将主查询条件写到子查询里执行效果会怎样？

--测试环境
create table t1 as select object_id,object_name from dba_objects;
create table t2 as select object_id,object_name from user_objects;
create table t3 as select rownum object_id,table_name object_name from user_tables;
analyze table t1 compute statistics for table for all columns;
analyze table t2 compute statistics for table for all columns;
analyze table t3 compute statistics for table for all columns;
t1--49830;t2--37;t3--16
开始测试：
1.比较简单的子查询：子查询中表直接和主表就有关联关系
select /*+gather_plan_statistics*/
 (select t2.object_name from t2 where t2.object_id = t1.object_id) object_name
  from t1;
执行计划：
---------------------------------------------------------------------------------------------
| Id  | Operation         | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers | Reads  |
---------------------------------------------------------------------------------------------
|*  1 |  TABLE ACCESS FULL| T2   |  49830 |      1 |     36 |00:00:01.75 |     149K|      0 |
|   2 |  TABLE ACCESS FULL| T1   |      1 |  49830 |  49830 |00:00:00.30 |    3546 |    231 |
---------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   1 - filter("T2"."OBJECT_ID"=:B1)
这个执行很简单，执行t1，然后根据t1返回行数决定t2扫描次数。

2.复杂点的子查询：子查询的子查询和主表有关联关系
select /*+gather_plan_statistics*/
 (select t2.object_name
    from t2
   where t2.object_name in
         (select t3.object_name from t3 where t1.object_id = t3.object_id)) object_name
  from t1;
执行计划：观看执行次数和实际返回行数  
----------------------------------------------------------------------------------------------
| Id  | Operation          | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers | Reads  |
----------------------------------------------------------------------------------------------
|*  1 |  FILTER            |      |  49830 |        |     15 |00:00:39.89 |    5680K|      2 |
|   2 |   TABLE ACCESS FULL| T2   |  49830 |     37 |   1843K|00:00:07.89 |     149K|      0 |
|*  3 |   TABLE ACCESS FULL| T3   |   1843K|      1 |     15 |00:00:27.95 |    5531K|      2 |
|   4 |  TABLE ACCESS FULL | T1   |      1 |  49830 |  49830 |00:00:00.20 |    3546 |      0 |
----------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   1 - filter( EXISTS (SELECT /*+ */ 0 FROM "T3" "T3" WHERE
              "T3"."OBJECT_NAME"=:B1 AND "T3"."OBJECT_ID"=:B2))
   3 - filter("T3"."OBJECT_NAME"=:B1 AND "T3"."OBJECT_ID"=:B2)
首先执行t1，然后执行t2，最后执行t3（从predicate中3可以看出，t3依赖于t1传过来的参数object_id和t2传过来的参数object_name）。
仔细分析下Starts和A-Rows发现，执行t1返回49830行，然后执行t2，由于t2和t1没有任何关联条件，导致执行t2后返回的行数是t2数据行数与t2执行次数（t1返回行数）相乘；
这也就直接影响到t3执行次数。也就是说不管t3中有多少行满足条件的数据，都会对表t3扫描t1和t2行数相乘次。而如果t1和t2返回行数相当大，那么结果就可想而知了~

来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/24496749/viewspace-731888/，如需转载，请注明出处，否则将追究法律责任。

转载于:http://blog.itpub.net/24496749/viewspace-731888/