Count(*)一个“空表”为什么会很慢呢？只有几十行数据为什么select xxx from table limit 1需要300ms？select min(pk) fromtable也会慢？
见下例：
dba@localhost : test 18:14:32> show create table test_hmw\G
*************************** 1. row ***************************
Table: test_hmw
Create Table: CREATE TABLE `test_hmw` (
`mail` varchar(1024)DEFAULT NULL,
`user_id` bigint(20)DEFAULT NULL,
`id` int(10) unsigned NOTNULL AUTO_INCREMENT,
PRIMARY KEY (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=3997636 DEFAULT CHARSET=utf8
1 row in set (0.00 sec)
dba@localhost : test 18:14:45> select count(*) from test_hmw;
+----------+
| count(*) |
+----------+
| 3813472 |
+----------+
1 row in set (0.69 sec)
dba@localhost : test 18:14:56> show status like '%last_query%';
+-----------------+---------------+
| Variable_name |Value |
+-----------------+---------------+
| Last_query_cost | 776677.599000 |
+-----------------+---------------+
1 row in set (0.00 sec)
dba@localhost : test 18:15:14> delete from test_hmw;
Query OK, 3813472 rows affected (15.11 sec)
dba@localhost : test 18:15:56> select count(*) fromtest_hmw;
+----------+
| count(*) |
+----------+
| 0 |
+----------+
1 row in set (0.31 sec)
dba@localhost : test 18:15:59> show status like '%last_query%';
+-----------------+---------------+
| Variable_name |Value |
+-----------------+---------------+
| Last_query_cost | 757879.799000 |
+-----------------+---------------+
1 row in set (0.00 sec)
我们发现count一个空表的cost和删除前是一样的（时间不一样是因为缓存）；其实在Oracle里也存在这个现象，对于select xx from table where rownum<xxx或者select count(*)之类的操作是从起始的block扫描到HMW；而在MySQL里不是HMW，MySQL的多版本是直接存储在表里的，因此MySQL在删除了大量的数据，而purge thread回收前进行上面的操作就会扫描大量的“空”page，因此也就有了这种现象；
既然知道原因在MySQL可以通过重整表的方式再来看看cost:
dba@localhost : test 18:16:01> optimize table test_hmw;
2 rows in set (1.85 sec)
dba@localhost : test 18:16:14> select count(*) fromtest_hmw;
+----------+
| count(*) |
+----------+
| 0 |
+----------+
1 row in set (0.00 sec)
dba@localhost : test 18:16:16> show status like '%last_query%';
+-----------------+----------+
| Variable_name |Value |
+-----------------+----------+
| Last_query_cost | 1.199000 |
+-----------------+----------+
1 row in set (0.00 sec)
本文转自MIKE老毕 51CTO博客，原文链接：http://blog.51cto.com/boylook/1299866，如需转载请自行联系原作者
