ORDERED 和 USE_NL() hint (zt)的用法

来源：互联网发布：镇江java培训学费编辑：程序博客网时间：2024/06/06 08:25

前两天,在给一只report写一个Validation Set.发现把这三个table PO_AGENTS PA, PER_PEOPLE_F PAPF,HR_LOCATIONS_ALL_TL HL 串起来后语句执行的像蜗牛一样慢啊，用户每次跑个报表是都会抱怨说在挑选参数时太慢。不得不优化语句啊。后来想用用oralce 的优化器rule(基于规则)来优化语句。但是依然很慢。无意间看到了Use_NL()。没想到还真的有效果。当然我们也可以通过新建Index的方式来实现语句的速度。主要是我们公司每上传一支程式到正式环境走的签核流程实在太多太慢。让人根本不想去走那繁琐的流程。后来上网查了一下Use_NL()的用法。现在整理一下，给有需要的兄弟分享一下。

ORDERED好理解，就是表示根据 from 后面表的顺序join，从左到右，左边的表做驱动表

USE_NL()，先看看oracle doc怎么说：
In this statement, the USE_NL hint explicitly chooses a nested loops join with the customers table as the inner table:

SELECT /*+ ORDERED USE_NL(customers) to get first row faster */
accounts.balance, customers.last_name, customers.first_name
FROM accounts, customers
WHERE accounts.customer_id = customers.customer_id;

customers 作为inner table，也就是说作为被驱动表。驱动表称为outer table。

也就是说use_nl如果只带了一个表名作为参数，则该表为被驱动表。

如果带了2个以上的参数，oracle 并没有指出 use_nl(a b) 中哪个是驱动表，所以常使用 ordered 或者 full() 或者 index() 来强化我们的目标

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

试验模拟：
t2: 27325 rows, object_id 上有索引in1
t3: 7326 rows,无索引
两表都已经分析

1.有ordered，按照t3驱动t2的顺序join；
并且USE_NL(t2)也表示t2作为inner table，也就是被驱动表；
没有矛盾
代码:
--------------------------------------------------------------------------------
SQL> select /*+ ordered USE_NL(t2) */ count(*)
2 from t3,t2
3 where t2.object_id=t3.object_id;

COUNT(*)
----------
      7325
Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=7365 Card=1 Bytes=8)
   1    0   SORT (AGGREGATE)
   2    1     NESTED LOOPS (Cost=7365 Card=7325 Bytes=58600)
   3    2       TABLE ACCESS (FULL) OF 'T3' (Cost=39 Card=7326 Bytes=29304)
   4    2       INDEX (RANGE SCAN) OF 'IN1' (NON-UNIQUE) (Cost=1 Card=1 Bytes=4)
--------------------------------------------------------------------------------

2.有ordered，按照t3驱动t2的顺序join；
但是USE_NL(t3)却表示t3作为inner table，也就是被驱动表；
有矛盾，所以oracle 忽视这个hint，执行hash join
代码:
--------------------------------------------------------------------------------
SQL> select /*+ ordered USE_NL(t3) */ count(*)
2 from t3,t2
3 where t2.object_id=t3.object_id;

COUNT(*)
----------
      7325
Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=48 Card=1 Bytes=8)
   1    0   SORT (AGGREGATE)
   2    1     HASH JOIN (Cost=48 Card=7325 Bytes=58600)
   3    2       TABLE ACCESS (FULL) OF 'T3' (Cost=39 Card=7326 Bytes=29304)
   4    2       INDEX (FAST FULL SCAN) OF 'IN1' (NON-UNIQUE) (Cost=4 Card=27325 Bytes=109300)
--------------------------------------------------------------------------------

3.单独使用ordered是执行hash join，也用来选择驱动表（join顺序）

代码:
--------------------------------------------------------------------------------
SQL> select /*+ ordered */ count(*)
2 from t3,t2
3 where t2.object_id=t3.object_id;

SQL> select /*+ ordered */ count(*)
2 from t2,t3
3 where t2.object_id=t3.object_id;

COUNT(*)
----------
      7325
Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=52 Card=1 Bytes=8)
   1    0   SORT (AGGREGATE)
   2    1     HASH JOIN (Cost=52 Card=7325 Bytes=58600)
   3    2       INDEX (FAST FULL SCAN) OF 'IN1' (NON-UNIQUE) (Cost=4 Card=27325 Bytes=109300)
   4    2       TABLE ACCESS (FULL) OF 'T3' (Cost=39 Card=7326 Bytes=29304)
--------------------------------------------------------------------------------

4。第2个例子中，如果t3.object_id也建索引，仍然走hash join，只不过都变成FAST FULL SCAN
原因同2。当然，变成FAST FULL SCAN是因为select count(*)，如果是select *的话就会是FTS了。
代码:
--------------------------------------------------------------------------------
SQL> create index in3 on t3(object_id);

索引已创建。

SQL> set autot on
SQL> select /*+ ordered USE_NL(t3) */ count(*)
2 from t3,t2
3 where t2.object_id=t3.object_id;

COUNT(*)
----------
      7325
Execution Plan
----------------------------------------------------------
   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=13 Card=1 Bytes=8)
   1    0   SORT (AGGREGATE)
   2    1     HASH JOIN (Cost=13 Card=7325 Bytes=58600)
   3    2       INDEX (FAST FULL SCAN) OF 'IN3' (NON-UNIQUE) (Cost=4 Card=7326 Bytes=29304)
   4    2       INDEX (FAST FULL SCAN) OF 'IN1' (NON-UNIQUE) (Cost=4 Card=27325 Bytes=109300)

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

之前对use_nl的理解一直很模糊，看下面的案例。

SQL> select *
2 from table_detail t,
3 table(cast(str2varlist('123') as vartabletype)) t2
4 where t.id = t2.column_value;

Execution Plan
----------------------------------------------------------
Plan hash value: 31598426
----------------------------------------------------------
Operation | Name |Rows |Bytes|Cost(%CPU)|Time|
----------------------------------------------------------
SELECT STATEMENT | |784K|124M|2132(1)|00:00:26|
NESTED LOOPS | |784K|124M|2132(1)|00:00:26|
TABLE ACCESS FULL | table_detail|96 |15744| 11(0)|00:00:01|
COLLECTION ITERATOR PICKLER FETCH| STR2VARLIST | | | | |
-------------------------------------------------------------

t2是个虚拟的表，没有准确的统计信息，很容易出现执行计划错误，这种sql最好要用hint来固化其执行计划。

1.使用/*+ ORDERED use_nl(t2,t) */提示

SQL> select /*+ ORDERED use_nl(t2,t) */*
2 from table_detail t,
3 table(cast(str2varlist('123') as vartabletype)) t2
4 where t.id = t2.column_value;

Execution Plan
----------------------------------------------------------
Plan hash value: 3775534098

----------------------------------------------------------------------------
Operation | Name |Rows |Bytes|Cost(%CPU)|Time|
----------------------------------------------------------------------------
SELECT STATEMENT | |784K|124M|2132(1)|00:00:26|
NESTED LOOPS | |784K|124M|2132(1)|00:00:26|
TABLE ACCESS FULL | table_detail|96 |15744| 11(0)|00:00:01|
COLLECTION ITERATOR PICKLER FETCH| STR2VARLIST | | | | |
----------------------------------------------------------------------------

执行计划是错误的，还是取table_detail为驱动表。
错误理解：使用order提示，执行计划会去取use_nl(t2,t)中的t2作为驱动表。
2.使用/*+ ORDERED use_nl(t,t2) */提示

SQL> select /*+ ORDERED use_nl(t,t2) */*
2 from table_detail t,
3 table(cast(str2varlist('123') as vartabletype)) t2
4 where t.id = t2.column_value;

Execution Plan
----------------------------------------------------------
Plan hash value: 3775534098

----------------------------------------------------------------------------
Operation | Name |Rows |Bytes|Cost(%CPU)|Time|
----------------------------------------------------------------------------
SELECT STATEMENT | |784K|124M|2132(1)|00:00:26|
NESTED LOOPS | |784K|124M|2132(1)|00:00:26|
TABLE ACCESS FULL | table_detail|96 |15744| 11(0)|00:00:01|
COLLECTION ITERATOR PICKLER FETCH| STR2VARLIST | | | | |
----------------------------------------------------------------------------

执行计划还是错误的，这到底是怎么回事？
3.使用/*+ ORDERED use_nl(t,t2) */提示，同时调整了表顺序。

SQL> select /*+ ORDERED use_nl(t2,t) */*
2 from table(cast(str2varlist('123') as vartabletype)) t2,
3 table_detail t
4 where t.id = t2.column_value;

Execution Plan
----------------------------------------------------------
Plan hash value: 2272521841

-------------------------------------------------------------------------------
Operation | Name Rows |Bytes |Cost (%CPU)|Time|
-------------------------------------------------- ----------------------------
SELECT STATEMENT | |784K| 124M|25 (0)|00:00:01|
TABLE ACCESS BY INDEX ROWID | table_detail | 96 |15744| 1 (0)|00:00:01|
NESTED LOOPS | |784K| 124M|25 (0)|00:00:01|
COLLECTION ITERATOR PICKLER FETCH| STR2VARLIST | | | | |
INDEX RANGE SCAN | IND_deail_id | 1 | | 1 (0)|00:00:01|
-------------------------------------------------- ----------------------------

这回执行计划是正确的，原来order是根据from表顺序来决定驱动表，而不是use_nl(t,t2)中表的先后顺序。
调整from后面的表顺序是很老土的调优，仿佛回到了rbo的年代，采用leading指定驱动表看是否可行。
4.使用leading提示来指定驱动表

SQL> select /*+ leading(t2) use_nl(t) */*
2 from table_detail t,
3 table(cast(str2varlist('123') as vartabletype)) t2
4 where t.id = t2.column_value;

Execution Plan
----------------------------------------------------------
Plan hash value: 2272521841
-------------------------------------------------------------------------------
Operation | Name Rows |Bytes |Cost (%CPU)|Time|
-------------------------------------------------- ----------------------------
SELECT STATEMENT | |784K| 124M|25 (0)|00:00:01|
TABLE ACCESS BY INDEX ROWID | table_detail | 96 |15744| 1 (0)|00:00:01|
NESTED LOOPS | |784K| 124M|25 (0)|00:00:01|
COLLECTION ITERATOR PICKLER FETCH| STR2VARLIST | | | | |
INDEX RANGE SCAN | IND_deail_id | 1 | | 1 (0)|00:00:01|
-------------------------------------------------- ----------------------------

1./*+ use_nl(t2,t) */提示走nest Loop,但是没有提示t2还是t为驱动表

2./*+ ordered user_nl(t2,t) */提示走 Nest Loop,order提示的是from 后面的第一个表为驱动表.

3./*+ leading(t2) use_nl(t) */直接提示t2为驱动表。

结论：use_NL不能让优化器确定谁是驱动表谁是被驱动表。use_nl(t,t2)也没有指出哪个是驱动表，这时候我们就需要使用Ordered ,Leading来强制指定驱动表，以达到我们的目的。