關於Oracle full outer join 的bug問題分析及處理

full (outer) join是用來全連線兩個表的語法。即希望將A表和B表關聯，能夠得到A表中有而B表中沒有的記錄，或者B表中有而A表中沒有的記錄。

如何判斷是否有該記錄，則透過on子句來關聯。

下面是一個例子：

SQL> with
  2  A as(select 1 a, 2 b from dual),
  3  B as(select 2 a, 3 b from dual)
  4  select * from A full join B
  5      on A.a=B.a
  6  /
 
         A          B          A          B
---------- ---------- ---------- ----------
         1          2           
                                  2          3
 

瞭解了以上基本原理後，我們應該知道，理論上講，A表和B表的在from子句中的順序是沒有關係的，也就是不影響結果。但是，實際上，卻出現了這樣的問題，下面是對這種情況的描述：

--------------------------------------------------------------------------------------------
| Id  | Operation                | Name            | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT         |                 | 12791 |   349K|    82   (3)| 00:00:01 |
|   1 |  VIEW                    |                 | 12791 |   349K|    82   (3)| 00:00:01 |
|   2 |   UNION-ALL              |                 |       |       |            |          |
|*  3 |    FILTER                |                 |       |       |            |          |
|*  4 |     HASH JOIN RIGHT OUTER|                 | 12790 |  1124K|    41   (3)| 00:00:01 |
|   5 |      TABLE ACCESS FULL   | JXC_RISHARESUM  |  1735 | 78075 |     7   (0)| 00:00:01 |
|   6 |      TABLE ACCESS FULL   | JXC_ALLTRADEDAY | 12790 |   562K|    33   (0)| 00:00:01 |
|*  7 |    HASH JOIN ANTI        |                 |     1 |    76 |    41   (3)| 00:00:01 |
|*  8 |     TABLE ACCESS FULL    | JXC_RISHARESUM  |     1 |    45 |     7   (0)| 00:00:01 |
|   9 |     TABLE ACCESS FULL    | JXC_ALLTRADEDAY | 12790 |   387K|    33   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------

 

從以上執行計劃來看，在第四步驟，使用的是hash join rigth outer連線方式。而透過改變兩表的擺放順序，得到如下的執行計劃：

 

-----------------------------------------------------------------------------------------
| Id  | Operation             | Name            | Rows  | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |                 |  1876 | 52528 |    82   (3)| 00:00:01 |
|   1 |  VIEW                 |                 |  1876 | 52528 |    82   (3)| 00:00:01 |
|   2 |   UNION-ALL           |                 |       |       |            |          |
|*  3 |    FILTER             |                 |       |       |            |          |
|*  4 |     HASH JOIN OUTER   |                 |  1874 |   164K|    41   (3)| 00:00:01 |
|   5 |      TABLE ACCESS FULL| JXC_RISHARESUM  |  1735 | 78075 |     7   (0)| 00:00:01 |
|   6 |      TABLE ACCESS FULL| JXC_ALLTRADEDAY | 12790 |   562K|    33   (0)| 00:00:01 |
|*  7 |    HASH JOIN ANTI     |                 |     2 |   152 |    41   (3)| 00:00:01 |
|*  8 |     TABLE ACCESS FULL | JXC_ALLTRADEDAY |     2 |    90 |    33   (0)| 00:00:01 |
|   9 |     TABLE ACCESS FULL | JXC_RISHARESUM  |  1735 | 53785 |     7   (0)| 00:00:01 |
-----------------------------------------------------------------------------------------

 

注意，此時，執行計劃中的第四個步驟，變成了：hash join outer方式。這個才是我們所期望的方式。那究竟是什麼導致了這個變化呢？檢視他們的謂詞連線邏輯：

hash join right outer的：

   3 - filter(NVL("T1"."C_SHARENO","T2"."C_SHARENO")='26200703200004969020')
   4 - access("T2"."D_TRADEDATE"="T1"."D_TRADEDATE"(+) AND
              "T2"."D_REGDATE"="T1"."D_REGDATE"(+) AND "T2"."C_FUNDCODE"="T1"."C_FUNDCODE"(+) AND
              "T2"."C_FUNDACCO"="T1"."C_FUNDACCO"(+))
   7 - access("T2"."C_FUNDACCO"="T1"."C_FUNDACCO" AND
              "T2"."C_FUNDCODE"="T1"."C_FUNDCODE" AND "T2"."D_REGDATE"="T1"."D_REGDATE" AND
              "T2"."D_TRADEDATE"="T1"."D_TRADEDATE")
   8 - filter(NVL("T1"."C_SHARENO",NULL)='26200703200004969020')

 

hash join outer的：

   3 - filter(NVL("T1"."C_SHARENO","T2"."C_SHARENO")='26200703200004969020')
   4 - access("T2"."D_TRADEDATE"(+)="T1"."D_TRADEDATE" AND
              "T2"."D_REGDATE"(+)="T1"."D_REGDATE" AND "T2"."C_FUNDCODE"(+)="T1"."C_FUNDCODE"
              AND "T2"."C_FUNDACCO"(+)="T1"."C_FUNDACCO")
   7 - access("T2"."C_FUNDACCO"="T1"."C_FUNDACCO" AND
              "T2"."C_FUNDCODE"="T1"."C_FUNDCODE" AND "T2"."D_REGDATE"="T1"."D_REGDATE" AND
              "T2"."D_TRADEDATE"="T1"."D_TRADEDATE")
   8 - filter("T2"."C_SHARENO"='26200703200004969020')

還是沒有發現明顯的區別。但是實際卻導致了結果的不同。

還原到原始的表連線順序，然後對兩表進行分析，再檢視執行計劃：

SQL> call dbms_stats.gather_table_stats(user, 'JXC_ALLTRADEDAY');

呼叫完成。

SQL> call dbms_stats.gather_table_stats(user, 'JXC_RISHARESUM');

.....

-----------------------------------------------------------------------------------------
| Id  | Operation             | Name            | Rows  | Bytes | Cost (%CPU)| Time     |
-----------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |                 |  1738 | 48664 |    82   (3)| 00:00:01 |
|   1 |  VIEW                 |                 |  1738 | 48664 |    82   (3)| 00:00:01 |
|   2 |   UNION-ALL           |                 |       |       |            |          |
|*  3 |    FILTER             |                 |       |       |            |          |
|*  4 |     HASH JOIN OUTER   |                 |  1735 |   191K|    41   (3)| 00:00:01 |
|   5 |      TABLE ACCESS FULL| JXC_RISHARESUM  |  1735 | 98895 |     7   (0)| 00:00:01 |
|   6 |      TABLE ACCESS FULL| JXC_ALLTRADEDAY | 12775 |   698K|    33   (0)| 00:00:01 |
|*  7 |    HASH JOIN ANTI     |                 |     3 |   276 |    41   (3)| 00:00:01 |
|*  8 |     TABLE ACCESS FULL | JXC_ALLTRADEDAY |     3 |   168 |    33   (0)| 00:00:01 |
|   9 |     TABLE ACCESS FULL | JXC_RISHARESUM  |  1735 | 62460 |     7   (0)| 00:00:01 |
-----------------------------------------------------------------------------------------

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

   3 - filter(NVL("T1"."C_SHARENO","T2"."C_SHARENO")='26200703200004969020')
   4 - access("T2"."D_TRADEDATE"(+)="T1"."D_TRADEDATE" AND
              "T2"."D_REGDATE"(+)="T1"."D_REGDATE" AND "T2"."C_FUNDCODE"(+)="T1"."C_FUNDCODE"
              AND "T2"."C_FUNDACCO"(+)="T1"."C_FUNDACCO")
   7 - access("T2"."C_FUNDACCO"="T1"."C_FUNDACCO" AND
              "T2"."C_FUNDCODE"="T1"."C_FUNDCODE" AND "T2"."D_REGDATE"="T1"."D_REGDATE" AND
              "T2"."D_TRADEDATE"="T1"."D_TRADEDATE")
   8 - filter("T2"."C_SHARENO"='26200703200004969020')

 

可發現，這時原來連線方式的錯誤執行計劃被修正了，改為hash join outer連線。