Hadoop 2.x 安装

1、安装配置

1.建立hadoop用户(我是添加到root group，也能够新增hadoop group)

[root@hftclclw0001 ~]# useradd hadoop
[root@hftclclw0001 ~]# usermod -g root

[root@hftclclw0001 ~]# cat /etc/passwd
...
...
hadoop:x:50295:0::/home/hadoop:/bin/bash

[root@hftclclw0001 ~]# chmod 644 /etc/suders
[root@hftclclw0001 ~]# vi 644 /etc/suders
...
...
root	ALL=(ALL) 	ALL
hadoop	ALL=(ALL) 	ALL
...

2.ssh免密码登录

[hadoop@hftclclw0001 hadoop]$ ssh-keygen -t rsa

[hadoop@hftclclw0001 hadoop]$ cat ~/.ssh/id_rsa.pub >> ~/.ssh/authorized_keys 

[hadoop@hftclclw0001 hadoop]$ tree ~/.ssh/
/home/hadoop/.ssh/
├── authorized_keys
├── id_rsa
├── id_rsa.pub
└── known_hosts

0 directories, 4 files

操做其余各个机器，并复制公钥(d_rsa.pub)到其余各个机器的authorized_keys中。我使用的scp，复制到其余机器，再使用cat追加到authorized_keys文件中

3. 下载hadoop-2.x.y.tar.gz

[root@hftclclw0001 hadoop]# pwd
/home/hadoop

[root@hftclclw0001 hadoop]# tar -zxvf hadoop-2.7.1.tar.gz

[root@hftclclw0001 hadoop]# ll
total 546584
drwx------ 11 hadoop root      4096 Oct 20 09:05 hadoop-2.7.1
-rw-------  1 hadoop root 210606807 Oct 20 09:00 hadoop-2.7.1.tar.gz
drwx------ 13 hadoop root      4096 Oct 20 09:22 spark-1.5.1-bin-hadoop2.6
-rw-------  1 hadoop root 280901736 Oct 20 09:19 spark-1.5.1-bin-hadoop2.6.tgz
drwx------ 22 hadoop root      4096 Oct 21 00:07 sqoop-1.99.6-bin-hadoop200
-rw-------  1 hadoop root  68177818 May  5 22:34 sqoop-1.99.6-bin-hadoop200.tar.gz

4.配置hadoop-2.x.y

[hadoop@hftclclw0001 hadoop]$ pwd
/home/hadoop/hadoop-2.7.1/etc/hadoop

[hadoop@hftclclw0001 hadoop]$ vi hadoop-env.sh 
# The java implementation to use.
export JAVA_HOME=/usr/java/latest            => 配置java_home

[hadoop@hftclclw0001 hadoop]$ vi core-site.xml
<configuration>
	<property>
		<name>hadoop.tmp.dir</name>
		<value>/home/hadoop/hadoop-2.7.1/tmp</value>    => 需建立，默认在/tmp下
	</property>
	<property>
		<name>fs.defaultFS</name>
		<value>hdfs://{master:IP}:9000</value>          
	</property>
</configuration>

[hadoop@hftclclw0001 hadoop]$ vi hdfs-site.xml
<configuration>
	<property>
		<name>dfs.http.address</name>
		<value>{master:ip}:50070</value>
	</property>
	<property>
		<name>dfs.replication</name>
		<value>2</value>                        =>我这有3台机器，2台datanode 1台Namenode
	</property>
</configuration>

[hadoop@hftclclw0001 hadoop]$ vi mapred-site.xml
<configuration>
	<property>
		<name>mapreduce.framework.name</name>
		<value>yarn</value>
	</property>
</configuration>

[hadoop@hftclclw0001 hadoop]$ vi yarn-env.sh
...
export JAVA_HOME=/usr/java/latest
...

[hadoop@hftclclw0001 hadoop]$ vi yarn-site.xml 
	<property>
                <name>yarn.resourcemanager.hostname</name>     =>须要配置，在启动时候nodemanager会访问resouremanager
                <value>{master:ip}</value>
        </property> 
	<property>
		<name>yarn.nodemanager.aux-services</name>
		<value>mapreduce_shuffle</value>
	</property>
</configuration>

[hadoop@hftclclw0001 hadoop]$ vi masters                        =>其实做用的secondary namenode在那个节点上
{master:ip}

[hadoop@hftclclw0001 hadoop]$ vi slaves                         =>做用的datanode再那些节点上
{slave-1:ip}
{slave-2:ip}

5. 复制到其余机器

[hadoop@hftclclw0001 ~]$ pwd
/home/hadoop

[hadoop@hftclclw0001 ~]$ scp -r hadoop-2.7.1 hadoop@{ip}:/home/hadoop

6.启动

[hadoop@hftclclw0001 hadoop-2.7.1]$ ./bin/hadoop namenode -format

[hadoop@hftclclw0001 hadoop-2.7.1]$ pwd
/home/hadoop/hadoop-2.7.1

[hadoop@hftclclw0001 hadoop-2.7.1]$ ./sbin/start-dfs.sh    => 启动dfs， jps查看进程

master:namenode ,secondary namenode
slave:datanode

[hadoop@hftclclw0001 hadoop-2.7.1]$ ./sbin/start-yarn.sh   =>启动yarn

7.验证

     a.jps       => 校验各个进程

    b.netstat => 校验端口

    c.webui   => 能够校验cluster总体情况 

    d.也能够操做hdfs、或是submit mr job

[hadoop@hftclclw0001 hadoop-2.7.1]$ pwd
/home/hadoop/hadoop-2.7.1

[hadoop@hftclclw0001 hadoop-2.7.1]$ ./bin/hdfs dfs -ls /
...
...

[hadoop@hftclclw0001 hadoop-2.7.1]$ ./bin/hdfs dfs -mkdir /test
...
...

[hadoop@hftclclw0001 hadoop-2.7.1]$ ./bin/hadoop jar share/hadoop/mapreduce/hadoop-mapreduce-examples-2.7.1.jar wordcount {in} {out}

[hadoop@hftclclw0001 hadoop-2.7.1]$ ./bin/hadoop jar share/hadoop/mapreduce/hadoop-mapreduce-examples-2.7.1.jar pi 10 10

2、trouble shooting

1. 文件写入权限问题

   当外部程序写入hdfs时，默认都要进行用户认证。如按照上述配置，只能hadoop帐户能够写hdfs

   dfs.premissions.enabled=true 即对用户进行认证。修改成false

   dfs.datanode.data.dir.perm=700 即本地目录的写入权限。修改成755