具有多个分组或顺序的mysql查询优化

时间:2013-09-13 09:39:13

标签: mysql sql query-optimization

更新:表和索引定义

desc activities;x
+----------------+--------------+------+-----+---------+  
| Field          | Type         | Null | Key | Default |  
+----------------+--------------+------+-----+---------+  
| id             | int(11)      | NO   | PRI | NULL    |  
| trackable_id   | int(11)      | YES  | MUL | NULL    |  
| trackable_type | varchar(255) | YES  |     | NULL    |  
| owner_id       | int(11)      | YES  | MUL | NULL    |  
| owner_type     | varchar(255) | YES  |     | NULL    |  
| key            | varchar(255) | YES  |     | NULL    |  
| parameters     | text         | YES  |     | NULL    |  
| recipient_id   | int(11)      | YES  | MUL | NULL    |  
| recipient_type | varchar(255) | YES  |     | NULL    |  
| created_at     | datetime     | NO   |     | NULL    |  
| updated_at     | datetime     | NO   |     | NULL    |  
+----------------+--------------+------+-----+---------+  

show indexes from activities;

+------------+------------+-----------------------------------------------------+--------------+----------------+-----------+-------------+----------+--------+------+------------+  
| Table      | Non_unique | Key_name                                            | Seq_in_index | Column_name    | Collation | Cardinality | Sub_part | Packed | Null | Index_type |  
+------------+------------+-----------------------------------------------------+--------------+----------------+-----------+-------------+----------+--------+------+------------+  
| activities |          0 | PRIMARY                                             |            1 | id             | A         |        7263 |     NULL | NULL   |      | BTREE      |  
| activities |          1 | index_activities_on_trackable_id_and_trackable_type |            1 | trackable_id   | A         |        7263 |     NULL | NULL   | YES  | BTREE      |  
| activities |          1 | index_activities_on_trackable_id_and_trackable_type |            2 | trackable_type | A         |        7263 |     NULL | NULL   | YES  | BTREE      |  
| activities |          1 | index_activities_on_owner_id_and_owner_type         |            1 | owner_id       | A         |        7263 |     NULL | NULL   | YES  | BTREE      |  
| activities |          1 | index_activities_on_owner_id_and_owner_type         |            2 | owner_type     | A         |        7263 |     NULL | NULL   | YES  | BTREE      |  
| activities |          1 | index_activities_on_recipient_id_and_recipient_type |            1 | recipient_id   | A         |        2421 |     NULL | NULL   | YES  | BTREE      |  
| activities |          1 | index_activities_on_recipient_id_and_recipient_type |            2 | recipient_type | A         |        3631 |     NULL | NULL   | YES  | BTREE      |  
+------------+------------+-----------------------------------------------------+--------------+----------------+-----------+-------------+----------+--------+------+------------+  

select count(id) from activities;  
+-----------+  
| count(id) |  
+-----------+  
|      7117 |  
+-----------+  

这是我当前的查询:

SELECT act.*, group_concat(act.owner_id order by act.created_at desc) as owner_ids 
FROM (select * from activities order by created_at desc) as act 
INNER JOIN users on users.id = act.owner_id 
WHERE (users.city_id = 1 and act.owner_type = 'User') 
GROUP BY trackable_type, recipient_id, recipient_type 
order by act.created_at desc 
limit 20 offset 0;

做一个解释 Explain

我已经玩了很多这个查询,包括索引等。有没有办法优化这个查询?

5 个答案:

答案 0 :(得分:1)

我认为你根本不需要offset 0,看起来你也可以没有子查询。如果您不使用users表格中的字段,则可以使用in(或exists)明确说明:

select
    a.trackable_type, a.recipient_id, a.recipient_type,
    max(a.created_at) as max_created_at,
    group_concat(a.owner_id order by a.created_at desc) as owner_ids 
from activities as a
where
    a.owner_type = 'User' and
    a.owner_id in (select u.id from users as u where u.city_id = 1)
group by a.trackable_type, a.recipient_id, a.recipient_type
order by max_created_at desc
limit 20;

同样对我而言,如果您在owner_type, owner_id activities上创建索引(您的索引owner_id, owner_type对您的查询不起作用)并且索引在city_id上的users

答案 1 :(得分:1)

MySQL有时候很奇怪,所以我会试一试。我假设ID是用户表上的主键。

SELECT 
    act.trackable_type, act.recipient_id, act.recipient_type,
max(act.created_at) as max_created_at,
    group_concat(act.owner_id order by act.created_at DESC) as owner_ids 
FROM  activities act 
WHERE act.owner_id in (select id from users where city_id = 1)
AND act.owner_Type = 'User'
GROUP BY trackable_type, recipient_id, recipient_type 
ORDER BY max_created_at
LIMIT 20

答案 2 :(得分:0)

首先,我会开始使查询更具可读性: - )

您不需要带有ORDER BY的派生表,而是使用列列表而不是ACT。*。

SELECT ACT.TRACKABLE_TYPE, ACT.RECIPIENT_ID, ACT.RECIPIENT_TYPE, MAX(ACT.CREATED_AT) AS max_created,
   GROUP_CONCAT(ACT.OWNER_ID ORDER BY ACT.CREATED_AT DESC) AS OWNER_IDS 
FROM ACTIVITIES AS ACT 
JOIN USERS ON USERS.ID = ACT.OWNER_ID 
WHERE (USERS.CITY_ID = 1 AND ACT.OWNER_TYPE = 'USER') 
GROUP BY ACT.TRACKABLE_TYPE, ACT.RECIPIENT_ID, ACT.RECIPIENT_TYPE
ORDER BY max_created DESC 
LIMIT 20 OFFSET 0;

当您将USERS上的WHERE条件移动到派生表时,它可能会有所帮助:

SELECT ACT.TRACKABLE_TYPE, ACT.RECIPIENT_ID, ACT.RECIPIENT_TYPE, MAX(ACT.CREATED_AT) AS max_created,
   GROUP_CONCAT(ACT.OWNER_ID ORDER BY ACT.CREATED_AT DESC) AS OWNER_IDS 
FROM ACTIVITIES AS ACT 
JOIN (SELECT ID FROM USERS WHERE CITY_ID = 1) USERS 
  ON USERS.ID = ACT.OWNER_ID 
WHERE ACT.OWNER_TYPE = 'USER'
GROUP BY ACT.TRACKABLE_TYPE, ACT.RECIPIENT_ID, ACT.RECIPIENT_TYPE
ORDER BY max_created DESC 
LIMIT 20 OFFSET 0;

答案 3 :(得分:0)

您能否告诉我们您的用户表的大小,例如以下查询的结果:

select count(id) from users WHERE users.city_id = 1;

如果这是一个小数字,我建议使用

SELECT act.trackable_type, act.recipient_id, act.recipient_type, max(act.created_at) as max_created_at,
    group_concat(act.owner_id order by act.created_at DESC) as owner_ids 
FROM  activities act 
WHERE act.owner_id in (select id from users where city_id = 1)
AND act.owner_Type = 'User'
GROUP BY trackable_type, recipient_id, recipient_type 
ORDER BY max_created_at
LIMIT 20

否则,使用join会更好

SELECT ACT.TRACKABLE_TYPE, ACT.RECIPIENT_ID, ACT.RECIPIENT_TYPE, MAX(ACT.CREATED_AT) AS max_created_at,
   GROUP_CONCAT(ACT.OWNER_ID ORDER BY ACT.CREATED_AT DESC) AS OWNER_IDS 
FROM ACTIVITIES ACT 
JOIN USERS ON (USERS.CITY_ID = 1 AND USERS.ID = ACT.OWNER_ID)
WHERE ACT.OWNER_TYPE = 'USER'
GROUP BY ACT.TRACKABLE_TYPE, ACT.RECIPIENT_ID, ACT.RECIPIENT_TYPE
ORDER BY max_created DESC 
LIMIT 20;

答案 4 :(得分:0)

首先,这是一个非常棘手的查询,并且基于explainig的含义以及如何改进它,可以为开发人员职位构建一个有趣的访谈=)。

  1. MySQL使用nested loop joins,这意味着当有连接时,MySQL从表开始,并且表中的每个匹配行循环通过连接中第二个表中的相关行。

  2. 如果没有索引,那么对于每一行,MySQL都会进入磁盘并获取条件中使用的字段,并对另一个表中的每一行执行相同操作。在磁盘上运行既昂贵又耗时,最好从内存中获取信息,这样就可以从索引中获取数据。

  3. MySQL优化器选择了连接的顺序。但是你可以通过创建特殊索引(有时提示)来暗示MySQL。

  4. 当你执行这样的事情(select * from activities order by created_at desc)时,你将整个表加载到一个临时的无索引表中,这在任何情况下都不是一件好事。但最糟糕的是,MySQL应该从表开始连接,否则它需要在嵌套循环中检查表的每一行上的条件。

  5. 使用索引进行排序或分组(也需要排序)意味着什么?这意味着您按索引的顺序读取数据。但由于MySQL使用嵌套循环连接,因此当您排序的字段的表来自连接中的第一个表时,您只能利用索引进行排序。

  6. created_at字段未包含在group by子句中,这意味着您不关心从哪个组中挑选(并且它们在组内可能是相同的)

  7. 在复杂查询中,特别是在具有分页的查询中,通常最好只选择所需行的ID,然后将id重新连接到表中以用于其余字段(排序的数据越少,它需要的速度越快。

  8. 总结一下,我们需要使用索引从activities表开始连接,在嵌套循环中连接到users并获取id,然后返回join活动表其余的价值观。

  9. 因此,您需要在活动(owner_type, trackable_type, recipient_id, recipient_type, owner_id, created_at)上使用相当长的复合索引,并且可能是奢侈的,但需要索引 用户(id, city_id)

    现在,将查询重写为:

    SELECT *
    FROM
      (SELECT a.id, group_concat(a.owner_id order by a.created_at desc) as owner_ids
       FROM activities a
       JOIN users u ON a.owner_id = u.id AND u.city_id = 1
       WHERE a.owner_type = 'User'
       GROUP BY trackable_type, recipient_id, recipient_type
       ORDER BY a.created_at desc
       limit 20 offset 0) as owners
    JOIN activities a USING (id);
    

    您应该查看EXPLAIN,并且可能在子查询中使用STRAIGHT_JOIN而不是JOIN来确保正确的连接顺序。

    这个解决方案似乎是一个需要资源的解决方案。但它应该是您以下实验的良好基线。您应该首先介绍一些其他字段进行分组(在索引中使用varchar 255效率不高,尤其是其中两个),因此您应该使用一些足够的前缀,并将它们显式地作为分拣器引入或强制使用索引带前缀。你可以在一个特殊的石斑鱼场上投入一个函数,它是一个函数(trackable_type,recipient_id,recipient_type)。这个owner_type = 'User'也不是那么好,比较整数等更好。