在Oracle10g之前,优化SQL是个比较费力的技术活,不停的分析执行计划,加hint,分析统计信息等等。在10g中,Oracle推出了自己的SQL优化辅助工具:SQL优化器(SQLTuningAdvisor:STA),它是新的DBMS_SQLTUNE包。使用STA一定要保证优化器是CBO模式下。
执行DBMS_SQLTUNE包进行sql优化需要有advisor的权限:
SQL>createuserdaveidentifiedbydave;
用户已创建。
SQL>grantconnect,resourcetodave;
授权成功。
SQL>grantadvisortodave;
授权成功。
下面简单介绍一下如何优化一条找到的问题语句。
createtablebigtabasselectrownumas"id",a.*fromsys.all_objectsa;
createtablesmalltabasselectrownumas"id",a.*FROMsys.all_tablesa;
然后多运行几次下面的脚本,增加表里的数据:
insertintobigtabselectrownumas"id",a.*fromsys.all_objectsa;
insertintosmalltabselectrownumas"id",a.*FROMsys.all_tablesa;
这里创建一张大表和一张小表,并且都没有索引,下面执行一个查询:
SQL>settimingon
SQL>setautoton
SQL>selectcount(*)frombigtaba,smalltabbwherea.object_name=b.table_name;
COUNT(*)
----------
2141537
已用时间:00:00:20.05
执行计划
----------------------------------------------------------
Planhashvalue:3089226980
--------------------------------------------------------------------------------
|Id|Operation|Name|Rows|Bytes|Cost(%CPU)|Time|
--------------------------------------------------------------------------------
|0|SELECTSTATEMENT||1|45|3146(1)|00:00:38|
|1|SORTAGGREGATE||1|45|||
|*2|HASHJOIN||447K|19M|3146(1)|00:00:38|
|3|TABLEACCESSFULL|SMALLTAB|27327|533K|264(1)|00:00:04|
|4|TABLEACCESSFULL|BIGTAB|712K|16M|2878(1)|00:00:35|
--------------------------------------------------------------------------------
PredicateInformation(identifiedbyoperationid):
---------------------------------------------------
2-access("A"."OBJECT_NAME"="B"."TABLE_NAME")
统计信息
----------------------------------------------------------
0recursivecalls
0dbblockgets
31149consistentgets
21058physicalreads
0redosize
426bytessentviaSQL*Nettoclient
416bytesreceivedviaSQL*Netfromclient
2SQL*Netroundtripsto/fromclient
0sorts(memory)
0sorts(disk)
1rowsprocessed
第一步:创建优化任务
通过调用函数CREATE_TUNING_TASK来创建优化任务,调用存储过程EXECUTE_TUNING_TASK执行该任务:
SQL>setautotoff
SQL>settimingoff
SQL>DECLARE
2my_task_nameVARCHAR2(30);
3my_sqltextCLOB;
4BEGIN
5my_sqltext:='selectcount(*)frombigtaba,smalltabbwherea.object_name=b.table_name';
6my_task_name:=DBMS_SQLTUNE.CREATE_TUNING_TASK(
7sql_text=>my_sqltext,
8user_name=>'DAVE',--注意是大写,不然会报错,用户无效
9scope=>'COMPREHENSIVE',
10time_limit=>60,
11task_name=>'tuning_sql_test',
12description=>'Tasktotuneaqueryonaspecifiedtable');
13
14DBMS_SQLTUNE.EXECUTE_TUNING_TASK(task_name=>'tuning_sql_test');
15END;
16/
PL/SQLproceduresuccessfullycompleted.
在函数CREATE_TUNING_TASK,sql_text是需要优化的语句,user_name是该语句通过哪个用户执行,scope是优化范围(limited或comprehensive),time_limit优化过程的时间限制,task_name优化任务名称,description优化任务描述。
第二步:执行优化任务
通过调用dbms_sqltune.execute_tuning_task过程来执行前面创建好的优化任务。
SQL>execdbms_sqltune.execute_tuning_task('tuning_sql_test');
PL/SQL过程已成功完成。
第三步:检查优化任务的状态
通过查看user_advisor_tasks/dba_advisor_tasks视图可以查看优化任务的当前状态。
SQL>SELECTtask_name,statusFROMUSER_ADVISOR_TASKSWHEREtask_name='tuning_sql_test';
TASK_NAMESTATUS
-----------------------------------------
tuning_sql_testCOMPLETED
第四步:查看优化结果
通过dbms_sqltune.report_tning_task函数可以获得优化任务的结果。
SQL>SETLONG999999
SQL>setserveroutputonsize999999
SQL>SETLINESIZE100
SQL>SELECTDBMS_SQLTUNE.REPORT_TUNING_TASK('tuning_sql_test')fromDUAL;
DBMS_SQLTUNE.REPORT_TUNING_TASK('TUNING_SQL_TEST')
----------------------------------------------------------------------------------------------------
GENERALINFORMATIONSECTION
-------------------------------------------------------------------------------
TuningTaskName:tuning_sql_test
TuningTaskOwner:DEMO
Scope:COMPREHENSIVE
TimeLimit(seconds):60
CompletionStatus:COMPLETED
Startedat:5/28/201013:16:43
Completedat:5/28/201013:16:44
NumberofIndexFindings:1
SchemaName:DEMO
SQLID:6p64dnnsqf9pm
SQLText:selectcount(*)frombigtaba,smalltabbwhere
a.object_name=b.table_name
-------------------------------------------------------------------------------
FINDINGSSECTION(1finding)
-------------------------------------------------------------------------------
1-IndexFinding(seeexplainplanssectionbelow)
Theexecutionplanofthisstatementcanbeimprovedbycreatingoneormore
indices.
Recommendation(estimatedbenefit:100%)
----------------------------------------
-ConsiderrunningtheAccessAdvisortoimprovethephysicalschemadesign
orcreatingtherecommendedindex.
createindexDEMO.IDX$$_06C50001onSYS.SMALLTAB('TABLE_NAME');
-ConsiderrunningtheAccessAdvisortoimprovethephysicalschemadesign
orcreatingtherecommendedindex.
createindexDEMO.IDX$$_06C50002onSYS.BIGTAB('OBJECT_NAME');
Rationale
---------
Creatingtherecommendedindicessignificantlyimprovestheexecutionplan
ofthisstatement.However,itmightbepreferabletorun"AccessAdvisor"
usingarepresentativeSQLworkloadasopposedtoasinglestatement.This
willallowtogetcomprehensiveindexrecommendationswhichtakesinto
accountindexmaintenanceoverheadandadditionalspaceconsumption.
EXPLAINPLANSSECTION
-------------------------------------------------------------------------------
1-Original
-----------
Planhashvalue:3089226980
--------------------------------------------------------------------------------
|Id|Operation|Name|Rows|Bytes|Cost(%CPU)|Time|
--------------------------------------------------------------------------------
|0|SELECTSTATEMENT||1|36|3550(2)|00:00:43|
|1|SORTAGGREGATE||1|36|||
|*2|HASHJOIN||155K|5462K|3550(2)|00:00:43|
|3|TABLEACCESSFULL|SMALLTAB|1223|22014|11(0)|00:00:01|
|4|TABLEACCESSFULL|BIGTAB|1205K|20M|3526(1)|00:00:43|
--------------------------------------------------------------------------------
PredicateInformation(identifiedbyoperationid):
---------------------------------------------------
2-access("A"."OBJECT_NAME"="B"."TABLE_NAME")
2-UsingNewIndices
--------------------
Planhashvalue:494801882
-----------------------------------------------------------------------------------------
|Id|Operation|Name|Rows|Bytes|Cost(%CPU)|Time|
-----------------------------------------------------------------------------------------
|0|SELECTSTATEMENT||1|36|1108(3)|00:00:14|
|1|SORTAGGREGATE||1|36|||
|*2|HASHJOIN||155K|5462K|1108(3)|00:00:14|
|3|INDEXFASTFULLSCAN|IDX$$_06C50001|1223|22014|3(0)|00:00:01|
|4|INDEXFASTFULLSCAN|IDX$$_06C50002|1205K|20M|1093(2)|00:00:14|
-----------------------------------------------------------------------------------------
PredicateInformation(identifiedbyoperationid):
---------------------------------------------------
2-access("A"."OBJECT_NAME"="B"."TABLE_NAME")
------------------------------------------------------------------------------
看一下这个优化建议报告:
第一部分是关于这次优化任务的基本信息:如任务名称、执行时间、范围、涉及到的语句等等。
第二部分是关于这次优化任务的所找到的问题以及给出的优化建议。前面先给出了问题描述:可以通过建立更多的所引来提高性能;然后是建议的具体内容:在表smalltab的字段table_name上创建索引,在表bigtab的字段object_name上创建索引;最后是相关注意事项:此次优化虽然给出了创建索引的建议,但是最好通过SQL访问建议器(SQLAccessAdvisorSAA)结合整个数据库的工作量来深入分析,那样就能给出考虑了索引维护和空间消耗等因素的更加合理的建议。
最后,报告还给出了原有的查询计划,以及采用优化建议以后的查询计划的对比。可以看出COST值大大下降。
五、删除优化任务
通过调用dbms_sqltuen.drop_tuning_task可以删除已经存在的优化任务
SQL>execdbms_sqltune.drop_tuning_task('tuning_sql_test');
PL/SQLproceduresuccessfullycompleted.
注:整理自网络