MySQL 将选择子查询重构为连接

Question

I have 4 separate select subqueries here that are just selecting 1 column I need – but I know this is inefficient.

How can the same be achieved but with joins? I've tried with subqueries inside joins with group by, but unfortunately haven't been able to get it working.

SELECT [columns needed],
(SELECT posts.created_at FROM posts WHERE posts.thread_id = threads.id ORDER BY posts.created_at DESC LIMIT 1) as lastpost_created_at,
(SELECT users.username FROM posts INNER JOIN users on posts.user_id = users.id WHERE posts.thread_id = threads.id ORDER BY posts.created_at DESC LIMIT 1) as lastpost_username,
(SELECT users.avatar FROM posts INNER JOIN users on posts.user_id = users.id WHERE posts.thread_id = threads.id ORDER BY posts.created_at DESC LIMIT 1) as lastpost_avatar,
(SELECT COUNT(*) FROM posts WHERE posts.thread_id = threads.id) as replies
FROM threads
INNER JOIN posts ON threads.post_id = posts.id
INNER JOIN users ON posts.user_id = users.id
ORDER BY lastpost_created_at DESC

Answer 1

I'm not sure if I've interpreted correctly, but it seems like you want to get:

• Every thread,
• With the most recent post per thread,
• Including the user details (avatar etc) for who made the post
• And the total number of posts for each thread

I've put together an SQLfiddle that I believe supplies all of these, query below:

select threads.id,
       lastpost_created_at.lastpost,
       posts.user_id,
       /* posts.other_fields, */
       users.avatar,
       /* users.other_fields, */
       post_counts.replies
from   threads
       left join 
       (
          /* first we aggregate the posts by thread_id 
             to get the most recent created timestamp. */
          /* Note that this assumes unique timestamp per post / thread_id */
          select  max( posts.created_at ) as lastpost, 
                  thread_id
          from    posts
          group by thread_id
       ) as lastpost_created_at
       on threads.id = lastpost_created_at.thread_id

       /* Now we join on to the posts table to get the associated user id */
       /* (and any other fields you want from posts)*/
       left join 
       posts
       on threads.id = posts.thread_id
       and posts.created_at = lastpost_created_at.lastpost

       /* Then grab the user avatar etc */
       left join
       users
       on posts.user_id = users.id

       /* and finally get the total number of replies per thread */
       left join
       (
          select  thread_id,
                  count(*) as replies
          from    posts
          group by thread_id
       ) as post_counts
       on threads.id = post_counts.thread_id

One note: If two posts for a single thread have the exact same created_at timestamp, this will result in a duplicate row for both posts.

EDIT

I've put together a query that only returns one result in the case of duplicate timestamps:

select threads.id,
       most_recent_post.created_at,
       most_recent_post.user_id,
       users.avatar,
       post_counts.replies
from   threads
       left join 
       (
          /* Get the most recent post per thread, ordered by created at and ID */
          select  di.id, di.user_id, di.created_at, di.thread_id
          from    (
                      select  thread_id, max(created_at) AS created_at
                      from    posts d
                      group by thread_id
                  ) as dd
                  inner join    
                  posts as di
                  on di.id = (
                      select  id
                      from    posts ds
                      where   ds.thread_id = dd.thread_id
                              and ds.created_at = dd.created_at
                      order by id desc
                      limit 1
                  )
       ) as most_recent_post
       on threads.id = most_recent_post.thread_id

       /* Then grab the user avatar etc */
       left join
       users
       on most_recent_post.user_id = users.id

       /* and finally get the total number of replies per thread */
       left join
       (
          select  thread_id,
                  count(*) as replies
          from    posts
          group by thread_id
       ) as post_counts
       on threads.id = post_counts.thread_id

Updated SQLfiddle here . This uses a technique for Groupwise Maximum found here .