Merge pull request #101 from colorant/yarn-client-scheduler

For SPARK-527, Support spark-shell when running on YARN

sync to trunk and resubmit here

In current YARN mode approaching, the application is run in the Application Master as a user program thus the whole spark context is on remote.

This approaching won't support application that involve local interaction and need to be run on where it is launched.

So In this pull request I have a YarnClientClusterScheduler and backend added.

With this scheduler, the user application is launched locally,While the executor will be launched by YARN on remote nodes with a thin AM which only launch the executor and monitor the Driver Actor status, so that when client app is done, it can finish the YARN Application as well.

This enables spark-shell to run upon YARN.

This also enable other Spark applications to have the spark context to run locally with a master-url "yarn-client". Thus e.g. SparkPi could have the result output locally on console instead of output in the log of the remote machine where AM is running on.

Docs also updated to show how to use this yarn-client mode.

1 files changed, 25 insertions, 2 deletions

diff --git a/docs/running-on-yarn.md b/docs/running-on-yarn.md
index 4056e9c15d..68fd6c2ab1 100644
--- a/docs/running-on-yarn.md
+++ b/docs/running-on-yarn.md
@@ -45,6 +45,10 @@ System Properties:
 Ensure that HADOOP_CONF_DIR or YARN_CONF_DIR points to the directory which contains the (client side) configuration files for the hadoop cluster.
 This would be used to connect to the cluster, write to the dfs and submit jobs to the resource manager.
 
+There are two scheduler mode that can be used to launch spark application on YARN.
+
+## Launch spark application by YARN Client with yarn-standalone mode.
+
 The command to launch the YARN Client is as follows:
 
     SPARK_JAR=<SPARK_ASSEMBLY_JAR_FILE> ./spark-class org.apache.spark.deploy.yarn.Client \
@@ -52,6 +56,7 @@ The command to launch the YARN Client is as follows:
       --class <APP_MAIN_CLASS> \
       --args <APP_MAIN_ARGUMENTS> \
       --num-workers <NUMBER_OF_WORKER_MACHINES> \
+      --master-class <ApplicationMaster_CLASS>
       --master-memory <MEMORY_FOR_MASTER> \
       --worker-memory <MEMORY_PER_WORKER> \
       --worker-cores <CORES_PER_WORKER> \
@@ -85,11 +90,29 @@ For example:
     $ cat $YARN_APP_LOGS_DIR/$YARN_APP_ID/container*_000001/stdout
     Pi is roughly 3.13794
 
-The above starts a YARN Client programs which periodically polls the Application Master for status updates and displays them in the console. The client will exit once your application has finished running.
+The above starts a YARN Client programs which start the default Application Master. Then SparkPi will be run as a child thread of Application Master, YARN Client will  periodically polls the Application Master for status updates and displays them in the console. The client will exit once your application has finished running.
+
+With this mode, your application is actually run on the remote machine where the Application Master is run upon. Thus application that involve local interaction will not work well, e.g. spark-shell.
+
+## Launch spark application with yarn-client mode.
+
+With yarn-client mode, the application will be launched locally. Just like running application or spark-shell on Local / Mesos / Standalone mode. The launch method is also the similar with them, just make sure that when you need to specify a master url, use "yarn-client" instead. And you also need to export the env value for SPARK_JAR and SPARK_YARN_APP_JAR
+
+In order to tune worker core/number/memory etc. You need to export SPARK_WORKER_CORES, SPARK_WORKER_MEMORY, SPARK_WORKER_INSTANCES e.g. by ./conf/spark-env.sh
+
+For example:
+
+    SPARK_JAR=./assembly/target/scala-{{site.SCALA_VERSION}}/spark-assembly-{{site.SPARK_VERSION}}-hadoop2.0.5-alpha.jar \
+    SPARK_YARN_APP_JAR=examples/target/scala-{{site.SCALA_VERSION}}/spark-examples-assembly-{{site.SPARK_VERSION}}.jar \
+    ./run-example org.apache.spark.examples.SparkPi yarn-client
+
+
+    SPARK_JAR=./assembly/target/scala-{{site.SCALA_VERSION}}/spark-assembly-{{site.SPARK_VERSION}}-hadoop2.0.5-alpha.jar \
+    SPARK_YARN_APP_JAR=examples/target/scala-{{site.SCALA_VERSION}}/spark-examples-assembly-{{site.SPARK_VERSION}}.jar \
+    MASTER=yarn-client ./spark-shell
 
 # Important Notes
 
-- When your application instantiates a Spark context it must use a special "yarn-standalone" master url. This starts the scheduler without forcing it to connect to a cluster. A good way to handle this is to pass "yarn-standalone" as an argument to your program, as shown in the example above.
 - We do not requesting container resources based on the number of cores. Thus the numbers of cores given via command line arguments cannot be guaranteed.
 - The local directories used for spark 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.