关于cursor_sharing = similar

摘要：本文通过简单实验来尝试说明cursor_sharing=similar的含义。

1.1. 实验现象

我们先看看在表没有分析无统计数据情况下的表现

SQL> alter session set cursor_sharing = similar;

Session altered.

SQL> select name,value from v$sysstat where name like @#%parse%@#;

NAME VALUE

---------------------------------------------------------------- ----------

parse time cpu 4948

parse time elapsed 4468

parse count (total) 170148

parse count (hard) 1619 (硬分析次数)

parse count (failures) 80

SQL> select count(*) from t where object_id = 1000;

COUNT(*)

----------

0

SQL> select name,value from v$sysstat where name like @#%parse%@#;

NAME VALUE

parse time cpu 4948

parse time elapsed 4468

parse count (total) 170172

parse count (hard) 1620

parse count (failures) 80

SQL> /

NAME VALUE

---------------------------------------------------------------- ----------

parse time cpu 4948

parse time elapsed 4468

parse count (total) 170176

parse count (hard) 1620

parse count (failures) 80

SQL> select count(*) from t where object_id = 1000;

COUNT(*)

----------

0

SQL> select name,value from v$sysstat where name like @#%parse%@#;

NAME VALUE

---------------------------------------------------------------- ----------

parse time cpu 4948

parse time elapsed 4468

parse count (total) 170178

parse count (hard) 1620

parse count (failures) 80

SQL> select count(*) from t where object_id = 1001;

COUNT(*)

----------

0

SQL> select name,value from v$sysstat where name like @#%parse%@#;

NAME VALUE

---------------------------------------------------------------- ----------

parse time cpu 4948

parse time elapsed 4468

parse count (total) 170180

parse count (hard) 1620(即使object_id发生变化依然没有硬解析)

parse count (failures) 80

我们再来看分析表和字段信息后的表现

SQL> analyze table t1 compute statistics for table for columns object_id;

Table analyzed.

SQL> select name,value from v$sysstat where name like @#%parse%@#;

NAME VALUE

---------------------------------------------------------------- ----------

parse time cpu 4973

parse time elapsed 4495

parse count (total) 170982

parse count (hard) 1640

parse count (failures) 80

SQL> select count(*) from t1 where object_id = 5000;

COUNT(*)

----------

0

SQL> select name,value from v$sysstat where name like @#%parse%@#;

NAME VALUE

---------------------------------------------------------------- ----------

parse time cpu 4973

parse time elapsed 4495

parse count (total) 170984

parse count (hard) 1641

parse count (failures) 80

SQL> select count(*) from t1 where object_id = 5000;

COUNT(*)

----------

0

SQL> select name,value from v$sysstat where name like @#%parse%@#;

NAME VALUE

---------------------------------------------------------------- ----------

parse time cpu 4973

parse time elapsed 4495

parse count (total) 171008

parse count (hard) 1641 (重复执行没发生变化)

parse count (failures) 80

SQL> select count(*) from t1 where object_id = 5001;

COUNT(*)

----------

0

SQL> select name,value from v$sysstat where name like @#%parse%@#;

NAME VALUE

---------------------------------------------------------------- ----------

parse time cpu 4973

parse time elapsed 4495

parse count (total) 171010

parse count (hard) 1642 (当object_id变化的时候产生硬分析)

parse count (failures) 80

SQL>

SQL> select sql_text,child_number from v$sql where sql_text like @#select count(*) from t1 where%@#;

SQL_TEXT

------------------------------------------------------------------------------

CHILD_NUMBER

------------

select count(*) from t1 where object_id = :"SYS_B_0"

0

select count(*) from t1 where object_id = :"SYS_B_0"

1

1.2. 结论

可以看出若存在object_id的 histograms ,则每次是不同的值的时候都产生硬解析 ,若不存在 histograms，则不产生硬解析。换句话说，当表的字段被分析过存在histograms的时候，similar 的表现和exact一样，当表的字段没被分析，不存在histograms的时候，similar的表现和force一样。这样避免了一味地如force一样转换成变量形式，因为有histograms的情况下转换成变量之后就容易产生错误的执行计划，没有利用上统计信息。而exact呢，在没有histograms的情况下也要分别产生硬解析，这样的话，由于执行计划不会受到数据分布的影响（因为没有统计信息）重新解析是没有实质意义的。而similar则综合了两者的优点。