MySQL分区如何迁移

ゝ一世哀愁。 2022-03-10 13:58 205阅读 0赞

### **| 背景** ###

**需求来源**

MySQL越来越流行，而且存储在MySQL的数据量也越来越大，单表数据达亿行已经是非常常见的现象，而这些表里面保存了大量的历史记录，严重影响SQL执行的效率。本文是针对客户需求，迁移MySQL Innodb大表分区中部分历史归档分区到其他实例或者其他库表，而且迁移过程尽量减少对业务环境的影响。

**环境介绍**

*  MySQL 5.7.21
 *  Centos 7.4
 *  innodb\_file\_per\_table=1

### **| MySQL常用的Innodb迁移方法** ###

*  MySQL Enterprise Backup（物理备份，类似于xtrabackup）
 *  Copying Data Files (冷备份)
 *  逻辑导出和导入(mysqldump,mydumper,mysqlpump)
 *  可传输的表空间

### **| 迁移方案（可传输的表空间）** ###

**准备工作**

*  MySQL版本必须是5.7
 *  迁移过程中存在短暂时间内业务不可写，建议提前做好准备

**操作步骤**

查看需要迁移表（原表）结构

root@localhost : testdba 02:03:18> use test
    Database changed
    
    root@localhost : test 08:37:50> show create table sbtest2;
    +---------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
    | Table | Create Table |
    +---------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
    | sbtest2 | CREATE TABLE `sbtest2` (
    `id` int(10) DEFAULT NULL,
    `name` varchar(20) COLLATE utf8_bin DEFAULT NULL,
    `date` int(20) DEFAULT NULL,
    KEY `idx_fenqu` (`date`)
    ) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_bin
    /*!50100 PARTITION BY RANGE (date)
    (PARTITION p0 VALUES LESS THAN (20161201) ENGINE = InnoDB,
     PARTITION p1 VALUES LESS THAN (20170101) ENGINE = InnoDB,
     PARTITION p2 VALUES LESS THAN (20170201) ENGINE = InnoDB,
     PARTITION p3 VALUES LESS THAN (20170301) ENGINE = InnoDB,
     PARTITION p4 VALUES LESS THAN (20170401) ENGINE = InnoDB,
     PARTITION p5 VALUES LESS THAN (20170501) ENGINE = InnoDB,
     PARTITION p6 VALUES LESS THAN (20170601) ENGINE = InnoDB,
     PARTITION p7 VALUES LESS THAN (20170701) ENGINE = InnoDB,
     PARTITION p8 VALUES LESS THAN (20170801) ENGINE = InnoDB,
     PARTITION p9 VALUES LESS THAN (20170901) ENGINE = InnoDB,
     PARTITION p10 VALUES LESS THAN (20171001) ENGINE = InnoDB,
     PARTITION p11 VALUES LESS THAN (20171101) ENGINE = InnoDB,
     PARTITION p12 VALUES LESS THAN (20171201) ENGINE = InnoDB,
     PARTITION p13 VALUES LESS THAN (20180101) ENGINE = InnoDB,
     PARTITION p14 VALUES LESS THAN (20180201) ENGINE = InnoDB,
     PARTITION p15 VALUES LESS THAN (20180301) ENGINE = InnoDB,
     PARTITION p16 VALUES LESS THAN (20180401) ENGINE = InnoDB,
     PARTITION p17 VALUES LESS THAN (20180501) ENGINE = InnoDB,
     PARTITION p18 VALUES LESS THAN (20180601) ENGINE = InnoDB,
     PARTITION p19 VALUES LESS THAN MAXVALUE ENGINE = InnoDB) */ |
    +---------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
    1 row in set (0.00 sec)
    
    
    root@localhost : test 12:04:03> SELECT PARTITION_NAME, TABLE_ROWS FROM INFORMATION_SCHEMA.PARTITIONS WHERE TABLE_NAME = 'sbtest2';
    +----------------+------------+
    | PARTITION_NAME | TABLE_ROWS |
    +----------------+------------+
    | p0 | 22 |
    | p1 | 2 |
    | p2 | 2 |
    | p3 | 2 |
    | p4 | 2 |
    | p5 | 2 |
    | p6 | 2 |
    | p7 | 2 |
    | p8 | 2 |
    | p9 | 2 |
    | p10 | 2 |
    | p11 | 2 |
    | p12 | 2 |
    | p13 | 2 |
    | p14 | 2 |
    | p15 | 2 |
    | p16 | 2 |
    | p17 | 2 |
    | p18 | 2 |
    | p19 | 14 |
    +----------------+------------+
    20 rows in set (0.00 sec)

按照个人迁移分区表需求，可以把历史分区迁移到其他MySQL实例，也可以迁移到同一MySQL实例的其他库中。首先创建与原表相同表结构的分区表，在创建分区表时，我们只需要创建我们需要迁移的表分区结构。例：下面是迁移案例，由于只迁移2017年数据，所以表结构只创建了存储2017年数据的分区（也就是分区p2-p13）。

root@localhost : test 01:59:36> create database testdba;
    Query OK, 1 row affected (0.12 sec)
    
    root@localhost : test 01:59:44> use testdba;
    Database changed
    
    root@localhost : testdba 06:04:26> CREATE TABLE `sbtest2` (
    -> id int(10),
    -> name varchar(20),
    -> date int(20),
    -> key idx_fenqu(date)
    -> )
    -> PARTITION BY RANGE (date) (
    -> PARTITION p2 VALUES LESS THAN (20170201),
    -> PARTITION p3 VALUES LESS THAN (20170301),
    -> PARTITION p4 VALUES LESS THAN (20170401),
    -> PARTITION p5 VALUES LESS THAN (20170501),
    -> PARTITION p6 VALUES LESS THAN (20170601),
    -> PARTITION p7 VALUES LESS THAN (20170701),
    -> PARTITION p8 VALUES LESS THAN (20170801),
    -> PARTITION p9 VALUES LESS THAN (20170901),
    -> PARTITION p10 VALUES LESS THAN (20171001),
    -> PARTITION p11 VALUES LESS THAN (20171101),
    -> PARTITION p12 VALUES LESS THAN (20171201),
    -> PARTITION p13 VALUES LESS THAN (20180101)
    -> );
    Query OK, 0 rows affected (0.22 sec)

清除新表所有的分区独立表空间，为导入原表的分区独立表空间做准备

root@localhost : testdba 02:00:05> use testdba;
    Database changed
    
    root@localhost : testdba 02:00:23> ALTER TABLE sbtest2 DISCARD PARTITION p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13 TABLESPACE;
    Query OK, 0 rows affected (0.27 sec)

在原表中执行FLUSH TABLES ... FOR EXPORT（在分区表空间传输没有完成之前，不要退出该会话或者执行unlock tables;操作），用来获取元数据校验文件.cfg和确保该表的脏页刷到磁盘，并加共享表锁

root@localhost : testdba 02:00:24> USE test;
    Database changed
    
    root@localhost : test 02:00:29> FLUSH TABLES test.sbtest2 FOR EXPORT;
    Query OK, 0 rows affected (0.00 sec)
    
    [root@slave test]# cd /var/lib/mysql/data/mydata/test
    
    [root@slave test]# ls 
    db.opt sbtest2#P#p10.cfg sbtest2#P#p12.ibd sbtest2#P#p15.cfg sbtest2#P#p17.ibd sbtest2#P#p2.cfg sbtest2#P#p4.ibd sbtest2#P#p7.cfg sbtest2#P#p9.ibd
    sbtest2#P#p0.cfg sbtest2#P#p10.ibd sbtest2#P#p13.cfg sbtest2#P#p15.ibd sbtest2#P#p18.cfg sbtest2#P#p2.ibd sbtest2#P#p5.cfg sbtest2#P#p7.ibd sbtest2.frm
    sbtest2#P#p0.ibd sbtest2#P#p11.cfg sbtest2#P#p13.ibd sbtest2#P#p16.cfg sbtest2#P#p18.ibd sbtest2#P#p3.cfg sbtest2#P#p5.ibd sbtest2#P#p8.cfg
    sbtest2#P#p1.cfg sbtest2#P#p11.ibd sbtest2#P#p14.cfg sbtest2#P#p16.ibd sbtest2#P#p19.cfg sbtest2#P#p3.ibd sbtest2#P#p6.cfg sbtest2#P#p8.ibd
    sbtest2#P#p1.ibd sbtest2#P#p12.cfg sbtest2#P#p14.ibd sbtest2#P#p17.cfg sbtest2#P#p19.ibd sbtest2#P#p4.cfg sbtest2#P#p6.ibd sbtest2#P#p9.cfg

进入到原表ibd所在的目录下，把原表需要迁移的分区表空间和元数据校验文件.cfg传输到新表所在的位置，并赋予权限

[root@slave test]# cp sbtest2#P#p2.* sbtest2#P#p3.* sbtest2#P#p4.* sbtest2#P#p5.* sbtest2#P#p6.* sbtest2#P#p7.* sbtest2#P#p8.* sbtest2#P#p9.* sbtest2#P#p10.* sbtest2#P#p11.* sbtest2#P#p12.* sbtest2#P#p13.* /var/lib/mysql/data/mydata/testdba/
    
    [root@slave test]# ls ../testdba/
    db.opt sbtest2#P#p11.cfg sbtest2#P#p12.ibd sbtest2#P#p2.cfg sbtest2#P#p3.ibd sbtest2#P#p5.cfg sbtest2#P#p6.ibd sbtest2#P#p8.cfg sbtest2#P#p9.ibd
    sbtest2#P#p10.cfg sbtest2#P#p11.ibd sbtest2#P#p13.cfg sbtest2#P#p2.ibd sbtest2#P#p4.cfg sbtest2#P#p5.ibd sbtest2#P#p7.cfg sbtest2#P#p8.ibd sbtest2.frm
    sbtest2#P#p10.ibd sbtest2#P#p12.cfg sbtest2#P#p13.ibd sbtest2#P#p3.cfg sbtest2#P#p4.ibd sbtest2#P#p6.cfg sbtest2#P#p7.ibd sbtest2#P#p9.cfg
    
    [root@slave test]# chown -R mysql:mysql /var/lib/mysql

切回到执行FLUSH TABLES ... FOR EXPORT语句窗口，释放共享表锁

root@localhost : test 02:00:29> USE test;
    Database changed
    
    root@localhost : test 02:01:07> UNLOCK TABLES;
    Query OK, 0 rows affected (0.00 sec)

进入新表所在的实例或新表所在的库，手动导入分区表空间，进行数据恢复（应用传输到新表的分区表空间）

root@localhost : test 02:01:07> USE testdba;
    Database changed
    
    root@localhost : testdba 02:01:14> ALTER TABLE sbtest2 IMPORT PARTITION p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13 TABLESPACE;
    Query OK, 0 rows affected (0.62 sec)

表空间迁移完成，数据恢复完成，最后校验数据准确性

root@localhost : testdba 02:03:16> SELECT PARTITION_NAME, TABLE_ROWS FROM INFORMATION_SCHEMA.PARTITIONS WHERE TABLE_NAME = 'sbtest2' and TABLE_SCHEMA='testdba';
    +----------------+------------+
    | PARTITION_NAME | TABLE_ROWS |
    +----------------+------------+
    | p2 | 2 |
    | p3 | 2 |
    | p4 | 2 |
    | p5 | 2 |
    | p6 | 2 |
    | p7 | 2 |
    | p8 | 2 |
    | p9 | 2 |
    | p10 | 2 |
    | p11 | 2 |
    | p12 | 2 |
    | p13 | 2 |
    +----------------+------------+
    12 rows in set (0.00 sec)

### **| 总结** ###

以上是我们使用MySQL的分区表空间传输方法，解决了分区表历史数据归档到其他实例或者同一实例其他库的问题。对比逻辑迁移方式mysqldump或者insert .. select ...方式速度更快，数据立即可用，而且对业务的影响更小。

### **| 作者简介** ###

**岳雷·沃趣科技数据库工程师**

熟悉MySQL体系结构和innodb存储引擎工作原理；以及MySQL备份恢复、复制、数据迁移等技术；专注于MySQL、MariaDB开源数据库，喜好开源技术。