Hive OLAP多维聚合函数

OLAP多维聚合函数

Hive OLAP多维聚合函数是增强版的Group By语句。一般和Group By同时使用，用来进行多维分析。
本文总结Hive三种常用的多维聚合函数:With Cube 任意维度聚合、Grouping Sets 指定维度聚合、With Rollup 层级维度聚合。

    1
    2

测试数据

-- 建表
create table student_scores(
id int,
studentId int,
language int,
math int,
english int,
classId string,
departmentId string
);
-- 写入数据
insert into table student_scores values
  (1,111,68,69,90,'class1','department1'),
  (2,112,73,80,96,'class1','department1'),
  (3,113,90,74,75,'class1','department1'),
  (4,114,89,94,93,'class1','department1'),
  (5,115,99,93,89,'class1','department1'),
  (6,121,96,74,79,'class2','department1'),
  (7,122,89,86,85,'class2','department1'),
  (8,123,70,78,61,'class2','department1'),
  (9,124,76,70,76,'class2','department1'),
  (10,211,89,93,60,'class1','department2'),
  (11,212,76,83,75,'class1','department2'),
  (12,213,71,94,90,'class1','department2'),
  (13,214,94,94,66,'class1','department2'),
  (14,215,84,82,73,'class1','department2'),
  (15,216,85,74,93,'class1','department2'),
  (16,221,77,99,61,'class2','department2'),
  (17,222,80,78,96,'class2','department2'),
  (18,223,79,74,96,'class2','department2'),
  (19,224,75,80,78,'class2','department2'),
  (20,225,82,85,63,'class2','department2');

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    17
    18
    19
    20
    21
    22
    23
    24
    25
    26
    27
    28
    29
    30
    31
    32

With Cube 任意维度聚合

语法:GROUP BY a,b WITH CUBE。
作用:根据Group BY维度的所有可能组合进行聚合。类似于Apache Kylin的Cube多维立方体概念。n个维度会有2的n次方种组合。

    1
    2

select departmentid,classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores group by departmentid,classid with cube;

结果
departmentid    classid summath uv
NULL            NULL    1654    20
NULL            class1  930     11
NULL            class2  724     9
department1     NULL    718     9
department1     class1  410     5
department1     class2  308     4
department2     NULL    936     11
department2     class1  520     6
department2     class2  416     5

等价于union all:

-- 0个维度-没有维度
select null as departmentid,null classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores
union all
-- 1个维度-classid
select null as departmentid,classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores group by classid
union all
-- 1个维度-departmentid
select departmentid,null as classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores group by departmentid
union all
-- 2个维度-departmentid、classid
select departmentid,classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores group by departmentid,classid;

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    17
    18
    19
    20
    21
    22
    23
    24
    25
    26
    27
    28
    29
    30
    31
    32
    33

Grouping Sets 指定维度聚合

语法:GROUP BY a,b GROUPING SETS (a,b)
作用:根据GROUPING SETS指定维度组合进行聚合。是Cube的一部分。Grouping Sets 分组(Grouping) 集(Sets)，是多个分组的并集。等价于Union ALL单个分组结果。如grouping sets(A,B) 等价于...group by null,B union all ...group by A,null。

    1
    2

select departmentid,classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores
group by departmentid,classid grouping sets((departmentid,classid),departmentid);

结果:
departmentid    classid summath uv
department1     NULL    718     9
department1     class1  410     5
department1     class2  308     4
department2     NULL    936     11
department2     class1  520     6
department2     class2  416     5

等价于union all:
select departmentid,classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores
group by departmentid,classid
union all
select departmentid,null as classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores
group by departmentid,null;

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    17
    18
    19
    20
    21
    22

With Rollup 层级维度聚合

语法:GROUP BY a,b,c WITH ROLLUP
作用:以GROUP BY最左侧的维度为主，从该维度的角度去上卷、下钻。是Cube的一部分。

    1
    2

select departmentid,classid,sum(math) as sumMath,count(distinct studentid) as uv
from student_scores group by departmentid,classid with rollup;

结果
departmentid    classid summath uv
NULL            NULL    1654    20
department1     NULL    718     9
department1     class1  410     5
department1     class2  308     4
department2     NULL    936     11
department2     class1  520     6
department2     class2  416     5

等价于union all:
-- 下钻
select departmentid,classid,sum(math) as sumMath,count(distinct studentid) as uv from student_scores
group by departmentid,classid
union all
select departmentid,null as classid,sum(math) as sumMath,count(distinct studentid) as uv from student_scores
group by departmentid,null
-- 上卷
union all
select null as departmentid,null as classid,sum(math) as sumMath,count(distinct studentid) as uv from student_scores
group by null,null;