16. 窗口函数 (Window Function) 的使用

从SQL Server 2005起，SQL Server开始支持窗口函数 (Window Function)，以及到SQL Server 2012，窗口函数功能增强，目前为止支持以下几种窗口函数：

1. 排序函数 (Ranking Function) ；

2. 聚合函数 (Aggregate Function) ；

3. 分析函数 (Analytic Function) ；

4. NEXT VALUE FOR Function， 这是给sequence专用的一个函数；

 

一. 排序函数(Ranking Function)

帮助文档里的代码示例很全。

排序函数中，ROW_NUMBER()较为常用，可用于去重、分页、分组中选择数据，生成数字辅助表等等；

排序函数在语法上要求OVER子句里必须含ORDER BY，否则语法不通过，对于不想排序的场景可以这样变通；

drop table if exists test_ranking

create table test_ranking
( 
id int not null,
name varchar(20) not null,
value int not null
) 

insert test_ranking 
select 1,'name1',1 union all 
select 1,'name2',2 union all 
select 2,'name3',2 union all 
select 3,'name4',2

select id , name, ROW_NUMBER() over (PARTITION by id ORDER BY name) as num
from test_ranking

select id , name, ROW_NUMBER() over (PARTITION by id) as num
from test_ranking
/*
Msg 4112, Level 15, State 1, Line 1
The function 'ROW_NUMBER' must have an OVER clause with ORDER BY.
*/

--ORDERY BY后面给一个和原表无关的派生列
select id , name, ROW_NUMBER() over (PARTITION by id ORDER BY GETDATE()) as num
from test_ranking

select id , name, ROW_NUMBER() over (PARTITION by id ORDER BY (select 0)) as num
from test_ranking

 

二. 聚合函数 (Aggregate Function)

SQL Server 2005中，窗口聚合函数仅支持PARTITION BY，也就是说仅能对分组的数据整体做聚合运算；

SQL Server 2012开始，窗口聚合函数支持ORDER BY，以及ROWS/RAGNE选项，原本需要子查询来实现的需求，如： 移动平均 (moving averages), 总计聚合 (cumulative aggregates), 累计求和 (running totals) 等，变得更加方便；

 

代码示例1：总计/小计/累计求和

drop table if exists test_aggregate;

create table test_aggregate
(
event_id      varchar(100),
rk            int,
price         int
)

insert into test_aggregate
values
('a',1,10),
('a',2,10),
('a',3,50),
('b',1,10),
('b',2,20),
('b',3,30)


--1. 没有窗口函数时，用子查询
select a.event_id, 
       a.rk,  --build ranking column if needed
       a.price, 
　　   (select sum(price) from test_aggregate b where b.event_id = a.event_id and b.rk <= a.rk) as totalprice 
  from test_aggregate a


--2. 从SQL Server 2012起，用窗口函数
--2.1 
--没有PARTITION BY, 没有ORDER BY，为全部总计；
--只有PARTITION BY, 没有ORDER BY，为分组小计；
--只有ORDER BY，没有PARTITION BY，为全部累计求和(RANGE选项，见2.2)
select *,
　　   sum(price) over() as TotalPrice,
　　   sum(price) over(partition by event_id) as SubTotalPrice,
       sum(price) over(order by rk) as RunningTotalPrice
  from test_aggregate a

--2.2 注意ORDER BY列的选择，可能会带来不同结果
select *,
　　   sum(price) over(partition by event_id order by rk) as totalprice 
  from test_aggregate a
/*
event_id    rk    price    totalprice
a    1    10    10
a    2    10    20
a    3    50    70
b    1    10    10
b    2    20    30
b    3    30    60
*/

select *,
　　   sum(price) over(partition by event_id order by price) as totalprice 
  from test_aggregate a
/*
event_id    rk    price    totalprice
a    1    10    20
a    2    10    20
a    3    50    70
b    1    10    10
b    2    20    30
b    3    30    60
*/

--因为ORDER BY还有个子选项ROWS/RANGE，不指定的情况下默认为RANGE UNBOUNDED PRECEDING AND CURRENT ROW 
--RANGE按照ORDER BY中的列值，将相同的值的行均视为当前同一行
select  *,sum(price) over(partition by event_id order by price) as totalprice from test_aggregate a
select  *,sum(price) over(partition by event_id order by price range between unbounded preceding and current row) as totalprice from test_aggregate a

--如果ORDER BY中的列值有重复值，手动改用ROWS选项即可实现逐行累计求和
select  *,sum(price) over(partition by event_id order by price rows between unbounded preceding and current row) as totalprice from test_aggregate a

 

代码示例2：移动平均

--移动平均,举个例子，就是求前N天的平均值，和股票市场的均线类似
drop table if exists test_moving_avg

create table test_moving_avg
(
ID    int, 
Value int,
DT    datetime
)

insert into test_moving_avg 
values
(1,10,GETDATE()-10),
(2,110,GETDATE()-9),
(3,100,GETDATE()-8),
(4,80,GETDATE()-7),
(5,60,GETDATE()-6),
(6,40,GETDATE()-5),
(7,30,GETDATE()-4),
(8,50,GETDATE()-3),
(9,20,GETDATE()-2),
(10,10,GETDATE()-1)

--1. 没有窗口函数时，用子查询
select *,
(select AVG(Value) from test_moving_avg a where a.DT >= DATEADD(DAY, -5, b.DT) AND a.DT < b.DT) AS avg_value_5days
from test_moving_avg b

--2. 从SQL Server 2012起，用窗口函数
--三个内置常量，第一行，最后一行，当前行：UNBOUNDED PRECEDING, UNBOUNDED FOLLOWING, CURRENT ROW 
--在行间移动，用BETWEEN m preceding AND n following (m, n > 0)
SELECT *,
       sum(value) over (ORDER BY DT ROWS BETWEEN 5 preceding AND CURRENT ROW) moving_sum,
       avg(value) over (ORDER BY DT ROWS BETWEEN 4 preceding AND CURRENT ROW) moving_avg1,
       avg(value) over (ORDER BY DT ROWS BETWEEN 5 preceding AND 1 preceding) moving_avg2,
       avg(value) over (ORDER BY DT ROWS BETWEEN 3 preceding AND 1 following) moving_avg3
FROM  test_moving_avg
ORDER BY DT

 

三. 分析函数 (Analytic Function)

代码示例1：取当前行某列的前一个/下一个值

drop table if exists test_analytic

create table test_analytic
(
SalesYear         varchar(10),
Revenue           int,
Offset            int
)

insert into test_analytic
values
(2013,1001,1),
(2014,1002,1),
(2015,1003,1),
(2016,1004,1),
(2017,1005,1),
(2018,1006,1)

--当年及去年的销售额
select *,lag(Revenue,1,null) over(order by SalesYear asc) as PreviousYearRevenue from test_analytic
select *,lag(Revenue,Offset,null) over(order by SalesYear asc) as PreviousYearRevenue from test_analytic
select *,lead(Revenue,1,null) over(order by SalesYear desc) as PreviousYearRevenue from test_analytic

--当年及下一年的销售额
select *,lead(Revenue,1,null) over(order by SalesYear asc) as NextYearRevenue from test_analytic
select *,lead(Revenue,Offset,null) over(order by SalesYear asc) as NextYearRevenue from test_analytic
select *,lag(Revenue,1,null) over(order by SalesYear desc) as NextYearRevenue from test_analytic

--可以根据offset调整跨度

 

代码示例2：分组中某列最大/最小值，对应的其他列值

假设有个门禁系统，在员工每次进门时写入一条记录，记录了“身份号码”，“进门时间”，“衣服颜色"，查询每个员工最后一次进门时的“衣服颜色”。

drop table if exists test_first_last

create table test_first_last
(
EmployeeID             int,
EnterTime              datetime,
ColorOfClothes         varchar(20)
)

insert into test_first_last
values
(1001, GETDATE()-9, 'GREEN'),
(1001, GETDATE()-8, 'RED'),
(1001, GETDATE()-7, 'YELLOW'),
(1001, GETDATE()-6, 'BLUE'),
(1002, GETDATE()-5, 'BLACK'),
(1002, GETDATE()-4, 'WHITE')

--1. 用子查询
--LastColorOfColthes
select * from test_first_last a
where not exists(select 1 from test_first_last b where a.EmployeeID = b.EmployeeID and a.EnterTime < b.EnterTime)

--LastColorOfColthes
select *
from 
(select *, ROW_NUMBER() over(partition by EmployeeID order by EnterTime DESC) num
from test_first_last ) t
where t.num =1


--2. 用窗口函数
--用LAST_VALUE时，必须加上ROWS/RANGE BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING，否则结果不正确
select *, 
       FIRST_VALUE(ColorOfClothes) OVER (PARTITION BY EmployeeID ORDER BY EnterTime DESC) as LastColorOfClothes,
       FIRST_VALUE(ColorOfClothes) OVER (PARTITION BY EmployeeID ORDER BY EnterTime ASC) as FirstColorOfClothes,
       LAST_VALUE(ColorOfClothes) OVER (PARTITION BY EmployeeID ORDER BY EnterTime ASC ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) as LastColorOfClothes,
       LAST_VALUE(ColorOfClothes) OVER (PARTITION BY EmployeeID ORDER BY EnterTime DESC ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) as FirstColorOfClothes
from test_first_last

--对于显示表中所有行，并追加Last/First字段时用窗口函数方便些
--对于挑选表中某一行/多行时，用子查询更方便

 

四. NEXT VALUE FOR Function

drop sequence if exists test_seq

create sequence test_seq
start with 1
increment by 1;

GO

drop table if exists test_next_value

create table test_next_value
(
ID         int,
Name       varchar(10)
)

insert into test_next_value(Name)
values
('AAA'),
('AAA'),
('BBB'),
('CCC')

--对于多行数据获取sequence的next value，是否使用窗口函数都会逐行计数
--窗口函数中ORDER BY用于控制不同列值的计数顺序
select *, NEXT VALUE FOR test_seq from test_next_value
select *, NEXT VALUE FOR test_seq OVER(ORDER BY Name DESC) from test_next_value

 

参考：

SELECT - OVER Clause (Transact-SQL)

https://docs.microsoft.com/en-us/sql/t-sql/queries/select-over-clause-transact-sql?view=sql-server-2017

SQL Server Windowing Functions: ROWS vs. RANGE

https://www.sqlpassion.at/archive/2015/01/22/sql-server-windowing-functions-rows-vs-range/