优化器生成正确执行计划的前提条件是要有正确的统计信息，不准确的统计信息往往会导致错误的执行计划。当通过SQL和基数推断出的执行计划和实际执行计划不同时，就可以借助10053事件。10053事件是用来诊断优化器如何估算成本和选择执行计划的，用它产生的trace文件提供了Oracle如何选择执行计划，为什么会得到这样的执行计划信息。和10046事件类似，它主要用于特殊情况下的分析和诊断。
1、测试环境：
SQL> select * from v$version;
BANNER
----------------------------------------------------------------------
Oracle Database 11g Enterprise Edition Release 11.2.0.1.0 - Production
2、建立测试对象
SQL> create table tabtemp as select * from dba_objects where object_id is not null;
Table created.
SQL> select count(object_id) from tabtemp;
COUNT(OBJECT_ID)
----------------
72764
测试表object_id列的数值分布：
SQL> select count(distinct object_id) from tabtemp;
COUNT(DISTINCTOBJECT_ID)
------------------------
72764
建立索引：
SQL> create index idx_tabtemp_id on tabtemp(object_id);
3、生成10053事件
统计表及索引信息：
SQL> exec dbms_stats.gather_table_stats(user,'TABTEMP',cascade=>true);
查看执行计划：
SQL> alter session set tracefile_identifier='plan';
SQL> set autotrace trace exp;
SQL> alter session set events '10053 trace name context forever,level 1';
SQL> select * from tabtemp where object_id=3;
Execution Plan
----------------------------------------------------------
Plan hash value: 2221486709
--------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 97 | 2 (0)| 00:00:01 |
| 1 | TABLE ACCESS BY INDEX ROWID| TABTEMP | 1 | 97 | 2 (0)| 00:00:01 |
|* 2 | INDEX RANGE SCAN | IDX_TABTEMP_ID | 1 | | 1 (0)| 00:00:01 |
-------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"=3)
由执行计划可知，查询走索引，这是非常高效的查询方式。
更新测试表，将object_id列数值全部设置为3：
SQL> update tabtemp set object_id=3 where object_id!=3;
72763 rows updated.
SQL> commit;
SQL> select count(distinct object_id) from tabtemp;
COUNT(DISTINCTOBJECT_ID)
------------------------
1
不收集统计数据，查看执行计划：
SQL> set autotrace trace exp;
SQL> select * from tabtemp where object_id=3;
Execution Plan
----------------------------------------------------------
Plan hash value: 2221486709
-------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 97 | 2 (0)| 00:00:01 |
| 1 | TABLE ACCESS BY INDEX ROWID| TABTEMP | 1 | 97 | 2 (0)| 00:00:01 |
|* 2 | INDEX RANGE SCAN | IDX_TABTEMP_ID | 1 | | 1 (0)| 00:00:01 |
-------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJECT_ID"=3)
由输出结果可知，本次查询沿用原来的执行计划，是错误的执行计划。
重新对更新后的测试对象进行数据分析：
SQL> set autotrace off;
SQL> exec dbms_stats.gather_table_stats(user,'TABTEMP',cascade=>true);
查看收集统计数据后的执行计划：
SQL> set autotrace trace exp;
SQL> select * from tabtemp where object_id=3;
Execution Plan
----------------------------------------------------------
Plan hash value: 3955501171
-----------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-----------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 72757 | 6749K| 293 (2)| 00:00:04 |
|* 1 | TABLE ACCESS FULL| TABTEMP | 72757 | 6749K| 293 (2)| 00:00:04 |
-----------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("OBJECT_ID"=3)
由输出可知，本次查询使用了正确的执行计划。所以，要注意在实际生产环境中对表、索引等进行及时有效的统计数据收集工作，避免因此带来性能问题。
SQL> alter session set events '10053 trace name context off';
SQL> select value from v$diag_info where name='Default Trace File';
VALUE
------------------------------------------------------------------------------
c:appadministratordiag
dbmsorcl11gorcl11g raceorcl11g_ora_5952_plan.trc
4、分析10053事件trace文件中CBO出错的位置
#more orcl11g_ora_5952_plan.trc
在前面模拟中有如下操作：
SQL> update tabtemp set object_id=3 where object_id!=3;
72763 rows updated.
SQL> commit;
SQL> select count(distinct object_id) from tabtemp;
COUNT(DISTINCTOBJECT_ID)
------------------------
1
SQL> select * from tabtemp where object_id=3; 此处没有重新进行统计信息收集，直接发起查询。
查看10053trace文件中相对应的内容：