Docker创建的集群下使用ansible部署hadoop

基础环境

MBP, Palallels Desktop, Centos7

关键词

docker, ansible, hadoop

集群架构

集群包含4台“虚拟主机”，采用Docker创建容器的方式创建，无需创建多台虚拟机，简单方便。

Docker

docker 安装

登录到Centos 7寄主机中安装docker

[root@centos-linux ~]# yum -y install docker.x86_64

启动docker服务

[root@centos-linux ~]# systemctl start docker

Centos镜像拉取

国内使用docker.io拉取镜像的时候非常慢，所以找了镜像仓库进行拉取。
本次用到的镜像仓库是http://hub.daocloud.io, 在搜索栏中输入centos并搜索

在检索到的镜像详情右侧给出了pull的命令，直接运行即可。

[root@centos-linux ~]# docker pull daocloud.io/library/centos:latest

拉取完成之后可以看到

[root@centos-linux ~]# docker images
REPOSITORY                   TAG                 IMAGE ID            CREATED             SIZE
daocloud.io/library/centos   latest              328edcd84f1b        3 weeks ago         192.5 MB

创建容器

按照集群的架构，创建容器时需要设置固定IP，所以先要在docker下创建固定IP的网络组

[root@centos-linux ~]# docker network create --subnet=172.18.0.0/16 netgroup

docker的网络组创建完成之后就可以创建固定IP的容器了

[root@centos-linux ~]# docker run -d --privileged -ti -v /sys/fs/cgroup:/sys/fs/cgroup --name cluster-master -h cluster-master --net netgroup --ip 172.18.0.2 daocloud.io/library/centos /usr/sbin/init
[root@centos-linux ~]# docker run -d --privileged -ti -v /sys/fs/cgroup:/sys/fs/cgroup --name cluster-slave1 -h cluster-slave1 --net netgroup --ip 172.18.0.3 daocloud.io/library/centos /usr/sbin/init
[root@centos-linux ~]# docker run -d --privileged -ti -v /sys/fs/cgroup:/sys/fs/cgroup --name cluster-slave2 -h cluster-slave2 --net netgroup --ip 172.18.0.4 daocloud.io/library/centos /usr/sbin/init
[root@centos-linux ~]# docker run -d --privileged -ti -v /sys/fs/cgroup:/sys/fs/cgroup --name cluster-slave3 -h cluster-slave3 --net netgroup --ip 172.18.0.5 daocloud.io/library/centos /usr/sbin/init

在Centos7下使用简单方式创建容器后遇到sshd启动失败的问题，所以需要添加参数--privileged和-v /sys/fs/cgroup:/sys/fs/cgroup，并在启动的时候运行/usr/sbin/init。

openssh

创建的centos容器中并没有预装ssh服务，需要登录到所有容器中手动安装。

登录容器

[root@centos-linux ~]# docker ps -a
CONTAINER ID        IMAGE                        COMMAND             CREATED             STATUS              PORTS               NAMES
328913df0d52        daocloud.io/library/centos   "/usr/sbin/init"    1 hours ago        Up 1 hours                             cluster-slave3
83361d4bf079        daocloud.io/library/centos   "/usr/sbin/init"    1 hours ago        Up 1 hours                             cluster-slave2
35d1e5732340        daocloud.io/library/centos   "/usr/sbin/init"    1 hours ago        Up 1 hours                             cluster-slave1
37fbcc24b0e3        daocloud.io/library/centos   "/usr/sbin/init"    1 hours ago        Up 1 hours                             cluster-master

[root@centos-linux ~]# docker exec -it 37fbcc24b0e3 /bin/bash

安装openssh

[root@cluster-master /]# yum -y install openssh openssh-server openssh-clients

启动ssh服务

[root@cluster-master /]# systemctl start sshd

ssh自动接受新的公钥

master设置ssh登录自动添加kown_hosts
vi编辑/etc/ssh/ssh_config配置文件
设置StrictHostKeyChecking为no

配置cluster-master的hosts

/etc/hosts文件在容器启动时被重写，直接修改内容在容器重启后不能保留，为了让容器在重启之后获取集群hosts，使用了一种启动容器后重写hosts的方法。
需要在~/.bashrc中追加以下指令

:>/etc/hosts
cat >>/etc/hosts<<EOF
127.0.0.1   localhost
172.18.0.2  cluster-master
172.18.0.3  cluster-slave1
172.18.0.4  cluster-slave2
172.18.0.5  cluster-slave3
EOF

运行source ~/.bashrc使之生效
这时可以看到/etc/hosts文件已经被改为需要的内容

[root@cluster-master /]# cat /etc/hosts
127.0.0.1   localhost
172.18.0.2  cluster-master
172.18.0.3  cluster-slave1
172.18.0.4  cluster-slave2
172.18.0.5  cluster-slave3

cluster-master公钥分发

在master机上执行ssh-keygen -t rsa并一路回车，完成之后会生成~/.ssh目录，目录下有id_rsa（私钥文件）和id_rsa.pub（公钥文件），再将id_rsa.pub重定向到文件authorized_keys

[root@cluster-master /]# cd ~/.ssh
[root@cluster-master /]# cat id_rsa.pub>authorized_keys

文件生成之后用scp将公钥文件分发到集群slave主机

[root@cluster-master /]# ssh root@cluster-slave1 'mkdir ~/.ssh'
[root@cluster-master /]# scp ~/.ssh/authorized_keys root@cluster-slave1:~/.ssh
[root@cluster-master /]# ssh root@cluster-slave2 'mkdir ~/.ssh'
[root@cluster-master /]# scp ~/.ssh/authorized_keys root@cluster-slave2:~/.ssh
[root@cluster-master /]# ssh root@cluster-slave3 'mkdir ~/.ssh'
[root@cluster-master /]# scp ~/.ssh/authorized_keys root@cluster-slave3:~/.ssh

分发完成之后测试是否已经可以免输入密码登录。另外本次实验使用到了root用户，如果在其他用户下使用免密码登录，需要确保用户对~/.ssh/authorized_keys文件有可操作权限。

Ansible

源码安装ansible

使用git将文件安装在cluster-master的/opt目录下，因为无需agent，所以安装在一台主控机上即可管理集群。容器中的centos没有预装git，在安装之前需要事先安装git:

[root@cluster-master /]# yum -y install git

在/opt下执行

[root@cluster-master opt]# git clone git://github.com/ansible/ansible.git --recursive

在执行ansible的env-setup之前需要安装一些容器中没有预装的软件包

[root@cluster-master opt]# yum -y install python-setuptools
[root@cluster-master opt]# easy_install pip
[root@cluster-master opt]# pip install paramiko PyYAML Jinja2 httplib2 six

进入ansible目录，执行安装脚本

[root@cluster-master opt]# cd ./ansible
[root@cluster-master ansible]# source ./hacking/env-setup

配置ansible hosts

创建/etc/ansible/hosts文件，并将cluster内的主机按组的方式写入该文件

[cluster]
cluster-master
cluster-slave1
cluster-slave2
cluster-slave3

[master]
cluster-master

[slaves]
cluster-slave1
cluster-slave2
cluster-slave3

Hadoop

在集群中安装openjdk

使用ansible在在集群中安装openjdk

[root@cluster-master ansible]# ansible cluster -m yum -a "name=java-1.8.0-openjdk,java-1.8.0-openjdk-devel state=latest"

在cluster-master上安装hadoop

将hadoop安装包下载至/opt目录下

[root@cluster-master opt]# wget http://mirrors.tuna.tsinghua.edu.cn/apache/hadoop/common/hadoop-2.7.4/hadoop-2.7.4.tar.gz

下载完成之后解压安装包，并创建链接文件

[root@cluster-master opt]# tar -xzvf hadoop-2.7.4.tar.gz
[root@cluster-master opt]# ln -s hadoop-2.7.4 hadoop

设置java和hadoop环境变量(.bashrc)

# java
export JAVA_HOME=/usr/lib/jvm/java-1.8.0-openjdk-1.8.0.141-1.b16.el7_3.x86_64
export PATH=$HADOOP_HOME/bin:$PATH
# hadoop
export HADOOP_HOME=/opt/hadoop
export PATH=$HADOOP_HOME/bin:$HADOOP_HOME/sbin:$PATH

修改slaves文件

[root@cluster-master opt]# vi /opt/hadoop/etc/hadoop/slaves
cluster-slave1
cluster-slave2
cluster-slave3

修改hadoop运行所需配置文件

core-site.xml

<configuration>
    <property>
        <name>hadoop.tmp.dir</name>
        <value>/home/hadoop/tmp</value>
        <description>A base for other temporary directories.</description>
    </property>
    <!-- file system properties -->
    <property>
        <name>fs.default.name</name>
       <value>hdfs://cluster-master:9000</value>
    </property>
</configuration>

hdfs-site.xml

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

mapred-site.xml

<configuration>
    <property>
        <name>mapred.job.tracker</name>
        <value>http://cluster-master:9001</value>
    </property>
</configuration>

yarn-site.xml

<configuration>
    <property>
        <name>yarn.nodemanager.aux-services</name>
        <value>mapreduce_shuffle</value>
    </property>
    <property>
        <name>yarn.resourcemanager.address</name>
        <value>cluster-master:18040</value>
    </property>
</configuration>

打包hadoop文件

将hadoop链接文件和hadoop-2.7.4打包成一个文件方便ansible分发到slave主机

[root@cluster-master opt]# tar -cvf hadoop-dis.tar hadoop hadoop-2.7.4

使用ansible-playbook分发.bashrc和hadoop-dis.tar至slave主机

---
- hosts: cluster
  tasks:
    - name: copy .bashrc to slaves
      copy: src=~/.bashrc dest=~/
      notify:
        - exec source
    - name: mkdir /home/hadoop/tmp
      shell: mkdir -p /home/hadoop/tmp
      
    - name: copy hadoop-dis.tar to slaves
      unarchive: src=/opt/hadoop-dis.tar dest=/opt


  handlers:
    - name: exec source
      shell: source ~/.bashrc

将以上yaml保存为hadoop-dis.yaml，并执行

[root@cluster-master opt]# ansible-playbook hadoop-dis.yaml

hadoop-dis.tar会自动解压到slave主机的/opt目录下。

格式化namenode

[root@cluster-master opt]# hadoop namenode -format

启动hadoop集群

到这一步已经可以开始hadoop之旅了，启动比较简单，在$HADOOP_HOME/sbin下有几个启动和停止的脚本如下：

[root@cluster-master opt]# cd $HADOOP_HOME/sbin
[root@cluster-master sbin]# ls -l
total 120
-rwxr-xr-x. 1 20415 101 2752 Aug  1 00:35 distribute-exclude.sh
-rwxr-xr-x. 1 20415 101 6452 Aug  1 00:35 hadoop-daemon.sh
-rwxr-xr-x. 1 20415 101 1360 Aug  1 00:35 hadoop-daemons.sh
-rwxr-xr-x. 1 20415 101 1640 Aug  1 00:35 hdfs-config.cmd
-rwxr-xr-x. 1 20415 101 1427 Aug  1 00:35 hdfs-config.sh
-rwxr-xr-x. 1 20415 101 2291 Aug  1 00:35 httpfs.sh
-rwxr-xr-x. 1 20415 101 3128 Aug  1 00:35 kms.sh
-rwxr-xr-x. 1 20415 101 4080 Aug  1 00:35 mr-jobhistory-daemon.sh
-rwxr-xr-x. 1 20415 101 1648 Aug  1 00:35 refresh-namenodes.sh
-rwxr-xr-x. 1 20415 101 2145 Aug  1 00:35 slaves.sh
-rwxr-xr-x. 1 20415 101 1779 Aug  1 00:35 start-all.cmd
-rwxr-xr-x. 1 20415 101 1471 Aug  1 00:35 start-all.sh
-rwxr-xr-x. 1 20415 101 1128 Aug  1 00:35 start-balancer.sh
-rwxr-xr-x. 1 20415 101 1401 Aug  1 00:35 start-dfs.cmd
-rwxr-xr-x. 1 20415 101 3734 Aug  1 00:35 start-dfs.sh
-rwxr-xr-x. 1 20415 101 1357 Aug  1 00:35 start-secure-dns.sh
-rwxr-xr-x. 1 20415 101 1571 Aug  1 00:35 start-yarn.cmd
-rwxr-xr-x. 1 20415 101 1347 Aug  1 00:35 start-yarn.sh
-rwxr-xr-x. 1 20415 101 1770 Aug  1 00:35 stop-all.cmd
-rwxr-xr-x. 1 20415 101 1462 Aug  1 00:35 stop-all.sh
-rwxr-xr-x. 1 20415 101 1179 Aug  1 00:35 stop-balancer.sh
-rwxr-xr-x. 1 20415 101 1455 Aug  1 00:35 stop-dfs.cmd
-rwxr-xr-x. 1 20415 101 3206 Aug  1 00:35 stop-dfs.sh
-rwxr-xr-x. 1 20415 101 1340 Aug  1 00:35 stop-secure-dns.sh
-rwxr-xr-x. 1 20415 101 1642 Aug  1 00:35 stop-yarn.cmd
-rwxr-xr-x. 1 20415 101 1340 Aug  1 00:35 stop-yarn.sh
-rwxr-xr-x. 1 20415 101 4295 Aug  1 00:35 yarn-daemon.sh
-rwxr-xr-x. 1 20415 101 1353 Aug  1 00:35 yarn-daemons.sh

主要使用HDFS，启动start-dfs.sh即可

[root@cluster-master sbin]# ./start-dfs.sh
Starting namenodes on [cluster-master]
cluster-master: starting namenode, logging to /opt/hadoop-2.7.4/logs/hadoop-root-namenode-cluster-master.out
cluster-slave1: starting datanode, logging to /opt/hadoop-2.7.4/logs/hadoop-root-datanode-cluster-slave1.out
cluster-slave3: starting datanode, logging to /opt/hadoop-2.7.4/logs/hadoop-root-datanode-cluster-slave3.out
cluster-slave2: starting datanode, logging to /opt/hadoop-2.7.4/logs/hadoop-root-datanode-cluster-slave2.out
Starting secondary namenodes [0.0.0.0]
0.0.0.0: starting secondarynamenode, logging to /opt/hadoop-2.7.4/logs/hadoop-root-secondarynamenode-cluster-master.out

启动完成之后会在master和slave看到以下进程

cluster-master

[root@cluster-master sbin]# jps
2484 SecondaryNameNode
2648 Jps
2315 NameNode

cluster-slave

[root@cluster-slave1 logs]# jps
27502 DataNode
27583 Jps

体验hadoop

在master端上传文件试试

[root@cluster-master sbin]# hadoop dfs -put start-dfs.sh /
DEPRECATED: Use of this script to execute hdfs command is deprecated.
Instead use the hdfs command for it.

[root@cluster-master sbin]# hadoop dfs -ls /
DEPRECATED: Use of this script to execute hdfs command is deprecated.
Instead use the hdfs command for it.

Found 1 items
-rw-r--r--   3 root supergroup       3734 2017-08-26 10:57 /start-dfs.sh

总结

docker下部署hadoop确实比较奇葩，但是对于个人学习来说确是十分方便，不用在虚拟机里复制多套环境，也简化了网络配置的步骤，加上ansible的使用使得集群的管理更加高效。