MHA原理：
（1）一主三从架构，主库挂了，但主库能被从库ssh上去的情况下，MHA从三个从库中选择同步最接近的作为新主，然后新主和s2，s3都ssh到原主上通过binlog补上还没有同步的数据，io_thread读取到binlog位置，传到save_binary_logs，然后回放，达到s1,s2,s3和原主一致。
（2）主库无法ssh上去的情况下，即主库的系统无法连接，假设有binlog server ，利用binlog补数据，和前面情况一样；
如果没有binlog server，假设s1靠前，s2和s3会通过relay-log和s1同步（所以从库的relay不能自动清除参数relay_log_purge = 0） ，提升s1为主，把s2和s3作为s1的从；
（3）5.6之后，如果有一主三从的架构，使用GTID复制，keepalived+一主两从，并且其中一个从库使用增强半同步复制，这样完全可以取代MHA，这就是为什么MHA代码为什么不更新，out的原因。
实验环境：
OS：CentOS release 6.6 (Final)
数据库：mysql 5.7.16
master ：192.168.91.23
slave1 ：192.168.91.22
slave2：192.168.91.21
VIP：192.168.91.200
| master && masterha_manager | Darren1 | 192.168.91.23 |
| slave1 | Darren2 | 192.168.91.22 |
| slave2 | Darren3 | 192.168.91.21 |
实验架构图：
提前做好一主两从传统的复制（这里不再演示）
提前安装依赖包：（一般yum源可能没有这些依赖包，可以安装epel）
[root@Darren1 tools]# rpm -ivh epel-release-latest-6.noarch.rpm
[root@Darren2 tools]# rpm -ivh epel-release-latest-6.noarch.rpm
[root@Darren3 tools]# rpm -ivh epel-release-latest-6.noarch.rpm
下载软件链接：
http://code.google.com/p/mysql-master-ha/wiki
#master上安装manager和node：
[root@Darren1 tools]# ls
mha4mysql-manager-0.56-0.el6.noarch.rpm
mha4mysql-node-0.56-0.el6.noarch.rpm
[root@Darren1 tools]# yum localinstall mha4mysql*
[root@Darren1 mysql]# rpm -qa |grep mha
mha4mysql-node-0.56-0.el6.noarch
mha4mysql-manager-0.56-0.el6.noarch
#如果yum也无法安装，那就先安装依赖包，再安装rpm包：
yum install perl-DBD-MySQL perl-Config-Tiny perl-Log-Dispatch perl-Parallel-ForkManager -y
rpm -ivh mha4mysql-node-0.56-0.el6.noarch.rpm
rpm -ivh mha4mysql-manager-0.56-0.el6.noarch.rpm
#slave1和slave2在从库上安装node软件：
[root@Darren2 app1]# yum localinstall mha4mysql-node-0.56-0.el6.noarch.rpm
[root@Darren3 app1]# yum localinstall mha4mysql-node-0.56-0.el6.noarch.rpm
物理机器之间信任机制建立
#三台机器同样配置：
[root@Darren1 tools]# vim /etc/hosts
192.168.91.23 Darren1
192.168.91.22 Darren2
192.168.91.21 Darren3
在master上面操作：
[root@Darren1 ]# cd /root/.ssh
[root@Darren1 .ssh]# ssh-keygen ＃一路回车
[root@Darren1 .ssh]# cat /etc/ssh/sshd_config |grep authorized
#AuthorizedKeysFile .ssh/authorized_keys
＃把生成的秘钥文件重定向到authorized_keys文件中：
[root@Darren1 .ssh]# cat id_rsa.pub >>authorized_keys
[root@Darren1 .ssh]# chmod 600 *
#将创建的文件拷贝到slave1和slave2上面：
[root@Darren1 .ssh]# scp * 192.168.91.22:/root/.ssh/
[root@Darren1 .ssh]# scp * 192.168.91.21:/root/.ssh/
#在slave1和slave2上面操作：
[root@Darren2 .ssh]#chmod 600 /root/.ssh/*
[root@Darren3 .ssh]#chmod 600 /root/.ssh/*
#测试互信是否成功：
[root@Darren1 ~]# ssh Darren2 hostname
[root@Darren1 ~]# ssh Darren3 hostname
[root@Darren2 ~]# ssh Darren1 hostname
[root@Darren2 ~]# ssh Darren3 hostname
[root@Darren3 ~]# ssh Darren1 hostname
[root@Darren3 ~]# ssh Darren2 hostname
#创建masterha日志文件目录：
[root@Darren1 masterha]#mkdir -p /var/log/masterha/app1
#在master上创建目录及配置文件：
[root@Darren1 ~]# mkdir -p /etc/masterha
[root@Darren1 ~]# cd /etc/masterha/
#上传五个文件，文件及脚本内容在文章末尾处贴出来：
[root@Darren1 masterha]# ls
app1.conf drop_vip.sh init_vip.sh masterha_default.conf master_ip_failover master_ip_online_change
[root@Darren1 masterha]# chmod +x drop_vip.sh init_vip.sh master_ip_failover master_ip_online_change
------------------------------------------------------------
[root@Darren1 masterha]# cat app1.conf
[server default]
#mha manager工作目录
manager_workdir = /var/log/masterha/app1
manager_log = /var/log/masterha/app1/app1.log
remote_workdir = /var/log/masterha/app1
[server1]
hostname=192.168.91.23
master_binlog_dir = /data/mysql/mysql3306/logs
candidate_master = 1
check_repl_delay = 0 #用防止master故障时，切换时slave有延迟，卡在那里切不过来。
[server2]
hostname=192.168.91.22
master_binlog_dir=/data/mysql/mysql3306/logs
candidate_master=1
check_repl_delay=0
[server3]
hostname=192.168.91.21
master_binlog_dir=/data/mysql/mysql3306/logs
candidate_master=1
check_repl_delay=0
[root@Darren1 masterha]# cat masterha_default.conf
[server default]
#MySQL的用户和密码
user=admin
password = 147258
#系统ssh用户
ssh_user=root
#复制用户
repl_user=repl
repl_password= 147258
#监控
ping_interval=1
#shutdown_script=""
#切换调用的脚本
master_ip_failover_script= /etc/masterha/master_ip_failover
master_ip_online_change_script= /etc/masterha/master_ip_online_change
-------------------------------------------
#分别传到slave1和slave2上相同位置：
[root@Darren1 masterha]# scp -r /etc/masterha 192.168.91.22:/etc/
[root@Darren1 masterha]# scp -r /etc/masterha 192.168.91.21:/etc/
#创建一个admin用户，用于启动masterha_manager ：
root@localhost [(none)]>create user admin@'192.168.91.%' identified by '147258';
root@localhost [(none)]>grant all on *.* to 'admin'@'192.168.91.%';
#启动VIP：
[root@Darren1 masterha]# sh init_vip.sh
#节点之间的主从关系检测：
[root@Darren1 masterha]# masterha_check_repl --global_conf=/etc/masterha/masterha_default.conf --conf=/etc/masterha/app1.conf
最后输出：MySQL Replication Health is OK.表示检测通过
#节点直接的互信检测：
[root@Darren1 masterha]# masterha_check_ssh --global_conf=/etc/masterha/masterha_default.conf --conf=/etc/masterha/app1.conf
Thu May 11 15:24:55 2017 - [info] All SSH connection tests passed successfully.
#启动masterha：
[root@Darren1 app1]# nohup masterha_manager --global_conf=/etc/masterha/masterha_default.conf --conf=/etc/masterha/app1.conf > /tmp/masterha_manager.log 2>&1 &
#查看masterha状态：
[root@Darren1 app1]# masterha_check_status --global_conf=/etc/masterha/masterha_default.conf --conf=/etc/masterha/app1.conf
app1 (pid:86957) is running(0:PING_OK), master:192.168.91.23
[root@Darren1 app1]# ps -ef |grep master
root 1371 1 0 May04 ? 00:00:02 /usr/libexec/postfix/master
root 78618 67225 0 08:02 pts/0 00:00:00 perl /usr/bin/masterha_manager --global_conf=/etc/masterha/masterha_default.conf --conf=/etc/masterha/app1.conf
root 79204 67246 0 08:07 pts/2 00:00:00 grep master
#通过日志文件查看启动过程：
-----------------------------------------------
[root@Darren1 app1]# tail -f app1.log
Thu May 11 15:47:15 2017 - [info] MHA::MasterMonitor version 0.56.
Thu May 11 15:47:16 2017 - [info] GTID failover mode = 0
Thu May 11 15:47:16 2017 - [info] Dead Servers:
Thu May 11 15:47:16 2017 - [info] Alive Servers: --罗列mha中的所有节点
Thu May 11 15:47:16 2017 - [info] 192.168.91.23(192.168.91.23:3306)
Thu May 11 15:47:16 2017 - [info] 192.168.91.22(192.168.91.22:3306)
Thu May 11 15:47:16 2017 - [info] Alive Slaves: --活跃的从库
Thu May 11 15:47:16 2017 - [info] 192.168.91.22(192.168.91.22:3306) Version=5.7.16-log (oldest major version between slaves) log-bin:enabled
Thu May 11 15:47:16 2017 - [info] Replicating from 192.168.91.23(192.168.91.23:3306)
Thu May 11 15:47:16 2017 - [info] Primary candidate for the new Master (candidate_master is set)
Thu May 11 15:47:16 2017 - [info] Current Alive Master: 192.168.91.23(192.168.91.23:3306) --现在的主库
Thu May 11 15:47:16 2017 - [info] Checking slave configurations..
Thu May 11 15:47:16 2017 - [info] read_only=1 is not set on slave 192.168.91.22(192.168.91.22:3306).
Thu May 11 15:47:16 2017 - [info] Checking replication filtering settings..
Thu May 11 15:47:16 2017 - [info] binlog_do_db= , binlog_ignore_db=
Thu May 11 15:47:16 2017 - [info] Replication filtering check ok.
Thu May 11 15:47:16 2017 - [info] GTID (with auto-pos) is not supported
Thu May 11 15:47:16 2017 - [info] Starting SSH connection tests..
Thu May 11 15:47:17 2017 - [info] All SSH connection tests passed successfully.
Thu May 11 15:47:17 2017 - [info] Checking MHA Node version..
Thu May 11 15:47:17 2017 - [info] Version check ok.
Thu May 11 15:47:17 2017 - [info] Checking SSH publickey authentication settings on the current master..
Thu May 11 15:47:17 2017 - [info] HealthCheck: SSH to 192.168.91.23 is reachable.
Thu May 11 15:47:17 2017 - [info] Master MHA Node version is 0.56.
Thu May 11 15:47:17 2017 - [info] Checking recovery script configurations on 192.168.91.23(192.168.91.23:3306)..
save_binary_logs
#检查binlog部分,如果主库挂了，从库需要用到binlog补数据，这个命令是主库故障转移时候用到，这里是command=test是测试，没有实际操作。
Thu May 11 15:47:17 2017 - [info] Executing command: save_binary_logs --command=test --start_pos=4 --binlog_dir=/data/mysql/mysql3306/logs --output_file=/var/log/masterha/app1/save_binary_logs_test --manager_version=0.56 --start_file=mysql-bin.000029
Thu May 11 15:47:17 2017 - [info] Connecting to root@192.168.91.23(192.168.91.23:22)..
Creating /var/log/masterha/app1 if not exists.. ok.
Checking output directory is accessible or not..
ok.
Binlog found at /data/mysql/mysql3306/logs, up to mysql-bin.000029
Thu May 11 15:47:17 2017 - [info] Binlog setting check done.
#查看从库的relay-log，使用到命令apply_diff_relay_logs，如果主库挂了，并且不能ssh上去也没有binlog server备份,那么选举一个从库作为新主，其他从库需要用到这里的relay-log补数据；
Thu May 11 15:47:17 2017 - [info] Checking SSH publickey authentication and checking recovery script configurations on all alive slave servers..
Thu May 11 15:47:17 2017 - [info] Executing command : apply_diff_relay_logs --command=test --slave_user='admin' --slave_host=192.168.91.22 --slave_ip=192.168.91.22 --slave_port=3306 --workdir=/var/log/masterha/app1 --target_version=5.7.16-log --manager_version=0.56 --relay_dir=/data/mysql/mysql3306/data --current_relay_log=relay-bin.000001 --slave_pass=xxx
Thu May 11 15:47:17 2017 - [info] Connecting to root@192.168.91.22(192.168.91.22:22)..
Checking slave recovery environment settings..
Relay log found at /data/mysql/mysql3306/data, up to relay-bin.000002
Temporary relay log file is /data/mysql/mysql3306/data/relay-bin.000002
Testing mysql connection and privileges..mysql: [Warning] Using a password on the command line interface can be insecure.
done.
Testing mysqlbinlog output.. done.
Cleaning up test file(s).. done.
Thu May 11 15:47:18 2017 - [info] Slaves settings check done.
Thu May 11 15:47:18 2017 - [info]
#现在的主库和从库
192.168.91.23(192.168.91.23:3306) (current master)
+--192.168.91.22(192.168.91.22:3306)
Thu May 11 15:47:18 2017 - [info] Checking master_ip_failover_script status:
Thu May 11 15:47:18 2017 - [info] /etc/masterha/master_ip_failover --command=status --ssh_user=root --orig_master_host=192.168.91.23 --orig_master_ip=192.168.91.23 --orig_master_port=3306
Thu May 11 15:47:18 2017 - [info] OK.
Thu May 11 15:47:18 2017 - [warning] shutdown_script is not defined.
#每隔1秒ping一次
Thu May 11 15:47:18 2017 - [info] Set master ping interval 1 seconds.
Thu May 11 15:47:18 2017 - [warning] secondary_check_script is not defined. It is highly recommended setting it to check master reachability from two or more routes.
Thu May 11 15:47:18 2017 - [info] Starting ping health check on 192.168.91.23(192.168.91.23:3306)..
Thu May 11 15:47:18 2017 - [info] Ping(SELECT) succeeded, waiting until MySQL doesn't respond..
------------------------------------
#停止mha：
[root@Darren1 app1]# masterha_stop --global_conf=/etc/masterha/masterha_default.conf --conf=/etc/masterha/app1.conf
Stopped app1 successfully.
当主库上的mysql挂了，这个时候会从slave1和slave2中选择一个作为新主，如果选择slave1作为新主，这个时候slave2会把slave1作为新主，以前挂掉的主等修复好之后可以作为slave1的从。
模拟切换
[root@Darren1 app1]# /etc/init.d/mysqld stop
Shutting down MySQL............ SUCCESS!
#MHA日志目录下会生成两个文件，其中app1.failover.complete 这个文件下次启动mha之前必须删除或者移走，否则无法启动：
[root@Darren1 app1]# ll
total 24
-rw-r--r-- 1 root root 0 May 10 08:12 app1.failover.complete
-rw-r--r-- 1 root root 17416 May 10 08:12 app1.log
-rw-r--r-- 1 root root 234 May 10 08:12 saved_master_binlog_from_192.168.91.23_3306_20170510081201.binlog
#可以看到ip已经切换到从库上面：
[root@Darren2 masterha]# ip addr|grep 192
inet 192.168.91.22/24 brd 192.168.91.255 scope global eth0
inet 192.168.91.200/32 scope global eth0
inet 192.168.91.100/32 scope global eth0
#slave1上的从库信息也被清除了：
root@localhost [(none)]>show slave status\G
Empty set (0.00 sec)
#slave2也把slave1作为了新主：
root@localhost [(none)]>show slave status\G
*************************** 1. row ***************************
Slave_IO_State: Waiting for master to send event
Master_Host: 192.168.91.22
Master_User: repl
挂掉的主库怎么办？
可以作为新主的从库：
[root@Darren1 app1]# /etc/init.d/mysqld start
#在MHA日志中可以找到master_log_file和master_log_pos：
[root@Darren1 app1]# vim app1.log|grep CHANGE
root@localhost [(none)]>change master to
master_host='192.168.91.22',
master_port=3306,
master_user='repl',
master_password='147258',
Master_Log_File='mysql-bin.000014',
Master_Log_Pos=234;
#重新启动mha：
[root@Darren1 app1]# nohup masterha_manager --global_conf=/etc/masterha/masterha_default.conf --conf=/etc/masterha/app1.conf > /tmp/masterha_manager.log 2>&1 &
[root@Darren1 app1]# cat app1.master_status.health
94112 0:PING_OK master:192.168.91.22
[root@Darren1 app1]# tail app1.log
Thu May 11 17:06:42 2017 - [warning] secondary_check_script is not defined. It is highly recommended setting it to check master reachability from two or more routes.
Thu May 11 17:06:42 2017 - [info] Starting ping health check on 192.168.91.22(192.168.91.22:3306).. --应为这个时候22已经是主库了，所以这里一直ping22是否存在
Thu May 11 17:06:42 2017 - [info] Ping(SELECT) succeeded, waiting until MySQL doesn't respond..
可能遇到的报错
（1）[root@Darren1 .ssh]# masterha_check_ssh --global_conf=/etc/masterha/masterha_default.conf --conf=/etc/masterha/app1.conf
......
Thu May 11 19:42:40 2017 - [info] Connecting to root@192.168.91.21(192.168.91.21:22)..
Can't exec "mysqlbinlog": No such file or directory at /usr/share/perl5/vendor_perl/MHA/BinlogManager.pm line 106.
或者执行同样的命令遇到这样的报错：
Testing mysql connection and privileges..sh: mysql: command not found
mysql command failed with rc 127:0!
解决方法：
需要做一个软连接
ln -s /usr/local/mysql/bin/mysqlbinlog /usr/bin/mysqlbinlog
ln -s /usr/local/mysql/bin/mysql /usr/bin/mysql
（2）在启动masterha_manager的时候可能报错：
[root@Darren1 app1]# tail app1.log
Thu May 11 21:40:18 2017 - [error][/usr/share/perl5/vendor_perl/MHA/MasterFailover.pm, ln309] Last failover was done at 2017/05/11 19:04:14. Current time is too early to do failover again. If you want to do failover, manually remove /var/log/masterha/app1/app1.failover.complete and run this script again.
Thu May 11 21:40:18 2017 - [error][/usr/share/perl5/vendor_perl/MHA/ManagerUtil.pm, ln177] Got ERROR: at /usr/bin/masterha_manager line 65
原因是没有删除app1.failover.complete，rm就好。
本文转自 Darren_Chen 51CTO博客，原文链接：http://blog.51cto.com/darrenmemos/1924835，如需转载请自行联系原作者
