優化擁有謂詞or的子查詢

於節點2上發現一個sql，其結構如下：
select distinct t.attr, t.item
 from a t
where attr = :1
  and (pro = :2 or exists
       (select 1
          from b hps
         where hps.pro = :3
           and t.pro = hps.sub_pro))
鑑於以往的經驗，一開始就覺得子查詢中的連線謂詞or有問題。
注:
表a有50多萬條記錄，attr和pro上分別有索引；表b有10萬條記錄，pro上有單獨索引，同時(pro,sub_pro)組成unique索引；
同時表a上的attr,pro欄位的分佈情況如下,對於表a，pro欄位的選擇性比attr高出了很多
SQL> select count(distinct pro),count(distinct attr) from a;
 
COUNT(DISTINCT pro)           COUNT(DISTINCT attr)
------------------------- ---------------------------
          164067                         716
          
 
先通過set autotrace traceonly檢視一把其執行計劃和consistent gets
 --------------------------------------------------------------------------------------------------------
| Id  | Operation                     | Name                   | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT              |                        |    63 |   945 |    83   (2)| 00:00:01 |
|   1 |  HASH UNIQUE                  |                        |    63 |   945 |    83   (2)| 00:00:01 |
|*  2 |   FILTER                      |                        |       |       |            |          |
|   3 |    TABLE ACCESS BY INDEX ROWID| a                      |  1268 | 19020 |    82   (0)| 00:00:01 |
|*  4 |     INDEX RANGE SCAN          | idx_a_attr             |  1268 |       |     8   (0)| 00:00:01 |
|*  5 |    INDEX UNIQUE SCAN          | idx_b_un_pro_sub       |     1 |    12 |     1   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------------------

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

   2 - filter("pro"=TO_NUMBER(:B) OR  EXISTS (SELECT 0 FROM "b" "HPS" WHERE
              "HPS"."SUB_pro"=:B1 AND "HPS"."pro"=TO_NUMBER(:C)))
   4 - access("attr"=TO_NUMBER(:A))
   5 - access("HPS"."pro"=TO_NUMBER(:C) AND "HPS"."SUB_pro"=:B1)

Statistics
----------------------------------------------------------
          0  recursive calls
          0  db block gets
       4690  consistent gets
          0  physical reads
          0  redo size
        589  bytes sent via SQL*Net to client
        487  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
          1  rows processed
 
從執行計劃看出，訪問表a的時候選擇了attr上的索引，實際效果不佳，邏輯讀有4690；
通過等價改寫後的sql如下，去除了謂詞or，改用了union
select distinct t.attr, t.item
       from a t
      where attr = :a
        and pro = :b
union
select distinct t.attr, t.item
       from a t
      where attr = :a
            and exists
             (select 1
                from b hps
               where hps.pro = :c
                 and t.pro = hps.sub_pro)
改寫後的sql的執行計劃如下               
--------------------------------------------------------------------------------------------------------
| Id  | Operation                     | Name                   | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT              |                        |     6 |   150 |    16  (75)| 00:00:01 |
|   1 |  SORT UNIQUE                  |                        |     6 |   150 |    16  (75)| 00:00:01 |
|   2 |   UNION-ALL                   |                        |       |       |            |          |
|*  3 |    TABLE ACCESS BY INDEX ROWID| a                      |     1 |    15 |     4   (0)| 00:00:01 |
|*  4 |     INDEX RANGE SCAN          | idx_a_pro              |     3 |       |     3   (0)| 00:00:01 |
|*  5 |    TABLE ACCESS BY INDEX ROWID| a                      |     1 |    15 |     3   (0)| 00:00:01 |
|   6 |     NESTED LOOPS              |                        |     5 |   135 |    10  (10)| 00:00:01 |
|   7 |      SORT UNIQUE              |                        |     5 |    60 |     2   (0)| 00:00:01 |
|*  8 |       INDEX RANGE SCAN        | idx_b_un_pro_sub       |     5 |    60 |     2   (0)| 00:00:01 |
|*  9 |      INDEX RANGE SCAN         | idx_a_pro              |     3 |       |     2   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------------------

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

   3 - filter("attr"=TO_NUMBER(:A))
   4 - access("pro"=TO_NUMBER(:B))
   5 - filter("attr"=TO_NUMBER(:A))
   8 - access("HPS"."pro"=TO_NUMBER(:C))
   9 - access("T"."pro"="HPS"."SUB_pro")

Statistics
----------------------------------------------------------
          0  recursive calls
          0  db block gets
         24  consistent gets
          0  physical reads
          0  redo size
        589  bytes sent via SQL*Net to client
        487  bytes received via SQL*Net from client
          2  SQL*Net roundtrips to/from client
          2  sorts (memory)
          0  sorts (disk)
          1  rows processed
訪問表a時選擇了pro上的索引，雖然增加了一次訪問次數，但是邏輯讀卻下降到了24.

來自 “ ITPUB部落格 ” ，連結：http://blog.itpub.net/15480802/viewspace-702062/，如需轉載，請註明出處，否則將追究法律責任。