一种特殊场景下的HASH JOIN的优化为NEST LOOP.

应用场景：

有如下的SQL:

select t.*, t1.owner
from t, t1
where t.id=t1.id;

表t ,t1的数据量比较大，比如200W行.但是两张表能关联的行数却很少，比如不到50条. T1表的行比较宽,且在id列上有单列索引.

这里限制t1的索引为单列索引是为了让访问t1表数据的时候要通过rowid回表.因为在实际应用中,我们通常会需要t1表的几个列，但是不可能对它们全都索引.

此时可以优化为下面的SQL.

select /*+ use_nl(t1, t) ordered */  t.*, t1.owner

from t, t1
where t.id=t1.id and t1.id is not null;
改写原理:

1. 当我们使用NEST LOOP的时候，t1表可以使用id列上的index走出INDEX RANGE SCAN，此时如果输入值不在索引范围内时，是能够很快返回的，理论上就无须访问索引的leaf块。只需访问root和部分branch即可.而这里我们能够真正匹配的数据很少，所以大部分时候都没有到达leaf.

不过这种场景实际应用中并不常见,所以在CBO看来, cost= rows of t * 3(假设index深度为3)， 这个cost远大于hash join的cost。如果不加hint，CBO是不会选这个执行计划的.

1. select  /*+ use_nl(p, l) ordered gather_plan_statistics */
2. l.l_suppkey    from partsupp p ,lineitem l    where
3. l.l_suppkey=p.ps_suppkey+10000010    and p.ps_suppkey>0
4. Plan hash value: 3264550197
5. -------------------------------------------------------------------------------------------------------------------------------
6. | Id  | Operation             | Name        | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time   | A-Rows |   A-Time   | Buffers |
7. -------------------------------------------------------------------------------------------------------------------------------
8. |   0 | SELECT STATEMENT      |             |      1 |        |       |  2405K(100)|          |      0 |00:00:00.91 |    1698 |
9. |   1 |  NESTED LOOPS         |             |      1 |    480M|  3662M|  2405K  (1)| 08:01:03 |      0 |00:00:00.91 |    1698 |
10. |*  2 |   INDEX FAST FULL SCAN| I_PARTSUPP1 |      1 |    800K|  3125K|   308   (2)| 00:00:04 |    800K|00:00:00.18 |    1685 |
11. |*  3 |   INDEX RANGE SCAN    | I_LINEITEM1 |    800K|    600 |  2400 |     3   (0)| 00:00:01 |      0 |00:00:00.62 |      13 |
12. -------------------------------------------------------------------------------------------------------------------------------

比如下图，要访问10051的数据，通过root节点就知道没这个key了，根本不需完成一次index range scan，而在loop join的下一次循环如果是要10052，直接在cache里快速匹配即可，这个是我猜测的，在CBO执行过程中这个应该很容易实现，相当于做预先处理，这样的预处理再结合storage index就形成了smart scan。

2. HASH JOIN时需要真正拿到索引中t1.id列全部数据后再做匹配，因此此时会走IFS或IFFS，直接遍历索引的所有叶子节点，然后建立hash表.

当索引里记录比较多同时能够匹配的记录有很少是，这种方法不如上面的NEST LOOP.

实验准备：

表的创建和数据的准备,需要的时间比较久，得耐心等待.

--表T

create table t (
id number,
col1 varchar2(30),
col2 varchar2(30),
col3 varchar2(30),
col4 varchar2(30),
col5 varchar2(30),
col6 varchar2(30),
col7 varchar2(30),
col8 varchar2(30),
col9 varchar2(30),
col10 varchar2(30)
);

--随机生成约200W行数据

insert into t select
rownum,
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30) 
from dual connect by rownum<2000000;

--T1变的行宽比较大，此时full table scan 读的块较多.

create table t1 (
id number,
id2 number,
id3 varchar2(30),
id4 varchar2(30),
id5 varchar2(30),
id6 varchar2(30),
id7 varchar2(30),
id8 varchar2(30),
id9 varchar2(30),
id10 varchar2(30),
id11 varchar2(30),
id12 varchar2(30),
id13 varchar2(30),
id14 varchar2(30),
id15 varchar2(30),
id16 varchar2(30),
id17 varchar2(30),
id18 varchar2(30),
id19 varchar2(30),
id20 varchar2(30),
id21 varchar2(30), 
owner varchar2(30));

--T1表上创建索引

create index t1_idx_id on t1(id);

--随机生成约200W行数据

insert into t1 select
rownum+2000000,
round(dbms_random.value(1,100000000)),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30),
dbms_random.string('u',30)
from dual connect by rownum<2000000;

--收集表的统计信息

exec dbms_stats.gather_table_stats(ownname=>'SYS',tabname=>'T',estimate_percent=>100,method_opt=>'for all columns size auto',cascade=>TRUE)
exec dbms_stats.gather_table_stats(ownname=>'SYS',tabname=>'T1',estimate_percent=>100,method_opt=>'for all columns size auto',cascade=>TRUE)

--让表T, T1 有少数的关联行,这里是制造10条关联记录.

update t1 set id = (select round(dbms_random.value(1,1999999)) from dual)
where id in (select round(dbms_random.value(2000001,4000000)) from dual connect by rownum<10) and id is not null;

实验开始:

SQL> select t.*, t1.owner
  2  from t, t1
  3  where t.id=t1.id and t1.id is not null;

已选择10行。

已用时间:  00: 00: 18.54

执行计划
----------------------------------------------------------
Plan hash value: 1444793974

-----------------------------------------------------------------------------------
| Id  | Operation          | Name | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     |
-----------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |      |     1 |   352 |       |   109K  (1)| 00:21:56 |
|*  1 |  HASH JOIN         |      |     1 |   352 |    93M|   109K  (1)| 00:21:56 |
|*  2 |   TABLE ACCESS FULL| T1   |  1999K|    70M|       | 49367   (1)| 00:09:53 |
|   3 |   TABLE ACCESS FULL| T    |  1999K|   600M|       | 24691   (1)| 00:04:57 |
-----------------------------------------------------------------------------------

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

1 - access("T"."ID"="T1"."ID")
   2 - filter("T1"."ID" IS NOT NULL)

统计信息
----------------------------------------------------------
         97  recursive calls
          0  db block gets
     454560  consistent gets
     283928  physical reads
          0  redo size
       2086  bytes sent via SQL*Net to client
        420  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
         10  rows processed

全表扫描的T1时，18S。

SQL> select /*+index(t1 T1_IDX_ID)*/t.*, t1.owner
  2  from t, t1
  3  where t.id=t1.id and t1.id is not null;

已选择10行。

已用时间:  00: 01: 52.29

执行计划
----------------------------------------------------------
Plan hash value: 2604990230

--------------------------------------------------------------------------------------------------
| Id  | Operation                    | Name      | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |           |     1 |   352 |       |   246K  (1)| 00:49:19 |
|*  1 |  HASH JOIN                   |           |     1 |   352 |    93M|   246K  (1)| 00:49:19 |
|   2 |   TABLE ACCESS BY INDEX ROWID| T1        |  1999K|    70M|       |   186K  (1)| 00:37:15 |
|*  3 |    INDEX FULL SCAN           | T1_IDX_ID |  1999K|       |       |  4279   (1)| 00:00:52 |
|   4 |   TABLE ACCESS FULL          | T         |  1999K|   600M|       | 24691   (1)| 00:04:57 |
--------------------------------------------------------------------------------------------------

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

1 - access("T"."ID"="T1"."ID")
   3 - filter("T1"."ID" IS NOT NULL)

统计信息
----------------------------------------------------------
         98  recursive calls
          0  db block gets
     458819  consistent gets
     271761  physical reads
   13092036  redo size
       2086  bytes sent via SQL*Net to client
        420  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
         10  rows processed

SQL> select /*+ use_nl(t1, t) ordered */t.*, t1.owner
  2  from t, t1
  3  where t.id=t1.id and t1.id is not null;

已选择10行。

已用时间:  00: 00: 04.81

执行计划
----------------------------------------------------------
Plan hash value: 3173170385

------------------------------------------------------------------------------------------
| Id  | Operation                    | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |           |     1 |   352 |  4026K  (1)| 13:25:15 |
|   1 |  NESTED LOOPS                |           |       |       |            |          |
|   2 |   NESTED LOOPS               |           |     1 |   352 |  4026K  (1)| 13:25:15 |
|   3 |    TABLE ACCESS FULL         | T         |  1999K|   600M| 24691   (1)| 00:04:57 |
|*  4 |    INDEX RANGE SCAN          | T1_IDX_ID |     1 |       |     2   (0)| 00:00:01 |
|   5 |   TABLE ACCESS BY INDEX ROWID| T1        |     1 |    37 |     2   (0)| 00:00:01 |
------------------------------------------------------------------------------------------

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

4 - access("T"."ID"="T1"."ID")
       filter("T1"."ID" IS NOT NULL)

统计信息
----------------------------------------------------------
          0  recursive calls
          0  db block gets
      90946  consistent gets
      90914  physical reads
          0  redo size
       2086  bytes sent via SQL*Net to client
        420  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
         10  rows processed

select /*+ use_nl(t1, t) ordered  gather_plan_statistics*/t.*, t1.owner
from t, t1 where t.id=t1.id and t1.id is not null

Plan hash value: 3173170385

---------------------------------------------------------------------------------------------------------------------------------------------
| Id  | Operation                    | Name      | Starts | E-Rows |E-Bytes| Cost (%CPU)| E-Time   | A-Rows |   A-Time   | Buffers | Reads  |
---------------------------------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |           |      4 |        |       |  4026K(100)|          |  40 |00:01:04.01 |     363K|       363K|
|   1 |  NESTED LOOPS                |           |      4 |        |       |            |          |  40 |00:01:04.01 |     363K|       363K|
|   2 |   NESTED LOOPS               |           |      4 |      1 |   352 |  4026K  (1)| 13:25:15 |  40 |00:01:04.01 |     363K|       363K|
|   3 |    TABLE ACCESS FULL         | T         |      4 |   1999K|   600M| 24691   (1)| 00:04:57 |   7999K|00:00:33.47 |     363K|    363K|
|*  4 |    INDEX RANGE SCAN          | T1_IDX_ID |   7999K|      1 |       |     2   (0)| 00:00:01 |  40 |00:00:51.27 | 60 |      0 |
|   5 |   TABLE ACCESS BY INDEX ROWID| T1        |     40 |      1 |    37 |     2   (0)| 00:00:01 |  40 |00:00:00.01 | 40 |      0 |
---------------------------------------------------------------------------------------------------------------------------------------------