我遇到了一些"复杂的"查询,如:
with q1 as (
select w.entity_id, p.actionable_group, p.error_type
from issue_analysis p
join log l on p.id = l.issue_analysis_id
join website w on l.website_id = w.id
where l.date >= '2016-06-01'
group by 1, 2, 3
),
q2 as (
select w.entity_id, p.actionable_group, p.error_type
from issue_analysis p
join log l on p.id = l.issue_analysis_id
join website w on l.website_id = w.id
where l.date >= '2016-06-01'
group by 1, 2, 3
having count(1) = 1
)
试图
SELECT q1.entity_id, count(q1.entity_id), count(q2.entity_id)
from q1, q2
group by 1
order by 1
但是结果为我提供了一个"错误"数据,因为它并不真正包含两个计数...
你可以请描述一下最清洁的"在没有大量嵌套查询的情况下解决此类问题的方法?如果它可能会有所帮助 - q2
与q1
类似,但最后与having count(1) = 1
相似。
P.S。 文档链接会很好,答案很简单。
答案 0 :(得分:0)
毫无疑问,您获得笛卡尔积,这会影响聚合。相反,在进行连接之前聚合:
select q1.entity_id, q1_cnt, q2_cnt
from (select q1.entity_id, count(*) as q1_cnt
from q1
group by q1.entity_id
) q1 join
(select q2.entity_id, count(*) as q2_cnt
from q2
group by q2.entity_id
) q2
on q1.entity_id = q2.entity_id;