Apache Hadoop 集群安装文档

简介：

Apache Hadoop 集群安装文档

软件：jdk-8u111-linux-x64.rpm、hadoop-2.8.0.tar.gz

http://www.apache.org/dyn/closer.cgi/hadoop/common/hadoop-2.8.0/hadoop-2.8.0.tar.gz

  系统：CentOS 6.8 x64
  
  主机列表及配置信息：
  
                    master.hadoop     datanode[01:03].hadoop

  CPU：                  8                     4

  MEM:                  16G                    8G

  DISK：               100G*2                100G*2

一、系统初始化

# master.hadoop

shell > vim /etc/hosts

192.168.1.25  master.hadoop
192.168.1.27  datanode01.hadoop
192.168.1.28  datanode02.hadoop
192.168.1.29  datanode03.hadoop

shell > yum -y install epel-release
shell > yum -y install ansible

shell > ssh-keygen  # 生成密钥
shell > ssh-copy-id -i ~/.ssh/id_rsa.pub "-p 22 root@datanode01.hadoop"
shell > ssh-copy-id -i ~/.ssh/id_rsa.pub "-p 22 root@datanode02.hadoop"
shell > ssh-copy-id -i ~/.ssh/id_rsa.pub "-p 22 root@datanode03.hadoop"

shell > vim /etc/ansible/hosts

# datanode.hadoop

[datanode]

datanode[01:03].hadoop

shell > ansible datanode -m shell -a 'useradd hadoop && echo hadoop | passwd --stdin hadoop'

shell > ansible datanode -m shell -a "echo '* - nofile 65536' >> /etc/security/limits.conf"

shell > ansible datanode -m copy -a 'src=/etc/hosts dest=/etc/hosts'  # 同步 hosts

shell > ansible datanode -m shell -a '/etc/init.d/iptables stop && chkconfig --del iptables'  # 关闭防火墙

shell > ansible datanode -m shell -a 'sed -i '/SELINUX/s/enforcing/disabled/' /etc/selinux/config'  # 关闭 SELinux

shell > ansible datanode -m shell -a 'echo 'vm.swappiness = 0' >> /etc/sysctl.conf'  # 修改内核参数

shell > ansible datanode -m shell -a 'echo 'echo never > /sys/kernel/mm/redhat_transparent_hugepage/defrag'  # 关闭透明大页

shell > ansible datanode -m shell -a 'echo 'echo never > /sys/kernel/mm/redhat_transparent_hugepage/defrag' >> /etc/rc.local'

shell > ansible datanode -m shell -a 'reboot'

# 上面的 ansible 操作，master.hadoop 也要执行

二、时间同步

# master.hadoop

shell > /bin/cp -f /usr/share/zoneinfo/Asia/Shanghai /etc/localtime

shell > yum -y install ntp

shell > /etc/init.d/ntpd stop && chkconfig --del ntpd

shell > ntpdate us.pool.ntp.org | hwclock -w

shell > vim /etc/ntp.conf
# 允许时间同步客户端
restrict 192.168.1.0 mask 255.255.255.0 nomodify
# Server 向谁同步时间
server us.pool.ntp.org prefer
# Server 无法向时间服务器同步时，使用本地时钟
server 127.127.1.0
fudge 127.127.1.0 stratum 10

shell > /etc/init.d/ntpd start

shell > echo -e '
/usr/sbin/ntpdate us.pool.ntp.org | hwclock -w > /dev/null' >> /etc/rc.local

shell > echo -e '
/etc/init.d/ntpd start > /dev/null' >> /etc/rc.local

shell > ansible datanode -m shell -a 'yum -y install ntpdate'

shell > ansible datanode -m shell -a '/bin/cp -f /usr/share/zoneinfo/Asia/Shanghai /etc/localtime'

shell > ansible datanode -m shell -a 'ntpdate master.hadoop | hwclock -w'

shell > ansible datanode -m cron -a "name='ntpdate master.hadoop' minute=0 hour=0 job='/usr/sbin/ntpdate master.hadoop | hwclock -w > /dev/null'"

三、集群部署

# master.hadoop

1、安装 jdk、下载、解压 apache hadoop、设置主机间 hadoop 用户无密码登录

shell > rpm -ivh /usr/local/src/jdk-8u111-linux-x64.rpm

shell > echo 'export JAVA_HOME=/usr/java/default' >> /etc/profile && source /etc/profile

shell > tar zxf /usr/local/src/hadoop-2.8.0.tar.gz -C /usr/local/

shell > chown -R hadoop.hadoop /usr/local/hadoop-2.8.0

shell > echo -e '
export PATH=$PATH:/usr/local/hadoop-2.8.0/bin' >> /etc/profile && source /etc/profile

shell > su - hadoop

hadoop shell > ssh-keygen

hadoop shell > cat .ssh/id_rsa.pub > .ssh/authorized_keys && chmod 600 .ssh/authorized_keys

hadoop shell > ssh-copy-id -i ~/.ssh/id_rsa.pub "-p 22 hadoop@datanode01.hadoop"
hadoop shell > ssh-copy-id -i ~/.ssh/id_rsa.pub "-p 22 hadoop@datanode02.hadoop"
hadoop shell > ssh-copy-id -i ~/.ssh/id_rsa.pub "-p 22 hadoop@datanode03.hadoop"

2、配置 apache hadoop

# 指定 Slave、即 DataNode、NodeManager 角色

hadoop shell > vim /usr/local/hadoop-2.8.0/etc/hadoop/slaves
datanode01.hadoop
datanode02.hadoop
datanode03.hadoop

# 修改 hadoop-env.sh

hadoop shell > vim /usr/local/hadoop-2.8.0/etc/hadoop/hadoop-env.sh

export JAVA_HOME=/usr/java/default

# 修改 core-site.xml

<configuration>

    <property>
        <name>fs.defaultFS</name>
        <value>hdfs://master.hadoop:8020</value>
    </property>

    <property>
        <name>hadoop.tmp.dir</name>
        <value>file:///data/hadoop/tmp</value>
    </property>

    <property>
        <name>fs.trash.interval</name>
        <value>1440</value>
    </property>

    <property>
        <name>io.file.buffer.size</name>
        <value>131072</value>
    </property>

</configuration>

# hadoop 核心配置文件
# 默认加载项 HADOOP_HOME/share/doc/hadoop/hadoop-project-dist/hadoop-common/core-default.xml

# fs.defaultFS NameNode IP:PORT，老版本为 fs.default.name
# hadoop.tmp.dir hadoop 临时目录，很多目录不明确配置时，都基于该目录 ( 默认 /tmp，系统重启时会被删除 )，很重要！
# fs.trash.interval 开启垃圾回收，1440 分钟，默认 0 关闭 ( 用户文件系统级删除的数据会被移到回收站，24小时后被删除 )
# io.file.buffer.size 读写流文件缓存大小，减少IO次数，默认 4096 字节

# 修改 hdfs-site.xml

hadoop shell > vim /usr/local/hadoop-2.8.0/etc/hadoop/hdfs-site.xml

<configuration>

    <property>
        <name>dfs.blocksize</name>
        <value>134217728</value>
    </property>

    <property>
        <name>dfs.replication</name>
        <value>3</value>
    </property>

    <property>
        <name>dfs.namenode.name.dir</name>
        <value>file:///data/dfs/nn</value>
    </property>

    <property>
        <name>dfs.namenode.checkpoint.dir</name>
        <value>file:///data/dfs/sn</value>
    </property>

    <property>
        <name>dfs.datanode.data.dir</name>
        <value>file:///data/dfs/dn</value>
    </property>

    <property>
        <name>dfs.namenode.handler.count</name>
        <value>20</value>
    </property>
    
</configuration>

# HDFS 配置文件
# 默认加载项 HADOOP_HOME/share/doc/hadoop/hadoop-project-dist/hadoop-hdfs/hdfs-default.xml

# dfs.hosts / dfs.hosts.exclude 允许或排除某些 DataNode 连接 NameNode

# dfs.blocksize 块大小，默认 134217728 ( 128M )
# dfs.replication 默认副本数，数据冗余
# dfs.namenode.name.dir NameNode 元数据存放位置，可以配置多个目录，以 , 分割，用作数据冗余！
# dfs.namenode.checkpoint.dir SecondaryNameNode 数据存储目录，该角色负责将 NameNode 的 edit log 合并到 fsimage
# dfs.datanode.data.dir DataNode 数据存放位置，可以配置多个目录，以 , 分割，数据轮询写入，增加写入速度 ( 多个目录应该对应多个设备 DISK )
# dfs.namenode.handler.count NameNode 线程数，用于跟 DataNode 通信，默认 10，增大该参数可以优化性能，但是资源也相应提升

# 修改 yarn-site.xml

hadoop shell > vim /usr/local/hadoop-2.8.0/etc/hadoop/yarn-site.xml

<configuration>

    <property>
        <name>yarn.resourcemanager.hostname</name>
        <value>master.hadoop</value>
    </property>

    <property>
      <name>yarn.resourcemanager.scheduler.class</name>
      <value>org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.CapacityScheduler</value>
    </property>

    <property>
        <name>yarn.nodemanager.log-dirs</name>
        <value>${yarn.log.dir}/userlogs</value>
    </property>

    <property>
        <name>yarn.nodemanager.remote-app-log-dir</name>
        <value>/tmp/logs</value>
    </property>

    <property>
        <name>yarn.nodemanager.aux-services</name>
        <value>mapreduce_shuffle</value>
    </property>

</configuration>

# YARN 配置文件
# 默认加载项 HADOOP_HOME/share/doc/hadoop/hadoop-yarn/hadoop-yarn-common/yarn-default.xml

# yarn.resourcemanager.hostname ReSourceManager 主机，其余关于端口的监听都基于该配置项
# yarn.resourcemanager.scheduler.class 资源调度算法，CapacityScheduler 计算能力调度、FairScheduler 公平调度、Fifo Scheduler 先进先出调度
# yarn.nodemanager.log-dirs nodemanager 日志目录
# yarn.nodemanager.remote-app-log-dir nodemanager 中间结果保持目录

# 修改 mapred-site.xml

hadoop shell > cat /usr/local/hadoop-2.8.0/etc/hadoop/mapred-site.xml.template > /usr/local/hadoop-2.8.0/etc/hadoop/mapred-site.xml
hadoop shell > vim /usr/local/hadoop-2.8.0/etc/hadoop/mapred-site.xml

<configuration>

    <property>
        <name>mapreduce.framework.name</name>
        <value>yarn</value>
    </property>

    <property>
        <name>mapreduce.jobhistory.address</name>
        <value>master.hadoop:10020</value>
    </property>

    <property>
        <name>mapreduce.jobhistory.webapp.address</name>
        <value>master.hadoop:19888</value>
    </property>

    <property>
        <name>yarn.app.mapreduce.am.staging-dir</name>
        <value>/tmp/hadoop-yarn/staging</value>
    </property>

</configuration>

# MAPREDUCE 配置文件
# 默认加载项 HADOOP_HOME/share/doc/hadoop/hadoop-mapreduce-client/hadoop-mapreduce-client-core/mapred-default.xml

# mapreduce.framework.name 使用 yarn 来管理资源
# yarn.app.mapreduce.am.staging-dir 提交作业时的临时目录，提交作业历史目录 mapreduce.jobhistory.done-dir、mapreduce.jobhistory.intermediate-done-dir 都基于该目录

hadoop shell > exit

3、部署 Slave

shell > ansible datanode -m copy -a 'src=/usr/local/src/jdk-8u111-linux-x64.rpm dest=/usr/local/src/'

shell > yum -y install rsync

shell > ansible datanode -m shell -a 'yum -y install rsync'

shell > ansible datanode -m synchronize -a 'src=/usr/local/hadoop-2.8.0 dest=/usr/local/'

# 我还傻傻的用 copy 模块，结果慢的要死，synchroize 为 rsync 模块，好快！

shell > ansible datanode -m shell -a 'rpm -ivh /usr/local/src/jdk-8u111-linux-x64.rpm'

shell > ansible datanode -m shell -a "echo -e '
export JAVA_HOME=/usr/java/default' >> /etc/profile && source /etc/profile"

四、启动集群

# master.hadoop

shell > chmod -R a+w /data
shell > ansible datanode -m shell -a 'chmod -R a+w /data'

# 需要给 /data 目录写入权限，否则无法初始化文件系统 hdfs namenode -format

shell > su - hadoop

hadoop shell > hdfs namenode -format  # 初次启动需要格式化文件系统

hadoop shell > sh /usr/local/hadoop-2.8.0/sbin/start-all.sh  # 启动所有服务 / stop-all.sh 关闭服务

hadoop shell > jps
4386 ResourceManager
4659 Jps
3990 NameNode
4204 SecondaryNameNode

# 这是 master.hadoop 启动的角色
# http://192.168.1.25:50070 # NameNode
# http://192.168.1.25:8088  # ReSourceManagerv
# http://192.168.1.25:10020  # MapReduce JobHistory Server :19888 webui

# datanode.hadoop

hadoop shell > jps
2508 Jps
2238 DataNode
2351 NodeManager

# 这是 datanode.hadoop 启动的角色

hadoop shell > hdfs dfs -ls
ls: `.': No such file or directory

hadoop shell > hdfs dfs -mkdir /user
hadoop shell > hdfs dfs -mkdir /user/hadoop

hadoop shell > hdfs dfs -ls

# 为 hadoop 用户创建家目录

五、运行示例

# master.hadoop

hadoop shell > hdfs dfs -put shakespeare.txt  # 上传本地文件到 hdfs
hadoop shell > hdfs dfs -ls
Found 1 items
-rw-r--r--   3 hadoop supergroup    5447165 2017-05-17 16:49 shakespeare.txt

hadoop shell > hadoop jar /usr/local/hadoop-2.8.0/share/hadoop/mapreduce/hadoop-mapreduce-examples-2.8.0.jar grep shakespeare.txt outfile what

# 执行官方示例，词频统计 ( 统计 what 出现次数 )

hadoop shell > hdfs dfs -ls
drwxr-xr-x   - hadoop supergroup          0 2017-04-11 19:38 outfile
-rw-r--r--   3 hadoop supergroup    5447165 2017-04-11 19:35 shakespeare.txt

hadoop shell > hdfs dfs -cat outfile/*
2309    what

报错管理：

1、bin/hdfs namenode -format # 初始化文件系统报错

17/04/01 19:04:29 ERROR namenode.NameNode: Failed to start namenode.
java.io.IOException: Cannot create directory /data/dfs/namenode/current
    at org.apache.hadoop.hdfs.server.common.Storage$StorageDirectory.clearDirectory(Storage.java:352)
    at org.apache.hadoop.hdfs.server.namenode.NNStorage.format(NNStorage.java:573)
    at org.apache.hadoop.hdfs.server.namenode.NNStorage.format(NNStorage.java:594)
    at org.apache.hadoop.hdfs.server.namenode.FSImage.format(FSImage.java:156)
    at org.apache.hadoop.hdfs.server.namenode.NameNode.format(NameNode.java:1102)
    at org.apache.hadoop.hdfs.server.namenode.NameNode.createNameNode(NameNode.java:1544)
    at org.apache.hadoop.hdfs.server.namenode.NameNode.main(NameNode.java:1671)

# 解决方法

shell > chmod -R a+w /data
shell > ansible datanode -m shell -a 'chmod -R a+w /data'

2、WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable # 迷之警告