深入解析MySQL分区(Partition)功能

jopen 11年前

自5.1开始对分区(Partition)有支持

= 水平分区(根据列属性按行分)=</span>
举个简单例子:一个包含十年发票记录的表可以被分区为十个不同的分区,每个分区包含的是其中一年的记录。

=== 水平分区的几种模式:===
* Range(范围) – 这种模式允许DBA将数据划分不同范围。例如DBA可以将一个表通过年份划分成三个分区,80年代(1980's)的数据,90年代(1990's)的数据以及任何在2000年(包括2000年)后的数据。 

* Hash(哈希) – 这中模式允许DBA通过对表的一个或多个列的Hash Key进行计算,最后通过这个Hash码不同数值对应的数据区域进行分区,。例如DBA可以建立一个对表主键进行分区的表。 

* Key(键值) – 上面Hash模式的一种延伸,这里的Hash Key是MySQL系统产生的。 

* List(预定义列表) – 这种模式允许系统通过DBA定义的列表的值所对应的行数据进行分割。例如:DBA建立了一个横跨三个分区的表,分别根据2004年2005年和2006年值所对应的数据。 

* Composite(复合模式) - 很神秘吧,哈哈,其实是以上模式的组合使用而已,就不解释了。举例:在初始化已经进行了Range范围分区的表上,我们可以对其中一个分区再进行hash哈希分区。 

= 垂直分区(按列分)=</span>
举个简单例子:一个包含了大text和BLOB列的表,这些text和BLOB列又不经常被访问,这时候就要把这些不经常使用的text和BLOB了划分到另一个分区,在保证它们数据相关性的同时还能提高访问速度。


[分区表和未分区表试验过程]

*创建分区表,按日期的年份拆分

mysql> CREATE TABLE part_tab ( c1 int default NULL, c2 varchar(30) default NULL, c3 date default NULL) engine=myisam   PARTITION BY RANGE (year(c3)) (PARTITION p0 VALUES LESS THAN (1995),  PARTITION p1 VALUES LESS THAN (1996) , PARTITION p2 VALUES LESS THAN (1997) ,  PARTITION p3 VALUES LESS THAN (1998) , PARTITION p4 VALUES LESS THAN (1999) ,  PARTITION p5 VALUES LESS THAN (2000) , PARTITION p6 VALUES LESS THAN (2001) ,  PARTITION p7 VALUES LESS THAN (2002) , PARTITION p8 VALUES LESS THAN (2003) ,  PARTITION p9 VALUES LESS THAN (2004) , PARTITION p10 VALUES LESS THAN (2010),  PARTITION p11 VALUES LESS THAN MAXVALUE ); 
注意最后一行,考虑到可能的最大值

*创建未分区表
mysql> create table no_part_tab (c1 int(11) default NULL,c2 varchar(30) default NULL,c3 date default NULL) engine=myisam;

*通过存储过程灌入800万条测试数据

mysql> set sql_mode=''; /* 如果创建存储过程失败,则先需设置此变量, bug? */

mysql> delimiter //   /* 设定语句终结符为 //,因存储过程语句用;结束 */


mysql> CREATE PROCEDURE load_part_tab()         begin      declare v int default 0;      while v < 8000000      do          insert into part_tab          values (v,'testing partitions',adddate('1995-01-01',(rand(v)*36520) mod 3652));           set v = v + 1;      end while;      end      //  mysql> delimiter ;  mysql> call load_part_tab();
Query OK, 1 row affected (8 min 17.75 sec)
mysql> insert into no_part_tab select * from part_tab;
Query OK, 8000000 rows affected (51.59 sec)
Records: 8000000 Duplicates: 0 Warnings: 0

* 测试SQL性能

mysql> select count(*) from part_tab where c3 > date '1995-01-01' and c3 < date '1995-12-31';

+----------+
| count(*) |
+----------+
|   795181 |
+----------+

1 row in set (0.55 sec)


mysql> select count(*) from no_part_tab where c3 > date '1995-01-01' and c3 < date '1995-12-31'; 
+----------+
| count(*) |
+----------+
|   795181 |
+----------+
1 row in set (4.69 sec)
结果表明分区表比未分区表的执行时间少90%。

* 通过explain语句来分析执行情况
mysql > explain select count(*) from no_part_tab where c3 > date '1995-01-01' and c3 < date '1995-12-31'\G
/* 结尾的\G使得mysql的输出改为列模式 */                    
*************************** 1. row ***************************
           id: 1
select_type: SIMPLE
        table: no_part_tab
         type: ALL
possible_keys: NULL
          key: NULL
      key_len: NULL
          ref: NULL
         rows: 8000000
        Extra: Using where
1 row in set (0.00 sec)


mysql> explain select count(*) from part_tab where c3 > date '1995-01-01' and c3 < date '1995-12-31'\G 

*************************** 1. row ***************************
           id: 1
select_type: SIMPLE
        table: part_tab
         type: ALL
possible_keys: NULL
          key: NULL
      key_len: NULL
          ref: NULL
         rows: 798458
        Extra: Using where
1 row in set (0.00 sec)
explain语句显示了SQL查询要处理的记录数目

* 试验创建索引后情况
mysql> create index idx_of_c3 on no_part_tab (c3);
Query OK, 8000000 rows affected (1 min 18.08 sec)
Records: 8000000 Duplicates: 0 Warnings: 0

mysql> create index idx_of_c3 on part_tab (c3);
Query OK, 8000000 rows affected (1 min 19.19 sec)
Records: 8000000 Duplicates: 0 Warnings: 0
创建索引后的数据库文件大小列表:
2008-05-24 09:23             8,608 no_part_tab.frm
2008-05-24 09:24       255,999,996 no_part_tab.MYD
2008-05-24 09:24        81,611,776 no_part_tab.MYI
2008-05-24 09:25                 0 part_tab#P#p0.MYD
2008-05-24 09:26             1,024 part_tab#P#p0.MYI
2008-05-24 09:26        25,550,656 part_tab#P#p1.MYD
2008-05-24 09:26         8,148,992 part_tab#P#p1.MYI
2008-05-24 09:26        25,620,192 part_tab#P#p10.MYD
2008-05-24 09:26         8,170,496 part_tab#P#p10.MYI
2008-05-24 09:25                 0 part_tab#P#p11.MYD
2008-05-24 09:26             1,024 part_tab#P#p11.MYI
2008-05-24 09:26        25,656,512 part_tab#P#p2.MYD
2008-05-24 09:26         8,181,760 part_tab#P#p2.MYI
2008-05-24 09:26        25,586,880 part_tab#P#p3.MYD
2008-05-24 09:26         8,160,256 part_tab#P#p3.MYI
2008-05-24 09:26        25,585,696 part_tab#P#p4.MYD
2008-05-24 09:26         8,159,232 part_tab#P#p4.MYI
2008-05-24 09:26        25,585,216 part_tab#P#p5.MYD
2008-05-24 09:26         8,159,232 part_tab#P#p5.MYI
2008-05-24 09:26        25,655,740 part_tab#P#p6.MYD
2008-05-24 09:26         8,181,760 part_tab#P#p6.MYI
2008-05-24 09:26        25,586,528 part_tab#P#p7.MYD
2008-05-24 09:26         8,160,256 part_tab#P#p7.MYI
2008-05-24 09:26        25,586,752 part_tab#P#p8.MYD
2008-05-24 09:26         8,160,256 part_tab#P#p8.MYI
2008-05-24 09:26        25,585,824 part_tab#P#p9.MYD
2008-05-24 09:26         8,159,232 part_tab#P#p9.MYI
2008-05-24 09:25             8,608 part_tab.frm
2008-05-24 09:25                68 part_tab.par

* 再次测试SQL性能

mysql> select count(*) from no_part_tab where c3 > date '1995-01-01' and c3 < date '1995-12-31'; 

+----------+
| count(*) |
+----------+
|   795181 |
+----------+

1 row in set (2.42 sec)   /* 为原来4.69 sec 的51%*/   


重启mysql ( net stop mysql, net start mysql)后,查询时间降为0.89 sec,几乎与分区表相同。

mysql> select count(*) from part_tab where c3 > date '1995-01-01' and c3 < date '1995-12-31'; 
+----------+
| count(*) |
+----------+
|   795181 |
+----------+
1 row in set (0.86 sec)

* 更进一步的试验
** 增加日期范围
mysql> select count(*) from no_part_tab where c3 > date '1995-01-01' and c3 < date '1997-12-31';
+----------+
| count(*) |
+----------+
| 2396524 |
+----------+
1 row in set (5.42 sec)

mysql> select count(*) from part_tab where c3 > date '1995-01-01' and c3 < date '1997-12-31';
+----------+
| count(*) |
+----------+
| 2396524 |
+----------+

1 row in set (2.63 sec)


** 增加未索引字段查询
mysql> select count(*) from part_tab where c3 > date '1995-01-01' and c3 < date  '1996-12-31' and c2='hello';
+----------+
| count(*) |
+----------+
|        0 |
+----------+
1 row in set (0.75 sec)

mysql> select count(*) from no_part_tab where c3 > date '1995-01-01' and c3 < date '1996-12-31' and c2='hello';
+----------+
| count(*) |
+----------+
|        0 |
+----------+
1 row in set (11.52 sec)


= 初步结论 =
* 分区和未分区占用文件空间大致相同 (数据和索引文件)
* 如果查询语句中有未建立索引字段,分区时间远远优于未分区时间
* 如果查询语句中字段建立了索引,分区和未分区的差别缩小,分区略优于未分区。


= 最终结论 =
* 对于大数据量,建议使用分区功能。
* 去除不必要的字段
* 根据手册, 增加myisam_max_sort_file_size 会增加分区性能

[分区命令详解]

= 分区例子 = 
* RANGE 类型
CREATE TABLE users (         uid INT UNSIGNED NOT NULL AUTO_INCREMENT PRIMARY KEY,         name VARCHAR(30) NOT NULL DEFAULT '',         email VARCHAR(30) NOT NULL DEFAULT ''  )  PARTITION BY RANGE (uid) (         PARTITION p0 VALUES LESS THAN (3000000)         DATA DIRECTORY = '/data0/data'         INDEX DIRECTORY = '/data1/idx',           PARTITION p1 VALUES LESS THAN (6000000)         DATA DIRECTORY = '/data2/data'         INDEX DIRECTORY = '/data3/idx',           PARTITION p2 VALUES LESS THAN (9000000)         DATA DIRECTORY = '/data4/data'         INDEX DIRECTORY = '/data5/idx',           PARTITION p3 VALUES LESS THAN MAXVALUE     DATA DIRECTORY = '/data6/data'          INDEX DIRECTORY = '/data7/idx'  );
在这里,将用户表分成4个分区,以每300万条记录为界限,每个分区都有自己独立的数据、索引文件的存放目录,与此同时,这些目录所在的物理磁盘分区可能也都是完全独立的,可以提高磁盘IO吞吐量。
      
* LIST 类型
CREATE TABLE category (       cid INT UNSIGNED NOT NULL AUTO_INCREMENT PRIMARY KEY,       name VARCHAR(30) NOT NULL DEFAULT ''  )  PARTITION BY LIST (cid) (       PARTITION p0 VALUES IN (0,4,8,12)       DATA DIRECTORY = '/data0/data'        INDEX DIRECTORY = '/data1/idx',              PARTITION p1 VALUES IN (1,5,9,13)       DATA DIRECTORY = '/data2/data'       INDEX DIRECTORY = '/data3/idx',              PARTITION p2 VALUES IN (2,6,10,14)       DATA DIRECTORY = '/data4/data'       INDEX DIRECTORY = '/data5/idx',              PARTITION p3 VALUES IN (3,7,11,15)       DATA DIRECTORY = '/data6/data'       INDEX DIRECTORY = '/data7/idx'  );   
分成4个区,数据文件和索引文件单独存放。

* HASH 类型     
CREATE TABLE users (       uid INT UNSIGNED NOT NULL AUTO_INCREMENT PRIMARY KEY,       name VARCHAR(30) NOT NULL DEFAULT '',       email VARCHAR(30) NOT NULL DEFAULT ''  )  PARTITION BY HASH (uid) PARTITIONS 4 (       PARTITION p0       DATA DIRECTORY = '/data0/data'       INDEX DIRECTORY = '/data1/idx',         PARTITION p1       DATA DIRECTORY = '/data2/data'       INDEX DIRECTORY = '/data3/idx',         PARTITION p2       DATA DIRECTORY = '/data4/data'       INDEX DIRECTORY = '/data5/idx',         PARTITION p3       DATA DIRECTORY = '/data6/data'       INDEX DIRECTORY = '/data7/idx'  );
分成4个区,数据文件和索引文件单独存放。

例子:
CREATE TABLE ti2 (id INT, amount DECIMAL(7,2), tr_date DATE)      ENGINE=myisam      PARTITION BY HASH( MONTH(tr_date) )      PARTITIONS 6;    CREATE PROCEDURE load_ti2()         begin      declare v int default 0;      while v < 80000      do          insert into ti2          values (v,'3.14',adddate('1995-01-01',(rand(v)*3652) mod 365));           set v = v + 1;      end while;      end      //

 

* KEY 类型
CREATE TABLE users (       uid INT UNSIGNED NOT NULL AUTO_INCREMENT PRIMARY KEY,       name VARCHAR(30) NOT NULL DEFAULT '',       email VARCHAR(30) NOT NULL DEFAULT ''  )  PARTITION BY KEY (uid) PARTITIONS 4 (       PARTITION p0       DATA DIRECTORY = '/data0/data'       INDEX DIRECTORY = '/data1/idx',              PARTITION p1       DATA DIRECTORY = '/data2/data'        INDEX DIRECTORY = '/data3/idx',              PARTITION p2        DATA DIRECTORY = '/data4/data'       INDEX DIRECTORY = '/data5/idx',              PARTITION p3        DATA DIRECTORY = '/data6/data'       INDEX DIRECTORY = '/data7/idx'  );   
分成4个区,数据文件和索引文件单独存放。

* 子分区
子分区是针对 RANGE/LIST 类型的分区表中每个分区的再次分割。再次分割可以是 HASH/KEY 等类型。例如:
CREATE TABLE users (       uid INT UNSIGNED NOT NULL AUTO_INCREMENT PRIMARY KEY,       name VARCHAR(30) NOT NULL DEFAULT '',       email VARCHAR(30) NOT NULL DEFAULT ''  )  PARTITION BY RANGE (uid) SUBPARTITION BY HASH (uid % 4) SUBPARTITIONS 2(       PARTITION p0 VALUES LESS THAN (3000000)       DATA DIRECTORY = '/data0/data'       INDEX DIRECTORY = '/data1/idx',         PARTITION p1 VALUES LESS THAN (6000000)       DATA DIRECTORY = '/data2/data'       INDEX DIRECTORY = '/data3/idx'  );
对 RANGE 分区再次进行子分区划分,子分区采用 HASH 类型。
或者
CREATE TABLE users (       uid INT UNSIGNED NOT NULL AUTO_INCREMENT PRIMARY KEY,       name VARCHAR(30) NOT NULL DEFAULT '',       email VARCHAR(30) NOT NULL DEFAULT ''  )  PARTITION BY RANGE (uid) SUBPARTITION BY KEY(uid) SUBPARTITIONS 2(       PARTITION p0 VALUES LESS THAN (3000000)       DATA DIRECTORY = '/data0/data'       INDEX DIRECTORY = '/data1/idx',         PARTITION p1 VALUES LESS THAN (6000000)       DATA DIRECTORY = '/data2/data'       INDEX DIRECTORY = '/data3/idx'  );
对 RANGE 分区再次进行子分区划分,子分区采用 KEY 类型。

= 分区管理 =

    * 删除分区  
ALERT TABLE users DROP PARTITION p0;

      删除分区 p0。


    * 重建分区
          o RANGE 分区重建
ALTER TABLE users REORGANIZE PARTITION p0,p1 INTO (PARTITION p0 VALUES LESS THAN (6000000));
            将原来的 p0,p1 分区合并起来,放到新的 p0 分区中。
          o LIST 分区重建
ALTER TABLE users REORGANIZE PARTITION p0,p1 INTO (PARTITION p0 VALUES IN(0,1,4,5,8,9,12,13));
            将原来的 p0,p1 分区合并起来,放到新的 p0 分区中。
          o HASH/KEY 分区重建
 ALTER TABLE users REORGANIZE PARTITION COALESCE PARTITION 2;
            用 REORGANIZE 方式重建分区的数量变成2,在这里数量只能减少不能增加。想要增加可以用 ADD PARTITION 方法。
    * 新增分区
          o 新增 RANGE 分区   
 ALTER TABLE category ADD PARTITION (PARTITION p4 VALUES IN (16,17,18,19)              DATA DIRECTORY = '/data8/data'              INDEX DIRECTORY = '/data9/idx');
            新增一个RANGE分区。
          o 新增 HASH/KEY 分区
  ALTER TABLE users ADD PARTITION PARTITIONS 8;
            将分区总数扩展到8个。

[ 给已有的表加上分区 ]
alter table results partition by RANGE (month(ttime))   (PARTITION p0 VALUES LESS THAN (1),  PARTITION p1 VALUES LESS THAN (2) , PARTITION p2 VALUES LESS THAN (3) ,  PARTITION p3 VALUES LESS THAN (4) , PARTITION p4 VALUES LESS THAN (5) ,  PARTITION p5 VALUES LESS THAN (6) , PARTITION p6 VALUES LESS THAN (7) ,  PARTITION p7 VALUES LESS THAN (8) , PARTITION p8 VALUES LESS THAN (9) ,  PARTITION p9 VALUES LESS THAN (10) , PARTITION p10 VALUES LESS THAN (11),  PARTITION p11 VALUES LESS THAN (12),  PARTITION P12 VALUES LESS THAN (13) ); 


默认分区限制分区字段必须是主键(PRIMARY KEY)的一部分,为了去除此
限制:
[方法1] 使用ID
mysql> ALTER TABLE np_pk      ->     PARTITION BY HASH( TO_DAYS(added) )      ->     PARTITIONS 4;
ERROR 1503 (HY000): A PRIMARY KEY must include all columns in the table's partitioning function

However, this statement using the id column for the partitioning column is valid, as shown here:

mysql> ALTER TABLE np_pk      ->     PARTITION BY HASH(id)      ->     PARTITIONS 4;
Query OK, 0 rows affected (0.11 sec)
Records: 0 Duplicates: 0 Warnings: 0

[方法2] 将原有PK去掉生成新PK
mysql> alter table results drop PRIMARY KEY;
Query OK, 5374850 rows affected (7 min 4.05 sec)
Records: 5374850 Duplicates: 0 Warnings: 0

mysql> alter table results add PRIMARY KEY(id, ttime);
Query OK, 5374850 rows affected (6 min 14.86 sec)
Records: 5374850 Duplicates: 0 Warnings: 0