MySQL多次加入同一个表使计数变得奇怪

您的问题是,您将所有超时都加入到所有确认中。因此,对于具有2个超时和3个确认的注册,您将获得2 x 3=6行,然后进行聚合。

聚合多个表时,请在加入前进行聚合:

select 
  user_id, 
  round(case when total_messages = 0 then 0 else total_confirmed / total_messages end, 2) as confirmation_rate
from
(
  select 
    s.user_id, 
    t.total_timeouts,
    c.total_confirmed,
    coalesce(t.timeoutCount,0) + coalesce(c.total_confirmed, 0) as total_messages
  from signups s
  left join 
  (
    select user_id, count(*) as total_timeouts
    from confirmations
    where action = 'timeout'
    group by user_id
  ) as t on t.user_id = s.user_id
  left join 
  (
    select user_id, count(*) as total_confirmed
    from confirmations
    where action = 'confirmed'
    group by user_id
  ) as c on c.user_id = s.user_id
)
order by user_id;

另一种选择是这里的条件聚合:

select 
  user_id, 
  round(case when total_messages = 0 then 0 else total_confirmed / total_messages end, 2) as confirmation_rate
from
(
  select 
    s.user_id, 
    sum(c.action = 'timeout') as total_timeouts,
    sum(c.action = 'confirmed') as total_confirmed,
    sum(c.action = 'timeout') + sum(c.action = 'confirmed') as total_messages
  from signups s
  left join confirmations c on c.user_id = s.user_id
  group by s.user_id
)
order by user_id;

要查看发生了什么,请查看不带分组依据的子查询。

select 
  Signups.user_id,
  timeoutTable.*,
  confirmedTable.*
from Signups
left join (
  select * from Confirmations where action = 'timeout'
) as timeoutTable on Signups.user_id = timeoutTable.user_id
left join (
  select * from Confirmations where action = 'confirmed'
) as confirmedTable on Signups.user_id = confirmedTable.user_id

            timeoutTable                                    confirmedTable
user_id     user_id     time_stamp              action      user_id     time_stamp              action
15          null        null                    null        null        null                    null
16          null        null                    null        null        null                    null
7           7           2021-03-25 07:40:25     timeout     7           2020-01-24 15:43:47     confirmed
7           7           2020-03-31 13:11:43     timeout     7           2020-01-24 15:43:47     confirmed
10          null        null                    null        null        null                    null
5           null        null                    null        null        null                    null
9           null        null                    null        null        null                    null
8           8           2021-03-07 19:48:06     timeout     8           2020-07-27 19:43:25     confirmed
12          null        null                    null        null        null                    null

请注意user_id 7有2行,每行都有一个超时 和 确认。这些行被计算为 二者都 超时和确认。这是因为您要同时连接两个表。

如果我们先进行聚合,然后再加入,计数就会计算出来。

select 
  Signups.user_id,
  timeoutTable.*,
  confirmedTable.*
from Signups
left join (
  select user_id, count(user_id) as timeouts
  from Confirmations
  where action = 'timeout'
  group by user_id
) as timeoutTable on Signups.user_id = timeoutTable.user_id
left join (
  select user_id, count(user_id) as confirmations
  from Confirmations
  where action = 'confirmed'
  group by user_id
) as confirmedTable on Signups.user_id = confirmedTable.user_id

            timeoutTable            confirmedTable
user_id     user_id     timeouts    user_id     confirmations
15          null        null        null        null
16          null        null        null        null
7           7           2           7           1
10          null        null        null        null
5           null        null        null        null
9           null        null        null        null
8           8           1           8           1
12          null        null        null        null

Demonstration

注:

left join (
  select * from Confirmations where action = 'timeout'
) as timeoutTable on Signups.user_id = timeoutTable.user_id

最好写成没有子查询。

left join Confirmations as timeoutTable
  on Signups.user_id = timeoutTable.user_id
 and where timeoutTable.action = 'timeout'

注意:当列名与SQL关键字冲突或非常接近时,这很好地表明您的名称过于通用。例如 time_stamp 。什么时候盖章?使用更具描述性的内容,例如 created_at 或 action_at 或 confirmed_at 。

这就引出了每一行确认代表什么的问题。它们不是确认,而是确认尝试。也许该表应该称为ConfirmationAttempts,时间戳为 attempted_at 。