oracle 9i,10g,11g执行计划的固定一直是oracle的心病之一,虽然CBO很强了,在10g里我们用outline,引入sql_profile,11g里用sql_profile,引入sql_plan_baseline
1 相关参数
optimizer_capture_sql_plan_baselines默认值false,即不采用自动捕获
optimizer_use_sql_plan_baselines 默认值true,即优先使用dba_sql_plan_baselines里的执行计划;
2 默认参数测试
有条sql很简单:select * from hr.emp where EMPLOYEE_ID=206;
此表无索引,无主键,就是一个最普通的堆表
执行计划无悬念:全表扫苗:

sql> select* from table(dbms_xplan.display_cursor('76rrg3a06j24n',0)); 
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
sql_ID  76rrg3a06j24n,child number 0
-------------------------------------
select * from hr.emp where EMPLOYEE_ID=206
Plan hash value: 3956160932
--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%cpu)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |       |       |     2 (100)|          |
|*  1 |  TABLE ACCESS FULL| EMP  |     1 |   133 |     2   (0)| 00:00:01 |
--------------------------------------------------------------------------

建立基线:

declare
v_ret  varchar2(100);
begin
v_ret := dbms_spm.load_plans_from_cursor_cache(sql_id=>'76rrg3a06j24n');
end;

检查基线:

select sql_handle,sql_text,origin,version,enabled,accepted,fixed from dba_sql_plan_baselines
sql_HANDLE  sql_TEXT           ORIGIN                   VERSION  ENABLED    ACCEPTED    FIXED
sql_4313a2ffff308f08    <CLOB>  MANUAL-LOAD 11.2.0.4.0      YES      YES            NO

建立索引:
create index i_employee_id on hr.emp(employee_Id);
再次执行,检查child_number:

select sql_id,child_number,child_Number from v$sql where sql_text like 'select %hr.emp%';
sql_ID  CHILD_NUMBER    sql_TEXT    CHILD_NUMBER
76rrg3a06j24n   1   select * from hr.emp where EMPLOYEE_ID=206  1

查看其执行计划:

Plan hash value: 3956160932 
--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%cpu)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |     2 |   266 |     2   (0)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| EMP  |     2 |   266 |     2   (0)| 00:00:01 |
-------------------------------------------------------------------------- 
Predicate Information (identified by operation id):
--------------------------------------------------- 
   1 - filter("EMPLOYEE_ID"=206) 
Note
-----
   - sql plan baseline "sql_PLAN_464x2zzzm13s8d8a279cc" used for this statement

依旧全表扫描,这是不科学的,理论上是走索引,他给出了说明,走的是sql_PLAN_464x2zzzm13s8d8a279cc,检查dba_sql_plan_baselines

sql_HANDLE sql_TEXT    ORIGIN             VERSION  ENABLED ACCEPTED    FIXED   PLAN_NAME
sql_4313a2ffff308f08    <CLOB>  MANUAL-LOAD 11.2.0.4.0  YES YES NO  sql_PLAN_464x2zzzm13s8d8a279cc
sql_4313a2ffff308f08    <CLOB>  AUTO-CAPTURE    11.2.0.4.0  YES NO  NO  sql_PLAN_464x2zzzm13s8f5e7e8ee

这里其实他是捕捉到另一个执行计划了,但是,由于没启用,就是全表扫描了,启用新的执行计划:

declare
tp clob;
begin
tp:=dbms_spm.evolve_sql_plan_baseline(sql_handle=>'sql_4313a2ffff308f08',plan_name=>'sql_PLAN_464x2zzzm13s8f5e7e8ee');
end;

再次运行:

Plan hash value: 1162342648

---------------------------------------------------------------------------------------------
| Id  | Operation                   | Name          | Rows  | Bytes | Cost (%cpu)| Time     |
---------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |               |     1 |   133 |     2   (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| EMP           |     1 |   133 |     2   (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN          | I_EMPLOYEE_ID |     1 |       |     1   (0)| 00:00:01 |
---------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   2 - access("EMPLOYEE_ID"=206)

Note
-----
   - dynamic sampling used for this statement (level=2)
   - sql plan baseline "sql_PLAN_464x2zzzm13s8f5e7e8ee" used for this statement

好了,走正常的执行计划了,再次删除较优的base_line:

declare v_ret  varchar2(100);
begin
  v_ret:= dbms_spm.drop_sql_plan_baseline(sql_handle=>'sql_4313a2ffff308f08',plan_name=>'sql_PLAN_464x2zzzm13s8f5e7e8ee');
  end  ;

毫无疑问,又走全表扫描了,但dba_hist_sql_plan_baselines里依旧是生成了的,只是没有启用而已;
修改状态:

declare
v_ret  varchar2(100);
begin
v_ret := dbms_spm.alter_sql_plan_baseline(
sql_handle=>'sql_4313a2ffff308f08',plan_name=>'sql_PLAN_464x2zzzm13s8d8a279cc',attribute_name=>'enabled',attribute_value=>'NO'
);
end;

查看执行计划,两个都没启用了,这和修改参数(optimizer_use_sql_plan_baselines=false)是一样的:
Execution Plan

----------------------------------------------------------
Plan hash value: 1162342648 
---------------------------------------------------------------------------------------------
| Id  | Operation                   | Name          | Rows  | Bytes | Cost (%cpu)| Time     |
---------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |               |     1 |   133 |     2   (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| EMP           |     1 |   133 |     2   (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN          | I_EMPLOYEE_ID |     1 |       |     1   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------- 
Predicate Information (identified by operation id):
--------------------------------------------------- 
   2 - access("EMPLOYEE_ID"=206) 
Note
-----
   - dynamic sampling used for this statement (level=2)