keepalived+mysql双主原理：

（1）主库1上的keepalived启动之后，会检查mysql服务是否活着，如果活着，keepalived进入master状态，获得VIP；

（2）主库2上的keepalived启动之后，也会检查mysql是否活着，然后检查keepalived组内是否有master状态，如果有，则主库2上的keepalived进入backup状态，处于随时接管VIP状态；

（3）如果主库1上的mysql挂了，keepalived进入fault状态，释放VIP，主库2上的keepalived会变成master状态，获得VIP；

实验环境：

OS：CentOS release 6.6 (Final)

数据库：mysql 5.7.14

A: master ：192.168.91.23

B: slave ：192.168.91.22

VIP：192.168.91.100

操作系统时间一致更改：

date -s "20170227 16:25"

hwclock --systohc

主从参数：

A：

server_id = 330623

gtid_mode=ON

log_slave_updates = 0

enforce_gtid_consistency = ON

auto_increment_offset =1

auto_increment_increment =2

B：

server_id = 330622

gtid_mode=ON

log_slave_updates = 0

enforce_gtid_consistency = ON

auto_increment_offset=2

auto_increment_increment=2

配置AB互为主从：

A：

创建复制账户：

create user rep@'192.168.91.%' identified by '147258';

grant replication slave on *.* to rep@'192.168.91.%';

把A做个全备，还原到B上（这里省略不写）

B：添加A为B的主库：

change master tomaster_host='192.168.91.23', master_port=3306, master_user='rep',master_password='147258', master_auto_position=1;

start slave；

A：添加B为A的主库：

change master tomaster_host='192.168.91.22', master_port=3306, master_user='rep',master_password='147258', master_auto_position=1;

start slave；

创建一个监控账户：（后面checkMySQL.py脚本会用到，用于检测mysql数据库状态，这个用户只要有usage权限即可）

GRANT REPLICATION CLIENT ON *.* TO 'monitor'@'%' IDENTIFIED BY 'm0n1tor';

Ａ和Ｂ都要安装keepalived软件：

yum install keepalived -y

yum install MySQL-python -y

A的keepalived配置文件：

[root@Darren1 keepalived]#cat << EOF > keepalived.conf

vrrp_script vs_mysql_23 { #可以根据实际情况命名

script "/etc/keepalived/checkMySQL.py -h 192.168.91.23 -P 3306"

interval 60 #切换时间

}

vrrp_instance VI_23 { #可以根据实际情况命名

state BACKUP#刚开始时使其处于backup状态

nopreempt #设置为不抢占，m1挂了，m2接管VIP,m1重启不会自动抢回VIP

interface eth0 #VIP用的网卡

virtual_router_id 23#路由id，范围是0-255，不能和路由器高可用的id一样，同一集群中该数值要相同

priority 100#优先级，同一个vrrp_instance的MASTER优先级必须比BACKUP高。

advert_int 5

authentication {

auth_type PASS#认证加密

auth_pass 1111# 认证密码，但密码不要超过 8 位

}

track_script {

vs_mysql_23 #调用这个脚本，返回0就持有VIP，返回1就释放VIP

}

virtual_ipaddress {

192.168.11.100 #VIP地址

}

}

EOF

B配置文件：

[root@Darren2 keepalived]#cat << EOF > keepalived.conf

vrrp_script vs_mysql_22 {

script "/etc/keepalived/checkMySQL.py -h 192.168.91.22 -P 3306"#此处和A不同，其他都相同

interval 60

}

vrrp_instance VI_22 {

state BACKUP

nopreempt

interface eth0

virtual_router_id 23

priority 90

advert_int 5

authentication {

auth_type PASS

auth_pass 1111

}

track_script {

vs_mysql_22

}

virtual_ipaddress {

192.168.91.100

}

}

EOF

checkMySQL.py脚本作用（这里省略不写）：

脚本的作用是判断mysql进程是否存在，如果存在返回0，如果不存在返回1；

A和B启用keepalived

/etc/init.d/keepalived start（开始开的时候，A和B谁先启动，VIP就先在谁上）

chkconfig –level 2345 keepalived on

keepalived启动过程：

此时A开启keepalived服务：

[root@Darren1 ~]# /etc/init.d/keepalived start

[root@Darren1 ~]# tail -f /var/log/messages

May 9 14:41:05 Darren1 Keepalived[28172]: Starting Keepalived v1.2.13 (03/19,2015)

May 9 14:41:05 Darren1 Keepalived[28173]: Starting Healthcheck child process, pid=28175

May 9 14:41:05 Darren1 Keepalived[28173]: Starting VRRP child process, pid=28176

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: Netlink reflector reports IP 192.168.91.23 added

May 9 14:41:05 Darren1 Keepalived_healthcheckers[28175]: Netlink reflector reports IP 192.168.91.23 added

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: Netlink reflector reports IP fe80::20c:29ff:fe56:5380 added

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: Registering Kernel netlink reflector

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: Registering Kernel netlink command channel

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: Registering gratuitous ARP shared channel

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: Opening file '/etc/keepalived/keepalived.conf'.

May 9 14:41:05 Darren1 Keepalived_healthcheckers[28175]: Netlink reflector reports IP fe80::20c:29ff:fe56:5380 added

May 9 14:41:05 Darren1 Keepalived_healthcheckers[28175]: Registering Kernel netlink reflector

May 9 14:41:05 Darren1 Keepalived_healthcheckers[28175]: Registering Kernel netlink command channel

May 9 14:41:05 Darren1 Keepalived_healthcheckers[28175]: Opening file '/etc/keepalived/keepalived.conf'.

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: Configuration is using : 62873 Bytes

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: Using LinkWatch kernel netlink reflector...

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) Entering BACKUP STATE

May 9 14:41:05 Darren1 Keepalived_healthcheckers[28175]: Configuration is using : 5173 Bytes

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: VRRP sockpool: [ifindex(2), proto(112), unicast(0), fd(10,11)]

May 9 14:41:05 Darren1 Keepalived_healthcheckers[28175]: Using LinkWatch kernel netlink reflector...

May 9 14:41:05 Darren1 Keepalived_vrrp[28176]: VRRP_Script(vs_mysql_23) succeeded

May 9 14:41:21 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) Transition to MASTER STATE

May 9 14:41:26 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) Entering MASTER STATE

May 9 14:41:26 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) setting protocol VIPs.

May 9 14:41:26 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) Sending gratuitous ARPs on eth0 for 192.168.91.100

May 9 14:41:26 Darren1 Keepalived_healthcheckers[28175]: Netlink reflector reports IP 192.168.91.100 added

May 9 14:41:31 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) Sending gratuitous ARPs on eth0 for 192.168.91.100

总结启动过程：

（1）启动keepalived三个进程，分别是主进程，健康检查子进程，VRRP协议子进程；

（2）启动结束后，VRRP_Instance开始进入backup状态；

（3）进入backup成功后，VRRP_Instance转变状态为master，然后进入master状态；

（4）获取VIP，并且用ARP广播告诉其他服务器；

keepalived切换过程：

停止A的mysql服务：

[root@Darren1 ~]# /etc/init.d/mysqld stop

Shutting down MySQL............ SUCCESS!

此时A的变化：

[root@Darren1 ~]# tail -f /var/log/messages

May 9 14:43:25 Darren1 Keepalived_vrrp[28176]: VRRP_Script(vs_mysql_23) failed

May 9 14:43:26 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) Entering FAULT STATE

May 9 14:43:26 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) removing protocol VIPs.

May 9 14:43:26 Darren1 Keepalived_vrrp[28176]: VRRP_Instance(VI_23) Now in FAULT state

May 9 14:43:26 Darren1 Keepalived_healthcheckers[28175]: Netlink reflector reports IP 192.168.91.100 removed

总结A的变化：VRRP_Instance进入fault状态，释放VIP；

此时B的变化：

[root@Darren2 ~]# tail -f /var/log/messages

May 9 14:43:26 Darren2 Keepalived_vrrp[35138]: VRRP_Instance(VI_22) Transition to MASTER STATE

May 9 14:43:31 Darren2 Keepalived_vrrp[35138]: VRRP_Instance(VI_22) Entering MASTER STATE

May 9 14:43:31 Darren2 Keepalived_vrrp[35138]: VRRP_Instance(VI_22) setting protocol VIPs.

May 9 14:43:31 Darren2 Keepalived_vrrp[35138]: VRRP_Instance(VI_22) Sending gratuitous ARPs on eth0 for 192.168.91.100

May 9 14:43:31 Darren2 Keepalived_healthcheckers[35137]: Netlink reflector reports IP 192.168.91.100 added

May 9 14:43:36 Darren2 Keepalived_vrrp[35138]: VRRP_Instance(VI_22) Sending gratuitous ARPs on eth0 for 192.168.91.100

总结B的变化：VRRP_Instance进入master状态，获得VIP，ARP广播通知；

使用VIP登陆数据库：

创建登陆用户：

create user'keepalived'@'%' identified by '147258';

grant all on *.* to 'keepalived'@'%';

此时VIP在B上：

[root@Darren2 ~]# ip addr |grep 192

inet 192.168.91.22/24 brd 192.168.91.255 scope global eth0

inet 192.168.91.100/32 scope global eth0

在A上用'keepalived'@'%'账户登陆，是可以登陆成功的，证明此VIP是有效的：

[root@Darren1 ~]# mysql -ukeepalived -p147258 -h292.168.91.100

keepalived@192.168.91.100 [(none)]>select user(),current_user();

+--------------------------+----------------+

| user() | current_user() |

+--------------------------+----------------+

| keepalived@192.168.91.23 | keepalived@% |

+--------------------------+----------------+

总结：

几种VIP切换情况：

（1）m1主机宕机，VIP会切换到m2;

（2）m1上的mysql挂了，VIP会切换到m2;

（3）m1上的keepalived服务挂了，又分为两种情况：

/etc/init.d/keepalived stop：正常切换

kill -9 keepalived_pid：因为keepalived是直接退出，m1和m2都有VIP,但是连接时候只有一个是生效的；

（4）脑裂的情况，m1和m2都各自认为自己是master状态，抢占VIP，VIP一会在m1上，一会在m2上；

在同一个交换机下不存在脑裂情况，这个在比较复杂的网络环境中会发生。

可以在脚本中防范：ping一下网关，如果连网关都ping不通，vrrp_script脚本就放回1，keepalived进入fault状态；

m1挂了，m2接管VIP，m1修复好了之后怎么办？

（1）如果是GTID复制直接把m1change master to m2上，如果是传统复制，需要找到binlog位置；

（2）等待m1同步完成；

（3）启动keepalived；

keepalived+mysql双主缺点和对应方法：

（1）数据库一致性难保障：

可以使用增强半同步，把主库等待从库回应的参数rpl_semi_sync_master_timeout 调大点，

出现master的日志没能实时的传到slave上，需要手工把binlog截取出来补到从库上；如果系统不存在了，可以通过binlog server 补日志；

（2）需要手动把出现故障的主库加入到原来的结构中；