使用hint优化Oracle的执行计划 以及 SQL Tune Advisor的使用
背景：
某表忽然出现查询非常缓慢的情况，cost 100+ 秒以上；严重影响生产。
原SQL：
explain plan for
select * from (
select ID id,RET_NO retNo, FROM_SYS fromSy, TO_SYS toSys, COMMAND_CODE commandCode, COMMAND, STATUS,
EXT_CODE, ORIGN_CODE orignCode,error_message errorMessage, RE_F, RET_MSG retMsg
from interface_table where ((command_code in('AASSS')
and status in('F','E') and (re_f = 'N') and FROM_SYS = 'MEE')
or (COMMAND_CODE in('XXXX','XXXX9') and FROM_SYS = 'EXT' and RE_F = 'N')
) and MOD(id, 1) = 0 order by id) where rownum <= 100 ;查看其执行计划：
SELECT plan_table_output FROM TABLE(DBMS_XPLAN.DISPLAY('PLAN_TABLE'));
Plan hash value: 1871549687 ---------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 99 | 382K| 637 (1)| 00:00:08 | |* 1 | COUNT STOPKEY | | | | | | | 2 | VIEW | | 100 | 386K| 637 (1)| 00:00:08 | |* 3 | TABLE ACCESS BY INDEX ROWID| INTERFACE_TABLE | 355 | 55735 | 637 (1)| 00:00:08 | |* 4 | INDEX FULL SCAN | PK_INTERFACE_TABLE | 1439 | | 280 (2)| 00:00:04 | ----------------------------------------------------------------------------------------------------
优化后的SQL：
explain plan for
select * from (
select /*+ index(INT_TABLE IX_INT_TABLE_2)*/ ID id,RET_NO retNo, FROM_SYS fromSy, TO_SYS toSys, COMMAND_CODE commandCode, COMMAND, STATUS,
EXT_CODE, ORIGN_CODE orignCode,error_message errorMessage, RE_F, RET_MSG retMsg
from interface_table where ((command_code in('AASSS')
and status in('F','E') and (re_f = 'N') and FROM_SYS = 'MEE')
or (COMMAND_CODE in('XXXX','XXXX9') and FROM_SYS = 'EXT' and RE_F = 'N')
) and MOD(id, 1) = 0 order by id) where rownum <= 100 ;
查看其执行计划：
SELECT plan_table_output FROM TABLE(DBMS_XPLAN.DISPLAY('PLAN_TABLE'));
Plan hash value: 3625182869 -------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 99 | 382K| 19105 (1)| 00:03:50 | |* 1 | COUNT STOPKEY | | | | | | | 2 | VIEW | | 356 | 1376K| 19105 (1)| 00:03:50 | |* 3 | SORT ORDER BY STOPKEY | | 356 | 55892 | 19105 (1)| 00:03:50 | | 4 | CONCATENATION | | | | | | |* 5 | TABLE ACCESS BY INDEX ROWID| INTERFACE_TABLE | 69 | 10833 | 9552 (1)| 00:01:55 | |* 6 | INDEX RANGE SCAN | IX_INTERFACE_TABLE_2 | 77145 | | 99 (0)| 00:00:02 | |* 7 | TABLE ACCESS BY INDEX ROWID| INTERFACE_TABLE | 287 | 45059 | 9552 (1)| 00:01:55 | |* 8 | INDEX RANGE SCAN | IX_INTERFACE_TABLE_2 | 77145 | | 99 (0)| 00:00:02 | --------------------------------------------------------------------------------------------------------
比较：
查看执行计划，原来是使用 full scan - 当数据量大时非常慢；优化后oracle优先走range scan，hint 的 index 是未处理标识字段的索引，正常情况下这个数据集合相对较小--------所以可以达到优化目的。
具体情况具体分析，我们必须要看实际的表存的业务数据，分析其业务关系找到最小业务集合；后者要看懂执行计划，根据rows, bytes, cost, time 找到最优项目。这个分析顺序不能倒置。
问题：为何使用 rownum 后，oracle执行计划会走full scan？
转：如何看懂执行计划：http://jadethao.iteye.com/blog/1613943
==== section2 ====
http://blog.chinaunix.net/uid-77311-id-3233190.html
begin
for i in 1 .. 100000 loop
insert into tb_test values (i, 'test');
commit;
end loop;
end;
/
dbms_stats.gather_table_stats(ownname => 'SCOTT', tabname => 'TB_TEST');
end;
3.编造一个执行计划不对的SQL
select/*+ full(t)*/ count(1) from scott.tb_test t where t.id=1
我们知道在ID列上有索引,这个SQL走索引是正确的执行计划,但这里强制oracle走全表扫描,然后通过STA,看oracle给出的执行计划是否正确.
4.创建TUNING_TASK并执行
declare
l_task_name varchar2(30);
l_sql clob;
begin
l_sql := 'select/*+ full(t)*/ count(1) from scott.tb_test t where t.id=1';
l_task_name := DBMS_SQLTUNE.CREATE_TUNING_TASK(sql_text => l_sql,
user_name => 'SCOTT',
scope => 'COMPREHENSIVE',
time_limit => 60,
task_name => 'task_name01',
description => null);
dbms_sqltune.Execute_tuning_task(task_name => 'task_name01');
end;
5.查看oracle给出的优化建议
SQL> set serveroutput on;
SQL> set long 999999999;
SQL> SELECT DBMS_SQLTUNE.REPORT_TUNING_TASK('task_name01') FROM DUAL;
DBMS_SQLTUNE.REPORT_TUNING_TASK('TASK_NAME01')
--------------------------------------------------------------------------------
GENERAL INFORMATION SECTION
-------------------------------------------------------------------------------
Tuning Task Name : task_name01
Tuning Task Owner : SYS
Scope : COMPREHENSIVE
Time Limit(seconds) : 60
Completion Status : COMPLETED
Started at : 06/03/2012 00:07:58
Completed at : 06/03/2012 00:07:59
Number of SQL Profile Findings : 1
DBMS_SQLTUNE.REPORT_TUNING_TASK('MY_TASK_NAME01')
--------------------------------------------------------------------------------
-------------------------------------------------------------------------------
Schema Name: SCOTT
SQL ID : ga3q5tqjgsj5u
SQL Text : select/*+ full(t)*/ count(1) from scott.tb_test t where t.id=1
-------------------------------------------------------------------------------
FINDINGS SECTION (1 finding)
-------------------------------------------------------------------------------
1- SQL Profile Finding (see explain plans section below)
--------------------------------------------------------
DBMS_SQLTUNE.REPORT_TUNING_TASK('MY_TASK_NAME01')
--------------------------------------------------------------------------------
A potentially better execution plan was found for this statement.
Recommendation (estimated benefit: 84.11%)
------------------------------------------
- Consider accepting the recommended SQL profile.
execute dbms_sqltune.accept_sql_profile(task_name => 'task_name01',
replace => TRUE);
-------------------------------------------------------------------------------
EXPLAIN PLANS SECTION
-------------------------------------------------------------------------------
DBMS_SQLTUNE.REPORT_TUNING_TASK('TASK_NAME01')
--------------------------------------------------------------------------------
1- Original With Adjusted Cost
------------------------------
Plan hash value: 1372292586
--------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------------
-- 当前的执行计划
DBMS_SQLTUNE.REPORT_TUNING_TASK('TASK_NAME01')
--------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 4 | 6 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 4 | | |
|* 2 | TABLE ACCESS FULL| TB_TEST | 1 | 4 | 6 (0)| 00:00:01 |
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
DBMS_SQLTUNE.REPORT_TUNING_TASK('TASK_NAME01')
--------------------------------------------------------------------------------
2 - filter("T"."ID"=1)
-- Oracle给出的执行计划
2- Using SQL Profile
--------------------
Plan hash value: 847665939
--------------------------------------------------------------------------------
---
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
DBMS_SQLTUNE.REPORT_TUNING_TASK('TASK_NAME01')
--------------------------------------------------------------------------------
--------------------------------------------------------------------------------
---
| 0 | SELECT STATEMENT | | 1 | 4 | 1 (0)| 00:00:0
1 |
| 1 | SORT AGGREGATE | | 1 | 4 | |
|
|* 2 | INDEX RANGE SCAN| IDX_TB_TEST | 1 | 4 | 1 (0)| 00:00:0
1 |
--------------------------------------------------------------------------------
---
DBMS_SQLTUNE.REPORT_TUNING_TASK('TASK_NAME01')
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("T"."ID"=1)
-------------------------------------------------------------------------------
从上面的输出可以看出Oracle给出的执行计划是正确的.可以使用如下方法使用正确的执行计划
begin
dbms_sqltune.accept_sql_profile(task_name => 'task_name01', replace => TRUE);
end;
-- The End --