MySQL并行DDL的实现方法_数据库_开发者_运维开发者技术经验分享

随着mysql版本的不断更新，对DDL操作的支持也在不断的完善和更新：比如从MySQL 5.6引入Online DDL，在MySQL 5.7对Online DDL进一步完善，到现在的8.0版本，则对DDL的实现重新进行了设计，比如DDL操作支持原子特性，在MySQL 8.0.27引入并行DDL。本篇就来探究一下MySQL 8.0.27的并行DDL对于DDL操作速度的提升。

MySQL 8.0.14引入了innodb_parallel_read_threads变量来控制扫描聚簇编程客栈索引的并行线程。MySQL 8.0.27引入了innodb_ddl_threads变量来控制用于创建二级索引时的并行线程数量，此参数一般和一并引入的innodb_ddl_buffer_size一起使用，innodb_ddl_buffer_size用于指定进行并行DDL操作时能够使用的buffer大小，buffer是在所有的DDL并行线程中平均分配的，所以一般如果调大innodb_ddl_threads变量时，也需要调大innodb_ddl_buffer_size的大小。

innodb_ddl_threads、innodb_ddl_buffer_size和innodb_parallel_read_threads的默认大小分别为：

mysql> select @@global.innodb_ddl_thread编程s;
+-----------------------------+
| @@global.innodb_ddl_threads |
+-----------------------------+
|                           4 |
+-----------------------------+
1 row in set (0.00 sec)
 
mysql> select @@global.innodb_ddl_buffer_size;
+---------------------------------+
| @@global.innodb_ddl_buffer_size |
+---------------------------------+
|                         1048576 |
+---------------------------------+
1 row in set (0.00 sec)
 
mysql> select @@global.innodb_parallel_read_threads;
+---------------------------------------+
| @@global.innodb_parallel_read_threads |
+---------------------------------------+
|                                     4 |
+---------------------------------------+
1 row in set (0.00 sec)

接下来测试一下调大innodb_ddl_threads、innodb_ddl_buffer_size和innodb_parallel_read_threads参数值对DDL操作的性能提升。

首先创建一张5000万的表：

-- 数据库版本为8.0.28
mysql> select @@version;
+----------+
| @@version|
+----------+
| 8.0.28   |
+----------+
1 row in set (0.00 sec)
 
-- buffer pool大小为24G
mysql> select @@global.innodb_buffer_pool_size;
+----------------------------------+
| @@global.innodb_buffer_pool_size |
+----------------------------------+
|                       1073741824 |
+----------------------------------+
1 row in set (0.001 sec)
 
mysql> create database action;
Query OK, 1 row affected (0.01 sec)
 
# sysbench /usr/share/sysbench/oltp_read_write.Lua --mysql-socket=/data/mysql/data/3306/mysqld.sock  --mysql-user=root --mysql-password='123' --mysql-db=action --tables=1 --table-size=50000000 --report-interval=1 --threads=8 prepare
 
mysql> select count(*) from action.sbtest1;
+----------+
| count(*) |
+----------+
| 50000000 |
+----------+
1 row in set (21.64 sec)
 
-- 表空间大小为12G
# ll -h
total 12G
-rw-r-----. 1 mysql mysql 12G Jan 20 17:38 sbtest1.ibd

分别测试不同的线程数量和缓冲区大小的DDL操作时间，例如：

-- 设置并发DDL线程为1
mysql> set innodb_ddl_threads = 1;
Query OK, 0 rows affected (0.01 sec)
 
-- 设置buffer大小为512M
mysql> set innodb_ddl_buffer_size = 536870912;
Query OK, 0 rows affected (0.00 sec)
 
-- 设置并行索引扫描线程为1
mysql> set innodb_parallel_read_pythonthreads = 1;
Query OK, 0 rows affected (0.01 sec)
 
-- 执行DDL操作
mysql> alter table action.sbtest1 add index idx_c(c);
Query OK, 0 rows affected (6 min 54.21 sec)
Records: 0  Duplicates: 0  Warnings: 0
 
-- 查看DDL的内存最大占用
mysql> select event_name,CURRENT_NUMBER_OF_BYTES_Uhttp://www.devze.comSED/1024/1024 from performance_schema.memory_summary_global_by_event_name where event_name='memory/innodb/ddl';
+-------------------+----------------------------------------+
| event_name        | CURRENT_NUMBER_OF_BYTES_USED/1024/1024 |
+-------------------+----------------------------------------+
| memory/innodb/ddl |                           513.08750916 |
+-------------------+----------------------------------------+
1 row in set (0.00 sec)

通过不断调整相关参数得到以下结果:

innodb_ddl_threads	innodb_ddl_buffer_size	innodb_parallel_read_threads	DDL占用最大内存	DDL时间
1	512M	1	513M	6 min 54.21 sec
2	1G	2	1230M	4 min 12.08 sec
4	2G	4	2735M	3 min 43.01 sec
8	4G	8	5791M	3 min 19.63 sec
16	8G	16	5975M	3 min 12.33 sec
32	16G	32	6084M	3 min 11.11 sec