Home  >  Article  >  Database  >  MySQL高可用-双主故障自动切换方案

MySQL高可用-双主故障自动切换方案

WBOY
WBOYOriginal
2016-06-07 14:56:072088browse

前言: (PS:前言是第二次修改本文时加的)对于这篇文章,有博友提出了一些疑问和见解,有了博友的关注,也促使我想把这套东西做的更实用、更安全。后来又经过思考,对脚本中一些条件和行为做了些改变。经过修改,现在终于敢说让小伙伴本使用这套东西了。

前言:(PS: 前言是第二次修改本文时加的)对于这篇文章,有博友提出了一些疑问和见解, 有了博友的关注,也促使我想把这套东西做的更实用、更安全。后来又经过思考, 对脚本中一些条件和行为做了些改变。经过修改,现在终于敢说让小伙伴本使用这套东西了。

主要目的:

    以双主结构配合keepalived解决MySQL主从结构中主库的单点故障;同时通过具体的查询语句提供更细粒度、更为真实的关于主库可用性的判断

基本思路:

    将DB1和DB2做成主动被动模式的双主结构:DB1主动、DB2被动,通过keepalived的VIP对外,将VIP设置成原DB1的IP,保证改造过程对代码透明

    三个前提:

            两台MySQL的配置文件里需要加上“log_slave_updates = 1”;

            并且“备用机”通过“read_only”参数实现除root用户之外的只读特性;

            分别在两个数据库创建test.test表,插入几条数据,供检测脚本使用。

    正常时,VIP在DB1,通过keepalived调用脚本定期检查mysql服务可用性(通过一个低权限用户连接mysql服务器并执行一个简单查询,根据返回结果来判定mysql是否可用)

若无法执行查询:

1. 第一次检测失败后,检查服务状态,:

  1. 若服务异常,则执行切换:关闭DB1的keepalived,使VIP漂移至DB2,通过DB2上keepalived的notify_master机制,触发脚本将DB2的mysql从被动状态(只读)切换到主动状态(可读写),并发送通知邮件。

  2. 若服务正常(则可能是一些临时性因素导致的监测失败),等待30s做第二次检查,这30s是对瞬时/短时因素造成检查失败的容忍时间,本着“能不切则不切”的原则。若第二次检查仍然失败

2.  开始执行系列切换动作

  1. 将DB1的MySQL设置为 read_only模式 (阻止写请继续求进入)

  2. kill掉当前客户端的线程。原来担心kill掉线程会对数据执行造成影响,后来查看了官方文档“mysql shutdown process”,发现mysql正常关闭过程也有一步是如此操作,所以这里可以放心了。然后 sleep 2,给kill命令一些时间(关于kill命令的机制,参考官方解释

  3. 关闭DB1的keepalived,使DB2接管VIP。通过DB2上keepalived的notify_master机制,触发脚本将DB2的

    mysql从被动状态(只读)切换到主动状态(可读写),并发送通知邮件。

3.  管理员修复DB1后,通过脚本“change_to_backup.sh”将主库切换回DB1。脚本思路如下:

     注:涉及到切换主备,就会有中断时间,所以推荐此步骤在业务低谷期执行

  1. 将DB2的read_only属性置为1

  2. kill掉DB2上的client线程,并重启DB2的keepalived使VIP漂移至DB1

  3. 确定DB1跟上了DB2的更新并将DB1上的read_only属性移除


关于“数据一致性”和“切换时间”:

      连续两次失败以后,通过对主MySQL设置read_only属性,同时kill掉用户线程来保证在DB2接管服务之前,DB1上已经没有写操作,避免主从数据不一致。并且切换时间基本上是可确定的:

      30s(两次检测间隔)+2s(等待kill命令时间)+约1s(keepalived 切换VIP),总时间不会超过35s。


以上是大致思路,具体实现看过下面的脚本,就会一目了然了。

DB1上keepalived 配置

! Configuration File for keepalived

vrrp_script chk_mysql {
    script "/etc/keepalived/check_mysql.sh"
    interval 30         #这里我的检查间隔设置的比较长,因为我们数据库前面有redis做缓存,数据库一两分钟级别的终端对整体可用性影响不大。这也是我没有采用成熟的方案而自己搞了这一套方案的“定心丸”
}
vrrp_instance VI_1 {
    state BACKUP        #通过下面的priority来区分MASTER和BACKUP,也只有如此,底下的nopreempt才有效
    interface em2
    virtual_router_id 51
    priority 100
    advert_int 1
    nopreempt           #防止切换到从库后,主keepalived恢复后自动切换回主库
    authentication {
        auth_type PASS
        auth_pass 1111
    }
    track_script {
        chk_mysql
    }
    
    virtual_ipaddress {
        192.168.1.5/24
    }
}


/etc/keepalived/check_mysql.sh脚本内容如下(主要的判断逻辑都在这里)

#!/bin/sh

###判断如果上次检查的脚本还没执行完,则退出此次执行
if [ `ps -ef|grep -w "$0"|grep "/bin/sh*"|grep "?"|grep "?"|grep -v "grep"|wc -l` -gt 2 ];then  #理论上这里应该是1,但是实验的结果却是2
    exit 0
fi

alias mysql_con='mysql -uxxxx -pxxxx'

###定义一个简单判断mysql是否可用的函数
function excute_query {
    mysql_con -e "select * from test.test;" 2>>/etc/keepalived/logs/check_mysql.err
}

###定义无法执行查询,且mysql服务异常时的处理函数
function service_error {
    echo -e "`date "+%F  %H:%M:%S"`    -----mysql service error,now stop keepalived-----" >> /etc/keepalived/logs/check_mysql.err
    /sbin/service keepalived stop &>> /etc/keepalived/logs/check_mysql.err
    echo -e "\n@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n" >> /etc/keepalived/logs/check_mysql.err
}

###定义无法执行查询,但mysql服务正常的处理函数
function query_error {
    echo -e "`date "+%F  %H:%M:%S"`    -----query error, but mysql service ok, retry after 30s-----" >> /etc/keepalived/logs/check_mysql.err
    sleep 30
    excute_query
    if [ $? -ne 0 ];then
        echo -e "`date "+%F  %H:%M:%S"`    -----still can't execute query-----" >> /etc/keepalived/logs/check_mysql.err

        ###对DB1设置read_only属性
        echo -e "`date "+%F  %H:%M:%S"`    -----set read_only = 1 on DB1-----" >> /etc/keepalived/logs/check_mysql.err
        mysql_con -e "set global read_only = 1;" 2>> /etc/keepalived/logs/check_mysql.err

        ###kill掉当前客户端连接
        echo -e "`date "+%F  %H:%M:%S"`    -----kill current client thread-----" >> /etc/keepalived/logs/check_mysql.err
        rm -f /tmp/kill.sql &>/dev/null
        ###这里其实是一个批量kill线程的小技巧
        mysql_con -e 'select concat("kill ",id,";") from  information_schema.PROCESSLIST where command="Query" or command="Execute" into outfile "/tmp/kill.sql";'
        mysql_con -e "source /tmp/kill.sql"
        sleep 2    ###给kill一个执行和缓冲时间
        ###关闭本机keepalived       
        echo -e "`date "+%F  %H:%M:%S"`    -----stop keepalived-----" >> /etc/keepalived/logs/check_mysql.err
        /sbin/service keepalived stop &>> /etc/keepalived/logs/check_mysql.err
        echo -e "\n@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n" >> /etc/keepalived/logs/check_mysql.err
    else
        echo -e "`date "+%F  %H:%M:%S"`    -----query ok after 30s-----" >> /etc/keepalived/logs/check_mysql.err
        echo -e "\n@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n" >> /etc/keepalived/logs/check_mysql.err
    fi
}

###检查开始: 执行查询
excute_query
if [ $? -ne 0 ];then
    /sbin/service mysql status &>/dev/null
    if [ $? -ne 0 ];then
        service_error
    else
        query_error
    fi
fi


DB2上keepalived配置:

! Configuration File for keepalived

vrrp_instance VI_1 {
    state BACKUP
    interface em2
    virtual_router_id 51
    priority 90
    advert_int 1
    authentication {
        auth_type PASS
        auth_pass 1111
    }
    notify_master /etc/keepalived/notify_master_mysql.sh    #此条指令告诉keepalived发现自己转为MASTER后执行的脚本
    virtual_ipaddress {
        192.168.1.5/24
    }
}

/etc/keepalived/notify_master_mysql.sh脚本内容:

#!/bin/bash

###当keepalived监测到本机转为MASTER状态时,执行该脚本
alias mysql_con='mysql -uxxxx -pxxxx'
echo -e "`date "+%F  %H:%M:%S"`   -----keepalived change to MASTER-----" >> /etc/keepalived/logs/state_change.log

###判断是否已经将从master接收到的binlog全部在本地执行(这么做仍然无法完全确定从库就已经追上了主库,因为虽然说从库延时一般情况都是慢在sql_thread上,但是也无法完全保证io_thread完全就没有延时。但至少能保证已经读取到的binlog在本地执行完毕)
pos=`mysql_con -e "show slave status\G;"|grep "Master_Log_Pos"|awk '{printf ("%s",$NF "\t")}'`
read_pos=`echo $pos|awk '{print $1}'`
exec_pos=`echo $pos|awk '{print $2}'`
until [ $read_pos = $exec_pos ]
do
    echo -e "`date "+%F  %H:%M:%S"`    -----Exec_Master_Log_Pos is behind Read_Master_Log_Pos, wait......" >> /etc/keepalived/logs/state_ch
ange.log
    sleep 1
done

###然后解除read_only属性
echo -e "`date "+%F  %H:%M:%S"`    -----set read_only = 0 on DB2-----" >> /etc/keepalived/logs/state_change.log
mysql_con -e "set global read_only = 0;" 2>> /etc/keepalived/logs/state_change.log

echo "DB2 keepalived changed to MASTER,online DB server changed to DB2"|/bin/mailx -s "DB2 keepalived change to MASTER" xxxx@xxxx.com 2>> /etc/keepalived/logs/state_change.log
echo -e "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n" >> /etc/keepalived/logs/state_change.log

DB2上手动切换回DB1的脚本change_to_backup.sh:

#!/bin/sh
###手动执行将主库切换回DB1的操作

alias mysql_con='mysql -uxxxx -pxxxx'

echo -e "`date "+%F  %H:%M:%S"`    -----change to BACKUP manually-----" >> /etc/keepalived/logs/state_change.log
echo -e "`date "+%F  %H:%M:%S"`    -----set read_only = 1 on DB2-----" >> /etc/keepalived/logs/state_change.log
mysql_con -e "set global read_only = 1;" 2>> /etc/keepalived/logs/state_change.log

###kill掉当前客户端连接
echo -e "`date "+%F  %H:%M:%S"`    -----kill current client thread-----" >> /etc/keepalived/logs/state_change.log
rm -f /tmp/kill.sql &>/dev/null
###这里其实是一个批量kill线程的小技巧
mysql_con -e 'select concat("kill ",id,";") from  information_schema.PROCESSLIST where command="Query" or command="Execute" into outfile "/tmp/kill.sql";'
mysql_con -e "source /tmp/kill.sql" 2>> /etc/keepalived/logs/state_change.log
sleep 2    ###给kill一个执行和缓冲时间

###重启DB2的keepalived使VIP漂移到DB1
echo -e "`date "+%F  %H:%M:%S"`    -----make VIP move to DB1-----" >> /etc/keepalived/logs/state_change.log
/sbin/service keepalived restart &>> /etc/keepalived/logs/state_change.log

###确保DB1已经追上了,下面的repl为复制所用的账户,-h后跟DB1的内网IP
pos=`mysql -urepl -pxxxx -h192.168.1.x -e "show slave status\G;"|grep "Master_Log_Pos"|awk '{printf ("%s",$NF "\t")}'`
read_pos=`echo $pos|awk '{print $1}'`
exec_pos=`echo $pos|awk '{print $2}'`
until [ $read_pos = $exec_pos ]
do
    echo -e "`date "+%F  %H:%M:%S"`    -----DB1 Exec_Master_Log_Pos($exec_pos) is behind Read_Master_Log_Pos($read_pos), wait......" >> /etc/keepalived/logs/state_change.log
    sleep 1
done

###然后解除DB1的read_only属性
echo -e "`date "+%F  %H:%M:%S"`    -----set read_only = 0 on DB1-----" >> /etc/keepalived/logs/state_change.log
ssh -pxxxx 192.168.1.x 'mysql -uxxxx -pxxxx -e "set global read_only = 0;"' 2>> /etc/keepalived/logs/state_change.log

echo "DB2 keepalived转为BACKUP状态,线上数据库切换至DB1"|/bin/mailx -s "DB2 keepalived change to BACKUP" xxx@xxxx.com 2>> /etc/keepalived/logs/state_change.log

echo -e "@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@\n" >> /etc/keepalived/logs/state_change.log


日志截图:

DB1 mysql服务故障

wKiom1Yfb07x-EW5AAD6D94EuSs047.jpg

DB1 mysql服务正常,查询失败

wKiom1YYad6QI607AAFDXJ38Zpk793.jpg

DB2 一次切换过程

wKioL1YYaOXjTQEDAAB9x-ob4rE048.jpg

DB2 执行脚本手动切回DB1:

wKiom1YYa97Rq_5_AAFRVWUeFVI819.jpg


总结:此方案适用于中小型企业,解决了主从中master节点的单点问题;同时,在此基础上,可以再增加从库实现读写分离等架构。

Statement:
The content of this article is voluntarily contributed by netizens, and the copyright belongs to the original author. This site does not assume corresponding legal responsibility. If you find any content suspected of plagiarism or infringement, please contact admin@php.cn