机器规划
本次实战用到了三台CentOS7的机器,身份信息如下所示:
IP地址 | hostname(主机名) | 身份 |
---|---|---|
192.168.119.163 | node0 | NameNode、ResourceManager、HistoryServer、Master |
192.168.119.164 | node1 | DataNode、NodeManager、Worker |
192.168.119.165 | node2 | DataNode、NodeManager、Worker 、SecondaryNameNode |
要注意的地方:
- spark的Master和hdfs的NameNode、Yarn的ResourceManager在同一台机器;
- spark的Worker和hdfs的DataNode、Yarn的NodeManager在同一台机器;
先部署和启动hadoop集群环境
部署spark2.2集群on Yarn模式的前提,是先搭建好hadoop集群环境,请参考《Linux部署hadoop2.7.7集群》一文,将hadoop集群环境部署并启动成功;
部署spark集群
- 本次实战的部署方式,是先部署standalone模式的spark集群,再做少量配置修改,即可改为on Yarn模式;
- standalone模式的spark集群部署,请参考《部署spark2.2集群(standalone模式)》一文,要注意的是spark集群的master和hadoop集群的NameNode是同一台机器,worker和DataNode在是同一台机器,并且建议spark和hadoop部署都用同一个账号来进行;
修改配置
如果您已经完成了hadoop集群和spark集群(standalone模式)的部署,接下来只需要两步设置即可:
- 假设hadoop的文件夹hadoop-2.7.7所在目录为/home/hadoop/,打开spark的spark-env.sh文件,在尾部追加一行:
export HADOOP_CONF_DIR=/home/hadoop/hadoop-2.7.7/etc/hadoop
- 打开hadoop-2.7.7/etc/hadoop/yarn-site.xml文件,在configuration节点中增加下面两个子节点,如果不做以下设置,在提交spark任务的时候,yarn可能将spark任务kill掉,导致"Failed to send RPC xxxxxx"异常:
<property>
<name>yarn.nodemanager.pmem-check-enabled</name>
<value>false</value>
</property>
<property>
<name>yarn.nodemanager.vmem-check-enabled</name>
<value>false</value>
</property>
本次实战一共有三台电脑,请确保在每台电脑上都做了上述配置;
启动hadoop和spark
hadoop和spark都部署在当前账号的家目录下,因此启动命令和顺序如下:
代码语言:javascript复制~/hadoop-2.7.7/sbin/start-dfs.sh
&& ~/hadoop-2.7.7/sbin/start-yarn.sh
&& ~/hadoop-2.7.7/sbin/mr-jobhistory-daemon.sh start historyserver
&& ~/spark-2.3.2-bin-hadoop2.7/sbin/start-all.sh
验证spark
- 在hdfs创建一个目录用于保存输入文件:
~/hadoop-2.7.7/bin/hdfs dfs -mkdir /input
- 准备一个txt文件(我这里是GoneWiththeWind.txt),提交到hdfs的/input目录下:
~/hadoop-2.7.7/bin/hdfs dfs -put ~/GoneWiththeWind.txt /input
- 以client模式启动spark-shell
~/spark-2.3.2-bin-hadoop2.7/bin/spark-shell --master yarn --deploy-mode client
以下信息表示启动成功:
代码语言:javascript复制2019-02-09 10:13:09 WARN NativeCodeLoader:62 - Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Setting default log level to "WARN".
To adjust logging level use sc.setLogLevel(newLevel). For SparkR, use setLogLevel(newLevel).
2019-02-09 10:13:15 WARN Client:66 - Neither spark.yarn.jars nor spark.yarn.archive is set, falling back to uploading libraries under SPARK_HOME.
Spark context Web UI available at http://node0:4040
Spark context available as 'sc' (master = yarn, app id = application_1549678248927_0001).
Spark session available as 'spark'.
Welcome to
____ __
/ __/__ ___ _____/ /__
_ / _ / _ `/ __/ '_/
/___/ .__/_,_/_/ /_/_ version 2.3.2
/_/
Using Scala version 2.11.8 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_191)
Type in expressions to have them evaluated.
Type :help for more information.
scala>
- 输入以下内容,即可统计之前提交的txt文件中的单词出现次数,然后将前十名打印出来:
sc.textFile("hdfs://node0:8020/input/GoneWiththeWind.txt").flatMap(line => line.split(" ")).map(word => (word, 1)).reduceByKey(_ _).sortBy(_._2,false).take(10).foreach(println)
控制台输出如下,可见任务执行成功:
代码语言:javascript复制scala> sc.textFile("hdfs://node0:8020/input/GoneWiththeWind.txt").flatMap(line => line.split(" ")).map(word => (word, 1)).reduceByKey(_ _).sortBy(_._2,false).take(10).foreach(println)
(the,18264)
(and,14150)
(to,10020)
(of,8615)
(a,7571)
(her,7086)
(she,6217)
(was,5912)
(in,5751)
(had,4502)
- 在网页上查看yarn信息,如下图:
java版本的任务提交
如果您的开发语言是java,请将应用编译构建为jar包,然后执行以下命令,就会以client模式提交任务到yarn:
代码语言:javascript复制~/spark-2.3.2-bin-hadoop2.7/bin/spark-submit
--master yarn
--deploy-mode client
--class com.bolingcavalry.sparkwordcount.WordCount
--executor-memory 512m
--total-executor-cores 2
~/jars/sparkwordcount-1.0-SNAPSHOT.jar
192.168.119.163
8020
GoneWiththeWind.txt
上述命令的最后三个参数是WorkCount类运行时需要用到的参数,该应用的详情请参考《第一个spark应用开发详解(java版)》;
停止hadoop和spark
如果需要停止hadoop和spark服务,命令和顺序如下:
代码语言:javascript复制~/spark-2.3.2-bin-hadoop2.7/sbin/stop-all.sh
&& ~/hadoop-2.7.7/sbin/mr-jobhistory-daemon.sh stop historyserver
&& ~/hadoop-2.7.7/sbin/stop-yarn.sh
&& ~/hadoop-2.7.7/sbin/stop-dfs.sh
至此,Spark on Yarn模式的集群部署和验证已经完成,希望能够带给您一些参考;