MHA就提供了這樣一種優雅的方式，只會堵塞業務0.5~2s的時間，在這段時間內，業務無法讀取和寫入。
叢集資訊
角色 IP地址 ServerID 型別
Master 192.168.244.10 1 寫入
Candicate master 192.168.244.20 2 讀
Slave 192.168.244.30 3 讀
Monitor host 192.168.244.40 監控叢集組
MHA具體的搭建步驟和原理，可參考另外一篇部落格：
線上切換的步驟
1. 關閉MHA監控
# masterha_stop --conf=/etc/masterha/app1.cnf
2. 線上切換
# /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是將原master切換為新主的slave，預設情況下，是不新增的。
--running_updates_limit預設為1s，即如果主從延遲時間（Seconds_Behind_Master），或master show processlist中dml操作大於1s，則不會執行切換。
線上切換的輸出
Tue Apr :: - [] MHA::MasterRotate version :: - [ :: - [ :: - [] * Phase :: - [ :: - [warning] Global configuration /etc/ :: - [] Reading application default configuration from /etc/masterha/ :: - [] Reading server configuration from /etc/masterha/ :: - [] GTID failover mode = :: - [] Current Alive Master: .(.: :: - [ :: - [] .(.:) Version=.--bin:enabledTue Apr :: - [] Replicating from .(.: :: - [] Primary candidate :: - [] .(.:) Version=.--bin:enabledTue Apr :: - [] Replicating from .(.:.(:)? (YES/no): yes Tue Apr :: - [] Executing FLUSH NO_WRITE_TO_BINLOG TABLES. This may take :: - [ :: - [ :: - [] Checking replication health on . :: - [ :: - [] Checking replication health on . :: - [ :: - [] . :: - [.(.:+--.(.:+--.(.:.(.:+--.(.:+--.(.:.(.:) to .(.:)? (yes/ :: - [] Checking whether .(.:) is ok :: - [ :: - [] .(.: :: - [] .(.: :: - [] ** Phase :: - [ :: - [] * Phase :: - [ :: - [] Executing master ip online change script to disable :: - [] /usr/local/bin/master_ip_online_change --command=stop --orig_master_host=. --=. --orig_master_port= --orig_master_user= --orig_master_password= --new_master_host=. --new_master_ip=. --new_master_port= --new_master_user= --new_master_password= --orig_master_ssh_user=root --new_master_ssh_user=root --orig_master_is_new_slaveTue Apr :: :: Set read_only= :: :: . :: - [ :: - [ :: - [ :: - [ :: - [] Orig master binlog:pos is mysql-bin.: :: - [] Waiting to execute all relay logs on .(.: :: - [] master_pos_wait(mysql-bin.:) completed on .(.:). Executed :: - [] :: - [] Getting new masterTue Apr :: - [] mysql-bin.: :: - [=, MASTER_PORT=, MASTER_LOG_FILE=, MASTER_LOG_POS=, MASTER_USER=, MASTER_PASSWORD=;Tue Apr :: - [] Executing master ip online change script to allow :: - [] /usr/local/bin/master_ip_online_change --command=start --orig_master_host=. --=. --orig_master_port= --orig_master_user= --orig_master_password= --new_master_host=. --new_master_ip=. --new_master_port= --new_master_user= --new_master_password= --orig_master_ssh_user=root --new_master_ssh_user=root --orig_master_is_new_slaveTue Apr :: Set read_only=. on the new master: . :: - [ :: - [ :: - [] * Switching slaves :: - [ :: - [] -- Slave switch on host .(.:) started, pid: :: - [ :: - [] Log messages from . :: - [ :: - [] Waiting to execute all relay logs on .(.: :: - [] master_pos_wait(mysql-bin.:) completed on .(.:). Executed :: - [] :: - [] Resetting slave .(.:) and starting replication from the new master .(.:)..Tue Apr :: - [ :: - [ :: - [] End of log messages from . :: - [ :: - [] -- Slave switch on host .(.: :: - [ :: - [ :: - [ :: - [ :: - [] Resetting slave .(.:) and starting replication from the new master .(.:)..Tue Apr :: - [ :: - [ :: - [ :: - [ :: - [] * Phase :: - [ :: - [] .: Resetting slave :: - [] Switching master to .(.:) completed successfully.
MHA線上切換的原理
1. 檢查當前的配置資訊及主從伺服器的資訊
包括讀取MHA的配置檔案/etc/masterha/app1.cnf及檢查當前slave的健康狀態
2. 阻止對當前master的更新
主要透過如下步驟：
1> 等待1.5s（$time_until_kill_threads*100ms），等待當前連線斷開。
2> 執行 read_only=1，阻止新的DML操作
3> 等待0.5s，等待當前DML操作完成。
4> kill掉所有連線。
5> FLUSH NO_WRITE_TO_BINLOG TABLES
6> FLUSH TABLES WITH READ LOCK
3. 等待新master執行完所有的relay log
Waiting to execute all relay logs on 192.168.244.20(192.168.244.20:3306)..
4. 將新master的read_only設定為off，並新增VIP
5. slave切換到新master上。
1> 等待slave（192.168.244.30）應用完原主從複製產生的relay log，然後執行change master操作切換到新master上。
2> 釋放原master上加的鎖。
3> 因masterha_master_switch命令列中帶有--orig_master_is_new_slave引數，故原master也切換為新master的從。
6. 清理新master的相關資訊。
主要是執行了reset slave all操作，清除之前的複製資訊。
MHA線上切換需滿足的條件
MHA在執行線上切換之前，會判斷當前的主從複製資訊，只有滿足了以下條件，才能執行切換動作：
1. 所有SLAVE的IO執行緒和SQL執行緒都在執行。
2. 所有slave的Seconds_Behind_Master小於或等於running_updates_limit的值，該引數如果沒有顯示指定的話，則預設為1s
3. 在master上，透過show processlist輸出，沒有一個DML操作的時間大於running_updates_limit的值。
線上切換時，開啟general log，各個伺服器的操作資訊
注：在執行masterha_master_switch命令時，會有兩次確認操作
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):
以下輸出中間都有兩次空白，其中第一次空白之前的輸出對應第一次確認之前，第二次之前的輸出對應第二次確認之前。
原master 192.168.244.10
170412 16:52:38 23 Connect monitor@node4 on
23 Query set autocommit=1
23 Query SELECT CONNECTION_ID() AS Value170412 16:52:39 24 Connect monitor@node4 on
24 Query set autocommit=1
24 Query SELECT CONNECTION_ID() AS Value 24 Query SET wait_timeout=86400
24 Query SELECT @@global.server_id As Value 24 Query SELECT VERSION() AS Value 24 Query SELECT @@global.gtid_mode As Value 24 Query SHOW GLOBAL VARIABLES LIKE 'log_bin'
24 Query SHOW MASTER STATUS 24 Query SELECT @@global.datadir AS Value 24 Query SELECT @@global.slave_parallel_workers AS Value 24 Query SHOW SLAVE STATUS 24 Query SELECT @@global.read_only As Value 24 Query SELECT @@global.relay_log_purge As Value
170412 16:54:06 24 Query FLUSH NO_WRITE_TO_BINLOG TABLES 24 Query SELECT GET_LOCK('MHA_Master_High_Availability_Monitor', '0') AS Value 24 Query SHOW PROCESSLIST
170412 16:55:51 24 Query SHOW SLAVE STATUS 24 Query CHANGE MASTER TO MASTER_HOST='dummy_host'170412 16:55:52 24 Query SHOW SLAVE STATUS 24 Query RESET SLAVE /*!50516 ALL */
24 Query SELECT RELEASE_LOCK('MHA_Master_High_Availability_Monitor') As Value 24 Quit
25 Connect monitor@node4 on
25 Query set autocommit=1
25 Query SELECT CONNECTION_ID() AS Value 25 Query SET sql_log_bin=0
25 Query SHOW PROCESSLIST 25 Query SELECT @@global.read_only As Value 25 Query SET GLOBAL read_only=1
25 Query SELECT @@global.read_only As Value 25 Query SHOW PROCESSLIST 25 Query SET sql_log_bin=1
25 Quit
26 Connect monitor@node4 on
26 Query set autocommit=1
26 Query SELECT CONNECTION_ID() AS Value 26 Query SET wait_timeout=86400
26 Query FLUSH TABLES WITH READ LOCK 26 Query SHOW MASTER STATUS170412 16:55:53 26 Query UNLOCK TABLES 26 Query CHANGE MASTER TO MASTER_HOST = '192.168.244.20' MASTER_USER = 'repl' MASTER_PASSWORD = <secret> MASTE
R_PORT = 3306 MASTER_LOG_FILE = 'mysql-bin.000010' MASTER_LOG_POS = 120 26 Query SET GLOBAL relay_log_purge=0
26 Query START SLAVE 27 Connect Out repl@192.168.244.20:3306
26 Query SHOW SLAVE STATUS 26 Query SELECT RELEASE_LOCK('MHA_Master_High_Availability_Failover') As Value 26 Quit
新master 192.168.244.20
170412 16:52:38 23 Connect monitor@node4 on
23 Query set autocommit=1
23 Query SELECT CONNECTION_ID() AS Value170412 16:52:39 24 Connect monitor@node4 on
24 Query set autocommit=1
24 Query SELECT CONNECTION_ID() AS Value 24 Query SET wait_timeout=86400
24 Query SELECT @@global.server_id As Value 24 Query SELECT VERSION() AS Value 24 Query SELECT @@global.gtid_mode As Value 24 Query SHOW GLOBAL VARIABLES LIKE 'log_bin'
24 Query SHOW MASTER STATUS 24 Query SELECT @@global.datadir AS Value 24 Query SELECT @@global.slave_parallel_workers AS Value 24 Query SHOW SLAVE STATUS 24 Query SELECT @@global.read_only As Value 24 Query SELECT @@global.relay_log_purge As Value 24 Query SELECT @@global.relay_log_info_repository AS Value 24 Query SELECT @@global.datadir AS Value 24 Query SELECT @@global.relay_log_info_file AS Value 24 Query SHOW SLAVE STATUS 24 Query SELECT Repl_slave_priv AS Value FROM mysql.user WHERE user = 'repl'
170412 16:54:06 24 Query SELECT GET_LOCK('MHA_Master_High_Availability_Failover', '0') AS Value 24 Query SHOW SLAVE STATUS 24 Query SHOW SLAVE STATUS
170412 16:55:52 24 Query SHOW PROCESSLIST 25 Connect monitor@node4 on
25 Query set autocommit=1
25 Query SELECT CONNECTION_ID() AS Value 25 Query SELECT @@global.read_only As Value 25 Query SELECT @@global.read_only As Value 25 Quit
24 Query SHOW SLAVE STATUS 24 Query SELECT MASTER_POS_WAIT('mysql-bin.000017','120',0) AS Result 24 Query STOP SLAVE SQL_THREAD 24 Query SHOW SLAVE STATUS 24 Query SHOW MASTER STATUS 26 Connect monitor@node4 on
26 Query set autocommit=1
26 Query SELECT CONNECTION_ID() AS Value 26 Query SET sql_log_bin=0
26 Query SELECT @@global.read_only As Value 26 Query SET GLOBAL read_only=0
26 Query SET sql_log_bin=1
26 Quit
24 Query SELECT @@global.read_only As Value 27 Connect repl@node3 on
27 Query SELECT UNIX_TIMESTAMP() 27 Query SHOW VARIABLES LIKE 'SERVER_ID'
27 Query SET @master_heartbeat_period= 1799999979520
27 Query SET @master_binlog_checksum= @@global.binlog_checksum 27 Query SELECT @master_binlog_checksum 27 Query SELECT @@GLOBAL.GTID_MODE 27 Query SHOW VARIABLES LIKE 'SERVER_UUID'
27 Query SET @slave_uuid= '8a1093c8-1d00-11e7-954f-000c299a5715'
27 Binlog Dump Log: 'mysql-bin.000010' Pos: 120170412 16:55:53 28 Connect repl@node1 on
28 Query SELECT UNIX_TIMESTAMP() 28 Query SHOW VARIABLES LIKE 'SERVER_ID'
28 Query SET @master_heartbeat_period= 1799999979520
28 Query SET @master_binlog_checksum= @@global.binlog_checksum 28 Query SELECT @master_binlog_checksum 28 Query SELECT @@GLOBAL.GTID_MODE 28 Query SHOW VARIABLES LIKE 'SERVER_UUID'
24 Query STOP SLAVE 28 Query SET @slave_uuid= '2a6365e0-1d05-11e7-956d-000c29c64704'
28 Binlog Dump Log: 'mysql-bin.000010' Pos: 120
24 Query SHOW SLAVE STATUS 24 Query RESET SLAVE /*!50516 ALL */
24 Query SHOW SLAVE STATUS 24 Query SELECT RELEASE_LOCK('MHA_Master_High_Availability_Failover') As Value 24 Quit
slave 192.168.244.30
170412 16:52:37 16 Connect monitor@node4 on
16 Query set autocommit=1
16 Query SELECT CONNECTION_ID() AS Value170412 16:52:38 17 Connect monitor@node4 on
17 Query set autocommit=1
17 Query SELECT CONNECTION_ID() AS Value 17 Query SET wait_timeout=86400
17 Query SELECT @@global.server_id As Value 17 Query SELECT VERSION() AS Value 17 Query SELECT @@global.gtid_mode As Value 17 Query SHOW GLOBAL VARIABLES LIKE 'log_bin'
17 Query SHOW MASTER STATUS 17 Query SELECT @@global.datadir AS Value 17 Query SELECT @@global.slave_parallel_workers AS Value 17 Query SHOW SLAVE STATUS 17 Query SELECT @@global.read_only As Value 17 Query SELECT @@global.relay_log_purge As Value 17 Query SELECT @@global.relay_log_info_repository AS Value 17 Query SELECT @@global.datadir AS Value 17 Query SELECT @@global.relay_log_info_file AS Value 17 Query SHOW SLAVE STATUS 17 Query SELECT Repl_slave_priv AS Value FROM mysql.user WHERE user = 'repl'
170412 16:54:05 17 Query SELECT GET_LOCK('MHA_Master_High_Availability_Failover', '0') AS Value 17 Query SHOW SLAVE STATUS 17 Query SHOW SLAVE STATUS
170412 16:55:50 17 Query SHOW SLAVE STATUS170412 16:55:51 17 Query SHOW SLAVE STATUS 17 Query SELECT MASTER_POS_WAIT('mysql-bin.000017','120',0) AS Result 17 Query STOP SLAVE SQL_THREAD 17 Query SHOW SLAVE STATUS 17 Query STOP SLAVE 17 Query STOP SLAVE 17 Query SHOW SLAVE STATUS 17 Query RESET SLAVE 17 Query CHANGE MASTER TO MASTER_HOST = '192.168.244.20' MASTER_USER = 'repl' MASTER_PASSWORD = <secret> MASTE
R_PORT = 3306 MASTER_LOG_FILE = 'mysql-bin.000010' MASTER_LOG_POS = 120 17 Query SET GLOBAL relay_log_purge=0
17 Query START SLAVE 18 Connect Out repl@192.168.244.20:3306
17 Query SHOW SLAVE STATUS170412 16:55:52 17 Query SELECT RELEASE_LOCK('MHA_Master_High_Availability_Failover') As Value 17 Quit