前言
最近同事发现了一个问题，在12c中跑的buffer get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not exists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：
create table t1 (id number,name varchar2(20),dep_id varchar2(10));
create table t2 (id number,dep_id varchar2(10));
insert into t1 select rownum,'a','kk' from dual connect by level <=3000000;
insert into t2 select rownum,'kk' from dual connect by level <=1000000;
insert into t2 select rownum,'mm' from dual;
commit;
--12c
drop table t1;
drop table t2;
create table t1 (id number,dep_id varchar2(10));
insert into t1 select rownum,'kk' from dual connect by level <=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。
DEP_ID COUNT(*)
kk 3000000
sql> select dep_id,count(*) from t2 group by dep_id;
DEP_ID COUNT(*)
mm 1
kk 1000000
sql>
--12c
sql> select dep_id,count(*) from t2 group by dep_id;
DEP_ID COUNT(*)
kk 1000000
sql>
我们将要执行的sql语句是：
我们先来看执行情况的差距，10g的bufferget小，12c多：
COUNT(*)
0
sql> select* from table(dbms_xplan.display_cursor(null,null,'ALLSTATS LAST'));
PLAN_TABLE_OUTPUT
sql_ID 22t5mb43w55pr,child number 0
select /+ gather_plan_statistics / count(*) from t1,t2 where t1.id=t2.id and t1.dep_id='kk' and not
exists (select 1 from t1,t2 where t1.id=t2.id and t2.dep_id='mm')
Plan hash value: 3404612428
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem |
| 0 | SELECT STATEMENT | | 1 | | 1 |00:00:00.02 | 2086 | | | |
| 1 | SORT AGGREGATE | | 1 | 1 | 1 |00:00:00.02 | 2086 | | | |
| 2 | FILTER | | 1 | | 0 |00:00:00.02 | 2086 | | | |
| 3 | HASH JOIN | | 0 | 901K| 0 |00:00:00.01 | 0 | 39M| 5518K| |
| 4 | TABLE ACCESS FULL| T2 | 0 | 901K| 0 |00:00:00.01 | 0 | | | |
| 5 | TABLE ACCESS FULL| T1 | 0 | 2555K| 0 |00:00:00.01 | 0 | | | |
| 6 | HASH JOIN | | 1 | 23 | 1 |00:00:00.02 | 2086 | 1517K| 1517K| 612K (0)|
|* 7 | TABLE ACCESS FULL| T2 | 1 | 23 | 1 |00:00:00.02 | 2082 | | | |
| 8 | TABLE ACCESS FULL| T1 | 1 | 2555K| 1 |00:00:00.01 | 4 | | | |
Predicate Information (identified by operation id):
2 - filter( IS NULL)
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
- dynamic sampling used for this statement
34 rows selected.
sql>
--12c
sql> select /+ gather_plan_statistics / count(*) from t1,t2 where t1.id=t2.id and t2.dep_id='mm');
COUNT(*)
1000000
sql> select* from table(dbms_xplan.display_cursor(null,'ALLSTATS LAST'));
PLAN_TABLE_OUTPUT
sql_ID 22t5mb43w55pr,t2 where
t1.id=t2.id and t1.dep_id='kk' and not exists (select 1 from t1,t2
where t1.id=t2.id and t2.dep_id='mm')
Plan hash value: 1692274438
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem |
| 0 | SELECT STATEMENT | | 1 | | 1 |00:00:00.79 | 10662 | | | |
| 1 | SORT AGGREGATE | | 1 | 1 | 1 |00:00:00.79 | 10662 | | | |
| 2 | FILTER | | 1 | | 1000K|00:00:00.74 | 10662 | | | |
| 3 | HASH JOIN | | 1 | 1215K| 1000K|00:00:00.52 | 8579 | 43M| 6111K| 42M (0)|
| 4 | TABLE ACCESS FULL | T2 | 1 | 1215K| 1000K|00:00:00.01 | 2083 | | | |
| 5 | TABLE ACCESS FULL | T1 | 1 | 2738K| 3000K|00:00:00.07 | 6496 | | | |
| 6 | HASH JOIN RIGHT SEMI| | 1 | 35 | 0 |00:00:00.02 | 2083 | 1245K| 1245K| 461K (0)|
|* 7 | TABLE ACCESS FULL | T2 | 1 | 23 | 0 |00:00:00.02 | 2083 | | | |
| 8 | TABLE ACCESS FULL | T1 | 0 | 2738K| 0 |00:00:00.01 | 0 | | | |
Predicate Information (identified by operation id):
2 - filter( IS NULL)
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
- dynamic statistics used: dynamic sampling (level=2)
35 rows selected.
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not exists的返回数据对外层的影响。子查询要返回0行记录，才满足not exist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
不满足not exists（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not exist的条件，所以还需要在外层查询中继续查询。
COUNT(*)
1000000
sql> set line 1000
sql> set pages 1000
sql> col PLAN_TABLE_OUTPUT for a250
sql>
sql>
sql> select /+ gather_plan_statistics / count(*) from t1,t2 where t1.id=t2.id and t2.dep_id='kk');
COUNT(*)
0
sql> select* from table(dbms_xplan.display_cursor(null,'ALLSTATS LAST'));
PLAN_TABLE_OUTPUT
sql_ID c5hj2p2jt1fxf,t2
where t1.id=t2.id and t2.dep_id='kk')
Plan hash value: 1692274438
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem |
| 0 | SELECT STATEMENT | | 1 | | 1 |00:00:00.28 | 2087 | | | |
| 1 | SORT AGGREGATE | | 1 | 1 | 1 |00:00:00.28 | 2087 | | | |
| 2 | FILTER | | 1 | | 0 |00:00:00.28 | 2087 | | | |
| 3 | HASH JOIN | | 0 | 1215K| 0 |00:00:00.01 | 0 | 69M| 7428K| |
| 4 | TABLE ACCESS FULL | T2 | 0 | 1215K| 0 |00:00:00.01 | 0 | | | |
| 5 | TABLE ACCESS FULL | T1 | 0 | 2738K| 0 |00:00:00.01 | 0 | | | |
| 6 | HASH JOIN RIGHT SEMI| | 1 | 2738K| 1 |00:00:00.28 | 2087 | 43M| 6111K| 42M (0)|
|* 7 | TABLE ACCESS FULL | T2 | 1 | 1215K| 1000K|00:00:00.12 | 2083 | | | |
| 8 | TABLE ACCESS FULL | T1 | 1 | 2738K| 1 |00:00:00.01 | 4 | | | |
Predicate Information (identified by operation id):
2 - filter( IS NULL)
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
- dynamic statistics used: dynamic sampling (level=2)
35 rows selected.
sql>
可以看到第38，39行的buffer为0.
