Spark on Yarn Client和Cluster模式详解

Spark在YARN中有yarn-cluster和yarn-client两种运行模式:

I. Yarn Cluster

Spark Driver首先作为一个ApplicationMaster在YARN集群中启动，客户端提交给ResourceManager的每一个job都会在集群的worker节点上分配一个唯一的ApplicationMaster，由该ApplicationMaster管理全生命周期的应用。因为Driver程序在YARN中运行，所以事先不用启动Spark Master/Client，应用的运行结果不能在客户端显示（可以在history server中查看），所以最好将结果保存在HDFS而非stdout输出，客户端的终端显示的是作为YARN的job的简单运行状况。

by @Sandy Ryza

by 明风@taobao
从terminal的output中看到任务初始化更详细的四个步骤：

1. 14/09/28 11:24:52 INFO RMProxy: Connecting to ResourceManager at hdp01/172.19.1.231:8032
   
2. 14/09/28 11:24:52 INFO Client: Got Cluster metric info from ApplicationsManager (ASM), number of NodeManagers: 4
   
3. 14/09/28 11:24:52 INFO Client: Queue info ... queueName: root.default, queueCurrentCapacity: 0.0, queueMaxCapacity: -1.0,
   
4. queueApplicationCount = 0, queueChildQueueCount = 0
   
5. 14/09/28 11:24:52 INFO Client: Max mem capabililty of a single resource in this cluster 8192
   
6. 14/09/28 11:24:53 INFO Client: Uploading file:/usr/lib/spark/examples/lib/spark-examples_2.10-1.0.0-cdh5.1.0.jar to hdfs://hdp01:8020/user/spark/.sparkStaging/application_1411874193696_0003/spark-examples_2.10-1.0.0-cdh5.1.0.jar
   
7. 14/09/28 11:24:54 INFO Client: Uploading file:/usr/lib/spark/assembly/lib/spark-assembly-1.0.0-cdh5.1.0-hadoop2.3.0-cdh5.1.0.jar to hdfs://hdp01:8020/user/spark/.sparkStaging/application_1411874193696_0003/spark-assembly-1.0.0-cdh5.1.0-hadoop2.3.0-cdh5.1.0.jar
   
8. 14/09/28 11:24:55 INFO Client: Setting up the launch environment
   
9. 14/09/28 11:24:55 INFO Client: Setting up container launch context
   
10. 14/09/28 11:24:55 INFO Client: Command for starting the Spark ApplicationMaster: List($JAVA_HOME/bin/java, -server, -Xmx512m, -Djava.io.tmpdir=$PWD/tmp, -Dspark.master="spark://hdp01:7077", -Dspark.app.name="org.apache.spark.examples.SparkPi", -Dspark.eventLog.enabled="true", -Dspark.eventLog.dir="/user/spark/applicationHistory", -Dlog4j.configuration=log4j-spark-container.properties, org.apache.spark.deploy.yarn.ApplicationMaster, --class, org.apache.spark.examples.SparkPi, --jar , file:/usr/lib/spark/examples/lib/spark-examples_2.10-1.0.0-cdh5.1.0.jar, , --executor-memory, 1024, --executor-cores, 1, --num-executors , 2, 1>, <LOG_DIR>/stdout, 2>, <LOG_DIR>/stderr)
    
11. 14/09/28 11:24:55 INFO Client: Submitting application to ASM
    
12. 14/09/28 11:24:55 INFO YarnClientImpl: Submitted application application_1411874193696_0003
    
13. 14/09/28 11:24:56 INFO Client: Application report from ASM:
    
14. application identifier: application_1411874193696_0003
    
15. appId: 3
    
16. clientToAMToken: null
    
17. appDiagnostics:
    
18. appMasterHost: N/A
    
19. appQueue: root.spark
    
20. appMasterRpcPort: -1
    
21. appStartTime: 1411874695327
    
22. yarnAppState: ACCEPTED
    
23. distributedFinalState: UNDEFINED
    
24. appTrackingUrl: http://hdp01:8088/proxy/application_1411874193696_0003/
    
25. appUser: spark

复制代码

1. 由client向ResourceManager提交请求，并上传jar到HDFS上
这期间包括四个步骤：
a). 连接到RM
b). 从RM ASM（ApplicationsManager ）中获得metric、queue和resource等信息。
c). upload app jar and spark-assembly jar
d). 设置运行环境和container上下文（launch-container.sh等脚本)

2. ResouceManager向NodeManager申请资源，创建Spark ApplicationMaster（每个SparkContext都有一个ApplicationMaster）
3. NodeManager启动Spark App Master，并向ResourceManager AsM注册
4. Spark ApplicationMaster从HDFS中找到jar文件，启动DAGscheduler和YARN Cluster Scheduler
5. ResourceManager向ResourceManager AsM注册申请container资源（INFO YarnClientImpl: Submitted application）
6. ResourceManager通知NodeManager分配Container，这时可以收到来自ASM关于container的报告。（每个container的对应一个executor）
7. Spark ApplicationMaster直接和container（executor）进行交互，完成这个分布式任务。
需要注意的是：
a). Spark中的localdir会被yarn.nodemanager.local-dirs替换
b). 允许失败的节点数(spark.yarn.max.worker.failures)为executor数量的两倍数量，最小为3.
c). SPARK_YARN_USER_ENV传递给spark进程的环境变量
d). 传递给app的参数应该通过–args指定。

部署：
环境介绍：
hdp0[1-4]四台主机
hadoop使用CDH 5.1版本: hadoop-2.3.0+cdh5.1.0+795-1.cdh5.1.0.p0.58.el6.x86_64
直接下载对应2.3.0的pre-build版本http://spark.apache.org/downloads.html
下载完毕后解压，检查spark-assembly目录：

1. file /home/spark/spark-1.1.0-bin-hadoop2.3/lib/spark-assembly-1.1.0-hadoop2.3.0.jar
   
2. /home/spark/spark-1.1.0-bin-hadoop2.3/lib/spark-assembly-1.1.0-hadoop2.3.0.jar: Zip archive data, at least v2.0 to extract

复制代码

然后输出环境变量HADOOP_CONF_DIR/YARN_CONF_DIR和SPARK_JAR(可以设置到spark-env.sh中）

1. export HADOOP_CONF_DIR=/etc/hadoop/etc
   
2. export SPARK_JAR=/home/spark/spark-1.1.0-bin-hadoop2.3/lib/spark-assembly-1.1.0-hadoop2.3.0.jar

复制代码

如果使用cloudera manager 5，在Spark Service的操作中可以找到Upload Spark Jar将spark-assembly上传到HDFS上。

Spark Jar Location (HDFS) 
spark_jar_hdfs_path

/user/spark/share/lib/spark-assembly.jar

默认值

The location of the Spark jar in HDFS

Spark History Location (HDFS) 
spark.eventLog.dir

/user/spark/applicationHistory

默认值

The location of Spark application history logs in HDFS. Changing this value will not move existing logs to the new location.

提交任务，此时在YARN的web UI和history Server上就可以看到运行状态信息。

1. spark-submit --class org.apache.spark.examples.SparkPi --master yarn-cluster /usr/lib/spark/examples/lib/spark-examples_2.10-1.0.0-cdh5.1.0.jar
   

复制代码

II. yarn-client

(YarnClientClusterScheduler)查看对应类的文件
在yarn-client模式下，Driver运行在Client上，通过ApplicationMaster向RM获取资源。本地Driver负责与所有的executor container进行交互，并将最后的结果汇总。结束掉终端，相当于kill掉这个spark应用。一般来说，如果运行的结果仅仅返回到terminal上时需要配置这个。

客户端的Driver将应用提交给Yarn后，Yarn会先后启动ApplicationMaster和executor，另外ApplicationMaster和executor都 是装载在container里运行，container默认的内存是1G，ApplicationMaster分配的内存是driver- memory，executor分配的内存是executor-memory。同时，因为Driver在客户端，所以程序的运行结果可以在客户端显 示，Driver以进程名为SparkSubmit的形式存在。
配置YARN-Client模式同样需要HADOOP_CONF_DIR/YARN_CONF_DIR和SPARK_JAR变量。
提交任务测试：

1. spark-submit --class org.apache.spark.examples.SparkPi --deploy-mode client /usr/lib/spark/examples/lib/spark-examples_2.10-1.0.0-cdh5.1.0.jar
   
2. terminal output：
   
3. 
   
4. 14/09/28 11:18:34 INFO Client: Command for starting the Spark ApplicationMaster: List($JAVA_HOME/bin/java, -server, -Xmx512m, -Djava.io.tmpdir=$PWD/tmp, -Dspark.tachyonStore.folderName="spark-9287f0f2-2e72-4617-a418-e0198626829b", -Dspark.eventLog.enabled="true", -Dspark.yarn.secondary.jars="", -Dspark.driver.host="hdp01", -Dspark.driver.appUIHistoryAddress="", -Dspark.app.name="Spark Pi", -Dspark.jars="file:/usr/lib/spark/examples/lib/spark-examples_2.10-1.0.0-cdh5.1.0.jar", -Dspark.fileserver.uri="http://172.19.17.231:53558", -Dspark.eventLog.dir="/user/spark/applicationHistory", -Dspark.master="yarn-client", -Dspark.driver.port="35938", -Dspark.httpBroadcast.uri="http://172.19.17.231:43804", -Dlog4j.configuration=log4j-spark-container.properties, org.apache.spark.deploy.yarn.ExecutorLauncher, --class, notused, --jar , null, --args 'hdp01:35938' , --executor-memory, 1024, --executor-cores, 1, --num-executors , 2, 1>, <LOG_DIR>/stdout, 2>, <LOG_DIR>/stderr)
   
5. 14/09/28 11:18:34 INFO Client: Submitting application to ASM
   
6. 14/09/28 11:18:34 INFO YarnClientSchedulerBackend: Application report from ASM:
   
7. appMasterRpcPort: -1
   
8. appStartTime: 1411874314198
   
9. yarnAppState: ACCEPTED
   
10. ......

复制代码

##最后将结果输出到terminal中