Oracle group by 基本及的拓展 ROLLUP, CUBE, GROUPING 功能and GROUPING 集合
/*
声明:本人不是专门学数据库的,也不是专门的翻译,只是因为碰到一个问题(SQL CookBook中)找了一下,发现一个英文网站的解释很清晰,特此翻译过来,mark.不喜勿砖,谢谢!
原文链接:
ROLLUP, CUBE, GROUPING Functions and GROUPING SETS
*/
环境:
DROP TABLE dimension_tab;
CREATE TABLE dimension_tab (
fact_1_id NUMBER NOT NULL,
fact_2_id NUMBER NOT NULL,
fact_3_id NUMBER NOT NULL,
fact_4_id NUMBER NOT NULL,
sales_value NUMBER(10,2) NOT NULL
);
INSERT INTO dimension_tab
SELECT TRUNC(DBMS_RANDOM.value(low => 1, high => 3)) AS fact_1_id,
TRUNC(DBMS_RANDOM.value(low => 1, high => 6)) AS fact_2_id,
TRUNC(DBMS_RANDOM.value(low => 1, high => 11)) AS fact_3_id,
TRUNC(DBMS_RANDOM.value(low => 1, high => 11)) AS fact_4_id,
ROUND(DBMS_RANDOM.value(low => 1, high => 100), 2) AS sales_value
FROM dual
CONNECT BY level <= 1000;
COMMIT;
1.Group by 基本用法
为了方便理解,先用一个聚合函数,求和.(未使用group by)
SELECT SUM(sales_value) AS sales_value
FROM dimension_tab;
SALES_VALUE
-----------
50528.39
1 row selected.
SQL>
SELECT SUM(sales_value) AS sales_value
FROM dimension_tab;
SALES_VALUE
-----------
50528.39
1 row selected.
SQL>
将想要分组的列放在group by 之后.结果的行数是我们目标列中包含不同值的个数.
SELECT fact_1_id,
COUNT(*) AS num_rows,
SUM(sales_value) AS sales_value
FROM dimension_tab
GROUP BY fact_1_id
ORDER BY fact_1_id;
FACT_1_ID NUM_ROWS SALES_VALUE
---------- ---------- -----------
1 478 24291.35
2 522 26237.04
2 rows selected.
SQL>
如果包含两列,将会产生聚合结果(解释为笛卡尔积?),返回10列(2*5)
SELECT fact_1_id,
fact_2_id,
COUNT(*) AS num_rows,
SUM(sales_value) AS sales_value
FROM dimension_tab
GROUP BY fact_1_id, fact_2_id
ORDER BY fact_1_id, fact_2_id;
FACT_1_ID FACT_2_ID NUM_ROWS SALES_VALUE
---------- ---------- ---------- -----------
1 1 83 4363.55
1 2 96 4794.76
1 3 93 4718.25
1 4 105 5387.45
1 5 101 5027.34
2 1 109 5652.84
2 2 96 4583.02
2 3 110 5555.77
2 4 113 5936.67
2 5 94 4508.74
10 rows selected.
SQL>
2.ROLLUP
出了正常的分组结果外,使用rollup还会,返回部分列分组,规则:从右向左,知道一个完整的分组.如果,rollup中的列数为n,那么将会有n+1个分组层次.
SELECT fact_1_id,
fact_2_id,
SUM(sales_value) AS sales_value
FROM dimension_tab
GROUP BY ROLLUP (fact_1_id, fact_2_id)
ORDER BY fact_1_id, fact_2_id;
FACT_1_ID FACT_2_ID SALES_VALUE
---------- ---------- -----------
1 1 4363.55
1 2 4794.76
1 3 4718.25
1 4 5387.45
1 5 5027.34
1 24291.35
2 1 5652.84
2 2 4583.02
2 3 5555.77
2 4 5936.67
2 5 4508.74
2 26237.04
50528.39
13 rows selected.
SQL>
以下语句结果:Click Here.当行中包含null的时候,这种并不是一种很好的方法,稍后会讨论.
SELECT fact_1_id,
fact_2_id,
fact_3_id,
SUM(sales_value) AS sales_value
FROM dimension_tab
GROUP BY ROLLUP (fact_1_id, fact_2_id, fact_3_id)
ORDER BY fact_1_id, fact_2_id, fact_3_id;
也可以在group by的时候使用rolluo做部分分组.结果:Click Here
SELECT fact_1_id,
fact_2_id,
fact_3_id,
SUM(sales_value) AS sales_value
FROM dimension_tab
GROUP BY fact_1_id, ROLLUP (fact_2_id, fact_3_id)
ORDER BY fact_1_id, fact_2_id, fact_3_id;
3.CUBE
Cube是rollup的拓展,Cube将会返回所有分组的一个统计,如果,Cube中有n组,那么将返回2的n次方组合结果.
SELECT fact_1_id,
fact_2_id,
SUM(sales_value) AS sales_value
FROM dimension_tab
GROUP BY CUBE (fact_1_id, fact_2_id)
ORDER BY fact_1_id, fact_2_id;
FACT_1_ID FACT_2_ID SALES_VALUE
---------- ---------- -----------
1 1 4363.55
1 2 4794.76
1 3 4718.25
1 4 5387.45
1 5 5027.34
1 24291.35
2 1 5652.84
2 2 4583.02
2 3 5555.77
2 4 5936.67
2 5 4508.74
2 26237.04
1 10016.39
2 9377.78
3 10274.02
4 11324.12
5 9536.08
50528.39
18 rows selected.
SQL>
如果cube中有三组(或n组)时,结果(或需计算):Click Here
SELECT fact_1_id,
fact_2_id,
fact_3_id,
SUM(sales_value) AS sales_value
FROM dimension_tab
GROUP BY CUBE (fact_1_id, fact_2_id, fact_3_id)
ORDER BY fact_1_id, fact_2_id, fact_3_id;
也有可能只是一部分进行cube分组,结果:Click Here
SELECT fact_1_id,
fact_2_id,
fact_3_id,
SUM(sales_value) AS sales_value
FROM dimension_tab
GROUP BY fact_1_id, CUBE (fact_2_id, fact_3_id)
ORDER BY fact_1_id, fact_2_id, fact_3_id;
4.Grouping 函数
上面曾经说过,如果某一列存在null值,而我们的rollup或者cube时,也将会出现列为null的时候,该怎样区分null到底是由数据null还是由于rollup活cube自己生成的null呢.这里使用grouping来解决这个问题.如果是数据本身的值(null或其他值),grouping(列名)将会在这一行返回0,如果这一行是由于rollup或者cube产生的话,将会返回1.
SELECT fact_1_id,
fact_2_id,
SUM(sales_value) AS sales_value,
GROUPING(fact_1_id) AS f1g,
GROUPING(fact_2_id) AS f2g
FROM dimension_tab
GROUP BY CUBE (fact_1_id, fact_2_id)
ORDER BY fact_1_id, fact_2_id;
FACT_1_ID FACT_2_ID SALES_VALUE F1G F2G
---------- ---------- ----------- ---------- ----------
1 1 4363.55 0 0
1 2 4794.76 0 0
1 3 4718.25 0 0
1 4 5387.45 0 0
1 5 5027.34 0 0
1 24291.35 0 1
2 1 5652.84 0 0
2 2 4583.02 0 0
2 3 5555.77 0 0
2 4 5936.67 0 0
2 5 4508.74 0 0
2 26237.04 0 1
1 10016.39 1 0
2 9377.78 1 0
3 10274.02 1 0
4 11324.12 1 0
5 9536.08 1 0
50528.39 1 1
18 rows selected.
SQL>
可以看出:(都是由rollup或cube所产生的)
- F1G=0,F2G=0 : 正常的group by结果
- F1G=0,F2G=1 : 一行
FACT_1_ID列的统计 - F1G=1,F2G=0 : 一行
FACT_2_ID列的统计 - F1G=1,F2G=1 : 由
FACT_1_ID和FACT_2_ID所产生的一个统计
grouping函数可以用来筛选排序结果.
SELECT fact_1_id,
fact_2_id,
SUM(sales_value) AS sales_value,
GROUPING(fact_1_id) AS f1g,
GROUPING(fact_2_id) AS f2g
FROM dimension_tab
GROUP BY CUBE (fact_1_id, fact_2_id)
HAVING GROUPING(fact_1_id) = 1 OR GROUPING(fact_2_id) = 1
ORDER BY GROUPING(fact_1_id), GROUPING(fact_2_id);
FACT_1_ID FACT_2_ID SALES_VALUE F1G F2G
---------- ---------- ----------- ---------- ----------
1 24291.35 0 1
2 26237.04 0 1
4 11324.12 1 0
3 10274.02 1 0
2 9377.78 1 0
1 10016.39 1 0
5 9536.08 1 0
50528.39 1 1
8 rows selected.
SQL>
5.GROUPING_ID函数
grouping_id函数也是提供了一个可以确定是否为统计的行的辨别方式.grouping_id返回的是group by分组的级别(或者说层次)
SELECT fact_1_id,
fact_2_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id) AS grouping_id
FROM dimension_tab
GROUP BY CUBE (fact_1_id, fact_2_id)
ORDER BY fact_1_id, fact_2_id;
FACT_1_ID FACT_2_ID SALES_VALUE GROUPING_ID
---------- ---------- ----------- -----------
1 1 4363.55 0
1 2 4794.76 0
1 3 4718.25 0
1 4 5387.45 0
1 5 5027.34 0
1 24291.35 1
2 1 5652.84 0
2 2 4583.02 0
2 3 5555.77 0
2 4 5936.67 0
2 5 4508.74 0
2 26237.04 1
1 10016.39 2
2 9377.78 2
3 10274.02 2
4 11324.12 2
5 9536.08 2
50528.39 3
18 rows selected.
SQL>
6.GROUP_ID
重复的统计的结果进行划分.第一次为0,依次出现相同结果,id 1开始递增.
SELECT fact_1_id,
fact_2_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id) AS grouping_id,
GROUP_ID() AS group_id
FROM dimension_tab
GROUP BY GROUPING SETS(fact_1_id, CUBE (fact_1_id, fact_2_id))
ORDER BY fact_1_id, fact_2_id;
FACT_1_ID FACT_2_ID SALES_VALUE GROUPING_ID GROUP_ID
---------- ---------- ----------- ----------- ----------
1 1 4363.55 0 0
1 2 4794.76 0 0
1 3 4718.25 0 0
1 4 5387.45 0 0
1 5 5027.34 0 0
1 24291.35 1 1
1 24291.35 1 0
2 1 5652.84 0 0
2 2 4583.02 0 0
2 3 5555.77 0 0
2 4 5936.67 0 0
2 5 4508.74 0 0
2 26237.04 1 1
2 26237.04 1 0
1 10016.39 2 0
2 9377.78 2 0
3 10274.02 2 0
4 11324.12 2 0
5 9536.08 2 0
50528.39 3 0
20 rows selected.
SQL>
也可对结果进行筛选.
SELECT fact_1_id,
fact_2_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id) AS grouping_id,
GROUP_ID() AS group_id
FROM dimension_tab
GROUP BY GROUPING SETS(fact_1_id, CUBE (fact_1_id, fact_2_id))
HAVING GROUP_ID() = 0
ORDER BY fact_1_id, fact_2_id;
FACT_1_ID FACT_2_ID SALES_VALUE GROUPING_ID GROUP_ID
---------- ---------- ----------- ----------- ----------
1 1 4363.55 0 0
1 2 4794.76 0 0
1 3 4718.25 0 0
1 4 5387.45 0 0
1 5 5027.34 0 0
1 24291.35 1 0
2 1 5652.84 0 0
2 2 4583.02 0 0
2 3 5555.77 0 0
2 4 5936.67 0 0
2 5 4508.74 0 0
2 26237.04 1 0
1 10016.39 2 0
2 9377.78 2 0
3 10274.02 2 0
4 11324.12 2 0
5 9536.08 2 0
50528.39 3 0
18 rows selected.
SQL>
7.GROUPING 集合
使用cube,特别是在多列时,分组会很多.以下为例,将会有8个组层次.结果:Click Here
SELECT fact_1_id,
fact_2_id,
fact_3_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id, fact_3_id) AS grouping_id
FROM dimension_tab
GROUP BY CUBE(fact_1_id, fact_2_id, fact_3_id)
ORDER BY fact_1_id, fact_2_id, fact_3_id;
如果我们只需要其中一些分组结果,用grouping sets来筛选,选出"FACT_1_ID, FACT_2_ID" and "FACT_1_ID, FACT_3_ID分组的统计
SELECT fact_1_id,
fact_2_id,
fact_3_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id, fact_3_id) AS grouping_id
FROM dimension_tab
GROUP BY GROUPING SETS((fact_1_id, fact_2_id), (fact_1_id, fact_3_id))
ORDER BY fact_1_id, fact_2_id, fact_3_id;
FACT_1_ID FACT_2_ID FACT_3_ID SALES_VALUE GROUPING_ID
---------- ---------- ---------- ----------- -----------
1 1 4363.55 1
1 2 4794.76 1
1 3 4718.25 1
1 4 5387.45 1
1 5 5027.34 1
1 1 2737.4 2
1 2 1854.29 2
1 3 2090.96 2
1 4 2605.17 2
1 5 2590.93 2
1 6 2506.9 2
1 7 1839.85 2
1 8 2953.04 2
1 9 2778.75 2
1 10 2334.06 2
2 1 5652.84 1
2 2 4583.02 1
2 3 5555.77 1
2 4 5936.67 1
2 5 4508.74 1
2 1 3512.69 2
2 2 2847.94 2
2 3 2972.5 2
2 4 2534.06 2
2 5 3115.99 2
2 6 2775.85 2
2 7 2208.19 2
2 8 2358.55 2
2 9 1884.11 2
2 10 2027.16 2
30 rows selected.
SQL>
8.组合列
使用rollup的列的组合(统计),
ROLLUP (a, b, c) (a, b, c) (a, b) (a) ()使用cube的列的组合(统计),
CUBE (a, b, c) (a, b, c) (a, b) (a, c) (a) (b, c) (b) (c) ()允许使用括号将列括起来.在使用rollup或cube,grouping sets时,会将其作为一个单独的整体,不会拆分括号里面的列再组合(统计).
ROLLUP ((a, b), c) (a, b, c) (a, b) () Not considered: (a)
CUBE ((a, b), c) (a, b, c) (a, b) (c) () Not considered: (a, c) (a) (b, c) (b)
以下使用cube时,括号与不带括号.结果:Click Here For Regular 和Click Here For braces
-- Regular Cube.
SELECT fact_1_id,
fact_2_id,
fact_3_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id, fact_3_id) AS grouping_id
FROM dimension_tab
GROUP BY CUBE(fact_1_id, fact_2_id, fact_3_id)
ORDER BY fact_1_id, fact_2_id, fact_3_id;
-- Cube with composite column.
SELECT fact_1_id,
fact_2_id,
fact_3_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id, fact_3_id) AS grouping_id
FROM dimension_tab
GROUP BY CUBE((fact_1_id, fact_2_id), fact_3_id)
ORDER BY fact_1_id, fact_2_id, fact_3_id;
9.ConcatenatedGroupings(个人理解:组之间的笛卡尔积)
在使用group by时,有GROUPING SETS, CUBE 和ROLLUP组合时,不同组会产生笛卡尔积.
SELECT fact_1_id,
fact_2_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id) AS grouping_id
FROM dimension_tab
GROUP BY GROUPING SETS(fact_1_id, fact_2_id)
ORDER BY fact_1_id, fact_2_id;
FACT_1_ID FACT_2_ID SALES_VALUE GROUPING_ID
---------- ---------- ----------- -----------
1 24291.35 1
2 26237.04 1
1 10016.39 2
2 9377.78 2
3 10274.02 2
4 11324.12 2
5 9536.08 2
7 rows selected.
SQL>
SELECT fact_3_id,
fact_4_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_3_id, fact_4_id) AS grouping_id
FROM dimension_tab
GROUP BY GROUPING SETS(fact_3_id, fact_4_id)
ORDER BY fact_3_id, fact_4_id;
FACT_3_ID FACT_4_ID SALES_VALUE GROUPING_ID
---------- ---------- ----------- -----------
1 6250.09 1
2 4702.23 1
3 5063.46 1
4 5139.23 1
5 5706.92 1
6 5282.75 1
7 4048.04 1
8 5311.59 1
9 4662.86 1
10 4361.22 1
1 4718.55 2
2 5439.1 2
3 4643.4 2
4 4515.3 2
5 5110.27 2
6 5910.78 2
7 4987.22 2
8 4846.25 2
9 5458.82 2
10 4898.7 2
20 rows selected.
SQL>
SELECT fact_1_id,
fact_2_id,
fact_3_id,
fact_4_id,
SUM(sales_value) AS sales_value,
GROUPING_ID(fact_1_id, fact_2_id, fact_3_id, fact_4_id) AS grouping_id
FROM dimension_tab
GROUP BY GROUPING SETS(fact_1_id, fact_2_id), GROUPING SETS(fact_3_id, fact_4_id)
ORDER BY fact_1_id, fact_2_id, fact_3_id, fact_4_id;
GROUPING SETS(fact_1_id, fact_2_id) (fact_1_id) (fact_2_id) GROUPING SETS(fact_3_id, fact_4_id) (fact_3_id) (fact_4_id) GROUPING SETS(fact_1_id, fact_2_id), GROUPING SETS(fact_3_id, fact_4_id) (fact_1_id, fact_3_id) (fact_1_id, fact_4_id) (fact_2_id, fact_3_id) (fact_2_id, fact_4_id)在两个GROUPING SETS中将产生笛卡尔积.
GROUPING SETS(a, b), GROUPING SETS(c, d) (a, c) (a, d) (b, c) (b, d)
[个人理解]
在使用group by的时候,特别是使用统计函数时,cube rollup和grouping sets是很有用的,可以分组进行统计.在分组的同时,我们还有grouping_id(返回分组的级别层次计数)和grouping(判断行是否由cube或rollup产生)进行一系列的操作.