MySQL 5.7版本新特性连载（四）

本文将和大家一起分享下5.7的新特性，不过我们要先从即将被删除的特性以及建议不再使用的特性说起。根据这些情况，我们在新版本及以后的版本中，应该不再使用，避免未来产生兼容性问题。

本文是基于MySQL-5.7.7-rc版本，未来可能 还会发生更多变化。

1、SQL MODE变化

a. 默认启用 STRICT_TRANS_TABLES 模式；

b. 对 ONLY_FULL_GROUP_BY 模式实现了更复杂的特性支持，并且也被默认启用；

c. 其他被默认启用的sql mode还有 NO_ENGINE_SUBSTITUTION；

【iMySQL建议】

对广大MySQL使用者而言，以往不是那么严格的模式还是很方便的，在5.7版本下可能会觉得略为不适，慢慢习惯吧。比如向一个20字符长度的VARCHAR列写入30个字符，在以前会自动阶段并给个提示告警，而在5.7版本下，则直接抛出错误了。个人认为这倒是一个好的做法，避免各种奇葩的写法。

【新特性实践】

-- 查看默认的 sql_mode

[[email protected]]> select @@sql_mode;

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

| @@sql_mode |

| ONLY_FULL_GROUP_BY,STRICT_TRANS_TABLES,NO_AUTO_CREATE_USER,NO_ENGINE_SUBSTITUTION |

-- 插入50个字符

[[email protected]]> insert into t_char select 0, repeat('x',50);

ERROR 1406 (22001): Data too long for column 'uname' at row 1

-- 修改本 session 的 sql_mode

[[email protected]]> set sql_mode = 'ONLY_FULL_GROUP_BY,NO_AUTO_CREATE_USER,NO_ENGINE_SUBSTITUTION';

Query OK, 0 rows affected (0.00 sec)

-- 去掉 STRICT_TRANS_TABLES 模式后

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

| ONLY_FULL_GROUP_BY,NO_AUTO_CREATE_USER,NO_ENGINE_SUBSTITUTION |

Query OK, 1 row affected, 1 warning (0.00 sec)  -- 提示有告警信息

Records: 1 Duplicates: 0 Warnings: 1

[[email protected]]> show warnings;

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

| Level | Code | Message |

| Warning | 1265 | Data truncated for column 'uname' at row 1 |

因为 uname 字段的长度为 40 个字符。

2、优化online操作，例如修改buffer pool、修改索引名（非主键）、修改REPLICATION FILTER、修改MATER而无需关闭SLAVE线程 等众多特性。

可以在线修改buffer pool对DBA来说实在太方便了，实例运行过程中可以动态调整，避免事先分配不合理的情况，不过 innodb_buffer_pool_instances 不能修改，而且在 innodb_buffer_pool_instances 大于 1 时，也不能将 buffer pool 调整到 1GB 以内，需要稍加注意。

如果是加大buffer pool，其过程大致是：

1、以innodb_buffer_pool_chunk_size为单位，分配新的内存pages；

2、扩展buffer pool的AHI(adaptive hash index)链表，将新分配的pages包含进来；

3、将新分配的pages添加到free list中；

如果是缩减buffer pool，其过程则大致是：

1、重整buffer pool，准备回收pages；

2、以innodb_buffer_pool_chunk_size为单位，释放删除这些pages（这个过程会有一点点耗时）；

3、调整AHI链表，使用新的内存地址。

实际测试时，发现在线修改 buffer poo 的代价并不大，SQL命令提交完毕后都是瞬间完成，而后台进程的耗时也并不太久。在一个并发128线程跑tpcc压测的环境中，将 buffer pool 从32G扩展到48G，后台线程耗时 3秒，而从 48G 缩减回 32G 则耗时 18秒，期间压测的事务未发生任何锁等待。

-- 演示1：从 1G 扩大到 16G

[[email protected]]> SET GLOBAL innodb_buffer_pool_size = 51539607552;

-- 看看日志记录

09:21:19.460543Z 0 [Note] InnoDB: Resizing buffer pool from 1073741824 to 17179869184. (unit=134217728)

09:21:19.468069Z 0 [Note] InnoDB: disabled adaptive hash index.

09:21:20.760724Z 0 [Note] InnoDB: buffer pool 0 : 60 chunks (491511 blocks) were added.

09:21:21.922869Z 0 [Note] InnoDB: buffer pool 1 : 60 chunks (491520 blocks) were added.

09:21:21.935114Z 0 [Note] InnoDB: buffer pool 0 : hash tables were resized.

09:21:21.947264Z 0 [Note] InnoDB: buffer pool 1 : hash tables were resized.

09:21:22.203031Z 0 [Note] InnoDB: Resized hash tables at lock_sys, adaptive hash index, dictionary.

09:21:22.203062Z 0 [Note] InnoDB: Completed to resize buffer pool from 1073741824 to 17179869184.

09:21:22.203075Z 0 [Note] InnoDB: Re-enabled adaptive hash index.

-- 演示2：从 16G 缩减到 1G

[[email protected]]> SET GLOBAL innodb_buffer_pool_size = 1073741824;

09:22:55.591669Z 0 [Note] InnoDB: Resizing buffer pool from 17179869184 to 1073741824. (unit=134217728)

09:22:55.680836Z 0 [Note] InnoDB: disabled adaptive hash index.

09:22:55.680864Z 0 [Note] InnoDB: buffer pool 0 : start to withdraw the last 491511 blocks.

09:22:55.765778Z 0 [Note] InnoDB: buffer pool 0 : withdrew 489812 blocks from free list. Tried to relocate 1698 pages (491510/491511).

09:22:55.774492Z 0 [Note] InnoDB: buffer pool 0 : withdrew 0 blocks from free list. Tried to relocate 1 pages (491511/491511).

09:22:55.782745Z 0 [Note] InnoDB: buffer pool 0 : withdrawn target 491511 blocks.

09:22:55.782786Z 0 [Note] InnoDB: buffer pool 1 : start to withdraw the last 491520 blocks.

09:22:55.892068Z 0 [Note] InnoDB: buffer pool 1 : withdrew 489350 blocks from free list. Tried to relocate 2166 pages (491517/491520).

09:22:55.900743Z 0 [Note] InnoDB: buffer pool 1 : withdrew 0 blocks from free list. Tried to relocate 2 pages (491519/491520).

09:22:55.908257Z 0 [Note] InnoDB: buffer pool 1 : withdrew 0 blocks from free list. Tried to relocate 0 pages (491519/491520).

09:22:55.915778Z 0 [Note] InnoDB: buffer pool 1 : withdrew 0 blocks from free list. Tried to relocate 1 pages (491520/491520).

09:22:55.923836Z 0 [Note] InnoDB: buffer pool 1 : withdrawn target 491520 blocks.

09:22:56.149172Z 0 [Note] InnoDB: buffer pool 0 : 60 chunks (491511 blocks) were freed.

09:22:56.308997Z 0 [Note] InnoDB: buffer pool 1 : 60 chunks (491520 blocks) were freed.

09:22:56.316258Z 0 [Note] InnoDB: buffer pool 0 : hash tables were resized.

09:22:56.324027Z 0 [Note] InnoDB: buffer pool 1 : hash tables were resized.

09:22:56.393589Z 0 [Note] InnoDB: Resized hash tables at lock_sys, adaptive hash index, dictionary.

09:22:56.393616Z 0 [Note] InnoDB: Completed to resize buffer pool from 17179869184 to 1073741824.

09:22:56.393628Z 0 [Note] InnoDB: Re-enabled adaptive hash index.

再来看下在线修改非主键索引名，直接用 ALTER TABLE RENAME INDEX 语法即可。

例如下面的SQL语法：

[[email protected]]> ALTER TABLE orders RENAME INDEX idx1 TO idxxx1;

Query OK, 0 rows affected (0.11 sec)

Records: 0 Duplicates: 0 Warnings: 0

可以看到，几乎瞬间完成，尽管我在执行这个SQL时正跑着64个并发tpcc压测。