Impala - 将列转换为行而不使用 UNION ALL

Question

我有一个包含用户和每周活动的表格，例如这是user_activity 表格：

userid | wk1 | wk2 | wk3
u1     |  1  |  0  |  1
u2     |  0  |  1  |  0
u3     |  1  |  0  |  1

我想把它改成：

week   | active
wk1    |   2
wk2    |   1
wk3    |   2

我可以像这样使用 UNION ALL 来实现这一点：

SELECT 'wk1' as week,
       SUM( wk1 ) AS active
FROM user_activity
UNION ALL
SELECT 'wk2' as week,
       SUM( wk2 ) AS active
FROM user_activity
UNION ALL
SELECT 'wk3' as week,
       SUM( wk3 ) AS active
FROM user_activity;

有没有办法在不使用 UNION ALL 的情况下达到同样的效果？

谢谢！

编辑： Impala 版本：2.6.0

没有 UNION ALL 的原因：对于每个 SELECT，将从 HDFS 扫描整个表。如果我有一个巨大的表，这将导致 OOM 错误。

Answer 1

这应该可以解决您的性能问题。
该表仅被扫描一次。
记录没有被复制 X3。
UNION ALL 在这里仅用于单个记录。

select  concat('wk',cast(c.i as string))    as week
       ,case c.i 
            when 1 then wk1 
            when 2 then wk2
            when 3 then wk3
        end                 as active

from   (select  sum(wk1) AS wk1
               ,sum(wk2) AS wk2
               ,sum(wk3) AS wk3 

        from    user_activity
        ) t

        cross join (            select 1 as i
                    union all   select 2 
                    union all   select 3
                    ) c
;                    

---

+------+--------+
| week | active |
+------+--------+
| wk1  | 2      |
| wk2  | 1      |
| wk3  | 2      |
+------+--------+

Answer 2

您可以尝试反透视和聚合。这样 user_activity 表只会被读取一次。

select
    w.week,
    sum(case w.week 
        when 'wk1' then wk1
        when 'wk2' then wk2
        when 'wk3' then wk3
    end) active
from user_activity u
cross join (
    select 'wk1' week union all
    select 'wk2' week union all
    select 'wk3' week 
) w group by w.week;

生产：

+------+--------+
| week | active |
+------+--------+
| wk1  | 2      |
| wk2  | 1      |
| wk3  | 2      |
+------+--------+

它只需要一个聚合而不是三个。我使用UNION ALL 仅用于构建自定义数据透视查询。我没有在用户表上使用它。

Answer 3

忽略琐碎的“使用UNION”，这个问题似乎有点荒谬。但这里有一种方法：

with nounionall as (
      select (case row_number() over (order by userid)
                  when 1 then 'wk1'
                  when 2 then 'wk2'
                  when 3 then 'wk3'
              end) as week
      from user_activity ua
      limit 3
     )
select nounionall.week,
       sum(case when nounionall.week = 'wk1' then wk1
                when nounionall.week = 'wk2' then wk2
                when nounionall.week = 'wk3' then wk3
           end) as actives
from nounionall cross join
     user_activity ua
group by nounionall.week

Answer 4

这个没有任何联合的非常简单的解决方案怎么样：-

SELECT [week],active
FROM (
    SELECT SUM(wk1)wk1,SUM(wk2)wk2,SUM(wk3)wk3
    FROM user_activity)pvt
UNPIVOT ([active] FOR [Week] IN  (wk1,wk2,wk3)) unpvt