执行计划绑定
需要绑定SQL执行计划常见的几种情况:
SQL执行计划突变,导致数据库性能下降,从历史执行计划找一个合理的,进行绑定。
SQL无法使用更优的执行计划,且无历史执行计划,可通过hint手工构造的方式,进行绑定。
某些Bug引起优化器生成较差的执行计划。在bug修复前,进行绑定。
ORACLE固定执行计划的3种方式:
Oracle 9i使用outline (可跨版本10,11g均可使用)
Oracle 10g使用sql profile (11g也可使用)
Oracle 11g使用sql plan manage
接下来简述如何使用这3种方式进行执行计划的固定,并举例说明3种固定执行计划的优缺点,通过对比选择合适的固定执行计划来应对不同的业务场景。也就是什么场景下使用何种执行计划固定比较合适。
一、大纲(Stored Outline)
语法:(手动创建outline)
CREATE [ OR REPLACE ]
[ PUBLIC | PRIVATE ] OUTLINE [ outline ]
[ FROM [ PUBLIC | PRIVATE ] source_outline ]
[ FOR CATEGORY category ]
[ ON statement ] ;
1、当SQL执行计划因新版本变更,统计信息不准确,新建索引,参数改变等发生改变时,存储大纲可以使SQL语句的执行计划保持不变。在创建某条语句的大纲时,ORACLE会将SQL语句的文本,执行计划和语句使用的hints存储在一个系统默认用户OUTLN的3个表OL$,OL$HINTS,OL$NODES上。
2、使用大纲(outline)固定执行计划
--环境构建,建立测试表
SQL> create table zw as select * from dba_objects where object_id is not null;
Table created.
SQL> explain plan for select count(*) from zw;
Explained.
SQL> set lines 200
SQL> select * from table(dbms_xplan.display());
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------------------------------
Plan hash value: 249608387
-------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
-------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 339 (1)| 00:00:05 |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | TABLE ACCESS FULL| ZW | 100K| 339 (1)| 00:00:05 |
-------------------------------------------------------------------
Note
-----
- dynamic sampling used for this statement (level=2)
13 rows selected.
--查询数据字典dba_outlines:
SQL> select NAME,OWNER,CATEGORY,USED,SQL_TEXT,ENABLED,TIMESTAMP from dba_outlines;
no rows selected
--创建大纲(全表扫描),默认是私有outline
SQL> create or replace outlinezwoutlinefor category mycate onselect count(*) from zw;
Outline created.
--再次查询dba_outlines;
col NAME for a10
col OWNER for a10
col CATEGORY for a10
col SQL_TEXT for a30
select NAME,OWNER,CATEGORY,USED,SQL_TEXT,ENABLED,TIMESTAMP from dba_outlines;
NAME OWNER CATEGORY USED SQL_TEXT ENABLED TIMESTAMP
---------- ---------- ---------- ------ ------------------------------ -------- -------------------
ZWOUTLINE SYS MYCATE UNUSEDselect count(*) from zw ENABLED 2017-09-02 15:36:33
--创建object_id列索引,将该列属性设置为非空
--索引不存储null值
SQL> alter table zw modify object_id not null;
Table altered.
SQL> create index idx_zw_obj_id on zw(object_id);
Index created.
SQL> analyze table zw compute statistics;
Table analyzed.
SQL> explain plan for select count(*) from zw;
Explained.
SQL> select * from table(dbms_xplan.display());
PLAN_TABLE_OUTPUT
-----------------------------------------------------------------------------------------------------------------------------
Plan hash value: 1836624960
-------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
-------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 54 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | INDEX FAST FULL SCAN| IDX_ZW_OBJ_ID | 100K| 54 (0)| 00:00:01 |
-------------------------------------------------------------------------------
--使用大纲固定执行计划:(alter system/session set use_stored_outlines=mycate;)系统级或会话级别;
SQL> alter systemset use_stored_outlines=mycate;
System altered.
--查询dba_ouitlines(sql还未应用)
SQL> select NAME,OWNER,CATEGORY,USED,SQL_TEXT,ENABLED,TIMESTAMP from dba_outlines;
NAME OWNER CATEGORY USED SQL_TEXT ENABLED TIMESTAMP
---------- ---------- ---------- ------ ------------------------------ -------- -------------------
ZWOUTLINE SYS MYCATE UNUSED select count(*) from zw ENABLED 2017-09-02 15:36:33
--固定执行计划之后,就会按照创大纲时的执行计划去执行。
实际执行验证:(实际执行后就会应用outline)
SQL> select count(*) from zw;
COUNT(*)
----------
87036
SQL> select * from table(dbms_xplan.display_cursor(null, null, 'advanced'));
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------------
SQL_ID 1f5n0rapts695, child number 0
-------------------------------------
select count(*) from zw
Plan hash value: 1836624960
-------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
-------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | 54 (100)| |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | INDEX FAST FULL SCAN| IDX_ZW_OBJ_ID | 87036 | 54 (0)| 00:00:01 |
-------------------------------------------------------------------------------
Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------
1 - SEL$1
2 - SEL$1 / ZW@SEL$1
Outline Data
-------------
/*+
BEGIN_OUTLINE_DATA
IGNORE_OPTIM_EMBEDDED_HINTS
OPTIMIZER_FEATURES_ENABLE('11.2.0.4')
DB_VERSION('11.2.0.4')
ALL_ROWS
OUTLINE_LEAF(@"SEL$1")
INDEX_FFS(@"SEL$1" "ZW"@"SEL$1" ("ZW"."OBJECT_ID"))
END_OUTLINE_DATA
*/
Column Projection Information (identified by operation id):
-----------------------------------------------------------
1 - (#keys=0) COUNT(*)[22]
39 rows selected.
--查询dba_outlines
select NAME,OWNER,CATEGORY,USED,SQL_TEXT,ENABLED,TIMESTAMP from dba_outlines;
NAME OWNER CATEGORY USED SQL_TEXT ENABLED TIMESTAMP
---------- ---------- ---------- ------ ------------------------------ -------- -------------------
ZWOUTLINE SYS MYCATE UNUSED select count(*) from zw ENABLED 2017-09-02 15:54:31
上述的建立的大纲为公有大纲,为了不影响其它用户的使用,可以建立私有大纲如下:
create or replace private outline zwoutline2 for category mycate2 on select count(*) from zw;
思考:为什么我构建测试时,固定的是全表扫描,而不是比较优化的索引扫描?
其实这里我想说明的是outline的缺点是比较死板的,当创建新的索引,或者数据量大幅度变化时是无法做出相应改变的,也就是说它是固定死的。
关于outline具体参考如下链接:
http://blog.csdn.net/whiteoldbig/article/details/17210079
- 这种方法不能指定outline的具体名字,由系统自动生成,可以通过alter outline SYS_OUTLINE_14061114223605901 rename to dh_test2修改
- 不指定类别的话默认为default,而且此处创建时,不能指定为default类别(会报错)。
- 我们使用outline固定执行计划时,一般都是选用此种方法
参考:
http://www.bubuko.com/infodetail-216529.html
实验如下:
SQL> create table dh_stat as select rownum id ,object_name name ,object_type type from dba_objects;
Table created.
SQL> create index ind_1 on dh_stat(id) compute statistics;
Index created.
SQL> alter system flush shared_pool;
System altered.
SQL> exec DBMS_STATS.GATHER_TABLE_STATS(OWNNAME=>'sys',TABNAME=>'dh_stat',ESTIMATE_PERCENT=>30,METHOD_OPT=>'FOR ALL COLUMNS SIZE 1',NO_INVALIDATE=>FALSE,CASCADE=>TRUE,DEGREE => 1);
PL/SQL procedure successfully completed.
SQL> col name format a15
SQL> col name format a30
SQL> col sql_text for a55
SQL> select /* outlinetest1 */ * from dh_stat where id=771;
ID NAME TYPE
---------- ------------------------------ -------------------
771 SYS_IL0000000772C00002$$ INDEX
SQL> set lines 200
SQL> select sql_text,sql_id,hash_value,child_number from v$sql a where sql_text like '%outlinetest1%' and sql_text not like '%v$sql%';
SQL_TEXT SQL_ID HASH_VALUE CHILD_NUMBER
------------------------------------------------------- ------------- ---------- ------------
select /* outlinetest1 */ * from dh_stat where id=771 9a69w18a8uuhk 344812050 0
SQL> select * from table(dbms_xplan.display_cursor('9a69w18a8uuhk',null));
PLAN_TABLE_OUTPUT
-----------------------------------------------------------------------------
SQL_ID 9a69w18a8uuhk, child number 0
-------------------------------------
select /* outlinetest1 */ * from dh_stat where id=771
Plan hash value: 2780970545
---------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | | 2 (100)| |
| 1 | TABLE ACCESS BY INDEX ROWID| DH_STAT | 1 | 39 | 2 (0)| 00:00:01 |
PLAN_TABLE_OUTPUT
----------------------------------------------------------------------------------------------------------
|* 2 | INDEX RANGE SCAN | IND_1 | 1 | | 1 (0)| 00:00:01 |
---------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("ID"=771)
19 rows selected.
---根据sql已经有的执行计划生成outline:
SQL> exec DBMS_OUTLN.create_outline(hash_value=>344812050,child_number => 0,category=>'TEST');
PL/SQL procedure successfully completed.
--查询outline情况:
SQL> select name,category,used,sql_text from dba_outlines;
NAME CATEGORY USED SQL_TEXT
------------------------------ ------------------------------ ------ -------------------------------------------------------
SYS_OUTLINE_17090216454529101 TEST UNUSED select /* outlinetest1 */ * from dh_stat where id=771
此处outline的USED状态没有改变,因为我们没有激活TEST类别的outline,再次申明,outline必须通过use_stored_outlines参数激活后,优化器才会使用outline
---下面这一步激活TEST类别的OUTLINE,也可以在系统级激活OUTLINE
SQL> alter session set use_stored_outlines=TEST;
验证省略。。。。。。。。。。。
创新互联专注于中大型企业的网站设计制作、网站设计和网站改版、网站营销服务,追求商业策划与数据分析、创意艺术与技术开发的融合,累计客户上千,服务满意度达97%。帮助广大客户顺利对接上互联网浪潮,准确优选出符合自己需要的互联网运用,我们将一直专注高端网站设计和互联网程序开发,在前进的路上,与客户一起成长!
二、SQL_PROFILE
1、DBMS_SQLTUNE是10g引入的一个新特性,它可以通过自动优化性能较差SQL,并给出合理的优化建议,其中优化建议中的sql_profile文件它是一个存储在数据字典中的信息集合。sql_profile不包含单独的执行计划,提供数据库配置、绑定变量、优化统计信息、数据集等信息供优化器选择执行计划。这里不对SQL优化建议工具SQL Tuning Advisor STA)进行介绍,有兴趣的童鞋研究一下DBMS_SQLTUNE包。
语法:
SQL text format:
DBMS_SQLTUNE.CREATE_TUNING_TASK( sql_text IN CLOB, bind_list IN sql_binds := NULL, user_name IN VARCHAR2 := NULL, scope IN VARCHAR2 := SCOPE_COMPREHENSIVE, time_limit IN NUMBER := TIME_LIMIT_DEFAULT, task_name IN VARCHAR2 := NULL, description IN VARCHAR2 := NULL) RETURN VARCHAR2;
SQL ID format:
DBMS_SQLTUNE.CREATE_TUNING_TASK( sql_id IN VARCHAR2, plan_hash_value IN NUMBER := NULL, scope IN VARCHAR2 := SCOPE_COMPREHENSIVE, time_limit IN NUMBER := TIME_LIMIT_DEFAULT, task_name IN VARCHAR2 := NULL, description IN VARCHAR2 := NULL) RETURN VARCHAR2;
Workload Repository format:
DBMS_SQLTUNE.CREATE_TUNING_TASK( begin_snap IN NUMBER, end_snap IN NUMBER, sql_id IN VARCHAR2, plan_hash_value IN NUMBER := NULL, scope IN VARCHAR2 := SCOPE_COMPREHENSIVE, time_limit IN NUMBER := TIME_LIMIT_DEFAULT, task_name IN VARCHAR2 := NULL, description IN VARCHAR2 := NULL) RETURN VARCHAR2; 实验1:sql_id format 1. sys用户创建oracle优化任务(v_sqlid,task_name根据实际情况改变) set long 10000000 longchunksize 1000000 linesize 150 pagesize 0 serveroutput on size 1000000 verify off declare my_task_name varchar2(30); v_sqlid varchar2(50); begin v_sqlid:='11wrxmug9y4a7'; my_task_name := dbms_sqltune.create_tuning_task (sql_id=> v_sqlid, scope => 'comprehensive', time_limit=>160, task_name=>'task_00000', description => 'tuning task'); dbms_sqltune.execute_tuning_task('task_00000'); end; / 注:dbms_sqltune.execute_tuning_task('task_00000'),是执行优化任务 2. 打印优化任务,里面有一些具体的改进措施 select dbms_sqltune.report_tuning_task('task_00000') from dual; 3.根据优化建议,绑定profile,理论上可提高语句执行效率 execute dbms_sqltune.accept_sql_profile(task_name => 'task_00000',task_owner => 'sys', replace => true); 实验2:sql_text format SQL> alter session set statistics_level=all; Session altered. SQL> set serveroutput off SQL> select * from scott.emp where ename='SCOTT' and DEPTNO=20; EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ---------- ---------- --------- ---------- ------------------- ---------- ---------- ---------- 7788 SCOTT ANALYST 7566 1987-04-19 00:00:00 3000 20 SQL> SELECT * FROM table(dbms_xplan.display_cursor(NULL,NULL,'runstats_last')); PLAN_TABLE_OUTPUT -------------------------------------------------------------------------------- SQL_ID 4dsqbp572auuu, child number 0 ------------------------------------- select * from scott.emp where ename='SCOTT' and DEPTNO=20 Plan hash value: 3956160932 ------------------------------------------------------------------------------------ | Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | ------------------------------------------------------------------------------------ | 0 | SELECT STATEMENT | | 1 | | 1 |00:00:00.01 | 7 | |* 1 | TABLE ACCESS FULL| EMP | 1 | 1 | 1 |00:00:00.01 | 7 | PLAN_TABLE_OUTPUT --------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 1 - filter(("ENAME"='SCOTT' AND "DEPTNO"=20)) 18 rows selected. SQL> DECLARE my_task_name VARCHAR2(30); my_sqltext CLOB; BEGIN my_sqltext := 'select * from scott.emp where ename= :name and DEPTNO= :deptno'; my_task_name := DBMS_SQLTUNE.CREATE_TUNING_TASK( sql_text => my_sqltext, bind_list => sql_binds(anydata.convertvarchar2(10),anydata.convertnumber(2)), user_name => 'SYS', scope => 'COMPREHENSIVE', time_limit => 60, task_name => 'test_sql_tuning', description => 'Task to tune a query on emp'); END; / PL/SQL procedure successfully completed. 参数说明: bind_list:多个绑定变量以','逗号分隔。参数值一定要根据绑定变量对应的列的类型书写.如:emp.ename类型是VARCHAR2(10),那么就要写成 bind_list =>sql_binds(anydata.convertvarchar2(10)), time_limit:执行的最长时间,默认是60。 scope:LIMITED,用大概1秒时间去优化SQL语句,但是并不进行SQL Profiling分析;COMPREHENSIVE,进行全面分析,包含SQL Profiling分析;比LIMITED用时更长。 --查看任务名 SQL> select task_name from dba_advisor_log where task_name='test_sql_tuning'; TASK_NAME ------------------------------ test_sql_tuning --执行sql tuning任务 SQL> exec DBMS_SQLTUNE.EXECUTE_TUNING_TASK( task_name => 'test_sql_tuning' ); PL/SQL procedure successfully completed. --查看sql tunning任务状态 SQL> select task_name,status from dba_advisor_log where task_name='test_sql_tuning'; TASK_NAME STATUS ------------------------------ ----------- test_sql_tuning COMPLETED ----展示sql tunning结果 set long 10000 set longchunksize 1000 set linesize 100 select dbms_sqltune.report_tuning_task('test_sql_tuning') from dual; ...........省略 --根据建议accept_sql_profile execute dbms_sqltune.accept_sql_profile(.....) --完成后删除sql tunning任务 exec dbms_sqltune.drop_tuning_task('test_sql_tuning'); --查看SQL Tuning Advisor的进展(task执行很久) set lines 200 col opname for a20 col ADVISOR_NAME for a20 select sid,serial#,username,opname,advisor_name,target_desc,start_time sofar,totalwork from v$advisor_progress where username = 'SYS';
三、coe_xfr_sql_profile.sql固定执行计划(也是生成sql_profile)
--环境构建,建立测试表,与outline测试一样
create table zw as select * from dba_objects where object_id is not null;
alter table zw modify object_id not null; --索引不存储null值
create index idx_zw_obj_id on zw(object_id);
analyze table zw compute statistics;
select count(*) from zw;
COUNT(*)
----------
87038
SQL> select * from table(dbms_xplan.display_cursor(null,0));
PLAN_TABLE_OUTPUT
-----------------------------------------------------------------------------------
SQL_ID 1f5n0rapts695, child number 0
-------------------------------------
select count(*) from zw
Plan hash value: 1836624960
-------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
-------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | 54 (100)| |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | INDEX FAST FULL SCAN| IDX_ZW_OBJ_ID | 87038 | 54 (0)| 00:00:01 |
-------------------------------------------------------------------------------
--使用hint提示,强制走全表,生成一个执行计划:
SQL> select /*+ full(zw) */ count(*)from zw;
COUNT(*)
----------
87038
SQL> select * from table(dbms_xplan.display_cursor(null,0));
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------
SQL_ID fp5ksbhww4594, child number 0
-------------------------------------
select /*+ full(zw) */ count(*)from zw
Plan hash value: 249608387
-------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
-------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | 339 (100)| |
| 1 | SORT AGGREGATE | | 1 | | |
| 2 | TABLE ACCESS FULL| ZW | 87038 | 339 (1)| 00:00:05 |
-------------------------------------------------------------------
14 rows selected.
--运行coe_xfr_sql_profile脚本:(固定执行计划)
SQL> @coe_xfr_sql_profile.sql
Parameter 1:
SQL_ID (required)
Enter value for 1: 1f5n0rapts695 --原sql的sql_id
PLAN_HASH_VALUE AVG_ET_SECS
--------------- -----------
1836624960 .02
Parameter 2:
PLAN_HASH_VALUE (required)
Enter value for 2: 249608387 --想要选择的执行计划的PLAN_HASH_VALUE值
Values passed to coe_xfr_sql_profile:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
SQL_ID : "1f5n0rapts695"
PLAN_HASH_VALUE: "249608387"
SQL>BEGIN
2 IF :sql_text IS NULL THEN
3 RAISE_APPLICATION_ERROR(-20100, 'SQL_TEXT for SQL_ID &&sql_id. was not found in memory (gv$sqltext_with_newlines) or AWR (dba_hist_sqltext).');
4 END IF;
5 END;
6 /
SQL>SET TERM OFF;
SQL>BEGIN
2 IF :other_xml IS NULL THEN
3 RAISE_APPLICATION_ERROR(-20101, 'PLAN for SQL_ID &&sql_id. and PHV &&plan_hash_value. was not found in memory (gv$sql_plan) or AWR (dba_hist_sql_plan).');
4 END IF;
5 END;
6 /
SQL>SET TERM OFF;
Execute coe_xfr_sql_profile_1f5n0rapts695_249608387.sql
on TARGET system in order to create a custom SQL Profile
with plan 249608387 linked to adjusted sql_text.
COE_XFR_SQL_PROFILE completed.
SQL> @coe_xfr_sql_profile_1f5n0rapts695_249608387.sql
省略.................
coe_xfr_sql_profile_1f5n0rapts695_249608387 completed
SQL> explain plan for select count(*) from zw;
Explained.
SQL> select * from table(dbms_xplan.display());
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Plan hash value: 249608387
-------------------------------------------------------------------
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
-------------------------------------------------------------------
| 0| SELECT STATEMENT | | 1 | 312 (1)| 00:00:04 |
| 1| SORT AGGREGATE | | 1 | | |
| 2| TABLE ACCESS FULL| ZW | 87038 | 312 (1)| 00:00:04 |
-------------------------------------------------------------------
Note
PLAN_TABLE_OUTPUT
-------------------------------------------------------------------------------------
-SQL profile "coe_1f5n0rapts695_249608387"used for this statement
--查询:
SQL> select name,category,status,sql_text fromdba_sql_profiles;
NAME CATEGORY STATUS SQL_TEXT
------------------------------ ------------ -------------------------------------------
coe_1f5n0rapts695_249608387 DEFAULE ENABLED select count(*) from zw
当存在只有一个执行计划时,而又不想让应用不改sql代码的情况下改变执行计划,请参照自动/手动类型sql_profile(dbms_sqltune/coe_xfr_sql_profile.sql)
使用coe_xfr_sql_profile.sql固定计划是不是很好用呢?是的,这一切都归功于oracle mos上的功劳,需要的童鞋可以到matelink上查找和下载。还有其它两个有关的脚本:coe_load_sql_baseline.sql,coe_load_sql_profile.sql,有兴趣的童鞋可以一起下载研究。
思考:当在使用SQL_PROFILE绑定之前,使用了OUTLINE进行固定的话,谁的优先级高呢?
根据网上的一些资料说是OUTLINE的优先级最高,但都是简短的一句话,没有证明。可是经过我无数次的测试,发现都是SQL_PROFILE的优先级较高,具体相关测试结果我就不粘贴出来了。(或许是我测试语句的特殊性,需再进一步验证)
值得一提的是,sql_profile并不会以outline方式存储冻结执行计划,当表中数据增长或索引被删除或重建时,在sql_profile不变的情况下执行计划也可以发生变化,信息的存储和与数据的分布或者访问路径有关。
四、SQL PLAN MANAGE(baseline)
1、从11g开始,oracle引入了SQL执行计划管理(SQLPlan Management)这个新特性,与Oracle 9i 的outline和10g 的profile相比,Oracle 11g的SPM相对更加的灵活,允许你同时接受多个执行计划。
2、使用SQL PlanManagement固定执行计划
--一条带有绑定变量的SQL语句,但数据分布不均,严重倾斜时,最好的执行计划会根据绑定变量的值而不同。执行时,根据不同的变量值,SPM会花费很少的运算从中选择一条最合适的。
SQL> select id,count(*) from test group by id order by 2;
ID COUNT(*)
---------- ----------
10 1100
88 10100
999 1000000
--接下来定义一个变量a,分别赋值999和10,看它的执行计划是如何的
SQL>alter system flush shared_pool;
SQL>var a1 number;
SQL>exec :a1:=999;
SQL>select t.* from test t wheret.id=:a1;
1000000 rows selected.
Elapsed: 00:00:25.30
SQL> select * from table(dbms_xplan.display_cursor(null,0));
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
SQL_ID cpsdn05zdq02p,child number 0
-------------------------------------
select t.* from test t where t.id=:a1
Plan hash value: 1357081020
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0| SELECT STATEMENT | | | | 424 (100)| |
|* 1| TABLE ACCESS FULL| TEST | 337K| 1316K| 424 (2)| 00:00:06 |
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Predicate Information (identified byoperation id):
---------------------------------------------------
1- filter("T"."ID"=:A1)
-##########################ID列上有个索引IDX_ID ################################
SQL>alter system flush shared_pool;
SQL>var a1 number;
SQL>exec :a1:=10;
SQL>select t.* from test t wheret.id=:a1;
1100 rows selected.
Elapsed: 00:00:00.04
SQL> select * fromtable(dbms_xplan.display_cursor(null,0));
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
SQL_ID cpsdn05zdq02p,child number 0
-------------------------------------
select t.* from test t where t.id=:a1
Plan hash value: 1357081020
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0| SELECT STATEMENT | | | | 424 (100)| |
|* 1| TABLE ACCESS FULL| TEST | 337K| 1316K| 424 (2)| 00:00:06 |
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Predicate Information (identified byoperation id):
---------------------------------------------------
1- filter("T"."ID"=:A1)
--这里可以看到,无论赋值是999还是10,其执行计划都是一样的,但根据理论来说,我们都知道,id=10时走索引效率是最好的。假设数据是均匀分布的,那么基数评估cardinality=density*num_rows。Density可通过user_tab_col_statistics查询。
select column_name,num_distinct,density from user_tab_col_statistics where table_name='TEST';
COLUMN_NAME NUM_DISTINCT DENSITY
------------------------------ ------------- ---------
ID 3 .333333333
--我们看到的Rows列预估的337k就是cardinality=density*num_rows=0.3333*1011200约等于337k行,
--但是我们都知道ID=10只有1100行,而ID=999有1000000行,所以当ID=10的时候走索引全扫描,ID=999的时候走全表扫描是最合理的执行计划。
那么面对这种情况,我们该如何让这种情况下的执行计划达到最优呢?方法有如下几个:
--1、去除绑定变量,直接硬解析的方式(非理想的,如果涉及要该程序代码这是很不可取的)
--2、启用11g的新特性ACS(自适应游标,这个BUG不是一般的多,不建议启用)
--3、收集直方图信息(如果在生产高峰期,收集直方图信息所占资源无法评估)
--4、使用SPM把不同的执行计划加入到SQLPlan Baseline中。
--使用手工捕获的方式
alter system flush shared_pool;
var a1 number;
exec :a1:=999;
select t.* from test t where t.id=:a1;
select * fromtable(dbms_xplan.display_cursor(null,0));
var temp varchar2(1000);
exec :temp:=dbms_spm.load_plans_from_cursor_cache(sql_id => 'cpsdn05zdq02p');
exec :temp :=dbms_spm.alter_sql_plan_baseline(sql_handle=>'SQL_d230ce970caa0077',plan_name=>'SQL_PLAN_d4c6fkw6an03r97bbe3d0',attribute_name=>'ENABLED',attribute_value=>'NO'); --先修改全表扫描的sql planbaselines的enabled属性为NO,不然捕获不了索引的。
exec :a1:=10;
select t.* from test t where t.id=:a1;
select * fromtable(dbms_xplan.display_cursor(null,0));
exec :temp:=dbms_spm.load_plans_from_cursor_cache(sql_id => 'cpsdn05zdq02p');
dbms_spm.alter_sql_plan_baseline(sql_handle=>'SQL_d230ce970caa0077',plan_name=>'SQL_PLAN_d4c6fkw6an03r97bbe3d0',attribute_name=>'ENABLED',attribute_value=>'YES');
SQL> select sql_handle,plan_name,origin,enabled,accepted,fixed from dba_sql_plan_baselines;
SQL_HANDLE PLAN_NAME ORIGIN ENA ACC FIX
-------------------------------------------------- -------------- --- --- ---
SQL_d230ce970caa0077SQL_PLAN_d4c6fkw6an03r97bbe3d0 MANUAL-LOAD YES YES NO
SQL_d230ce970caa0077SQL_PLAN_d4c6fkw6an03rf98b55bb MANUAL-LOAD YES YES NO
--验证结果:
SQL> var a1 number;
SQL> exec :a1:=10;
SQL> select t.* from test t wheret.id=:a1;
SQL> select * fromtable(dbms_xplan.display_cursor(null,0));
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
SQL_ID cpsdn05zdq02p,child number 0
-------------------------------------
select t.* from test t where t.id=:a1
Plan hash value: 578627003
---------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------
| 0| SELECT STATEMENT | | | | 5(100)| |
|* 1| INDEX RANGE SCAN| IDX_ID | 1280 | 5120 | 5 (0)| 00:00:01 |
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Predicate Information (identified byoperation id):
---------------------------------------------------
1- access("T"."ID"=:A1)
Note
-----
-SQL plan baseline SQL_PLAN_d4c6fkw6an03rf98b55bbused for this statement
22 rows selected.
SQL> var a1 number;
SQL> exec :a1:=999;
SQL> select t.* from test t wheret.id=:a1;
SQL> select * fromtable(dbms_xplan.display_cursor(null,0));
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
SQL_ID cpsdn05zdq02p,child number 0
-------------------------------------
select t.* from test t where t.id=:a1
Plan hash value: 1357081020
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0| SELECT STATEMENT | | | | 424 (100)| |
|* 1| TABLE ACCESS FULL| TEST | 1001K| 3912K| 424 (2)| 00:00:06 |
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Predicate Information (identified byoperation id):
---------------------------------------------------
1- filter("T"."ID"=:A1)
Note
-----
-SQL plan baseline SQL_PLAN_d4c6fkw6an03r97bbe3d0used for this statement
22 rows selected.
SPM的灵活之处在于,可以动态管理,不像存储大纲(stored outline)和SQL Profile需要DBA手工创建,当然SPM也可以,因为我在以上演示中也没让它自动捕获。
思考:1、何种情况下使用什么固定执行计划的方法更加有效?2、在各种固定执行计划都使用的情况下,那种优先级更高?
名称栏目:执行计划绑定
本文路径:http://ybzwz.com/article/jsejii.html