永远不会使用以 DATETIME 作为复合键第一部分的主键索引

Question

我在将日期时间（甚至日期）作为主键的第一部分进行索引时遇到问题。

我使用 MySQL 5.5

这是我的两个表：

-- This is my standard table with dateDim as a dateTime

CREATE TABLE `stats` (
 `dateDim` datetime NOT NULL,
 `accountDim` mediumint(8) unsigned NOT NULL,
 `execCodeDim` smallint(5) unsigned NOT NULL,
 `operationTypeDim` tinyint(3) unsigned NOT NULL,
 `junkDim` tinyint(3) unsigned NOT NULL,
 `ipCountryDim` smallint(5) unsigned NOT NULL,
 `count` int(10) unsigned NOT NULL,
 `amount` bigint(20) NOT NULL,
 PRIMARY KEY (`dateDim`,`accountDim`,`execCodeDim`,`operationTypeDim`,`junkDim`,`ipCountryDim`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8


-- Here is a copy with datDim as an integer

CREATE TABLE `stats_todays` (
`dateDim` int(11) unsigned NOT NULL,
 `accountDim` mediumint(8) unsigned NOT NULL,
 `execCodeDim` smallint(5) unsigned NOT NULL,
 `operationTypeDim` tinyint(3) unsigned NOT NULL,
 `junkDim` tinyint(3) unsigned NOT NULL,
 `ipCountryDim` smallint(5) unsigned NOT NULL,
 `count` int(10) unsigned NOT NULL,
 `amount` bigint(20) NOT NULL,
 PRIMARY KEY (`dateDim`,`accountDim`,`execCodeDim`,`operationTypeDim`,`junkDim`,`ipCountryDim`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8

我用完全相同的数据填充两个表（接近 10 000 000）

但：

• 统计表对 dateDim 使用 DATETIME
• stats_todays 将 un INTEGER 与 TO_DAYS() 用于 dateDim

我的问题是：当索引的第一部分是日期时间时，为什么 MySQL 不使用 PRIMARY KEY ？？？这是非常奇怪的，因为使用相同的数据但用 INTEGER 和 TO_DAYS(dateDim) 合并相同的请求会晃动......

带有统计表（和日期时间）的示例：

SELECT * 
FROM `stats`  
WHERE 
   dateDim = '2014-04-03 00:00:00' 
   AND accountDim = 4
   AND execCodeDim = 9
   AND operationTypeDim = 1
   AND junkDim = 5
   AND ipCountryDim = 3

=> 1 result (4.5sec)

Explain:

id  select_type     table   type    possible_keys   key     key_len     ref     rows           Extra
1   SIMPLE          stats   ALL           NULL     NULL       NULL      NULL    8832329     Using where

对另一个表 stats_todays 的相同请求（使用 INTEGER 和 TO_DAYS() ）

EXPLAIN SELECT * 
FROM `stats_todays`  
WHERE 
   dateDim = TO_DAYS('2014-04-03 00:00:00')
   AND accountDim = 4
   AND execCodeDim = 9
   AND operationTypeDim = 1
   AND junkDim = 5
   AND ipCountryDim = 3

=> Result 1 row (0.0003 sec) 

Explain:

id  select_type     table          type     possible_keys   key     key_len     ref                               rows  Extra
1   SIMPLE         stats_todays     const   PRIMARY     PRIMARY     13  const,const,const,const,const,const     1    

如果您阅读了完整的帖子，您就会明白这不是一个低基数问题，因为请求使用与 INTEGER dateDim 字段完全相同的基数....

以下是一些高级细节：

SELECT COUNT( DISTINCT dateDim )
FROM stats_todays
UNION ALL
SELECT COUNT( DISTINCT dateDim )
FROM stats;

Result:


COUNT(DISTINCT dateDim)
2192
2192

这是索引描述：

SHOW INDEXES FROM `stats` 

Table   Non_unique  Key_name    Seq_in_index    Column_name     Collation   Cardinality     Sub_part    Packed  Null    Index_type  Comment     Index_comment
stats   0            PRIMARY          1         dateDim           A     6921           NULL                 NULL        BTREE        
stats   0            PRIMARY          2         accountDim        A     883232         NULL                 NULL        BTREE        
stats   0            PRIMARY          3         execCodeDim       A     8832329     NULL                NULL        BTREE        
stats   0            PRIMARY          4         operationTypeDim  A     8832329     NULL                NULL        BTREE        
stats   0            PRIMARY          5         junkDim           A     8832329     NULL                NULL        BTREE        
stats   0            PRIMARY          6         ipCountryDim      A     8832329     NULL                NULL        BTREE       

SHOW INDEXES FROM `stats_todays` 

Table   Non_unique  Key_name    Seq_in_index    Column_name     Collation   Cardinality     Sub_part    Packed  Null    Index_type  Comment     Index_comment
stats_todays    0   PRIMARY     1              dateDim              A        7518   NULL                   NULL         BTREE        
stats_todays    0   PRIMARY     2              accountDim           A        4022582    NULL                   NULL         BTREE        
stats_todays    0   PRIMARY     3              execCodeDim          A        8045164    NULL                   NULL         BTREE        
stats_todays    0   PRIMARY     4              operationTypeDim     A        8045164    NULL                   NULL         BTREE        
stats_todays    0   PRIMARY     5              junkDim              A        8045164    NULL                   NULL         BTREE        
stats_todays    0   PRIMARY     6              ipCountryDim         A        8045164    NULL                   NULL         BTREE        

SELECT dateDim, COUNT(*) FROM stats GROUP BY dateDim WITH ROLLUP

• 告诉有 2192 个不同的日期，并且重新分区是平滑的（按日期大约 3000 - 4000 行）
• 表中有 8 831 990 行
• 另一张桌子也一样
• 我尝试使用 COVERING INDEX（用所有 PK 列替换 *）=> 没有任何改变
• 我试过 force|use index => 没有任何改变
• 与日期字段相同，而不是日期时间
• 同 INDEX 或 UNIQUE 而不是主键

Answer 1

这是 5.5.x 中的错误。看这里

这表明您的查询应该是

SELECT * 
FROM `stats`  
WHERE 
   dateDim = CAST('2014-04-03 00:00:00' as datetime)
   AND accountDim = 4
   AND execCodeDim = 9
   AND operationTypeDim = 1
   AND junkDim = 5
   AND ipCountryDim = 3