bind peeking--绑定变量窥视

绑定变量窥视，就是指oracle在第一次解析SQL语句的时候（也就是说该SQL第一次传入shared pool），会将你输入的绑定变量的值带入SQL语句里，从而参考你的字面值来猜测该SQL大概会返回多少条记录，从而得到优化的执行计划。然后，以后再次执行相同的SQL语句时，不再考虑你所输入的绑定变量的值，直接取出第一次生成的绑定变量。但是，很可惜的是，使用绑定变量从而共享游标与SQL优化是两个矛盾的目标。Oracle使用绑定变量的前提，是oracle认为大部分的列的数据都是分布比较均匀的。从而，使用第一次的绑定变量的值所得到的执行计划，大多数情况下都能适用于该绑定变量的其他的值。很明显，如果第一次传入的绑定变量的值恰好占整个数据量的百分比较高，从而导致全表扫描的执行计划。而后来传入的绑定变量的值都占整个数据量的百分比都很低，则应该走索引扫描会更好的，但是由于使用了绑定变量，从而oracle并不会再去看你的绑定变量的值，而是直接拿全表扫描的执行计划来用。这时，由于使用了绑定变量，虽然我们达到了游标共享，从而节省CPU的目的，但是SQL的执行计划却不够优化了。

实验

SCOTT@ fyl>create table t1 as select * from dba_objects;
SCOTT@ fyl>create index t1_id on t1(object_id);
SCOTT@ fyl>execute dbms_stats.gather_table_stats(ownname => 'SCOTT',tabname => 'T1' ,method_opt => 'for all indexed columns' ,cascade => true);

SCOTT@ fyl>var id number;
SCOTT@ fyl>exec :id :=100;
SCOTT@ fyl>set autotrace traceonly;
SCOTT@ fyl>select * from t1 where object_id<:id;
98 rows selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 2288890262
-------------------------------------------------------------------------------------
| Id  | Operation                   | Name  | Rows  | Bytes | Cost (%CPU)| Time     |
-------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |       |  3272 |   313K|    17   (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| T1    |  3272 |   313K|    17   (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN          | T1_ID |   589 |       |     3   (0)| 00:00:01 |
-------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   2 - access("OBJECT_ID"<TO_NUMBER(:ID))

Statistics
----------------------------------------------------------
          1  recursive calls
          0  db block gets
         18  consistent gets
          0  physical reads
          0  redo size
      10120  bytes sent via SQL*Net to client
        485  bytes received via SQL*Net from client
          8  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
         98  rows processed

SCOTT@ fyl>exec :id :=50000;
SCOTT@ fyl>select * from t1 where object_id<:id;
49588 rows selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 2288890262
-------------------------------------------------------------------------------------
| Id  | Operation                   | Name  | Rows  | Bytes | Cost (%CPU)| Time     |
-------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |       |  3272 |   313K|    17   (0)| 00:00:01 |
|   1 |  TABLE ACCESS BY INDEX ROWID| T1    |  3272 |   313K|    17   (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN          | T1_ID |   589 |       |     3   (0)| 00:00:01 |
-------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   2 - access("OBJECT_ID"<TO_NUMBER(:ID))

Statistics
----------------------------------------------------------
          0  recursive calls
          0  db block gets
       7887  consistent gets
          0  physical reads
          0  redo size
    5495043  bytes sent via SQL*Net to client
      36774  bytes received via SQL*Net from client
       3307  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
      49588  rows processed

由于使用了绑定变量导致两次执行计划一样，Plan hash value: 2288890262
SCOTT@ fyl>select /*+full(t1) */ * from t1 where object_id<:50000;
49588 rows selected.
Execution Plan
----------------------------------------------------------
Plan hash value: 3617692013
--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |  3272 |   313K|   262   (1)| 00:00:04 |
|*  1 |  TABLE ACCESS FULL| T1   |  3272 |   313K|   262   (1)| 00:00:04 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   1 - filter("OBJECT_ID"<TO_NUMBER(:ID))

Statistics
----------------------------------------------------------
          1  recursive calls
          0  db block gets
       4198  consistent gets
          0  physical reads
          0  redo size
    2349310  bytes sent via SQL*Net to client
      36774  bytes received via SQL*Net from client
       3307  SQL*Net roundtrips to/from client
          0  sorts (memory)
          0  sorts (disk)
      49588  rows processed
走索引时    object_id<:50000--  7887  consistent gets
全表扫描时  object_id<:50000--  4198  consistent gets

 OPT_PARAM('_optim_peek_user_binds' 'false')

select /*+opt_param('_optim_peek_user_binds', 'false')*/ id,name from test1 where id=:fid;

11G--ACS其实就是根据不同绑定变量的值为同一个SQL生成更多更优的执行计划，来适应data skew的不同情况
具体可参见 http://blog.itpub.net/15415488/viewspace-621535