Oracle中connect by语句的优化

很多应用中都会有类似组织机构的表，组织机构的表又通常是典型的层次结构（没有循环节点）。于是通过组织控制数据权限的时候，许多人都喜欢通过connect by获得组织信息，然后再过滤目标数据。

在有些情况下，这样写并没有什么问题，但有些情况下，这个就是一个大问题。

归根结底，这是connect by特性导致的，Oracle无法知道connect by之后到底返回多少数据，所以有可能采取一些你所不期望的算法，结果自然不是你所期望的---非常慢。

下面，我就讨论在Oracle 12.1.0.2中如果遇到这样的语句应该如何处理。

为了很好理解，我做了3表：

执行SQL:

SELECT A.CI, A.ENBAJ02 AS CELL_NAME
  FROM TDL_CM_CELL A, T_ORG_CELL_SCOPE S
 WHERE S.REGION_NAME = A.REGION_NAME
  AND S.CITY_NAME = A.CITY_NAME
  AND (S.ORG_ID) IN (SELECT ID
                        FROM T_ORG O
                      START WITH ID = 101021003 --1010210 
                      --START WITH ID=1
                      CONNECT BY PARENT_ID = PRIOR ID)

实际使用的执行计划:

而不会采用自适应计划（adaptive plan)：

Plan Hash Value  : 2596385940 

-------------------------------------------------------------------------------------------------------------------
| Id  | Operation                                       | Name                  | Rows | Bytes  | Cost | Time     |
-------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                                |                       | 2622 | 228114 |  227 | 00:00:01 |
|   1 |   NESTED LOOPS                                  |                       | 2622 | 228114 |  227 | 00:00:01 |
|   2 |    NESTED LOOPS                                 |                       | 2622 | 228114 |  227 | 00:00:01 |
| * 3 |     HASH JOIN                                   |                       |    1 |     31 |    7 | 00:00:01 |
|   4 |      VIEW                                       | VW_NSO_1              |    1 |     13 |    4 | 00:00:01 |
|   5 |       HASH UNIQUE                               |                       |    1 |     20 |    4 | 00:00:01 |
| * 6 |        CONNECT BY NO FILTERING WITH SW (UNIQUE) |                       |      |        |      |          |
|   7 |         TABLE ACCESS FULL                       | T_ORG                 |   75 |    825 |    3 | 00:00:01 |
|   8 |      TABLE ACCESS FULL                          | T_ORG_CELL_SCOPE      |   85 |   1530 |    3 | 00:00:01 |
| * 9 |     INDEX RANGE SCAN                            | IDX_TDL_CM_CELL_SCOPE |  257 |        |    8 | 00:00:01 |
|  10 |    TABLE ACCESS BY INDEX ROWID                  | TDL_CM_CELL           | 2313 | 129528 |  220 | 00:00:01 |
-------------------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
------------------------------------------
* 3 - access("S"."ORG_ID"="ID")
* 6 - access("PARENT_ID"=PRIOR "ID")
* 6 - filter("ID"=101021003)
* 9 - access("S"."REGION_NAME"="A"."REGION_NAME" AND "S"."CITY_NAME"="A"."CITY_NAME")


Notes
-----
- This is an adaptive plan

原因在于，oracle无法知道connect by之后的数量，所以只能认为是很大的量

--

有一种方式就是，就是使用提示来解决：

SELECT /*+ no_merge(x) use_nl(a x) */
  A.CI, A.ENBAJ02 AS CELL_NAME
   FROM TDL_CM_CELL A,
        (select s.city_name, s.region_name
           from T_ORG_CELL_SCOPE S
          WHERE (S.ORG_ID) IN
                (SELECT ID
                   FROM T_ORG O
                  START WITH ID = 101021003 --1010210  
                 --START WITH ID=1
                 CONNECT BY PARENT_ID = PRIOR ID)
         
         ) x
  where x.REGION_NAME = A.REGION_NAME
    AND x.CITY_NAME = A.CITY_NAME

这样计划就是：

Plan Hash Value  : 37846894 

---------------------------------------------------------------------------------------------------------------------
| Id   | Operation                                        | Name                  | Rows | Bytes  | Cost | Time     |
---------------------------------------------------------------------------------------------------------------------
|    0 | SELECT STATEMENT                                 |                       | 2313 | 277560 |  227 | 00:00:01 |
|    1 |   NESTED LOOPS                                   |                       | 2313 | 277560 |  227 | 00:00:01 |
|    2 |    NESTED LOOPS                                  |                       | 2313 | 277560 |  227 | 00:00:01 |
|    3 |     VIEW                                         |                       |    1 |     64 |    7 | 00:00:01 |
|  * 4 |      HASH JOIN                                   |                       |    1 |     31 |    7 | 00:00:01 |
|    5 |       VIEW                                       | VW_NSO_1              |    1 |     13 |    4 | 00:00:01 |
|    6 |        HASH UNIQUE                               |                       |    1 |     20 |    4 | 00:00:01 |
|  * 7 |         CONNECT BY NO FILTERING WITH SW (UNIQUE) |                       |      |        |      |          |
|    8 |          TABLE ACCESS FULL                       | T_ORG                 |   75 |    825 |    3 | 00:00:01 |
|    9 |       TABLE ACCESS FULL                          | T_ORG_CELL_SCOPE      |   85 |   1530 |    3 | 00:00:01 |
| * 10 |     INDEX RANGE SCAN                             | IDX_TDL_CM_CELL_SCOPE |  257 |        |    8 | 00:00:01 |
|   11 |    TABLE ACCESS BY INDEX ROWID                   | TDL_CM_CELL           | 2313 | 129528 |  220 | 00:00:01 |
---------------------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
------------------------------------------
* 4 - access("S"."ORG_ID"="ID")
* 7 - access("PARENT_ID"=PRIOR "ID")
* 7 - filter("ID"=101021003)
* 10 - access("X"."REGION_NAME"="A"."REGION_NAME" AND "X"."CITY_NAME"="A"."CITY_NAME") 

如果一个应用的start id可能是一个很大的范围，如果强制使用提示，也会出现问题，所以如果有这样的应用，可以考虑使用oracle 12c的adaptive特性。

如果不行，就必须把不同范围的查询，定义为不同的功能提交给用户。