作者：杨涛涛
一直以来，MySQL 只有针对聚合函数的汇总类功能，比如MAX, AVG 等，没有从 SQL 层针对聚合类每组展开处理的功能。不过 MySQL 开放了 UDF 接口，可以用 C 来自己写UDF，这个就增加了功能行难度。
这种针对每组展开处理的功能就叫窗口函数，有的数据库叫分析函数。
在 MySQL 8.0 之前，我们想要得到这样的结果，就得用以下几种方法来实现：
1. session 变量
2. group_concat 函数组合
3. 自己写 store routines
接下来我们用经典的 学生/课程/成绩 来做窗口函数演示
学生表
mysql> show create table student \G *************************** 1. row *************************** Table: student Create Table: CREATE TABLE student ( sid int(10) unsigned NOT NULL, sname varchar(64) DEFAULT NULL, PRIMARY KEY (sid) ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci 1 row in set (0.00 sec)课程表
mysql> show create table course\G *************************** 1. row *************************** Table: course Create Table: CREATE TABLE `course` ( `cid` int(10) unsigned NOT NULL, `cname` varchar(64) DEFAULT NULL, PRIMARY KEY (`cid`) ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci 1 row in set (0.00 sec)成绩表
mysql> show create table score\G *************************** 1. row *************************** Table: score Create Table: CREATE TABLE `score` ( `sid` int(10) unsigned NOT NULL, `cid` int(10) unsigned NOT NULL, `score` tinyint(3) unsigned DEFAULT NULL, PRIMARY KEY (`sid`,`cid`) ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci 1 row in set (0.00 sec)测试数据
mysql> select * from student; +-----------+--------------+ | sid | sname | +-----------+--------------+ | 201910001 | 张三 | | 201910002 | 李四 | | 201910003 | 武松 | | 201910004 | 潘金莲 | | 201910005 | 菠菜 | | 201910006 | 杨发财 | | 201910007 | 欧阳修 | | 201910008 | 郭靖 | | 201910009 | 黄蓉 | | 201910010 | 东方不败 | +-----------+--------------+ 10 rows in set (0.00 sec) mysql> select * from score;; +-----------+----------+-------+ | sid | cid | score | +-----------+----------+-------+ | 201910001 | 20192001 | 50 | | 201910001 | 20192002 | 88 | | 201910001 | 20192003 | 54 | | 201910001 | 20192004 | 43 | | 201910001 | 20192005 | 89 | | 201910002 | 20192001 | 79 | | 201910002 | 20192002 | 97 | | 201910002 | 20192003 | 82 | | 201910002 | 20192004 | 85 | | 201910002 | 20192005 | 80 | | 201910003 | 20192001 | 48 | | 201910003 | 20192002 | 98 | | 201910003 | 20192003 | 47 | | 201910003 | 20192004 | 41 | | 201910003 | 20192005 | 34 | | 201910004 | 20192001 | 81 | | 201910004 | 20192002 | 69 | | 201910004 | 20192003 | 67 | | 201910004 | 20192004 | 99 | | 201910004 | 20192005 | 61 | | 201910005 | 20192001 | 40 | | 201910005 | 20192002 | 52 | | 201910005 | 20192003 | 39 | | 201910005 | 20192004 | 74 | | 201910005 | 20192005 | 86 | | 201910006 | 20192001 | 42 | | 201910006 | 20192002 | 52 | | 201910006 | 20192003 | 36 | | 201910006 | 20192004 | 58 | | 201910006 | 20192005 | 84 | | 201910007 | 20192001 | 79 | | 201910007 | 20192002 | 43 | | 201910007 | 20192003 | 79 | | 201910007 | 20192004 | 98 | | 201910007 | 20192005 | 88 | | 201910008 | 20192001 | 45 | | 201910008 | 20192002 | 65 | | 201910008 | 20192003 | 90 | | 201910008 | 20192004 | 89 | | 201910008 | 20192005 | 74 | | 201910009 | 20192001 | 73 | | 201910009 | 20192002 | 42 | | 201910009 | 20192003 | 95 | | 201910009 | 20192004 | 46 | | 201910009 | 20192005 | 45 | | 201910010 | 20192001 | 58 | | 201910010 | 20192002 | 52 | | 201910010 | 20192003 | 55 | | 201910010 | 20192004 | 87 | | 201910010 | 20192005 | 36 | +-----------+----------+-------+ 50 rows in set (0.00 sec) mysql> select * from course; +----------+------------+ | cid | cname | +----------+------------+ | 20192001 | mysql | | 20192002 | oracle | | 20192003 | postgresql | | 20192004 | mongodb | | 20192005 | dble | +----------+------------+ 5 rows in set (0.00 sec)比如我们求成绩排名前三的学生排名，我来举个用 session 变量和 group_concat 函数来分别实现的例子：
session 变量方式
每组开始赋一个初始值序号和初始分组字段。
SELECT b.cname, a.sname, c.score, c.ranking_score FROM student a, course b, ( SELECT c.*, IF( @cid = c.cid, @rn := @rn + 1, @rn := 1 ) AS ranking_score, @cid := c.cid AS tmpcid FROM ( SELECT * FROM score ORDER BY cid, score DESC ) c, ( SELECT @rn := 0 rn, @cid := '' ) initialize_table ) c WHERE a.sid = c.sid AND b.cid = c.cid AND c.ranking_score <= 3 ORDER BY b.cname,c.ranking_score; +------------+-----------+-------+---------------+ | cname | sname | score | ranking_score | +------------+-----------+-------+---------------+ | dble | 张三 | 89 | 1 | | dble | 欧阳修 | 88 | 2 | | dble | 菠菜 | 86 | 3 | | mongodb | 潘金莲 | 99 | 1 | | mongodb | 欧阳修 | 98 | 2 | | mongodb | 郭靖 | 89 | 3 | | mysql | 李四 | 100 | 1 | | mysql | 潘金莲 | 81 | 2 | | mysql | 欧阳修 | 79 | 3 | | oracle | 武松 | 98 | 1 | | oracle | 李四 | 97 | 2 | | oracle | 张三 | 88 | 3 | | postgresql | 黄蓉 | 95 | 1 | | postgresql | 郭靖 | 90 | 2 | | postgresql | 李四 | 82 | 3 | +------------+-----------+-------+---------------+ 15 rows in set, 5 warnings (0.01 sec)group_concat 函数方式
利用 findinset 内置函数来返回下标作为序号使用。
SELECT * FROM ( SELECT b.cname, a.sname, c.score, FIND_IN_SET(c.score, d.gp) score_ranking FROM student a, course b, score c, ( SELECT cid, GROUP_CONCAT( score ORDER BY score DESC SEPARATOR ',' ) gp FROM score GROUP BY cid ORDER BY score DESC ) d WHERE a.sid = c.sid AND b.cid = c.cid AND c.cid = d.cid ORDER BY d.cid, score_ranking ) ytt WHERE score_ranking <= 3； +------------+-----------+-------+---------------+ | cname | sname | score | score_ranking | +------------+-----------+-------+---------------+ | dble | 张三 | 89 | 1 | | dble | 欧阳修 | 88 | 2 | | dble | 菠菜 | 86 | 3 | | mongodb | 潘金莲 | 99 | 1 | | mongodb | 欧阳修 | 98 | 2 | | mongodb | 郭靖 | 89 | 3 | | mysql | 李四 | 100 | 1 | | mysql | 潘金莲 | 81 | 2 | | mysql | 欧阳修 | 79 | 3 | | oracle | 武松 | 98 | 1 | | oracle | 李四 | 97 | 2 | | oracle | 张三 | 88 | 3 | | postgresql | 黄蓉 | 95 | 1 | | postgresql | 郭靖 | 90 | 2 | | postgresql | 李四 | 82 | 3 | +------------+-----------+-------+---------------+ 15 rows in set (0.00 sec)
MySQL 8.0 后提供了原生的窗口函数支持，语法和大多数数据库一样，比如还是之前的例子：
用 row_number() over () 直接来检索排名。
mysql> SELECT * FROM ( SELECT b.cname, a.sname, c.score, row_number() over ( PARTITION BY b.cname ORDER BY c.score DESC ) score_rank FROM student AS a, course AS b, score AS c WHERE a.sid = c.sid AND b.cid = c.cid ) ytt WHERE score_rank <= 3; +------------+-----------+-------+------------+ | cname | sname | score | score_rank | +------------+-----------+-------+------------+ | dble | 张三 | 89 | 1 | | dble | 欧阳修 | 88 | 2 | | dble | 菠菜 | 86 | 3 | | mongodb | 潘金莲 | 99 | 1 | | mongodb | 欧阳修 | 98 | 2 | | mongodb | 郭靖 | 89 | 3 | | mysql | 李四 | 100 | 1 | | mysql | 潘金莲 | 81 | 2 | | mysql | 欧阳修 | 79 | 3 | | oracle | 武松 | 98 | 1 | | oracle | 李四 | 97 | 2 | | oracle | 张三 | 88 | 3 | | postgresql | 黄蓉 | 95 | 1 | | postgresql | 郭靖 | 90 | 2 | | postgresql | 李四 | 82 | 3 | +------------+-----------+-------+------------+ 15 rows in set (0.00 sec)那我们再找出课程 MySQL 和 DBLE 里不及格的倒数前两名学生名单。
mysql> SELECT * FROM ( SELECT b.cname, a.sname, c.score, row_number () over ( PARTITION BY b.cid ORDER BY c.score ASC ) score_ranking FROM student AS a, course AS b, score AS c WHERE a.sid = c.sid AND b.cid = c.cid AND b.cid IN (20192005, 20192001) AND c.score < 60 ) ytt WHERE score_ranking < 3; +-------+--------------+-------+---------------+ | cname | sname | score | score_ranking | +-------+--------------+-------+---------------+ | mysql | 菠菜 | 40 | 1 | | mysql | 杨发财 | 42 | 2 | | dble | 武松 | 34 | 1 | | dble | 东方不败 | 36 | 2 | +-------+--------------+-------+---------------+ 4 rows in set (0.00 sec)到此为止，我们只是演示了row_number() over() 函数的使用方法，其他的函数有兴趣的朋友可以自己体验体验，方法都差不多。
