非中断业务数据库迁移一例

下面是某业务库的迁移过程, 从Source Master/Slave集群迁移至Target MHA集群.

成都创新互联坚持“要么做到，要么别承诺”的工作理念，服务领域包括：成都网站制作、做网站、企业官网、英文网站、手机端网站、网站推广等服务，满足客户于互联网时代的鄯善网站设计、移动媒体设计的需求，帮助企业找到有效的互联网解决方案。努力成为您成熟可靠的网络建设合作伙伴！

前期准备, 配置数据从Source Master复制到Target Master, 即Target Master成为Source Master的Slave(若Target Master是新库的话, 还可添加replicate-do-db=name过滤条件).

迁移开始时, 在Source Master上执行flush no_write_to_binlog tables, 和flush tables with read lock.

在Target Master上查看延时为0时, 执行flush binary logs. 然后stop slave, 和reset slave all, 清除Source Master至Target Master的复制关系. 

再后, 依次重启各应用节点, 切换至Target Master新数据源.

最后在Source Master上unlock tables, 至此迁移结束.

该过程中的主角flush tables with read lock, 其含义为Closes all open tables and locks all tables for all databases with a global read lock, 即在MySQL数据库Server层, 获取一个全局读锁.

那么配角flush no_write_to_binlog tables, 其含义为Closes all open tables, forces all tables in use to be closed, 起到加速flush tables with read lock操作完成的作用.

在数据库管理的过程中, 经常会见到flush tables with read lock的身影.

如mysqldump备份时, 通过general log会看到如下输出.

mysqldump --login-path=mytest --default-character-set=utf8mb4 --single-transaction --routines --triggers --events --quick --force --master-data=2 --hex-blob --all-databases > full.sql

25 Query FLUSH /*!40101 LOCAL */ TABLES

25 Query FLUSH TABLES WITH READ LOCK

25 Query SET SESSION TRANSACTION ISOLATION LEVEL REPEATABLE READ

25 Query START TRANSACTION /*!40100 WITH CONSISTENT SNAPSHOT */

25 Query SHOW VARIABLES LIKE 'gtid\_mode'

25 Query SELECT @@GLOBAL.GTID_EXECUTED

25 Query SHOW MASTER STATUS

25 Query UNLOCK TABLES

又如xtrabackup备份时, 也可见类似日志.

xtrabackup --defaults-file=/etc/my.cnf --user=abc --password=xyz --socket=/3306/mysql.sock --target-dir=./pxb_full --backup

34 QuerySET SESSION lock_wait_timeout=31536000

34 QueryFLUSH NO_WRITE_TO_BINLOG TABLES

34 QueryFLUSH TABLES WITH READ LOCK

34 QuerySHOW MASTER STATUS

34 QuerySHOW VARIABLES

34 QueryFLUSH NO_WRITE_TO_BINLOG ENGINE LOGS

34 QueryUNLOCK TABLES

MHA在线手动切换时, 也用到了flush tables with read lock, 详见如下输出.

masterha_master_switch --conf=/etc/masterha/app2/app2.cnf --master_state=alive --orig_master_is_new_slave

...

It is better to execute FLUSH NO_WRITE_TO_BINLOG TABLES on the master before switching. Is it ok to execute on 192.168.4.33(192.168.4.33:3307)? (YES/no): yes

Wed Feb 28 21:58:41 2018 - [info] Executing FLUSH NO_WRITE_TO_BINLOG TABLES. This may take long time..

Wed Feb 28 21:58:41 2018 - [info]  ok.

...

Wed Feb 28 21:58:43 2018 - [info] Locking all tables on the orig master to reject updates from everybody (including root):

Wed Feb 28 21:58:43 2018 - [info] Executing FLUSH TABLES WITH READ LOCK..

Wed Feb 28 21:58:43 2018 - [info]  ok.