Hadoop入门学习笔记——四、MapReduce的框架配置和YARN的部署

视频课程地址：https://www.bilibili.com/video/BV1WY4y197g7
课程资料链接：https://pan.baidu.com/s/15KpnWeKpvExpKmOC8xjmtQ?pwd=5ay8

Hadoop入门学习笔记（汇总）

四、MapReduce的框架配置和YARN的部署

本次YARN的部署结构如下图所示：

当前，共有三台服务器（虚拟机）构成集群，集群规划如下所示：

主机	部署的服务
node1	ResourceManager、NodeManager、ProxyServer、JobHistoryServer
node2	NodeManager
node3	NodeManager

MapReduce是运行在YARN上的，所以MapReduce只需要配置，YARN需要部署并启动。

4.1. 配置MapReduce和YARN

1、在node1节点，修改mapred-env.sh文件：

# 进入hadoop配置文件目录
cd /export/server/hadoop-3.3.4/etc/hadoop/
# 打开mapred-env.sh文件
vim mapred-env.sh

打开后，在文件中加入以下内容：

# 设置JDK路径
export JAVA_HOME=/export/server/jdk
# 设置JobHistoryServer进程的内存为1G
export HADOOP_JOB_HISTORYSERVER_HEAPSIZE=1000
# 设置日志级别为INFO
export HADOOP_MAPRED_ROOT_LOGGER=INFO,RFA

2、再修改同目录下的mapred-site.xml配置文件，在其configuration标签内增加以下内容：

  <property><name>mapreduce.framework.name</name><value>yarn</value><description></description></property><property><name>mapreduce.jobhistory.address</name><value>node1:10020</value><description></description></property><property><name>mapreduce.jobhistory.webapp.address</name><value>node1:19888</value><description></description></property><property><name>mapreduce.jobhistory.intermediate-done-dir</name><value>/data/mr-history/tmp</value><description></description></property><property><name>mapreduce.jobhistory.done-dir</name><value>/data/mr-history/done</value><description></description></property><property><name>yarn.app.mapreduce.am.env</name><value>HADOOP_MAPRED_HOME=$HADOOP_HOME</value></property><property><name>mapreduce.map.env</name><value>HADOOP_MAPRED_HOME=$HADOOP_HOME</value></property><property><name>mapreduce.reduce.env</name><value>HADOOP_MAPRED_HOME=$HADOOP_HOME</value></property>

其中，
mapreduce.framework.name 表示MapReduce的运行框架，这里设置为Yarn；
mapreduce.jobhistory.address 表示历史服务器通讯地址和端口号，这里为node1:10020；
mapreduce.jobhistory.webapp.address 表示历史服务器Web端地址和端口号，这里为node1:19888；
mapreduce.jobhistory.intermediate-done-dir 表示历史信息在HDFS的记录临时路径，这里是/data/mr-history/tmp；
mapreduce.jobhistory.done-dir 表示历史信息在HDFS的记录路径，这里是/data/mr-history/done；
yarn.app.mapreduce.am.env 表示MapReduce HOME的路径，这里设置为HADOOP_HOME相同路径；
mapreduce.map.env 表示Map HOME的路径，这里设置为HADOOP_HOME相同路径；
mapreduce.reduce.env 表示Reduce HOME的路径，这里设置为HADOOP_HOME相同路径；

至此，MapReduce的配置完成。

3、接下来，配置YARN。在node1节点，修改yarn-env.sh文件：

# 进入hadoop配置文件目录
cd /export/server/hadoop-3.3.4/etc/hadoop/
# 打开yarn-env.sh文件
vim yarn-env.sh

在文件中添加以下内容：

# 设置JDK路径的环境变量
export JAVA_HOME=/export/server/jdk
# 设置HADOOP_HOME的环境变量
export HADOOP_HOME=/export/server/hadoop
# 设置配置文件路径的环境变量
export HADOOP_CONF_DIR=$HADOOP_HOME/etc/hadoop
# 设置日志文件路径的环境变量
export HADOOP_LOG_DIR=$HADOOP_HOME/logs

4、修改同目录下的yarn-site.xml配置文件，在其configuration节点中添加以下内容：

  <!-- Site specific YARN configuration properties --><property><name>yarn.log.server.url</name><value>http://node1:19888/jobhistory/logs</value><description></description></property><property><name>yarn.web-proxy.address</name><value>node1:8089</value><description>proxy server hostname and port</description></property><property><name>yarn.log-aggregation-enable</name><value>true</value><description>Configuration to enable or disable log aggregation</description></property><property><name>yarn.nodemanager.remote-app-log-dir</name><value>/tmp/logs</value><description>Configuration to enable or disable log aggregation</description></property><!-- Site specific YARN configuration properties --><property><name>yarn.resourcemanager.hostname</name><value>node1</value><description></description></property><property><name>yarn.resourcemanager.scheduler.class</name><value>org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler</value><description></description></property><property><name>yarn.nodemanager.local-dirs</name><value>/data/nm-local</value><description>Comma-separated list of paths on the local filesystem where intermediate data is written.</description></property><property><name>yarn.nodemanager.log-dirs</name><value>/data/nm-log</value><description>Comma-separated list of paths on the local filesystem where logs are written.</description></property><property><name>yarn.nodemanager.log.retain-seconds</name><value>10800</value><description>Default time (in seconds) to retain log files on the NodeManager Only applicable if log-aggregation is disabled.</description></property><property><name>yarn.nodemanager.aux-services</name><value>mapreduce_shuffle</value><description>Shuffle service that needs to be set for Map Reduce applications.</description></property>

其中，核心配置如下：
yarn.resourcemanager.hostname 表示ResourceManager设置在哪个节点，这里是node1节点；
yarn.nodemanager.local-dirs 表示NodeManager中间数据Linux系统本地存储的路径；
yarn.nodemanager.log-dirs 表示NodeManager数据Linux系统日志本地存储的路径；
yarn.nodemanager.aux-services 表示为MapReduce程序开启Shuffle服务；
额外配置如下：
yarn.log.server.url 表示历史服务器的URL；
yarn.web-proxy.address 表示代理服务器的主机和端口号；
yarn.log-aggregation-enable 表示是否开启日志聚合；
yarn.nodemanager.remote-app-log-dir 表示程序日志在HDFS中的存放路径；
yarn.resourcemanager.scheduler.class 表示选择Yarn使用的调度器，这里选的是公平调度器；

5、完成上述配置后，需要将MapReduce和YARN的配置文件分发到node2和node3服务器相同位置中，使用hadoop用户身份执行以下命令

# 将mapred-env.sh、mapred-site.xml、yarn-env.sh、yarn-site.xml四个配置文件，复制到node2的相同路径下
scp mapred-env.sh mapred-site.xml yarn-env.sh yarn-site.xml node2:`pwd`/
# 将mapred-env.sh、mapred-site.xml、yarn-env.sh、yarn-site.xml四个配置文件，复制到node3的相同路径下
scp mapred-env.sh mapred-site.xml yarn-env.sh yarn-site.xml node2:`pwd`/

4.2. YARN集群启停脚本

在启动YARN集群前，需要确保HDFS集群已经启动。同样，启停YARN集群也必须使用hadoop用户身份。

4.2.1. 一键启停脚本

1. $HADOOP_HOME/sbin/start-yarn.sh或start-yarn.sh 一键启动YARN集群

• 会基于yarn-site.xml中配置的yarn.resourcemanager.hostname来决定在哪台机器上启动resourcemanager；
• 会基于workers文件配置的主机启动NodeManager;
• 在当前机器启动ProxyServer（代理服务器）。
  命令执行效果如下图所示：
  
  此时通过jps命令查看进程，可以看到如下效果：
  
  此时，可以看到ResourceManager、NodeManager和WebAppProxyServer都已经启动，还需要启动HistoryServer，可以通过后续章节介绍的mapred --daemon start historyserver命令启动。
  至此，整个YARN集群启动完成。
  此时，可以通过访问http://node1:8088/ 即可看到YARN集群的监控页面（即ResourceManager的WebUI）
  

2. $HADOOP_HOME/sbin/stop-yarn.sh 或 stop-yarn.sh 一键关闭YARN集群。
3. 配置部署好YARN集群后，可以关闭YARN集群、关闭JobHistoryServer、关闭HDFS集群、关闭虚拟机之后，对虚拟机创建快照，保存好当前环境。

4.2.2. 单独进程启停

1. 在每一台机器，单独启动或停止进程，可以通过如下命令执行：

$HADOOP_HOME/bin/yarn --daemon start|stop resourcemanager|nodemanager|proxyserver

start和stop决定启动和停止；
可控制resourcemanager、nodemanager、webappproxyserver三种进程。
例如：

# 在node1启动ResourceManager
yarn --daemon start resourcemanager
# 在node1、node2、node3分别启动NodeManager
yarn --daemon start nodemanager
# 在node1启动WebProxyServer
yarn --daemon start proxyserver

2. 历史服务器（JobHistoryServer）的启动和停止

$HADOOP_HOME/bin/mapred --daemon start|stop historyserver

用法：

# 启动JobHistoryServer
mapred --daemon start historyserver
# 停止JobHistoryServer
mapred --daemon stop historyserver

4.3. 提交MapReduce示例程序到YARN运行

YARN作为资源调度管控框架，其本身提供资供许多程序运行，常见的有：

• MapReduce程序
• Spark程序
• Flink程序

Hadoop官方提供了一些预置的MapReduce程序代码，存放于$HADOOP_HOME/share/hadoop/mapreduce/hadoop-mapreduce-examples-3.3.1.jar文件内。
上述程序可使用hadoop har命令提交至YARN运行，其命令语法为：

hadoop jar 程序文件 java类名 [程序参数] ... [程序参数]

4.3.1. 提交wordcount（单词统计）示例程序

1、程序内容

• 给定数据输入的路径（HDFS）、给定结果输出的路径（HDFS）
• 将输入路径内的数据中的单词进行计数，将结果写到输出路径

2、准备一份待统计的数据文件并上传至HDFS中
使用vim words.txt命令，在Linux本地创建words.txt文件，其内容如下：

itheima itcast itheima itcast
hadoop hdfs hadoop hdfs
hadoop mapreduce hadoop yarn
itheima hadoop itcast hadoop
itheima itcast hadoop yarn mapreduce

使用命令hdfs dfs -mkdir -p /input在HDFS根目录创建input文件夹（用于存储待统计的文件），使用hdfs dfs -mkdir -p /output命令在HDFS根目录创建output文件夹（用于存储统计结果），使用hdfs dfs -put words.txt /input命令将本地的words.txt文件上传至HDFS系统中。

3、提交MapReduce程序
使用如下命令：

hadoop jar /export/server/hadoop-3.3.4/share/hadoop/mapreduce/hadoop-mapreduce-examples-3.3.4.jar wordcount hdfs://node1:8020/input/ hdfs://8020/output/wc

其中，
hadoop jar 表示向YARN提交一个Java程序；
/export/server/hadoop-3.3.4/share/hadoop/mapreduce/hadoop-mapreduce-examples-3.3.4.jar 表示所要提交的程序路径；
wordcount 表示要运行的java类名；
hdfs://node1:8020/input/ 表示参数1，在本程序中是待统计的文件夹，这里写了hdfs协议头，指明了是HDFS文件系统的路径（经测试，不写也可以，默认读取HDFS文件系统路径）；
hdfs://8020/output/wc 表示参数2，在本程序中是统计结果输出的文件夹，这里写明了hdfs协议头，指明了是HDFS文件系统的路径（经测试，不写也可以，默认读取HDFS文件系统路径），这里需要确保该文件夹不存在，否则会报错。
运行日志如下所示：

[hadoop@node1 ~]$ hadoop jar /export/server/hadoop/share/hadoop/mapreduce/hadoop-mapreduce-examples-3.3.4.jar wordcount hdfs://node1:8020/input hdfs://node1:8020/output/wc
2023-12-14 15:31:53,988 INFO client.DefaultNoHARMFailoverProxyProvider: Connecting to ResourceManager at node1/192.168.88.101:8032
2023-12-14 15:31:55,818 INFO mapreduce.JobResourceUploader: Disabling Erasure Coding for path: /tmp/hadoop-yarn/staging/hadoop/.staging/job_1702538855741_0001
2023-12-14 15:31:56,752 INFO input.FileInputFormat: Total input files to process : 1
2023-12-14 15:31:57,040 INFO mapreduce.JobSubmitter: number of splits:1
2023-12-14 15:31:57,607 INFO mapreduce.JobSubmitter: Submitting tokens for job: job_1702538855741_0001
2023-12-14 15:31:57,607 INFO mapreduce.JobSubmitter: Executing with tokens: []
2023-12-14 15:31:58,167 INFO conf.Configuration: resource-types.xml not found
2023-12-14 15:31:58,170 INFO resource.ResourceUtils: Unable to find 'resource-types.xml'.
2023-12-14 15:31:59,119 INFO impl.YarnClientImpl: Submitted application application_1702538855741_0001
2023-12-14 15:31:59,406 INFO mapreduce.Job: The url to track the job: http://node1:8089/proxy/application_1702538855741_0001/
2023-12-14 15:31:59,407 INFO mapreduce.Job: Running job: job_1702538855741_0001
2023-12-14 15:32:23,043 INFO mapreduce.Job: Job job_1702538855741_0001 running in uber mode : false
2023-12-14 15:32:23,045 INFO mapreduce.Job:  map 0% reduce 0%
2023-12-14 15:32:37,767 INFO mapreduce.Job:  map 100% reduce 0%
2023-12-14 15:32:50,191 INFO mapreduce.Job:  map 100% reduce 100%
2023-12-14 15:32:51,220 INFO mapreduce.Job: Job job_1702538855741_0001 completed successfully
2023-12-14 15:32:51,431 INFO mapreduce.Job: Counters: 54File System CountersFILE: Number of bytes read=84FILE: Number of bytes written=553527FILE: Number of read operations=0FILE: Number of large read operations=0FILE: Number of write operations=0HDFS: Number of bytes read=248HDFS: Number of bytes written=54HDFS: Number of read operations=8HDFS: Number of large read operations=0HDFS: Number of write operations=2HDFS: Number of bytes read erasure-coded=0Job CountersLaunched map tasks=1Launched reduce tasks=1Data-local map tasks=1Total time spent by all maps in occupied slots (ms)=11593Total time spent by all reduces in occupied slots (ms)=9650Total time spent by all map tasks (ms)=11593Total time spent by all reduce tasks (ms)=9650Total vcore-milliseconds taken by all map tasks=11593Total vcore-milliseconds taken by all reduce tasks=9650Total megabyte-milliseconds taken by all map tasks=11871232Total megabyte-milliseconds taken by all reduce tasks=9881600Map-Reduce FrameworkMap input records=6Map output records=21Map output bytes=233Map output materialized bytes=84Input split bytes=98Combine input records=21Combine output records=6Reduce input groups=6Reduce shuffle bytes=84Reduce input records=6Reduce output records=6Spilled Records=12Shuffled Maps =1Failed Shuffles=0Merged Map outputs=1GC time elapsed (ms)=300CPU time spent (ms)=2910Physical memory (bytes) snapshot=353423360Virtual memory (bytes) snapshot=5477199872Total committed heap usage (bytes)=196218880Peak Map Physical memory (bytes)=228843520Peak Map Virtual memory (bytes)=2734153728Peak Reduce Physical memory (bytes)=124579840Peak Reduce Virtual memory (bytes)=2743046144Shuffle ErrorsBAD_ID=0CONNECTION=0IO_ERROR=0WRONG_LENGTH=0WRONG_MAP=0WRONG_REDUCE=0File Input Format CountersBytes Read=150File Output Format CountersBytes Written=54

4、查看运行结果

运行完毕后，使用hadoop fs -ls /output/wc可以看到运行结果输出的文件

使用hadoop fs -cat /output/wc/part-r-00000命令，可以看到程序运行的结果

除此之外，在YARN集群的监控页面http://node1:8088/ 点击左侧的Applications菜单，可以看到刚才运行过的任务

再点击任务的ID，可以进入任务详情页面

再点击某一个阶段的Logs链接，可以看到对应阶段的运行的客户端日志（在配置yarn-site.xml文件时，配置了开启日志聚合），这个页面本质上是JobHistoryServer提供的页面（19888端口）

在任务详情页面点击History链接，可以看到任务的历史运行状态，在其中可以看到其Map任务和Reduce任务，也可以继续点进Map和Reduce任务查看相关的日志等信息，对于程序出错时的排查很有帮助。

4.3.2. 提交根据Monte Carlo蒙特卡罗算法求圆周率的示例程序

1、提交程序

hadoop jar /export/server/hadoop/share/hadoop/mapreduce/hadoop-mapreduce-examples-3.3.4.jar pi 3 1000

hadoop jar 表示向YARN提交一个Java程序；
/export/server/hadoop-3.3.4/share/hadoop/mapreduce/hadoop-mapreduce-examples-3.3.4.jar 表示所要提交的程序路径；
pi 表示运行的Java类名；
3 表示使用3个Map任务；
1000 表示样本数为1000，样本数越多，求得的圆周率越准确，但是程序运行时长越长。
运行日志如下所示：

[hadoop@node1 ~]$ hadoop jar /export/server/hadoop/share/hadoop/mapreduce/hadoop-mapreduce-examples-3.3.4.jar pi 3 1000
Number of Maps  = 3
Samples per Map = 1000
Wrote input for Map #0
Wrote input for Map #1
Wrote input for Map #2
Starting Job
2023-12-14 16:06:12,042 INFO client.DefaultNoHARMFailoverProxyProvider: Connecting to ResourceManager at node1/192.168.88.101:8032
2023-12-14 16:06:13,550 INFO mapreduce.JobResourceUploader: Disabling Erasure Coding for path: /tmp/hadoop-yarn/staging/hadoop/.staging/job_1702538855741_0002
2023-12-14 16:06:13,888 INFO input.FileInputFormat: Total input files to process : 3
2023-12-14 16:06:14,149 INFO mapreduce.JobSubmitter: number of splits:3
2023-12-14 16:06:14,658 INFO mapreduce.JobSubmitter: Submitting tokens for job: job_1702538855741_0002
2023-12-14 16:06:14,659 INFO mapreduce.JobSubmitter: Executing with tokens: []
2023-12-14 16:06:15,065 INFO conf.Configuration: resource-types.xml not found
2023-12-14 16:06:15,065 INFO resource.ResourceUtils: Unable to find 'resource-types.xml'.
2023-12-14 16:06:15,256 INFO impl.YarnClientImpl: Submitted application application_1702538855741_0002
2023-12-14 16:06:15,403 INFO mapreduce.Job: The url to track the job: http://node1:8089/proxy/application_1702538855741_0002/
2023-12-14 16:06:15,404 INFO mapreduce.Job: Running job: job_1702538855741_0002
2023-12-14 16:06:32,155 INFO mapreduce.Job: Job job_1702538855741_0002 running in uber mode : false
2023-12-14 16:06:32,156 INFO mapreduce.Job:  map 0% reduce 0%
2023-12-14 16:06:47,156 INFO mapreduce.Job:  map 67% reduce 0%
2023-12-14 16:06:50,188 INFO mapreduce.Job:  map 100% reduce 0%
2023-12-14 16:06:57,275 INFO mapreduce.Job:  map 100% reduce 100%
2023-12-14 16:06:58,328 INFO mapreduce.Job: Job job_1702538855741_0002 completed successfully
2023-12-14 16:06:58,589 INFO mapreduce.Job: Counters: 54File System CountersFILE: Number of bytes read=72FILE: Number of bytes written=1108329FILE: Number of read operations=0FILE: Number of large read operations=0FILE: Number of write operations=0HDFS: Number of bytes read=786HDFS: Number of bytes written=215HDFS: Number of read operations=17HDFS: Number of large read operations=0HDFS: Number of write operations=3HDFS: Number of bytes read erasure-coded=0Job CountersLaunched map tasks=3Launched reduce tasks=1Data-local map tasks=3Total time spent by all maps in occupied slots (ms)=39354Total time spent by all reduces in occupied slots (ms)=7761Total time spent by all map tasks (ms)=39354Total time spent by all reduce tasks (ms)=7761Total vcore-milliseconds taken by all map tasks=39354Total vcore-milliseconds taken by all reduce tasks=7761Total megabyte-milliseconds taken by all map tasks=40298496Total megabyte-milliseconds taken by all reduce tasks=7947264Map-Reduce FrameworkMap input records=3Map output records=6Map output bytes=54Map output materialized bytes=84Input split bytes=432Combine input records=0Combine output records=0Reduce input groups=2Reduce shuffle bytes=84Reduce input records=6Reduce output records=0Spilled Records=12Shuffled Maps =3Failed Shuffles=0Merged Map outputs=3GC time elapsed (ms)=699CPU time spent (ms)=11980Physical memory (bytes) snapshot=775233536Virtual memory (bytes) snapshot=10945183744Total committed heap usage (bytes)=466890752Peak Map Physical memory (bytes)=227717120Peak Map Virtual memory (bytes)=2734153728Peak Reduce Physical memory (bytes)=113000448Peak Reduce Virtual memory (bytes)=2742722560Shuffle ErrorsBAD_ID=0CONNECTION=0IO_ERROR=0WRONG_LENGTH=0WRONG_MAP=0WRONG_REDUCE=0File Input Format CountersBytes Read=354File Output Format CountersBytes Written=97
Job Finished in 46.895 seconds
Estimated value of Pi is 3.14133333333333333333

2、查看运行情况
在在YARN集群的监控页面，可以查看对应任务的History信息，可以看到当前任务使用了3个Map任务和1个Reduce任务，同时，也可以查看相应的运行日志信息。