你的位置:首页 > 数据库

[数据库]MySQL基于时间字段进行分区的方案总结


MySQL支持的分区类型一共有四种:RANGE,LIST,HASH,KEY。其中,RANGE又可分为原生RANGE和RANGE COLUMNS,LIST分为原生LIST和LIST COLUMNS,HASH分为原生HASH和LINEAR HASH,KEY包含原生KEY和LINEAR HASH。关于这些分区之间的差别,改日另写文章进行阐述。

最近,碰到一个需求,要对表的时间字段(类型:datetime)基于天进行分区。于是遍历MySQL官方文档分区章节,总结如下:

 

实现方式

主要是以下几种:

1. 基于RANGE

2. 基于RANGE COLUMNS

3. 基于HASH

 

测试数据 

为了测试以上三种方案,特构造了100万的测试数据,放在test表中,test表只有两列:id和hiredate,其中hiredate只包含10天的数据,从2015-12-01到2015-12-10。具体信息如下:

mysql> show create table test\G*************************** 1. row ***************************    Table: testCreate Table: CREATE TABLE `test` ( `id` int(11) DEFAULT NULL, `hiredate` datetime DEFAULT NULL) ENGINE=InnoDB DEFAULT CHARSET=latin11 row in set (0.00 sec)mysql> select min(hiredate),max(hiredate) from test;+---------------------+---------------------+| min(hiredate)    | max(hiredate)    |+---------------------+---------------------+| 2015-12-01 00:00:00 | 2015-12-10 23:59:56 |+---------------------+---------------------+1 row in set (0.44 sec)mysql> select date(hiredate),count(*) from test group by date(hiredate);+----------------+----------+| date(hiredate) | count(*) |+----------------+----------+| 2015-12-01   |  99963 || 2015-12-02   |  100032 || 2015-12-03   |  100150 || 2015-12-04   |  99989 || 2015-12-05   |  99908 || 2015-12-06   |  99897 || 2015-12-07   |  100137 || 2015-12-08   |  100171 || 2015-12-09   |  99851 || 2015-12-10   |  99902 |+----------------+----------+10 rows in set (0.98 sec)

 

测试的维度

测试的维度主要从两个方面进行,

一、分区剪裁

针对特定的查询,是否能进行分区剪裁(即只查询相关的分区,而不是所有分区)

二、查询时间

鉴于该批测试数据是静止的(即没有并发进行的insert,update和delete操作),数据量也不太大,从这个维度来考量貌似意义也不是很大。

因此,重点测试第一个维度。

 

基于RANGE的分区方案

在这里,选用了TO_DAYS函数

CREATE TABLE range_datetime(  id INT,  hiredate DATETIME)PARTITION BY RANGE (TO_DAYS(hiredate) ) (  PARTITION p1 VALUES LESS THAN ( TO_DAYS('20151202') ),  PARTITION p2 VALUES LESS THAN ( TO_DAYS('20151203') ),  PARTITION p3 VALUES LESS THAN ( TO_DAYS('20151204') ),  PARTITION p4 VALUES LESS THAN ( TO_DAYS('20151205') ),  PARTITION p5 VALUES LESS THAN ( TO_DAYS('20151206') ),  PARTITION p6 VALUES LESS THAN ( TO_DAYS('20151207') ),  PARTITION p7 VALUES LESS THAN ( TO_DAYS('20151208') ),  PARTITION p8 VALUES LESS THAN ( TO_DAYS('20151209') ),  PARTITION p9 VALUES LESS THAN ( TO_DAYS('20151210') ),  PARTITION p10 VALUES LESS THAN ( TO_DAYS('20151211') ));

插入数据并查看特定查询的执行计划

mysql> insert into range_datetime select * from test;                                  Query OK, 1000000 rows affected (8.15 sec)Records: 1000000 Duplicates: 0 Warnings: 0mysql> explain partitions select * from range_datetime where hiredate >= '20151207124503' and hiredate<='20151210111230'; +----+-------------+----------------+--------------+------+---------------+------+---------+------+--------+-------------+| id | select_type | table     | partitions  | type | possible_keys | key | key_len | ref | rows  | Extra    |+----+-------------+----------------+--------------+------+---------------+------+---------+------+--------+-------------+| 1 | SIMPLE   | range_datetime | p7,p8,p9,p10 | ALL | NULL     | NULL | NULL  | NULL | 400061 | Using where |+----+-------------+----------------+--------------+------+---------------+------+---------+------+--------+-------------+1 row in set (0.03 sec)

注意执行计划中的partitions的内容,只查询了p7,p8,p9,p10三个分区,由此来看,使用to_days函数确实可以实现分区裁剪。

 

基于RANGE COLUMNS的分区方案

RANGE COLUMNS可以直接基于列,而无需像上述RANGE那种,分区的对象只能为整数。

创表语句如下:

CREATE TABLE range_columns (   id INT,  hiredate DATETIME)PARTITION BY RANGE COLUMNS(hiredate) (  PARTITION p1 VALUES LESS THAN ( '20151202' ),  PARTITION p2 VALUES LESS THAN ( '20151203' ),  PARTITION p3 VALUES LESS THAN ( '20151204' ),  PARTITION p4 VALUES LESS THAN ( '20151205' ),  PARTITION p5 VALUES LESS THAN ( '20151206' ),  PARTITION p6 VALUES LESS THAN ( '20151207' ),  PARTITION p7 VALUES LESS THAN ( '20151208' ),  PARTITION p8 VALUES LESS THAN ( '20151209' ),  PARTITION p9 VALUES LESS THAN ( '20151210' ),  PARTITION p10 VALUES LESS THAN ('20151211' ));

插入数据并查看上述查询的执行计划

mysql> insert into range_columns select * from test;                                  Query OK, 1000000 rows affected (9.20 sec)Records: 1000000 Duplicates: 0 Warnings: 0mysql> explain partitions select * from range_columns where hiredate >= '20151207124503' and hiredate<='20151210111230'; +----+-------------+---------------+--------------+------+---------------+------+---------+------+--------+-------------+| id | select_type | table     | partitions  | type | possible_keys | key | key_len | ref | rows  | Extra    |+----+-------------+---------------+--------------+------+---------------+------+---------+------+--------+-------------+| 1 | SIMPLE   | range_columns | p7,p8,p9,p10 | ALL | NULL     | NULL | NULL  | NULL | 400210 | Using where |+----+-------------+---------------+--------------+------+---------------+------+---------+------+--------+-------------+1 row in set (0.11 sec)

同样,使用该分区方案也实现了分区剪裁。

 

基于HASH的分区方案

因HASH分区对象同样只能为整数,所以我们无法像上述RANGE COLUMNS那种直接引用列,在这里,同样用了TO_DAYS函数进行转换。

创表语句如下:

CREATE TABLE hash_datetime (  id INT,  hiredate DATETIME)PARTITION BY HASH( TO_DAYS(hiredate) )PARTITIONS 10;

插入数据并查看上述查询的执行计划

mysql> insert into hash_datetime select * from test;Query OK, 1000000 rows affected (9.43 sec)Records: 1000000 Duplicates: 0 Warnings: 0mysql> explain partitions select * from hash_datetime where hiredate >= '20151207124503' and hiredate<='20151210111230';+----+-------------+---------------+-------------------------------+------+---------------+------+---------+------+---------+-------------+| id | select_type | table     | partitions          | type | possible_keys | key | key_len | ref | rows  | Extra    |+----+-------------+---------------+-------------------------------+------+---------------+------+---------+------+---------+-------------+| 1 | SIMPLE   | hash_datetime | p0,p1,p2,p3,p4,p5,p6,p7,p8,p9 | ALL | NULL     | NULL | NULL  | NULL | 1000500 | Using where |+----+-------------+---------------+-------------------------------+------+---------------+------+---------+------+---------+-------------+1 row in set (0.00 sec)

不难看出,使用hash分区并不能有效的实现分区裁剪,至少在本例,基于天的需求中如此。

 

以上三种方案都是基于datetime的,那么,对于timestamp类型,又该如何选择呢?

事实上,MySQL提供了一种基于UNIX_TIMESTAMP函数的RANGE分区方案,而且,只能使用UNIX_TIMESTAMP函数,如果使用其它函数,譬如to_days,会报如下错误:“ERROR 1486 (HY000): Constant, random or timezone-dependent expressions in (sub)partitioning function are not allowed”。

而且官方文档中也提到“Any other expressions involving TIMESTAMP values are not permitted. (See Bug #42849.)”。

下面来测试一下基于UNIX_TIMESTAMP函数的RANGE分区方案,看其能否实现分区裁剪。

 

针对TIMESTAMP的分区方案

创表语句如下:

CREATE TABLE range_timestamp (  id INT,  hiredate TIMESTAMP)PARTITION BY RANGE ( UNIX_TIMESTAMP(hiredate) ) (  PARTITION p1 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-02 00:00:00') ),  PARTITION p2 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-03 00:00:00') ),  PARTITION p3 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-04 00:00:00') ),  PARTITION p4 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-05 00:00:00') ),  PARTITION p5 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-06 00:00:00') ),  PARTITION p6 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-07 00:00:00') ),  PARTITION p7 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-08 00:00:00') ),  PARTITION p8 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-09 00:00:00') ),  PARTITION p9 VALUES LESS THAN ( UNIX_TIMESTAMP('2015-12-10 00:00:00') ),  PARTITION p10 VALUES LESS THAN (UNIX_TIMESTAMP('2015-12-11 00:00:00') ));

插入数据并查看上述查询的执行计划

mysql> insert into range_timestamp select * from test;Query OK, 1000000 rows affected (13.25 sec)Records: 1000000 Duplicates: 0 Warnings: 0mysql> explain partitions select * from range_timestamp where hiredate >= '20151207124503' and hiredate<='20151210111230';+----+-------------+-----------------+--------------+------+---------------+------+---------+------+--------+-------------+| id | select_type | table      | partitions  | type | possible_keys | key | key_len | ref | rows  | Extra    |+----+-------------+-----------------+--------------+------+---------------+------+---------+------+--------+-------------+| 1 | SIMPLE   | range_timestamp | p7,p8,p9,p10 | ALL | NULL     | NULL | NULL  | NULL | 400448 | Using where |+----+-------------+-----------------+--------------+------+---------------+------+---------+------+--------+-------------+1 row in set (0.00 sec)

同样也能实现分区裁剪。

 

总结:

1. 经过对比,个人倾向于第二种方案,即基于RANGE COLUMNS的分区实现。

2. 在5.7版本之前,对于DATA和DATETIME类型的列,如果要实现分区裁剪,只能使用YEAR() 和TO_DAYS()函数,在5.7版本中,又新增了TO_SECONDS()函数。

3. 其实LIST也能实现基于天的分区方案,但在这个需求上,相比于RANGE,还是显得很鸡肋。

4. TIMESTAMP类型的列,只能基于UNIX_TIMESTAMP函数进行分区,切记!

 

参考:

http://dev.mysql.com/doc/refman/5.7/en/partitioning.html