Oozie 提交Spark On Yarn示例

大家好，我是crazy_老中医，我写程序就像老中医一样，全屏感觉和经验，但是有用！

废话不多说，现在开始正文，本文将阐述如何将一个Spark程序通过oozie提交到hadoop的Yarn上运行。

准备工作

集群规划

	hdp-master	hdp-slave1	hdp-slave2
hadoop	NameNode DataNode SecondaryNameNode ResourceManager NodeManager	DataNode NodeManager	DataNode NodeManager
oozie	Y	N	N
spark	Master Worker	Worker	Worker

hadoop2.6.0

安装路径：
/opt/hadoop-2.6.0
环境变量：
export HADOOP_HOME=/opt/hadoop-2.6.0
export PATH=$PATH:$HADOOP_HOME/bin
export HADOOP_CONF_DIR=/opt/hadoop-2.6.0/etc/hadoop
启动hadoop：
$:cd /opt/hadoop-2.6.0
$:./sbin/start-dfs.sh
$:./sbin/start-yarn.sh
启动成功：
通过jps可看到集群规划中的相关进程。

oozie4.2.0

安装路径：
/opt/oozie-4.2.0
环境变量：
export OOZIE_HOME=/opt/oozie-4.2.0

export PATH=$PATH:$OOZIE_HOME/bin

启动oozie：
$:cd /opt/oozie-4.2.0
$:./bin/oozied.sh start

spark1.4.1

安装路径：
/opt/spark-1.4.1-bin-hadoop2.6

环境变量：
export SPARK_HOME=/opt/spark-1.4.1-bin-hadoop2.6

export PATH=$PATH:$SPARK_HOME/bin

启动spark：
$:cd /opt/spark-1.4.1-bin-hadoop2.6
$:./sbin/start-all.sh

启动成功：
通过jps可看到集群规划中的相关进程。

主节点jps截图

分节点jps截图

编写Spark程序

目的
这里用的是tomcat产生的access日志文件，日志格式如下：

118.118.118.30 - - [15/Jun/2016:18:34:19 +0800] "GET /WebAnalytics/js/ds.js?v=3 HTTP/1.1" 304 -
118.118.118.30 - - [15/Jun/2016:18:34:19 +0800] "GET /WebAnalytics/maindomain? HTTP/1.1" 200 3198
118.118.118.30 - - [15/Jun/2016:18:34:19 +0800] "GET /WebAnalytics/tracker/1.0/tpv? HTTP/1.1" 200 631
118.118.118.30 - - [15/Jun/2016:18:34:22 +0800] "GET /WebAnalytics/tracker/1.0/bind? HTTP/1.1" 200 631
118.118.118.30 - - [15/Jun/2016:18:34:46 +0800] "GET /WebAnalytics/js/ds.js?v=3 HTTP/1.1" 304 -

对日志中ip进行统计，统计结果如下图：



详细代码

package com.simple.spark.oozie.action;

import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.Function2;
import org.apache.spark.api.java.function.PairFunction;
import scala.Tuple2;

/**
* Created by Administrator on 2016/8/25.
*/
public class OozieAction {
public static void main(String[] args){
String input = args[0];
String output = args[1];
String master = "local[*]";
if(args.length >= 3){
master = args[2];
}

SparkConf sparkConf = new SparkConf();
sparkConf.setAppName("OozieAction - " + System.currentTimeMillis());
if(!"none".equals(master))
sparkConf.setMaster(master);
JavaSparkContext context = new JavaSparkContext(sparkConf);
JavaRDD<String> stringJavaRDD = context.textFile(input);
stringJavaRDD.mapToPair(new PairFunction<String, String, Integer>() {
public Tuple2<String, Integer> call(String s) throws Exception {
String key = s.split(" ")[0];
return new Tuple2<String, Integer>(key, 1);
}
}).reduceByKey(new Function2<Integer, Integer, Integer>() {
public Integer call(Integer v1, Integer v2) throws Exception {
return v1 + v2;
}
}).saveAsTextFile(output);
context.close();
}
}

Oozie工作流

示例工作流只有一个spark节点，节点工作流配置参照：http://oozie.apache.org/docs/4.2.0/DG_SparkActionExtension.html#Spark_on_YARN
workflow.xml

<workflow-app name="Spark_Workflow" xmlns="uri:oozie:workflow:0.5">
<start to="spark-SparkOozieAction"/>
<kill name="Kill">
<message>Action failed, error message[${wf:errorMessage(wf:lastErrorNode())}]</message>
</kill>
<action name="spark-SparkOozieAction">
<spark xmlns="uri:oozie:spark-action:0.1">
<job-tracker>${jobTracker}</job-tracker>
<name-node>${nameNode}</name-node>
<master>${jobmaster}</master>
<mode>${jobmode}</mode>
<name>${jobname}</name>
<class>${jarclass}</class>
<jar>${jarpath}</jar>
<spark-opts>${sparkopts}</spark-opts>
<arg>${jararg1}</arg>
<arg>${jararg2}</arg>
<arg>${jararg3}</arg>
</spark>
<ok to="End"/>
<error to="Kill"/>
</action>
<end name="End"/>
</workflow-app>

job.properties

oozie.use.system.libpath=True
oozie.wf.application.path=/user/root/oozie/t1/workflow.xml
security_enabled=False
dryrun=False
jobTracker=hdp-master:8032
nameNode=hdfs://hdp-master:8020
jobmaster=yarn-cluster
jobmode=cluster
jobname=SparkOozieAction
jarclass=com.simple.spark.oozie.action.OozieAction
jarpath=hdfs://hdp-master:8020/user/root/oozie/t1/SparkOozieAction-jar-with-dependencies.jar
sparkopts=--executor-memory 128M --total-executor-cores 2 --driver-memory 256M --conf spark.yarn.jar=hdfs://hdp-master:8020/system/spark/lib/spark-assembly-1.4.1-hadoop2.6.0.jar --conf spark.yarn.historyServer.address=http://hdp-master:18088 --conf spark.eventLog.dir=hdfs://hdp-master:8020/user/spark/applicationHistory --conf spark.eventLog.enabled=true
jararg1=/data/access_log.txt
jararg2=/out/oozie/t1
jararg3=cluster

本文重点：oozie spark on yarn官方文档有说明，要提交spark到yarn上，需要配置以下属性

1.确保spark-assembly-1.4.1-hadoop2.6.0.jar在oozie中可用，这个我理解了很久都没有参透，后面查阅相关文章后才知道，这里的意思是在提交任务时指定要应用的jar包，本文通过spark-opts参数--conf spark.yarn.jar=hdfs://hdp-master:8020/system/spark/lib/spark-assembly-1.4.1-hadoop2.6.0.jar指定。
2.master只能指定为yarn-client或者yarn-cluster，本文为：yarn-cluster

3. spark.yarn.historyServer.address=http://SPH-HOST:18088

4. spark.eventLog.dir=hdfs://NN:8020/user/spark/applicationHistory 这里的目录

必须事先创建好，spark会提示这个目录不存在的错误

5. spark.eventLog.enabled=true

提交workflow.xml到hdfs

$:hadoop fs -put /user/root/oozie/t1/workflow.xml

提交spark引用jar到hdfs

$:hadoop fs -put /user/root/oozie/t1/SparkOozieAction-jar-with-dependencies.jar

创建spark.eventLog.dir目录

hadoop fs -mkdir put /user/spark/applicationHistory

提交程序jar包到hdfs

hadoop fs -put /user/root/oozie/t1/SparkOozieAction-jar-with-dependencies.jar

提交程序到oozie

$:cd /opt/oozie-4.2.0
$:./bin/oozie job -oozie http://hdp-master:11000/oozie -config /opt/myapps/spark/t1/job.properties -run

到oozie UI界面查看任务情况
http://hdp-master:11000/oozie/



到yarn UI查看spark在yarn上的运行情况



至此，spark 通过oozie提交到hadoop yarn上完毕。
谢谢！