hadoop-2.3.0-cdh5.1.0完全分布式集群配置HA配置

一、安装前准备：

操作系统：CentOS 6.5 64位操作系统

环境：jdk1.7.0_45以上，本次采用jdk-7u55-linux-x64.tar.gz

master01 10.10.2.57 namenode 节点

master02 10.10.2.58 namenode 节点

slave01:10.10.2.173 datanode 节点

slave02:10.10.2.59 datanode 节点

slave03: 10.10.2.60 datanode 节点

注：Hadoop2.0以上采用的是jdk环境是1.7,Linux自带的jdk卸载掉，重新安装

下载地址：http://www.oracle.com/technetwork/java/javase/downloads/index.html

软件版本：hadoop-2.3.0-cdh5.1.0.tar.gz, zookeeper-3.4.5-cdh5.1.0.tar.gz

下载地址：http://archive.cloudera.com/cdh5/cdh/5/

开始安装：

二、jdk安装

1、检查是否自带jdk

rpm -qa | grep jdk

java-1.6.0-openjdk-1.6.0.0-1.45.1.11.1.el6.i686 

2、卸载自带jdk

yum -y remove java-1.6.0-openjdk-1.6.0.0-1.45.1.11.1.el6.i686

3、安装jdk-7u55-linux-x64.tar.gz

在usr/目录下创建文件夹java,在java文件夹下运行tar –zxvf jdk-7u55-linux-x64.tar.gz

解压到java目录下

[root@master01 java]# ls

jdk1.7.0_55

三、配置环境变量

远行vi /etc/profile

# /etc/profile

# System wide environment and startup programs, for login setup

# Functions and aliases go in /etc/bashrc

export JAVA_HOME=/usr/java/jdk1.7.0_55

export JRE_HOME=/usr/java/jdk1.7.0_55/jre

export CLASSPATH=/usr/java/jdk1.7.0_55/lib

export PATH=$JAVA_HOME/bin: $PATH

保存修改，运行source /etc/profile 重新加载环境变量

运行java -version

[root@master01 java]# java -version

java version "1.7.0_55"

Java(TM) SE Runtime Environment (build 1.7.0_55-b13)

Java HotSpot(TM) 64-Bit Server VM (build 24.55-b03, mixed mode)

Jdk配置成功

四、系统配置

预先准备5台机器，并配置IP

关闭防火墙

chkconfig iptables off(永久性关闭)

配置主机名和hosts文件

[root@master01 java]# vi /etc/hosts

127.0.0.1   localhost localhost.localdomain localhost4 localhost4.localdomain4

::1         localhost localhost.localdomain localhost6 localhost6.localdomain6

10.10.2.57 master01

10.10.2.58 master02

10.10.2.173 slave01

10.10.2.59 slave02

10.10.2.60 slave03

按照不同机器IP配置不同的主机名

3、SSH无密码验证配置

因为Hadoop运行过程需要远程管理Hadoop的守护进程，NameNode节点需要通过SSH（Secure Shell）链接各个DataNode节点，停止或启动他们的进程，所以SSH必须是没有密码的，所以我们要把NameNode节点和DataNode节点配制成无秘密通信，同理DataNode也需要配置无密码链接NameNode节点。

在每一台机器上配置：

vi /etc/ssh/sshd_config打开

RSAAuthentication yes # 启用 RSA 认证，PubkeyAuthentication yes # 启用公钥私钥配对认证方式

Master01:运行：ssh-keygen -t rsa -P ""  不输入密码直接enter

默认存放在 ／root／.ssh目录下,

cat ~/.ssh/id_rsa.pub >> ~/.ssh/authorized_keys

[root@master01 .ssh]# ls

authorized_keys  id_rsa  id_rsa.pub  known_hosts

slave01执行相同的操作，然后将master01 ／root／.ssh/目录下的id_rsa.pub放到 slave01 相同目录下的authorized_keys这样slave01就持有了master01的公钥 然后直接ssh slave01测试是否可以无密码连接到slave01上，然后将slave01 上的id_rsa.pub 追加到master01的authorized_keys中，测试ssh master01 是否可以直接连上slave01.

[root@master01 ~]# ssh slave01

Last login: Tue Aug 19 14:28:15 2014 from master01

[root@slave01 ~]# 

Master01-master02

Master01-slave01

Master01-slave02

Master01-slave03

Master02-slave01

Master02-slave02

Master02-slave03

执行相同的操作。

  

五、安装Hadoop

建立文件目录 /usr/local/cloud 创建文件夹data，存放数据、日志文件，haooop原文件，zookeeper原文件

[root@slave01 cloud]# ls

data  hadoop  tar  zookeeper

5.1、配置hadoop-env.sh

进入到/usr/local/cloud/hadoop/etc/hadoop目录下

配置vi hadoop-env.sh hadoop运行环境加载

export JAVA_HOME=/usr/java/jdk1.7.0_55

5.2、配置core-site.xml

<!—hadoop.tmp.dir:hadoop很多路径都依赖他，namenode节点该目录不可以删除，否则需要重新格式化-->

<property>

    <name>hadoop.tmp.dir</name>

    <value>/usr/local/cloud/data/hadoop/tmp</value>

</property>

<!—这个配置文件描述了集群的namenode节点的url,这里采用HA代表默认逻辑名，集群中的每个datanode节点都需要知道namenode的地址，数据才可以被使用-->

<property>

    <name>fs.defaultFS</name>

    <value>hdfs://zzg</value>

</property>

<!-- zookeeper集群的地址和端口,最好保持基数个至少3台-->

 <property>

    <name>ha.zookeeper.quorum</name>

    <value>master01:2181,slave01:2181,slave02:2181</value>

</property>

  

(2)hdfs-site.xml配置

<!—hadoop namenode数据的存储目录，只是针对与namenode，包含了namenode的系统信息元数据信息-->

<property>

    <name>dfs.namenode.name.dir</name>

    <value>/usr/local/cloud/data/hadoop/dfs/nn</value>

</property>

<!—datanode 要存储到数据到本地的路径，不必每一台机器都一样，但是为了方便管理最好还是一样-->

<property>

    <name>dfs.datanode.data.dir</name>

    <value>/usr/local/cloud/data/hadoop/dfs/dn</value>

</property>

<!—系统中文件备份数量，系统默认是3分-->

<property>

    <name>dfs.replication</name>

    <value>3</value>

</property>

<!-- dfs.webhdfs.enabled 置为true，否则一些命令无法使用如：webhdfs的LISTSTATUS -->

<property>

    <name>dfs.webhdfs.enabled</name>

    <value>true</value>

</property>

<!—可选，关闭权限带来一些不必要的麻烦-->

<property>

     <name>dfs.permissions</name>

     <value>false</value>

</property>

<!—可选，关闭权限带来一些不必要的麻烦-->

<property>

     <name>dfs.permissions.enabled</name>

     <value>false</value>

</property>

<!—HA配置-->

<!—设置集群的逻辑名-->

<property>

    <name>dfs.nameservices</name>

    <value>zzg</value>

</property>

<!—hdfs联邦集群中的namenode节点逻辑名-->

<property>

    <name>dfs.ha.namenodes.zzg</name>

    <value>nn1,nn2</value>

</property>

<!—hdfs namenode逻辑名中RPC配置，rpc 简单理解为序列化文件上传输出文件要用到-->

<property>

    <name>dfs.namenode.rpc-address.zzg.nn1</name>

    <value>master01:9000</value>

</property>

<property>

    <name>dfs.namenode.rpc-address.zzg.nn2</name>

    <value>master02:9000</value>

</property>

<!—配置hadoop页面访问端口端口-->

<property>

    <name>dfs.namenode.http-address.zzg.nn1</name>

    <value>master01:50070</value>

</property>

<property>

    <name>dfs.namenode.http-address.zzg.nn2</name>

    <value>master02:50070</value>

</property>

<!—建立与namenode的通信-->

<property>

    <name>dfs.namenode.servicerpc-address.zzg.nn1</name>

    <value>master01:53310</value>

</property>

<property>

    <name>dfs.namenode.servicerpc-address.zzg.nn2</name>

    <value>master02:53310</value>

</property>

<!—journalnode 共享文件集群-->

<property>

    <name>dfs.namenode.shared.edits.dir</name>

    <value>qjournal://master01:8485;slave01:8485;slave02:8485/zzg</value>

</property>

 <!—journalnode对namenode的进行共享设置-->

<property>

    <name>dfs.journalnode.edits.dir</name>

    <value>/usr/local/cloud/data/hadoop/ha/journal</value>

</property>

<!—设置故障处理类-->

<property>

    <name>dfs.client.failover.proxy.provider.zzg</name>

    <value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>

</property>

<!—开启自动切换-->

<property>

    <name>dfs.ha.automatic-failover.enabled</name>

    <value>true</value>

</property>

<property>

        <name>ha.zookeeper.quorum</name>

        <value>master01:2181,slave01:2181,slave02:2181</value>

</property>

<!—使用ssh方式进行故障切换-->

<property>

    <name>dfs.ha.fencing.methods</name>

    <value>sshfence</value>

</property>

<!—ssh通信密码通信位置-->

<property>

    <name>dfs.ha.fencing.ssh.private-key-files</name>

    <value>/root/.ssh/id_rsa</value>

</property>

5.3 配置maped-site.xml

<property>

                <name>mapreduce.framework.name</name>

                <value>yarn</value>

</property>

5.4配置yarn HA 

配置yarn-en.sh java环境

# some Java parameters

  export JAVA_HOME=/usr/java/jdk1.7.0_55

5.5配置yarn-site.xml

        <!—rm失联后重新链接的时间-->

        <property>

                <name>yarn.resourcemanager.connect.retry-interval.ms</name>

                <value>2000</value>

        </property>

        <!—开启resource manager HA,默认为false-->

         <property>

                <name>yarn.resourcemanager.ha.enabled</name>

                <value>true</value>

        </property>

        <!—开启故障自动切换-->

        <property>

                <name>yarn.resourcemanager.ha.automatic-failover.enabled</name>

                <value>true</value>

        </property>

        <!—配置resource manager -->

        <property>

                <name>yarn.resourcemanager.ha.rm-ids</name>

                <value>rm1,rm2</value>

        </property>

        <!—在master01上配置rm1,在master02上配置rm2,-->

        <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>

        <!—开启自动恢复功能-->

         <property>

                <name>yarn.resourcemanager.recovery.enabled</name>

                 <value>true</value>

        </property>

        <!—配置与zookeeper的连接地址-->

        <property>

                <name>yarn.resourcemanager.zk-state-store.address</name>

                <value>localhost: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>localhost:2181</value>

        </property>

        <property>

                <name>yarn.resourcemanager.cluster-id</name>

                <value>yarn-cluster</value>

        </property>

        <!—schelduler失联等待连接时间-->

         <property>

                <name>yarn.app.mapreduce.am.scheduler.connection.wait.interval-ms</name>

                <value>5000</value>

        </property>

        <!—配置rm1-->

        <property>

                <name>yarn.resourcemanager.address.rm1</name>

                <value>master01:23140</value>

        </property>

        <property>

                <name>yarn.resourcemanager.scheduler.address.rm1</name>

                <value>master01:23130</value>

        </property>

        <property>

                <name>yarn.resourcemanager.webapp.address.rm1</name>

                <value>master01:23188</value>

        </property>

        <property>

                <name>yarn.resourcemanager.resource-tracker.address.rm1</name>

                <value>master01:23125</value>

        </property>

         <property>

                <name>yarn.resourcemanager.admin.address.rm1</name>

                <value>master01:23141</value>

        </property>

        <property>

                <name>yarn.resourcemanager.ha.admin.address.rm1</name>

                <value>master01:23142</value>

        </property>

        <!—配置rm2-->

         <property>

                <name>yarn.resourcemanager.address.rm2</name>

                <value>master02:23140</value>

        </property>

        <property>

                <name>yarn.resourcemanager.scheduler.address.rm2</name>

                <value>master02:23130</value>

        </property>

        <property>

                <name>yarn.resourcemanager.webapp.address.rm2</name>

                <value>master02:23188</value>

        </property>

        <property>

                <name>yarn.resourcemanager.resource-tracker.address.rm2</name>

                <value>master02:23125</value>

        </property>

        <property>

                <name>yarn.resourcemanager.admin.address.rm2</name>

                <value>master02:23141</value>

        </property>

        <property>

                <name>yarn.resourcemanager.ha.admin.address.rm2</name>

                <value>master02:23142</value>

        </property>

        <!—配置nodemanager-->

        <property>

                <description>Address where the localizer IPC is.</description>

                <name>yarn.nodemanager.localizer.address</name>

                <value>0.0.0.0:23344</value>

        </property>

        <!—nodemanager http访问端口-->

         <property>

                <description>NM Webapp address.</description>

                <name>yarn.nodemanager.webapp.address</name>

                <value>0.0.0.0:23999</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>/usr/local/cloud/data/hadoop/yarn/local</value>

        </property>

        <property>

                <name>yarn.nodemanager.log-dirs</name>

                <value>/usr/local/cloud/data/logs/hadoop</value>

        </property>

        <property>

                <name>mapreduce.shuffle.port</name>

                <value>23080</value>

        </property>

        <!—故障处理类-->

         <property>

                <name>yarn.client.failover-proxy-provider</name>

                 <value>org.apache.hadoop.yarn.client.ConfiguredRMFailoverProxyProvider</value>

         </property>

六、配置zookeeper集群

在zookeeper目录下建立data目录 和logs目录，

配置zoo.cnf

dataDir=/usr/local/cloud/zookeeper/data

dataLogDir=/usr/local/cloud/zookeeper/logs

# the port at which the clients will connect

clientPort=2181

server.1=master01:2888:3888

server.2=master02:2888:3888

server.3=slave01:2888:3888

server.4=slave02:2888:3888

server.5=slave03:2888:3888

在data目录下创建myid文件，并在对应的机器上填写数字，如上配置master01 server01 的myid写入1，

master02 中的data的myid写入2，依次在其他机子上执行相同操作。

在各个机器下zookeeper目录下的bin目录下执行zkServer.sh start命令

再运行zkServer.sh status如果出现leader 或fllower 则说明集群配置正确。

  

到此各个配置文件配置完毕

七、启动Hadoop集群严格按照以下顺序执行（第一次）

(1)各个节点启动zookeeper，在zookeeper/bin/zkServer.sh start

(2) 在hadoop/bin/hdfs zkfc –formatZK 进行格式化创建命名空间

(3)在配置了journalnode的节点启动，master01，slave01,slave02

   在hadoop/sbin/hadoop-daemon.sh  journalnode

（4）在主namenode节点执行格式化

./bin/hadoop namenode -format zzg

 主机器上启动namenode

 hadoop/sbin/ hadoop-daemon.sh start namenode

(5)将主namenode节点格式化的目录拷贝到从主namenode节点上

hadoop/bin/hdfs namenode –bootstrapStandby

hadoop/sbin/hadoop-daemon.sh start namenode

(6) 在两个namenode节点都执行以下命令

./sbin/hadoop-daemon.sh start zkfc

(7) 在所有datanode节点都执行以下命令启动datanode

./sbin/hadoop-daemon.sh start datanode

(8)在主namenode节点启动yarn，运行yarn-start.sh命令

jps可以看到

namenode节点

[root@master01 ~]# jps

38972 JournalNode

38758 NameNode

39166 DFSZKFailoverController

37473 QuorumPeerMain

39778 ResourceManager

42620 Jps

datanode节点

[root@slave01 ~]# jps

33440 DataNode

35277 Jps

32681 QuorumPeerMain

33568 JournalNode

34231 NodeManager