标量子查询 外表有多少重复值决定filter次数 转化 left join

1. 标量子查询的执行计划
SORT  和 Table 是齐平的，上面没有类似HASH JOIN之类的连接方式。

2. 外表驱动内表的时候 starts 代表执行次数，由内表的distinct值决定的，相当于内表做了一个view

3. 内表做的view，可以直接和外表做外链接，这样可以改为join,12C 后可以自动转化了。

---------标量子查询  case when 用  left join 的改写-----

 --Oracle标量子查询执行计划,性能为王：SQL标量子查询的优化案例分析-CSDN博客
       SELECT 
       CASE    WHEN OFFER_SPEC_ID IN      ----A的distinct   OFFER_SPEC_ID 值决定循环次数 
              (SELECT LOCAL_ITEM_CODE    FROM PU_meta_DIM.CODE_ITEM                      
                      WHERe PROV_TYPE_ID = 49AND PROV_ITEM_ID = 64 改写成view )          
          THEN   1    ELSE    0 END
 FROM PU_base_IND.DM_SP_SUB_OFFER_SERV_D_201407       
 PARTITION(P20140727) A
 WHERe LOCAL_CODE = '028'      

SELECT 
CASE WHEN a.OFFER_SPEC_ID = b.LOCAL_ITEM_CODE THEN 1 ELSE 0 END
 FROM PU_base_IND.DM_SP_SUB_OFFER_SERV_D_201407 PARTITION (P20140727) A,
(SELECt DISTINCT LOCAL_ITEM_CODE
FROM PU_meta_DIM.CODE_ITEM
WHERe PROV_TYPE_ID = 49 AND PROV_ITEM_ID = 64) b
AND a.OFFER_SPEC_ID = b.LOCAL_ITEM_CODE(+ )
AND TO_CHAr (a.OFFER_SPEC_ID) = c.OFFER_SPEC_ID(+ )
AND TO_CHAr (a.OFFER_SPEC_ID) = d.OFFER_SPEC_ID(+ )
AND TO_CHAr (a.OFFER_SPEC_ID) = e.LOCAL_ITEM_CODE(+ )  

标量子查询特点及特殊执行计划

1.1 标量子查询执行计划

1）标量子查询的计划和普通计划的执行顺序不同，执行计划中标量子查询虽然在上面，但是它由下面的节点驱动，每行驱动查询一次标量子查询，而普通执行计划是下面的节点受上面节点驱动。（具体见下面示例分析）

2）标量子查询和FILTER类似，如果是它们引起的性能问题，要重点关注是否是执行子查询的次数过多导致查询的效率不高。3）标量子查询和FILTER类似，它会根据输入和输出构建HASH表缓存键值对，已经查询过的键值对直接从缓存中查找，不用再次执行子查询，从而减少子查询的次数达到优化的目的，10g和11g缓存的是255个HASH BUCKETS，12C是1024个HASH BUCKETS。如下所示：

从上面可以看出，标量子查询执行计划和普通执行计划不一样，它上面节点的是被下面节点驱动的，类似FILTER，谓词部分有系统绑定变量:B1，本例中ID 1-9属于标量子查询部分，它由ID=10的节点驱动，ID=10的结果集影响ID 1-9标量子查询的执行次数，标量子查询执行次数还和缓存有关，也就是ID=10的关联条件CUST_ID的DISTINCT数目。

下面探讨下标量子查询特点以及优化方法，对标量子查询的优化首选方式就是改写。

1.2 标量子查询特点

子查询执行次数依赖于关键条件DISTINCT数目，重复值越多，效率越好，反之，效率差。

如下所示：

从上面执行计划可以看出，虽然驱动的A-ROWS有108K行，但是因为distinct数目是11，实际上标量子查询只执行11次，由此可以看出，ORACLE内部构建了缓存存储键值对，从而减少子查询执行次数，达到优化目的。

2

标量子查询优化方法

标量子查询的改写：减少标量子查询次数、改为外连接使用JOIN。

2.1 减少子查询执行次数

2.2 将标量子查询改为外连接

其中改为外连接也是12C对标量子查询优化的新特性。

3

12C标量子查询优化新特性 

如前面所示，在12c之前，对于scalary subquery是无法unnest的，这样就导致标量子查询的执行次数受驱动表行数的影响，类似FILTER操作，而且scalary subquery的执行计划和普通执行计划不一样。

select dname,       (select max(b.sal) from scott.emp b where b.deptno = a.deptno)  from scott.dept a;

12C对标量子查询的优化如下：

  

12C优化改进其实将标量子查询内部改成OUTER JOIN，从而充分利用JOIN算法进行优化提高效率，注意12C针对标量子查询的优化只对于max,min,avg单个有效（类似max||min等组合的无效），对count、查询列等无效，并且受参数_optimizer_unnest_scalar_sq控制。

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

oracle标量子查询和自己定义函数有时用起来比較方便，并且开发者也常常使用。数据量小还无所谓。数据量大，往往存在性能问题。

下面測试帮助大家彻底搞懂标量子查询。 SQL> create table a (id int,name varchar2(10)); Table created. SQL> create table b (id int,name varchar2(10)); Table created. SQL> insert into a values (1,'a1'); 1 row created. SQL> insert into a values (2,'a2'); 1 row created. SQL> insert into b values (1,'b1'); 1 row created. SQL> insert into b values (2,'b2'); 1 row created. SQL> commit; Commit complete. SQL> @getlvall Session altered. SQL> select a.*,(select name from b where b.id=a.id) from a; ID NAME (SELECTNAMEFROMBWHER ---------- -------------------- -------------------- 1 a1 b1 2 a2 b2 SQL> @getplanspe PLAN_TABLE_OUTPUT -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- SQL_ID 8rv825dykpx1m, child number 0 ------------------------------------- select a.*,(select name from b where b.id=a.id) from a Plan hash value: 2657529235 ------------------------------------------------------------------------------------ | Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | ------------------------------------------------------------------------------------ |* 1 | TABLE ACCESS FULL| B | 2 | 1 | 2 |00:00:00.01 | 14 | | 2 | TABLE ACCESS FULL| A | 1 | 2 | 2 |00:00:00.01 | 8 | ------------------------------------------------------------------------------------ Predicate Information (identified by operation id): --------------------------------------------------- 1 - filter("B"."ID"=:B1) Note ----- - dynamic sampling used for this statement 22 rows selected.

复制

<strong>--由上面的运行计划能够知道，b表运行2次。返回2行</strong>

SQL> insert into a values (3,'a3');
1 row created.
SQL> commit;
Commit complete.
SQL> select a.*,(select name from b where b.id=a.id) from a;ID NAME                 (SELECTNAMEFROMBWHER
---------- -------------------- --------------------1 a1                   b12 a2                   b23 a3
SQL> @getplanspe
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  8rv825dykpx1m, child number 0
-------------------------------------
select a.*,(select name from b where b.id=a.id) from a
Plan hash value: 2657529235
------------------------------------------------------------------------------------
| Id  | Operation         | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------
|*  1 |  TABLE ACCESS FULL| B    |      3 |      1 |      2 |00:00:00.01 |      21 |
|   2 |  TABLE ACCESS FULL| A    |      1 |      2 |      3 |00:00:00.01 |       8 |
------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------1 - filter("B"."ID"=:B1)
Note
------ dynamic sampling used for this statement
22 rows selected.

<strong>--由上面的运行计划能够知道，b表运行3次。返回2行</strong>

SQL> insert into a values (4,'a4');
1 row created.
SQL> insert into a values (5,'a5');
1 row created.
SQL> insert into a values (6,'a6');
1 row created.
SQL> insert into a values (7,'a7');
1 row created.
SQL> insert into a values (8,'a8');
1 row created.
SQL> insert into a values (9,'a9');
1 row created.
SQL> commit;
Commit complete.
SQL> select a.*,(select name from b where b.id=a.id) from a;ID NAME                 (SELECTNAMEFROMBWHER
---------- -------------------- --------------------1 a1                   b12 a2                   b23 a34 a45 a56 a67 a78 a89 a9
9 rows selected.
SQL> @getplanspe
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  8rv825dykpx1m, child number 0
-------------------------------------
select a.*,(select name from b where b.id=a.id) from a
Plan hash value: 2657529235
------------------------------------------------------------------------------------
| Id  | Operation         | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------
|*  1 |  TABLE ACCESS FULL| B    |      9 |      1 |      2 |00:00:00.01 |      63 |
|   2 |  TABLE ACCESS FULL| A    |      1 |      2 |      9 |00:00:00.01 |       8 |
------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------1 - filter("B"."ID"=:B1)
Note
------ dynamic sampling used for this statement
22 rows selected.

<strong>
--由上面的运行计划能够知道，b表运行9次，返回2行</strong>

SQL> update b set name='b1';
2 rows updated.
SQL> commit;
Commit complete.
SQL> select a.*,(select name from b where b.id=a.id) from a;ID NAME                 (SELECTNAMEFROMBWHER
---------- -------------------- --------------------1 a1                   b12 a2                   b13 a34 a45 a56 a67 a78 a89 a9
9 rows selected.
SQL> @getplanspe
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  8rv825dykpx1m, child number 0
-------------------------------------
select a.*,(select name from b where b.id=a.id) from a
Plan hash value: 2657529235
------------------------------------------------------------------------------------
| Id  | Operation         | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------
|*  1 |  TABLE ACCESS FULL| B    |      9 |      1 |      2 |00:00:00.01 |      63 |
|   2 |  TABLE ACCESS FULL| A    |      1 |      2 |      9 |00:00:00.01 |       8 |
------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------1 - filter("B"."ID"=:B1)
Note
------ dynamic sampling used for this statement
22 rows selected.

<strong>--由上面的运行计划能够知道，b表运行2次，返回2行</strong>  原作者这里应该错了

SQL> insert into b values (3,'b1');
1 row created.
SQL> insert into b values (4,'b1');
1 row created.
SQL> insert into b values (5,'b1');
1 row created.
insert into b values (6,'b1');b1');
1 row created.
SQL> insert into b values (7,'b1');
1 row created.
SQL> insert into b values (8,'b1');
1 row created.
SQL> insert into b values (9,'b1');
1 row created.
SQL> commit;
Commit complete.
SQL> select a.*,(select name from b where b.id=a.id) from a;ID NAME                 (SELECTNAMEFROMBWHER
---------- -------------------- --------------------1 a1                   b12 a2                   b13 a3                   b14 a4                   b15 a5                   b16 a6                   b17 a7                   b18 a8                   b19 a9                   b1
9 rows selected.
SQL> @getplanspe
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  8rv825dykpx1m, child number 0
-------------------------------------
select a.*,(select name from b where b.id=a.id) from a
Plan hash value: 2657529235
------------------------------------------------------------------------------------
| Id  | Operation         | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------
|*  1 |  TABLE ACCESS FULL| B    |      9 |      1 |      9 |00:00:00.01 |      63 |
|   2 |  TABLE ACCESS FULL| A    |      1 |      2 |      9 |00:00:00.01 |       8 |
------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------1 - filter("B"."ID"=:B1)
Note
------ dynamic sampling used for this statement
22 rows selected.

<strong>--b.name字段所有为‘b1’。由上面的运行计划能够知道。b表运行9次。返回9行</strong>

SQL> update a set id=1;
9 rows updated.
SQL> commit;
Commit complete.
SQL> select * from a;ID NAME
---------- --------------------1 a11 a21 a31 a41 a51 a61 a71 a81 a9
9 rows selected.
SQL> select * from b;ID NAME
---------- --------------------1 b12 b13 b14 b15 b16 b17 b18 b19 b1
9 rows selected.
SQL> select a.*,(select name from b where b.id=a.id) from a;ID NAME                 (SELECTNAMEFROMBWHER
---------- -------------------- --------------------1 a1                   b11 a2                   b11 a3                   b11 a4                   b11 a5                   b11 a6                   b11 a7                   b11 a8                   b11 a9                   b1
9 rows selected.
SQL> @getplanspe
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  8rv825dykpx1m, child number 0
-------------------------------------
select a.*,(select name from b where b.id=a.id) from a
Plan hash value: 2657529235
------------------------------------------------------------------------------------
| Id  | Operation         | Name | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------
|*  1 |  TABLE ACCESS FULL| B    |      1 |      1 |      1 |00:00:00.01 |       7 |
|   2 |  TABLE ACCESS FULL| A    |      1 |      2 |      9 |00:00:00.01 |       8 |
------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------1 - filter("B"."ID"=:B1)
Note
------ dynamic sampling used for this statement
22 rows selected.
SQL> 

关联字段a.id所有为1。a表有9行，标量子查询相当于运行9次select name from b where b.id=1 ，oracle也不傻，starts=1。说明仅仅运行了1次。

所以是看外表有多少重复值决定filter次数的，和内表重复值无关。