Enhance kernels with magic commands in EMR Studio

Overview

EMR Studio and EMR Notebooks support magic commands. Magic commands, or magics, are enhancements that the IPython kernel provides to help you run and analyze data. IPython is an interactive shell environment that is built with Python.

Amazon EMR also supports Sparkmagic, a package that provides Spark-related kernels (PySpark, SparkR, and Scala kernels) with specific magic commands and that uses Livy on the cluster to submit Spark jobs.

You can use magic commands as long as you have a Python kernel in your EMR notebook. Similarly, any Spark-related kernel supports Sparkmagic commands.

Magic commands, also called magics, come in two varieties:

For all available magics, see List magic and Sparkmagic commands.

Considerations and limitations

List magic and Sparkmagic commands

Use the following commands to list the available magic commands:

Use %%configure to configure Spark

One of the most useful Sparkmagic commands is the %%configure command, which configures the session creation parameters. Using conf settings, you can configure any Spark configuration that's mentioned in the configuration documentation for Apache Spark.

%%configure -f
{"conf": {
    "spark.jars.packages": "com.jsuereth:scala-arm_2.11:2.0,ml.combust.bundle:bundle-ml_2.11:0.13.0,com.databricks:dbutils-api_2.11:0.0.3",
    "spark.jars": "s3://amzn-s3-demo-bucket/my-jar.jar"
    }
}        

%%configure
{ "conf": {
     "spark.jars": "hdfs://apps/hudi/lib/hudi-spark-bundle.jar,hdfs:///apps/hudi/lib/spark-spark-avro.jar", 
     "spark.serializer": "org.apache.spark.serializer.KryoSerializer",
     "spark.sql.hive.convertMetastoreParquet":"false"
     }
}

Use %%sh to run spark-submit

The %%sh magic runs shell commands in a subprocess on an instance of your attached cluster. Typically, you'd use one of the Spark-related kernels to run Spark applications on your attached cluster. However, if you want to use a Python kernel to submit a Spark application, you can use the following magic, replacing the bucket name with your bucket name in lowercase.

%%sh
spark-submit --master yarn --deploy-mode cluster s3://amzn-s3-demo-bucket/test.py

In this example, the cluster needs access to the location of s3://amzn-s3-demo-bucket/test.py, or the command will fail.

You can use any Linux command with the %%sh magic. If you want to run any Spark or YARN commands, use one of the following options to create an emr-notebook Hadoop user and grant the user permissions to run the commands:

Use %%display to visualize Spark dataframes

You can use the %%display magic to visualize a Spark dataframe. To use this magic, run the following command.

%%display df

Choose to view the results in a table format, as the following image shows.

You can also choose to visualize your data with five types of charts. Your options include pie, scatter, line, area, and bar charts.

Use EMR Notebooks magics

Amazon EMR provides the following EMR Notebooks magics that you can use with Python3 and Spark-based kernels:

Prerequisites

Before you install EMR Notebooks magics, complete the following tasks:

Install EMR Notebooks magics

Complete the following steps to install EMR Notebooks magics.

Mount a Workspace directory with %mount_workspace_dir

The %mount_workspace_dir magic lets you mount your Workspace directory onto your EMR cluster so that you can import and run other files, modules, or packages stored in your directory.

The following example mounts the entire Workspace directory onto a cluster, and specifies the optional <--fuse-type> argument to use goofys for mounting the directory.

%mount_workspace_dir . <--fuse-type goofys>

To verify that your Workspace directory is mounted, use the following example to display the current working directory with the ls command. The output should display all of the files in your Workspace.

%%sh
ls

When you're done making changes in your Workspace, you can unmount the Workspace directory with the following command:

%umount_workspace_dir

Download an Amazon S3 object with %generate_s3_download_url

The generate_s3_download_url command creates a presigned URL for an object stored in Amazon S3. You can use the presigned URL to download the object to your local machine. For example, you might run generate_s3_download_url to download the result of a SQL query that your code writes to Amazon S3.

The presigned URL is valid for 60 minutes by default. You can change the expiration time by specifying a number of seconds for the --expires-in flag. For example, --expires-in 1800 creates a URL that is valid for 30 minutes.

The following example generates a download link for an object by specifying the full Amazon S3 path: s3://EXAMPLE-DOC-BUCKET/path/to/my/object.

%generate_s3_download_url s3://EXAMPLE-DOC-BUCKET/path/to/my/object

To learn more about using generate_s3_download_url, run the following command to display help text.

%generate_s3_download_url?

Run a notebook in headless mode with %execute_notebook

With %execute_notebook magic, you can run another notebook in headless mode and view the output for each cell that you've run. This magic requires additional permissions for the instance role that Amazon EMR and Amazon EC2 share. For more details on how to grant additional permissions, run the command %execute_notebook?.

During a long-running job, your system might go to sleep because of inactivity, or might temporarily lose internet connectivity. This might disrupt the connection between your browser and the Jupyter Server. In this case, you might lose the output from the cells that you've run and sent from the Jupyter Server.

If you run the notebook in headless mode with %execute_notebook magic, EMR Notebooks captures output from the cells that have run, even if the local network experiences disruption. EMR Notebooks saves the output incrementally in a new notebook with the same name as the notebook that you've run. EMR Notebooks then places the notebook into a new folder within the workspace. Headless runs occur on the same cluster and uses service role EMR_Notebook_DefaultRole, but additional arguments can alter the default values.

To run a notebook in headless mode, use the following command:

%execute_notebook <relative-file-path>

To specify a cluster ID and service role for a headless run, use the following command:

%execute_notebook <notebook_name>.ipynb --cluster-id <emr-cluster-id> --service-role <emr-notebook-service-role>

When Amazon EMR and Amazon EC2 share an instance role, the role requires the following additional permissions:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "elasticmapreduce:StartNotebookExecution",
                "elasticmapreduce:DescribeNotebookExecution",
                "ec2:DescribeInstances"
            ],
            "Resource": "*"
        },
        {
            "Effect": "Allow",
            "Action": [
                "iam:PassRole"
            ],
            "Resource": "arn:aws:iam::<AccoundId>:role/EMR_Notebooks_DefaultRole"
        }
    ]
}