在写 a full join b full join c 时,发现数据主键有重复,所以在此处记录下.
1. 问题描述
在Hive中(其他类似SQL,比如PostgreSQL可能也存在此问题),当对多张表(3张及以上)进行full join时,会存在每张表的主键都是唯一,但当full join后,会发现主键可能有重复。
2. 问题复现
2.1. 插入数据
with temp1 as (
select '1' as id ,'张三' as name
union all
select '2' as id ,'李四' as name
union all
select '3' as id ,'王五' as name
),
temp2 as (
select '1' as id ,'深圳' as city
union all
select '3' as id ,'北京' as city
union all
select '4' as id ,'上海' as city
),
temp3 as (
select '1' as id ,'5000' as salary
union all
select '4' as id ,'10000' as salary
union all
select '5' as id ,'12000' as salary
)
2.2. 查询SQL以及问题
select
coalesce(a.id, b.id, c.id) as id
, a.name
, b.city
, c.salary
from temp1 as a
full join temp2 as b
on a.id = b.id
full join temp3 as c
on a.id = c.id
当执行如上查询SQL时,会发现其中 id = 4 的数据有重复,如下图所示:
3. 问题原因
之所以会出现这样的问题,是因为是以a表为主表,但a表中只有 id 为 1、2、3 的数据,但在b表中有id为4,c表中也有id为4,此时full join时,a表会以空值和b表、c表中id为4的数据join,这样关联后的表中就会出现2条id为4的数据;
4. 问题解决
在后续的表full join时,不单单使用第一张表的主键full join(因为是full join,所以肯定会存在第一张表为null,而其他表有值的数据),而使用 coalesce 方法对所有前面已经full join了的主键进行条件关联,如下代码:
方法1:
select
coalesce(a.id, b.id, c.id) as id
, a.name
, b.city
, c.salary
from temp1 as a
full join temp2 as b
on a.id = b.id
full join temp3 as c
on coalesce(a.id, b.id) = c.id
方法2:
select
temp.id
, temp.name
, temp.city
, c.salary
from
(select
coalesce(a.id, b.id) as id
, a.name
, b.city
from temp1 as a
full join temp2 as b
on a.id = b.id) temp
full join
temp3 as c
on temp.id = c.id
方法3:
select
temp.id
,temp1.name
,temp2.city
,temp3.salary
from
(select id
from
(select
id
from temp1
union all
select
id
from temp2
union all
select
id
from temp3
) tt
group by id
) temp
left join temp1
on temp.id = temp1.id
left join temp2
on temp.id = temp2.id
left join temp3
on temp.id = temp3.id
结果如下: