Spark Submit — `spark-submit` shell script

spark-submit shell script allows you to manage your Spark applications.

You can submit your Spark application to a Spark deployment environment for execution, kill or request status of Spark applications.

You can find spark-submit script in bin directory of the Spark distribution.

$ ./bin/spark-submit
Usage: spark-submit [options] <app jar | python file> [app arguments]
Usage: spark-submit --kill [submission ID] --master [spark://...]
Usage: spark-submit --status [submission ID] --master [spark://...]
Usage: spark-submit run-example [options] example-class [example args]
...

When executed, spark-submit script first checks whether SPARK_HOME environment variable is set and sets it to the directory that contains bin/spark-submit shell script if not. It then executes spark-class shell script to run SparkSubmit standalone application.

Caution

FIXME Add Cluster Manager and Deploy Mode to the table below (see options value)

Table 1. Command-Line Options, Spark Properties and Environment Variables (from `SparkSubmitArguments`'s loadEnvironmentArguments and handle)
Command-Line Option	Spark Property	Environment Variable	Description	Internal Property
`action`			Defaults to `SUBMIT`
`--archives`				`archives`
`--conf`				`sparkProperties`
`--deploy-mode`	`spark.submit.deployMode`	`DEPLOY_MODE`	Deploy mode	`deployMode`
`--driver-class-path`	`spark.driver.extraClassPath`		The driver’s class path	`driverExtraClassPath`
`--driver-java-options`	`spark.driver.extraJavaOptions`		The driver’s JVM options	`driverExtraJavaOptions`
`--driver-library-path`	`spark.driver.extraLibraryPath`		The driver’s native library path	`driverExtraLibraryPath`
`--driver-memory`	`spark.driver.memory`	`SPARK_DRIVER_MEMORY`	The driver’s memory	`driverMemory`
`--driver-cores`	`spark.driver.cores`			`driverCores`
`--exclude-packages`	`spark.jars.excludes`			`packagesExclusions`
`--executor-cores`	`spark.executor.cores`	`SPARK_EXECUTOR_CORES`	The number of executor CPU cores	`executorCores`
`--executor-memory`	`spark.executor.memory`	`SPARK_EXECUTOR_MEMORY`	An executor’s memory	`executorMemory`
`--files`	`spark.files`			`files`
`ivyRepoPath`	`spark.jars.ivy`
`--jars`	`spark.jars`			`jars`
`--keytab`	`spark.yarn.keytab`			`keytab`
`--kill`			`submissionToKill` and `action` set to `KILL`
`--master`	`spark.master`	`MASTER`	Master URL. Defaults to `local[*]`	`master`
`--class`				`mainClass`
`--name`	`spark.app.name`	`SPARK_YARN_APP_NAME` (YARN only)	Uses `mainClass` or the directory off `primaryResource` when no other ways set it	`name`
`--num-executors`	`spark.executor.instances`			`numExecutors`
`--packages`	`spark.jars.packages`			`packages`
`--principal`	`spark.yarn.principal`			`principal`
`--properties-file`	`spark.yarn.principal`			`propertiesFile`
`--proxy-user`				`proxyUser`
`--py-files`				`pyFiles`
`--queue`				`queue`
`--repositories`				`repositories`
`--status`			`submissionToRequestStatusFor` and `action` set to `REQUEST_STATUS`
`--supervise`				`supervise`
`--total-executor-cores`	`spark.cores.max`			`totalExecutorCores`
`--verbose`				`verbose`
`--version`			`SparkSubmit.printVersionAndExit()`
`--help`			`printUsageAndExit(0)`
`--usage-error`			`printUsageAndExit(1)`

Tip

Set SPARK_PRINT_LAUNCH_COMMAND environment variable to have the complete Spark command printed out to the console, e.g.

$ SPARK_PRINT_LAUNCH_COMMAND=1 ./bin/spark-shell
Spark Command: /Library/Ja...

Refer to Print Launch Command of Spark Scripts (or org.apache.spark.launcher.Main Standalone Application where this environment variable is actually used).

Tip	Avoid using `scala.App` trait for a Spark application’s main class in Scala as reported in SPARK-4170 Closure problems when running Scala app that "extends App". Refer to Executing Main — `runMain` internal method in this document.

Preparing Submit Environment — `prepareSubmitEnvironment` Internal Method

prepareSubmitEnvironment(args: SparkSubmitArguments)
  : (Seq[String], Seq[String], Map[String, String], String)

prepareSubmitEnvironment creates a 4-element tuple, i.e. (childArgs, childClasspath, sysProps, childMainClass).

Table 2. `prepareSubmitEnvironment`'s Four-Element Return Tuple
Element	Description
`childArgs`	Arguments
`childClasspath`	Classpath elements
`sysProps`	Spark properties
`childMainClass`	Main class

prepareSubmitEnvironment uses options to…

Caution

FIXME

Note	`prepareSubmitEnvironment` is used in `SparkSubmit` object.

Tip	See the elements of the return tuple using `--verbose` command-line option.

Custom Spark Properties File — `--properties-file` command-line option

--properties-file [FILE]

--properties-file command-line option sets the path to a file FILE from which Spark loads extra Spark properties.

Tip	Spark uses conf/spark-defaults.conf by default.

Driver Cores in Cluster Deploy Mode — `--driver-cores` command-line option

--driver-cores NUM

--driver-cores command-line option sets the number of cores to NUM for the driver in the cluster deploy mode.

Note	`--driver-cores` switch is only available for cluster mode (for Standalone, Mesos, and YARN).

Note	It corresponds to spark.driver.cores setting.

Note	It is printed out to the standard error output in verbose mode.

Additional JAR Files to Distribute — `--jars` command-line option

--jars JARS

--jars is a comma-separated list of local jars to include on the driver’s and executors' classpaths.

Caution

FIXME

Additional Files to Distribute `--files` command-line option

--files FILES

Caution

FIXME

Additional Archives to Distribute — `--archives` command-line option

--archives ARCHIVES

Caution

FIXME

Specifying YARN Resource Queue — `--queue` command-line option

--queue QUEUE_NAME

With --queue you can choose the YARN resource queue to submit a Spark application to. The default queue name is default.

Caution

FIXME What is a queue?

Note	It corresponds to spark.yarn.queue Spark’s setting.

Tip	It is printed out to the standard error output in verbose mode.

Actions

Submitting Applications for Execution — `submit` method

The default action of spark-submit script is to submit a Spark application to a deployment environment for execution.

Tip	Use --verbose command-line switch to know the main class to be executed, arguments, system properties, and classpath (to ensure that the command-line arguments and switches were processed properly).

When executed, spark-submit executes submit method.

submit(args: SparkSubmitArguments): Unit

If proxyUser is set it will…FIXME

Caution

FIXME Review why and when to use proxyUser.

It passes the execution on to runMain.

Executing Main — `runMain` internal method

runMain(
  childArgs: Seq[String],
  childClasspath: Seq[String],
  sysProps: Map[String, String],
  childMainClass: String,
  verbose: Boolean): Unit

runMain is an internal method to build execution environment and invoke the main method of the Spark application that has been submitted for execution.

Note	It is exclusively used when submitting applications for execution.

When verbose input flag is enabled (i.e. true) runMain prints out all the input parameters, i.e. childMainClass, childArgs, sysProps, and childClasspath (in that order).

Main class:
[childMainClass]
Arguments:
[childArgs one per line]
System properties:
[sysProps one per line]
Classpath elements:
[childClasspath one per line]

Note	Use `spark-submit`'s --verbose command-line option to enable `verbose` flag.

runMain builds the context classloader (as loader) depending on spark.driver.userClassPathFirst flag.

Caution

FIXME Describe spark.driver.userClassPathFirst

It adds the jars specified in childClasspath input parameter to the context classloader (that is later responsible for loading the childMainClass main class).

Note	`childClasspath` input parameter corresponds to --jars command-line option with the primary resource if specified in client deploy mode.

It sets all the system properties specified in sysProps input parameter (using Java’s System.setProperty method).

It creates an instance of childMainClass main class (as mainClass).

Note	`childMainClass` is the main class `spark-submit` has been invoked with.

Tip	Avoid using `scala.App` trait for a Spark application’s main class in Scala as reported in SPARK-4170 Closure problems when running Scala app that "extends App".

If you use scala.App for the main class, you should see the following warning message in the logs:

Warning: Subclasses of scala.App may not work correctly. Use a main() method instead.

Finally, runMain executes the main method of the Spark application passing in the childArgs arguments.

Any SparkUserAppException exceptions lead to System.exit while the others are simply re-thrown.

Adding Local Jars to ClassLoader — `addJarToClasspath` internal method

addJarToClasspath(localJar: String, loader: MutableURLClassLoader)

addJarToClasspath is an internal method to add file or local jars (as localJar) to the loader classloader.

Internally, addJarToClasspath resolves the URI of localJar. If the URI is file or local and the file denoted by localJar exists, localJar is added to loader. Otherwise, the following warning is printed out to the logs:

Warning: Local jar /path/to/fake.jar does not exist, skipping.

For all other URIs, the following warning is printed out to the logs:

Warning: Skip remote jar hdfs://fake.jar.

Note	`addJarToClasspath` assumes `file` URI when `localJar` has no URI specified, e.g. `/path/to/local.jar`.

Caution

FIXME What is a URI fragment? How does this change re YARN distributed cache? See Utils#resolveURI.

Killing Applications — `--kill` command-line option

--kill

Requesting Application Status — `--status` command-line option

--status

Command-line Options

Execute spark-submit --help to know about the command-line options supported.

➜  spark git:(master) ✗ ./bin/spark-submit --help
Usage: spark-submit [options] <app jar | python file> [app arguments]
Usage: spark-submit --kill [submission ID] --master [spark://...]
Usage: spark-submit --status [submission ID] --master [spark://...]
Usage: spark-submit run-example [options] example-class [example args]

Options:
  --master MASTER_URL         spark://host:port, mesos://host:port, yarn, or local.
  --deploy-mode DEPLOY_MODE   Whether to launch the driver program locally ("client") or
                              on one of the worker machines inside the cluster ("cluster")
                              (Default: client).
  --class CLASS_NAME          Your application's main class (for Java / Scala apps).
  --name NAME                 A name of your application.
  --jars JARS                 Comma-separated list of local jars to include on the driver
                              and executor classpaths.
  --packages                  Comma-separated list of maven coordinates of jars to include
                              on the driver and executor classpaths. Will search the local
                              maven repo, then maven central and any additional remote
                              repositories given by --repositories. The format for the
                              coordinates should be groupId:artifactId:version.
  --exclude-packages          Comma-separated list of groupId:artifactId, to exclude while
                              resolving the dependencies provided in --packages to avoid
                              dependency conflicts.
  --repositories              Comma-separated list of additional remote repositories to
                              search for the maven coordinates given with --packages.
  --py-files PY_FILES         Comma-separated list of .zip, .egg, or .py files to place
                              on the PYTHONPATH for Python apps.
  --files FILES               Comma-separated list of files to be placed in the working
                              directory of each executor.

  --conf PROP=VALUE           Arbitrary Spark configuration property.
  --properties-file FILE      Path to a file from which to load extra properties. If not
                              specified, this will look for conf/spark-defaults.conf.

  --driver-memory MEM         Memory for driver (e.g. 1000M, 2G) (Default: 1024M).
  --driver-java-options       Extra Java options to pass to the driver.
  --driver-library-path       Extra library path entries to pass to the driver.
  --driver-class-path         Extra class path entries to pass to the driver. Note that
                              jars added with --jars are automatically included in the
                              classpath.

  --executor-memory MEM       Memory per executor (e.g. 1000M, 2G) (Default: 1G).

  --proxy-user NAME           User to impersonate when submitting the application.
                              This argument does not work with --principal / --keytab.

  --help, -h                  Show this help message and exit.
  --verbose, -v               Print additional debug output.
  --version,                  Print the version of current Spark.

 Spark standalone with cluster deploy mode only:
  --driver-cores NUM          Cores for driver (Default: 1).

 Spark standalone or Mesos with cluster deploy mode only:
  --supervise                 If given, restarts the driver on failure.
  --kill SUBMISSION_ID        If given, kills the driver specified.
  --status SUBMISSION_ID      If given, requests the status of the driver specified.

 Spark standalone and Mesos only:
  --total-executor-cores NUM  Total cores for all executors.

 Spark standalone and YARN only:
  --executor-cores NUM        Number of cores per executor. (Default: 1 in YARN mode,
                              or all available cores on the worker in standalone mode)

 YARN-only:
  --driver-cores NUM          Number of cores used by the driver, only in cluster mode
                              (Default: 1).
  --queue QUEUE_NAME          The YARN queue to submit to (Default: "default").
  --num-executors NUM         Number of executors to launch (Default: 2).
  --archives ARCHIVES         Comma separated list of archives to be extracted into the
                              working directory of each executor.
  --principal PRINCIPAL       Principal to be used to login to KDC, while running on
                              secure HDFS.
  --keytab KEYTAB             The full path to the file that contains the keytab for the
                              principal specified above. This keytab will be copied to
                              the node running the Application Master via the Secure
                              Distributed Cache, for renewing the login tickets and the
                              delegation tokens periodically.

--class
--conf or -c
--deploy-mode (see Deploy Mode)
--driver-class-path (see --driver-class-path command-line option)
--driver-cores (see Driver Cores in Cluster Deploy Mode)
--driver-java-options
--driver-library-path
--driver-memory
--executor-memory
--files
--jars
--kill for Standalone cluster mode only
--master
--name
--packages
--exclude-packages
--properties-file (see Custom Spark Properties File)
--proxy-user
--py-files
--repositories
--status for Standalone cluster mode only
--total-executor-cores

List of switches, i.e. command-line options that do not take parameters:

--help or -h
--supervise for Standalone cluster mode only
--usage-error
--verbose or -v (see Verbose Mode)
--version (see Version)

YARN-only options:

--archives
--executor-cores
--keytab
--num-executors
--principal
--queue (see Specifying YARN Resource Queue (--queue switch))

`--driver-class-path` command-line option

--driver-class-path command-line option sets the extra class path entries (e.g. jars and directories) that should be added to a driver’s JVM.

Tip	You should use `--driver-class-path` in `client` deploy mode (not SparkConf) to ensure that the CLASSPATH is set up with the entries. `client` deploy mode uses the same JVM for the driver as `spark-submit`'s.

--driver-class-path sets the internal driverExtraClassPath property (when SparkSubmitArguments.handle called).

It works for all cluster managers and deploy modes.

If driverExtraClassPath not set on command-line, the spark.driver.extraClassPath setting is used.

Note	Command-line options (e.g. `--driver-class-path`) have higher precedence than their corresponding Spark settings in a Spark properties file (e.g. `spark.driver.extraClassPath`). You can therefore control the final settings by overriding Spark settings on command line using the command-line options.

Table 3. Spark Settings in Spark Properties File and on Command Line
Setting / System Property	Command-Line Option	Description
spark.driver.extraClassPath	`--driver-class-path`	Extra class path entries (e.g. jars and directories) to pass to a driver’s JVM.

Version — `--version` command-line option

$ ./bin/spark-submit --version
Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _\ \/ _ \/ _ `/ __/  '_/
   /___/ .__/\_,_/_/ /_/\_\   version 2.1.0-SNAPSHOT
      /_/

Branch master
Compiled by user jacek on 2016-09-30T07:08:39Z
Revision 1fad5596885aab8b32d2307c0edecbae50d5bd7a
Url https://github.com/apache/spark.git
Type --help for more information.

Verbose Mode — `--verbose` command-line option

When spark-submit is executed with --verbose command-line option, it enters verbose mode.

In verbose mode, the parsed arguments are printed out to the System error output.

FIXME

It also prints out propertiesFile and the properties from the file.

FIXME

Deploy Mode — `--deploy-mode` command-line option

You use spark-submit’s --deploy-mode command-line option to specify the deploy mode for a Spark application.

Environment Variables

The following is the list of environment variables that are considered when command-line options are not specified:

MASTER for --master
SPARK_DRIVER_MEMORY for --driver-memory
SPARK_EXECUTOR_MEMORY (see Environment Variables in the SparkContext document)
SPARK_EXECUTOR_CORES
DEPLOY_MODE
SPARK_YARN_APP_NAME
_SPARK_CMD_USAGE

External packages and custom repositories

The spark-submit utility supports specifying external packages using Maven coordinates using --packages and custom repositories using --repositories.

./bin/spark-submit \
  --packages my:awesome:package \
  --repositories s3n://$aws_ak:$aws_sak@bucket/path/to/repo

FIXME Why should I care?

`SparkSubmit` Standalone Application — `main` method

Tip	The source code of the script lives in https://github.com/apache/spark/blob/master/bin/spark-submit.

When executed, spark-submit script simply passes the call to spark-class with org.apache.spark.deploy.SparkSubmit class followed by command-line arguments.

Tip	`spark-class` uses the class name — `org.apache.spark.deploy.SparkSubmit` — to parse command-line arguments appropriately. Refer to `org.apache.spark.launcher.Main` Standalone Application

It creates an instance of SparkSubmitArguments.

If in verbose mode, it prints out the application arguments.

It then relays the execution to action-specific internal methods (with the application arguments):

When no action was explicitly given, it is assumed submit action.
kill (when --kill switch is used)
requestStatus (when --status switch is used)

Note	The action can only have one of the three available values: `SUBMIT`, `KILL`, or `REQUEST_STATUS`.

spark-env.sh - load additional environment settings

spark-env.sh consists of environment settings to configure Spark for your site.

export JAVA_HOME=/your/directory/java
export HADOOP_HOME=/usr/lib/hadoop
export SPARK_WORKER_CORES=2
export SPARK_WORKER_MEMORY=1G

spark-env.sh is loaded at the startup of Spark’s command line scripts.
SPARK_ENV_LOADED env var is to ensure the spark-env.sh script is loaded once.
SPARK_CONF_DIR points at the directory with spark-env.sh or $SPARK_HOME/conf is used.
spark-env.sh is executed if it exists.
$SPARK_HOME/conf directory has spark-env.sh.template file that serves as a template for your own custom configuration.

Consult Environment Variables in the official documentation.

Spark Submit — spark-submit shell script