选择在有很多行的表上花费大量时间

Question

我有这个选择：

select 'like' prefix
     , l.post
     , l.data as data
     , l.user
     , concat(k.user, ' liked you') as logs 
  from likes l 

inner join posts p on l.post = p.id 
inner join cadastro k on l.user = k.id 
where p.user = 1 and l.user <> p.user

order by data desc
limit 10

耗时 2.3993 秒。

有什么改进的想法吗？

`likes` (
  `id` int(11) UNSIGNED NOT NULL AUTO_INCREMENT,
  `user` int(11) UNSIGNED NOT NULL,
  `post` int(11) UNSIGNED NOT NULL,
  `data` datetime NOT NULL,
  PRIMARY KEY (`id`),
  UNIQUE KEY `user_post` (`user`,`post`),
  KEY `post_user` (post, user),
  FOREIGN KEY (`user`) REFERENCES cadastro (`id`),
  FOREIGN KEY (`post`) REFERENCES posts (`id`) ON DELETE CASCADE
)

`posts` (
  `id` int(11) UNSIGNED NOT NULL AUTO_INCREMENT,

`cadastro` (
  `id` int(11) UNSIGNED NOT NULL AUTO_INCREMENT,

有什么想法可以加快速度吗？我也尝试在喜欢数据上添加索引，但没有发现差异。

Answer 1

这是您的查询：

select 'like' as prefix, l.post, l.data as data, l.user,
        concat(k.user, ' liked you') as logs
from likes l join
     posts p
     on l.post = p.id oin
     cadastro k
     on l.user = k.id 
where p.user = 1 and l.user <> p.user
order by data desc
limit 10;

对于此查询，您需要post(user, id)、likes(post, user, data) 和cadastro(id, user) 上的索引。

您不能做太多的事情来消除order by 的开销，但这应该会加快查询的连接和过滤部分。

Answer 2

select  'like' prefix ,
        l.post ,
        l.data as data ,
        l.user ,
        ( SELECT concat(user, ' liked you')
              FROM cadastro  WHERE id = l.user ) AS logs
    from  likes l
    inner join  posts p  ON l.post = p.id
    where  p.user = 1
      and  l.user <> p.user
    order by  data desc
    limit  10

索引：

p: (user, data, id)  -- 'covering'; helps WHERE; may help ORDER BY
l: (post)
cadastro: I assume you have PRIMARY KEY(id)

进一步改进：将likes 索引从

PRIMARY KEY (`id`),
UNIQUE KEY `user_post` (`user`,`post`),
KEY `post_user` (post, user),

到

PRIMARY KEY(post, user),
INDEX(user, post)

并摆脱id。

警告：如果没有人“喜欢你”，则此查询的结果集可能会有所不同。如果这是一个问题，我将重新制定它以使用“派生”表。

通过将 cadastro 查找移动到子查询中，我认为该操作的发生频率低于 cadastro 在JOIN 中的情况。这不是一种通用优化——注意WHERE 子句很复杂，因为它涉及多个表。我不清楚p 上的索引是否会一直到data，并且对LIMIT 有帮助。如需进一步调查，请提供EXPLAIN FORMAT=JSON SELECT ...