??? 對于數(shù)據(jù)的匯總，是數(shù)據(jù)庫經(jīng)常用到的任務(wù)之一，除了我們通常使用的GROUP BY分組配合聚合函數(shù)對數(shù)據(jù)匯總，以及使用UNION ALL 對數(shù)據(jù)匯總之外，SQL還提供了 GROUP BY Col1,Col2.. WITH CUBE | ROLLUP，以及COMPUTE BY 等匯總方式，本文主要介紹了使用CUBE 與ROLLUP運(yùn)算符來實(shí)現(xiàn)數(shù)據(jù)的分級匯總。

IF OBJECT_ID('tb') IS NOT NULL
?DROP TABLE tb;
GO ?
CREATE TABLE tb
(
??? Provider??? VARCHAR(10)
??? ,MaterialNo VARCHAR(3)
??? ,Quantity INT
???
);?
INSERT INTO tb SELECT 'Canon',? '001',? 500?
UNION ALL SELECT 'Canon',? '001',???? 200?
UNION ALL SELECT 'Canon',? '002',???? 100
UNION ALL SELECT 'Canon',? '002',???? 300
UNION ALL SELECT 'Sony',? '001',???? 200?
UNION ALL SELECT 'IBM',? '002',???? 100?
UNION ALL SELECT 'IBM',? '001',???? 600?
UNION ALL SELECT 'IBM',? '001',???? 200??

?

--1.使用ROLLUP來對數(shù)據(jù)進(jìn)行匯總
SELECT Provider,MaterialNo,SUM(Quantity) AS Sum_Quantity FROM tb
GROUP BY Provider,MaterialNo
WITH ROLLUP
/*? --結(jié)果
Provider?? MaterialNo Sum_Quantity
---------- ---------- ------------
Canon????? 001??????? 700
Canon????? 002??????? 400
Canon????? NULL?????? 1100
IBM??????? 001??????? 800
IBM??????? 002??????? 100
IBM??????? NULL?????? 900
Sony?????? 001??????? 200
Sony?????? NULL?????? 200
NULL?????? NULL?????? 2200
*/
由結(jié)果可以得知，ROLLUP運(yùn)算符實(shí)現(xiàn)了對Provider級別進(jìn)行匯總，結(jié)果集中Provider字段不為空，MaterialNo字段為空的時候?qū)崿F(xiàn)了對不同的Provider分類匯總，而當(dāng)Provider與MaterialNo都為空時則是對所有的Provider來實(shí)現(xiàn)數(shù)據(jù)匯總。

?

--2.使用CUBE來對數(shù)據(jù)進(jìn)行匯總
SELECT Provider,MaterialNo,SUM(Quantity) AS Sum_Quantity FROM tb
GROUP BY Provider,MaterialNo
WITH CUBE
/*? --結(jié)果
Provider?? MaterialNo Sum_Quantity
---------- ---------- ------------
Canon????? 001??????? 700
Canon????? 002??????? 400
Canon????? NULL?????? 1100
IBM??????? 001??????? 800
IBM??????? 002??????? 100
IBM??????? NULL?????? 900
Sony?????? 001??????? 200
Sony?????? NULL?????? 200
NULL?????? NULL?????? 2200
NULL?????? 001??????? 1700
NULL?????? 002??????? 500
*/
與ROLLUP相比，從結(jié)果可以看出CUBE的結(jié)果集在ROLLUP結(jié)果集的基礎(chǔ)上增加了兩行，
NULL?????? 001??????? 1700
NULL?????? 002??????? 200
即對不同MaterialNo也實(shí)現(xiàn)了匯總。

?

--3.ROLLUP與CUBE的差異
?. ROLLUP 生成的結(jié)果集為所選列中值的某一維度的聚合。如以上示例中實(shí)現(xiàn)了對Provider維度進(jìn)行匯總。
?. CUBE? 生成的結(jié)果集為所選列中值的所有維度的聚合。如以上示例中實(shí)現(xiàn)了對Provider和MaterialNo所有維度進(jìn)行匯總。
?
--4.使用GROUPING函數(shù)來處理匯總產(chǎn)生的NULL值
???? 對于使用ROLLUP與CUBE匯總數(shù)據(jù)所產(chǎn)生的NULL值，容易引起與實(shí)際數(shù)據(jù)本身為NULL容易引起歧義，對此我們可以使用GROUPING函數(shù)的進(jìn)行區(qū)分。
???? 當(dāng)NULL為ROLLUP或CUBE所產(chǎn)生時，則GROUPING函數(shù)返回的值為1，當(dāng)NULL來自實(shí)際數(shù)據(jù)本身的話，GROUPING函數(shù)返回的值為0。
SELECT
??? CASE WHEN (GROUPING(Provider) = 1) THEN 'ALL'
???????? ELSE Provider END AS Provider,
??? CASE WHEN (GROUPING(MaterialNo) = 1) THEN 'ALL'
???????? ELSE MaterialNo END AS MaterialNo,
??? SUM(Quantity) AS Sum_Quantity FROM tb
GROUP BY Provider,MaterialNo
WITH CUBE?
/*? --結(jié)果
Provider?? MaterialNo Sum_Quantity
---------- ---------- ------------
Canon????? 001??????? 700
Canon????? 002??????? 400
Canon????? ALL??????? 1100
IBM??????? 001??????? 800
IBM??????? 002??????? 100
IBM??????? ALL??????? 900
Sony?????? 001??????? 200
Sony?????? ALL??????? 200
ALL??????? ALL??????? 2200
ALL??????? 001??????? 1700
ALL??????? 002??????? 500
*/

?

--5.使用Having子句來顧慮掉無用的數(shù)據(jù)，比如本例過濾掉了總計(jì)以及MaterialNo匯總的數(shù)據(jù)。
SELECT
??? CASE WHEN (GROUPING(Provider) = 1) THEN 'ALL'
???????? ELSE Provider END AS Provider,
??? CASE WHEN (GROUPING(MaterialNo) = 1) THEN 'ALL'
???????? ELSE MaterialNo END AS MaterialNo,
??? SUM(Quantity) AS Sum_Quantity FROM tb
GROUP BY Provider,MaterialNo WITH CUBE
HAVING GROUPING(Provider) <> 1
/*? --結(jié)果
Provider?? MaterialNo Sum_Quantity
---------- ---------- ------------
Canon????? 001??????? 700
Canon????? 002??????? 400
Canon????? ALL??????? 1100
IBM??????? 001??????? 800
IBM??????? 002??????? 100
IBM??????? ALL??????? 900
Sony?????? 001??????? 200
Sony?????? ALL??????? 200
*/

?

--過濾掉了相同Provider，不同的MaterialNo 的小計(jì)數(shù)據(jù)和最終的總計(jì)數(shù)據(jù)
SELECT
??? CASE WHEN (GROUPING(Provider) = 1) THEN 'ALL'
???????? ELSE Provider END AS Provider,
??? CASE WHEN (GROUPING(MaterialNo) = 1) THEN 'ALL'
???????? ELSE MaterialNo END AS MaterialNo,
??? SUM(Quantity) AS Sum_Quantity FROM tb
GROUP BY Provider,MaterialNo WITH CUBE
HAVING GROUPING(MaterialNo) = 0

?

總結(jié):CUBE，ROLLUP為多維數(shù)據(jù)集的匯總提供了可能，當(dāng)需要對所有維度進(jìn)行匯總，應(yīng)當(dāng)使用CUBE運(yùn)算符，對某一維度進(jìn)行匯總則使用ROLLUP運(yùn)算法。
???? 需要注意的是，WITH CUBE | ROLLUP必須跟在GROUP BY Col1,Col2列之后，然后可以通過使用HAVING子句配合GROUPING函數(shù)來過濾不需要的結(jié)果集。

ROLLUP 與 CUBE 運(yùn)算符的使用