GreatSQL The original content of the community shall not be used without authorization , For reprint, please contact the editor and indicate the source .

Introduction to semi synchronization

MASTER The node does not immediately return the result to the client after executing the transaction submitted by the client , But waiting for at least one SLAVE The node receives and writes relay log Is returned to the client .
Semi synchronous versus asynchronous replication , Improved data security , At the same time, it also causes a certain degree of delay , The delay is at least one TCP Round-trip time . therefore , Semi synchronous replication is best used in low latency networks .
MySQL from 5.5 Support semi synchronous replication from the beginning , stay 5.7.2 An improvement to semi synchronous replication was made in version ; The original semi synchronization strategy is AFTER_COMMIT The improved strategy is AFTER_SYNC The difference between the two is SLAVE node ACK The reply MASTER The timing is different .

Two modes are introduced

AFTER_COMMIT Model is introduced

MASTER Write each transaction to the binary log and save it on the disk , Send the transaction to SLAVE, Then submit the transaction to the storage engine for processing and submission , And then wait SLAVE Return confirmation , After receiving the confirmation message ,MASTER Returns the result to the client , Then the current client can continue to work .

AFTER_SYNC Model is introduced

MASTER Write each transaction to the binary log and save it on the disk , Send the transaction to SLAVE, And then wait SLAVE Return confirmation , After receiving the confirmation message , Submit the transaction to the storage engine for processing and submission , And return the result to the client , Then the current client can continue to work .

The two ways are compared

For the first AFTER_COMMIT The way , At present, the client only submits data to the storage engine and receives SLAVE After the confirmation of the return , Will receive the return result of the transaction . Received after transaction commit SLAVE Before returning the confirmation message , At this moment, other clients can see the transaction information submitted by the current client .
If SLAVE The node did not receive due to network and other reasons MASTER Transactions passed by nodes , and MASTER Node now crash 了 .HA failover , Clients are connected to SLAVE Node , At this time, I was in MASTER The transactions seen by the node are SLAVE The node does not see , There will be a loss of transactions .
For the second AFTER_SYNC The way , When the transaction is SLAVE After the confirmation MASTER After committing the transaction at the storage engine level , Only all clients can see the data changes caused by the transaction . therefore , All clients in MASTER See the same data at the same time .
When MASTER node crash Under the circumstances , All in MASTER All transactions committed on are copied to SLAVE( Save to relay log ). MASTER Server unexpected crash. At the moment HA Fail over to SALVE after , The data seen by the client is lossless , because SLAVE Is the latest .
Be careful , However , under these circumstances ,MASTER Cannot be directly restored to use , Because its binary log may contain uncommitted transactions , At this point, when binary logs are recovered and used for business needs , May result in contact with SLAVE The conflict of .

How to turn on semi synchronization

The way 1： Semi synchronization exists in the form of plug-ins , We can open it online directly （ This time use this method ）

#  The master node is turned on 
[[email protected]][(none)]>INSTALL PLUGIN rpl_semi_sync_master SONAME 'semisync_master.so';
Query OK, 0 rows affected (0.02 sec)

#  Open from node 
[[email protected]][(none)]>INSTALL PLUGIN rpl_semi_sync_slave SONAME 'semisync_slave.so';
Query OK, 0 rows affected (0.02 sec)

# PS： Generally, all nodes are deployed synchronously master and slave plug-in unit , In this way, it will be more convenient to handle the fault switching

The way 2： stay my.cnf Start in configuration

#  Both the master and slave nodes are configured synchronously 
plugin-load="rpl_semi_sync_master=semisync_master.so;rpl_sync_slave=semisync_slave.so"
rpl_semi_sync_master_enabled=1
rpl_semi_sync_slave_enabled=1

Check the opening of the plug-in

The way 1： Inquire about plugin

#  Master node view 
[[email protected]][test]>show plugins;
| rpl_semi_sync_master | ACTIVE | REPLICATION | semisync_master.so | GPL |

#  View from the node 
[[email protected]][(none)]>show plugins;
| rpl_semi_sync_slave | ACTIVE | REPLICATION | semisync_slave.so | GPL |

Mode two ： Inquire about information_schema.plugins More comprehensive information

#  Master node information 
(Thu Feb 17 03:03:12 2022)[[email protected]][(none)]>select * from information_schema.plugins where plugin_name like "%semi%"\G;
*************************** 1. row ***************************
           PLUGIN_NAME: rpl_semi_sync_master
        PLUGIN_VERSION: 1.0
         PLUGIN_STATUS: ACTIVE
           PLUGIN_TYPE: REPLICATION
   PLUGIN_TYPE_VERSION: 4.0
        PLUGIN_LIBRARY: semisync_master.so
PLUGIN_LIBRARY_VERSION: 1.10
         PLUGIN_AUTHOR: Oracle Corporation
    PLUGIN_DESCRIPTION: Semi-synchronous replication master
        PLUGIN_LICENSE: GPL
           LOAD_OPTION: ON
1 row in set (0.00 sec)

ERROR:
No query specified

#  From node information 
(Thu Feb 17 16:05:19 2022)[[email protected]][(none)]>select * from information_schema.plugins where plugin_name like "%semi%"\G;
*************************** 1. row ***************************
           PLUGIN_NAME: rpl_semi_sync_slave
        PLUGIN_VERSION: 1.0
         PLUGIN_STATUS: ACTIVE
           PLUGIN_TYPE: REPLICATION
   PLUGIN_TYPE_VERSION: 4.0
        PLUGIN_LIBRARY: semisync_slave.so
PLUGIN_LIBRARY_VERSION: 1.10
         PLUGIN_AUTHOR: Oracle Corporation
    PLUGIN_DESCRIPTION: Semi-synchronous replication slave
        PLUGIN_LICENSE: GPL
           LOAD_OPTION: ON
1 row in set (0.00 sec

Turn on the semi synchronous function

Because the plug-ins are installed online , So after the plug-in installation is completed , The service needs to be started

#  The master node enables semi synchronous replication 
[[email protected]][test]>SET GLOBAL rpl_semi_sync_master_enabled = on;
Query OK, 0 rows affected (0.00 sec)

#  Enable semi synchronous replication from node 
[[email protected]][(none)]>SET GLOBAL rpl_semi_sync_slave_enabled = on;
Query OK, 0 rows affected (0.00 sec)

After the above settings are completed , Restart from node IO Threads

(Mon Feb 14 15:19:58 2022)[[email protected]][(none)]>
(Mon Feb 14 15:19:58 2022)[[email protected]][(none)]>STOP SLAVE IO_THREAD;
Query OK, 0 rows affected, 1 warning (0.01 sec)

(Mon Feb 14 15:21:41 2022)[[email protected]][(none)]>START SLAVE IO_THREAD;
Query OK, 0 rows affected, 1 warning (0.01 sec)

See if semi synchronization is running

#  Master node 
[[email protected]][test]>show status like 'Rpl_semi_sync_master_status';
+-----------------------------+-------+
| Variable_name               | Value |
+-----------------------------+-------+
| Rpl_semi_sync_master_status | ON    |
+-----------------------------+-------+
1 row in set (0.00 sec)

#  From the node 
[[email protected]][(none)]>show status like 'Rpl_semi_sync_slave_status';
+----------------------------+-------+
| Variable_name              | Value |
+----------------------------+-------+
| Rpl_semi_sync_slave_status | ON    |
+----------------------------+-------+
1 row in set (0.01 sec)

Look at the master node error.log, It can be seen that the slave node has enabled semi synchronous replication

#  The key information  Start semi-sync binlog_dump to slave (server_id: 3306)
2022-02-14T02:16:35.411061-05:00 13 [Note] [MY-010014] [Repl] While initializing dump thread for slave with UUID <652ade08-8b1c-11ec-9f62-00155dcff90a>, found a zombie dump thread with the same UUID. Master is killing the zombie dump thread(12).
2022-02-14T02:16:35.411236-05:00 13 [Note] [MY-010462] [Repl] Start binlog_dump to master_thread_id(13) slave_server(3306), pos(, 4)
2022-02-14T02:16:35.411263-05:00 13 [Note] [MY-011170] [Repl] Start asynchronous binlog_dump to slave (server_id: 3306), pos(, 4).
2022-02-14T02:16:35.411419-05:00 12 [Note] [MY-011171] [Repl] Stop asynchronous binlog_dump to slave (server_id: 3306).
2022-02-14T02:19:33.913084-05:00 9 [Note] [MY-011130] [Repl] Semi-sync replication initialized for transactions.
2022-02-14T02:19:33.913133-05:00 9 [Note] [MY-011142] [Repl] Semi-sync replication enabled on the master.
2022-02-14T02:19:33.913638-05:00 0 [Note] [MY-011166] [Repl] Starting ack receiver thread.
2022-02-14T02:21:46.899725-05:00 14 [Note] [MY-010014] [Repl] While initializing dump thread for slave with UUID <652ade08-8b1c-11ec-9f62-00155dcff90a>, found a zombie dump thread with the same UUID. Master is killing the zombie dump thread(13).
2022-02-14T02:21:46.899894-05:00 14 [Note] [MY-010462] [Repl] Start binlog_dump to master_thread_id(14) slave_server(3306), pos(, 4)
2022-02-14T02:21:46.899953-05:00 14 [Note] [MY-011170] [Repl] Start semi-sync binlog_dump to slave (server_id: 3306), pos(, 4).

above ,MySQL Semi synchronous replication is completed ！

Semi synchronous parameter information

#  Master node parameter information 
[[email protected]][test]>show variables like '%Rpl%';
+-------------------------------------------+------------+
| Variable_name                             | Value      |
+-------------------------------------------+------------+
| rpl_read_size                             | 8192       |
| 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_stop_slave_timeout                    | 31536000   |
+-------------------------------------------+------------+
8 rows in set (0.00 sec)

Brief description of some parameters

Parameter name	purpose
rpl_semi_sync_master_enabled	Whether to enable semi synchronous replication ,ON For opening ,OFF To close
rpl_semi_sync_master_timeout	Master wait for Slave Of ack Timeout for reply , The default is 10 second （ The basic unit is milliseconds ）
rpl_semi_sync_master_trace_level	Debug level
rpl_semi_sync_master_wait_for_slave_count	Master Required after submission ACK Number of replies . If Slave The quantity is greater than or equal to this value , that Master It's normal , If it's below this value ,Master A timeout wait may occur during the transaction commit phase , When the wait exceeds the parameter （rpl_semi_sync_master_timeout） Set time ,Master It goes to asynchronous mode
rpl_semi_sync_master_wait_no_slave	Is in the timeout period , If Slave The number of downtime exceeds what should be received ack Number ,Master Whether to demote to asynchronous replication
rpl_semi_sync_master_wait_point	What is the way of semi synchronization , The default is AFTER_SYNC ,5.7.2 It used to be AFTER_COMMIT

#  From node parameter information 
[[email protected]][test]>show variables like '%Rpl%';
+---------------------------------+----------+
| Variable_name                   | Value    |
+---------------------------------+----------+
| rpl_read_size                   | 8192     |
| rpl_semi_sync_slave_enabled     | ON       |
| rpl_semi_sync_slave_trace_level | 32       |
| rpl_stop_slave_timeout          | 31536000 |
+---------------------------------+----------+
4 rows in set (0.00 sec)

Brief description of some parameters

Parameter name	purpose
rpl_semi_sync_master_enabled	Whether to enable semi synchronous replication ,ON For opening ,OFF To close
rpl_semi_sync_slave_trace_level	Debug level

Semi synchronous status information

Master node view

[[email protected]][test]> show status like '%Rpl_semi%';
+--------------------------------------------+-------+
| 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)

Brief description of the use of some parameters

Parameter name	purpose
Rpl_semi_sync_master_clients	How many slave libraries are currently set to semi synchronous mode , Does not contain asynchronous replication Slave example
Rpl_semi_sync_master_net_avg_wait_time	Waiting to return from the library ACK The average time to acknowledge a message ( Company ：ms)
Rpl_semi_sync_master_net_wait_time	Wait until you return from the library ACK Total time to acknowledge messages ( Company ：ms)
Rpl_semi_sync_master_net_waits	Waiting to return from the library ACK Total number of acknowledgement messages
Rpl_semi_sync_master_no_tx	The number of times a transaction is committed without a confirmation message returned from the library
Rpl_semi_sync_master_status	When semi synchronous replication is on , This variable is used to dynamically display the status of semi synchronous replication . The duty of OFF, Indicates that you are currently under asynchronous replication ; When waiting for a confirmation message to be returned from the library for more than rpl_semi_sync_master_timeout Change the value to OFF. When the value is set to ON, Indicates that you are currently under semi synchronous replication . When the slave database catches up with the master database, it is modified to ON.
Rpl_semi_sync_master_timefunc_failures	Call function gettimeofday() The number of failures
Rpl_semi_sync_master_tx_avg_wait_time	MASTER The average waiting time per transaction on the
Rpl_semi_sync_master_tx_wait_time	MASTER Total waiting time of transaction on
Rpl_semi_sync_master_tx_waits	MASTER Number of transactions waiting on
Rpl_semi_sync_master_wait_pos_backtraverse	MASTER The binary coordinate of the waiting event is lower than the total number of previous waiting events . When the order in which a transaction starts waiting for a reply is different from the order in which its binary log events are written , That's what happens . This situation is closing binlog_order_commit In this case, there may be .
Rpl_semi_sync_master_wait_sessions	The current number of replies waiting to return confirmation messages from the library
Rpl_semi_sync_master_yes_tx	Slave The number of successfully committed transactions confirmed by the node

#  State transition information from node 
show global status like '%semi%';
+----------------------------+-------+
| Variable_name              | Value |
+----------------------------+-------+
| Rpl_semi_sync_slave_status | ON    |
+----------------------------+-------+
1 row in set (0.00 sec)

Brief description of parameters

Parameter name	purpose
Rpl_semi_sync_slave_status	Dynamic parameters ,ON Indicates that semi synchronous replication is in use ,OFF Indicates asynchronous replication

Test the synchronization of semi synchronization

Whether semi synchronous will be degraded to asynchronous replication ？ Yes, it will .
When a semi synchronous replication timeout occurs （ from rpl_semi_sync_master_timeout Parameter control , In milliseconds , The default is 10000, namely 10s）, Will temporarily turn off semi synchronous replication , Switch to asynchronous replication .
When MASTER DUMP After the thread has sent all the events of a transaction , If in rpl_semi_sync_master_timeout Inside , Received a response from the library , Then the master and slave are restored to semi synchronous replication .

# 1. Turn off the slave node temporarily IO Threads 
[[email protected]][(none)]>STOP SLAVE IO_THREAD;
Query OK, 0 rows affected, 1 warning (0.02 sec)

# 2. The master node writes several pieces of test data 
[[email protected]][test]>insert into ptype values(4,'4','4'),(5,'5','5'),(6,'6','6');
Query OK, 3 rows affected (0.02 sec)

# 3. wait for 10s View replication status after , Semi synchronous has been turned off 
[[email protected]][test]>show status like 'Rpl_semi_sync_slave_status';
+----------------------------+-------+
| Variable_name              | Value |
+----------------------------+-------+
| Rpl_semi_sync_slave_status | OFF   |
+----------------------------+-------+
1 row in set (0.00 sec)

# 4. Open from node IO Threads 
[[email protected]][(none)]>START SLAVE IO_THREAD;
Query OK, 0 rows affected, 1 warning (0.02 sec)

# 5. Check the replication status again , Semi synchronous replication is automatically turned on 
[[email protected]][test]>show status like 'Rpl_semi_sync_slave_status';
+----------------------------+-------+
| Variable_name              | Value |
+----------------------------+-------+
| Rpl_semi_sync_slave_status | ON    |
+----------------------------+-------+
1 row in set (0.00 sec)

Enjoy GreatSQL :)

Article recommendation ：

GreatSQL Quarterly Report （2021.12.26）
https://mp.weixin.qq.com/s/FZ...

Technology sharing |sysbench Usage analysis of pressure measuring tools
https://mp.weixin.qq.com/s/m1...

Fault analysis | linux disk io High utilization , Analyze the correct posture
https://mp.weixin.qq.com/s/7c...

Technology sharing | Flashback at MySQL Implementation and improvement in
https://mp.weixin.qq.com/s/6j...

Wan Da #20, How to filter data in index push down
https://mp.weixin.qq.com/s/pt...

About GreatSQL

GreatSQL It is maintained by Wanli database MySQL Branch , Focus on Improvement MGR Reliability and performance , Support InnoDB Parallel query feature , It is suitable for financial grade applications MySQL Branch version .

Gitee:
https://gitee.com/GreatSQL/Gr...

GitHub:
https://github.com/GreatSQL/G...

Bilibili：
https://space.bilibili.com/13...

WeChat &QQ Group ：
Searchable add GreatSQL Community assistant wechat friend , Send verification information “ Add group ” Join in GreatSQL/MGR Exchange wechat group

QQ Group ：533341697
Wechat assistant ：wanlidbc