最近要在10几台机器上安装hadoop。对于这种繁复而重复的工作,一步步的打命令行,对于程序员来说是一件不能忍的事情。所以我就琢磨着怎么写一个脚本来自动安装hadoop。
任务: 在10几台机器上中的任意一台执行脚本,即可安装好hadoop。
条件: 每台机器的用户名和密码都是一样的。每台机器都配置好了ssh,能够远程登录。
解决思路:
1. 首先读取配置文件,读取到节点的ip和想要的机器名字,然后修改好本地hadoop的配置文件。
2. 然后读取配置文件,复制所有文件到每个节点的安装的路径。(利用scp命令可以远程复制)
3. 然后读取配置文件,自动ssh到每个节点做一些配置工作,包括配置hadoop和JDK 环境变量、生成ssh-key。
4. ssh到主节点,将每个节点(包括主节点自己)生成的public key 都复制并追加到自己的authorized_keys. 然后把这个authorized_keys分发到每个节点。
这样就配置好了hadoop。
题外话,介绍一下 ssh-keygen。ssh-keygen是一个ssh用于生成密钥的命令。用途是用于免密码登录。它会生成两个密钥,一个是公钥,一个是私钥。比如A 机器生成了pubKeyA,priKeyB。然后A 把pubKeyA给了机器B ,然后机器B 就可以无密码登录机器A 了。
在上面的步骤中,主要的难题是。
1. 对于步骤一,主要难题在于怎么用shell读取配置文件。由于我也之前没写过shell,所以Shell的循环和if 语句和字符串处理也卡了不少时间。
# 这段代码是怎么从hosts的节点配置文件中读取节点的信息 # hosts的格式如下 # 192.168.1.100 master # 192.168.1.101 slave1 # 192.168.1.102 slave2 # ... while read line do echo $line ip=`echo $line | cut -d" " -f1` name=`echo $line | cut -d" " -f2` if [ ! -z $ip ]; then echo $name if [[ $name == maste* ]]; then echo "$name" >> ../hadoop-1.2.1/conf/masters elif [[ $name == slave* ]]; then echo "$name" >> ../hadoop-1.2.1/conf/slaves fi fi done < hosts
2. 对于步骤2,由于刚开始节点直接没有实现无密码ssh,所以scp命令需要输入密码,所以怎么实现自动输入密码实在是一个棘手的问题。我搜索之后,发现一个工具叫expect。
expect工具就像bash一样有自己的语法,然后有自己的命令。它的语法是基于TCL这种脚本语言(我也没听过),看帮助可以直接man expect。我觉得主要需要知道的expect命令是spawn,expect,exp_continue这三个。
#!/usr/bin/expect # expect 定义函数的方式如下 proc usage {} { puts stderr "usage: $::argv0 ip usrname password" exit 1 } if {$argc != 3} { usage } #利用脚本传参数 set hostip [lindex $argv 0] set username [lindex $argv 1] set password [lindex $argv 2] set timeout 100000 # 利用expect的spawn命令来代理执行命令 spawn scp -r ../../hadoop ${username}@${hostip}:~ #获取期望的输出 expect { #如果输出是 要输入密码 #注意下面的大括号 不能换行写,必须用Java风格,而且与前面的“之间要有一个空格,我当时犯了这错误,程序执行的结果很奇怪却不报错。 "*assword:" { send "$password\n" #输入密码后期待spawn代理的命令结束 expect eof } #如果不需要输入密码,那也是一样期待命令结束 expect eof }
对于步骤3、4已经没什么挑战性了,很快就完成了。
下面我把所有代码贴上来
1. setHadoopOnce.sh 这个文件是脚本执行的起点
1 #!/bin/bash 2 #修改密码 3 pw=123456 4 loginName=hadoop 5 master=master 6 slave=slave 7 slaveNum=1 8 set timeout 100000 9 > ../hadoop-1.2.1/conf/masters 10 > ../hadoop-1.2.1/conf/slaves 11 #update local file 12 while read line 13 do 14 echo $line 15 ip=`echo $line | cut -d" " -f1` 16 name=`echo $line | cut -d" " -f2` 17 if [ ! -z $ip ]; then 18 echo $name 19 if [[ $name == maste* ]]; then 20 echo "$name" >> ../hadoop-1.2.1/conf/masters 21 elif [[ $name == slave* ]]; then 22 echo "$name" >> ../hadoop-1.2.1/conf/slaves 23 fi 24 fi 25 done < hosts 26 #upload file to all nodes 27 while read line 28 do 29 ip=`echo $line | cut -d" " -f1` 30 name=`echo $line | cut -d" " -f2` 31 if [ ! -z $ip ]; then 32 expect copyDataToAll.exp $ip $loginName $pw 33 expect setForAll.exp $ip $loginName $pw 34 fi 35 done < hosts 36 37 while read line 38 do 39 ip=`echo $line | cut -d" " -f1` 40 name=`echo $line | cut -d" " -f2` 41 if [ ! -z $ip ]; then 42 if [[ $name == maste* ]]; then 43 expect setForMaster.exp $ip $loginName $pw 44 fi 45 fi 46 done < hosts
2. copyDataToAll.exp 这个在setHadoopOnce.sh中的32行被调用,以复制文件到所有节点。
1 #!/usr/bin/expect 2 proc usage {} { 3 puts stderr "usage: $::argv0 ip usrname password" 4 exit 1 5 } 6 if {$argc != 3} { usage } 7 set hostip [lindex $argv 0] 8 set username [lindex $argv 1] 9 set password [lindex $argv 2] 10 set timeout 100000 11 spawn scp -r ../../hadoop ${username}@${hostip}:~ 12 expect { 13 "*assword:" { 14 send "$password\n" 15 expect eof 16 } 17 expect eof 18 }
3. setForAll.exp 为所有节点进行一些配置工作,在setHadoopOnce.sh中的33行被调用.
#!/usr/bin/expect proc usage {} { puts stderr "usage: $::argv0 ip usrname password" exit 1 } proc connect {pwd} { expect { "*(yes/no)?" { send "yes\n" expect "*assword:" { send "$pwd\n" expect { "*Last login:*" { return 0 } } } } "*assword:" { send "$pwd\n" expect { "*Last login:*" { return 0 } } } "*Last login:*" { return 0 } } return 1 } if {$argc != 3} { usage } set hostip [lindex $argv 0] set username [lindex $argv 1] set password [lindex $argv 2] set timeout 100000 spawn ssh ${username}@${hostip} if {[connect $password]} { exit 1 } #set host send "sudo bash ~/hadoop/setup/addHosts.sh\r" expect "*assword*" send "$password\r" expect "*ddhostsucces*" sleep 1 send "ssh-agent bash ~/hadoop/setup/sshGen.sh\n" expect { "*(yes/no)?" { send "yes\n" exp_continue } "*verwrite (y/n)?" { send "n\n" exp_continue } "*nter file in which to save the key*" { send "\n" exp_continue } "*nter passphrase*" { send "\n" exp_continue } "*nter same passphrase again*" { send "\n" exp_continue } "*our public key has been saved*" { exp_continue } "*etsshGenSucces*" { sleep 1 } } send "bash ~/hadoop/setup/setEnvironment.sh\n" expect "*etEnvironmentSucces*" sleep 1 send "exit\n" expect eof
3.1 addHosts.sh 在setForAll.exp中被调用,用于设置节点的hosts文件
#!/bin/bash hadoopRoot=~/hadoop hadoopPath=$hadoopRoot/hadoop-1.2.1 setupPath=$hadoopRoot/setup localip="`ifconfig |head -n 2|tail -n1 |cut -f2 -d: |cut -f1 -d" " `" hostline="`grep "$localip$" $hadoopRoot/setup/hosts`" sed -i /$hostline/\d $hadoopRoot/setup/hosts #cp /etc/hosts /etc/hosts.hadoop.bak for delip in `cat $hadoopRoot/setup/hosts` do delipline="`grep -n "$delip[[:space:]]" /etc/hosts |cut -f1 -d:`" #echo $delipline if [ -n "$delipline" ]; then sed -i $delipline\d /etc/hosts sleep 1s #else #echo "Your List have no the ip $delip" fi done cat $hadoopRoot/setup/hosts >> /etc/hosts rm -f "$setupPath"/sed* echo "addhostsuccess"
3.2 sshGen.sh 在setForAll.sh中被调用,用于生成sshkey。
#!/bin/bash sshPath=~/.ssh setupPath=~/hadoop/setup rm "$sshPath"/authorized_keys sleep 1 ssh-keygen -t rsa cat "$sshPath"/id_rsa.pub >> "$sshPath"/authorized_keys ssh-add echo "setsshGenSuccess"
3.3 setEnvironment.sh 在setForAll.sh中被调用,用于设置环境变量
#!/bin/bash hadoopRoot=~/hadoop hadoopPath=$hadoopRoot/hadoop-1.2.1 setupPath=$hadoopRoot/setup JAVA_VERSION=`java -version 2>&1 | awk '/java version/ {print $3}'|sed 's/"//g'|awk '{if ($1>=1.6) print "ok"}'` if [ "$JAVA_VERSION"x != "okx" ]; then cat "$setupPath"/jdkenv >> ~/.bashrc sleep 1 source ~/.bashrc sleep 1 fi Hadoop_Version=`hadoop version|awk '/Hadoop/ {print $2}'|awk '{if ($1>=1.0) print "ok"}'` if [ "$Hadoop_Version"x != "okx" ]; then cat "$setupPath"/hadoopenv >> ~/.bashrc sleep 1 source ~/.bashrc sleep 1 fi echo "setEnvironmentSuccess"
4. setForMaster.exp 远程ssh调用setForMaster.sh,以配置无密码登录的功能。
#!/usr/bin/expect proc usage {} { puts stderr "usage: $::argv0 ip usrname password" exit 1 } proc connect {pwd} { expect { "*(yes/no)?" { send "yes\n" expect "*assword:" { send "$pwd\n" expect { "*Last login:*" { return 0 } } } } "*assword:" { send "$pwd\n" expect { "*Last login:*" { return 0 } } } "*Last login:*" { return 0 } } return 1 } if {$argc != 3} { usage } set hostip [lindex $argv 0] set username [lindex $argv 1] set password [lindex $argv 2] set timeout 100000 spawn ssh ${username}@${hostip} if {[connect $password]} { exit 1 } send "ssh-agent bash ~/hadoop/setup/setForMaster.sh\n" expect { "*etForMasterSucces*" { sleep 1 send "exit\n" } "*assword*" { send "$password\n" exp_continue } "*(yes/no)?" { send "yes\n" exp_continue } }
4.1 setForMaster.sh
#!/bin/bash while read line do ip=`echo $line | cut -d" " -f1` name=`echo $line | cut -d" " -f2` if [ ! -z $ip ]; then if [[ $name == slave* ]]; then scp $ip:~/.ssh/authorized_keys ~/tmpkey cat ~/tmpkey >> ~/.ssh/authorized_keys fi fi done < ~/hadoop/setup/hosts sleep 1 rm -f ~/tmpkey while read line do ip=`echo $line | cut -d" " -f1` name=`echo $line | cut -d" " -f2` if [ ! -z $ip ]; then if [[ $name == slave* ]]; then scp ~/.ssh/authorized_keys $ip:~/.ssh/authorized_keys fi fi done < ~/hadoop/setup/hosts echo "setForMasterSuccess"