
You can deploy and run a single PySpark file on EMR Serverless with the EMR Serverless: Deploy and run PySpark job command. You'll be prompted for the following information:
- S3 URI - Your PySpark file will be copied here
- IAM Role - A job runtime role that can be used to run your EMR Serverless job
- EMR Serverless Application ID - The ID of an existing EMR Serverless Spark application
- Filename - The name of the local PySpark file you want to run on EMR Serverless
https://user-images.githubusercontent.com/1512/195953681-4e7e7102-4974-45b1-a695-195e91d45124.mp4
Future Considerations
- Allow for the ability to select different profiles
- Persist state (region selection)
- Create a Java environment
- Automate deployments to EMR
- Create virtualenv and upload to S3
- Pack pom into jar file
- Link to open logs in S3 or CloudWatch
- Testing :) https://vscode.rocks/testing/
Feedback Notes
I'm looking for feedback in a few different areas:
- How do you use Spark on EMR today?
- EMR on EC2, EMR on EKS, or EMR Serverless
- PySpark, Scala Spark, or SparkSQL
- Does the tool work as expected for browsing your EMR resources
- Do you find the devcontainer useful for local development
- What functionality is missing that you would like to see
Security
See CONTRIBUTING for more information.
License
This project is licensed under the Apache-2.0 License.