第一点， 比基于这个列的全部索引占用空间来的小。
第二点， 特别是基于FULL INDEX SCAN 的时候，占用空间小的索引对内存占用也小很多。
PostgreSQL,SqlServer等都支持条件索引，所以我们先来看下条件索引的实际情况。
表结构如下，记录大概有10W行：
Table "ytt.girl1"
Column | Type | Modifiers
--------+---------+--------------------
id | integer | not null
rank | integer | not null default 0
Indexes:
"girl1_pkey" PRIMARY KEY, btree (id)
"idx_girl1_rank" btree (rank) WHERE rank >= 10 AND rank <= 100
执行的查询语句为：
select * from girl1 where rank between 20 and 60 limit 20;
用了全部索引的查询计划：
QUERY PLAN
---------------------------------------------------------------------------------------------------------------------------------
Limit (cost=0.29..36.58 rows=20 width=8) (actual time=0.024..0.054 rows=20 loops=1)
-> Index Scan using idx_girl1_rank on girl1 (cost=0.29..421.26 rows=232 width=8) (actual time=0.023..0.044 rows=20 loops=1)
Index Cond: ((rank >= 20) AND (rank <= 60))
Total runtime: 0.087 ms
(4 rows)
Time: 1.881 ms
用了条件索引的查询计划：
QUERY PLAN
---------------------------------------------------------------------------------------------------------------------------------
Limit (cost=0.28..35.54 rows=20 width=8) (actual time=0.036..0.068 rows=20 loops=1)
-> Index Scan using idx_girl1_rank on girl1 (cost=0.28..513.44 rows=291 width=8) (actual time=0.033..0.061 rows=20 loops=1)
Index Cond: ((rank >= 20) AND (rank <= 60))
Total runtime: 0.106 ms
(4 rows)
Time: 0.846 ms
可以看出，在扫描的记录数以及时间上，条件索引的优势都很明显。
接下来，我们在MySQL 模拟下这样的过程。
由于MySQL 不支持这样的索引， 在SQL层面上，只能创建一个索引表来保存对应条件的主键以及索引键。
ytt>show create table girl1_filtered_index;
+----------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| Table | Create Table |
+----------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| girl1_filtered_index | CREATE TABLE `girl1_filtered_index` (
`id` int(11) NOT NULL,
`rank` int(11) NOT NULL DEFAULT ‘0‘,
PRIMARY KEY (`id`),
KEY `idx_rank` (`rank`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1 |
+----------------------+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1 row in set (0.00 sec)
接下来，对基础表的更新操作做下修改，创建了三个触发器。
DELIMITER $$
USE `t_girl`$$
DROP TRIGGER /*!50032 IF EXISTS */ `filtered_insert`$$
CREATE
/*!50017 DEFINER = ‘root‘@‘localhost‘ */
TRIGGER `filtered_insert` AFTER INSERT ON `girl1`
FOR EACH ROW BEGIN
IF new.rank BETWEEN 10 AND 100 THEN
INSERT INTO girl1_filtered_index VALUES (new.id,new.rank);
END IF;
END;
$$
DELIMITER ;
DELIMITER $$
USE `t_girl`$$
DROP TRIGGER /*!50032 IF EXISTS */ `filtered_update`$$
CREATE
/*!50017 DEFINER = ‘root‘@‘localhost‘ */
TRIGGER `filtered_update` AFTER UPDATE ON `girl1`
FOR EACH ROW BEGIN
IF new.rank BETWEEN 10 AND 100 THEN
REPLACE girl1_filtered_index VALUES (new.id,new.rank);
ELSE
DELETE FROM girl1_filtered_index WHERE id = old.id;
END IF;
END;
$$
DELIMITER ;
DELIMITER $$
USE `t_girl`$$
DROP TRIGGER /*!50032 IF EXISTS */ `filtered_delete`$$
CREATE
/*!50017 DEFINER = ‘root‘@‘localhost‘ */
TRIGGER `filtered_delete` AFTER DELETE ON `girl1`
FOR EACH ROW BEGIN
DELETE FROM girl1_filtered_index WHERE id = old.id;
END;
$$
DELIMITER ;
OK，我们导入测试数据。
ytt>load data infile ‘girl1.txt‘ into table girl1 fields terminated by ‘,‘;
Query OK, 100000 rows affected (1.05 sec)
Records: 100000 Deleted: 0 Skipped: 0 Warnings: 0
ytt>select count(*) from girl1;
+----------+
| count(*) |
+----------+
| 100000 |
+----------+
1 row in set (0.04 sec)
ytt>select count(*) from girl1_filtered_index;
+----------+
| count(*) |
+----------+
| 640 |
+----------+
1 row in set (0.00 sec)
这里，我们把查询语句修改成基础表和条件索引表的JOIN。
select a.id,a.rank from girl1 as a where a.id in (select b.id from girl1_filtered_index as b where b.rank between 20 and 60) limit 20;
当然这只是功能上的一个演示。 最终实现得靠MySQL 5.8了。^____^