percona-tools工具在线DDL的使用

很久没有写博客了，今天发表一篇吧：）

一般的在线进行表字段的增减都会形成表所，若是表较小还能接受，若是过大则这个锁持续的时间会让人比较烦恼，对业务持续性影响较大。

Percona 提供了一款关于MySQL管理的工具集很强大，包括了在线DDL工具 pt-online-schema-change

安装就不介绍了，自行百度吧

下面的原理介绍摘自网络http://www.it165.net/pro/html/201312/8928.html，写的比较清晰，就不重复造轮子了。

+++++++++++++++++++++

模仿MySQL的alter，但不一样的是在alter操做更改表结构的时候不用锁定表，也就是说执行alter的时候不会阻塞写和读取操做，客户端能够继续都和修改数据。注意执行这个工具的时候必须作好备份，操做以前最好详细读一下官方文档。

一、若是存在外键，根据alter-foreign-keys-method参数的值，检测外键相关的表，作相应设置的处理。没有使用 --alter-foreign-keys-method 指定特定的值，该工具不予执行

二、建立一个新的表，表结构为修改后的数据表，用于从源数据表向新表中导入数据。

三、建立触发器，用于记录从拷贝数据开始以后，对源数据表继续进行数据修改的操做记录下来，用于数据拷贝结束后，执行这些操做，保证数据不会丢失。若是表中已经定义了触发器这个工具就不能工做了。

四、拷贝数据，从源数据表中拷贝数据到新表中。

五、修改外键相关的子表，根据修改后的数据，修改外键关联的子表。

六、rename源数据表为old表，把新表rename为源表名，并将old表删除。

七、删除触发器。

+++++++++++++++++++

相关的参数请 --help

1. 进行--dry-run测试

[root@master ~]#  pt-online-schema-change  --user=root --password=root --host=localhost  --alter='ENGINE=Innodb' D=baby,t=baby_account --dry-run

Operation, tries, wait:

  analyze_table, 10, 1

  copy_rows, 10, 0.25

  create_triggers, 10, 1

  drop_triggers, 10, 1

  swap_tables, 10, 1

  update_foreign_keys, 10, 1

Child tables:

  `baby`.`baby_account_billing_contract` (approx. 21 rows)

  `baby`.`baby_account_tactics` (approx. 13308 rows)

  `baby`.`baby_account_tactics_bak` (approx. 1672 rows)

  `baby`.`baby_billing` (approx. 11856 rows)

  `baby`.`baby_billing_apply` (approx. 128 rows)

  `baby`.`baby_order_bak_haochongpei` (approx. 1 rows)

You did not specify --alter-foreign-keys-method, but there are foreign keys that reference the table. Please read the tool's documentation carefully.

##表有主键，对主键的处理方式须要明确指出

--alter-foreign-keys-method 说明：

如何把外键引用到新表?须要特殊处理带有外键约束的表,以保证它们能够应用到新表.当重命名表的时候,外键关系会带到重命名后的表上。

该工具备两种方法,能够自动找到子表,并修改约束关系。

auto 在rebuild_constraints和drop_swap两种处理方式中选择一个

rebuild_constraints使用 ALTER TABLE语句先删除外键约束,而后再添加.若是子表很大的话,会致使长时间的阻塞。

drop_swap 执行FOREIGN_KEY_CHECKS=0,禁止外键约束,删除原表,再重命名新表。这种方式很快,也不会产生阻塞,可是有风险：

1, 在删除原表和重命名新表的短期内,表是不存在的,程序会返回错误。

2, 若是重命名表出现错误,也不能回滚了.由于原表已经被删除。

none 相似"drop_swap"的处理方式,可是它不删除原表,而且外键关系会随着重命名转到老表上面。

我这里指定的是auto自动选择，能够看到下文的提示

[root@master ~]#  pt-online-schema-change  --user=root --password=root --host=localhost  --alter='ENGINE=Innodb' D=baby,t=baby_account --alter-foreign-keys-method=auto --dry-run

Operation, tries, wait:

  analyze_table, 10, 1

  copy_rows, 10, 0.25

  create_triggers, 10, 1

  drop_triggers, 10, 1

  swap_tables, 10, 1

  update_foreign_keys, 10, 1

Child tables:

  `baby`.`baby_account_billing_contract` (approx. 21 rows)

  `baby`.`baby_account_tactics` (approx. 13308 rows)

  `baby`.`baby_account_tactics_bak` (approx. 1672 rows)

  `baby`.`baby_billing` (approx. 11856 rows)

  `baby`.`baby_billing_apply` (approx. 128 rows)

  `baby`.`baby_order_bak_haochongpei` (approx. 1 rows)

Will automatically choose the method to update foreign keys.

Starting a dry run.  `baby`.`baby_account` will not be altered.  Specify --execute instead of --dry-run to alter the table.

Creating new table...

Created new table baby._baby_account_new OK.

Altering new table...

Altered `baby`.`_baby_account_new` OK.

Not creating triggers because this is a dry run.

Not copying rows because this is a dry run.

Not determining the method to update foreign keys because this is a dry run.

Not swapping tables because this is a dry run.

Not updating foreign key constraints because this is a dry run.

Not dropping old table because this is a dry run.

Not dropping triggers because this is a dry run.

2016-08-29T14:31:59 Dropping new table...

2016-08-29T14:31:59 Dropped new table OK.

Dry run complete.  `baby`.`baby_account` was not altered.

执行操做，看到报错，由于我这环境是主从复制的环境，因此--check-replication-filters参数须要明确指出

[root@master ~]#  pt-online-schema-change  --user=root --password=root --host=localhost  --alter='ENGINE=Innodb' D=baby,t=baby_account --alter-foreign-keys-method=auto --execute

Found 2 slaves:

  client.wboy.com

  client1.wboy.com

Will check slave lag on:

  client.wboy.com

  client1.wboy.com

Replication filters are set on these hosts:

  client.wboy.com

    binlog_ignore_db = mysql,test,information_schema,performance_schema

  client1.wboy.com

    binlog_ignore_db = mysql,test,information_schema,performance_schema

Please read the --check-replication-filters documentation to learn how to solve this problem. at /usr/local/bin/pt-online-schema-change line 8493.

[root@master ~]#  pt-online-schema-change  --user=root --password=root --host=localhost  --alter='ENGINE=Innodb' D=baby,t=baby_account --alter-foreign-keys-method=auto --no-check-replication-filters --execute

Found 2 slaves:

  client.wboy.com

  client1.wboy.com

Will check slave lag on:

  client.wboy.com

  client1.wboy.com

Operation, tries, wait:

  analyze_table, 10, 1

  copy_rows, 10, 0.25

  create_triggers, 10, 1

  drop_triggers, 10, 1

  swap_tables, 10, 1

  update_foreign_keys, 10, 1

Child tables:

  `baby`.`baby_account_billing_contract` (approx. 21 rows)

  `baby`.`baby_account_tactics` (approx. 13308 rows)

  `baby`.`baby_account_tactics_bak` (approx. 1672 rows)

  `baby`.`baby_billing` (approx. 11856 rows)

  `baby`.`baby_billing_apply` (approx. 128 rows)

  `baby`.`baby_order_bak_haochongpei` (approx. 1 rows)

Will automatically choose the method to update foreign keys.

Altering `baby`.`baby_account`...

Creating new table...

Created new table baby._baby_account_new OK.

Waiting forever for new table `baby`.`_baby_account_new` to replicate to client1.wboy.com...

Waiting for client1.wboy.com:   0% 00:00 remain

Waiting for client1.wboy.com:   0% 00:00 remain

Altering new table...

Altered `baby`.`_baby_account_new` OK.

2016-08-29T14:54:50 Creating triggers...

2016-08-29T14:54:50 Created triggers OK.

2016-08-29T14:54:50 Copying approximately 20189 rows...

Replica lag is 32 seconds on client1.wboy.com.  Waiting.

Replica lag is 60 seconds on client1.wboy.com.  Waiting.

Copying `baby`.`baby_account`:  13% 06:43 remain

Replica lag is 32 seconds on client1.wboy.com.  Waiting.

Replica lag is 60 seconds on client1.wboy.com.  Waiting.

Copying `baby`.`baby_account`:  58% 01:33 remain

Replica lag is 31 seconds on client1.wboy.com.  Waiting.

Replica lag is 59 seconds on client1.wboy.com.  Waiting.

Copying `baby`.`baby_account`:  99% 00:01 remain

2016-08-29T14:58:04 Copied rows OK.

2016-08-29T14:58:04 Max rows for the rebuild_constraints method: 6168

Determining the method to update foreign keys...

2016-08-29T14:58:04   `baby`.`baby_account_billing_contract`: 21 rows; can use rebuild_constraints

2016-08-29T14:58:04   `baby`.`baby_account_tactics`: too many rows: 13308; must use drop_swap

2016-08-29T14:58:04 Drop-swapping tables...

2016-08-29T14:58:04 Analyzing new table...

2016-08-29T14:58:05 Dropped and swapped tables OK.

Not dropping old table because --no-drop-old-table was specified.

2016-08-29T14:58:05 Dropping triggers...

2016-08-29T14:58:05 Dropped triggers OK.

Successfully altered `baby`.`baby_account`.