根据Oracle-L邮件列表里主题「

Full scan vs index

」的讨论而来。

1、测试环境创建

SYS@HEMESRHTDB2(1.206)> select * from v$version;

BANNER

--------------------------------------------------------------------------------

Oracle Database 11g Enterprise Edition Release 11.2.0.3.0 - 64bit Production

PL/SQL Release 11.2.0.3.0 - Production

CORE11.2.0.3.0Production

TNS for Linux: Version 11.2.0.3.0 - Production

NLSRTL Version 11.2.0.3.0 - Production

SYS@HEMESRHTDB2(1.206)>

create table t2

with generator as (

select --+ materialize

rownum pk

from all_objects

where rownum<=4000

)

select

/*+ ordered use_nl(v2)*/

rownum pk,

round(dbms_random.value(1,2)) a,

round(dbms_random.value(1,5)) b,

round(dbms_random.value(1,10)) c,

round(dbms_random.value(1,100)) d,

round(dbms_random.value(1,1000000)) e

from

generator v1,

generator v2

where

rownum<=600000

Table created.

SYS@HEMESRHTDB2(1.206)> select * from t2 where rownum<=100;

PK A B C DE

---------- ---------- ---------- ---------- ---------- ----------

1 2 1 3 80 296354

2 2 3 9 47 531531

3 2 3 1 10 330623

4 2 5 2 35 21138

5 2 5 7 50 425066

6 2 3 9 75 322065

7 2 4 1 93 55360

8 2 1 8 99 378844

9 2 5 8 72 869863

10 2 5 2 63 373369

11 1 4 4 37 313221

12 1 5 8 68 40918

13 1 2 8 48 457786

14 2 3 2 83 316507

15 1 4 2 14 734118

16 1 4 7 59 47266

……

SYS@HEMESRHTDB2(1.206)> create index ix_t2_key on t2(PK,A) online nologging;

Index created.

SYS@HEMESRHTDB2(1.206)> create index ix_t2_D on t2(D) online nologging;

Index created.

SYS@HEMESRHTDB2(1.206)> alter session set statistics_level ='ALL' ;

Session altered.

2、测试4种情况

索引情况。

ix_t2_key on t2(PK,A)

ix_t2_D on t2(D)

Select count(*) from T2 where pk > 520000;

select count(*) from t2 where A=1;

Select count(D) from T2 where pk > 520000;

3、测试

实验1

SYS@HEMESRHTDB2(1.206)> select count(*) from T2 where pk > 520000;

SYS@HEMESRHTDB2(1.206)> select * from TABLE(dbms_xplan.display_cursor(NULL,NULL,'allstats last ADVANCED PEEKED_BINDS'));

PLAN_TABLE_OUTPUT

-------------------------------------

SQL_ID9c98xbdbfww9r, child number 0

-------------------------------------

Select count(*) from T2 where pk > 520000

Plan hash value: 2050414396

实验2

SYS@HEMESRHTDB2(1.206)> select count(*) from t2 where A=1;

COUNT(*)

----------

299737

SYS@HEMESRHTDB2(1.206)> select * from TABLE(dbms_xplan.display_cursor(NULL,NULL,'allstats last ADVANCED PEEKED_BINDS'));

PLAN_TABLE_OUTPUT

-------------------------------------

SQL_ID651cjf8pmhb51, child number 0

-------------------------------------

select count(*) from t2 where A=1

Plan hash value: 2933116225

实验3

SYS@HEMESRHTDB2(1.206)> select count(D) from t2 where pk>=520000;

COUNT(D)

----------

80001

SYS@HEMESRHTDB2(1.206)> select * from TABLE(dbms_xplan.display_cursor(NULL,NULL,'allstats last ADVANCED PEEKED_BINDS'));

PLAN_TABLE_OUTPUT

-------------------------------------

SQL_ID41uuvuyutgn6q, child number 0

-------------------------------------

select count(D) from t2 where pk>=520000

Plan hash value: 3321871023

4、结论和延伸

实验1

SYS@HEMESRHTDB2(1.206)> select count(*) from T2 where pk > 520000;

按照预期走索引IX_T2_KEY。

实验2

SYS@HEMESRHTDB2(1.206)> select count(*) from t2 where A=1;

也同预期。

实验3

SYS@HEMESRHTDB2(1.206)> select count(D) from t2 where pk>=520000;

为啥变成count(D)就完全不一样了？

我们加上index hint看看效果如何？

延伸实验 Index hint

SYS@HEMESRHTDB2(1.206)> select /*+ index(T2 IX_T2_KEY) */ count(D) from t2 where pk>=520000;

COUNT(D)

----------

80001

SYS@HEMESRHTDB2(1.206)> select * from TABLE(dbms_xplan.display_cursor(NULL,NULL,'allstats last ADVANCED PEEKED_BINDS'));

PLAN_TABLE_OUTPUT

-------------------------------------

SQL_ID4s4zzmrzdzrbt, child number 0

-------------------------------------

select /*+ index(T2 IX_T2_KEY) */ count(D) from t2 where pk>=520000

Plan hash value: 948933721

这儿看起来好像有些眉目了，Cost虽然比全表扫描的要大，但真正耗用的buffers根本就不大。

问题出在统计信息！！？

SYS@HEMESRHTDB2(1.206)>

select

column_name,

num_distinct,

histogram, num_buckets,

to_char(LAST_ANALYZED, 'yyyy-mm-dd hh24:mi:ss')

from all_tab_col_statistics

where upper(table_name)='T2';

no rows selected

果然咧，这时收集下统计信息。

SYS@HEMESRHTDB2(1.206)>

begin

dbms_stats.gather_table_stats(

ownname => user,

tabname => 'T2',

estimate_percent =>100,

cascade => true);

end;

PL/SQL procedure successfully completed.

「for all columns 」或者「for all indexed columns」都OK，重要的是覆盖count(D) 字段，CBO才能够计算出正确的cost。

可以加个/*1*/之类的使执行计划重新解析，或是alter system flush shared_pool;

嗯，随你喜欢。

SYS@HEMESRHTDB2(1.206)> select /*1*/ count(D) from t2 where pk>=520000;

SYS@HEMESRHTDB2(1.206)> select * from TABLE(dbms_xplan.display_cursor(NULL,NULL,'allstats last ADVANCED PEEKED_BINDS'));

PLAN_TABLE_OUTPUT

-------------------------------------

SQL_ID6z1bc6xv0anrn, child number 0

-------------------------------------

select /*1*/ count(D) from t2 where pk>=520000

Plan hash value: 948933721

这个时候，便会选择正确的索引了，

当然，你也可以把D字段包含入索引IX_T2_KEY中。

最后：

如果扫描的范围再大一些？会发生什么？

10万/60万

SYS@HEMESRHTDB2(1.206)> select /*2*/ count(D) from t2 where pk>=500000;

COUNT(D)

----------

100001

SYS@HEMESRHTDB2(1.206)> select * from TABLE(dbms_xplan.display_cursor(NULL,NULL,'allstats last ADVANCED PEEKED_BINDS'));

PLAN_TABLE_OUTPUT

-------------------------------------

SQL_ID7shudyqdb5nbk, child number 0

-------------------------------------

select /*2*/ count(D) from t2 where pk>=500000

Plan hash value: 3321871023

转载请注明出处
作者:daaprk
链接:http://www.cnblogs.com/dap570/p/3396513.html
本作品采用知识共享署名-非商业性使用-禁止演绎 3.0 未本地化版本许可协议进行许可。