Question

我有一个MySQL（实际上是MariaDB）查询，当我尝试按DATETIME字段（sighted）而不是主键进行排序时，它运行得慢得多并且回退到filesort上。我有sighted的索引，但由于某种原因它似乎没有使用它。如何对此表进行索引，以便可以有效地对DATETIME字段进行排序？

select 
    user_id , t.name, f.value, y.name 
    from posts p 
        left join taxon_sightings t on (t.post_id = p.id) 
        left join fields f on (f.foreign_id = t.id) 
        left join fieldtypes y on (y.id = f.fieldtype_id) 
    group by p.id 
    order by p.sighted desc 
    limit 10;

当我按sighted DATETIME字段排序时：

MariaDB [dbase]> explain [yadda yadda, as above]
+----+-------------+-------+--------+---------------+------------+---------+--------------------------------+------+---------------------------------+
| id | select_type | table | type   | possible_keys | key        | key_len | ref                            | rows | Extra                           |
+----+-------------+-------+--------+---------------+------------+---------+--------------------------------+------+---------------------------------+
|  1 | SIMPLE      | p     | ALL    | NULL          | NULL       | NULL    | NULL                           | 1759 | Using temporary; Using filesort |
|  1 | SIMPLE      | t     | ref    | post_id       | post_id    | 4       | dataportal_test.p.id           |   12 |                                 |
|  1 | SIMPLE      | f     | ref    | foreign_id    | foreign_id | 4       | dataportal_test.t.id           |    6 | Using where                     |
|  1 | SIMPLE      | y     | eq_ref | PRIMARY       | PRIMARY    | 4       | dataportal_test.f.fieldtype_id |    1 | Using where                     |
+----+-------------+-------+--------+---------------+------------+---------+--------------------------------+------+---------------------------------+

比较我按ID排序的时间：

+----+-------------+-------+--------+---------------+------------+---------+--------------------------------+------+-------------+
| id | select_type | table | type   | possible_keys | key        | key_len | ref                            | rows | Extra       |
+----+-------------+-------+--------+---------------+------------+---------+--------------------------------+------+-------------+
|  1 | SIMPLE      | p     | index  | NULL          | PRIMARY    | 4       | NULL                           |    1 |             |
|  1 | SIMPLE      | t     | ref    | post_id       | post_id    | 4       | dataportal_test.p.id           |   12 |             |
|  1 | SIMPLE      | f     | ref    | foreign_id    | foreign_id | 4       | dataportal_test.t.id           |    6 | Using where |
|  1 | SIMPLE      | y     | eq_ref | PRIMARY       | PRIMARY    | 4       | dataportal_test.f.fieldtype_id |    1 | Using where |
+----+-------------+-------+--------+---------------+------------+---------+--------------------------------+------+-------------+

Answer 1

我认为你不能。我能想到的唯一技巧是试试这个：

select user_id, t.name, f.value, y.name 
from posts p left join
     taxon_sightings t
     on (t.post_id = p.id) left join
     fields f
     on (f.foreign_id = t.id) left join
     fieldtypes y
     on (y.id = f.fieldtype_id) 
group by p.sighted, p.id 
---------^ very important
order by p.sighted desc 
limit 10;

然后在posts(sighted, id)上添加一个索引。带group by的{{1}}可能会利用索引。如果你很幸运。。

Answer 2

假设p.id中的posts是唯一的，请让我们摆脱GROUP BY p.id。我认为它存在的唯一原因是因为JOINs正在膨胀行数？

除非你需要，否则摆脱LEFT。

SELECT  x.user_id, t.name, f.value, y.name
    FROM  
      ( SELECT  id, sighted
            FROM  posts
            -- No GROUP BY needed if id is UNIQUE
            ORDER BY  sighted DESC
            LIMIT  10 
      ) AS p
    JOIN  taxon_sightings t ON (t.post_id = p.id)
    JOIN  fields f ON (f.foreign_id = t.id)
    JOIN  fieldtypes y ON (y.id = f.fieldtype_id)
    GROUP BY  p.id
    ORDER BY  p.sighted DESC, p.id DESC;

并且（正如戈登建议的那样），INDEX(sighted, id)。

按日期时间排序时，加入大量的MySQL查询会变慢，而不是按ID排序

2 个答案: