深入解析：半连接与反连接的原理和等价改写方法

zoukankan html css js c++ java

深入解析：半连接与反连接的原理和等价改写方法
作者简介
刘志永
云和恩墨交付工程师，有5年的运营商及金融行业 DBA 运维工作经验，对故障处理、性能优化、迁移&升级有丰富的经验。

本文由恩墨大讲堂152期线上分享整理而成。课程回看可点击文末“阅读原文”。
半连接的原理及等价改写

1. 什么是半连接

当两张表进行关联，只返回匹配上的数据并且只会返回一张的表的数据，半连接一般就是指的在子查询中出现 IN 和 EXISTS。

执行计划最常见的有下面的两个：
NESTED LOOPS SEMI
HASH JOIN SEMI

1.1 NESTED LOOP SEMI

执行步骤：
优化器选择主查询中的表为驱动表；
选择子查询中的表为被驱动表；
把驱动表的每一行根据连接列去被驱动表中去查找匹配的行；
如果在被驱动表中匹配上一个或者多个，则返回驱动表中的数据。

HINT /*+ use_nl(b,a) leading(b) */
看下这个 SQL：
explain plan for
Select a.deptno, a.dname
from dept a
where exists (select 1 from emp b where a.deptno =b.deptno);

执行计划：
1* select * from table(dbms_xplan.display)

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 3429995275

---------------------------------------------------------------------------------
| Id | Operation          | Name       | Rows | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------------
|   0| SELECT STATEMENT   |            |     3 |  105 |     3   (0)| 00:00:01 |
|   1| NESTED LOOPSSEMI |            |     3 |  105 |     3  (0)| 00:00:01 |
|   2|   TABLE ACCESS FULL| DEPT       |    4 |    88 |     3  (0)| 00:00:01 |
|* 3|   INDEX RANGE SCAN | IDX_DEPTNO |    11 |  143 |     0   (0)| 00:00:01 |
---------------------------------------------------------------------------------

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified byoperation id):
---------------------------------------------------

   3- access("A"."DEPTNO"="B"."DEPTNO")

用 hint 换下驱动表：
跑下这个 SQL：
Explain plan for
Select /*+ use_nl(b,a) leading(b) */   a.deptno, a.dname
from dept a
where exists (select    1 from emp b where a.deptno = b.deptno);

执行计划：
SQL> select * fromtable(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 1511694371

----------------------------------------------------------------------------------------
| Id | Operation                    |Name    | Rows | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------------------
|   0| SELECT STATEMENT             |         |    3 |   105 |     7 (15)| 00:00:01 |
|   1| NESTED LOOPS                |         |    3 |   105 |     7 (15)| 00:00:01 |
|   2|   NESTED LOOPS               |         |    3 |   105 |     7 (15)| 00:00:01 |
|   3|   SORT UNIQUE               |         |   14 |   182 |     3  (0)| 00:00:01 |
|   4|     TABLE ACCESS FULL        | EMP     |   14 |   182 |     3  (0)| 00:00:01 |
|* 5|    INDEX UNIQUE SCAN         | PK_DEPT |     1 |      |     0   (0)| 00:00:01 |

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
|   6|   TABLE ACCESS BY INDEX ROWID|DEPT    |     1 |   22 |     1   (0)| 00:00:01 |
----------------------------------------------------------------------------------------

Predicate Information (identified byoperation id):
---------------------------------------------------

   5- access("A"."DEPTNO"="B"."DEPTNO")

等价改写
用 IN 来改写：
Explain plan for
select  a.deptno, a.dname from dept a where a.deptno in (select deptno from empb);

执行计划：
SQL> select * fromtable(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 1505539579

----------------------------------------------------------------------------------
| Id | Operation          | Name        | Rows | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------------
|   0| SELECT STATEMENT   |             |     3 |  105 |     3   (0)| 00:00:01 |
|   1| NESTED LOOPSSEMI |             |     3 |  105 |     3   (0)| 00:00:01 |
|   2|   TABLE ACCESS FULL| DEPT        |    4 |    88 |     3  (0)| 00:00:01 |
|* 3|   INDEX RANGE SCAN | IDEX_DEPTNO |    11 |  143 |     0   (0)| 00:00:01 |
----------------------------------------------------------------------------------

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified byoperation id):
---------------------------------------------------

   3- access("A"."DEPTNO"="DEPTNO")

改写成 inner join：
Explain plan for
select a.deptno, a.dname from dept a,(select b.deptno from emp b group by b.deptno) c where a.deptno=c.deptno;

执行计划：
SQL> select * fromtable(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 1766026621

-----------------------------------------------------------------------------------
| Id | Operation           | Name        | Rows | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------------
|   0| SELECT STATEMENT    |             |    14 |  658 |     4 (25)| 00:00:01 |
|   1| HASH GROUP BY      |             |   14 |   658 |     4 (25)| 00:00:01 |
|   2|   NESTED LOOPS      |             |    14 |  658 |     3   (0)| 00:00:01 |
|   3|    TABLE ACCESS FULL| DEPT        |    4 |   136 |     3  (0)| 00:00:01 |
|* 4|    INDEX RANGE SCAN | IDEX_DEPTNO|     4 |    52 |    0   (0)| 00:00:01 |
-----------------------------------------------------------------------------------

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

Predicate Information (identified byoperation id):
---------------------------------------------------

   4- access("A"."DEPTNO"="B"."DEPTNO")

Inner join 多了 group by 的操作，emp 的 depno 值不是唯一的，要保证逻辑上跟半连接的一致就需要把 emp 的连接进行去重操作，跟上面 emp 作为驱动表是一致的。

1.2 HASH JOIN SEMI

执行步骤：
优化器把主查询中的表当作驱动表，把 select 列和 join 列读入 pga 中，并把 join 列进行 hash 运算构建 hash table；
把另外一张表的连接列也进行 hash 运算，然后到 pga 中去探测 hash table ，如果关联上就返回驱动表的数据。

Hint /*+ use_hash(a,b) leading(a)*/
看下这个 SQL：
explainplan for
Select a.*
fromdept a
whereexists (select   1 from emp b wherea.deptno = b.deptno);

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Planhash value: 1754319153

---------------------------------------------------------------------------
|Id | Operation          | Name | Rows | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |     |     3 |   129 |    6   (0)| 00:00:01 |
|* 1 | HASH JOIN SEMI   |      |     3 |  129 |     6   (0)| 00:00:01 |
|   2 |  TABLE ACCESS FULL| DEPT |     4|   120 |     3  (0)| 00:00:01 |
|   3 |   TABLEACCESS FULL| EMP |    14 |  182 |     3   (0)| 00:00:01 |
---------------------------------------------------------------------------

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
PredicateInformation (identified by operation id):
---------------------------------------------------

   1 -access("A"."DEPTNO"="B"."DEPTNO")

改变下驱动表：
explain plan for
Select/*+ use_hash(a,b) leading(b)*/ a.*
from dept a
where exists (select     1 from emp b wherea.deptno = b.deptno);

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Planhash value: 300394613

----------------------------------------------------------------------------
|Id | Operation           | Name | Rows | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------
|   0 | SELECT STATEMENT    |     |     3 |   129 |    7 (15)| 00:00:01 |
|* 1 | HASH JOIN          |      |    3 |   129 |     7 (15)| 00:00:01 |
|   2 |   SORT UNIQUE      |     |    14 |   182 |    3   (0)| 00:00:01 |
|   3 |   TABLE ACCESS FULL| EMP |    14 |  182 |     3   (0)| 00:00:01 |
|   4 |  TABLE ACCESS FULL | DEPT |     4|   120 |     3  (0)| 00:00:01 |
----------------------------------------------------------------------------

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

PredicateInformation (identified by operation id):
---------------------------------------------------

   1 - access("A"."DEPTNO"="B"."DEPTNO")

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

Emp 连接列有重复值，需要进行去重操作，保证逻辑一致。

等价改写
用 IN 来改写：
explain plan for
select   a.* from dept a where a.deptno in (selectdeptno from emp b);

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Planhash value: 1754319153

---------------------------------------------------------------------------
|Id | Operation          | Name | Rows | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |     |     3 |   129 |    6   (0)| 00:00:01 |
|* 1 | HASH JOIN SEMI    |     |     3 |   129 |    6   (0)| 00:00:01 |
|   2 | TABLE ACCESS FULL| DEPT |     4 |  120 |     3   (0)| 00:00:01 |
|   3 |  TABLE ACCESS FULL| EMP |    14 |  182 |     3   (0)| 00:00:01 |
---------------------------------------------------------------------------

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

   1 -access("A"."DEPTNO"="DEPTNO")

用内连接改写：
explain plan for
select a.* from dept a ,(select b.deptno fromemp b group by b.deptno) c where a.deptno=c.deptno;

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 3977374068

-----------------------------------------------------------------------------
|Id | Operation            | Name | Rows | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------
|   0 | SELECT STATEMENT     |     |    14 |   602 |    7 (15)| 00:00:01 |
|* 1 | HASH JOIN           |      |   14 |   602 |     7 (15)| 00:00:01 |
|   2 |  TABLE ACCESS FULL | DEPT |     4 |  120 |     3   (0)| 00:00:01 |
|   3 |  VIEW               |      |   14 |   182 |     4 (25)| 00:00:01 |
|   4 |   HASH GROUP BY    |      |   14 |   182 |     4 (25)| 00:00:01 |
|   5 |    TABLE ACCESS FULL| EMP |    14 |  182 |     3   (0)| 00:00:01 |

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

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

   1 -access("A"."DEPTNO"="C"."DEPTNO")

反连接的原理及等价改写

1. 什么是反连接

两表关联只返回主表的数据，并且只返回主表与子表没关联上的数据，这种连接就叫反连接。反连接一般就是指的 NOT IN 和 NOT EXISTS。

执行计划：
NESTED LOOP ANTI
HASH JOIN ANTI

1.1 NESTED LOOP ANTI

NESTED LOOP ANTI 执行步骤：
优化器选择主查询中的表为驱动表；
选择子查询中的表为被驱动表；
把驱动表的每一行根据连接列去被驱动表中去查找匹配的行；
如果在被驱动表中没有匹配上，则返回驱动表中的没有匹配上的数据。

Hint /*+ use_nl(a,b) leading(b) */
在反连接中驱动表不会反生变化，因为反连接等价于外连接+子表连接条件 is null，使用嵌套循环进行关联的时候无法更改驱动表，驱动表会被固定为主表。

我们来看下下面这个执行计划：
explain plan for
Select      a.*
from dept a
where not exists (select     1 from emp b where a.deptno =b.deptno);

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Plan hash value: 2795798678

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

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

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

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |           |     4 |   172 |    3   (0)| 00:00:01
|

|   1 | NESTED LOOPS ANTI |           |     4 |   172 |    3   (0)| 00:00:01
|

|   2 |  TABLE ACCESS FULL| DEPT       |    4 |   120 |     3  (0)| 00:00:01
|

|* 3 |  INDEX RANGE SCAN | IDX_DEPTNO |    1 |    13 |     0  (0)| 00:00:01
|

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

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

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

   3 -access("A"."DEPTNO"="B"."DEPTNO")

使用/*+ use_nl(a,b) leading(b) */来更改驱动表看下：
explain plan for
Select    /*+ use_nl(a,b) leading(b) */   a.*
from dept a
where not exists (select   1 from emp b where a.deptno =b.deptno);

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Planhash value: 2795798678

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

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

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

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |           |     4 |   172 |    3   (0)| 00:00:01
|

|   1 | NESTED LOOPS ANTI |           |     4 |   172 |    3   (0)| 00:00:01
|

|   2 | TABLE ACCESS FULL|DEPT       |    4 |   120 |     3  (0)| 00:00:01
|

|* 3 |  INDEX RANGE SCAN | IDX_DEPTNO |    1 |    13 |     0  (0)| 00:00:01
|

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

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

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

   3 -access("A"."DEPTNO"="B"."DEPTNO")

等价改写
Not exists 改写为 not in，要注意子查询要排除 null 的情况，因为 not in 后面如果有 null 值整个查询都为空。

看下面这条 SQL：
explain plan for
select   a.* from dept a where a.deptno not in(select deptno from emp b where deptno is not null);

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Plan hash value: 2795798678

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

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

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

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |           |     4 |   172 |    3   (0)| 00:00:01
|

|   1 | NESTED LOOPS ANTI |            |     4 |  172 |     3   (0)| 00:00:01
|

|   2 |  TABLE ACCESS FULL| DEPT      |     4 |   120 |    3   (0)| 00:00:01
|

|* 3 |  INDEX RANGE SCAN | IDX_DEPTNO |    1 |    13 |     0  (0)| 00:00:01
|

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

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

PredicateInformation (identified by operation id):
---------------------------------------------------

   3 -access("A"."DEPTNO"="DEPTNO")
       filter("DEPTNO" IS NOT NULL)

用外连接改写：
根据反连接定义，翻译为标准 SQL 写法就是外连接+子表连接条件 is null。外连接是只返回主表的数据，子表只返回关联的数据，加上子表条件 is null 就是只返回没有关联上的数据。

我们看下这条 SQL：
selecta.*,b.*
from dept a
   left join emp b on a.deptno = b.deptno;

explainplan for
select a.*
from dept a
   left join emp b on a.deptno = b.deptno
where b.deptno is null;

加上 b.deptno is null 就返回了没有关联上的数据。

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
Planhash value: 2795798678

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

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

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

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |           |     4 |  172 |     3   (0)| 00:00:01
|

|   1 | NESTED LOOPS ANTI |            |     4 |  172 |     3   (0)| 00:00:01
|

|   2 |  TABLE ACCESS FULL| DEPT      |     4 |   120 |    3   (0)| 00:00:01
|

|* 3 |  INDEX RANGE SCAN | IDX_DEPTNO |    1 |   13 |     0   (0)| 00:00:01
|

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

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

PredicateInformation (identified by operation id):
---------------------------------------------------

   3 -access("A"."DEPTNO"="B"."DEPTNO")

1.2 HASH JOIN ANTI

执行步骤：
优化器把主查询中的表当作驱动表，把 select 列和 join 列读入 pga 中，并把 join 列进行 hash 运算构建 hash table；
把另外一张表的连接列也进行 hash 运算，然后到 pga 中去探测 hash table ，如果没有关联上就返回驱动表的数据。

两表关联如果是外连接，要改变 hash 连接的驱动表必须使用 swap_join_inputs。
看下面这个 SQL：
explainplan for
Select     a.*
fromdept a
wherenot exists (select     1 from emp b wherea.deptno =b.deptno);

执行计划：
PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Planhash value: 474461924

---------------------------------------------------------------------------
|Id | Operation          | Name | Rows | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |     |     4 |   172 |    6   (0)| 00:00:01 |
|* 1 | HASH JOIN ANTI    |      |    4 |   172 |     6  (0)| 00:00:01 |
|   2 |  TABLE ACCESS FULL| DEPT |     4 |  120 |     3   (0)| 00:00:01 |
|   3 |  TABLE ACCESS FULL| EMP |    14 |  182 |     3   (0)| 00:00:01 |
---------------------------------------------------------------------------

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
PredicateInformation (identified by operation id):
---------------------------------------------------

   1 -access("A"."DEPTNO"="B"."DEPTNO")

加 hint，更改驱动表
explainplan for
Select     /*+ use_hash(dept,emp@c) swap_join_inputs(emp@c)*/   *
fromdept
wherenot exists (select   /*+ qb_name(c)*/ 1from emp   where emp.deptno = dept.deptno);

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Plan hashvalue: 152508289

-----------------------------------------------------------------------------
|Id | Operation            | Name | Rows | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------
|   0 | SELECT STATEMENT     |     |     4 |   172 |    6   (0)| 00:00:01 |
|* 1 | HASH JOIN RIGHT ANTI|      |     4 |  172 |     6   (0)| 00:00:01 |
|   2 |  TABLE ACCESS FULL | EMP |    14|   182 |     3  (0)| 00:00:01 |
|   3 |  TABLE ACCESS FULL | DEPT |     4 |  120 |     3   (0)| 00:00:01 |
-----------------------------------------------------------------------------

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
PredicateInformation (identified by operation id):
---------------------------------------------------

   1 -access("EMP"."DEPTNO"="DEPT"."DEPTNO")

等价改写
Not in 改写：
explain plan for
select   a.* from dept a where a.deptno not in(select deptno from emp b where deptno is not null);

执行计划：
SQL>select * from table(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 474461924

---------------------------------------------------------------------------
|Id | Operation          | Name | Rows | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |     |     4 |   172 |    6   (0)| 00:00:01 |
|* 1 | HASH JOIN ANTI    |      |    4 |   172 |     6  (0)| 00:00:01 |
|   2 |  TABLE ACCESS FULL| DEPT |     4|   120 |     3  (0)| 00:00:01 |
|* 3 |  TABLE ACCESS FULL| EMP |    14 |  182 |     3   (0)| 00:00:01 |
---------------------------------------------------------------------------

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

   1 -access("A"."DEPTNO"="DEPTNO")
   3 - filter("DEPTNO" IS NOT NULL)

改为外连接：
explain plan for
select a.*
from dept a
  left join emp b on a.deptno = b.deptno
where b.deptno is null;

执行计划：
SQL> select * fromtable(dbms_xplan.display);

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 474461924

---------------------------------------------------------------------------
| Id | Operation          | Name |Rows | Bytes | Cost (%CPU)| Time     |
---------------------------------------------------------------------------
|   0| SELECT STATEMENT   |      |    4 |   172 |     6  (0)| 00:00:01 |
|* 1| HASH JOIN ANTI    |     |     4 |   172 |    6   (0)| 00:00:01 |
|   2|   TABLE ACCESS FULL| DEPT |     4 |  120 |     3  (0)| 00:00:01 |
|   3|   TABLE ACCESS FULL| EMP |    14|   182 |     3  (0)| 00:00:01 |
---------------------------------------------------------------------------

PLAN_TABLE_OUTPUT
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified byoperation id):
---------------------------------------------------

   1- access("A"."DEPTNO"="B"."DEPTNO")

总结

通过半连接和反连接的学习，我们可以知道半连接一般是出现在子查询有 IN 或者 EXISTS 的情况，并且只返回主表连接列关联上的数据，如果连接列是唯一值的话可以直接转化为内连接，反连接一般是出现在子查询有 NOT IN 或者 NOT EXISTS 的情况，并且只返回主表连接列没有关联上的数据，NEST LOOP ANTI 不能改驱动表。

资源下载
关注公众号：数据和云（OraNews）回复关键字获取
‘2017DTC’，2017 DTC 大会 PPT
‘DBALIFE’，“DBA 的一天”海报
‘DBA04’，DBA 手记4 经典篇章电子书
‘RACV1’, RAC 系列课程视频及 PPT
‘122ARCH’，Oracle 12.2 体系结构图
‘2017OOW’，Oracle OpenWorld 资料
‘PRELECTION’，大讲堂讲师课程资料

  戳原文，回看视频课程！
查看全文

相关阅读:
Silverlight Control（五）TimePicker
《ASP.NET Core 3框架揭秘》笔记
 Hyper-V初体验
 debian初体验
 异步编程-编程指南
 Oracle查询优化读书笔记
 设计原则
 日志管理
 Git在Vs中的使用
 SVN使用备注

原文地址：https://www.cnblogs.com/hzcya1995/p/13312413.html

深入解析：半连接与反连接的原理和等价改写方法

1.1 NESTED LOOP SEMI

1.2 HASH JOIN SEMI

1. 什么是反连接

1.1 NESTED LOOP ANTI

1.2 HASH JOIN ANTI