MySQL半同步（四）流程和相关参数

迈不过友情╰ 2022-10-22 07:36 12阅读 0赞

作者：**八怪（高鹏）**中亦科技数据库专家

## 源码版本5.7.29 ##

本文作为新书《深入理解MySQL主从原理》的补充部分，供大家参考，如有误请见谅。

![3906e7b5b73abf843395c20ff9666435.png][]

数据结构和流程参考:

*  MySQL：半同步（一）主库端初始化流程 https://www.jianshu.com/p/9437e9059697
 *  MySQL：半同步（二）主库端回调函数 https://www.jianshu.com/p/d55d6383434d
 *  MySQL：半同步（三）从库端初始化和回调函数 https://www.jianshu.com/p/9fecae89f82c

此处不再单独列出

### 一、半同步的总体流程图 ###

![5f117ffb641bf81b9ec0f5c32a87dbbb.png][]

查看大图请移步 https://www.jianshu.com/p/5ebf4f4c1cf8

### 二、半同步中几个重要的状态值（主库） ###

半同步插件是否注册：半同步插件注册状态

ReplSemiSyncMaster::init_done_                                                                                                                                                                                                                                                                                               
     /* True when initObject has been called */

半同步插件是否启用：半同步开启状态

ReplSemiSyncMaster::master_enabled_                                                                                                                                                                                                                                                                                          
    /* semi-sync is enabled on the master */                                                                                                                                                                                                                                                                                     
    getMasterEnabled用于获取变量                                                                                                                                                                                                                                                                                                 
    set_master_enabled设置变量

完全取决于参数rpl\_semi\_sync\_master\_enabled

是否使用半同步(异步)：半同步复制方式

ReplSemiSyncMaster::state_                                                                                                                                                                                                                                                                                                   
    /* whether semi-sync is switched */

本状态非常重要，决定了是否是半同步复制方式，在函数ReplSemiSyncMaster::commitTrx中会完全依赖这个状态决定是否进行用户线程事务的等待，也是Rpl\_semi\_sync\_master\_status信息的来源。

### 三、相关参数（主库） ###

研究的参数包括

| rpl_semi_sync_master_enabled              | ON         |                                                                                                                                                                                                                                                                   
    | rpl_semi_sync_master_timeout              | 10000      |                                                                                                                                                                                                                                                                   
    | rpl_semi_sync_master_trace_level          | 32         |                                                                                                                                                                                                                                                                   
    | rpl_semi_sync_master_wait_for_slave_count | 1          |                                                                                                                                                                                                                                                                   
    | rpl_semi_sync_master_wait_no_slave        | ON         |                                                                                                                                                                                                                                                                   
    | rpl_semi_sync_master_wait_point           | AFTER_SYNC |

##### rpl\_semi\_sync\_master\_enabled #####

默认为false

生效点：

*  插件初始化初始化生效，参考第一节初始化函数ReplSemiSyncMaster::initObject
 *  Ack\_receiver线程初始化生效，参考第一节初始化函数Ack\_receiver::init
 *  参数动态生效，参考函数fix\_rpl\_semi\_sync\_master\_enabled

fix\_rpl\_semi\_sync\_master\_enabled函数会动态转换为半同步开启状态，并且开启Ack\_receiver线程，当然根据参数rpl\_semi\_sync\_master\_wait\_for\_slave\_countrpl\_semi\_sync\_master\_wait\_no\_slave  
的设置，考虑是否使用同步复制方式，并且本参数会设置到 ReplSemiSyncMaster::master\_enabled\_ 全局变量。如果是关闭则会转化为异步复制switch\_off()，并且做一些清理操作，比如active\_tranxs\_。

##### rpl\_semi\_sync\_master\_timeout #####

默认为10000毫秒，即10秒

生效点：

*  等待需要的ack反馈的事务超时，生效函数为ReplSemiSyncMaster::commitTrx，如果超时那么切换为异步复制，调用switch\_off函数
 *  插件初始化生效
 *  参数动态生效，参考函数fix\_rpl\_semi\_sync\_master\_timeout设置方式就是赋值给ReplSemiSyncMaster::wait\_timeout\_

##### rpl\_semi\_sync\_master\_wait\_for\_slave\_count #####

默认为on  
本参数可以和参数rpl\_semi\_sync\_master\_wait\_no\_slave共同作用共同决定半同步复制还是异步复制，这里先说rpl\_semi\_sync\_master\_wait\_no\_slave设置on的情况  
生效点：

*  初始化Ack\_recevier线程的AckContainer的元素个数，调用函数setWaitSlaveCount
 *  等待需要的ack反馈的事务超时，会通过AckContainer遍历找到最小的ack反馈点，参考repl\_semi\_report\_binlog\_sync函数
 *  动态参数修改fix\_rpl\_semi\_sync\_master\_wait\_for\_slave\_count，这里也会重新设置AckContainer的元素个数，调用函数setWaitSlaveCount

因此和rpl\_semi\_sync\_master\_timeout参数配合，可以决定收到多少个半同步从库的ack后进行反馈。

##### rpl\_semi\_sync\_master\_wait\_no\_slave #####

默认为on

如果为on则遵守如上rpl\_semi\_sync\_master\_wait\_for\_slave\_count的说明，这里重点说一下它设置为false有什么作用，如果其设置为false那么会有如下影响

生效点：

*  每次有半连接从库退出的时候，dump线程退出之前会通过回调函数repl\_semi\_binlog\_dump\_end，调入ReplSemiSyncMaster::remove\_slave函数进行判断rpl\_semi\_sync\_master\_clients(现有存在的半同步从库数量)是否小于了rpl\_semi\_sync\_master\_wait\_for\_slave\_count的个数，如果小于了则直接切换为异步复制方式。而不需要等到无ack反馈才进行切换。
 *  如果设置半同步开启状态，需要进行判断是否rpl\_semi\_sync\_master\_clients(现有存在的半同步从库数量)是否小于了pl\_semi\_sync\_master\_wait\_for\_slave\_count的个数，如果小于则不开启同步复制。参考函数ReplSemiSyncMaster::enableMaster
 *  动态参数修改fix\_rpl\_semi\_sync\_master\_wait\_no\_slave，根据rpl\_semi\_sync\_master\_clients(现有存在的半同步从库数量)是否为0来决定是否开启半同步复制方式，但是这里有一个疑问判断条件为rpl\_semi\_sync\_master\_clients==0。

因此本参数设置为false，设置rpl\_semi\_sync\_master\_timeout为一个很大的值，也不能保证主从一定是半同步复制。

##### rpl\_semi\_sync\_master\_trace\_level #####

默认为32

我们在前面已经多次说过，如果这个参数设置16，将会有大量的半同步日志输出。这里不做描述

##### rpl\_semi\_sync\_master\_wait\_point #####

默认为AFTER\_SYNC，这个参数也是在前面说过了，这里不做详细描述。

生效点：

*  主要位于回调函数repl\_semi\_report\_commit和repl\_semi\_report\_binlog\_sync中，用于确认是在commit队列处理之前进行用户的ack反馈还是commit队列处理之后进行ack反馈。

### 四、重点统计值（主库） ###

主要研究的统计值包含

| Rpl_semi_sync_master_clients               | 0     |                                                                                                                                                                                                                                                                       
    | Rpl_semi_sync_master_no_times              | 0     |                                                                                                                                                                                                                                                                       
    | Rpl_semi_sync_master_no_tx                 | 0     |                                                                                                                                                                                                                                                                       
    | Rpl_semi_sync_master_status                | OFF   |                                                                                                                                                                                                                                                                       
    | 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     |

##### Rpl\_semi\_sync\_master\_clients：当前处于半同步复制的从库个数 #####

*  增加：repl\_semi\_binlog\_dump\_start回调函数调用ReplSemiSyncMaster::add\_slave，及dump线程启动时刻
 *  减少：repl\_semi\_binlog\_dump\_end回调函数调用ReplSemiSyncMaster::remove\_slave，及dump线程退出时刻

##### Rpl\_semi\_sync\_master\_no\_times：关闭半同步复制方式的次数 #####

内部为rpl\_semi\_sync\_master\_off\_times

*  增加：ReplSemiSyncMaster::switch\_off函数

##### Rpl\_semi\_sync\_master\_no\_tx：没有在半同步复制方式状态下，事务commit队列处理的次数(after\_sync)/事务处理的次数(after\_commit) #####

内部为rpl\_semi\_sync\_master\_no\_transactions。

*  增加：用户线程repl\_semi\_report\_commit/repl\_semi\_report\_binlog\_sync调用ReplSemiSyncMaster::commitTrx函数的末尾

##### Rpl\_semi\_sync\_master\_yes\_tx：半同步复制方式状态下，事务commit队列处理的次数(after\_sync)/事务处理的次数(after\_commit) #####

内部为rpl\_semi\_sync\_master\_yes\_transactions。

*  增加：同上

##### Rpl\_semi\_sync\_master\_status：内部表示就是ReplSemiSyncMaster::state\_，也就是当前主库状态是否处于半同步复制方式下。 #####

参考ReplSemiSyncMaster::setExportStats函数

*  更改就是ReplSemiSyncMaster::switch\_off函数。

##### Rpl\_semi\_sync\_master\_wait\_sessions：主要用于表示当前处于堵塞的session数量。 #####

*  增加：用户线程repl\_semi\_report\_commit/repl\_semi\_report\_binlog\_sync调用ReplSemiSyncMaster::commitTrx函数，进入等待前
 *  减少：用户线程repl\_semi\_report\_commit/repl\_semi\_report\_binlog\_sync调用ReplSemiSyncMaster::commitTrx函数，退出等待后但是需要注意的能够进入这个函数的是commit的leader线程，因此实际堵塞的会话数量可能是远远大于这个数字。

##### Rpl\_semi\_sync\_master\_tx\_wait\_time：主要用于表示由于等待ack反馈而等待的时间，单位微秒(百万分之1秒) #####

内部表示为rpl\_semi\_sync\_master\_trx\_wait\_time。

*  增加：用户线程repl\_semi\_report\_commit/repl\_semi\_report\_binlog\_sync调用ReplSemiSyncMaster::commitTrx函数，等待ack之后（未超时）

##### Rpl\_semi\_sync\_master\_tx\_waits：同上，表示次数 #####

*  增加：用户线程repl\_semi\_report\_commit/repl\_semi\_report\_binlog\_sync调用ReplSemiSyncMaster::commitTrx函数，等待ack之后（未超时）

##### rpl\_semi\_sync\_master\_wait\_pos\_backtraverse：降低wait\_file\_name\_/wait\_file\_pos\_的次数 #####

*  增加：用户线程repl\_semi\_report\_commit/repl\_semi\_report\_binlog\_sync调用ReplSemiSyncMaster::commitTrx函数。

### 五、从库参数和统计值 ###

从库的参数和统计值很简单，和主库不同开启rpl\_semi\_sync\_slave\_enabled参数，不会动态让从库的Rpl\_semi\_sync\_slave\_status生效，启用半同步复制方式

##### 参数rpl\_semi\_sync\_slave\_enabled：用于表示是否从库开启了半同步状态 #####

生效点为

*  ReplSemiSyncSlave::initObject初始化的时候
 *  fix\_rpl\_semi\_sync\_slave\_enabled动态修改，但是需要注意这里的动态修改并不影响统计值Rpl\_semi\_sync\_slave\_status

##### 状态值Rpl\_semi\_sync\_slave\_status：用于表示从库是否是半同步复制方式 #####

无法动态开启这点和主库不同，并且此状态不会由于主库问题在运行期间改变，因此不能做判断半同步复制方式的根据，应该在主库判断。  
改变：

*  dump线程启动时刻根据，rpl\_semi\_sync\_slave\_enabled参数和主库是否安装半同步插件来共同判定是否开启。
 *  dump线程关闭时刻。

下面是一个测试（主库未安装半同步插件）

mysql> show variables like '%semi%slave_enabled';                                                                                                                                                                                                                                                                            
    +-----------------------------+-------+                                                                                                                                                                                                                                                                                      
    | Variable_name               | Value |                                                                                                                                                                                                                                                                                      
    +-----------------------------+-------+                                                                                                                                                                                                                                                                                      
    | rpl_semi_sync_slave_enabled | ON    |                                                                                                                                                                                                                                                                                      
    +-----------------------------+-------+                                                                                                                                                                                                                                                                                      
    1 row in set (0.00 sec)                                                                                                                                                                                                                                                                                                      
                                                                                                                                                                                                                                                                                                                                 
    mysql> start slave;                                                                                                                                                                                                                                                                                                          
    Query OK, 0 rows affected (0.02 sec)                                                                                                                                                                                                                                                                                         
                                                                                                                                                                                                                                                                                                                                 
    mysql> show status like '%Rpl_semi_sync_slave_status';                                                                                                                                                                                                                                                                       
    +----------------------------+-------+                                                                                                                                                                                                                                                                                       
    | Variable_name              | Value |                                                                                                                                                                                                                                                                                       
    +----------------------------+-------+                                                                                                                                                                                                                                                                                       
    | Rpl_semi_sync_slave_status | OFF   |                                                                                                                                                                                                                                                                                       
    +----------------------------+-------+

可以看到Rpl\_semi\_sync\_slave\_status并未启动。

下面是另外一个测试

mysql> show slave status \G;                                                                                                                                                                                                                                                                                                 
    *************************** 1. row ***************************                                                                                                                                                                                                                                                               
                   Slave_IO_State: Reconnecting after a failed master event read                                                                                                                                                                                                                                                 
                      Master_Host: 192.168.1.63                                                                                                                                                                                                                                                                                  
                      Master_User: repsem                                                                                                                                                                                                                                                                                        
                      Master_Port: 3325                                                                                                                                                                                                                                                                                          
                    Connect_Retry: 60                                                                                                                                                                                                                                                                                            
                  Master_Log_File: log_bin.000004                                                                                                                                                                                                                                                                                
              Read_Master_Log_Pos: 194                                                                                                                                                                                                                                                                                           
                   Relay_Log_File: relaylog.000020                                                                                                                                                                                                                                                                               
                    Relay_Log_Pos: 363                                                                                                                                                                                                                                                                                           
            Relay_Master_Log_File: log_bin.000004                                                                                                                                                                                                                                                                                
                 Slave_IO_Running: Connecting                                                                                                                                                                                                                                                                                    
                Slave_SQL_Running: Yes                                                                                                                                                                                                                                                                                           
                                                                                                                                                                                                                                                                                                                                 
    ERROR:                                                                                                                                                                                                                                                                                                                       
    No query specified                                                                                                                                                                                                                                                                                                           
                                                                                                                                                                                                                                                                                                                                 
    mysql> show status like '%Rpl_semi_sync_slave_status';                                                                                                                                                                                                                                                                       
    +----------------------------+-------+                                                                                                                                                                                                                                                                                       
    | Variable_name              | Value |                                                                                                                                                                                                                                                                                       
    +----------------------------+-------+                                                                                                                                                                                                                                                                                       
    | Rpl_semi_sync_slave_status | ON    |                                                                                                                                                                                                                                                                                       
    +----------------------------+-------+                                                                                                                                                                                                                                                                                       
    1 row in set (0.02 sec)

可以看到即便IO线程已经处于重连状态，Rpl\_semi\_sync\_slave\_status依旧为on。

### 六、运行状态下主库半同步状态切换时刻 ###

*  我们熟知的超过参数rpl\_semi\_sync\_master\_timeout的设置  
    调用为

ReplSemiSyncMaster::commitTrx                                                                                                                                                                                                                                                                                                
      ->switch_off

*  如果参数rpl\_semi\_sync\_master\_wait\_no\_slave设置false，那么当dump线程退出之前会判断是否小于rpl\_semi\_sync\_master\_wait\_for\_slave\_count的设置，小于切换为异步  
    调用为

repl_semi_binlog_dump_end                                                                                                                                                                                                                                                                                                    
     ->ReplSemiSyncMaster::remove_slave                                                                                                                                                                                                                                                                                          
       ->switch_off

*  如果Ack\_recevier线程发现反馈ack的半同步从库数量满足rpl\_semi\_sync\_master\_wait\_for\_slave\_count的设置，尝试使用半同步复制方式  
    调用为

ReplSemiSyncMaster::reportReplyPacket                                                                                                                                                                                                                                                                                        
     ->handleAck                                                                                                                                                                                                                                                                                                                 
       ->ReplSemiSyncMaster::reportReplyBinlog                                                                                                                                                                                                                                                                                   
        ->try_switch_on

*  设置参数rpl\_semi\_sync\_master\_enabled参数的时候，会尝试使用半同步复制模式，参考上文
 *  设置参数rpl\_semi\_sync\_master\_wait\_no\_slave参数的时候，会尝试改变半同步复制模式，参考上文

全文完。

Enjoy MySQL :)

[3906e7b5b73abf843395c20ff9666435.png]: /images/20221022/ea91feee001d4166926b75f5b1380967.png
[5f117ffb641bf81b9ec0f5c32a87dbbb.png]: /images/20221022/0be2bbe3935d40408ef53f04456fe382.png