包含union all的view子查询无法展开表连接的模拟

之前小鱼写过一些关于子查询无法展开的文章，其中主要涉及到以下几点： 1关联的列没有not null约束，在子查询中使用not in去查找（oracle 10g无法展开为子查询，oracle 11g由于有新的hash join算法已经可以），如下 SQL> select a.* from ta01 a where a.obj

之前小鱼写过一些关于子查询无法展开的文章，其中主要涉及到以下几点：<br> 1关联的列没有not null约束，在子查询中使用not in去查找（oracle 10g无法展开为子查询，oracle 11g由于有新的hash join算法已经可以），如下

SQL> select a.* from ta01 a where a.object_id not in (select b.object_id from ta02 b)
Ta01和ta02表的object_id列上没有not null的约束

2 子查询中包含主表的关联列的谓词过滤（oracle 10g也无法展开为子查询，oracle 11g可以展开为子查询），如下

SQL> select a.* from ta01 a where a.object_id in （select b.object_id from ta02 b where a.object_type=’TABLE’）；

3 子查询中包含有下列形式的or的查询（oracle 10g同样无法展开为子查询，oracle 11g可以展开）

SQL> select a.* from ta01 a where a.object_id in (select b.object_id from ta02 b
where object_type='TABLE') or a.object_id in (select c.object_id from ta02 c wh
ere c.object_type='INDEX');

4 子查询包含有主表和子表列的like关联，比如下列sql语句（oracle 10g和11g都无法展开为子查询）

SQL> select a.* from ta01 a where a.object_id in (select b.object_id from ta02 b
where a.data_object_id like b.data_object_id);

http://www.eygle.com/archives/2013/04/2013_oracle_dtcc_dbsnake.html

eygle这篇文章里面又提到了一个比较常见的子查询无法展开的case，借鉴作者的案例进行一点简单的模拟来进行验证。

SQL> select * from v$version;

BANNER
----------------------------------------------------------------
Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 - 64bi
PL/SQL Release 10.2.0.4.0 - Production
CORE 10.2.0.4.0 Production
TNS for 64-bit Windows: Version 10.2.0.4.0 - Production
NLSRTL Version 10.2.0.4.0 - Production

SQL> create table ta01 as select * from dba_objects;

Table created.

SQL> create table ta02 as select * from dba_objects;

Table created.

SQL> create table tb01 as select * from dba_objects;

Table created.

SQL> create table tb02 as select * from dba_objects;

Table created.

SQL> create view ta_view as
2 select object_id,data_object_id,object_type from ta01 union all
3 select object_id,data_object_id,object_type from ta02;

View created.

SQL> create view tb_view as
2 select object_id,data_object_id,object_type from tb01 union all
3 select object_id,data_object_id,object_type from tb02;

View created.

SQL> select a.* from ta_view a where a.object_id in (select b.object_id from tb_
view b where b.object_type='TABLE');

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
------------------------------------------------------------

Plan hash value: 3623909176

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

| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |

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

| 0 | SELECT STATEMENT | | 3738K| 131M| 617 (1)| 00:00:08 |

|* 1 | FILTER | | | | | |

| 2 | VIEW | TA_VIEW | 88401 | 3194K| 308 (1)| 00:00:04 |

| 3 | UNION-ALL | | | | | |

| 4 | TABLE ACCESS FULL| TA01 | 42509 | 1535K| 154 (1)| 00:00:02 |

| 5 | TABLE ACCESS FULL| TA02 | 45892 | 1658K| 154 (1)| 00:00:02 |

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
------------------------------------------------------------
| 6 | VIEW | TB_VIEW | 42 | 1008 | 309 (1)| 00:00:04 |

| 7 | UNION-ALL | | | | | |

|* 8 | TABLE ACCESS FULL| TB01 | 17 | 408 | 154 (1)| 00:00:02 |

|* 9 | TABLE ACCESS FULL| TB02 | 25 | 600 | 154 (1)| 00:00:02 |

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

Predicate Information (identified by operation id):
---------------------------------------------------

1 - filter( EXISTS (SELECT /*+ */ 0 FROM ( (SELECT /*+ */
"OBJECT_ID" "OBJECT_ID","DATA_OBJECT_ID" "DATA_OBJECT_ID","OBJECT_
TYPE"

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
------------------------------------------------------------
"OBJECT_TYPE" FROM SYS."TB01" "TB01" WHERE "OBJECT_ID"=:B1 AND
"OBJECT_TYPE"='TABLE') UNION ALL (SELECT /*+ */ "OBJECT_ID"
"OBJECT_ID","DATA_OBJECT_ID" "DATA_OBJECT_ID","OBJECT_TYPE"
"OBJECT_TYPE" FROM SYS."TB02" "TB02" WHERE "OBJECT_ID"=:B2 AND
"OBJECT_TYPE"='TABLE')) "B"))
8 - filter("OBJECT_ID"=:B1 AND "OBJECT_TYPE"='TABLE')
9 - filter("OBJECT_ID"=:B1 AND "OBJECT_TYPE"='TABLE')

Note
-----
- dynamic sampling used for this statement

33 rows selected.

Statistics
----------------------------------------------------------
146 recursive calls
0 db block gets
136371691 consistent gets
0 physical reads
0 redo size
83259 bytes sent via SQL*Net to client
2868 bytes received via SQL*Net from client
218 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
3247 rows processed

这里由于主查询和子查询的表都是两个表相关联的视图，这个导致in之后的括号内的子查询并不是一个常量的集合，所以oracle这里没有对该子查询展开为表连接。

而我们可以改写为下列的sql语句，但是这里好像结果集不一样，这个是因为我们选取的表不是1对1引起的，这个不在我们考虑的范围内。
SQL> SELECT a.*
2 FROM ta_view a,
3 (SELECT object_id
4 FROM tb_view
5 WHERE object_type = 'TABLE') b
6 WHERE a.object_id = b.object_id;

6494 rows selected.

Execution Plan
----------------------------------------------------------
Plan hash value: 48097912

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

| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |

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

| 0 | SELECT STATEMENT | | 3328K| 158M| 640 (5)| 00:00:08 |

|* 1 | HASH JOIN | | 3328K| 158M| 640 (5)| 00:00:08 |

| 2 | VIEW | TB_VIEW | 3306 | 42978 | 309 (1)| 00:00:04 |

| 3 | UNION-ALL | | | | | |

|* 4 | TABLE ACCESS FULL| TB01 | 1756 | 24584 | 154 (1)| 00:00:02 |

|* 5 | TABLE ACCESS FULL| TB02 | 1550 | 21700 | 154 (1)| 00:00:02 |

| 6 | VIEW | TA_VIEW | 100K| 3637K| 308 (1)| 00:00:04 |

| 7 | UNION-ALL | | | | | |

| 8 | TABLE ACCESS FULL| TA01 | 50333 | 786K| 154 (1)| 00:00:02 |

| 9 | TABLE ACCESS FULL| TA02 | 50334 | 786K| 154 (1)| 00:00:02 |

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

Predicate Information (identified by operation id):
---------------------------------------------------

1 - access("A"."OBJECT_ID"="OBJECT_ID")
4 - filter("OBJECT_TYPE"='TABLE')
5 - filter("OBJECT_TYPE"='TABLE')

Statistics
----------------------------------------------------------
1083 recursive calls
0 db block gets
3412 consistent gets
2782 physical reads
196 redo size
139022 bytes sent via SQL*Net to client
5244 bytes received via SQL*Net from client
434 SQL*Net roundtrips to/from client
31 sorts (memory)
0 sorts (disk)
6494 rows processed

当然如果业务也是1对多，那么这个改写是不符合业务逻辑的，那么是否就不能进行改写了，其实也不是这样的，我们只需要在构造的这个表中加上一个distinct去掉b表的重复的值，这样不就是类似为半连接了吗。
SQL> SELECT a.*
2 FROM ta_view a,
3 (SELECT distinct object_id
4 FROM tb_view
5 WHERE object_type = 'TABLE') b
6 WHERE a.object_id = b.object_id;

3247 rows selected.

Execution Plan
----------------------------------------------------------
Plan hash value: 1607119383

--------------------------------------------------------------------------------
--
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
--------------------------------------------------------------------------------
--
| 0 | SELECT STATEMENT | | 3328K| 158M| 642 (5)| 00:00:08
|
|* 1 | HASH JOIN | | 3328K| 158M| 642 (5)| 00:00:08
|
| 2 | VIEW | | 3306 | 42978 | 310 (2)| 00:00:04
|
| 3 | HASH UNIQUE | | 3306 | 42978 | 310 (2)| 00:00:04
|
| 4 | VIEW | TB_VIEW | 3306 | 42978 | 309 (1)| 00:00:04
|
| 5 | UNION-ALL | | | | |
|
|* 6 | TABLE ACCESS FULL| TB01 | 1756 | 24584 | 154 (1)| 00:00:02
|
|* 7 | TABLE ACCESS FULL| TB02 | 1550 | 21700 | 154 (1)| 00:00:02
|
| 8 | VIEW | TA_VIEW | 100K| 3637K| 308 (1)| 00:00:04
|
| 9 | UNION-ALL | | | | |
|
| 10 | TABLE ACCESS FULL | TA01 | 50333 | 786K| 154 (1)| 00:00:02
|
| 11 | TABLE ACCESS FULL | TA02 | 50334 | 786K| 154 (1)| 00:00:02
|
--------------------------------------------------------------------------------
--

Predicate Information (identified by operation id):
---------------------------------------------------

1 - access("A"."OBJECT_ID"="B"."OBJECT_ID")
6 - filter("OBJECT_TYPE"='TABLE')
7 - filter("OBJECT_TYPE"='TABLE')

Statistics
----------------------------------------------------------
15 recursive calls
0 db block gets
3007 consistent gets
0 physical reads
0 redo size
83259 bytes sent via SQL*Net to client
2868 bytes received via SQL*Net from client
218 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
3247 rows processed

关于我们常见的select a.* from a,b where a.id=b.id和select a.* from a where a.id in (select b.id from b)这两种查询，在业务层面有不同的是如果b表没有重复的id，那么这个查询业务逻辑是相同的，而如果b表有重复的，子查询是hash join的半连接方式，会对b.id进行去重，所以如果b.id存在多个重复值，可能会引起查询结果不一致。

并不是说包含union all的veiw会引起子查询无法展开为表连接，如上单个union all的view是不会影响子查询展开为表连接的。
SQL> select a.* from ta01 a where a.object_id in (select b.object_id from tb_vie
w b where b.object_type='TABLE');

1623 rows selected.

Execution Plan
----------------------------------------------------------
Plan hash value: 1785931674

--------------------------------------------------------------------------------
--
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time
|
--------------------------------------------------------------------------------
--
| 0 | SELECT STATEMENT | | 1 | 190 | 464 (2)| 00:00:06
|
|* 1 | HASH JOIN RIGHT SEMI | | 1 | 190 | 464 (2)| 00:00:06
|
| 2 | VIEW | VW_NSO_1 | 4229 | 54977 | 309 (1)| 00:00:04
|
| 3 | VIEW | TB_VIEW | 4229 | 99K| 309 (1)| 00:00:04
|
| 4 | UNION-ALL | | | | |
|
|* 5 | TABLE ACCESS FULL| TB01 | 1692 | 40608 | 154 (1)| 00:00:02
|
|* 6 | TABLE ACCESS FULL| TB02 | 2537 | 60888 | 154 (1)| 00:00:02
|
| 7 | TABLE ACCESS FULL | TA01 | 42509 | 7347K| 154 (1)| 00:00:02
|
--------------------------------------------------------------------------------
--

Predicate Information (identified by operation id):
---------------------------------------------------

1 - access("A"."OBJECT_ID"="$nso_col_1")
5 - filter("OBJECT_TYPE"='TABLE')
6 - filter("OBJECT_TYPE"='TABLE')

Note
-----
- dynamic sampling used for this statement

Statistics
----------------------------------------------------------
53 recursive calls
0 db block gets
3166 consistent gets
0 physical reads
0 redo size
85130 bytes sent via SQL*Net to client
1680 bytes received via SQL*Net from client
110 SQL*Net roundtrips to/from client
3 sorts (memory)
0 sorts (disk)
1623 rows processed

而如果我们换到oracle 11.2.0.1的版本，再来看看包含两个union all的view的子查询是否能够展开
SQL> select * from v$version;

BANNER
--------------------------------------------------------------------------------
Oracle Database 11g Enterprise Edition Release 11.2.0.1.0 - 64bit Production
PL/SQL Release 11.2.0.1.0 - Production
CORE 11.2.0.1.0 Production
TNS for Linux: Version 11.2.0.1.0 - Production
NLSRTL Version 11.2.0.1.0 – Production

SQL> select a.* from ta_view a where a.object_id in (select b.object_id from tb_view b where b.object_type='TABLE');

7629 rows selected.

Execution Plan
----------------------------------------------------------
Plan hash value: 1656093151

-----------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-----------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 6779K| 323M| 1206 (4)| 00:00:15 |
|* 1 | HASH JOIN | | 6779K| 323M| 1206 (4)| 00:00:15 |
| 2 | VIEW | VW_NSO_1 | 4191 | 54483 | 586 (1)| 00:00:08 |
| 3 | HASH UNIQUE | | 4191 | 54483 | 586 (1)| 00:00:08 |
| 4 | VIEW | TB_VIEW | 4191 | 54483 | 585 (1)| 00:00:08 |
| 5 | UNION-ALL | | | | | |
|* 6 | TABLE ACCESS FULL| TB01 | 2036 | 48864 | 293 (1)| 00:00:04 |
|* 7 | TABLE ACCESS FULL| TB02 | 2155 | 51720 | 292 (1)| 00:00:04 |
| 8 | VIEW | TA_VIEW | 161K| 5845K| 585 (1)| 00:00:08 |
| 9 | UNION-ALL | | | | | |
| 10 | TABLE ACCESS FULL | TA01 | 82194 | 2969K| 292 (1)| 00:00:04 |
| 11 | TABLE ACCESS FULL | TA02 | 79581 | 2875K| 292 (1)| 00:00:04 |
-----------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

1 - access("A"."OBJECT_ID"="OBJECT_ID")
6 - filter("OBJECT_TYPE"='TABLE')
7 - filter("OBJECT_TYPE"='TABLE')

Note
-----
- dynamic sampling used for this statement (level=2)

Statistics
----------------------------------------------------------
63 recursive calls
0 db block gets
5917 consistent gets
3072 physical reads
0 redo size
200965 bytes sent via SQL*Net to client
6108 bytes received via SQL*Net from client
510 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
7629 rows processed

这里oracle选择了将子查询展开为了表连接，优化器确实在做着很多的改变，这些改变需要我们通过文档 测试用例和实际的案例来验证，可以这么说脱离了oracle的版本来谈性能优化 sql调优其实往往是不严谨的。

最近注册了朋友介绍的一个论坛 http://www.bi168.cn/forum.php ，之前也注册了很多论坛比如itpub、otn、ask maclean等等，都没有怎么过多的去关注论坛的东西，记得maclean liu说过每天能遇见的案例是有限的，但是网络的同行遇见的case确是很多的，如果能够将别人遇见的模拟 理解 解决那么这个就是自己的。

原文地址：包含union all的view子查询无法展开表连接的模拟, 感谢原作者分享。