EMR Serverless PySpark job

This example shows how to run a PySpark job on EMR Serverless that analyzes data from the NOAA Global Surface Summary of Day dataset from the Registry of Open Data on AWS.

The script analyzes data from a given year and finds the weather location with the most extreme rain, wind, snow, and temperature.

ℹ️ Throughout this demo, I utilize environment variables to allow for easy copy/paste

⚠️ The included script is only intended to be run in the us-east-1 region. You can run in other regions by setting up EMR Serverless with a VPC, which may incur additional cost, or copy some sampe data from the s3://noaa-gsod-pds/2022/ bucket.

Setup

You should have already completed the pre-requisites in this repo's README.

• Define some environment variables to be used later

export S3_BUCKET=<YOUR_BUCKET_NAME>
export JOB_ROLE_ARN=arn:aws:iam::<ACCOUNT_ID>:role/emr-serverless-job-role

• First, make sure the extreme_weather.py script is uploaded to an S3 bucket in the `us-east-1 region.

aws s3 cp extreme_weather.py s3://${S3_BUCKET}/code/pyspark/

• Now, let's create and start an Application on EMR Serverless. Applications are where you submit jobs and are associated with a specific open source framework and release version. For this application, we'll configure pre-initialized capacity to ensure this application can begin running jobs immediately.

aws emr-serverless create-application \
  --type SPARK \
  --name serverless-demo \
  --release-label "emr-6.6.0" \
    --initial-capacity '{
        "DRIVER": {
            "workerCount": 2,
            "workerConfiguration": {
                "cpu": "2vCPU",
                "memory": "4GB"
            }
        },
        "EXECUTOR": {
            "workerCount": 10,
            "workerConfiguration": {
                "cpu": "4vCPU",
                "memory": "8GB"
            }
        }
    }' \
    --maximum-capacity '{
        "cpu": "200vCPU",
        "memory": "200GB",
        "disk": "1000GB"
    }'

This will return information about your application. In this case, we've created an application that can handle 2 simultaneous Spark apps with an initial set of 10 executors, each with 4vCPU and 4GB of memory, that can scale up to 200vCPU or 50 executors.

{
  "applicationId": "00et0dhmhuokmr09",
  "arn": "arn:aws:emr-serverless:us-east-1:123456789012:/applications/00et0dhmhuokmr09",
  "name": "serverless-demo"
}

We'll set an APPLICATION_ID environment variable to reuse later.

export APPLICATION_ID=00et0dhmhuokmr09

• Get the state of your application

aws emr-serverless get-application \
    --application-id $APPLICATION_ID

Once your application is in CREATED state, you can go ahead and start it.

aws emr-serverless start-application \
    --application-id $APPLICATION_ID

Once your application is in STARTED state, you can submit jobs.

With pre-initialized capacity, you can define a minimum amount of resources that EMR Serverless keeps ready to respond to interactive queries. EMR Serverless will scale your application up as necessary to respond to workloads, but return to the pre-initialized capacity when there is no activity. You can start or stop an application to effectively pause your application so that you are not billed for resources you're not using. If you don't need second-level response times in your workloads, you can use the default capacity and EMR Serverless will decomission all resources when a job is complete and scale back up as more workloads come in.

Run your job

Now that you've created your application, you can submit jobs to it at any time.

We define our sparkSubmitParameters with resources that match our pre-initialized capacity, but EMR Serverless will still automatically scale as necessary.

ℹ️ Note that with Spark jobs, you must account for Spark overhead and configure our executor with less memory than the application.

In this case, we're also configuring Spark logs to be delivered to our S3 bucket.

aws emr-serverless start-job-run \
    --application-id $APPLICATION_ID \
    --execution-role-arn $JOB_ROLE_ARN \
    --job-driver '{
        "sparkSubmit": {
            "entryPoint": "s3://'${S3_BUCKET}'/code/pyspark/extreme_weather.py",
            "sparkSubmitParameters": "--conf spark.driver.cores=1 --conf spark.driver.memory=3g --conf spark.executor.cores=4 --conf spark.executor.memory=3g --conf spark.executor.instances=10"
        }
    }' \
    --configuration-overrides '{
        "monitoringConfiguration": {
            "s3MonitoringConfiguration": {
                "logUri": "s3://'${S3_BUCKET}'/logs/"
            }
        }
    }'

{
  "applicationId": "00esprurjpeqpq09",
  "arn": "arn:aws:emr-serverless:us-east-1:123456789012:/applications/00esprurjpeqpq09/jobruns/00esps8ka2vcu801",
  "jobRunId": "00esps8ka2vcu801"
}

Let's set our JOB_RUN_ID variable so we can use it to monitor the job progress.

export JOB_RUN_ID=00esps8ka2vcu801

aws emr-serverless get-job-run \
    --application-id $APPLICATION_ID \
    --job-run-id $JOB_RUN_ID

The job should start within a few seconds since we're making use of pre-initialized capacity.

We can also look at our logs while the job is running.

aws s3 ls s3://${S3_BUCKET}/logs/applications/$APPLICATION_ID/jobs/$JOB_RUN_ID/

Or copy the stdout of the job.

aws s3 cp s3://${S3_BUCKET}/logs/applications/$APPLICATION_ID/jobs/$JOB_RUN_ID/SPARK_DRIVER/stdout.gz - | gunzip

Clean up

When you're all done, make sure to call stop-application to decommission your capacity and delete-application if you're all done.

aws emr-serverless stop-application \
    --application-id $APPLICATION_ID

aws emr-serverless delete-application \
    --application-id $APPLICATION_ID

Spark UI Debugging

• Follow the steps in building the Spark UI Docker container to build the container locally

• Get credentials and set LOG_DIR

export AWS_ACCESS_KEY_ID=AKIAaaaa
export AWS_SECRET_ACCESS_KEY=bbbb
export AWS_SESSION_TOKEN=yyyy

export LOG_DIR=s3://${S3_BUCKET}/logs/applications/$APPLICATION_ID/jobs/$JOB_RUN_ID/sparklogs/

• Fire up Docker in the background

docker run --rm -d \
    --name emr-serverless-spark-ui \
    -p 18080:18080 \
    -e SPARK_HISTORY_OPTS="-Dspark.history.fs.logDirectory=$LOG_DIR -Dspark.hadoop.fs.s3.customAWSCredentialsProvider=com.amazonaws.auth.DefaultAWSCredentialsProviderChain" \
    -e AWS_REGION=us-east-1 \
    -e AWS_ACCESS_KEY_ID -e AWS_SECRET_ACCESS_KEY -e AWS_SESSION_TOKEN \
    emr/spark-ui

• Access the Spark UI via http://localhost:18080

• When you're done, stop the Docker image

docker stop emr-serverless-spark-ui

Using the AWS Glue Data Catalog with EMR Serverless

You can use the Glue Data Catalog along with SparkSQL in EMR Serverless by setting the proper Hive metastore config item.

You can do this when creating a new SparkSession in your PySpark code - make sure you also call enableHiveSupport().

from pyspark.sql import SparkSession

spark = (
    SparkSession.builder.appName("SparkSQL")
    .config(
        "hive.metastore.client.factory.class",
        "com.amazonaws.glue.catalog.metastore.AWSGlueDataCatalogHiveClientFactory",
    )
    .enableHiveSupport()
    .getOrCreate()
)

# We can query tables with SparkSQL
spark.sql("SHOW TABLES").show()

# Or we can also them with native Spark code
print(spark.catalog.listTables())