千家信息网

半同步复制(Semi-synchronous Replication)

发表于:2025-11-10 作者:千家信息网编辑
千家信息网最后更新 2025年11月10日,MySQL5.5之前,MySQL的复制是异步操作,主库和从库的数据之间存在一定的延迟。这样存在一定的隐患:当主库上写入一个事务并交成功,而从库尚未得到主库推送的Binlog日志时,主库宕机了,例如主库
千家信息网最后更新 2025年11月10日半同步复制(Semi-synchronous Replication)

MySQL5.5之前,MySQL的复制是异步操作,主库和从库的数据之间存在一定的延迟。这样存在一定的隐患:当主库上写入一个事务并

交成功,而从库尚未得到主库推送的Binlog日志时,主库宕机了,例如主库可能因磁盘损坏、内存故障等造成主库上该事务Binlog丢失,此时从库就可能损失这个事务,从而造成主从不一致。


为了解决这个问题,MySQL5.5引入了半同步复制机制。在MySQL5.5之前的异步复制时,主库完成了Commit提交操作之后,在主库写

Binlog日志后即可成功返回客户端,无需等待Binlog日志传送给从库。


而半同步复制时,为了保证主库上的每一个Binlog事务都能够被可靠的复制到从库上,主库在每次事务提交时,并不及时反馈给前

台应用用户,而是等待其中一个从库也接收到Binlog事务并成功写入中继日志后,主库才返回Commit操作成功给客户端。半同步复制保证

了事务成功提交后,至少有两份日志记录,一份在主库的Binlog日志上,另一份在至少一台从库的中继日志Relay Log上,从而更进一步

保证了数据的完整性。


半同步复制模式下,如果主库在事务提交前的任何一个环节故障,则事务提交不成功,从库上也没收到事务对应的Binlog日志,所

以主从数据是一致的;假如在主库传送Binlog日志到从库过程中,从库宕机或网络故障,导致Binlog并没有及时地传送到从库上,此时

主库的事务会等待一段时间(时间长短由rpl_semi_sync_master_timeout设置的秒数决定),如果Binlog在这段时间内无法成功推送到从

库上,则MySQL自动调整复制为异步模式,事务正常返回提交结果给客户端


半同步很大程度上取决于主从库之间的网络状况,往返时延RTT越小决定了从库实时性越好。通俗的说主从之间的网络越快从库越实

时。

半同步模式是作为MySQL5.5的一个插件来实现的,主库和重开使用不同的插件,安装比较简单,在上一节异步复制的基础上,安装

半同步复制插件即可。


(1)、首先,判断MySQL服务器是否支持动态增加插件:


MariaDB [(none)]> select @@have_dynamic_loading;

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

| @@have_dynamic_loading |

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

| YES |

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

1 row in set (0.01 sec)


(2)、确认支持增加插件后,检查MySQL的安装目录下是否存在插件,一般默认在$MYSQL_HOME/lib/plugin目录下存在主库插件

semisync_master.so和从库插件semisync_slave.so


[root@www ~]# ls /usr/local/mysql/lib/plugin/semisync_*

-rwxr-xr-x 1 root root 510355 Jun 13 09:44 /usr/local/mysql/lib/plugin/semisync_master.so

-rwxr-xr-x 1 root root 427031 Jun 13 09:45 /usr/local/mysql/lib/plugin/semisync_slave.so


在主库上安装插件semisync_master.so:

MariaDB [(none)]> install plugin rpl_semi_sync_master soname 'semisync_master.so';

Query OK, 0 rows affected (0.07 sec)


从库上则安装semisync_slave.so:


MariaDB [(none)]> install plugin rpl_semi_sync_slave soname 'semisync_slave.so';

Query OK, 0 rows affected (0.08 sec)


安装完成后,从plugin表中看到刚才安装的插件:


MariaDB [(none)]> select * from mysql.plugin;

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

| name | dl |

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

| rpl_semi_sync_master | semisync_master.so |

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

1 row in set (0.00 sec)

MariaDB [(none)]> select * from mysql.plugin;

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

| name | dl |

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

| rpl_semi_sync_slave | semisync_slave.so |

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

1 row in set (0.00 sec)


也就是说,安装完成后,MySQL会在系统表plugin中记录刚才安装的插件,下次重启后会自动加载插件。


(3)、需要分别在主库和从库上配置参数打开半同步 semi-sync,默认半同步设置是不打开的,


主库上配置全局参数:


MariaDB [(none)]> set global rpl_semi_sync_master_enabled=1;

Query OK, 0 rows affected (0.01 sec)


MariaDB [(none)]> set global rpl_semi_sync_master_timeout=30000;

Query OK, 0 rows affected (0.00 sec)


MariaDB [pancou]> show variables like '%rpl_semi%';

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

| Variable_name | Value |

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

| rpl_semi_sync_master_enabled | ON |

| rpl_semi_sync_master_timeout | 30000 |

| rpl_semi_sync_master_trace_level | 32 |

| rpl_semi_sync_master_wait_no_slave | ON |

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

4 rows in set (0.00 sec)


从库上一样的配置参数:


MariaDB [(none)]> set global rpl_semi_sync_slave_enabled=1;

Query OK, 0 rows affected (0.00 sec)

MariaDB [pancou]> show variables like 'rpl_semi_sync%';

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

| Variable_name | Value |

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

| rpl_semi_sync_slave_enabled | ON |

| rpl_semi_sync_slave_trace_level | 32 |

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

2 rows in set (0.00 sec)


注意,由于之前配置的复制是异步复制,所以要重启一下从库上的I/O线程(如果是全新配置的半同步复制则不需要):


MariaDB [(none)]> stop slave io_thread;start slave io_thread;

Query OK, 0 rows affected (0.00 sec)


Query OK, 0 rows affected (0.00 sec)


至此半同步配置完毕,下面可以来验证一下。主库上通过show status命令能够看到当前半同步复制的一些状态值:


MariaDB [(none)]> show status like '%semi_sync%';

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

| Variable_name | Value |

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

| Rpl_semi_sync_master_clients | 1 |

| Rpl_semi_sync_master_net_avg_wait_time | 0 |

| Rpl_semi_sync_master_net_wait_time | 0 |

| Rpl_semi_sync_master_net_waits | 0 |

| Rpl_semi_sync_master_no_times | 0 |

| Rpl_semi_sync_master_no_tx | 0 |

| Rpl_semi_sync_master_status | ON |

| Rpl_semi_sync_master_timefunc_failures | 0 |

| Rpl_semi_sync_master_tx_avg_wait_time | 0 |

| Rpl_semi_sync_master_tx_wait_time | 0 |

| Rpl_semi_sync_master_tx_waits | 0 |

| Rpl_semi_sync_master_wait_pos_backtraverse | 0 |

| Rpl_semi_sync_master_wait_sessions | 0 |

| Rpl_semi_sync_master_yes_tx | 0 |

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

14 rows in set (0.00 sec)

着重关注以下3个状态值:

Rpl_semi_sync_master_status:值为ON,表示半同步复制目前处于打开状态。

Rpl_semi_sync_master_yes_tx:值为0,表示主库当前尚未有任何一个事务时通过半同步复制到从库。

Rpl_semi_sync_master_no_tx: 值为0,表示当前有0个事务不是通过半同步模式下从库及时响应的(记住只给值,后面有对比)。


执行一个事务,再检查一下:


MariaDB [pancou]> create table emp(emp_id smallint,emp_name varchar(10));

Query OK, 0 rows affected (0.10 sec)

MariaDB [pancou]> insert into emp values(1,'tom'),(2,'jerry');

Query OK, 2 rows affected (0.01 sec)

Records: 2 Duplicates: 0 Warnings: 0


MariaDB [pancou]> show status like '%semi_sync%';

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

| Variable_name | Value |

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

| Rpl_semi_sync_master_clients | 1 |

| Rpl_semi_sync_master_net_avg_wait_time | 1496 |

| Rpl_semi_sync_master_net_wait_time | 2992 |

| Rpl_semi_sync_master_net_waits | 2 |

| Rpl_semi_sync_master_no_times | 0 |

| Rpl_semi_sync_master_no_tx | 0 |

| Rpl_semi_sync_master_status | ON |

| Rpl_semi_sync_master_timefunc_failures | 0 |

| Rpl_semi_sync_master_tx_avg_wait_time | 1203 |

| Rpl_semi_sync_master_tx_wait_time | 2407 |

| Rpl_semi_sync_master_tx_waits | 2 |

| Rpl_semi_sync_master_wait_pos_backtraverse | 0 |

| Rpl_semi_sync_master_wait_sessions | 0 |

| Rpl_semi_sync_master_yes_tx | 2 |

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

14 rows in set (0.00 sec)


此时会发现Rpl_semi_sync_master_yes_tx值变为2,即刚才的create 和 insert事务通过半同步复制到从库上了,

Rpl_semi_sync_master_yes_tx 基数增加到2,到从库上确认一下:


MariaDB [pancou]> select * from emp;

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

| emp_id | emp_name |

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

| 1 | tom |

| 2 | jerry |

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

2 rows in set (0.00 sec)


再尝试一下网络异常的场景下,主库等待rpl_semi_sync_master_timeout毫秒超时后,自动转成异步复制的场景。

(1)、首先,在主库上确认半同步会等待30秒超时;


MariaDB [pancou]> show variables like '%semi%time%';

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

| Variable_name | Value |

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

| rpl_semi_sync_master_timeout | 30000 |

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

1 row in set (0.00 sec)


(2)、在主库上通过iptables命令模拟从库宕机或网络故障:


第二天

MariaDB [pancou]> set global rpl_semi_sync_master_enabled=1;

Query OK, 0 rows affected (0.00 sec)

show variables like 'rpl_semi_%';

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

| Variable_name | Value |

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

| rpl_semi_sync_master_enabled | ON |

| rpl_semi_sync_master_timeout | 10000 |

| rpl_semi_sync_master_trace_level | 32 |

| rpl_semi_sync_master_wait_no_slave | ON |

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

4 rows in set (0.01 sec)

MariaDB [pancou]> show status like '%semi_sync%';

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

| Variable_name | Value |

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

| Rpl_semi_sync_master_clients | 0 |

| Rpl_semi_sync_master_net_avg_wait_time | 0 |

| Rpl_semi_sync_master_net_wait_time | 0 |

| Rpl_semi_sync_master_net_waits | 0 |

| Rpl_semi_sync_master_no_times | 0 |

| Rpl_semi_sync_master_no_tx | 0 |

| Rpl_semi_sync_master_status | ON |

| Rpl_semi_sync_master_timefunc_failures | 0 |

| Rpl_semi_sync_master_tx_avg_wait_time | 0 |

| Rpl_semi_sync_master_tx_wait_time | 0 |

| Rpl_semi_sync_master_tx_waits | 0 |

| Rpl_semi_sync_master_wait_pos_backtraverse | 0 |

| Rpl_semi_sync_master_wait_sessions | 0 |

| Rpl_semi_sync_master_yes_tx | 0 |

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

14 rows in set (0.00 sec)

MariaDB [pancou]> delete from emp where emp_id=2;

Query OK, 1 row affected (0.01 sec)


MariaDB [pancou]> show status like '%semi_sync%';

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

| Variable_name | Value |

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

| Rpl_semi_sync_master_clients | 1 |

| Rpl_semi_sync_master_net_avg_wait_time | 1494 |

| Rpl_semi_sync_master_net_wait_time | 1494 |

| Rpl_semi_sync_master_net_waits | 1 |

| Rpl_semi_sync_master_no_times | 0 |

| Rpl_semi_sync_master_no_tx | 0 |

| Rpl_semi_sync_master_status | ON |

| Rpl_semi_sync_master_timefunc_failures | 0 |

| Rpl_semi_sync_master_tx_avg_wait_time | 0 |

| Rpl_semi_sync_master_tx_wait_time | 0 |

| Rpl_semi_sync_master_tx_waits | 0 |

| Rpl_semi_sync_master_wait_pos_backtraverse | 0 |

| Rpl_semi_sync_master_wait_sessions | 0 |

| Rpl_semi_sync_master_yes_tx | 1 |

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

14 rows in set (0.00 sec)

在从库上


iptables -I INPUT -s 192.168.110.128 -j DROP


(3)、在主库上执行一个事务并提交(默认提交即可),主库上的提交操作会被阻塞30秒:


MariaDB [pancou]> insert into emp values(2,'jerry');


---等待


新打开一个主库的线程


MariaDB [(none)]> show processlist\G

*************************** 1. row ***************************

Id: 7

User: root

Host: localhost

db: NULL

Command: Query

Time: 0

State: init

Info: show processlist

Progress: 0.000

*************************** 2. row ***************************

Id: 10

User: root

Host: localhost

db: pancou

Command: Query

Time: 4

State: Waiting for semi-sync ACK from slave

Info: insert into emp values(2,'jerry')

Progress: 0.000

*************************** 3. row ***************************

Id: 11

User: repl_user

Host: 192.168.110.130:52165

db: NULL

Command: Binlog Dump

Time: 2037

State: Reading from net

Info: NULL

Progress: 0.000

3 rows in set (0.00 sec)


MariaDB [(none)]> show processlist\G

*************************** 1. row ***************************

Id: 6

User: root

Host: localhost

db: pancou

Command: Sleep

Time: 248

State:

Info: NULL

Progress: 0.000

*************************** 2. row ***************************

Id: 7

User: root

Host: localhost

db: NULL

Command: Query

Time: 0

State: init

Info: show processlist

Progress: 0.000

2 rows in set (0.00 sec)

(4)、主库上等待30秒后


Query OK, 1 row affected (10.00 sec)


MariaDB [pancou]> insert into emp values(2,'jerry');

Query OK, 1 row affected (0.01 sec)


检查半同步的状态值


MariaDB [pancou]> show status like '%semi_sync%';

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

| Variable_name | Value |

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

| Rpl_semi_sync_master_clients | 0 |

| Rpl_semi_sync_master_net_avg_wait_time | 15016390 |

| Rpl_semi_sync_master_net_wait_time | 30032780 |

| Rpl_semi_sync_master_net_waits | 2 |

| Rpl_semi_sync_master_no_times | 1 |

| Rpl_semi_sync_master_no_tx | 2 |

| Rpl_semi_sync_master_status | OFF |

| Rpl_semi_sync_master_timefunc_failures | 0 |

| Rpl_semi_sync_master_tx_avg_wait_time | 0 |

| Rpl_semi_sync_master_tx_wait_time | 0 |

| Rpl_semi_sync_master_tx_waits | 0 |

| Rpl_semi_sync_master_wait_pos_backtraverse | 0 |

| Rpl_semi_sync_master_wait_sessions | 0 |

| Rpl_semi_sync_master_yes_tx | 1 |

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

14 rows in set (0.00 sec)


Rpl_semi_sync_master_status | OFF

Rpl_semi_sync_master_yes_tx | 1 刚才并没有通过半同步复制完成,所以半同步并没有累加

Rpl_semi_sync_master_no_tx | 2

继续之前的场景下从库正常连接上主机之后,确认是否会自动切换回半同步复制模式。

iptables -D INPUT 1


检查从库上的状态:


MariaDB [pancou]> show slave status\G

*************************** 1. row ***************************

Slave_IO_State: Waiting for master to send event

Master_Host: 192.168.110.128

Master_User: repl_user

Master_Port: 3306

Connect_Retry: 60

Master_Log_File: mysql-bin.000005

Read_Master_Log_Pos: 1940

Relay_Log_File: www-relay-bin.000011

Relay_Log_Pos: 1024

Relay_Master_Log_File: mysql-bin.000005

Slave_IO_Running: Yes

Slave_SQL_Running: Yes

Replicate_Do_DB:

Replicate_Ignore_DB:


(2)、检查半同步复制状态值


MariaDB [pancou]> show status like '%semi_sync%';

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

| Variable_name | Value |

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

| Rpl_semi_sync_master_clients | 1 |

| Rpl_semi_sync_master_net_avg_wait_time | 10019692 |

| Rpl_semi_sync_master_net_wait_time | 30059077 |

| Rpl_semi_sync_master_net_waits | 3 |

| Rpl_semi_sync_master_no_times | 1 |

| Rpl_semi_sync_master_no_tx | 3 |

| Rpl_semi_sync_master_status | ON |

| Rpl_semi_sync_master_timefunc_failures | 0 |

| Rpl_semi_sync_master_tx_avg_wait_time | 0 |

| Rpl_semi_sync_master_tx_wait_time | 0 |

| Rpl_semi_sync_master_tx_waits | 0 |

| Rpl_semi_sync_master_wait_pos_backtraverse | 0 |

| Rpl_semi_sync_master_wait_sessions | 0 |

| Rpl_semi_sync_master_yes_tx | 1 |

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

14 rows in set (0.00 sec)


Rpl_semi_sync_master_status 值自动由OFF变为 | ON ,说明主库检测到正常以后主库到从库的复制方式自动切换为

半同步方式

(3)、在主库上做一个测试

MariaDB [pancou]> insert into emp values(3,'bob');

Query OK, 1 row affected (0.01 sec)


MariaDB [pancou]> show status like '%semi_sync%';

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

| Variable_name | Value |

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

| Rpl_semi_sync_master_clients | 1 |

| Rpl_semi_sync_master_net_avg_wait_time | 7515331 |

| Rpl_semi_sync_master_net_wait_time | 30061326 |

| Rpl_semi_sync_master_net_waits | 4 |

| Rpl_semi_sync_master_no_times | 1 |

| Rpl_semi_sync_master_no_tx | 3 |

| Rpl_semi_sync_master_status | ON |

| Rpl_semi_sync_master_timefunc_failures | 0 |

| Rpl_semi_sync_master_tx_avg_wait_time | 742 |

| Rpl_semi_sync_master_tx_wait_time | 742 |

| Rpl_semi_sync_master_tx_waits | 1 |

| Rpl_semi_sync_master_wait_pos_backtraverse | 0 |

| Rpl_semi_sync_master_wait_sessions | 0 |

| Rpl_semi_sync_master_yes_tx | 2 |

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

14 rows in set (0.01 sec)

提交一个事务后,Rpl_semi_sync_master_yes_tx 由1变为 2,确认刚才事务是半同步复制。



从半同步复制的流程会发现,半同步复制的 "半" 就体现在:虽然主库和从库的Binlog日志是同步的,但是主库并不等待从库应用这部分

日志就返回提交结果,这部分操作是异步的,从库的数据并不是和主库实时同步的,所以只能称之为半同步而不是完全的实时同步。


0