全索引扫描和覆盖索引范围扫描

今天在review 一个SQL的时候，发现即使在列前面有函数操作，查询也能够使用到索引，如下：（OLD）

1 mysql> desc SELECT sql_no_cache COUNT(1) FROM Appointment WHERE YEAR(ScheduledDate)=YEAR(NOW()) AND MONTH(ScheduledDate)=MONTH(NOW());
+----+-------------+-------------+-------+---------------+---------------+---------+------+---------+--------------------------+
2 | id | select_type | table       | type  | possible_keys | key           | key_len | ref  | rows    | Extra                    |
3 +----+-------------+-------------+-------+---------------+---------------+---------+------+---------+--------------------------+
4 |  1 | SIMPLE      | Appointment | index | NULL          | ScheduledDate | 9       | NULL | 1812978 | Using where; Using index |
5 +----+-------------+-------------+-------+---------------+---------------+---------+------+---------+--------------------------+
6 1 row in set (0.00 sec)

用的是全索引扫描(type=index)，用到了索引(key=ScheduledDate),并且用的是覆盖索引(Using index).看起来这个SQL没有问题，唯一不足之处在1)type列，INDEX是效率很差的连接类型，尽优于全表扫描(full). 2)扫描的行数太多(row=1812978).

按照最常见的优化方法改写SQL后的执行计划如下：（NEW）

1 mysql> desc SELECT sql_no_cache COUNT(1) FROM Appointment WHERE  ScheduledDate > DATE_FORMAT(NOW(),'%Y-%m-01 00:00:00') AND ScheduledDate < DATE_FORMAT(DATE_ADD(NOW(),INTERVAL 1 MONTH),'%Y-%m-01 00:00:01');
2 +----+-------------+-------------+-------+---------------+---------------+---------+------+--------+--------------------------+
3 | id | select_type | table       | type  | possible_keys | key           | key_len | ref  | rows   | Extra                    |
4 +----+-------------+-------------+-------+---------------+---------------+---------+------+--------+--------------------------+
5 |  1 | SIMPLE      | Appointment | range | ScheduledDate | ScheduledDate | 9       | NULL | 727166 | Using where; Using index |
6 +----+-------------+-------------+-------+---------------+---------------+---------+------+--------+--------------------------+
7 1 row in set (0.00 sec)

连接方式变成了范围查找(type=range),扫描的行数也减少了很多(rows=727166) 。

2个SQL的实际执行效率，在结果集很大的时候，还是存在比较大的差别

	1	2	3	4	5
OLD	2.75	2.51	2.35	2.56	2.77
NEW	0.49	0.65	0.50	0.54	0.60

 

 

             比较两个查询的平均执行时间OLD=(2.75+2.51+2.35+2.56+2.77)/5=2.588，NEW=0.556，效率相差 2.588/0.556=4.65.

分析一下1）为什么使用到函数了，还会用到索引？ 2）同样是使用了覆盖索引，为什么会出现严重的性能偏差？

 

这里举的例子有特殊性，select count(1),获取全部满足条件的总记录数，只需要扫描符合条件的索引页，就能够得到正确的结果集，不需要回表取数据。这是问题1的解释。在OLD里面，需要扫描了全部的索引页，然后对里面的数据进行计算比较。在NEW里面，只需要扫描少数的索引页，就能够得到结果集。这些差别在row里面体现了出来。这是问题2的解释。

在非覆盖索引下，一般的情况查询应该是这样的：

1 mysql> desc SELECT sql_no_cache * FROM Appointment WHERE YEAR(ScheduledDate)=YEAR(NOW()) AND MONTH(ScheduledDate)=MONTH(NOW());
2 +----+-------------+-------------+------+---------------+------+---------+------+---------+-------------+
3 | id | select_type | table       | type | possible_keys | key  | key_len | ref  | rows    | Extra       |
4 +----+-------------+-------------+------+---------------+------+---------+------+---------+-------------+
5 |  1 | SIMPLE      | Appointment | ALL  | NULL          | NULL | NULL    | NULL | 1541817 | Using where |
6 +----+-------------+-------------+------+---------------+------+---------+------+---------+-------------+
7 1 row in set (0.00 sec)

此时只通过索引是无法检索出结果集，所以使用了全表扫描。
优化后的执行计划是这样的：

1 mysql> desc SELECT sql_no_cache * FROM Appointment WHERE  ScheduledDate > DATE_FORMAT(NOW(),'%Y-%m-01 00:00:00') AND ScheduledDate < DATE_FORMAT(DATE_ADD(NOW(),INTERVAL 1 MONTH),'%Y-%m-01 00:00:01');
2 +----+-------------+-------------+------+---------------+------+---------+------+---------+-------------+
3 | id | select_type | table       | type | possible_keys | key  | key_len | ref  | rows    | Extra       |
4 +----+-------------+-------------+------+---------------+------+---------+------+---------+-------------+
5 |  1 | SIMPLE      | Appointment | ALL  | ScheduledDate | NULL | NULL    | NULL | 1541817 | Using where |
6 +----+-------------+-------------+------+---------------+------+---------+------+---------+-------------+
7 1 row in set (0.00 sec)

这里还是使用了全表扫描。MYSQL中查询返回的结果行占表总行数超过15%（测试下来是15%左右是优化器走索引还是全部扫描的分割线，手册里写的是30%）时，就会选择全表扫描。在这个案例里，上述条件返回337711，总记录数1698272，337711/1698272=0.1989，超过门限值，所以用了全表扫描。缩小结果集

 1 mysql> SELECT sql_no_cache count(*) FROM Appointment WHERE  ScheduledDate > DATE_FORMAT(NOW(),'%Y-%m-%d 00:00:00') AND ScheduledDate < DATE_FORMAT(DATE_ADD(NOW(),INTERVAL 1 day),'%Y-%m-%d 00:00:01');
 2 +----------+
 3 | count(*) |
 4 +----------+
 5 |    10169 |
 6 +----------+
 7 1 row in set (0.01 sec)
 8  
 9 mysql> desc SELECT sql_no_cache * FROM Appointment WHERE  ScheduledDate > DATE_FORMAT(NOW(),'%Y-%m-%d 00:00:00') AND ScheduledDate < DATE_FORMAT(DATE_ADD(NOW(),INTERVAL 1 day),'%Y-%m-%d 00:00:01');
10 +----+-------------+-------------+-------+---------------+---------------+---------+------+-------+-------------+
11 | id | select_type | table       | type  | possible_keys | key           | key_len | ref  | rows  | Extra       |
12 +----+-------------+-------------+-------+---------------+---------------+---------+------+-------+-------------+
13 |  1 | SIMPLE      | Appointment | range | ScheduledDate | ScheduledDate | 9       | NULL | 20924 | Using where |
14 +----+-------------+-------------+-------+---------------+---------------+---------+------+-------+-------------+
15 1 row in set (0.00 sec)

这里执行计划和预期相同。

 

 

 

转载于:https://www.cnblogs.com/zuoxingyu/archive/2012/09/27/2706020.html