左外连接 Postgres 中深两层的结果是笛卡尔积

Question

给定以下 4 个表：

CREATE TABLE events ( id, name )
CREATE TABLE profiles ( id, event_id )
CREATE TABLE donations ( amount, profile_id )
CREATE TABLE event_members( id, event_id, user_id )

我正在尝试获取所有事件的列表，以及任何成员的计数以及任何捐赠的总和。问题是捐赠的总和是错误的（似乎是捐赠的笛卡尔结果 * # of event_members）。

这是 SQL 查询 (Postgres)

SELECT events.name, COUNT(DISTINCT event_members.id), SUM(donations.amount)
FROM            events
LEFT OUTER JOIN profiles      ON events.id = profiles.event_id
LEFT OUTER JOIN donations     ON donations.profile_id = profiles.id
LEFT OUTER JOIN event_members ON event_members.event_id = events.id
GROUP BY events.name

sum(donations.amount) 返回 = 实际捐款总额 * event_members 中的行数。如果我注释掉 count(distinct event_members.id) 和 event_members 左外连接，则总和是正确的。

编辑：埃尔文指出我正确的方向。查询重写为：

SELECT events.name, COUNT(DISTINCT event_members.id), 
  从捐赠中选择（SUM（donations.amount），其中捐赠.profile_id =profiles.id和profiles.event_id = events.id）作为total_donations
    FROM 事件
    左外连接 event_members ON event_members.event_id = events.id
    GROUP BY events.name

Answer 1

正如我在引用的问题下详细解释的那样，您需要先聚合，然后加入表以避免代理CROSS JOIN。像这样：

SELECT e.name, e.sum_donations, m.ct_members
FROM (
    SELECT e.id, e.name, SUM(d.amount) AS sum_donations
    FROM   events             e
    LEFT   JOIN profiles      p ON p.event_id = e.id
    LEFT   JOIN donations     d ON d.profile_id = p.id
    GROUP  BY 1, 2
    ) e
LEFT   JOIN (
    SELECT event_id, COUNT(DISTINCT id) AS ct_members
    FROM   event_members
    GROUP  BY 1
    ) m ON m.event_id = e.id

IFevent_members.id是主键（正如人们可能假设的那样），您可以简化为

COUNT(*) AS ct_members

因为id保证是UNIQUE NOT NULL. 这样快了一点。

Answer 2

你似乎有这两个独立的结构（-[意味着1-N关联）：

events -[ profiles -[ donations
events -[ event members

我将第二个包装到子查询中：

SELECT events.name,
  member_count.the_member_count
  COUNT(DISTINCT event_members.id),
  SUM(donations.amount)

FROM            events
LEFT OUTER JOIN profiles      ON events.id = profiles.event_id
LEFT OUTER JOIN donations     ON donations.profile_id = profiles.id

LEFT OUTER JOIN (
  SELECT
    event_id,
    COUNT(*) AS the_member_count
  FROM event_members
  GROUP BY event_id
) AS member_count
  ON member_count.event_id = events.id

GROUP BY events.name