Linux Cluster环境下批量分发执行补丁

转自：http://blog.csdn.net/napolunyishi/article/details/18219867

这两天做了一个需求，因为上一个版本的/tmp空间默认只分配了5G，而升级程序上传解压路径也是/tmp，且解压过程要占用5G左右的空间，这样 就导致/tmp空间占满，解压失败，升级无法进行。后来想了一个办法，就是从/var建了一个软链接到/tmp，这样实际上升级文件上传解压到/var路 径下了，解决了/tmp路径空间不足的问题。但是因为客户那里是一个cluster环境，由6个node组成，最好能够在一个机器上执行一次补丁，所有 node都生效。

经过几次尝试，写了两个shell脚本，一个run.sh负责调度和分发补丁，另外一个doFix.sh，真正给系统打补丁的脚本。这里只介绍调度分发的部分

基本思路就是从主node上将doFix.sh脚本分发到其他各个node上，然后再从主node上ssh到各node执行补丁脚本。

调度脚本run.sh：

    #!/bin/bash  

    #check version
    cat /etc/issue | grep "V3.5"
    if [ $? -ne  ]; then
        echo "This is not V3.5, exit!"
        exit
    fi  

    ISFABRIC=
    ADMPW=""  

    echo  

    read -rsn1 -p "Is this a fabric with same admin password? [y/n]"
    echo
    if [ "$REPLY" = "y" -o "$REPLY" = "Y" ]; then
        ISFABRIC=
        read -esp "Please input the password: " ADMPW
    else
        echo "Please run the script doFix.sh manually"
        exit
    fi
    echo
    echo  

    echo "Patch current node..."
    sh ./doFix.sh  

    NODEIP=$(ifconfig eth0 | grep 'inet addr' | cut -d':' -f2 | awk '{print $1}')  

    echo "Begin to patch other nodes..."
    echo  

    egrep "MAGICNODE-" /etc/hosts | awk '{print $1}' | sort | uniq | while read LINE
    do
        if [ "$NODEIP" = "$LINE" ]; then
           continue
        fi  

        echo "Patching node $LINE"
        # scp doFix.sh to remote node
        ./scpExpect $ADMPW./doFix.sh admin@$LINE:/tmp/
        if [ $? -ne  ]; then
           echo "SCP script to node $LINE failed! skip"
           continue
        fi  

        # execute patch script
        ./sshExpectSudo admin $ADMPW $LINE "sudo bash /tmp/doFix.sh"
        if [ $? -ne  ]; then
            echo "Remote patch node $LINE failed! skip"
            continue
        fi
        echo
        echo
    done  

    echo "Patch finished."  

从/etc/hosts文件中获取cluster环境的所有node IP，然后将doFix.sh脚本通过scp送过去，再执行ssh打补丁。

这里两个关键点是scpExpect和sshExpect，何为Expect文件？请参考http://expect.sourceforge.net/，网上有很多Expect的介绍文章，归结一句话就是一种实现自动化交互的脚本语言，从TCL扩展出来。

scpExpect

    #!/usr/bin/expect -f
    # usage: scpExpect user password ip file target  

    system "rm -rf  ~/.ssh/known_hosts"
    set password [lindex $argv ]
    set from  [lindex $argv ]
    set to  [lindex $argv ]
    set timeout   

    # now connect to remote UNIX box (ipaddr) with given script to execute
    spawn scp -r -o StrictHostKeyChecking=no $from $to
    match_max
    expect {
                  "*assword:*" {
                            send -- "$password\r"
                            set timeout -
                            expect {
                                     "*assword:*" {
                                            exit
                                     }
                            }
                } "yes/no)?" {
                           send "yes\r"
                           set timeout -
                } -re . {
                           exp_continue
                } timeout {
                           exit
                } eof {
                           exit
                }
     }  

sshExpect

#!/usr/bin/expect -f
# usage:  ssh user password ipaddre command target
# set Variables
set user [lrange $argv  ]
set password [lrange $argv  ]
set ipaddr [lrange $argv  ]
set command  [lrange $argv  ]
set target  [lrange $argv  ]
set timeout -  

# now connect to remote UNIX box (ipaddr) with given script to execute
spawn ssh -o StrictHostKeyChecking=no $user@$ipaddr $command $target
match_max
# Look for passwod prompt
expect "*?assword:*"
# Send password aka $password
send -- "$password\r"
# send blank line (\r) to make sure we get back to gui
send -- "\r"
expect eof