Oracle学习笔记（二）——嵌套循环连接
连接方式
Oracle的表连接方式有以下三种：
- 嵌套循环链接（Nested Loops Join）
- 哈希链接（Hash Join）
- 排序合并（Merge Sort Join）
连接涉及到两个表A和B，通俗的讲嵌套循环链接相当于遍历A中的每一条记录（满足A表条件），然后再在B表中遍历记录直至找到匹配的记录，等同于两层循环。而哈希链接和排序合并，可以看作是先各自处理自身的记录（排序或hash），处理完成后再批量的进行匹配。哈希链接和排序合并链接适合处理返回大量数据的操作，大吞吐量的查询对于嵌套循环来说效率不会太高。
准备工作是建立两个表test_join_t1和test_join_t2，两个表分别有记录100条和10000条。
create table test_join_t1 as select rownum as col1, rownum + 1 as col2, rownum + 2 as col3 from dual connect by level <= 100;
create table test_join_t2 as select rownum as col1, rownum + 1 as col2, rownum + 2 as col3 from dual connect by level <= 10000;
嵌套循环连接
嵌套循环连接，在一般的应用中占有绝大多数的比例。首先执行最基本的连接查询语句
select
*
from
test_join_t1 t1, test_join_t2 t2
where
t1.col1 = t2.col1;
执行计划
----------------------------------------------------------
Plan hash value: 1262213046
-----------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-----------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 100 | 7800 | 11 (10)| 00:00:01 |
|* 1 | HASH JOIN | | 100 | 7800 | 11 (10)| 00:00:01 |
| 2 | TABLE ACCESS FULL| TEST_JOIN_T1 | 100 | 3900 | 2 (0)| 00:00:01 |
| 3 | TABLE ACCESS FULL| TEST_JOIN_T2 | 10000 | 380K| 8 (0)| 00:00:01 |
-----------------------------------------------------------------------------------
select
/*+ leading(t1) use_nl(t2) */
*
from
test_join_t1 t1, test_join_t2 t2
where
t1.col1 = t2.col1;
Plan hash value: 3805247585
---------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers |
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | 100 |00:00:00.10 | 2717 |
| 1 | NESTED LOOPS | | 1 | 100 | 100 |00:00:00.10 | 2717 |
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
| 2 | TABLE ACCESS FULL| TEST_JOIN_T1 | 1 | 100 | 100 |00:00:00.01 | 10 |
|* 3 | TABLE ACCESS FULL| TEST_JOIN_T2 | 100 | 1 | 100 |00:00:00.10 | 2707 |
---------------------------------------------------------------------------------------------
select
/*+ leading(t2) use_nl(t1) */
*
from
test_join_t1 t1, test_join_t2 t2
where
t1.col1 = t2.col1;
Plan hash value: 2586695766
---------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers |
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | 100 |00:00:00.18 | 30041 |
| 1 | NESTED LOOPS | | 1 | 100 | 100 |00:00:00.18 | 30041 |
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
| 2 | TABLE ACCESS FULL| TEST_JOIN_T2 | 1 | 10000 | 10000 |00:00:00.01 | 34 |
|* 3 | TABLE ACCESS FULL| TEST_JOIN_T1 | 10000 | 1 | 100 |00:00:00.17 | 30007 |
---------------------------------------------------------------------------------------------
先介绍一下执行计划中的各个字段含义：
- Starts：表访问次数
- E-Rows：Oracle预估返回条数
- A-Rows：Oracle真实返回记录条数
- A-Time：真实执行时间
通过对比以上两次执行计划可以得出一下结论：
- 驱动表只被访问一次
- 被驱动表访问的次数跟驱动表的记录数一致（应该是驱动表返回记录数一致，后面会再做测试）
接下来让t1表作为驱动表加上where限制条件col2在[10,50)范围内
select
/*+ leading(t1) use_nl(t2) */
*
from
test_join_t1 t1, test_join_t2 t2
where
t1.col1 = t2.col1 and
t1.col2 >= 10 and
t1.col2 < 50;
Plan hash value: 3805247585
---------------------------------------------------------------------------------------------
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers |
---------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | 40 |00:00:00.04 | 1089 |
| 1 | NESTED LOOPS | | 1 | 40 | 40 |00:00:00.04 | 1089 |
|* 2 | TABLE ACCESS FULL| TEST_JOIN_T1 | 1 | 40 | 40 |00:00:00.01 | 6 |
|* 3 | TABLE ACCESS FULL| TEST_JOIN_T2 | 40 | 1 | 40 |00:00:00.04 | 1083 |
---------------------------------------------------------------------------------------------
select
/*+ leading(t1) use_nl(t2) */
*
from
test_join_t1 t1, test_join_t2 t2
where
t1.col1 = t2.col1 and
t2.col2 >= 10 and
t2.col2 < 50;
Plan hash value: 3805247585
---------------------------------------------------------------------------------------------
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers |
---------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | 40 |00:00:00.08 | 2709 |
| 1 | NESTED LOOPS | | 1 | 40 | 40 |00:00:00.08 | 2709 |
| 2 | TABLE ACCESS FULL| TEST_JOIN_T1 | 1 | 100 | 100 |00:00:00.01 | 6 |
|* 3 | TABLE ACCESS FULL| TEST_JOIN_T2 | 100 | 1 | 40 |00:00:00.08 | 2703 |
---------------------------------------------------------------------------------------------
从执行计划可以看出，驱动表t1访问次数仍然是1，被驱动表访问次数为100，而非限制条件所限定的40条。因此，可以得出嵌套循环的特点：
- 驱动表只被访问1次
- 被驱动表访问的次数为驱动表的“实际返回记录数”，被驱动表添加查询条件是不会影响被访问次数的。
接下来在t2表的col2上建立索引，重新执行上面的sql看执行计划如何
create index index_t2_col2 on test_join_t2(col2);
select
/*+ leading(t1) use_nl(t2) */
*
from
test_join_t1 t1, test_join_t2 t2
where
t1.col1 = t2.col1 and
t2.col2 >= 10 and
t2.col2 < 50;
Plan hash value: 3428653669
-----------------------------------------------------------------------------------------------------------------
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows| A-Time | Buffers | Reads |
-----------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | 40|00:00:00.01 | 24 | 2|
| 1 | NESTED LOOPS | | 1 | | 40|00:00:00.01 | 24 | 2|
| 2 | NESTED LOOPS | | 1 | 40 | 4000|00:00:00.01 | 20 | 2|
| 3 | TABLE ACCESS FULL | TEST_JOIN_T1 | 1 | 100 | 100|00:00:00.01 | 6 | 0|
|* 4 | INDEX RANGE SCAN | INDEX_T2_COL2 | 100 | 45 | 4000|00:00:00.01 | 14 | 2|
|* 5 | TABLE ACCESS BY INDEX ROWID| TEST_JOIN_T2 | 4000 | 1 | 40|00:00:00.01 | 4 | 0|
----------------------------------------------------------------------------------------------------------------
select
/*+ leading(t2) use_nl(t1) */
*
from
test_join_t1 t1, test_join_t2 t2
where
t1.col1 = t2.col1 and
t2.col2 >= 10 and
t2.col2 < 50;
Plan hash value: 2941912635
--------------------------------------------------------------------------------------------------------
PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows| A-Time | Buffers |
--------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | 40|00:00:00.01 | 132 |
| 1 | NESTED LOOPS | | 1 | 40 | 40|00:00:00.01 | 132 |
| 2 | TABLE ACCESS BY INDEX ROWID| TEST_JOIN_T2 | 1 | 40 | 40|00:00:00.01 | 9 |
|* 3 | INDEX RANGE SCAN | INDEX_T2_COL2 | 1 | 45 | 40|00:00:00.01 | 5 |
|* 4 | TABLE ACCESS FULL | TEST_JOIN_T1 | 40 | 1 | 40|00:00:00.01 | 123 |
--------------------------------------------------------------------------------------------------------
对于驱动表t2先走利用了索引获取到最终的记录，然后再进入循环访问被驱动表t1。经过上面的测试，可以看出嵌套循环就是由两层for循环操作构成：
- 查询驱动表记录获取满足查询条件的所有记录
- 遍历第一步得到的驱动表记录，带入链接条件值获取满足链接条件的所有被驱动表记录
- 遍历第二步中得到的被驱动表记录，判断是否满足查询条件，如果是则返回结果，如果不是则跳过记录
到此可以总结一下嵌套循环的主要特点以及适用场景：
- 嵌套循环有驱动表和被驱动表的概念，驱动顺序不同执行计划差异非常大
- 驱动表只被访问一次，被驱动表被访问多次。嵌套循环访问表的次数直接受驱动表的返回记录数的影响。因此应当让实际返回记录数（A-Rows）小的表作为驱动表，返回记录数大的表作为被驱动表。
- 在驱动表的查询条件上建立索引可以改善查询效率
- 在连接条件上建立索引也可以改善查询效率
- 在被驱动表查询条件建立索引对查询效率的影响要视情况而定，不一定带来好处