If you want to test out the YARN deployment mode, you can use the current Spark examples. A `spark-examples_{{site.SCALA_BINARY_VERSION}}-{{site.SPARK_VERSION}}` file can be generated by running `sbt/sbt assembly`. NOTE: since the documentation you're reading is for Spark version {{site.SPARK_VERSION}}, we are assuming here that you have downloaded Spark {{site.SPARK_VERSION}} or checked it out of source control. If you are using a different version of Spark, the version numbers in the jar generated by the sbt package command will obviously be different.
Most of the configs are the same for Spark on YARN as for other deployment modes. See the Configuration page for more information on those. These are configs that are specific to Spark on YARN.
*`SPARK_YARN_USER_ENV`, to add environment variables to the Spark processes launched on YARN. This can be a comma separated list of environment variables, e.g. `SPARK_YARN_USER_ENV="JAVA_HOME=/jdk64,FOO=bar"`.
*`spark.yarn.applicationMaster.waitTries`, property to set the number of times the ApplicationMaster waits for the the spark master and then also the number of tries it waits for the Spark Context to be intialized. Default is 10.
*`spark.yarn.submit.file.replication`, the HDFS replication level for the files uploaded into HDFS for the application. These include things like the spark jar, the app jar, and any distributed cache files/archives.
*`spark.yarn.preserve.staging.files`, set to true to preserve the staged files(spark jar, app jar, distributed cache files) at the end of the job rather then delete them.
*`spark.yarn.scheduler.heartbeat.interval-ms`, the interval in ms in which the Spark application master heartbeats into the YARN ResourceManager. Default is 5 seconds.
*`spark.yarn.max.executor.failures`, the maximum number of executor failures before failing the application. Default is the number of executors requested times 2 with minimum of 3.
There are two deploy modes that can be used to launch Spark applications on YARN. In yarn-cluster mode, the Spark driver runs inside an application master process which is managed by YARN on the cluster, and the client can go away after initiating the application. In yarn-client mode, the driver runs in the client process, and the application master is only used for requesting resources from YARN.
Unlike in Spark standalone and Mesos mode, in which the master's address is specified in the "master" parameter, in YARN mode the ResourceManager's address is picked up from the Hadoop configuration. Thus, the master parameter is simply "yarn-client" or "yarn-cluster".
The spark-submit script described in the [cluster mode overview](cluster-overview.html) provides the most straightforward way to submit a compiled Spark application to YARN in either deploy mode. For info on the lower-level invocations it uses, read ahead. For running spark-shell against YARN, skip down to the yarn-client section.
The above starts a YARN client program which starts the default Application Master. Then SparkPi will be run as a child thread of Application Master. The client will periodically poll the Application Master for status updates and display them in the console. The client will exit once your application has finished running. Refer to the "Viewing Logs" section below for how to see driver and executor logs.
Because the application is run on a remote machine where the Application Master is running, applications that involve local interaction, such as spark-shell, will not work.
With yarn-client mode, the application will be launched locally, just like running an application or spark-shell on Local / Mesos / Standalone client mode. The launch method is also the same, just make sure to specify the master URL as "yarn-client". You also need to export the env value for SPARK_JAR.
In order to tune executor cores/number/memory etc., you need to export environment variables or add them to the spark configuration file (./conf/spark_env.sh). The following are the list of options.
In YARN terminology, executors and application masters run inside "containers". YARN has two modes for handling container logs after an application has completed. If log aggregation is turned on (with the yarn.log-aggregation-enable config), container logs are copied to HDFS and deleted on the local machine. These logs can be viewed from anywhere on the cluster with the "yarn logs" command.
yarn logs -applicationId <appID>
will print out the contents of all log files from all containers from the given application.
When log aggregation isn't turned on, logs are retained locally on each machine under YARN_APP_LOGS_DIR, which is usually configured to /tmp/logs or $HADOOP_HOME/logs/userlogs depending on the Hadoop version and installation. Viewing logs for a container requires going to the host that contains them and looking in this directory. Subdirectories organize log files by application ID and container ID.
- Before Hadoop 2.2, YARN does not support cores in container resource requests. Thus, when running against an earlier version, the numbers of cores given via command line arguments cannot be passed to YARN. Whether core requests are honored in scheduling decisions depends on which scheduler is in use and how it is configured.
- The local directories used by Spark executors will be the local directories configured for YARN (Hadoop YARN config yarn.nodemanager.local-dirs). If the user specifies spark.local.dir, it will be ignored.
- The --files and --archives options support specifying file names with the # similar to Hadoop. For example you can specify: --files localtest.txt#appSees.txt and this will upload the file you have locally named localtest.txt into HDFS but this will be linked to by the name appSees.txt, and your application should use the name as appSees.txt to reference it when running on YARN.
- The --addJars option allows the SparkContext.addJar function to work if you are using it with local files. It does not need to be used if you are using it with HDFS, HTTP, HTTPS, or FTP files.