hadoop2.2.0 centos 编译安装详解

搭建环境:Centos x 6.4 64bit

1、安装JDK

我这里用的是64位机，要下载对应的64位的JDK,下载地址：http://www.oracle.com/technetwork/cn/java/javase/downloads/jdk7-downloads-1880260-zhs.html，选择对应的JDK版本，解压JDK，然后配置环境变量,

[html] view plaincopy
vi /etc/profile 

注：这里有的人喜欢配置在当前用户里，我这里是配置的全局。
[html] view plaincopy
export PATH 
export JAVA_HOME=/opt/jdk1.7 
export PATH=$PATH:$JAVA_HOME/bin 
 
source /etc/profile 

测试下JDK是否安装成功: java -version
[html] view plaincopy
java version "1.7.0_45" 
Java(TM) SE Runtime Environment (build 1.7.0_45-b18) 
Java HotSpot(TM) 64-Bit Server VM (build 24.45-b08, mixed mode) 

2、编译前的准备(maven)

maven官方下载地址，可以选择源码编码安装，这里就直接下载编译好的 就可以了


[html] view plaincopy
wget http://mirror.bit.edu.cn/apache/maven/maven-3/3.1.1/binaries/apache-maven-3.1.1-bin.zip 
解压文件后，同样在/etc/profie里配置环境变量


[html] view plaincopy
export MAVEN_HOME=/opt/maven3.1.1 
export PATH=$PATH:$MAVEN_HOME/bin 
验证配置是否成功: mvn -version

[html] view plaincopy
Apache Maven 3.1.1 (0728685237757ffbf44136acec0402957f723d9a; 2013-09-17 23:22:22+0800) 
Maven home: /opt/maven3.1.1 
Java version: 1.7.0_45, vendor: Oracle Corporation 
Java home: /opt/jdk1.7/jre 
Default locale: en_US, platform encoding: UTF-8 
OS name: "linux", version: "2.6.32-358.el6.x86_64", arch: "amd64", family: "unix" 

3、编译hadoop
这个地方你将会遇到各式各样的头疼问题

首先官方下载hadoop源码

[html] view plaincopy
wget http://mirrors.cnnic.cn/apache/hadoop/common/hadoop-2.2.0/hadoop-2.2.0-src.tar.gz 
  如果是你32bit的机器，可以直接下载官方已经编译好的包，64bit的机子跑编译好的包跑不了。
        由于maven国外服务器可能连不上，先给maven配置一下国内镜像，在maven目录下，conf/settings.xml,在<mirrors></mirros>里添加，原本的不要动

[html] view plaincopy
<mirror> 
     <id>nexus-osc</id> 
      <mirrorOf>*</mirrorOf> 
  <name>Nexusosc</name> 
  <url>http://maven.oschina.net/content/groups/public/</url> 
</mirror> 

    同样，在<profiles></profiles>内新添加
[html] view plaincopy
<profile> 
       <id>jdk-1.7</id> 
       <activation> 
         <jdk>1.7</jdk> 
       </activation> 
       <repositories> 
         <repository> 
           <id>nexus</id> 
           <name>local private nexus</name> 
           <url>http://maven.oschina.net/content/groups/public/</url> 
           <releases> 
             <enabled>true</enabled> 
           </releases> 
           <snapshots> 
             <enabled>false</enabled> 
           </snapshots> 
         </repository> 
       </repositories> 
       <pluginRepositories> 
         <pluginRepository> 
           <id>nexus</id> 
          <name>local private nexus</name> 
           <url>http://maven.oschina.net/content/groups/public/</url> 
           <releases> 
             <enabled>true</enabled> 
           </releases> 
           <snapshots> 
             <enabled>false</enabled> 
           </snapshots> 
         </pluginRepository> 
       </pluginRepositories> 
     </profile> 
编译clean
[html] view plaincopy
cd hadoop2.2.0-src 
mvn clean install –DskipTests 

发现异常
[html] view plaincopy
[ERROR] Failed to execute goal org.apache.hadoop:hadoop-maven-plugins:2.2.0:protoc (compile-protoc) on project hadoop-common: org.apache.maven.plugin.MojoExecutionException: 'protoc --version' did not return a version -> [Help 1] 
[ERROR]  
[ERROR] To see the full stack trace of the errors, re-run Maven with the -e switch. 
[ERROR] Re-run Maven using the -X switch to enable full debug logging. 
[ERROR]  
[ERROR] For more information about the errors and possible solutions, please read the following articles: 
[ERROR] [Help 1] http://cwiki.apache.org/confluence/display/MAVEN/MojoExecutionException 
[ERROR]  
[ERROR] After correcting the problems, you can resume the build with the command 
[ERROR]   mvn <goals> -rf :hadoop-common 

hadoop2.2.0编译需要protoc2.5.0的支持，所以还要下载protoc，下载地址:https://code.google.com/p/protobuf/downloads/list，要下载2.5.0版本噢
对protoc进行编译安装前先要装几个依赖包:gcc,gcc-c++,make 如果已经安装的可以忽略

[html] view plaincopy
yum install gcc 
yum intall gcc-c++ 
yum install make 

安装protoc
[html] view plaincopy
tar -xvf protobuf-2.5.0.tar.bz2 
cd protobuf-2.5.0 
./configure --prefix=/opt/protoc/ 
make && make install 
安装完配置下环境变量，就不多说了，跟上面过程一样。

别急，还不要着急开始编译安装，不然又是各种错误,需要安装cmake,openssl-devel,ncurses-devel依赖

[html] view plaincopy
yum install cmake 
yum install openssl-devel 
yum install ncurses-devel 

目前的2.2.0 的Source Code 压缩包解压出来的code有个bug 需要patch后才能编译。否则编译hadoop-auth 会提示下面错误：

[ERROR] Failed to execute goal org.apache.maven.plugins:maven-compiler-plugin:2.5.1:testCompile (default-testCompile) on project hadoop-auth: Compilation failure: Compilation failure:
[ERROR] /home/chuan/trunk/hadoop-common-project/hadoop-auth/src/test/java/org/apache/hadoop/security/authentication/client/AuthenticatorTestCase.java:[84,13] cannot access org.mortbay.component.AbstractLifeCycle
[ERROR] class file for org.mortbay.component.AbstractLifeCycle not found
Patch :https://issues.apache.org/jira/browse/HADOOP-10110


ok,现在可以进行编译了，
[html] view plaincopy
mvn package -Pdist,native -DskipTests -Dtar 

现在可以拿出你的手机，玩会游戏了，慢慢等吧！
[html] view plaincopy
[INFO] ------------------------------------------------------------------------ 
[INFO] Reactor Summary: 
[INFO]  
[INFO] Apache Hadoop Main ................................ SUCCESS [3.709s] 
[INFO] Apache Hadoop Project POM ......................... SUCCESS [2.229s] 
[INFO] Apache Hadoop Annotations ......................... SUCCESS [5.270s] 
[INFO] Apache Hadoop Assemblies .......................... SUCCESS [0.388s] 
[INFO] Apache Hadoop Project Dist POM .................... SUCCESS [3.485s] 
[INFO] Apache Hadoop Maven Plugins ....................... SUCCESS [8.655s] 
[INFO] Apache Hadoop Auth ................................ SUCCESS [7.782s] 
[INFO] Apache Hadoop Auth Examples ....................... SUCCESS [5.731s] 
[INFO] Apache Hadoop Common .............................. SUCCESS [1:52.476s] 
[INFO] Apache Hadoop NFS ................................. SUCCESS [9.935s] 
[INFO] Apache Hadoop Common Project ...................... SUCCESS [0.110s] 
[INFO] Apache Hadoop HDFS ................................ SUCCESS [1:58.347s] 
[INFO] Apache Hadoop HttpFS .............................. SUCCESS [26.915s] 
[INFO] Apache Hadoop HDFS BookKeeper Journal ............. SUCCESS [17.002s] 
[INFO] Apache Hadoop HDFS-NFS ............................ SUCCESS [5.292s] 
[INFO] Apache Hadoop HDFS Project ........................ SUCCESS [0.073s] 
[INFO] hadoop-yarn ....................................... SUCCESS [0.335s] 
[INFO] hadoop-yarn-api ................................... SUCCESS [54.478s] 
[INFO] hadoop-yarn-common ................................ SUCCESS [39.215s] 
[INFO] hadoop-yarn-server ................................ SUCCESS [0.241s] 
[INFO] hadoop-yarn-server-common ......................... SUCCESS [15.601s] 
[INFO] hadoop-yarn-server-nodemanager .................... SUCCESS [21.566s] 
[INFO] hadoop-yarn-server-web-proxy ...................... SUCCESS [4.754s] 
[INFO] hadoop-yarn-server-resourcemanager ................ SUCCESS [20.625s] 
[INFO] hadoop-yarn-server-tests .......................... SUCCESS [0.755s] 
[INFO] hadoop-yarn-client ................................ SUCCESS [6.748s] 
[INFO] hadoop-yarn-applications .......................... SUCCESS [0.155s] 
[INFO] hadoop-yarn-applications-distributedshell ......... SUCCESS [4.661s] 
[INFO] hadoop-mapreduce-client ........................... SUCCESS [0.160s] 
[INFO] hadoop-mapreduce-client-core ...................... SUCCESS [36.090s] 
[INFO] hadoop-yarn-applications-unmanaged-am-launcher .... SUCCESS [2.753s] 
[INFO] hadoop-yarn-site .................................. SUCCESS [0.151s] 
[INFO] hadoop-yarn-project ............................... SUCCESS [4.771s] 
[INFO] hadoop-mapreduce-client-common .................... SUCCESS [24.870s] 
[INFO] hadoop-mapreduce-client-shuffle ................... SUCCESS [3.812s] 
[INFO] hadoop-mapreduce-client-app ....................... SUCCESS [15.759s] 
[INFO] hadoop-mapreduce-client-hs ........................ SUCCESS [6.831s] 
[INFO] hadoop-mapreduce-client-jobclient ................. SUCCESS [8.126s] 
[INFO] hadoop-mapreduce-client-hs-plugins ................ SUCCESS [2.320s] 
[INFO] Apache Hadoop MapReduce Examples .................. SUCCESS [9.596s] 
[INFO] hadoop-mapreduce .................................. SUCCESS [3.905s] 
[INFO] Apache Hadoop MapReduce Streaming ................. SUCCESS [7.118s] 
[INFO] Apache Hadoop Distributed Copy .................... SUCCESS [11.651s] 
[INFO] Apache Hadoop Archives ............................ SUCCESS [2.671s] 
[INFO] Apache Hadoop Rumen ............................... SUCCESS [10.038s] 
[INFO] Apache Hadoop Gridmix ............................. SUCCESS [6.062s] 
[INFO] Apache Hadoop Data Join ........................... SUCCESS [4.104s] 
[INFO] Apache Hadoop Extras .............................. SUCCESS [4.210s] 
[INFO] Apache Hadoop Pipes ............................... SUCCESS [9.419s] 
[INFO] Apache Hadoop Tools Dist .......................... SUCCESS [2.306s] 
[INFO] Apache Hadoop Tools ............................... SUCCESS [0.037s] 
[INFO] Apache Hadoop Distribution ........................ SUCCESS [21.579s] 
[INFO] Apache Hadoop Client .............................. SUCCESS [7.299s] 
[INFO] Apache Hadoop Mini-Cluster ........................ SUCCESS [7.347s] 
[INFO] ------------------------------------------------------------------------ 
[INFO] BUILD SUCCESS 
[INFO] ------------------------------------------------------------------------ 
[INFO] Total time: 11:53.144s 
[INFO] Finished at: Fri Nov 22 16:58:32 CST 2013 
[INFO] Final Memory: 70M/239M 
[INFO] ------------------------------------------------------------------------ 

直到看到上面的内容那就说明编译完成了。
编译后的路径在:hadoop-2.2.0-src/hadoop-dist/target/hadoop-2.2.0

[html] view plaincopy
[root@localhost bin]# ./hadoop version 
Hadoop 2.2.0 
Subversion Unknown -r Unknown 
Compiled by root on 2013-11-22T08:47Z 
Compiled with protoc 2.5.0 
From source with checksum 79e53ce7994d1628b240f09af91e1af4 
This command was run using /data/hadoop-2.2.0-src/hadoop-dist/target/hadoop-2.2.0/share/hadoop/common/hadoop-common-2.2.0.jar 

可以看出hadoop的版本
[html] view plaincopy
[root@localhost hadoop-2.2.0]# file lib//native/* 
lib//native/libhadoop.a:        current ar archive 
lib//native/libhadooppipes.a:   current ar archive 
lib//native/libhadoop.so:       symbolic link to `libhadoop.so.1.0.0' 
lib//native/libhadoop.so.1.0.0: <span style="color:#ff0000;">ELF 64-bit LSB shared object, x86-64, version 1</span> (SYSV), dynamically linked, not stripped 
lib//native/libhadooputils.a:   current ar archive 
lib//native/libhdfs.a:          current ar archive 
lib//native/libhdfs.so:         symbolic link to `libhdfs.so.0.0.0' 
lib//native/libhdfs.so.0.0.0:   <span style="color:#ff0000;">ELF 64-bit LSB shared object, x86-64, version 1</span> (SYSV), dynamically linked, not stripped 

注意红色字体部分，如果下载官网的编译好的包，这里显示的是32-bit。

hadoop编译成功，下面可以来部署集群。



5、部署集群准备

     两台以上机器，修改hostname, ssh免登陆，关闭防火墙等

5.1、创建新用户

[html] view plaincopy
useradd hadoop 
su hadoop 
  注意以下操作有些需要root权限
5.2、修改主机名

[html] view plaincopy
vi /etc/sysconfig/network 
[html] view plaincopy
hostname master 
注销一下系统
[html] view plaincopy
[root@master ~]#  
变成master了,修改生效
5.3、修改hosts

[html] view plaincopy
vi /etc/hosts 
新增你的主机IP和HOSTNAME 
 
192.168.10.10  master 
192.168.10.11  slave1 
    5.4、ssh免登陆

查看ssh

[html] view plaincopy
[root@localhost data]# rpm -qa|grep ssh 
libssh2-1.4.2-1.el6.x86_64 
openssh-5.3p1-84.1.el6.x86_64 
openssh-server-5.3p1-84.1.el6.x86_64 
缺少openssh-clients,
[html] view plaincopy
yum install openssh-clients 
     

修改/etc/ssh/sshd_config

        RSAAuthentication yes

PubkeyAuthentication yes

        AuthorizedKeysFile      .ssh/authorized_keys



把这三行放开保存

然后service sshd restart 

现在开始配置无密登录

[html] view plaincopy
[hadoop@master ~]$ cd /home/hadoop/ 
[hadoop@master ~]$ ssh-keygen -t rsa 
一路回车
[html] view plaincopy
[hadoop@master ~]$ cd .ssh/ 
[hadoop@master .ssh]$ cp id_rsa.pub authorized_keys 
[hadoop@master .ssh]$ chmod 600 authorized_keys 
把authorized_keys复制到其他要无密的机器上
[html] view plaincopy
[hadoop@master .ssh]$ scp authorized_keys [email protected]:/home/hadoop/.ssh/ 
记得这里是以要以root权限过去，不然会报权限错误
一般情况到这里就可以无密登录了，可是我怎么还是需要密码，经过一翻搜寻才知道这是centos6.4版本的问题，《关于centos ssh无密登录失败的记录》
[html] view plaincopy
[hadoop@master .ssh]$ ssh slave1 
Last login: Mon Nov 25 14:49:25 2013 from master 
[hadoop@slave1 ~]$ 
看到已经变成slave1了，说明成功鸟

6、开始集群配置工作
配置之前在要目录下创建三个目录，用来放hadooop文件和日志数据


[html] view plaincopy
[hadoop@master ~]$mkdir -p dfs/name 
[hadoop@master ~]$mkdir -p dfs/data 
[hadoop@master ~]$mkdir -p temp 
把之前编译成功的版本移到hadoop目录下，注意目录权限问题
下面就开始配置文件

6.1  hadoop-env.sh

找到JAVA_HOME，把路径改为实际地址

        6.2  yarn-env.sh

同6.1

6.3  slave

配置所有slave节点

6.4 core-site.xml


[html] view plaincopy
              <property> 
                              <name>fs.defaultFS</name> 
                              <value>hdfs://master:9000</value>   //系统分布式URL 
              </property> 
<property> 
                              <name>io.file.buffer.size</name> 
                              <value>131072</value> 
              </property> 
             <property> 
                              <name>hadoop.tmp.dir</name> 
                              <value>file:/home/hadoop/temp</value> 
              </property> 
             <property> 
                         <name>hadoop.proxyuser.hadoop.hosts</name> 
                        <value>*</value> 
            </property> 
               <property> 
                         <name>hadoop.proxyuser.hadoop.groups</name> 
                        <value>*</value> 
             </property> 
  注意fs.defaultFS为2.2.0新的变量，代替旧的：fs.default.name

6.5、hdfs-site.xml

配置namenode、datanode的本地目录信息

[html] view plaincopy
<property> 
                <name>dfs.namenode.secondary.http-address</name> 
               <value>master:9001</value> 
        </property> 
         <property> 
                  <name>dfs.namenode.name.dir</name> 
                 <value>/home/hadoop/dfs/name</value> 
            </property> 
           <property> 
                    <name>dfs.datanode.data.dir</name> 
                    <value>/home/hadoop/dfs/data,/mnt/d1,/mnt/d2,/mnt/d3<table border="1" cellspacing="0" cellpadding="0" style="background:black;"><tbody><tr><td valign="top"><span style="background-color: rgb(240, 240, 240);"></value></span></td></tr></tbody></table>            </property> 
<span style="white-space:pre">  </span><property> 
                     <name>dfs.replication</name> 
                     <value>3</value> 
             </property> 
<span style="white-space:pre">      </span><property> 
                     <name>dfs.webhdfs.enabled</name> 
<span style="white-space:pre">          </span><value>true</value> 
         </property> 
新的：dfs.namenode.name.dir，旧：dfs.name.dir，新：dfs.datanode.name.dir，旧：dfs.data.dir
dfs.replication确定 data block的副本数目，hadoop基于rackawareness(机架感知)默认复制3份分block,（同一个rack下两个，另一个rack下一 份，按照最短距离确定具体所需block, 一般很少采用跨机架数据块，除非某个机架down了）

6.6、mapred-site.xml

配置其使用 Yarn 框架执行 map-reduce 处理程序
这个地方需要把mapred-site.xml.template复制重新命名


[html] view plaincopy
<property> 
                              <name>mapreduce.framework.name</name> 
                              <value>yarn</value> 
              </property> 
              <property> 
                              <name>mapreduce.jobhistory.address</name> 
                              <value><span style="font-family: Consolas, 'Courier New', Courier, mono, serif; line-height: 18px;">master</span>:10020</value> 
              </property> 
              <property> 
             <name>mapreduce.jobhistory.webapp.address</name> 
             <value>master:19888</value> 
 
     </property> 

新的计算框架取消了实体上的jobtracker, 故不需要再指定mapreduce.jobtracker.addres，而是要指定一种框架，这里选择yarn. 备注2：hadoop2.2.还支持第三方的计算框架，但没怎么关注过。
配置好以后将$HADOOP_HOME下的所有文件，包括hadoop目录分别copy到其它3个节点上。
    6.7、yarn-site.xml

配置ResourceManager,NodeManager的通信端口，WEB监控端口等
[html] view plaincopy
    <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.resourcemanager.address</name> 
              <value>master:8032</value> 
      </property> 
       <property> 
              <name>yarn.resourcemanager.scheduler.address</name> 
              <value>master:8030</value> 
       </property> 
       <property> 
                      <name>yarn.resourcemanager.resource-tracker.address</name> 
                       <value>master:8031</value> 
        </property> 
        <property> 
                      <name>yarn.resourcemanager.admin.address</name> 
                       <value>master:8033</value> 
         </property> 
         <property> 
              <name>yarn.resourcemanager.webapp.address</name> 
              <value>master:8088</value> 
          </property> 
[html] view plaincopy
<table cellspacing="0" cellpadding="0" class="t1   "><tbody><tr><td valign="middle" class="td1"><p class="p1"><property></p><p class="p1"><name>yarn.nodemanager.resource.memory-mb</name> //配置内存</p><p class="p1"><value>15360</value></p><p class="p1"></property></p></td></tr></tbody></table> 
[html] view plaincopy
<table cellspacing="0" cellpadding="0" class="t1   "><tbody><tr><td valign="middle" class="td1"><p class="p1"><span style="background-color: rgb(255, 255, 255);">到这里基本配置好了，把所有复制到其他的slave节点。</span></p></td></tr></tbody></table> 

7、启动hadoop

这里你可以进行环境变量设置，不举例了

7.1、格式化namenode

[html] view plaincopy
[hadoop@master hadoop]$ cd /home/hadoop/hadoop-2.2.0/bin/ 
[hadoop@master bin]$ ./hdfs namenode -format 
7.2、启动hdfs
[html] view plaincopy
[hadoop@master bin]$ cd ../sbin/ 
[hadoop@master sbin]$ ./start-dfs.sh 
这时候在master中输入jps应该看到namenode和secondarynamenode服务启动，slave中看到datanode服务启动

7.3、启动yarn
[html] view plaincopy
[hadoop@master sbin]$ ./start-yarn.sh 

master中应该有ResourceManager服务，slave中应该有nodemanager服务

查看集群状态：./bin/hdfs dfsadmin –report

查看文件块组成：  ./bin/hdfsfsck / -files -blocks

查看各节点状态:    http://192.168.10.10:50070

查看resourcemanager上cluster运行状态:    http:// 192.168.10.11:8088

8、安装中要注意的事项

8.1、注意版本，机器是32bit还是64位

8.2、注意依赖包的安装

8.3、写配置文件注意”空格“，特别是从别的地方copy的时候

8.4、关闭所有节点的防火墙

如果有看到类似"no route to host"这样的异常，基本就是防火墙没关

记得关的时候要切换到root帐号


[html] view plaincopy
（1） 重启后永久性生效： 
 
开启：chkconfig iptables on 
 
关闭：chkconfig iptables off 
 
（2） 即时生效，重启后失效： 
 
开启：service iptables start 
 
关闭：service iptables stop 
8.5、开启datanode后自动关闭
基本是因为namenode和datanode的clusterID不一致，可以参考《解决hadoop集群中datanode启动后自动关闭的问题》


其他一些特殊异常只能google之了

8.6 no datanode to stop

删除/tmp目录下的

adoop-daemon.sh代码，脚本是通过pid文件来停止hadoop服务的，而集群配置是使用的默认配置，pid文件位于/tmp目录下，对比/tmp目录下hadoop pid文件中的进程id和ps ax查出来的进程id，发现两个进程id不一致，终于找到了问题的根源。
赶紧去更新hadoop的配置吧！
修改hadoop-env.sh中的：HADOOP_PID_DIR = hadoop安装路径
9、运行测试例子


[html] view plaincopy
[hadoop@master bin]$ ./yarn jar ../share/hadoop/mapreduce/hadoop-mapreduce-examples-2.2.0.jar randomwriter /home/hadoop/dfs/input/ 
这里要注意不要用 -jar,不然会报异常“Exception in thread "main" java.lang.NoClassDefFoundError: org/apache/hadoop/util/ProgramDriver”
[html] view plaincopy
[hadoop@master bin]$ ./yarn jar ../share/hadoop/mapreduce/hadoop-mapreduce-examples-2.2.0.jar wordcount /home/hadoop/dfs/input/ /home/hadoop/dfs/output/ 
在input下面新建两个文件
[html] view plaincopy
$mkdir /dfs/input %echo ‘hello,world’ >> input/file1.in 
$echo ‘hello, ruby’ >> input/file2.in 
 
./bin/hadoop fs -mkdir -p /home/hadoop/dfs/input 
./bin/hadoop fs –put /home/hadoop/dfs/input /home/hadoop/test/test_wordcount/in 
 
查看word count的计算结果： 
$bin/hadoop fs -cat /home/hadoop/test/test_wordcount/out/* 
hadoop 1 
hello  1 
ruby