回收InnoDB表空间

以下论述均假定innodb_file_per_table开启

先用常规optimize回收：

 

mysql> select count(*) from t;

+----------+

| count(*) |

+----------+

|  1200096 |

+----------+



[mysql@even employees]$ ls -alh t.ibd

-rw-rw---- 1 mysql dba 72M 10-08 17:39 t.ibd



mysql> delete from t limit 800000;



mysql> select count(*) from t;

+----------+

| count(*) |

+----------+

|   400096 |

+----------+



[mysql@even employees]$ ls -alh t.ibd

-rw-rw---- 1 mysql dba 72M 10-08 17:41 t.ibd





mysql> optimize table t;

+-------------+----------+----------+-------------------------------------------------------------------+

| Table       | Op       | Msg_type | Msg_text                                                          |

+-------------+----------+----------+-------------------------------------------------------------------+

| employees.t | optimize | note     | Table does not support optimize, doing recreate + analyze instead |

| employees.t | optimize | status   | OK                                                                |

+-------------+----------+----------+-------------------------------------------------------------------+



[mysql@even employees]$ ls -alh t.ibd

-rw-rw---- 1 mysql dba 29M 10-08 17:43 t.ibd

来看下这种方法的缺陷

 

 

mysql> show processlist;

+----+------+-----------+-----------+---------+------+-------------------+------------------+

| Id | User | Host      | db        | Command | Time | State             | Info             |

+----+------+-----------+-----------+---------+------+-------------------+------------------+

|  5 | root | localhost | NULL      | Query   |    0 | NULL              | show processlist |

|  6 | root | localhost | employees | Query   |   64 | copy to tmp table | optimize table t |

+----+------+-----------+-----------+---------+------+-------------------+------------------+

mysqladmin debug结果如下

 

 

Thread database.table_name          Locked/Waiting        Lock_type



6       employees.t                 Locked - read         Read lock without concurrent inserts

用optimize缺点显而易见：读锁，特别当你有大表时性能恐怕会深受影响
这里推荐使用percona公司：pt-online-schema-change，避免加锁

 

 

$ pt-online-schema-change -uroot -poracle --alter "ENGINE=InnoDB" D=employees,t=t --execute

 

执行pt工具时有可能发生的类似错误：

 

Cannot chunk the original table  There is no good index and the table is oversized

这是因为被作用的表需要含有主键或者唯一索引，这或许也能成为这款工具的小bug吧

 

 

参考文章：
http://hidba.org/?p=795
http://www.mysqlperformanceblog.com/2013/09/25/how-to-reclaim-space-in-innodb-when-innodb_file_per_table-is-on/

 

By 迦夜

2013-10-8

All life is a game of luck！