[20181220]使用提示OR_EXPAND优化.txt
[20181220]使用提示OR_EXPAND优化.txt
--//链接http://www.itpub.net/thread-2107240-2-1.html,http://www.itpub.net/thread-2107231-2-1.html的讨论.
--//ZALBB建议在18c下尝试看看,我们这里仅仅1台18c,而且还是生产系统,正好前几天在办公机器重新安装12c,在12c测试看看.
--//主要问题感觉oracle对于这样的sql有点奇怪....
1.环境:
SCOTT@test01p> @ ver1
PORT_STRING VERSION BANNER CON_ID
------------------------------ -------------- -------------------------------------------------------------------------------- ----------
IBMPC/WIN_NT64-9.1.0 12.2.0.1.0 Oracle Database 12c Enterprise Edition Release 12.2.0.1.0 - 64bit Production 0
create table t1 as select rownum id1 ,rownum id2 ,lpad('x',100,'x') name from dual connect by level<=6000;
create table t2 as select rownum id1 ,rownum id2 ,lpad('x',100,'x') name from dual connect by level<=6000;
create index i_t1_id1 on t1(id1);
create index i_t1_id2 on t1(id2);
create index i_t2_id1 on t2(id1);
--//分析略.
2.测试:
SCOTT@test01p> alter session set statistics_level = all;
Session altered.
SCOTT@test01p> select * from t1 where t1.id1 in (select t2.id1 from t2 where t2.id1=11 ) or (t1.id2=10 );
ID1 ID2 NAME
---------- ---------- ----------------------------------------------------------------------------------------------------
10 10 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
11 11 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
SCOTT@test01p> @ dpc '' ''
PLAN_TABLE_OUTPUT
-------------------------------------
SQL_ID gz5pqkg6svm7k, child number 0
-------------------------------------
select * from t1 where t1.id1 in (select t2.id1 from t2 where
t2.id1=11 ) or (t1.id2=10 )
Plan hash value: 1962644737
-------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
-------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 30 (100)| | 2 |00:00:00.01 | 115 |
|* 1 | FILTER | | 1 | | | | | 2 |00:00:00.01 | 115 |
| 2 | TABLE ACCESS FULL| T1 | 1 | 6000 | 638K| 30 (0)| 00:00:01 | 6000 |00:00:00.01 | 113 |
|* 3 | FILTER | | 5999 | | | | | 1 |00:00:00.01 | 2 |
|* 4 | INDEX RANGE SCAN| I_T2_ID1 | 1 | 1 | 4 | 1 (0)| 00:00:01 | 1 |00:00:00.01 | 2 |
-------------------------------------------------------------------------------------------------------------------------
Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------
1 - SEL$1
2 - SEL$1 / T1@SEL$1
3 - SEL$2
4 - SEL$2 / T2@SEL$2
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter(("T1"."ID2"=10 OR IS NOT NULL))
3 - filter(11=:B1)
4 - access("T2"."ID1"=:B1)
32 rows selected.
--//执行计划存在1个全表扫描.里面的索引选择性很好,oracle并没有选择合理的执行计划.
--//而且有1个小小的细节,id=4的starts=1,而前面的id=3的starts=5999.你可以看出这里oracle显示执行计划有1个小小的bug.
--//id=4的starts应该是5999.这样看到的逻辑读不应该是后面的2而是2*5999 = 11998.
--//而且你可以看出oracle忽略的id=4多次INDEX RANGE SCAN的成本.
--//链接http://www.itpub.net/thread-2107240-2-1.html里面的显示倒是正确的.它的版本是11.2.0.4.180717.
3.是否通过提示优化sql语句:
--//首先想到的是USE_CONCAT.
select /*+ USE_CONCAT(@"SEL$1" 8 OR_PREDICATES(1)) */ * from t1 where t1.id1 in (select /*+unnest */ t2.id1 from t2 where t2.id1=11 ) or (t1.id2=10 );
--//执行计划如下:
-------------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
-------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 33 (100)| | 2 |00:00:00.01 | 118 |
| 1 | CONCATENATION | | 1 | | | | | 2 |00:00:00.01 | 118 |
| 2 | TABLE ACCESS BY INDEX ROWID BATCHED| T1 | 1 | 1 | 109 | 2 (0)| 00:00:01 | 1 |00:00:00.01 | 4 |
|* 3 | INDEX RANGE SCAN | I_T1_ID2 | 1 | 1 | | 1 (0)| 00:00:01 | 1 |00:00:00.01 | 3 |
|* 4 | FILTER | | 1 | | | | | 1 |00:00:00.01 | 114 |
|* 5 | TABLE ACCESS FULL | T1 | 1 | 5999 | 638K| 30 (0)| 00:00:01 | 5999 |00:00:00.01 | 112 |
|* 6 | FILTER | | 5999 | | | | | 1 |00:00:00.01 | 2 |
|* 7 | INDEX RANGE SCAN | I_T2_ID1 | 1 | 1 | 4 | 1 (0)| 00:00:01 | 1 |00:00:00.01 | 2 |
-------------------------------------------------------------------------------------------------------------------------------------------
Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------
1 - SEL$1
2 - SEL$1_1 / T1@SEL$1
3 - SEL$1_1 / T1@SEL$1
5 - SEL$1_2 / T1@SEL$1_2
6 - SEL$2
7 - SEL$2 / T2@SEL$2
Predicate Information (identified by operation id):
---------------------------------------------------
3 - access("T1"."ID2"=10)
4 - filter( IS NOT NULL)
5 - filter(LNNVL("T1"."ID2"=10))
6 - filter(11=:B1)
7 - access("T2"."ID1"=:B1)
--//很奇怪id=4,依旧选择过滤,unnest提示没有用.实际上使用USE_CONCAT相当每个or分支加入LNNVL(条件)来排他符合条件的记录.
--//也就是oracle依旧选择的执行计划不是很理想,甚至比前面还要差.
4.尝试OR_EXPAND提示:
select /*+ OR_EXPAND */ * from t1 where t1.id1 in (select /*+ unnest */ t2.id1 from t2 where t2.id1=11 ) or (t1.id2=10 );
--//执行计划如下:
Plan hash value: 1716482303
----------------------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time | A-Rows | A-Time | Buffers |
----------------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 5 (100)| | 2 |00:00:00.01 | 9 |
| 1 | VIEW | VW_ORE_BA8ECEFB | 1 | 2 | 156 | 5 (0)| 00:00:01 | 2 |00:00:00.01 | 9 |
| 2 | UNION-ALL | | 1 | | | | | 2 |00:00:00.01 | 9 |
| 3 | TABLE ACCESS BY INDEX ROWID BATCHED | T1 | 1 | 1 | 109 | 2 (0)| 00:00:01 | 1 |00:00:00.01 | 4 |
|* 4 | INDEX RANGE SCAN | I_T1_ID2 | 1 | 1 | | 1 (0)| 00:00:01 | 1 |00:00:00.01 | 3 |
| 5 | NESTED LOOPS SEMI | | 1 | 1 | 113 | 3 (0)| 00:00:01 | 1 |00:00:00.01 | 5 |
|* 6 | TABLE ACCESS BY INDEX ROWID BATCHED| T1 | 1 | 1 | 109 | 2 (0)| 00:00:01 | 1 |00:00:00.01 | 3 |
|* 7 | INDEX RANGE SCAN | I_T1_ID1 | 1 | 1 | | 1 (0)| 00:00:01 | 1 |00:00:00.01 | 2 |
|* 8 | INDEX RANGE SCAN | I_T2_ID1 | 1 | 1 | 4 | 1 (0)| 00:00:01 | 1 |00:00:00.01 | 2 |
----------------------------------------------------------------------------------------------------------------------------------------------------
Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------
1 - SET$9162BF3C / VW_ORE_BA8ECEFB@SEL$BA8ECEFB
2 - SET$9162BF3C
3 - SET$9162BF3C_1 / T1@SEL$1
4 - SET$9162BF3C_1 / T1@SEL$1
5 - SEL$C90BA1D5
6 - SEL$C90BA1D5 / T1@SEL$1
7 - SEL$C90BA1D5 / T1@SEL$1
8 - SEL$C90BA1D5 / T2@SEL$2
Outline Data
-------------
/*+
BEGIN_OUTLINE_DATA
IGNORE_OPTIM_EMBEDDED_HINTS
OPTIMIZER_FEATURES_ENABLE('12.2.0.1')
DB_VERSION('12.2.0.1')
ALL_ROWS
OUTLINE_LEAF(@"SEL$C90BA1D5")
UNNEST(@"SEL$2")
OUTLINE_LEAF(@"SET$9162BF3C_1")
OUTLINE_LEAF(@"SET$9162BF3C")
OR_EXPAND(@"SEL$1" (1) (2))
OUTLINE_LEAF(@"SEL$BA8ECEFB")
OUTLINE(@"SET$9162BF3C_2")
OUTLINE(@"SEL$2")
OUTLINE(@"SET$9162BF3C")
OR_EXPAND(@"SEL$1" (1) (2))
OUTLINE(@"SEL$1")
NO_ACCESS(@"SEL$BA8ECEFB" "VW_ORE_BA8ECEFB"@"SEL$BA8ECEFB")
INDEX_RS_ASC(@"SET$9162BF3C_1" "T1"@"SEL$1" ("T1"."ID2"))
BATCH_TABLE_ACCESS_BY_ROWID(@"SET$9162BF3C_1" "T1"@"SEL$1")
INDEX_RS_ASC(@"SEL$C90BA1D5" "T1"@"SEL$1" ("T1"."ID1"))
BATCH_TABLE_ACCESS_BY_ROWID(@"SEL$C90BA1D5" "T1"@"SEL$1")
INDEX(@"SEL$C90BA1D5" "T2"@"SEL$2" ("T2"."ID1"))
LEADING(@"SEL$C90BA1D5" "T1"@"SEL$1" "T2"@"SEL$2")
USE_NL(@"SEL$C90BA1D5" "T2"@"SEL$2")
END_OUTLINE_DATA
*/
Predicate Information (identified by operation id):
---------------------------------------------------
4 - access("T1"."ID2"=10)
6 - filter(LNNVL("T1"."ID2"=10))
7 - access("T1"."ID1"=11)
8 - access("T2"."ID1"=11)
filter("T1"."ID1"="T2"."ID1")
--//12c下oracle选择正确的执行计划.可以发现id=2使用UNION-ALL,也就是oracle做了查询转换成union all的形式.
--//另外我曾经尝试将ounline date的提示信息加入到11g环境,执行计划依旧没有选择OR_EXPAND.
--//通过10053事件看看.
SCOTT@test01p> @ 10053x cg5kmfhgczjfd 0
PL/SQL procedure successfully completed.
ORE: after OR Expansion:******* UNPARSED QUERY IS *******
SELECT "VW_ORE_BA8ECEFB"."ITEM_1" "ID1","VW_ORE_BA8ECEFB"."ITEM_2" "ID2","VW_ORE_BA8ECEFB"."ITEM_3" "NAME" FROM ( (SELECT "T1"."ID1" "ITEM_1","T1"."ID2" "ITEM_2","T1"."NAME" "ITEM_3" FROM "SCOTT"."T1" "T1" WHERE "T1"."ID2"=10) UNION ALL (SELECT "T1"."ID1" "ITEM_1","T1"."ID2" "ITEM_2","T1"."NAME" "ITEM_3" FROM "SCOTT"."T1" "T1" WHERE "T1"."ID1"=ANY (SELECT /*+ UNNEST */ "T2"."ID1" "ID1" FROM "SCOTT"."T2" "T2" WHERE "T2"."ID1"=11) AND LNNVL("T1"."ID2"=10))) "VW_ORE_BA8ECEFB"
--//格式化显示如下:
SELECT "VW_ORE_BA8ECEFB"."ITEM_1" "ID1"
,"VW_ORE_BA8ECEFB"."ITEM_2" "ID2"
,"VW_ORE_BA8ECEFB"."ITEM_3" "NAME"
FROM ( (SELECT "T1"."ID1" "ITEM_1"
,"T1"."ID2" "ITEM_2"
,"T1"."NAME" "ITEM_3"
FROM "SCOTT"."T1" "T1"
WHERE "T1"."ID2" = 10)
UNION ALL
(SELECT "T1"."ID1" "ITEM_1"
,"T1"."ID2" "ITEM_2"
,"T1"."NAME" "ITEM_3"
FROM "SCOTT"."T1" "T1"
WHERE "T1"."ID1" = ANY (SELECT /*+ UNNEST */
"T2"."ID1" "ID1"
FROM "SCOTT"."T2" "T2"
WHERE "T2"."ID1" = 11)
AND LNNVL ("T1"."ID2" = 10))) "VW_ORE_BA8ECEFB";
--//也就是oracle查询转换为 UNION ALL的形式.
--//你可以看到第2个条件人为的加入LNNVL ("T1"."ID2" = 10).
--// OR_EXPAND 提示 与 USE_CONCAT 提示到底有什么不同?
5.补充使用USE_CONCAT看到的情况:
select /*+ USE_CONCAT(@"SEL$1" 8 OR_PREDICATES(1)) */ * from t1 where t1.id1 in (select /*+unnest */ t2.id1 from t2 where t2.id1=11 ) or (t1.id2=10 );
SCOTT@test01p> @ 10053x 18h6hkqcqq3w2 0
PL/SQL procedure successfully completed.
--//看这些太烦,不过可以发现如下:
LORE: Or-Expansion validity checks failed on query block SEL$2 (#2) because Cost based OR expansion enabled
SYS@test01p> @ hide or_exp
old 10: and lower(a.ksppinm) like lower('%&1%')
new 10: and lower(a.ksppinm) like lower('%or_exp%')
NAME DESCRIPTION DEFAULT_VALUE SESSION_VALUE SYSTEM_VALUE
---------------------------------- ------------------------------------------------- ------------- ------------- ------------
_no_or_expansion OR expansion during optimization disabled TRUE FALSE FALSE
_optimizer_cbqt_or_expansion enables cost based OR expansion TRUE ON ON
_optimizer_interleave_or_expansion interleave OR Expansion during CBQT TRUE TRUE TRUE
_optimizer_or_expansion control or expansion approach used TRUE DEPTH DEPTH
_optimizer_or_expansion_subheap Use subheap for optimizer or-expansion TRUE TRUE TRUE
_or_expand_nvl_predicate enable OR expanded plan for NVL/DECODE predicate TRUE TRUE TRUE
6 rows selected.
--//也就是12c缺省打开因为以上原因.不过我尝试"_optimizer_cbqt_or_expansion"=off也无效.放弃!!
--//我也尝试提高全表扫描的成本看看是否执行计划会发生改变,不过依旧没用.
SCOTT@test01p> exec dbms_stats.SET_TABLE_STATS(user,'T1',NUMBLKS=>800000000000);
PL/SQL procedure successfully completed.
[20181220]使用提示OR_EXPAND优化.txt的更多相关文章
- [20190524]使用use_concat or_expand提示优化.txt
[20190524]使用use_concat or_expand提示优化.txt --//上午看了链接https://connor-mcdonald.com/2019/05/22/being-gene ...
- [20181116]18c DML 日志优化.txt
[20181116]18c DML 日志优化.txt 1.环境:xxxxxxxx> select banner_full from v$version;BANNER_FULL---------- ...
- [20200223]关于latch and mutext的优化.txt
[20200223]关于latch and mutext的优化.txt --//前一段时间一直在测试使用DBMS_SHARED_POOL.MARKHOT标识热对象以及sql语句的优化.--//有别人问 ...
- solr入门之參考淘宝搜索提示功能优化拼音加汉字搜索功能
首先看一下从淘宝输入搜索keyword获取到的一些数据信息: 第一张:使用拼音的全程来查询 能够看到提示的是匹配的转换的拼音的方式,看最后一个提示项 这里另一个在指定分类文件夹下搜索的功能,难道后台还 ...
- dense_rank()+hash提示改写优化SQL
数据库环境:SQL SERVER 2005 今天看到一条SQL,返回10条数据,执行了50多S.刚好有空,就对它进行了优化,优化后1S出结果. 先看下原始SQL SELECT t1.line_no , ...
- nginx-502错误，老是提示busy.优化php-fpm如下
[global]pid = /usr/local/php/var/run/php-fpm.piderror_log = /usr/local/php/var/log/php-fpm.loglog_le ...
- 一百零三：CMS系统之使用sweetalert提示框优化返回结果
在base模板中引用 在修改密码的js中使用 $(function () { $('#submit').click(function (evnet) { evnet.preventDefault(); ...
- sql 优化 链接提示 查询提示 标提示
SQL Server的查询优化器在select查询执行的时候产生一个高效的查询执行计划.如果优化器不能选择最优的计划,那么就需要检查查询计划.统计信息.支持的索引等,而通过使用提示可以改变优化器选择查 ...
- MySQL优化篇（一），我可以和面试官多聊几句吗？——SQL优化流程与优化数据库对象
我可以和面试官多聊几句吗?只是想偷点技能过来.MySQL优化篇(基于MySQL8.0测试验证),上部分:优化SQL语句.数据库对象,MyISAM表锁和InnoDB锁问题. MyISAM表锁和InnoD ...
随机推荐
- java 8 双冒号运算符
前言 java8增加了双冒号运算符.lambda本质上都是语法糖,学习过C#委托.匿名委托再理解java8中的双冒号运算符就容易多了.双冒号就是把方法当作参数传递给需要的方法,或者说是传递到strea ...
- MySQL之实现Oracle中的rank()函数的功能
假设表格为student, 数据如下: 我们要在MySQL中实现Oracle中的rank()函数功能,即组内排序,具体来说: 就是对student表中按照课程(course)对学生(name) ...
- HTML学习感悟
HTML是一个超文本语言,原本并不打算做网站的我发现学习信息根本离不开web前端的掌握,因此需要对HTML进行一定程度的学习.对了,它可以说是网页的一个标志,打开任何网页我们看到的都是HTML的文本, ...
- 阿里云redis映射到阿里云服务器
参考文档:https://help.aliyun.com/document_detail/43850.html?spm=a2c4g.11186623.2.3.7yg9VH ECS Windows 篇 ...
- [PHP]算法-拼接最小字典序的实现
拼接最小字典序: 给定一个字符串类型的数组strs,请找到一种拼接顺序,使得将所有字符串拼接起来组成的大字符串是所有可能性中字典顺序最小的并放回这个大字符串. 思路: 1.字典序,12345这五个数, ...
- linux内核里的字符串转换 ,链表操作常用函数(转)
1.对双向链表的具体操作如下: list_add ———向链表添加一个条目 list_add_tail ———添加一个条目到链表尾部 __list_del_entry ———从链表中删除相应的条目 l ...
- P9架构师讲解从单机至亿级流量大型网站系统架构的演进过程
阶段一.单机构建网站 网站的初期,我们经常会在单机上跑我们所有的程序和软件.此时我们使用一个容器,如tomcat.jetty.jboos,然后直接使用JSP/servlet技术,或者使用一些开源的框架 ...
- Java学习笔记之——枚举类
枚举可以限定类的值只是有限个,例如:星期,只有星期一到星期天 语法案例:
- 【转】AtomicReference与volatile的区别
来源:AtomicReference与volatile的区别 AtomicReference与volatile的在可见性上的意义是一致的. volatile不能保证原子性,AutomicReferen ...
- java关于字符串的一些实用操作工具类方法
package cn.edu.stdu; import java.util.ArrayList; import java.util.LinkedHashSet; import java.util.Se ...