Skip to content

Latest commit

 

History

History
220 lines (165 loc) · 7.08 KB

README.md

File metadata and controls

220 lines (165 loc) · 7.08 KB
Raw

EMR Serverless Hive query

This example shows how to run a Hive query 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.

  • This example requires that you have access to the AWS Glue Data Catalog.

  • 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.sql and create_table.sql scripts are uploaded to an S3 bucket in the us-east-1 region.
aws s3 cp extreme_weather.sql s3://${S3_BUCKET}/code/hive/
aws s3 cp create_table.sql s3://${S3_BUCKET}/code/hive/
  • 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.

ℹ️ Please note that leaving a pre-initialized application running WILL incur costs in your AWS Account.

aws emr-serverless create-application \
  --type HIVE \
  --name serverless-demo \
  --release-label "emr-6.6.0" \
  --initial-capacity '{
        "DRIVER": {
            "workerCount": 1,
            "workerConfiguration": {
                "cpu": "2vCPU",
                "memory": "4GB",
                "disk": "30gb"
            }
        },
        "TEZ_TASK": {
            "workerCount": 10,
            "workerConfiguration": {
                "cpu": "4vCPU",
                "memory": "8GB",
                "disk": "30gb"
            }
        }
    }' \
    --maximum-capacity '{
        "cpu": "400vCPU",
        "memory": "1024GB",
        "disk": "1000GB"
    }'

This will return information about your application.

{
    "applicationId": "00et0f0b79s06o09",
    "arn": "arn:aws:emr-serverless:us-east-1:123456789012:/applications/00et0f0b79s06o09",
    "name": "serverless-demo"
}

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

export APPLICATION_ID=00et0f0b79s06o09
  • 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.

aws emr-serverless start-job-run \
    --application-id $APPLICATION_ID \
    --execution-role-arn $JOB_ROLE_ARN \
    --job-driver '{
        "hive": {
            "initQueryFile": "s3://'${S3_BUCKET}'/code/hive/create_table.sql",
            "query": "s3://'${S3_BUCKET}'/code/hive/extreme_weather.sql",
            "parameters": "--hiveconf hive.exec.scratchdir=s3://'${S3_BUCKET}'/hive/scratch --hiveconf hive.metastore.warehouse.dir=s3://'${S3_BUCKET}'/hive/warehouse"
        }
    }' \
    --configuration-overrides '{
        "applicationConfiguration": [
            {
                "classification": "hive-site",
                "properties": {
                    "hive.driver.cores": "2",
                    "hive.driver.memory": "4g",
                    "hive.tez.container.size": "8192",
                    "hive.tez.cpu.vcores": "4"
                }
            }
        ],
        "monitoringConfiguration": {
            "s3MonitoringConfiguration": {
                "logUri": "s3://'${S3_BUCKET}'/hive-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://${S3_BUCKET}/hive-logs/applications/$APPLICATION_ID/jobs/$JOB_RUN_ID/

Or copy the stdout of the job.

aws s3 cp s3://${S3_BUCKET}/hive-logs/applications/$APPLICATION_ID/jobs/$JOB_RUN_ID/HIVE_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

Tez UI Debugging

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

export S3_LOG_URI=s3://${S3_BUCKET}/hive-logs
  • Fire up Docker
docker run --rm -d \
    --name emr-serverless-tez-ui \
    -p 8088:8088 -p 8188:8188 -p 9999:9999 \
    -e AWS_REGION -e AWS_ACCESS_KEY_ID -e AWS_SECRET_ACCESS_KEY -e AWS_SESSION_TOKEN \
    -e S3_LOG_URI -e JOB_RUN_ID -e APPLICATION_ID \
    emr/tez-ui
docker stop emr-serverless-tez-ui

JSON Data

The example job in this repository uses OpenCSVSerde to process CSV data. If you want to process JSON data, you can use the org.apache.hadoop.hive.serde2.JsonSerDe with EMR 6.x and org.apache.hive.hcatalog.data.JsonSerDe on EMR 5.x. org.openx.data.jsonserde.JsonSerDe can be used in either EMR release.