關於cursor_sharing = similar(ZT)

關於cursor_sharing = similar

摘要：本文透過簡單實驗來嘗試說明cursor_sharing=similar的含義。

[@more@]

1.1. 實驗現象

我們先看看在表沒有分析無統計資料情況下的表現

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 (硬分析次數)

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發生變化依然沒有硬解析)

parse count (failures) 80

我們再來看分析表和欄位資訊後的表現

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 (重複執行沒發生變化)

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 (當object_id變化的時候產生硬分析)

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. 結論

可以看出若存在object_id的 histograms ,則每次是不同的值的時候都產生硬解析 ,若不存在 histograms，則不產生硬解析。換句話說，當表的欄位被分析過存在histograms的時候，similar 的表現和exact一樣，當表的欄位沒被分析，不存在histograms的時候，similar的表現和force一樣。這樣避免了一味地如force一樣轉換成變數形式，因為有histograms的情況下轉換成變數之後就容易產生錯誤的執行計劃，沒有利用上統計資訊。而exact呢，在沒有histograms的情況下也要分別產生硬解析，這樣的話，由於執行計劃不會受到資料分佈的影響（因為沒有統計資訊）重新解析是沒有實質意義的。而similar則綜合了兩者的優點。