在日常工作中，会碰到如下的场景，如mysql数据库升级，主服务器硬件升级等，这个时候就需要将写操作切换到另外一台服务器上，那么如何进行在线切换呢？同时，要求切换过程短，对业务的影响比较小。
MHA就提供了这样一种优雅的方式，只会堵塞业务0.5~2s的时间，在这段时间内，业务无法读取和写入。
集群信息
角色 IP地址 ServerID 类型
Master 192.168.244.10 1 写入
Candicate master 192.168.244.20 2 读
Slave 192.168.244.30 3 读
Monitor host 192.168.244.40 监控集群组
MHA具体的搭建步骤和原理，可参考另外一篇博客：
在线切换的步骤
1. 关闭MHA监控
# masterha_stop --conf=/etc/masterha/app1.cnf
2. 在线切换
# /usr/local/bin/masterha_master_switch --conf=/etc/masterha/app1.cnf --master_state=alive --new_master_host=192.168.244.20 --new_master_port=3306 --orig_master_is_new_slave --running_updates_limit=10000
其中，
--orig_master_is_new_slave是将原master切换为新主的slave，默认情况下，是不添加的。
--running_updates_limit默认为1s，即如果主从延迟时间（Seconds_Behind_Master），或master show processlist中dml操作大于1s，则不会执行切换。
在线切换的输出
Tue Apr :: - [info] MHA::MasterRotate version 0.56.
Tue Apr :: - [info] Starting online master switch..
Tue Apr :: - [info]
Tue Apr :: - [info] * Phase : Configuration Check Phase..
Tue Apr :: - [info]
Tue Apr :: - [warning] Global configuration file /etc/masterha_default.cnf not found. Skipping.
Tue Apr :: - [info] Reading application default configuration from /etc/masterha/app1.cnf..
Tue Apr :: - [info] Reading server configuration from /etc/masterha/app1.cnf..
Tue Apr :: - [info] GTID failover mode =
Tue Apr :: - [info] Current Alive Master: 192.168.244.10(192.168.244.10:)
Tue Apr :: - [info] Alive Slaves:
Tue Apr :: - [info] 192.168.244.20(192.168.244.20:) Version=5.6.-log (oldest major version between slaves) log
-bin:enabledTue Apr :: - [info] Replicating from 192.168.244.10(192.168.244.10:)
Tue Apr :: - [info] Primary candidate for the new Master (candidate_master is set)
Tue Apr :: - [info] 192.168.244.30(192.168.244.30:) Version=5.6.-log (oldest major version between slaves) log
-bin:enabledTue Apr :: - [info] Replicating from 192.168.244.10(192.168.244.10:) It is better to execute FLUSH NO_WRITE_TO_BINLOG TABLES on the master before switching. Is it ok to execute on 192.168.244.10(192.168
.244.10:)? (YES/no): yes
Tue Apr :: - [info] Executing FLUSH NO_WRITE_TO_BINLOG TABLES. This may take long time..
Tue Apr :: - [info] ok.
Tue Apr :: - [info] Checking MHA is not monitoring or doing failover..
Tue Apr :: - [info] Checking replication health on 192.168.244.20..
Tue Apr :: - [info] ok.
Tue Apr :: - [info] Checking replication health on 192.168.244.30..
Tue Apr :: - [info] ok.
Tue Apr :: - [info] 192.168.244.20 can be new master.
Tue Apr :: - [info]
From:
192.168.244.10(192.168.244.10:) (current master)
+--192.168.244.20(192.168.244.20:)
+--192.168.244.30(192.168.244.30:) To:
192.168.244.20(192.168.244.20:) (new master)
+--192.168.244.30(192.168.244.30:)
+--192.168.244.10(192.168.244.10:) Starting master switch from 192.168.244.10(192.168.244.10:) to 192.168.244.20(192.168.244.20:)? (yes/NO): yes
Tue Apr :: - [info] Checking whether 192.168.244.20(192.168.244.20:) is ok for the new master..
Tue Apr :: - [info] ok.
Tue Apr :: - [info] 192.168.244.10(192.168.244.10:): SHOW SLAVE STATUS returned empty result. To check replication
filtering rules, temporarily executing CHANGE MASTER to a dummy host.Tue Apr :: - [info] 192.168.244.10(192.168.244.10:): Resetting slave pointing to the dummy host.
Tue Apr :: - [info] ** Phase : Configuration Check Phase completed.
Tue Apr :: - [info]
Tue Apr :: - [info] * Phase : Rejecting updates Phase..
Tue Apr :: - [info]
Tue Apr :: - [info] Executing master ip online change script to disable write on the current master:
Tue Apr :: - [info] /usr/local/bin/master_ip_online_change --command=stop --orig_master_host=192.168.244.10 --orig_ma
ster_ip=192.168.244.10 --orig_master_port= --orig_master_user='monitor' --orig_master_password='monitor123' --new_master_host=192.168.244.20 --new_master_ip=192.168.244.20 --new_master_port= --new_master_user='monitor' --new_master_password='monitor123' --orig_master_ssh_user=root --new_master_ssh_user=root --orig_master_is_new_slaveTue Apr :: Set read_only on the new master.. ok.
Tue Apr :: Set read_only= on the orig master.. ok.
Tue Apr :: Killing all application threads..
Tue Apr :: done.
Disabling the VIP an old master: 192.168.244.10
SIOCSIFFLAGS: Cannot assign requested address
Tue Apr :: - [info] ok.
Tue Apr :: - [info] Locking all tables on the orig master to reject updates from everybody (including root):
Tue Apr :: - [info] Executing FLUSH TABLES WITH READ LOCK..
Tue Apr :: - [info] ok.
Tue Apr :: - [info] Orig master binlog:pos is mysql-bin.:.
Tue Apr :: - [info] Waiting to execute all relay logs on 192.168.244.20(192.168.244.20:)..
Tue Apr :: - [info] master_pos_wait(mysql-bin.:) completed on 192.168.244.20(192.168.244.20:). Executed
events.Tue Apr :: - [info] done.
Tue Apr :: - [info] Getting new master's binlog name and position..
Tue Apr :: - [info] mysql-bin.:
Tue Apr :: - [info] All other slaves should start replication from here. Statement should be: CHANGE MASTER TO MASTER_
HOST='192.168.244.20', MASTER_PORT=, MASTER_LOG_FILE='mysql-bin.000009', MASTER_LOG_POS=, MASTER_USER='repl', MASTER_PASSWORD='xxx';Tue Apr :: - [info] Executing master ip online change script to allow write on the new master:
Tue Apr :: - [info] /usr/local/bin/master_ip_online_change --command=start --orig_master_host=192.168.244.10 --orig_m
aster_ip=192.168.244.10 --orig_master_port= --orig_master_user='monitor' --orig_master_password='monitor123' --new_master_host=192.168.244.20 --new_master_ip=192.168.244.20 --new_master_port= --new_master_user='monitor' --new_master_password='monitor123' --orig_master_ssh_user=root --new_master_ssh_user=root --orig_master_is_new_slaveTue Apr :: Set read_only= on the new master.
Enabling the VIP 192.168.244.188 on the new master: 192.168.244.20
Tue Apr :: - [info] ok.
Tue Apr :: - [info]
Tue Apr :: - [info] * Switching slaves in parallel..
Tue Apr :: - [info]
Tue Apr :: - [info] -- Slave switch on host 192.168.244.30(192.168.244.30:) started, pid:
Tue Apr :: - [info]
Tue Apr :: - [info] Log messages from 192.168.244.30 ...
Tue Apr :: - [info]
Tue Apr :: - [info] Waiting to execute all relay logs on 192.168.244.30(192.168.244.30:)..
Tue Apr :: - [info] master_pos_wait(mysql-bin.:) completed on 192.168.244.30(192.168.244.30:). Executed
events.Tue Apr :: - [info] done.
Tue Apr :: - [info] Resetting slave 192.168.244.30(192.168.244.30:) and starting replication from the new master
92.168.244.20(192.168.244.20:)..Tue Apr :: - [info] Executed CHANGE MASTER.
Tue Apr :: - [info] Slave started.
Tue Apr :: - [info] End of log messages from 192.168.244.30 ...
Tue Apr :: - [info]
Tue Apr :: - [info] -- Slave switch on host 192.168.244.30(192.168.244.30:) succeeded.
Tue Apr :: - [info] Unlocking all tables on the orig master:
Tue Apr :: - [info] Executing UNLOCK TABLES..
Tue Apr :: - [info] ok.
Tue Apr :: - [info] Starting orig master as a new slave..
Tue Apr :: - [info] Resetting slave 192.168.244.10(192.168.244.10:) and starting replication from the new master
92.168.244.20(192.168.244.20:)..Tue Apr :: - [info] Executed CHANGE MASTER.
Tue Apr :: - [info] Slave started.
Tue Apr :: - [info] All new slave servers switched successfully.
Tue Apr :: - [info]
Tue Apr :: - [info] * Phase : New master cleanup phase..
Tue Apr :: - [info]
Tue Apr :: - [info] 192.168.244.20: Resetting slave info succeeded.
Tue Apr :: - [info] Switching master to 192.168.244.20(192.168.244.20:) completed successfully.
MHA在线切换的原理
1. 检查当前的配置信息及主从服务器的信息
包括读取MHA的配置文件/etc/masterha/app1.cnf及检查当前slave的健康状态
2. 阻止对当前master的更新
主要通过如下步骤：
1> 等待1.5s（$time_until_kill_threads*100ms），等待当前连接断开。
2> 执行 read_only=1，阻止新的DML操作
3> 等待0.5s，等待当前DML操作完成。
4> kill掉所有连接。
5> FLUSH NO_WRITE_TO_BINLOG TABLES
6> FLUSH TABLES WITH READ LOCK
3. 等待新master执行完所有的relay log
Waiting to execute all relay logs on 192.168.244.20(192.168.244.20:)..
4. 将新master的read_only设置为off，并添加VIP
5. slave切换到新master上。
1> 等待slave（192.168.244.30）应用完原主从复制产生的relay log，然后执行change master操作切换到新master上。
2> 释放原master上加的锁。
3> 因masterha_master_switch命令行中带有--orig_master_is_new_slave参数，故原master也切换为新master的从。
6. 清理新master的相关信息。
主要是执行了reset slave all操作，清除之前的复制信息。
MHA在线切换需满足的条件
MHA在执行在线切换之前，会判断当前的主从复制信息，只有满足了以下条件，才能执行切换动作：
1. 所有SLAVE的IO线程和SQL线程都在运行。
2. 所有slave的Seconds_Behind_Master小于或等于running_updates_limit的值，该参数如果没有显示指定的话，则默认为1s
3. 在master上，通过show processlist输出，没有一个DML操作的时间大于running_updates_limit的值。
在线切换时，打开general log，各个服务器的操作信息
注：在执行masterha_master_switch命令时，会有两次确认操作
1. It is better to execute FLUSH NO_WRITE_TO_BINLOG TABLES on the master before switching. Is it ok to execute on 192.168.244.10(192.168
.244.10:3306)? (YES/no):
2. Starting master switch from 192.168.244.10(192.168.244.10:3306) to 192.168.244.20(192.168.244.20:3306)? (yes/NO):
以下输出中间都有两次空白，其中第一次空白之前的输出对应第一次确认之前，第二次之前的输出对应第二次确认之前。
原master 192.168.244.10
:: Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
:: Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
Query SET wait_timeout=
Query SELECT @@global.server_id As Value
Query SELECT VERSION() AS Value
Query SELECT @@global.gtid_mode As Value
Query SHOW GLOBAL VARIABLES LIKE 'log_bin'
Query SHOW MASTER STATUS
Query SELECT @@global.datadir AS Value
Query SELECT @@global.slave_parallel_workers AS Value
Query SHOW SLAVE STATUS
Query SELECT @@global.read_only As Value
Query SELECT @@global.relay_log_purge As Value :: Query FLUSH NO_WRITE_TO_BINLOG TABLES
Query SELECT GET_LOCK('MHA_Master_High_Availability_Monitor', '') AS Value
Query SHOW PROCESSLIST :: Query SHOW SLAVE STATUS
Query CHANGE MASTER TO MASTER_HOST='dummy_host'
:: Query SHOW SLAVE STATUS
Query RESET SLAVE /*!50516 ALL */
Query SELECT RELEASE_LOCK('MHA_Master_High_Availability_Monitor') As Value
Quit
Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
Query SET sql_log_bin=
Query SHOW PROCESSLIST
Query SELECT @@global.read_only As Value
Query SET GLOBAL read_only=
Query SELECT @@global.read_only As Value
Query SHOW PROCESSLIST
Query SET sql_log_bin=
Quit
Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
Query SET wait_timeout=
Query FLUSH TABLES WITH READ LOCK
Query SHOW MASTER STATUS
:: Query UNLOCK TABLES
Query CHANGE MASTER TO MASTER_HOST = '192.168.244.20' MASTER_USER = 'repl' MASTER_PASSWORD = <secret> MASTE
R_PORT = MASTER_LOG_FILE = 'mysql-bin.000010' MASTER_LOG_POS = Query SET GLOBAL relay_log_purge=
Query START SLAVE
Connect Out repl@192.168.244.20:
Query SHOW SLAVE STATUS
Query SELECT RELEASE_LOCK('MHA_Master_High_Availability_Failover') As Value
Quit
新master 192.168.244.20
:: Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
:: Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
Query SET wait_timeout=
Query SELECT @@global.server_id As Value
Query SELECT VERSION() AS Value
Query SELECT @@global.gtid_mode As Value
Query SHOW GLOBAL VARIABLES LIKE 'log_bin'
Query SHOW MASTER STATUS
Query SELECT @@global.datadir AS Value
Query SELECT @@global.slave_parallel_workers AS Value
Query SHOW SLAVE STATUS
Query SELECT @@global.read_only As Value
Query SELECT @@global.relay_log_purge As Value
Query SELECT @@global.relay_log_info_repository AS Value
Query SELECT @@global.datadir AS Value
Query SELECT @@global.relay_log_info_file AS Value
Query SHOW SLAVE STATUS
Query SELECT Repl_slave_priv AS Value FROM mysql.user WHERE user = 'repl' :: Query SELECT GET_LOCK('MHA_Master_High_Availability_Failover', '') AS Value
Query SHOW SLAVE STATUS
Query SHOW SLAVE STATUS :: Query SHOW PROCESSLIST
Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
Query SELECT @@global.read_only As Value
Query SELECT @@global.read_only As Value
Quit
Query SHOW SLAVE STATUS
Query SELECT MASTER_POS_WAIT('mysql-bin.000017','',) AS Result
Query STOP SLAVE SQL_THREAD
Query SHOW SLAVE STATUS
Query SHOW MASTER STATUS
Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
Query SET sql_log_bin=
Query SELECT @@global.read_only As Value
Query SET GLOBAL read_only=
Query SET sql_log_bin=
Quit
Query SELECT @@global.read_only As Value
Connect repl@node3 on
Query SELECT UNIX_TIMESTAMP()
Query SHOW VARIABLES LIKE 'SERVER_ID'
Query SET @master_heartbeat_period=
Query SET @master_binlog_checksum= @@global.binlog_checksum
Query SELECT @master_binlog_checksum
Query SELECT @@GLOBAL.GTID_MODE
Query SHOW VARIABLES LIKE 'SERVER_UUID'
Query SET @slave_uuid= '8a1093c8-1d00-11e7-954f-000c299a5715'
Binlog Dump Log: 'mysql-bin.000010' Pos:
:: Connect repl@node1 on
Query SELECT UNIX_TIMESTAMP()
Query SHOW VARIABLES LIKE 'SERVER_ID'
Query SET @master_heartbeat_period=
Query SET @master_binlog_checksum= @@global.binlog_checksum
Query SELECT @master_binlog_checksum
Query SELECT @@GLOBAL.GTID_MODE
Query SHOW VARIABLES LIKE 'SERVER_UUID'
Query STOP SLAVE
Query SET @slave_uuid= '2a6365e0-1d05-11e7-956d-000c29c64704'
Binlog Dump Log: 'mysql-bin.000010' Pos:
Query SHOW SLAVE STATUS
Query RESET SLAVE /*!50516 ALL */
Query SHOW SLAVE STATUS
Query SELECT RELEASE_LOCK('MHA_Master_High_Availability_Failover') As Value
Quit
slave 192.168.244.30
:: Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
:: Connect monitor@node4 on
Query set autocommit=
Query SELECT CONNECTION_ID() AS Value
Query SET wait_timeout=
Query SELECT @@global.server_id As Value
Query SELECT VERSION() AS Value
Query SELECT @@global.gtid_mode As Value
Query SHOW GLOBAL VARIABLES LIKE 'log_bin'
Query SHOW MASTER STATUS
Query SELECT @@global.datadir AS Value
Query SELECT @@global.slave_parallel_workers AS Value
Query SHOW SLAVE STATUS
Query SELECT @@global.read_only As Value
Query SELECT @@global.relay_log_purge As Value
Query SELECT @@global.relay_log_info_repository AS Value
Query SELECT @@global.datadir AS Value
Query SELECT @@global.relay_log_info_file AS Value
Query SHOW SLAVE STATUS
Query SELECT Repl_slave_priv AS Value FROM mysql.user WHERE user = 'repl' :: Query SELECT GET_LOCK('MHA_Master_High_Availability_Failover', '') AS Value
Query SHOW SLAVE STATUS
Query SHOW SLAVE STATUS :: Query SHOW SLAVE STATUS
:: Query SHOW SLAVE STATUS
Query SELECT MASTER_POS_WAIT('mysql-bin.000017','',) AS Result
Query STOP SLAVE SQL_THREAD
Query SHOW SLAVE STATUS
Query STOP SLAVE
Query STOP SLAVE
Query SHOW SLAVE STATUS
Query RESET SLAVE
Query CHANGE MASTER TO MASTER_HOST = '192.168.244.20' MASTER_USER = 'repl' MASTER_PASSWORD = <secret> MASTE
R_PORT = MASTER_LOG_FILE = 'mysql-bin.000010' MASTER_LOG_POS = Query SET GLOBAL relay_log_purge=
Query START SLAVE
Connect Out repl@192.168.244.20:
Query SHOW SLAVE STATUS
:: Query SELECT RELEASE_LOCK('MHA_Master_High_Availability_Failover') As Value
Quit
参考
《深入浅出MySQL》