oracle中not exists对外层查询的影响详解

2024-02-29 00:49:03 69

前言

最近同事发现了一个问题，在12c中跑的bufferget很高，但是在10g中跑的buffer很低。怀疑是不是12c的优化器有问题。

这个10g的环境和12c的环境，数据量大致一样，只是有很少部分的不同，但是就是这个很少部分不同，造成了notexists中的子查询返回不同的值，进而对外层查询产生不同的影响。

我们来用如下的代码模拟一下。

初始化数据：

--10g
droptablet1;
droptablet2;

createtablet1(idnumber,namevarchar2(20),dep_idvarchar2(10));
createtablet2(idnumber,namevarchar2(20),dep_idvarchar2(10));

insertintot1selectrownum,'a','kk'fromdualconnectbylevel<=3000000;
insertintot2selectrownum,'a','kk'fromdualconnectbylevel<=1000000;
insertintot2selectrownum,'a','mm'fromdual;

commit;


--12c
droptablet1;
droptablet2;

createtablet1(idnumber,namevarchar2(20),dep_idvarchar2(10));
createtablet2(idnumber,namevarchar2(20),dep_idvarchar2(10));


insertintot1selectrownum,'a','kk'fromdualconnectbylevel<=3000000;
insertintot2selectrownum,'a','kk'fromdualconnectbylevel<=1000000;

commit;

我们看到，12c的数据和10g只是有很少的差别，t1表12c和10g都一样，t2表在12c只是少了一行数据。

--10g
SQL>selectdep_id,count(*)fromt1groupbydep_id;

DEP_IDCOUNT(*)
------------------------------
kk3000000

SQL>selectdep_id,count(*)fromt2groupbydep_id;

DEP_IDCOUNT(*)
------------------------------
mm1
kk1000000

SQL>


--12c
SQL>selectdep_id,count(*)fromt1groupbydep_id;

DEP_IDCOUNT(*)
------------------------------
kk3000000

SQL>selectdep_id,count(*)fromt2groupbydep_id;

DEP_IDCOUNT(*)
------------------------------
kk1000000

SQL>

我们将要执行的sql语句是：

selectcount(*)
fromt1,t2
wheret1.id=t2.id
andt1.dep_id='kk'
andnotexists(select1
fromt1,t2
wheret1.id=t2.id
andt2.dep_id='mm');

我们先来看执行情况的差距，10g的bufferget小，12c多：

--10g
SQL>select/*+gather_plan_statistics*/count(*)fromt1,t2wheret1.id=t2.idandt1.dep_id='kk'andnotexists(select1fromt1,t2wheret1.id=t2.idandt2.dep_id='mm');

COUNT(*)
----------
0

SQL>select*fromtable(dbms_xplan.display_cursor(null,null,'ALLSTATSLAST'));

PLAN_TABLE_OUTPUT
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID22t5mb43w55pr,childnumber0
-------------------------------------
select/*+gather_plan_statistics*/count(*)fromt1,t2wheret1.id=t2.idandt1.dep_id='kk'andnot
exists(select1fromt1,t2wheret1.id=t2.idandt2.dep_id='mm')

Planhashvalue:3404612428

------------------------------------------------------------------------------------------------------------------
|Id|Operation|Name|Starts|E-Rows|A-Rows|A-Time|Buffers|OMem|1Mem|Used-Mem|
------------------------------------------------------------------------------------------------------------------
|0|SELECTSTATEMENT||1||1|00:00:00.02|2086||||
|1|SORTAGGREGATE||1|1|1|00:00:00.02|2086||||
|*2|FILTER||1||0|00:00:00.02|2086||||
|*3|HASHJOIN||0|901K|0|00:00:00.01|0|39M|5518K||
|4|TABLEACCESSFULL|T2|0|901K|0|00:00:00.01|0||||
|*5|TABLEACCESSFULL|T1|0|2555K|0|00:00:00.01|0||||
|*6|HASHJOIN||1|23|1|00:00:00.02|2086|1517K|1517K|612K(0)|
|*7|TABLEACCESSFULL|T2|1|23|1|00:00:00.02|2082||||
|8|TABLEACCESSFULL|T1|1|2555K|1|00:00:00.01|4||||
------------------------------------------------------------------------------------------------------------------

PredicateInformation(identifiedbyoperationid):
---------------------------------------------------

2-filter(ISNULL)
3-access("T1"."ID"="T2"."ID")
5-filter("T1"."DEP_ID"='kk')
6-access("T1"."ID"="T2"."ID")
7-filter("T2"."DEP_ID"='mm')

Note
-----
-dynamicsamplingusedforthisstatement


34rowsselected.

SQL>


--12c
SQL>select/*+gather_plan_statistics*/count(*)fromt1,t2wheret1.id=t2.idandt1.dep_id='kk'andnotexists(select1fromt1,t2wheret1.id=t2.idandt2.dep_id='mm');

COUNT(*)
----------
1000000

SQL>select*fromtable(dbms_xplan.display_cursor(null,null,'ALLSTATSLAST'));

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID22t5mb43w55pr,childnumber0
-------------------------------------
select/*+gather_plan_statistics*/count(*)fromt1,t2where
t1.id=t2.idandt1.dep_id='kk'andnotexists(select1fromt1,t2
wheret1.id=t2.idandt2.dep_id='mm')

Planhashvalue:1692274438

--------------------------------------------------------------------------------------------------------------------
|Id|Operation|Name|Starts|E-Rows|A-Rows|A-Time|Buffers|OMem|1Mem|Used-Mem|
--------------------------------------------------------------------------------------------------------------------
|0|SELECTSTATEMENT||1||1|00:00:00.79|10662||||
|1|SORTAGGREGATE||1|1|1|00:00:00.79|10662||||
|*2|FILTER||1||1000K|00:00:00.74|10662||||
|*3|HASHJOIN||1|1215K|1000K|00:00:00.52|8579|43M|6111K|42M(0)|
|4|TABLEACCESSFULL|T2|1|1215K|1000K|00:00:00.01|2083||||
|*5|TABLEACCESSFULL|T1|1|2738K|3000K|00:00:00.07|6496||||
|*6|HASHJOINRIGHTSEMI||1|35|0|00:00:00.02|2083|1245K|1245K|461K(0)|
|*7|TABLEACCESSFULL|T2|1|23|0|00:00:00.02|2083||||
|8|TABLEACCESSFULL|T1|0|2738K|0|00:00:00.01|0||||
--------------------------------------------------------------------------------------------------------------------

PredicateInformation(identifiedbyoperationid):
---------------------------------------------------

2-filter(ISNULL)
3-access("T1"."ID"="T2"."ID")
5-filter("T1"."DEP_ID"='kk')
6-access("T1"."ID"="T2"."ID")
7-filter("T2"."DEP_ID"='mm')

Note
-----
-dynamicstatisticsused:dynamicsampling(level=2)


35rowsselected.

SQL>
SQL>

可以看到第23，24行，在10g中运行时，buffers是0，而在12c中，即78，79行，buffer是2083+6496。

也就是说在10g中，外层查询不进行t1和t2的扫描，直接返回结果了，而在12c中，外层查询还要进行t1表和t2表层扫描才返回结果。

这其实不是10g和12c的差别，而是notexists的返回数据对外层的影响。子查询要返回0行记录，才满足notexist的条件，从而返回外层查询结果。

在10g中，子查询返回了一行记录

--10g
SQL>select1fromt1,t2wheret1.id=t2.idandt2.dep_id='mm';

1
----------
1

SQL>

不满足notexists（即0行才满足），所以，也就不用在外层继续查询了。直接返回记录0行。

在12c中，子查询返回0行记录，满足notexist的条件，所以还需要在外层查询中继续查询。

--12c
SQL>selectcount(*)fromt1,t2wheret1.id=t2.idandt2.dep_id='kk';

COUNT(*)
----------
1000000

SQL>setline1000
SQL>setpages1000
SQL>colPLAN_TABLE_OUTPUTfora250
SQL>
SQL>
SQL>select/*+gather_plan_statistics*/count(*)fromt1,t2wheret1.id=t2.idandt1.dep_id='kk'andnotexists(select1fromt1,t2wheret1.id=t2.idandt2.dep_id='kk');

COUNT(*)
----------
0

SQL>select*fromtable(dbms_xplan.display_cursor(null,null,'ALLSTATSLAST'));

PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_IDc5hj2p2jt1fxf,childnumber0
-------------------------------------
select/*+gather_plan_statistics*/count(*)fromt1,t2where
t1.id=t2.idandt1.dep_id='kk'andnotexists(select1fromt1,t2
wheret1.id=t2.idandt2.dep_id='kk')

Planhashvalue:1692274438

--------------------------------------------------------------------------------------------------------------------
|Id|Operation|Name|Starts|E-Rows|A-Rows|A-Time|Buffers|OMem|1Mem|Used-Mem|
--------------------------------------------------------------------------------------------------------------------
|0|SELECTSTATEMENT||1||1|00:00:00.28|2087||||
|1|SORTAGGREGATE||1|1|1|00:00:00.28|2087||||
|*2|FILTER||1||0|00:00:00.28|2087||||
|*3|HASHJOIN||0|1215K|0|00:00:00.01|0|69M|7428K||
|4|TABLEACCESSFULL|T2|0|1215K|0|00:00:00.01|0||||
|*5|TABLEACCESSFULL|T1|0|2738K|0|00:00:00.01|0||||
|*6|HASHJOINRIGHTSEMI||1|2738K|1|00:00:00.28|2087|43M|6111K|42M(0)|
|*7|TABLEACCESSFULL|T2|1|1215K|1000K|00:00:00.12|2083||||
|8|TABLEACCESSFULL|T1|1|2738K|1|00:00:00.01|4||||
--------------------------------------------------------------------------------------------------------------------

PredicateInformation(identifiedbyoperationid):
---------------------------------------------------

2-filter(ISNULL)
3-access("T1"."ID"="T2"."ID")
5-filter("T1"."DEP_ID"='kk')
6-access("T1"."ID"="T2"."ID")
7-filter("T2"."DEP_ID"='kk')

Note
-----
-dynamicstatisticsused:dynamicsampling(level=2)


35rowsselected.

SQL>

可以看到第38，39行的buffer为0.

总结

以上就是这篇文章的全部内容了，希望本文的内容对大家的学习或者工作能带来一定的帮助，如果有疑问大家可以留言交流。

oracle中not exists对外层查询的影响详解

热门推荐

随机推荐