DX Spark Submit Utility

Dx-spark-submit is a utility script that can be used in DNAnexus Spark applications to more easily submit and monitor a Spark job.

DNAnexus is working to phase out outdated terminology and change scripts using those terms to remove inappropriate language. The terms master and slave will be replaced with driver and clusterWorker in Spark documentation. DNAnexus will also eventually replace the older terms in the codebase. For now, variable names and scripts containing the older terms will still be used in the actual code.

A license is required to access Spark functionality on the DNAnexus Platform. Contact DNAnexus Sales for more information.

Usage

usage: dx-spark-submit [-h | --help] [--log-level {INFO,WARN,TRACE,DEBUG}]
                       [--collect-logs] [--log-collect-dir LOG_COLLECT_DIR]
                       [--app-config APP_CONFIG] [--user-config USER_CONFIG]
                       spark-driver-args

positional arguments:
  spark-driver-args     Options to be passed directly to spark-submit, including 
                        Spark application, properties, and driver options

optional arguments:
  -h, --help            show this help message and exit
  --log-level {INFO,WARN,TRACE,DEBUG}
                        Log level for driver and executor
  --collect-logs        Collect logs to a project in the platform
  --log-collect-dir LOG_COLLECT_DIR
                        Directory in project to upload logs
  --app-config APP_CONFIG
                        Application configuration json string or file
  --user-config USER_CONFIG
                        User configuration json string or file

How does it work?

The dx-spark-submit utility simplifies common Spark application tasks.

• Allows easy overrides of Spark properties at the app developer and user level.

• Sets the driver and executor log level.

• Submits and sets up the UI to monitor Spark jobs.

• Initiates log collection once the job is done (success or failure).

Spark Property Overrides

Spark apps depend on specific configurations like spark-defaults.conf, hive-site.xml which will set up the environment for your application. In certain scenarios, an application developer or the user of the application may want to override a default setting.

dx-spark-submit allows you to specify two configuration inputs.

• Application configuration

• User configuration

Application Configuration JSON

Application config JSON (--app-config) contains the list of configurations the app developer may want to restrict or override.

{
  "spark-defaults.conf": [         
    {
      "name": "spark.ui.port",
      "value": 8081,
      "override_allowed": true
    },
    {
      "name": "spark.sql.parquet.filterPushdown",
      "value": false
    }

  ]
}

User Configuration JSON

User config JSON (--user-config) contains the list of configurations the app user may want to add or override. If you want to offer this override ability to users of your app you will need to reference this file in the app input spec, so it's available to dx-spark-submit.

{
  "spark-defaults.conf": [         
    {
      "name": "spark.ui.port",
      "value": 8080
    },
    {
      "name": "spark.sql.shuffle.partitions",
      "value": 1
    }
  ]
}

Note: These Spark configurations cannot be overridden as they affect the basic functioning of the cluster application:

spark.driver.host
spark.driver.bindAddress
spark.driver.port
spark.driver.blockManager.port
spark.blockManager.port
spark.port.maxRetries
spark.master
spark.driver.extraClassPath
spark.jars
spark.hadoop.mapreduce.fileoutputcommitter.algorithm.version

Log Collection

When --collect-logs option is set, log collection will be triggered by the script. It will collect the logs from clusterWorker and driver and upload to the project by default. If --log-collect-dir is specified, the logs are copied to the specified folder in the project.

Note: Subjobs cannot use the log collection feature.

Log Level

--log-level can be used to set the driver and executor log level (INFO, WARN, TRACE, DEBUG).

Spark Arguments

spark-driver-args should contain the Spark application and any arguments you want to pass to spark-submit.

Example

    dx-spark-submit \
        --log-level INFO \
        --collect-logs \
        --log-collect-dir pitestlogs \
        --app-config /app.json \
        --user-config /user.json \
        --class org.apache.spark.examples.SparkPi /cluster/spark/examples/jars/spark-examples*.jar 10

Note: dx-spark-submit is located in /cluster/dnax/bin on the Spark cluster worker container.