Question

架构就像这样visits_table：

+---------------------------+----------------------+------+-----+---------+----------------+
| Field                     | Type                 | Null | Key | Default | Extra          |
+---------------------------+----------------------+------+-----+---------+----------------+
| idvisit                   | int(10) unsigned     | NO   | PRI | NULL    | auto_increment |
| idsite                    | int(10) unsigned     | NO   | MUL | NULL    |                |
| idvisitor                 | binary(8)            | NO   |     | NULL    |                |
| visit_time                | datetime             | NO   |     | NULL    |                |
| user_id                   | varchar(200)         | YES  |     | NULL    |                |
| config_cookie             | tinyint(1)           | NO   |     | NULL    |                |
| custom_var_k1             | varchar(200)         | YES  |     | NULL    |                |
| custom_var_v1             | varchar(200)         | YES  |     | NULL    |                |
+---------------------------+----------------------+------+-----+---------+----------------+

索引：

+----------------------+------------+------------------------------+--------------+------------------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| Table                | Non_unique | Key_name                     | Seq_in_index | Column_name            | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment |
+----------------------+------------+------------------------------+--------------+------------------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| visits_table         |          0 | PRIMARY                      |            1 | idvisit                | A         |        1502 | NULL     | NULL   |      | BTREE      |         |               |
| visits_table         |          1 | index_idsite_datetime        |            1 | idsite                 | A         |           5 | NULL     | NULL   |      | BTREE      |         |               |
| visits_table         |          1 | index_idsite_datetime        |            2 | visit_time             | A         |        1502 | NULL     | NULL   |      | BTREE      |         |               |
| visits_table         |          1 | index_idsite_idvisitor       |            1 | idsite                 | A         |           1 | NULL     | NULL   |      | BTREE      |         |               |
| visits_table         |          1 | index_idsite_idvisitor       |            2 | idvisitor              | A         |         500 | NULL     | NULL   |      | BTREE      |         |               |
+----------------------+------------+------------------------------+--------------+------------------------+-----------+-------------+----------+--------+------+------------+---------+---------------+

我准备了两个问题：

SELECT 
    COUNT(`idvisit`) AS `visits_count`, 
    DATE(`visit_time`) AS `date` 
FROM (
    SELECT * 
    FROM 
        `visits_table` 
    WHERE 
        `idsite` = 2 
        AND `visit_time` >= '2015-04-01 00:00:00' 
        AND `visit_time` <= '2015-04-30 23:59:59'
) AS `visits`
WHERE 1 
GROUP BY 
    DATE(`visit_time`); 



+----+-------------+----------------------+------+----------------------------------------------+------+---------+------+------+---------------------------------+
| id | select_type | table                | type | possible_keys                                | key  | key_len | ref  | rows | Extra                           |
+----+-------------+----------------------+------+----------------------------------------------+------+---------+------+------+---------------------------------+
|  1 | PRIMARY     | <derived2>           | ALL  | NULL                                         | NULL | NULL    | NULL | 1469 | Using temporary; Using filesort |
|  2 | DERIVED     | visits_table         | ALL  | index_idsite_datetime,index_idsite_idvisitor | NULL | NULL    | NULL | 1502 | Using where                     |
+----+-------------+----------------------+------+----------------------------------------------+------+---------+------+------+---------------------------------+

在MySQL 5.6的第2行中type = ref，key = index_idsite_datetime，key_len = 4，ref = const，Extra = Using index

    SELECT 
        COUNT(`idvisit`) AS `visits_count`, 
        DATE(`visit_time`) AS `date` 
    FROM 
        `visits_table` 
    WHERE 
        `idsite` = 2 
        AND `visit_time` >= '2015-04-01 00:00:00' 
        AND `visit_time` <= '2015-04-30 23:59:59'
    GROUP BY 
        DATE(`visit_time`);

+----+-------------+----------------------+-------+----------------------------------------------+-----------------------+---------+------+------+-----------------------------------------------------------+
| id | select_type | table                | type  | possible_keys                                | key                   | key_len | ref  | rows | Extra                                                     |
+----+-------------+----------------------+-------+----------------------------------------------+-----------------------+---------+------+------+-----------------------------------------------------------+
|  1 | SIMPLE      | visits_table         | range | index_idsite_datetime,index_idsite_idvisitor | index_idsite_datetime | 12      | NULL | 1468 | Using where; Using index; Using temporary; Using filesort |
+----+-------------+----------------------+-------+----------------------------------------------+-----------------------+---------+------+------+-----------------------------------------------------------+

我有86M行的表，两个查询大约需要2个小时才能执行。我可以做些什么来加速这些查询？

Answer 1

我建议将查询编写为：

SELECT COUNT(*) AS `visits_count`, 
       DATE(`visit_time`) AS `date` 
FROM  `visits_table` 
WHERE  `idsite` = 2 AND
       `visit_time` >= '2015-04-01' AND
       `visit_time` < '2015-05-01'
GROUP BY DATE(`visit_time`);

这可能会节省大量时间，因为索引现在是覆盖索引。

我认为改进查询的一种方法是摆脱group by。尝试这样的查询：

select dte,
       (select count(*)
        from visits_table
        where idsite = 2 and
              visit_time >= dates.dte AND visit_time < dates.dte + interval 1 day
from (select date('2015-04-01') as dte union all
      select date('2015-04-02') as dte
     ) dates;

MySQL使用相关子查询的索引比使用索引进行聚合要好得多。这种方法的缺点是时间会随着结果集中的天数线性增加。

是否可以再优化这些查询？

1 个答案: