在Hive中,所有的表生成函数,包括用户自定义的和内置的,都统称为用户自定义表生成函数(user defined table generating functions),简称udtf。本文只介绍Hive自带的内置表生成函数。
1、explode(array)
功能:返回n行,每行对应数组中的一个元素。
spark-sql> select explode(array(1,2,3,4));
col
1
2
3
4
Time taken: 0.048 seconds, Fetched 4 row(s)
2、explode(map)
功能:返回n行两列,每行对应每个map键-值,第一列是map的键,第二列是map的值。(不常用)
spark-sql> select explode(map(1,2,3,4));
key value
1 2
3 4
Time taken: 0.039 seconds, Fetched 2 row(s)
3、posexplode(array)
功能:与explode类似,但除了数组元素本身(第二列),还返回各元素在数组中的位置(从0开始,第一列)。(不常用)
spark-sql> select posexplode(array(2,4,6,8));
pos col
0 2
1 4
2 6
3 8
Time taken: 0.034 seconds, Fetched 4 row(s)
4、stack(int n, v_1, v_2, ..., v_k)
功能:把k列数据转换成n行,k/n列,其中n必须是正整数,后面的v_1到v_k必须是元素,不能是列名。(不常用)
4.1、n设为3,将后面6个元素按顺序分为3行2列
select stack(3,1,2,3,4,5,6);
spark-sql> select stack(3,1,2,3,4,5,6);
col0 col1
1 2
3 4
5 6
Time taken: 0.032 seconds, Fetched 3 row(s)
4.2、n设为2,将后面6个元素按顺序分为2行3列
select stack(2,1,2,3,4,5,6);
spark-sql> select stack(2,1,2,3,4,5,6);col0 col1 col2
1 2 3
4 5 6
Time taken: 0.039 seconds, Fetched 2 row(s)
4.3、n设为3,将后面7个元素按顺序分为3行3列
select stack(3,1,2,3,4,5,6,7);
spark-sql> select stack(3,1,2,3,4,5,6,7);
col0 col1 col2
1 2 3
4 5 6
7 NULL NULL
Time taken: 0.035 seconds, Fetched 3 row(s)
4.4、n设为6,将后面6个元素转为为6行1列
select stack(6,1,2,3,4,5,6);
spark-sql> select stack(6,1,2,3,4,5,6);
col0
1
2
3
4
5
6
Time taken: 0.03 seconds, Fetched 6 row(s)
5、表生成函数必备:lateral view
udtf有一个很大的限制,在使用udtf时,select后面只能跟udtf,不能跟其他任何字段,否则会报错,如下:
hive 报错,spark-sql不报错,如下所示:
hive> select 1 as flag,explode(array(1,2,3,4));
FAILED: SemanticException [Error 10081]: UDTF's are not supported outside the SELECT clause, nor nested in expressions
spark-sql> select 1 as flag,explode(array(1,2,3,4));
flag col
1 1
1 2
1 3
1 4
Time taken: 0.044 seconds, Fetched 4 row(s)
2022-11-07 21:15:01,476 INFO thriftserver.SparkSQLCLIDriver: Time taken: 0.044 seconds, Fetched 4 row(s)
1)语法:
select
列别名1[ ,列别名2,列别名3……]
from 表名 lateral view udtf(expression) 虚拟表别名 as 列别名1[ ,列别名2,列别名3……]
lateral view跟在from后面,然后跟要使用的udtf,为生成的虚拟表起一个表别名,不写会报错。然后跟as 列别名,有些udtf会产生多个列,所以有时要跟多个列别名。
2)应用1:行转列
原始数据:
select *
from
(
select "a" as shop,"1,2,4" as uid_array
union
select "b" as shop,"4,5,6" as uid_array
)tb_final
a 1,2,4
b 4,5,6
Time taken: 5.345 seconds, Fetched 2 row(s)
首先使用split函数对uid_array进行切割,返回一个数组,然后使用lateral view explode进行行转列
select shop
,uid --这里是下面生成的列别名
from
( --准备原始数据
select *
from
(
select "a" as shop,"1,2,4" as uid_array
union
select "b" as shop,"4,5,6" as uid_array
)tb_final
)temp_test6
lateral view explode(split(uid_array, ',')) tb_a as uid;
a 1
a 2
a 4
b 4
b 5
b 6
Time taken: 1.479 seconds, Fetched 6 row(s)
3)应用2:求总聚合结果
举例:
此时想一句hql求出每个商店的来客数,以及两个商店去重后的来客数。由于uid为4的用户同时出现在两家商店,所以统计total维度时用户数为5。最终要得到如下结果:
a 3
b 3
总 5
explode实现,关键点在于构造一个array数组,将原本的聚合维度字段放入,然后任意自定义一个词例如‘total’作为总聚合的维度名称,再自定义一个别名(这里使用total_shop),代表包含total的字段名,聚合时使用新自定义的别名进行聚合,如下:
select total_shop
,count(distinct uid) as uid_num
from
( --准备原始数据
select shop
,uid --这里是下面生成的列别名
from
( --准备原始数据
select *
from
(
select "a" as shop,"1,2,4" as uid_array
union
select "b" as shop,"4,5,6" as uid_array
)tb_final
)temp_test6
lateral view explode(split(uid_array, ',')) tb_a as uid
)temp_test7
lateral view explode(array(shop, '总')) tb_a as total_shop
group by total_shop
order by total_shop;
a 3
b 3
总 5
Time taken: 2.14 seconds, Fetched 3 row(s)
这样的写法等价于:
select shop
,count(distinct uid) as uid_num
from
( --准备原始数据
select shop
,uid --这里是下面生成的列别名
from
( --准备原始数据
select *
from
(
select "a" as shop,"1,2,4" as uid_array
union
select "b" as shop,"4,5,6" as uid_array
)tb_final
)temp_test6
lateral view explode(split(uid_array, ',')) tb_a as uid
)temp_test7
group by shop
union all
select '总' as shop
,count(distinct uid) as uid_num
from
( --准备原始数据
select shop
,uid --这里是下面生成的列别名
from
( --准备原始数据
select *
from
(
select "a" as shop,"1,2,4" as uid_array
union
select "b" as shop,"4,5,6" as uid_array
)tb_final
)temp_test6
lateral view explode(split(uid_array, ',')) tb_a as uid
)temp_test7 ;
a 3
b 3
总 5
Time taken: 2.715 seconds, Fetched 3 row(s)
6、json_tuple(jsonstr, k1, k2, ...)
功能:从一个json字符串中获取多个key对应的value并作为一个元组返回。
举例:具体案例见
7、parse_url_tuple(url, p1, p2, ...)
功能:返回从url中抽取指定n部分的内容并作为一个元组返回,参数url是url字符串,而参数p1,p2,....是要抽取的部分。
举例:具体案例见