知识点:Mysql 数据库索引优化实战(4)

 

知识点:Mysql 索引原理完全手册(1)

知识点:Mysql 索引原理完全手册(2)

知识点:Mysql 索引优化实战(3)

知识点:Mysql 数据库索引优化实战(4)

一:插入订单

业务逻辑:插入订单数据,为了避免重复导单,一般会通过交易号去数据库中查询,判断该订单是否已经存在。

最基础的sql语句

mysql> select * from book_order where order_id = "10000";
+-------+--------------------+-------+------+----------+--------------+----------+------------------+-------------+-------------+------------+---------------------+
| id    | order_id     | general | net  | stock_id | order_status | description | finance_desc | create_type | order_state | creator | create_time          |
+-------+--------------------+-------+------+----------+--------------+----------+------------------+-------------+-------------+------------+---------------------+
| 10000 | 10000 |   6.6 | 6.13 |        1 |           10 | ok       | ok               | auto        |           1 | itdragon   | 2018-06-18 17:01:52 |
+-------+--------------------+-------+------+----------+--------------+----------+------------------+-------------+-------------+------------+---------------------+

mysql> explain select * from book_order where order_id = "10000";
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table               | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | book_order | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |    33.33 | Using where |
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+-------------+

几百上千万的订单,全表扫描!

通过explain命令可以清楚MySQL是如何处理sql语句的。打印的内容分别表示:

  • id : 查询序列号为1。
  • select_type : 查询类型是简单查询,简单的select语句没有union和子查询。
  • table : 表是 book_order。
  • partitions : 没有分区。
  • type : 连接类型,all表示采用全表扫描的方式。
  • possible_keys : 可能用到索引为null。
  • key : 实际用到索引是null。
  • key_len : 索引长度当然也是null。
  • ref : 没有哪个列或者参数和key一起被使用。
  • Extra : 使用了where查询。

是type为ALL,全表扫描,假设数据库中有几百万条数据,在没有索引的帮助下会异常卡顿。

初步优化:为order_id创建索引

mysql> create unique index idx_order_transaID on book_order (order_id);
mysql> explain select * from book_order where order_id = "10000";
+----+-------------+---------------------+------------+-------+--------------------+--------------------+---------+-------+------+----------+-------+
| id | select_type | table               | partitions | type  | possible_keys      | key                | key_len | ref   | rows | filtered | Extra |
+----+-------------+---------------------+------------+-------+--------------------+--------------------+---------+-------+------+----------+-------+
|  1 | SIMPLE      | book_order | NULL       | const | idx_order_transaID | idx_order_transaID | 453     | const |    1 |      100 | NULL  |
+----+-------------+---------------------+------------+-------+--------------------+--------------------+---------+-------+------+----------+-------+

这里创建的索引是唯一索引,而非普通索引。

唯一索引打印的type值是const。表示通过索引一次就可以找到。即找到值就结束扫描返回查询结果。

普通索引打印的type值是ref。表示非唯一性索引扫描。找到值还要继续扫描,直到将索引文件扫描完为止。(这里没有贴出代码),显而易见,const的性能要远高于ref。并且根据业务逻辑来判断,创建唯一索引是合情合理的。

再次优化:覆盖索引

mysql> explain select order_id from book_order where order_id = "10000";
+----+-------------+---------------------+------------+-------+--------------------+--------------------+---------+-------+------+----------+-------------+
| id | select_type | table               | partitions | type  | possible_keys      | key                | key_len | ref   | rows | filtered | Extra       |
+----+-------------+---------------------+------------+-------+--------------------+--------------------+---------+-------+------+----------+-------------+
|  1 | SIMPLE      | book_order | NULL       | const | idx_order_transaID | idx_order_transaID | 453     | const |    1 |      100 | Using index |
+----+-------------+---------------------+------------+-------+--------------------+--------------------+---------+-------+------+----------+-------------+

这里将select * from改为了select order_id from后,Extra 显示 Using index,表示该查询使用了覆盖索引,说明该sql语句的性能很好。若提示的是Using filesort(使用内部排序)和Using temporary(使用临时表)则表明该sql需要立即优化了。

根据业务逻辑来的,查询结构返回order_id 是可以满足业务逻辑要求的。

二:查询订单

业务逻辑:优先处理订单级别高,录入时间长的订单。

通过订单级别和订单录入时间排序

最基础的sql语句

mysql> explain select * from book_order order by order_state,create_time;
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+----------------+
| id | select_type | table               | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra          |
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+----------------+
|  1 | SIMPLE      | book_order | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |      100 | Using filesort |
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+----------------+

首先,采用全表扫描就不合理,还使用了文件排序Using filesort,更加拖慢了性能。 MySQL在4.1版本之前文件排序是采用双路排序的算法,由于两次扫描磁盘,I/O耗时太长。后优化成单路排序算法。其本质就是用空间换时间,但如果数据量太大,buffer的空间不足,会导致多次I/O的情况。其效果反而更差。与其找运维同事修改MySQL配置,还不如自己乖乖地建索引。

初步优化:为order_state,create_time 创建复合索引

mysql> create index idx_order_stateDate on book_order (order_state,create_time);
mysql> explain select * from book_order order by order_state,create_time;
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+----------------+
| id | select_type | table               | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra          |
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+----------------+
|  1 | SIMPLE      | book_order | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    3 |      100 | Using filesort |
+----+-------------+---------------------+------------+------+---------------+------+---------+------+------+----------+----------------+

创建复合索引后你会惊奇的发现,和没创建索引一样???都是全表扫描,都用到了文件排序。是索引失效?还是索引创建失败?

我们试着看看下面打印情况

mysql> explain select order_state,create_time from book_order order by order_state,create_time;
+----+-------------+---------------------+------------+-------+---------------+---------------------+---------+------+------+----------+-------------+
| id | select_type | table               | partitions | type  | possible_keys | key                 | key_len | ref  | rows | filtered | Extra       |
+----+-------------+---------------------+------------+-------+---------------+---------------------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | book_order | NULL       | index | NULL          | idx_order_stateDate | 68      | NULL |    3 |      100 | Using index |
+----+-------------+---------------------+------------+-------+---------------+---------------------+---------+------+------+----------+-------------+

将select * from 换成了 select order_state,create_time from 后。type从all升级为index,表示(full index scan)全索引文件扫描,Extra也显示使用了覆盖索引。可是不对啊!!!!检索虽然快了,但返回的内容只有order_state和create_time 两个字段,让业务同事怎么用?难道把每个字段都建一个复合索引?

MySQL没有这么笨,可以使用force index 强制指定索引。在原来的sql语句上修改 force index(idx_order_stateDate) 即可。

mysql> explain select * from book_order force index(idx_order_stateDate) order by order_state,create_time;
+----+-------------+---------------------+------------+-------+---------------+---------------------+---------+------+------+----------+-------+
| id | select_type | table               | partitions | type  | possible_keys | key                 | key_len | ref  | rows | filtered | Extra |
+----+-------------+---------------------+------------+-------+---------------+---------------------+---------+------+------+----------+-------+
|  1 | SIMPLE      | book_order | NULL       | index | NULL          | idx_order_stateDate | 68      | NULL |    3 |      100 | NULL  |
+----+-------------+---------------------+------------+-------+---------------+---------------------+---------+------+------+----------+-------+

再次优化:订单级别真的要排序么?

对于这种重复且分布平均的字段,排序和加索引的作用不大。

我们能否先固定 order_state 的值,然后再给 create_time 排序?

mysql> explain select * from book_order where order_state=3 order by create_time;
+----+-------------+---------------------+------------+------+---------------------+---------------------+---------+-------+------+----------+-----------------------+
| id | select_type | table               | partitions | type | possible_keys       | key                 | key_len | ref   | rows | filtered | Extra                 |
+----+-------------+---------------------+------------+------+---------------------+---------------------+---------+-------+------+----------+-----------------------+
|  1 | SIMPLE      | book_order | NULL       | ref  | idx_order_stateDate | idx_order_stateDate | 5       | const |    1 |      100 | Using index condition |
+----+-------------+---------------------+------------+------+---------------------+---------------------+---------+-------+------+----------+-----------------------+

和之前的sql比起来,type从index 升级为 ref(非唯一性索引扫描)。索引的长度从68变成了5,说明只用了一个索引。ref也是一个常量。Extra 为Using index condition 表示自动根据临界值,选择索引扫描还是全表扫描。总的来说性能远胜于之前的sql。

小结

建索引:

    1. 主键,唯一索引
    1. 经常用作查询条件的字段需要创建索引
    1. 经常需要排序、分组和统计的字段需要建立索引
    1. 查询中与其他表关联的字段,外键关系建立索引

不要建索引:

    1. 百万级以下的数据不需要创建索引
    1. 经常增删改的表不需要创建索引
    1. 数据重复且分布平均的字段不需要创建索引
    1. 频发更新的字段不适合创建索引
    1. where条件里用不到的字段不需要创建索引

talk is esay , show me the code

posted on 2018-06-25 20:26  一只阿木木  阅读(343)  评论(0编辑  收藏  举报