emr-bootstrap-spark

This package contains the AWS bootstrap scripts for Mozilla's flavoured Spark setup. The deployed scripts in S3 are referenced by ATMO clusters and Airflow jobs.

Interactive job

export SPARK_PROFILE=telemetry-spark-cloudformation-TelemetrySparkInstanceProfile-1SATUBVEXG7E3
export SPARK_BUCKET=telemetry-spark-emr-2
export KEY_NAME=20161025-dataops-dev
aws emr create-cluster \
  --region us-west-2 \
  --name SparkCluster \
  --instance-type c3.4xlarge \
  --instance-count 1 \
  --service-role EMR_DefaultRole \
  --ec2-attributes KeyName=${KEY_NAME},InstanceProfile=${SPARK_PROFILE} \
  --release-label emr-5.2.1 \
  --applications Name=Spark Name=Hive Name=Zeppelin \
  --bootstrap-actions Path=s3://${SPARK_BUCKET}/bootstrap/telemetry.sh \
  --configurations https://s3-us-west-2.amazonaws.com/${SPARK_BUCKET}/configuration/configuration.json \
  --steps Type=CUSTOM_JAR,Name=CustomJAR,ActionOnFailure=TERMINATE_JOB_FLOW,Jar=s3://us-west-2.elasticmapreduce/libs/script-runner/script-runner.jar,Args=\["s3://${SPARK_BUCKET}/steps/zeppelin/zeppelin.sh"\]

Batch job

# Also export the vars from the 'interactive' section above.
export DATA_BUCKET=telemetry-public-analysis-2 # Or use the private bucket.
export CODE_BUCKET=telemetry-analysis-code-2
aws emr create-cluster \
  --region us-west-2 \
  --name SparkCluster \
  --instance-type c3.4xlarge \
  --instance-count 1 \
  --service-role EMR_DefaultRole \
  --ec2-attributes KeyName=${KEY_NAME},InstanceProfile=${SPARK_PROFILE} \
  --release-label emr-5.2.1 \
  --applications Name=Spark Name=Hive \
  --bootstrap-actions Path=s3://${SPARK_BUCKET}/bootstrap/telemetry.sh \
  --configurations https://s3-us-west-2.amazonaws.com/${SPARK_BUCKET}/configuration/configuration.json \
  --auto-terminate \
  --steps Type=CUSTOM_JAR,Name=CustomJAR,ActionOnFailure=TERMINATE_JOB_FLOW,Jar=s3://us-west-2.elasticmapreduce/libs/script-runner/script-runner.jar,Args=\["s3://${SPARK_BUCKET}/steps/batch.sh","--job-name","foo","--notebook","s3://${CODE_BUCKET}/jobs/foo/Telemetry Hello World.ipynb","--data-bucket","${DATA_BUCKET}"\]

Deploy to AWS via ansible

To deploy to the staging location:

ansible-playbook ansible/deploy.yml -e '@ansible/envs/stage.yml' -i ansible/inventory

Once deployed, you can see the effects in action by launching a cluster via ATMO stage.

To deploy for production clusters:

ansible-playbook ansible/deploy.yml -e '@ansible/envs/production.yml' -i ansible/inventory

The Spark Jupyter notebook configuration is hosted at https://s3-us-west-2.amazonaws.com/telemetry-spark-emr-2/credentials/jupyter_notebook_config.py. At the moment, this is only needed for the GitHub Gist export option in the Jupyter notebook. The credentials it contains are managed under the Mozilla GitHub account by :whd. This file should not be made public.

Contributing to emr-bootstrap-spark

You may set up a development environment to test and verify modifications applied to this repository.

Install prerequisite packages

pip install ansible boto boto3

Create and bootstrap the development environment

• Define a new ansible environment in env/dev-<username>.yml
  • Set spark_emr_bucket to a unique bucket e.g. telemetry-spark-emr-2-dev-<username>
  • Set stack_name to a unique name e.g. telemetry-spark-cloudformation-dev-<username>
• Recursively copy assets from staging to dev
  • aws s3 cp --recursive s3://telemetry-spark-emr-2-stage s3://telemetry-spark-emr-2-dev-<username>
• Deploy to AWS using ansible-playbook on the new environment
• Launch a new instance using the appropriate SPARK_PROFILE and SPARK_BUCKET keys
  • Set SPARK_PROFILE to the cloudformation instance profile
    • This can be found as an output on the cloudformation dashboard
    • Alternatively:

         aws cloudformation describe-stacks --stack-name telemetry-spark-cloudformation-dev-<username> |
         jq '.Stacks[0].Outputs[0].OutputValue'
      

  • Set SPARK_BUCKET to spark_emr_bucket value in env/dev-<username>.yml