8.1.1 构造数据环境
drop table test1;
drop table test2;
drop table test3;
create table test1 as select * from dba_objects;
create table test2 as select * from test1;
create table test3 as select * from test1;
insert into test2 select * from test2;
insert into test2 select * from test2;
insert into test3 select * from test3;
insert into test3 select * from test3;
create index ix_test1_01 on test1(object_id);
create index ix_test2_01 on test2(object_id);
create index ix_test3_01 on test3(object_id);
8.1.2 问题SQL
select count(t1.OWNER),count(t.OBJECT_TYPE)
from test1 t1
inner join (select * from test2 t2
union all
select * from test3 t3) t
on t1.object_id=t.object_id
where t1.OWNER='SCOTT';
在最后count之前只返回208条数据，但是逻辑读时9000+，还有20个物理读，因为t1.owner=’SCOTT’返回数据量很少，object_id列还有索引，完全可以走嵌套循环，但是，连接列谓词没有推入，造成了大表走Hash。
8.1.3 优化方法1 hint
select /*+ PUSH_PRED(t)*/count(t1.OWNER),count(t.OBJECT_TYPE)
from test1 t1
inner join (select /*+ index(t2 ix_test2_01)*/ * from test2 t2
union all
select /*+ index(t3 ix_test3_01)*/ * from test3 t3) t
on t1.object_id=t.object_id
where t1.OWNER='SCOTT';
8.1.4 优化方法2 拆分union
select count(OWNER),count(OBJECT_TYPE) from
(select t1.OWNER,t2.OBJECT_TYPE
from test1 t1
inner join test2 t2
on t1.object_id=t2.object_id
where t1.OWNER='SCOTT'
union all
select t1.OWNER,t3.OBJECT_TYPE
from test1 t1
inner join test3 t3
on t1.object_id=t3.object_id
where t1.OWNER='SCOTT');
因为要访问test1两次，性能比直接加hint稍微差一些。
8.1.5 直接在视图内部连接
select count(t1.OWNER),count(t.OBJECT_TYPE)
from test1 t1
inner join (select t2.*
from test2 t2 ,test1 t1
where t1.OWNER='SCOTT'
and t1.object_id=t2.object_id
union all
select t3.*
from test3 t3,test1 t1
where t1.OWNER='SCOTT'
and t1.object_id=t3.object_id) t
on t1.object_id=t.object_id
where t1.OWNER='SCOTT';
这个要访问test1 三次，比直接hint也差一些。
8.1.6 共同部分用with
with tmp as
(select /*+ NO_MERGE*/ * from test1 t1 where t1.OWNER='SCOTT')
select count(OWNER),count(OBJECT_TYPE) from
(select t1.OWNER,t2.OBJECT_TYPE
from tmp t1
inner join test2 t2
on t1.object_id=t2.object_id
union all
select t1.OWNER,t3.OBJECT_TYPE
from tmp t1
inner join test3 t3
on t1.object_id=t3.object_id);
这个执行计划比较长，就不贴了。逻辑读更低了，但是有1个物理读，不知道为什么。