Hadoop2.6.0自动化部署脚本（一）

1 概述

最近自己写了一个Hadoop自动化部署脚本，包括Hadoop集群自动化部署脚本和Hadoop增加单节点自动化部署脚本。需要快速部署Hadoop集群的童鞋可以使用该脚本。这些脚本我在用5台虚拟机进行了测试，如果在使用中还有bug，欢迎指出。本文主要介绍Hadoop集群自动化部署脚本，安装的Hadoop版本为2.6.0。

2 依赖

安装Hadoop2.6.0集群需要依赖JDK和Zookeeper。本文安装的JDK版本为jdk-7u60-linux-x64，Zookeeper版本为zookeeper-3.4.6。

3 各文件及配置说明

该部署脚本由两部分构成：root用户下执行的脚本和Hadoop启动用户下执行的脚本。这些脚本都只需要在一台服务器上执行即可，执行脚本的服务器作为Hadoop的Master服务器。下面分别进行说明。

3.1 root脚本说明

root脚本的目录结构如下：

conf — 配置文件目录

init.conf

expect — expect脚本目录

password.expect

scp.expect

otherInstall.expect

file — 安装文件目录

hadoop-2.6.0.tar.gz

jdk-7u60-linux-x64.tar.gz

zookeeper-3.4.6.tar.gz

installRoot.sh — 脚本执行文件

3.1.1 conf目录

该目录下的init.conf文件为root执行脚本使用的配置文件，在执行脚本之前需要对该配置文件进行修改。文件内容如下：

#jdk file and version
JDK_FILE_TAR=jdk-7u60-linux-x64.tar.gz

#jdk unpack name
JDK_FILE=jdk1.7.0_60

#java home
JAVAHOME=/usr/java

#Whether install the package for dependence,0 means no,1 means yes
IF_INSTALL_PACKAGE=1

#host conf
ALLHOST="hadoop1master hadoop1masterha hadoop1slave1 hadoop1slave2 hadoop1slave3"
ALLIP="192.168.0.180 192.168.0.184 192.168.0.181 192.168.0.182 192.168.0.183"

#zookeeper conf
ZOOKEEPER_TAR=zookeeper-3.4.6.tar.gz
ZOOKEEPERHOME=/usr/local/zookeeper-3.4.6
SLAVELIST="hadoop1slave1 hadoop1slave2 hadoop1slave3"

#hadoop conf
HADOOP_TAR=hadoop-2.6.0.tar.gz
HADOOPHOME=/usr/local/hadoop-2.6.0
HADOOP_USER=hadoop2
HADOOP_PASSWORD=hadoop2

#root conf: $MASTER_HA $SLAVE1 $SLAVE2 $SLAVE3
ROOT_PASSWORD="hadoop hadoop hadoop hadoop"

下面是个别参数的解释及注意事项：

ALLHOST为Hadoop集群各个服务器的hostname，使用空格分隔；ALLIP为Hadoop集群各个服务器的ip地址，使用空格分隔。要求ALLHOST和ALLIP要一一对应。

SLAVELIST为zookeeper集群部署的服务器的hostname。

ROOT_PASSWORD为除了Master服务器以外的其他服务器root用户的密码，使用逗号隔开。（在实际情况下，可能各个服务器的root密码并不相同。）

3.1.2 expect目录

该目录下包含password.expect、scp.expect、otherInstall.expect三个文件。password.expect用来设置hadoop启动用户的密码；scp.expect用来远程传输文件；otherInstall.expect用来远程执行其他服务器上的installRoot.sh。这三个文件都在installRoot.sh中被调用。

password.expect文件内容如下：

#!/usr/bin/expect -f
set user [lindex $argv 0]
set password [lindex $argv 1]
spawn passwd $user
expect "New password:"
send "$password\r"
expect "Retype new password:"
send "$password\r"
expect eof

其中argv 0和argv 1都是在installRoot.sh脚本中进行传值的。其他两个文件argv *也是这样传值的。

scp.expect文件内容如下：

#!/usr/bin/expect -f
# set dir, host, user, password
set dir [lindex $argv 0]
set host [lindex $argv 1]
set user [lindex $argv 2]
set password [lindex $argv 3]
set timeout -1
spawn scp -r $dir $user@$host:/root/
expect {
"(yes/no)?"
{
send "yes\n"
expect "*assword:" { send "$password\n"}
}
"*assword:"
{
send "$password\n"
}
}
expect eof

otherInstall.expect文件内容如下：

#!/usr/bin/expect -f
# set dir, host, user, password
set dir [lindex $argv 0]
set name [lindex $argv 1]
set host [lindex $argv 2]
set user [lindex $argv 3]
set password [lindex $argv 4]
set timeout -1
spawn ssh -q $user@$host "$dir/$name"
expect {
"(yes/no)?"
{
send "yes\n"
expect "*assword:" { send "$password\n"}
}
"*assword:"
{
send "$password\n"
}
}
expect eof

3.1.3 file目录

这里就是安装Hadoop集群及其依赖所需的安装包。

3.1.4 installRoot.sh脚本

该脚本是在root用户下需要执行的脚本，文件内容如下：

#!/bin/bash

if [ $USER != "root" ]; then
echo "[ERROR]:Must run as root";  exit 1
fi
# Get absolute path and name of this shell
readonly PROGDIR=$(readlink -m $(dirname $0))
readonly PROGNAME=$(basename $0)
hostname=`hostname`

source /etc/profile
# import init.conf
source $PROGDIR/conf/init.conf
echo "install start..."
# install package for dependence
if [ $IF_INSTALL_PACKAGE -eq 1 ]; then
yum -y install expect >/dev/null 2>&1
echo "expect install successful."
# yum install openssh-clients #scp
fi

#stop iptables or open ports, now stop iptables
service iptables stop
chkconfig iptables off
FF_INFO=`service iptables status`
if [ -n "`echo $FF_INFO | grep "Firewall is not running"`" ]; then
echo "Firewall is already stop."
else
echo "[ERROR]:Failed to shut down the firewall.Exit shell."
exit 1
fi
#stop selinux
setenforce 0
SL_INFO=`getenforce`
if [ $SL_INFO == "Permissive" -o $SL_INFO == "disabled" ]; then
echo "selinux is already stop."
else
echo "[ERROR]:Failed to shut down the selinux. Exit shell."
exit 1
fi

#host config
hostArr=( $ALLHOST )
IpArr=( $ALLIP )
for (( i=0; i <= ${#hostArr[@]}; i++ ))
{
if [ -z "`grep "${hostArr[i]}" /etc/hosts`" -o -z "`grep "${IpArr[i]}" /etc/hosts`" ]; then
echo "${IpArr[i]} ${hostArr[i]}" >> /etc/hosts
fi
}

#user config
groupadd $HADOOP_USER && useradd -g $HADOOP_USER $HADOOP_USER && $PROGDIR/expect/password.expect $HADOOP_USER $HADOOP_PASSWORD >/dev/null 2>&1

# check jdk
checkOpenJDK=`rpm -qa | grep java`
# already install openJDK ,uninstall
if [ -n "$checkOpenJDK" ]; then
rpm -e --nodeps $checkOpenJDK
echo "uninstall openJDK successful"
fi
# A way of exception handling. `java -version` perform error then perform after ||.
`java -version` || (
[ ! -d $JAVAHOME ] && ( mkdir $JAVAHOME )
tar -zxf $PROGDIR/file/$JDK_FILE_TAR -C $JAVAHOME
echo "export JAVA_HOME=$JAVAHOME/$JDK_FILE" >> /etc/profile
echo 'export JAVA_BIN=$JAVA_HOME/bin' >> /etc/profile
echo 'export PATH=$PATH:$JAVA_HOME/bin' >> /etc/profile
echo 'export CLASSPATH=.:$JAVA_HOME/lib/dt.jar:$JAVA_HOME/lib/tools.jar' >> /etc/profile
echo 'export JAVA_HOME JAVA_BIN PATH CLASSPATH' >> /etc/profile
echo "sun jdk done"
)

# check zookeeper
slaveArr=( $SLAVELIST )
if [[ "${slaveArr[@]}" =~ $hostname ]]; then
`zkServer.sh status` || [ -d  $ZOOKEEPERHOME ] || (
tar -zxf $PROGDIR/file/$ZOOKEEPER_TAR -C /usr/local/
chown -R $HADOOP_USER:$HADOOP_USER $ZOOKEEPERHOME
echo "export ZOOKEEPER_HOME=$ZOOKEEPERHOME" >> /etc/profile
echo 'PATH=$PATH:$ZOOKEEPER_HOME/bin' >> /etc/profile
echo "zookeeper done"
)
fi

# check hadoop2
`hadoop version` || [ -d  $HADOOPHOME ] || (
tar -zxf $PROGDIR/file/$HADOOP_TAR -C /usr/local/
chown -R $HADOOP_USER:$HADOOP_USER $HADOOPHOME
echo "export HADOOP_HOME=$HADOOPHOME" >> /etc/profile
echo 'PATH=$PATH:$HADOOP_HOME/bin' >> /etc/profile
echo 'HADOOP_HOME_WARN_SUPPRESS=1' >> /etc/profile
echo "hadoop2 done"
)
source /etc/profile

#ssh config
sed -i "s/^#RSAAuthentication\ yes/RSAAuthentication\ yes/g" /etc/ssh/sshd_config
sed -i "s/^#PubkeyAuthentication\ yes/PubkeyAuthentication\ yes/g" /etc/ssh/sshd_config
sed -i "s/^#AuthorizedKeysFile/AuthorizedKeysFile/g" /etc/ssh/sshd_config
sed -i "s/^GSSAPIAuthentication\ yes/GSSAPIAuthentication\ no/g" /etc/ssh/sshd_config
sed -i "s/^#UseDNS\ yes/UseDNS\ no/g" /etc/ssh/sshd_config
service sshd restart

# install other servers
rootPasswdArr=( $ROOT_PASSWORD )
if [ $hostname == ${hostArr[0]} ]; then
i=0
for node in $ALLHOST; do
if [ $hostname == $node ]; then
echo "this server, do nothing"
else
# cope install dir to other server
$PROGDIR/expect/scp.expect $PROGDIR $node $USER ${rootPasswdArr[$i]}
$PROGDIR/expect/otherInstall.expect $PROGDIR $PROGNAME $node $USER ${rootPasswdArr[$i]}
i=$(($i+1)) #i++
echo $node" install successful."
fi
done
# Let the environment variables take effect
su - root
fi

这个脚本主要干了下面几件事：

如果在配置文件中设置了IF_INSTALL_PACKAGE=1，则安装expect，默认是安装expect。如果服务器上已经有了expect，则可以设置IF_INSTALL_PACKAGE=0。

关闭防火墙，停止selinux。

将Hadoop集群的各个机器host及ip对应关系写到/etc/hosts文件中。

新建Hadoop启动用户及用户组。

安装jdk、zookeeper、hadoop并设置环境变量。

修改ssh配置文件/etc/ssh/sshd_config。

如果判断执行脚本的机器是Master机器，则拷贝本机的root脚本到其他机器上并执行。

注意：在执行该脚本之前，需要确保Hadoop集群安装的各个服务器上能够执行scp命令，如果不能执行，需要在各个服务器上安装openssh-clients，执行脚本为：yum –y install openssh-clients。

3.2 hadoop脚本说明

hadoop脚本的目录结构如下：

bin — 脚本目录

config_hadoop.sh

config_ssh.sh

config_zookeeper.sh

ssh_nopassword.expect

start_all.sh

conf — 配置文件目录

init.conf

template — 配置文件模板目录

core-site.xml

hadoop-env.sh

hdfs-site.xml

mapred-site.xml

mountTable.xml

myid

slaves

yarn-env.sh

yarn-site.xml

zoo.cfg

installCluster.sh — 脚本执行文件

3.2.1 bin脚本目录

该目录中包含installCluster.sh脚本中调用的所有脚本，下面一一说明。

3.2.1.1 config_hadoop.sh

该脚本主要是创建Hadoop所需目录，以及配置文件的配置，其中的参数均在init.conf中。

#!/bin/bash

# Get absolute path of this shell
readonly PROGDIR=$(readlink -m $(dirname $0))
# import init.conf
source $PROGDIR/../conf/init.conf

for node in $ALL; do
# create dirs
ssh -q $HADOOP_USER@$node "
mkdir -p $HADOOPDIR_CONF/hadoop2/namedir
mkdir -p $HADOOPDIR_CONF/hadoop2/datadir
mkdir -p $HADOOPDIR_CONF/hadoop2/jndir
mkdir -p $HADOOPDIR_CONF/hadoop2/tmp
mkdir -p $HADOOPDIR_CONF/hadoop2/hadoopmrsys
mkdir -p $HADOOPDIR_CONF/hadoop2/hadoopmrlocal
mkdir -p $HADOOPDIR_CONF/hadoop2/nodemanagerlocal
mkdir -p $HADOOPDIR_CONF/hadoop2/nodemanagerlogs
"
echo "$node create dir done."
for conffile in $CONF_FILE; do
# copy
scp $PROGDIR/../template/$conffile $HADOOP_USER@$node:$HADOOPHOME/etc/hadoop
# update
ssh -q $HADOOP_USER@$node "
sed -i 's%MASTER_HOST%${MASTER_HOST}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%MASTER_HA_HOST%${MASTER_HA_HOST}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%SLAVE1%${SLAVE1}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%SLAVE2%${SLAVE2}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%SLAVE3%${SLAVE3}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%HDFS_CLUSTER_NAME%${HDFS_CLUSTER_NAME}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%VIRTUAL_PATH%${VIRTUAL_PATH}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%DFS_NAMESERVICES%${DFS_NAMESERVICES}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%NAMENODE1_NAME%${NAMENODE1_NAME}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%NAMENODE2_NAME%${NAMENODE2_NAME}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%NAMENODE_JOURNAL%${NAMENODE_JOURNAL}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%HADOOPDIR_CONF%${HADOOPDIR_CONF}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%ZOOKEEPER_ADDRESS%${ZOOKEEPER_ADDRESS}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%YARN1_NAME%${YARN1_NAME}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%YARN2_NAME%${YARN2_NAME}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%HADOOPHOME%${HADOOPHOME}%g' $HADOOPHOME/etc/hadoop/$conffile
sed -i 's%JAVAHOME%${JAVAHOME}%g' $HADOOPHOME/etc/hadoop/$conffile
# update yarn.resourcemanager.ha.id for yarn_ha
if [ $conffile == 'yarn-site.xml' ]; then
if [ $node == $MASTER_HA_HOST ]; then
sed -i 's%YARN_ID%${YARN2_NAME}%g' $HADOOPHOME/etc/hadoop/$conffile
else
sed -i 's%YARN_ID%${YARN1_NAME}%g' $HADOOPHOME/etc/hadoop/$conffile
fi
fi
"
done
echo "$node copy hadoop template done."
done

3.2.1.2 config_ssh.sh和ssh_nopassword.expect

这两个文件是配置ssh无密码登录的，ssh_nopassword.expect被config_ssh.sh调用。

config_ssh.sh文件如下：

#!/bin/bash

# Get absolute path of this shell
readonly PROGDIR=$(readlink -m $(dirname $0))
# import init.conf
source $PROGDIR/../conf/init.conf
# Get hostname
HOSTNAME=`hostname`

# Config ssh nopassword login
echo "Config ssh on master"
# If the directory "~/.ssh" is not exist, then execute mkdir and chmod
[ ! -d ~/.ssh ] && ( mkdir ~/.ssh ) && ( chmod 700 ~/.ssh )
# If the file "~/.ssh/id_rsa.pub" is not exist, then execute ssh-keygen and chmod
[ ! -f ~/.ssh/id_rsa.pub ] && ( yes|ssh-keygen -t rsa -P '' -f ~/.ssh/id_rsa ) && ( chmod 600 ~/.ssh/id_rsa.pub )

echo "Config ssh nopassword for cluster"
# For all node, including master and slaves
for node in $ALL; do
# execute bin/ssh_nopassword.expect
$PROGDIR/ssh_nopassword.expect $node $HADOOP_USER $HADOOP_PASSWORD $HADOOPDIR_CONF/.ssh/id_rsa.pub >/dev/null 2>&1
echo "$node done."
done
echo "Config ssh successful."

ssh_nopassword.expect文件如下：

#!/usr/bin/expect -f

set host [lindex $argv 0]
set user [lindex $argv 1]
set password [lindex $argv 2]
set dir [lindex $argv 3]
spawn ssh-copy-id -i $dir $user@$host
expect {
yes/no
{
send "yes\r";exp_continue
}
-nocase "password:"
{
send "$password\r"
}
}
expect eof

3.2.1.3 config_zookeeper.sh

该文件主要是对zookeeper的配置，文件内容如下：

#!/bin/bash

# Get absolute path of this shell
readonly PROGDIR=$(readlink -m $(dirname $0))
# import init.conf
source $PROGDIR/../conf/init.conf

#update conf
sed -i "s%ZOOKEEPERHOME%${ZOOKEEPERHOME}%g" $PROGDIR/../template/zoo.cfg
sed -i "s%ZOOKEEPER_SLAVE1%${ZOOKEEPER_SLAVE1}%g" $PROGDIR/../template/zoo.cfg
sed -i "s%ZOOKEEPER_SLAVE2%${ZOOKEEPER_SLAVE2}%g" $PROGDIR/../template/zoo.cfg
sed -i "s%ZOOKEEPER_SLAVE3%${ZOOKEEPER_SLAVE3}%g" $PROGDIR/../template/zoo.cfg

zookeeperArr=( "$ZOOKEEPER_SLAVE1" "$ZOOKEEPER_SLAVE2" "$ZOOKEEPER_SLAVE3" )
myid=1
for node in ${zookeeperArr[@]}; do
scp $PROGDIR/../template/zoo.cfg $HADOOP_USER@$node:$ZOOKEEPERHOME/conf
echo $myid > $PROGDIR/../template/myid
ssh -q $HADOOP_USER@$node "
[ ! -d $ZOOKEEPERHOME/data ] && ( mkdir $ZOOKEEPERHOME/data )
[ ! -d $ZOOKEEPERHOME/log ] && ( mkdir $ZOOKEEPERHOME/log )
"
scp $PROGDIR/../template/myid $HADOOP_USER@$node:$ZOOKEEPERHOME/data
myid=`expr $myid + 1` #i++
echo "$node copy zookeeper template done."
done

3.2.1.4 start_all.sh

该脚本主要用来启动zookeeper及Hadoop全部组件，文件内容如下：

#!/bin/bash

source /etc/profile
# Get absolute path of this shell
readonly PROGDIR=$(readlink -m $(dirname $0))
# import init.conf
source $PROGDIR/../conf/init.conf

# start zookeeper
zookeeperArr=( "$ZOOKEEPER_SLAVE1" "$ZOOKEEPER_SLAVE2" "$ZOOKEEPER_SLAVE3" )
for znode in ${zookeeperArr[@]}; do
ssh -q $HADOOP_USER@$znode "
source /etc/profile
$ZOOKEEPERHOME/bin/zkServer.sh start
"
echo "$znode zookeeper start done."
done

# start journalnode
journalArr=( $JOURNALLIST )
for jnode in ${journalArr[@]}; do
ssh -q $HADOOP_USER@$jnode "
source /etc/profile
$HADOOPHOME/sbin/hadoop-daemon.sh start journalnode
"
echo "$jnode journalnode start done."
done

# format zookeeper
$HADOOPHOME/bin/hdfs zkfc -formatZK

# format hdfs
$HADOOPHOME/bin/hdfs namenode -format -clusterId $DFS_NAMESERVICES

# start namenode
$HADOOPHOME/sbin/hadoop-daemon.sh start namenode

# sign in master_ha, sync from namenode to namenode_ha
ssh -q $HADOOP_USER@$MASTER_HA_HOST "
$HADOOPHOME/bin/hdfs namenode -bootstrapStandby
"

# start zkfc on master
$HADOOPHOME/sbin/hadoop-daemon.sh start zkfc

# start namenode_ha and datanode
$HADOOPHOME/sbin/start-dfs.sh

# start yarn
$HADOOPHOME/sbin/start-yarn.sh

# start yarn_ha
ssh -q $HADOOP_USER@$MASTER_HA_HOST "
source /etc/profile
$HADOOPHOME/sbin/yarn-daemon.sh start resourcemanager
"
echo "start all done."

4 集群自动化部署流程

4.1 root脚本的执行

选择一台服务器作为Hadoop2.6.0的主节点，使用root用户执行。

确保Hadoop集群所在服务器可以执行scp命令：在各个服务器上执行scp，如果提示命令没有找到，执行安装命令：yum –y install openssh-clients。

执行以下操作：

执行cd ~，进入/root目录下

将root脚本所在目录打成tar包（假设打包后的文件名为root_install.tar.gz），执行rz -y，上传root_install.tar.gz（若无法找到rz命令，执行安装命令：yum -y install lrzsz）

执行tar -zxvf root_install.tar.gz解压

执行cd root_install，进入到root_install目录中

执行. /installRoot.sh，开始安装jdk、zookeeper、Hadoop，等待安装结束

检查/etc/hosts、/etc/profile的配置，执行java -version、hadoop version命令检查jdk和Hadoop的安装情况。若出现java、hadoop命令找不到的情况，重新登录一次服务器再进行检查。

4.2 hadoop脚本的执行

在主节点使用Hadoop启动用户执行（该启动用户是在root中执行的脚本里创建的，下面假设该用户为hadoop2）：

在root用户中直接进入hadoop2用户，执行su - hadoop2

执行以下操作：

执行cd~，进入/home/hadoop2目录下

将hadoop脚本所在目录打成tar包（假设打包后的文件名为hadoop_install.tar.gz），执行rz -y，上传hadoop_install.tar.gz（若无法找到rz命令，执行安装命令：yum -y install lrzsz）

执行tar -zxvf hadoop_install.tar.gz解压

执行cd hadoop_install，进入到hadoop_install目录中

执行./installCluster.sh，开始配置并启动zookeeper、Hadoop，等待脚本执行结束

检查zookeeper、Hadoop启动日志，检查是否安装成功。通过Hadoop本身提供的监控页面来检查Hadoop集群的状态。

最后根据mountTable.xml中fs.viewfs.mounttable.hCluster.link./tmp的配置，执行如下命令创建该name对应的value目录：

hdfs dfs -mkdir hdfs://hadoop-cluster1/tmp

如果不创建，执行hdfs dfs -ls /tmp时会提示找不到目录。

5 总结

Hadoop2.6.0部署脚本仍有缺陷，比如配置文件中参数较多，有部分重复，脚本的编写也有待改进。权当抛砖引玉。如有错误请童鞋们指正，谢谢。