Converting to Columnar Formats

Your Amazon Athena query performance improves if you convert your data into open source columnar formats such as Apache Parquet or ORC.

You can do this to existing Amazon S3 data sources by creating a cluster in Amazon EMR and converting it using Hive. The following example using the AWS CLI shows you how to do this with a script and data stored in Amazon S3.


The process for converting to columnar formats using an EMR cluster is as follows:

  • Create an EMR cluster with Hive installed.

  • In the step section of the cluster create statement, you can specify a script stored in Amazon S3, which points to your input data and creates output data in the columnar format in an Amazon S3 location. In this example, the cluster auto-terminates.

    For more information, here's an example script beginning with the CREATE TABLE snippet:

    ADD JAR /usr/lib/hive-hcatalog/share/hcatalog/hive-hcatalog-core-1.0.0-amzn-5.jar;
    CREATE EXTERNAL TABLE impressions (
      requestBeginTime string,
      adId string,
      impressionId string,
      referrer string,
      userAgent string,
      userCookie string,
      ip string,
      number string,
      processId string,
      browserCookie string,
      requestEndTime string,
      timers struct<modelLookup:string, requestTime:string>,
      threadId string,
      hostname string,
      sessionId string)
    PARTITIONED BY (dt string)
    ROW FORMAT  serde ''
    with serdeproperties ( 'paths'='requestBeginTime, adId, impressionId, referrer, userAgent, userCookie, ip' )
    LOCATION 's3://${REGION}.elasticmapreduce/samples/hive-ads/tables/impressions' ;


    Replace REGION in the LOCATION clause with the region where you are running queries. For example, if your console is in us-east-1, REGION will be s3://us-east-1.elasticmapreduce/samples/hive-ads/tables/.

    This creates the table in Hive on the cluster which uses samples located in the Amazon EMR samples bucket. On Amazon EMR release 4.7.0, you need to include the ADD JAR line to find the appropriate JsonSerDe. The prettified sample data looks like the following:

        "number": "977680",
        "referrer": "",
        "processId": "1823",
        "adId": "TRktxshQXAHWo261jAHubijAoNlAqA",
        "browserCookie": "mvlrdwrmef",
        "userCookie": "emFlrLGrm5fA2xLFT5npwbPuG7kf6X",
        "requestEndTime": "1239714001000",
        "impressionId": "1I5G20RmOuG2rt7fFGFgsaWk9Xpkfb",
        "userAgent": "Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 6.0; SLCC1; .NET CLR 2.0.50727; Media Center PC 5.0; .NET CLR 3.0.04506; InfoPa",
        "timers": {
            "modelLookup": "0.3292",
            "requestTime": "0.6398"
        "threadId": "99",
        "ip": "",
        "modelId": "bxxiuxduad",
        "hostname": "",
        "sessionId": "J9NOccA3dDMFlixCuSOtl9QBbjs6aS",
        "requestBeginTime": "1239714000000"

    In Hive, you need to load the data from the partitions, so the script runs the following:

    MSCK REPAIR TABLE impressions;

    The script then creates a table that stores your data in a Parquet-formatted file on Amazon S3:

    CREATE EXTERNAL TABLE  parquet_hive (
        requestBeginTime string,
        adId string,
        impressionId string,
        referrer string,
        userAgent string,
        userCookie string,
        ip string
    LOCATION 's3://myBucket/myParquet/';

    The data are inserted from the impressions table into parquet_hive:

    ip FROM impressions WHERE dt='2009-04-14-04-05';

    The script stores the above impressions table columns from the date, 2009-04-14-04-05, into s3://myBucket/myParket/ in a Parquet-formatted file. The full script is located on Amazon S3 at:


  • After your EMR cluster is terminated, you then create your table in Athena, which uses the data in the format produced by the cluster.


Example: Converting data to Parquet using an EMR cluster#

  1. Use the AWS CLI to create a cluster. If you need to install the AWS CLI, see Installing the AWS Command Line Interface in the AWS CLI User Guide.
  2. You need roles to use Amazon EMR, so if you haven't used Amazon EMR before, create the default roles using the following command:
aws emr create-default-roles

For more information, see Create and Use IAM Roles for Amazon EMR in the Amazon EMR Management Guide.

  1. Create an Amazon EMR cluster using the emr-4.7.0 release to convert the data using the following AWS CLI emr create-cluster command:
export REGION=us-east-1
export SAMPLEURI=s3://${REGION}.elasticmapreduce/samples/hive-ads/tables/impressions/
export S3BUCKET=myBucketName

aws emr create-cluster --applications Name=Hadoop Name=Hive Name=HCatalog \
--ec2-attributes KeyName=myKey,InstanceProfile=EMR_EC2_DefaultRole,SubnetId=subnet-mySubnetId \
--service-role EMR_DefaultRole --release-label emr-4.7.0 --instance-type \
m4.large --instance-count 1 --steps Type=HIVE,Name="Convert to Parquet",\
s3://athena-examples/conversion/write-parquet-to-s3.q,-hiveconf,INPUT=${SAMPLEURI},-hiveconf,OUTPUT=s3://{$S3BUCKET}/myParquet,-hiveconf,REGION=${REGION}] \
--region ${REGION} --auto-terminate

A successful request gives you a cluster ID. You can monitor the progress of your cluster using the AWS Management Console or using the cluster ID with the list-steps subcommand in the AWS CLI:

aws emr list-steps --cluster-id myClusterID

Look for the script step status. If it is COMPLETED, then the conversion is done and you are ready to query the data.

  1. Now query the data in Athena. First, you need to create the same table that you created on the EMR cluster.

    You can use the same statement as above. Log into Athena and enter the statement in the Query Editor window:

    CREATE EXTERNAL TABLE  parquet_hive (
        requestBeginTime string,
        adId string,
        impressionId string,
        referrer string,
        userAgent string,
        userCookie string,
        ip string
    LOCATION 's3://myBucket/myParquet/';

    Choose Run Query.

  2. Run the following query to show that you can query this data:

    SELECT * FROM parquet_hive LIMIT 10;

    Alternatively, you can select the view (eye) icon by the parquet_hive table in Catalog:


    The results should show output similar to this: