搭建Hadoop的HDFS HA及YARN HA集群,基于2.7.1版本安装。

HadoopHA工作原理图.png

YarnHA工作原理.png

安装规划

安装用户

garrison/zdh1234

配置IP对应节点名称

cat /etc/hosts
vi /etc/hosts
10.43.159.237 zdh-237
10.43.159.238 zdh-238
10.43.159.239 zdh-239
10.43.159.240 zdh-240
10.43.159.245 zdh-245

1.创建用户（所有的用户必须同名）

groupadd hadoop
useradd -g hadoop -s /bin/bash -md /home/garrison garrison
passwd garrison

2.设置本地无密码登陆

ssh-keygen -t dsa -P '' -f ~/.ssh/id_dsa
cat ~/.ssh/id_dsa.pub >> ~/.ssh/authorized_keys
验证免密登陆
ssh localhost

设置远程无密码登陆，需要把本机的公钥放到对方的机器authorized_keys，才能免密登陆其他机器。
进入zdh-238的garrison
scp ~/.ssh/authorized_keys garrison@zdh-237:~/.ssh/authorized_keys_from_zdh-238
进入zdh-237 garrison的.ssh目录,注意备份，否则下面步骤存在重复的ywmaster公钥。
cat authorized_keys_from_zdh-238  >>  authorized_keys
进入zdh-238的garrison，这样就能在zdh-238上面免密登陆zdh-237
ssh zdh-237

其他zdh-239等同理复制到zdh-237上面,实现其他机器免密登陆zdh-237

再把zdh-237上面的authorized_keys分发到其他zdh-238等上面,实现几台机器都能免密登陆
scp ~/.ssh/authorized_keys garrison@zdh-238:~/.ssh/authorized_keys

3.拷贝安装包

scp root@10.43.159.41:/home/xiehh/.tar.gz .
scp root@10.43.159.41:/home/ling/java/jdk-7u80-linux-x64.tar.gz .
scp garrison@zdh-237:/home/garrison/.tar.gz .

4.安装jdk

tar -zxvf jdk-7u80-linux-x64.tar.gz
mv jdk-7u80-linux-x64.tar.gz backup/
/home/garrison/jdk1.7.0_80
配置环境变量，必须放在.bashrc里面，否则通过后台执行找不到环境变量。
创建.bash_profile,加载.bashrc
vi .bash_profile

# .bash_profile# Get the aliases and functionsif [ -f ~/.bashrc ]; then
        . ~/.bashrcfi# User specific environment and startup programs

vi .bashrc

export JAVA_HOME=~/jdk1.7.0_80export PATH=$PATH:$JAVA_HOME/binexport CLASSPATH=.:$JAVA_HOME/jre/lib/rt.jar:$JAVA_HOME/lib/dt.jar:$JAVA_HOME/lib/tools.jar

source .bashrc
验证jdk
java -version

5.安装Zookeeper

解压zookeeper包
tar -zxvf zookeeper-3.5.1-alpha.tar.gz
mv zookeeper-3.5.1-alpha.tar.gz  backup/
在zookeeper-3.5.1-alpha/conf/目录执行
mv zoo_sample.cfg zoo.cfg
修改zoo.cfg文件,（默认服务端口2181，zookeeper修改源数据的地方，包括myid文件）：
dataDir=/home/garrison/zookeeper-3.5.1-alpha/tmp
文件最后添加，配置zookeeper集群通信端口:

server.1=zdh-237:2888:3888
server.2=zdh-238:2888:3888
server.3=zdh-239:2888:3888

然后分别在zdh-237三个节点中创建一个tmp文件夹：
mkdir ~/zookeeper-3.5.1-alpha/tmp：
再创建一个空文件：
touch /tmp/myid

把zdh-237的zookeeper拷贝到zdh-238等节点
scp -r garrison@zdh-237:/home/garrison/zookeeper-3.5.1-alpha .

最后向该文件写入ID:
zdh-237执行echo 1 > /tmp/myid
zdh-238执行echo 2 > /tmp/myid
zdh-239执行echo 3 > /tmp/myid

配置环境变量方便以后操作：
export ZOOKEEPER_HOME=:~/zookeeper-3.5.1-alpha
export PATH=ZOOKEEPER_HOME/bin:$ZOOKEEPER_HOME/conf

启动zookeeper，进入到 zookeeper-3.4.5/bin/
./zkServer.sh start
查看状态：
./zkServer.sh status
停止zookeeper:
./zkServer.sh stop

验证，客户端登陆：
./zkCli.sh -server zdh-237:2181
列出目录：
ls /

6.安装hadoop

tar -zxvf hadoop-2.7.1.tar.gz
mv  hadoop-2.7.1.tar.gz backup/

配置环境变量
export HADOOP_HOME=~/hadoop-2.7.1
export PATH=HADOOP_HOME/sbin:HADOOP_HOME/etc/hadoop

别名方便进入配置操作
alias conf='cd /home/garrison/hadoop-2.7.1/etc/hadoop'

修改文件列表：

• 6.1 core-site.xml

• 6.2 hdfs-site.xml

• 6.3 yarn-site.xml

• 6.4 mapred-site.xml

• 6.5 slaves

zdh-237:2181,zdh-238:2181,zdh-239:2181
/data/hadoop/dfs/name/current 等目录如果不存在则需要修改，一般加上前缀/home/garrison/hadoop-2.7.1

6.1-6.4的配置可以参考文档：

搭建hadoop2.6.0 HA及YARN HA
http://www.aboutyun.com/thread-10572-1-1.html

6.4

新增mapred-site.xml

6.5

修改slaves （配置所有slave节点）
zdh-237
zdh-238
zdh-239

拷贝zdh-240的hadoop到其他节点。
scp -r garrison@zdh-240:/home/garrison/hadoop-2.7.1 .
rm -r /home/garrison/hadoop-2.7.1/etc/hadoop
scp -r garrison@zdh-240:~/hadoop-2.7.1/etc/hadoop ~/hadoop-2.7.1/etc/hadoop
集群机器配置hadoop环境变量

修改zdh-245的yarn-site.xml的rm1为rm2

<!--在namenode1上配置rm1,在namenode2上配置rm2,
注意：一般都喜欢把配置好的文件远程复制到其它机器上，
但这个在YARN的另一个机器上一定要修改--> <property> 
  <name>yarn.resourcemanager.ha.id</name> 
  <value>rm1</value> <description>If we want to launch more than one RM in single node, we need this configuration</description> </property>

启动journalnode（在namenode1上启动所有journalnode）
进入到hadoop-2.6.0
sbin/hadoop-daemons.sh start journalnode
或者单独进入到datanode1,datanode2,datanode3执行
sbin/hadoop-daemon.sh start journalnode
停止journalnode：
hadoop-daemon.sh stop journalnode
(运行jps命令检验，多了JournalNode进程)

7.启动hadoop和yarn

格式化HDFS,在namenode1上执行命令:
hadoop namenode -format
格式化后会在根据core-site.xml中的hadoop.tmp.dir配置生成个文件

启动namenode进程,在namenode1上执行
sbin/hadoop-daemon.sh start namenode

在namenode2上执行，完成主备节点同步信息
hdfs namenode -bootstrapStandby

格式化ZK(在namenode1上执行即可)，会在zookeeper集群上面创建节点hadoop-ha，
用于管理切换主备namenode
hdfs zkfc -formatZK

启动HDFS(在namenode1上执行)
sbin/start-dfs.sh

启动YARN(在namenode1和namenode2上执行)
sbin/start-yarn.sh
注意在namenode2上执行此命令时会提示NodeManager已存在等信息不用管这些，
主要是启动namenode2上的resourceManager完成与namenode1的互备作用，目前没有找到单独启动resourceManager的方法

8.查看结点状态

启动后查看namenode分别为Active和Standby
http://10.43.159.240:50070
http://10.43.159.245:50070
在namenode1上查看nm1和nm2状态：
hdfs haadmin -getServiceState nn1
hdfs haadmin -getServiceState nn2

在namenode1上查看rm1和rm2分别为active和standby状态
yarn rmadmin -getServiceState rm1
yarn rmadmin -getServiceState rm2
或者查看状态：
http://10.43.159.240:8188
http://10.43.159.245:8188 会重定向到zdh-240
hadoop
zdh-240 active
zdh-245 standby
yarn
zdh-240 rm1 active
zdh-245 rm2 standby
简单验证hadoop
hadoop jar ~/hadoop-2.7.1/share/hadoop/mapreduce/hadoop-mapreduce-examples-2.7.1.jar wordcount hdfs://gagcluster/usr/wordcout.txt /usr/wordresult_001
hadoop fs -text /usr/wordresult_001/part-r-00000

9.验证HA

yarn kill zdh-240 上面的ResourceManager
http://10.43.159.240:8188/cluster/apps 无法访问；rm1 链接失败
http://10.43.159.245:8188/cluster/apps 可以访问；rm2 active

hadoop kill zdh-240 上面的NameNode
手动切换主备时，确定要转为active的namenode的id，这里将namenode1设为active：
hdfs haadmin -failover --forcefence --forceactive nn2 nn1
自动切换主备时，需要重新启动被kill的active节点，standby节点才会变为active,
原来被kill的active节点变成standby。

另外一种方法就是，关闭当前为 active namenode 状态的上的 DFSZKFailoverController 进程，
在需要变成standby的hdfs上面执行：
hadoop-daemon.sh stop zkfc

10.配置免密登陆的其他方法

10.1.生成ssh公私钥文件

操作机器：
在zte-1、zte-2、zte-3上，使用hdfs用户 ，家目录下
操作命令：
ssh-keygen
操作说明：
该命令执行完后应按三次Enter键，即三次需要输入的皆为空即可

10.2.为hdfs用户配置ssh免密码登录

操作机器：
在zte-1、zte-2、zte-3上，使用hdfs用户
操作命令：
ssh-copy-id hdfs@zdh-7
ssh-copy-id hdfs@zdh-9
ssh-copy-id hdfs@zdh-11
操作说明：
交互(yes/no)需要输入yes，提示输入密码需要输入密码。
当出现类似：
Now try logging into the machine, with "ssh 'hdfs@zdh-1'",
and check in:.ssh/authorized_keys
则表示成功，如果显示….can’t established则表示发生错误。

10.3.上述操作可以优化

分别在zdh-7，zdh-9,zdh-11上面
ssh-copy-id -i hdfs@zdh-7，
再把zdh-7上面的.ssh/authorized_keys,.ssh/known_hosts拷贝到其他机器即可。

参考：

6.1 core-site.xml中增加如下配置项

<configuration>
    <property>
            <name>fs.defaultFS</name>
            <value>hdfs://gagcluster</value>
    </property>
    <property>
            <name>io.file.buffer.size</name>
            <value>131072</value>
    </property>
    <property>
            <name>hadoop.tmp.dir</name>
            <value>file:/home/garrison/hadoop-2.7.1/data/hadoop/tmp</value>
            <description>Abase for other temporary directories.</description>
    </property>
    <property>
            <name>hadoop.proxyuser.hduser.hosts</name>
            <value>*</value>
    </property>
    <property>
            <name>hadoop.proxyuser.hduser.groups</name>
            <value>*</value>
    </property>
    <property>
            <name>ha.zookeeper.quorum</name>
            <value>zdh-237:2181,zdh-238:2181,zdh-239:2181</value>
    </property></configuration>

6.2 hdfs-site.xml增加如下配置项

<configuration>
    <property>
            <name>dfs.nameservices</name>
            <value>gagcluster</value>
    </property>
    <property>
            <name>dfs.ha.namenodes.gagcluster</name>
            <value>nn1,nn2</value>
    </property>
    <property>
            <name>dfs.namenode.rpc-address.gagcluster.nn1</name>
            <value>zdh-240:9000</value>
    </property>
    <property>
            <name>dfs.namenode.rpc-address.gagcluster.nn2</name>
            <value>zdh-245:9000</value>
    </property>
    <property>
            <name>dfs.namenode.http-address.gagcluster.nn1</name>
            <value>zdh-240:50070</value>
    </property>
    <property>
            <name>dfs.namenode.http-address.gagcluster.nn2</name>
            <value>zdh-245:50070</value>
    </property>
    <property>
            <name>dfs.namenode.shared.edits.dir</name>
            <value>qjournal://zdh-237:8485;zdh-238:8485;zdh-239:8485/gagcluster</value>
    </property>
    <property>
            <name>dfs.client.failover.proxy.provider.gagcluster</name>
            <value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>
    </property>
    <property>
            <name>dfs.ha.fencing.methods</name>
            <value>sshfence</value>
    </property>
    <property>
            <name>dfs.ha.fencing.ssh.private-key-files</name>
            <value>/home/garrison/.ssh/id_rsa</value>
    </property>
    <property>
            <name>dfs.journalnode.edits.dir</name>
            <value>/home/garrison/hadoop-2.7.1/data/hadoop/tmp/journal</value>
    </property>
    <property>
            <name>dfs.ha.automatic-failover.enabled</name>
            <value>true</value>
    </property>
    <property>   
            <name>dfs.namenode.name.dir</name>   
            <value>file:/home/garrison/hadoop-2.7.1/data/hadoop/dfs/name</value>  
    </property> 
    <property>   
            <name>dfs.datanode.data.dir</name>   
            <value>file:/home/garrison/hadoop-2.7.1/data/hadoop/dfs/data</value>  
    </property> 
    <property>   
            <name>dfs.replication</name>   
            <value>3</value> 
    </property> 
    <property>  
            <name>dfs.webhdfs.enabled</name>  
            <value>true</value> 
    </property>
    <property>  
            <name>dfs.journalnode.http-address</name>  
            <value>0.0.0.0:8480</value>  
    </property>  
    <property>  
            <name>dfs.journalnode.rpc-address</name>  
            <value>0.0.0.0:8485</value>  
     </property> 
     <property>
            <name>ha.zookeeper.quorum</name>
            <value>zdh-237:2181,zdh-238:2181,zdh-239:2181</value>
    </property></configuration>

6.3 yarn-site.xml增加如下配置项

<configuration><property> 
   <name>yarn.resourcemanager.connect.retry-interval.ms</name> 
   <value>2000</value> </property><property> 
   <name>yarn.resourcemanager.ha.enabled</name> 
   <value>true</value> </property> <property>
  <name>yarn.resourcemanager.ha.rm-ids</name>
  <value>rm1,rm2</value></property><property>
  <name>ha.zookeeper.quorum</name>
  <value>zdh-237:2181,zdh-238:2181,zdh-239:2181</value> </property><property> 
   <name>yarn.resourcemanager.ha.automatic-failover.enabled</name> 
   <value>true</value> </property> <property>
  <name>yarn.resourcemanager.hostname.rm1</name>
  <value>zdh-240</value></property>               <property>
   <name>yarn.resourcemanager.hostname.rm2</name>
   <value>zdh-245</value></property><property> 
  <name>yarn.resourcemanager.ha.id</name> 
  <value>rm1</value> <description>If we want to launch more than one RM in single node, we need this configuration</description> </property> <property>
  <name>yarn.resourcemanager.recovery.enabled</name> 
  <value>true</value> </property><property> 
  <name>yarn.resourcemanager.zk-state-store.address</name> 
  <value>zdh-237:2181,zdh-238:2181,zdh-239:2181</value></property> <property> 
  <name>yarn.resourcemanager.store.class</name> 
  <value>org.apache.hadoop.yarn.server.resourcemanager.recovery.ZKRMStateStore</value> </property> <property>
  <name>yarn.resourcemanager.zk-address</name>
  <value>zdh-237:2181,zdh-238:2181,zdh-239:2181</value></property><property> 
  <name>yarn.resourcemanager.cluster-id</name> 
  <value>gagcluster-yarn</value> </property> <property> 
  <name>yarn.app.mapreduce.am.scheduler.connection.wait.interval-ms</name> 
  <value>5000</value> </property> <property> 
  <name>yarn.resourcemanager.address.rm1</name> 
  <value>zdh-240:8132</value> </property> <property> 
  <name>yarn.resourcemanager.scheduler.address.rm1</name> 
  <value>zdh-240:8130</value> </property> <property> 
  <name>yarn.resourcemanager.webapp.address.rm1</name> 
  <value>zdh-240:8188</value> </property> <property>
   <name>yarn.resourcemanager.resource-tracker.address.rm1</name> 
   <value>zdh-240:8131</value> </property> <property> 
  <name>yarn.resourcemanager.admin.address.rm1</name> 
  <value>zdh-240:8033</value> </property> <property> 
  <name>yarn.resourcemanager.ha.admin.address.rm1</name> 
  <value>zdh-240:23142</value> </property> <property> 
  <name>yarn.resourcemanager.address.rm2</name> 
  <value>zdh-245:8132</value> </property> <property> 
  <name>yarn.resourcemanager.scheduler.address.rm2</name> 
  <value>zdh-245:8130</value> </property> <property> 
  <name>yarn.resourcemanager.webapp.address.rm2</name> 
  <value>zdh-245:8188</value> </property> <property> 
  <name>yarn.resourcemanager.resource-tracker.address.rm2</name> 
  <value>zdh-245:8131</value> </property> <property> 
  <name>yarn.resourcemanager.admin.address.rm2</name> 
  <value>zdh-245:8033</value> </property> <property> 
  <name>yarn.resourcemanager.ha.admin.address.rm2</name> 
  <value>zdh-245:23142</value> </property> <property> 
  <name>yarn.nodemanager.aux-services</name> 
  <value>mapreduce_shuffle</value> </property> <property> 
  <name>yarn.nodemanager.aux-services.mapreduce.shuffle.class</name> 
  <value>org.apache.hadoop.mapred.ShuffleHandler</value> </property> <property> 
  <name>yarn.nodemanager.local-dirs</name> 
  <value>/home/garrison/hadoop-2.7.1/data/hadoop/yarn/local</value> </property> <property> 
  <name>yarn.nodemanager.log-dirs</name> 
  <value>/home/garrison/hadoop-2.7.1/data/log/hadoop</value> </property> <property> 
  <name>mapreduce.shuffle.port</name> 
  <value>23080</value> </property> <property> 
  <name>yarn.client.failover-proxy-provider</name> 
  <value>org.apache.hadoop.yarn.client.ConfiguredRMFailoverProxyProvider</value> </property> <property>
  <name>yarn.resourcemanager.ha.automatic-failover.zk-base-path</name>
  <value>/yarn-leader-election</value>
  <description>Optional setting. The default value is /yarn-leader-election</description></property></configuration>

6.4 mapred-site.xml增加如下配置项

<configuration>
    <property>
        <name>mapreduce.framework.name</name>
        <value>yarn</value>
    </property>
    <property>
        <name>mapreduce.jobhistory.address</name>
        <value>0.0.0.0:10020</value>
    </property>
    <property>
        <name>mapreduce.jobhistory.webapp.address</name>
        <value>0.0.0.0:19888</value>
     </property></configuration>

作者：木木与呆呆
链接：https://www.jianshu.com/p/d2538cde7d87