简介:masterha_master_switch作为一个切换工具被集成在MHA程序包中,

安装:编译安装MHA manager后会在/usr/local/bin/中生成二进制可执行程序masterha_master_switch。

使用:

$masterha_master_switch --help
Usage:
    # For master failover

    masterha_master_switch --master_state=dead
    --global_conf=/etc/masterha_default.cnf
    --conf=/usr/local/masterha/conf/app1.cnf --dead_master_host=host1

    # For online master switch

    masterha_master_switch --master_state=alive
    --global_conf=/etc/masterha_default.cnf
    --conf=/usr/local/masterha/conf/app1.cnf

    See online reference
    (http://code.google.com/p/mysql-master-ha/wiki/masterha_master_switch)
    for details.

在这里,我习惯将一套主从的配置都放在app1.cnf中,并且更改为业务相关的名称,如mainBusiness.cnf

分析:

目标:获取masterha_master_switch的在线切换逻辑

环境:MHA manager 192.168.1.8

       MHA node1+MySQL5.7+GTID 192.168.1.109+PORT3109     主

       MHA node1+MySQL5.7+GTID 192.168.1.110+PORT3110     从

配置文件内容:

manager_workdir=/data/mha/mainBusiness                 #设置MHA的工作目录
manager_log=/data/mha/mainBusiness/manager.log         #MHA manager的日志输出
remote_workdir=/data/mha/                              #预设MHA node端的工作目录

master_binlog_dir= /data/mysql/3109/log/,/data/mysql/3110/log/  #预设MHA node端的binlog目录                                                                                       
#secondary_check_script= masterha_secondary_check -s 192.168.1.109 -s 192.168.1.110                                                                    
secondary_check_script= masterha_secondary_check -s 192.168.1.109 -s 192.168.1.110 --user=root --master_host=192.168.1.109 --master_port=3109          

ping_interval=1                                       #设置MHA manager的检测间隔(1秒)

[server1]
hostname=MySQL-Cent7-IP001109
ip=192.168.1.109
port=3109
ssh_user=root
ssh_port=22
candidate_master=1                                    #设置该节点是否可以提升为主,1为是,0否
check_repl_delay=0                                    #发生故障后是否检查本实例主从落后程度,0否,1是

[server2]
hostname=MySQL-Cent7-IP001110
ip=192.168.1.110
port=3110
ssh_user=root
ssh_port=22
candidate_master=1                                    #设置该节点是否可以提升为主,1为是,0否
check_repl_delay=0                                    #发生故障后是否检查本实例主从落后程度,0否,1是
    在MHA manager端上执行:

$masterha_master_switch --master_state=alive --conf=/etc/mha/mainBusiness.cnf  --orig_master_is_new_slave
#--master_state             指明在线切换
#--orig_master_is_new_slave 指定原先的主作为从库挂到新的主上

MHA manager端输出如下

#####################输出段1###########################
[info] MHA::MasterRotate version 0.57.
[info] Starting online master switch..        #开始在线切换
[info] 
[info] * Phase 1: Configuration Check Phase.. #阶段1,检查配置
[info] 
[warning] Global configuration file /etc/masterha_default.cnf not found. Skipping..             #我这里没有使用全局参数文件,会有报错跳过,没关系
[info] Reading application default configuration from /etc/mha/mainBusiness.cnf..               #程序从mainBusiness文件中读取配置                                         
[info] Reading server configuration from /etc/mha/mainBusiness.cnf..
[info] GTID failover mode = 1                #启用GTID故障转移模式
[info] Current Alive Master: MySQL-Cent7-IP001109(192.168.1.109:3109)  #列出当前存活的主实例
[info] Alive Slaves:                         #列出当前存活的从实例
[info]   MySQL-Cent7-IP001110(192.168.1.110:3110)  Version=5.7.19-log (oldest major version between slaves) log-bin:enabled
[info]   GTID ON                             #从采用了GTID模式
[info]   Replicating from 192.168.1.109(192.168.1.109:3109)
[info]   Primary candidate for the new Master (candidate_master is set)

It is better to execute FLUSH NO_WRITE_TO_BINLOG TABLES on the master before switching. Is it ok to execute on MySQL-Cent7-IP001109(192.168.1.109:3109)? (YES/no): 
#处于事务一致性考虑,程序询问是否临时关闭非事务型的表

原master,3109端显示如下

#####################输出段1###########################
SET wait_timeout=86400;                                #设置连接超时时间,防止切换时翻车
SELECT @@global.server_id As Value;                    
SELECT VERSION() AS Value;                             #获取用于复制的server—id
SELECT @@global.gtid_mode As Value;                    #获取自身是否使用了GTID复制
SHOW GLOBAL VARIABLES LIKE 'log_bin';                  #检查自身是否启用了binlog
SHOW MASTER STATUS;                                    #获取自身的事务执行情况
SELECT @@global.datadir AS Value;                      #获取自身数据文件的存储位置
SELECT @@global.slave_parallel_workers AS Value;       #检查是否采用了多线程复制
SHOW SLAVE STATUS;                                     #获取自身作为从库时的事务执行情况
SELECT @@global.read_only As Value;                    #获取自身是否开启了只读
SELECT @@global.relay_log_purge As Value;              #检查自身是否开启了relay log自动清除

原slave,3110端显示如下

SELECT @@global.server_id As Value;
SELECT VERSION() AS Value;
SELECT @@global.gtid_mode As Value;
SHOW GLOBAL VARIABLES LIKE 'log_bin';
SHOW MASTER STATUS;
SELECT @@global.datadir AS Value;
SELECT @@global.slave_parallel_workers AS Value;
SHOW SLAVE STATUS;
SELECT @@global.read_only As Value;
SELECT @@global.relay_log_purge As Value;
SELECT @@global.relay_log_info_repository AS Value;     #差异处,获取自身relay信息保存形式(table)
SELECT Relay_log_name FROM mysql.slave_relay_log_info;  #差异处,获取正在使用的relay文件名称
SELECT @@global.datadir AS Value;                       
SHOW SLAVE STATUS;
SELECT Repl_slave_priv AS Value FROM mysql.user WHERE user = 'repl';  #差异处,检查复制用户是否具有复制权限

第一部分小结:

        读取配置文件,确认主从关系与复制方式;

        根据主从关系复制方式,连接主库:设置必要参数,获取的复制详细信息/

                            连接从库:获取复制的详细信息,获取relay信息,获取repl账号并确认权限

接MHA manager端输出确认输入yes后

MHA manager端输出如下:

#####################输出段2###########################
[info] Executing FLUSH NO_WRITE_TO_BINLOG TABLES. This may take long time..
[info]  ok.               #注意,这里虽然显示关闭了非事务表,但是上面的抓包语句里面并没有出现相关语句
[info] Checking MHA is not monitoring or doing failover..  #检查MHA是否工作,切换时要求MHA manager停止运行
[info] Checking replication health on MySQL-Cent7-IP001110.. #检查主从健康程度
[info]  ok.
[info] Searching new master from slaves..                    #开始在从库中选取一个新主
[info]  Candidate masters from the configuration file:       #列出候选从库
[info]   MySQL-Cent7-IP001109(192.168.1.109:3109)  Version=5.7.19-log log-bin:enabled
[info]     GTID ON                                          #检查GTID开启情况
[info]   MySQL-Cent7-IP001110(192.168.1.110:3110)  Version=5.7.19-log (oldest major version between slaves) log-bin:enabled
[info]     GTID ON                                          #检查GTID开启情况
[info]     Replicating from 192.168.1.109(192.168.1.109:3109)
[info]     Primary candidate for the new Master (candidate_master is set)
[info]  Non-candidate masters:
[info]  Searching from candidate_master slaves which have received the latest relay log events..
[info] #在所有从库中选取relay log最新的一个作为新的主库
From:
MySQL-Cent7-IP001109(192.168.1.109:3109) (current master)
 +--MySQL-Cent7-IP001110(192.168.1.110:3110)

To:
MySQL-Cent7-IP001110(192.168.1.110:3110) (new master)
 +--MySQL-Cent7-IP001109(192.168.1.109:3109)

Starting master switch from MySQL-Cent7-IP001109(192.168.1.109:3109) to MySQL-Cent7-IP001110(192.168.1.110:3110)? (yes/NO): 
#程序询问是否可以进行切换了

master,3109端显示如下

#####################输出段2###########################
USE `unknown_database`;
FLUSH NO_WRITE_TO_BINLOG TABLES;                              -- 刷新表缓存到硬盘,同时不记录这条到binlog
SELECT GET_LOCK('MHA_Master_High_Availability_Monitor', '0') AS Value;   -- 加一个模拟锁,防止出现切换程序多开的问题,这样多开的程序以为获取不到同名锁就会失败退出,
#GET_LOCK(str,time) -- str为锁的名称,0表示持续锁,5.7.5之后可以存在多个名称不同的GET_LOCK
SHOW PROCESSLIST;

原slave,3110端显示如下

#####################输出段2###########################
USE `unknown_database`;
SELECT GET_LOCK('MHA_Master_High_Availability_Failover', '0') AS Value;
SHOW SLAVE STATUS;
SHOW SLAVE STATUS;

第二部分小结:

        1.确认可以保存关闭表后,刷新表缓存到磁盘, 避免丢数据,加模拟锁防止切换程序多开造成切换异常

        2.从候选从库中选出拥有最新数据的从库,并将其设为要切换到的主库

接MHA manager端输出确认可以切换后

MHA manager端输出如下:

#####################输出段3###########################
[info] Checking whether MySQL-Cent7-IP001110(192.168.1.110:3110) is ok for the new master..
[info]  ok.   #检查上一步中被选定的新主库的是否真正可以成为主
[info] MySQL-Cent7-IP001109(192.168.1.109:3109): SHOW SLAVE STATUS returned empty result. To check replication filtering rules, temporarily executing CHANGE MASTER to a dummy host.          #没有检查到原来主(现在被作为从)作为从库的残留信息,不管3721先将其挂到一个空主上
[info] MySQL-Cent7-IP001109(192.168.1.109:3109): Resetting slave pointing to the dummy host.
[info] ** Phase 1: Configuration Check Phase completed.
[info] # 配置检查阶段结束
[info] * Phase 2: Rejecting updates Phase..
[info] #
master_ip_online_change_script is not defined. If you do not disable writes on the current master manually, applications keep writing on the current master. Is it ok to proceed? (yes/NO):
#切换程序检测到主机虚拟IP地址切换的地址没有定义,如果只切换主从身份的话,应用还会写到原来的主上,需要设置只读。询问是否真的要切换,这里我们只是做切换实验,先观察下两个实例的输出,然后直接切换即可。

原master,3109端显示如下

USE `unknown_database`;
FLUSH NO_WRITE_TO_BINLOG TABLES;                                -- 刷新表缓存到磁盘, 避免同步丢数据的问题

SELECT GET_LOCK('MHA_Master_High_Availability_Monitor', '0') AS Value;   
SHOW PROCESSLIST;
SHOW SLAVE STATUS;
CHANGE MASTER TO MASTER_HOST='dummy_host';                               #将原主作为从切换到一个莫须有的主机上
SHOW SLAVE STATUS;
RESET SLAVE /*!50516 ALL */;                                             #尝试清除自己之前残存的slave属性的信息,若本机不为主时,需要!
SELECT RELEASE_LOCK('MHA_Master_High_Availability_Monitor') As Value;    

原slave,3110端显示如下

USE `unknown_database`;
SELECT GET_LOCK('MHA_Master_High_Availability_Failover', '0') AS Value;
SHOW SLAVE STATUS;
SHOW SLAVE STATUS;
SHOW PROCESSLIST;

第三部分总结:

    1.检查选定的从库的详细信息,确认真的是否可以作为新的主库

    2.处理从库之间的数据比对,若只有一个主和从,则将原主库先挂到一个莫须有的空主上

                            若有两个以上的从库,则需清理从库原来的slave记录

接MHA manager端输出确认,强制切换后:

MHA manager端输出如下:

#####################输出段4###########################
[info] Locking all tables on the orig master to reject updates from everybody (including root):
[info] Executing FLUSH TABLES WITH READ LOCK..
[info]  ok.   #所有的表都禁止写操作
[info] Orig master binlog:pos is 3109binlog.000070:536.
[info]  Waiting to execute all relay logs on MySQL-Cent7-IP001110(192.168.1.110:3110)..
[info]  master_pos_wait(3109binlog.000070:536) completed on MySQL-Cent7-IP001110(192.168.1.110:3110). Executed 0 events.
[info]   done.#将从库已经获得,但还未来得及执行的事务应用到自身,和其他没有跟上的从库
[info] Getting new master`s binlog name and position..
[info]  3110binlog.000049:536
[info]  All other slaves should start replication from here. Statement should be: CHANGE MASTER TO MASTER_HOST='MySQL-Cent7-IP001110 or 192.168.1.110', MASTER_PORT=3110, MASTER_AUTO_POSITION=1, MASTER_USER='repl', MASTER_PASSWORD='xxx';
[info] Setting read_only=0 on MySQL-Cent7-IP001110(192.168.1.110:3110)..
[info]  ok.   #即将所有其他从库都用change master的语句切换到新的主,即原来的从110上
[info] 
[info] * Switching slaves in parallel..
[info]        #即将并行切换所有从库
[info] Unlocking all tables on the orig master: 
[info] Executing UNLOCK TABLES..
[info]  ok.   #释放原主库的表锁
[info] Starting orig master as a new slave..
[info]  Resetting slave MySQL-Cent7-IP001109(192.168.1.109:3109) and starting replication from the new master MySQL-Cent7-IP001110(192.168.1.110:3110)..
[info]  Executed CHANGE MASTER.
[info]  Slave started.
[info] All new slave servers switched successfully.
[info]        
[info] * Phase 5: New master cleanup phase..
[info]        #即将清理新从库的主从信息,切换到主库
[info]  MySQL-Cent7-IP001110: Resetting slave info succeeded.
[info] Switching master to MySQL-Cent7-IP001110(192.168.1.110:3110) completed successfully.
              #切换完成

新slave,3109端显示如下:

#####################输出段4###########################
USE ``;
SELECT CONNECTION_ID() AS Value;
SET wait_timeout=86400;
SET GLOBAL read_only=1;                           #将自己设为只读
SHOW MASTER STATUS;
UNLOCK TABLES;                                    #释放表锁
CHANGE MASTER TO MASTER_HOST='192.168.1.110', MASTER_PORT=3110, MASTER_USER='repl', MASTER_PASSWORD='123456', MASTER_AUTO_POSITION=1; #把自己指向MHA选定的新主
START SLAVE;
SHOW SLAVE STATUS;                                #差异处,启动slave身份
SELECT RELEASE_LOCK('MHA_Master_High_Availability_Failover') As Value;   #释放锁止同应用的并发锁

新master,3110端显示如下:

#####################输出段4###########################
SHOW SLAVE STATUS;
SELECT MASTER_POS_WAIT('3109binlog.000070','536',0) AS Result;      #检查自己是否执行完差异事务
STOP SLAVE SQL_THREAD;                                              #停止从库SQL线程
SHOW SLAVE STATUS;                                                  
SHOW MASTER STATUS;                                                 
SELECT @@global.read_only As Value;                                 
SELECT @@global.read_only As Value;                                 
SET GLOBAL read_only=0;                                             #取消只读状态
USE ``;
SELECT UNIX_TIMESTAMP();                                            
SELECT @@GLOBAL.SERVER_ID;                                          
SET @master_heartbeat_period= 30000001024;                          
SET @master_binlog_checksum= @@global.binlog_checksum;              
SELECT @master_binlog_checksum;                                     
SELECT @@GLOBAL.GTID_MODE;                                          
SELECT @@GLOBAL.SERVER_UUID;                                        
SET @slave_uuid= '28ea40ab-9bbd-11e7-8cd1-000c29c31069';            #写入从库的UUID
STOP SLAVE;                                                         #停止自身作为从库的身份
SHOW SLAVE STATUS;                                              
RESET SLAVE /*!50516 ALL */;                                        #清除自身作为从库的所有记录
SHOW SLAVE STATUS;
SELECT RELEASE_LOCK('MHA_Master_High_Availability_Failover') As Value; #释放锁止同应用的并发锁

第四部分总结:

    1.检查选定的新主实例relay日志是否存在未执行完的原主事务,并应用到新主

    原主库:锁止,记录主从信息,释放锁,将自己指向新的主,启动从库身份

    新主库:  自己执行完未来得及执行完的事务后停止自己从库身份,取消只读,清除自身作为从的记录

    最后释放切换程序的并发锁

全文总结:

1.读取配置文件,确认主从关系与复制方式,根据主从关系复制方式,连接主库:设置必要参数,获取的复制详细信息;  连接从库:获取复制的详细信息,获取relay信息,获取repl账号并确认权限。

2.确认可以保存关闭非事务表后,关闭非事务表,加模拟锁防止切换程序多开造成切换异常,从候选从库中选出拥有最新数据的从库,并将其设为要切换到的主库。

3.检查选定的从库的详细信息,确认真的是否可以作为新的主库,处理从库之间的数据比对,若只有一个主和从,则将原主库先挂到一个莫须有的空主上;若有两个以上的从库,则需清理从库原来的slave记录。

4.检查选定的新主实例relay日志是否存在未执行完的原主事务,并应用到新主,应用完后清理自己的从库信息,取消只读,转为主库;锁止原主库,记录主从信息,切换到新主库。