11G新特性 -- Multicolumn Statistics (Column groups)

默认oracle会收集表中各个列的统计信息，但是会忽略列之间的关联关系。在大多情况下，优化器假设在复杂查询中的列之间是独立的。当where子句后指定了一个表的多个列条件时，优化器通常会将多个列的选择性（selectivity）相乘得到where语句的选择性，导致优化器做出错误判断！
Oracle 11g引入了多列统计信息概念，如果上面情况列关联性很好，可以做多列统计信息收集，让优化器做出正确判断。

在oracle 10g中，只有在一些特殊场合，优化器才会考虑列之间的关联关系：
-The optimizer used the number of distinct keys in an index to estimate selectivity provided all columns of a conjunctive predicate match all columns of a concatenated index key. In addition, the predicates must be equalities used in equijoins.
- If you set DYNAMIC_SAMPLING to level 4, the optimizer used dynamic sampling to estimate the selectivity of predicates involving multiple columns from a table. Because the sampling size is quite small, the results are dubious in most cases.

创建Column Groups:

DECLARE
  cg_name varchar2();
BEGIN
  cg_name := dbms_stats.create_extended_stats(null,'customers', '(cust_state_province,country_id)');
END;
/

查看Column Groups:

SQL> select extension_name, extension from dba_stat_extensions where table_name='CUSTOMERS';

EXTENSION_NAME                 EXTENSION
------------------------------ --------------------------------------------------------------------------------
SYS_STU#S#WF25Z#QAHIHE#MOFFMM_ ("CUST_STATE_PROVINCE","COUNTRY_ID")

或者
SQL>  select sys.dbms_stats.show_extended_stats_name ('sh','customers','(cust_state_province,country_id)') col_group_name from dual;

COL_GROUP_NAME
--------------------------------------------------
SYS_STU#S#WF25Z#QAHIHE#MOFFMM_

删除:

SQL> exec dbms_stats.drop_extended_stats('sh','customers','(cust_state_province, country_id)');

收集Column Groups的统计信息:

SQL> exec dbms_stats.gather_table_stats('sh','customers',method_opt =>'for all columns size skewonly for columns (cust_state_province,country_id) size skewonly');

监控Column Groups:

--查询多列统计信息
SQL> Select extension_name, extension from user_stat_extensions where table_name='CUSTOMERS';

EXTENSION_NAME                 EXTENSION
------------------------------ --------------------------------------------------------------------------------
SYS_STU#S#WF25Z#QAHIHE#MOFFMM_ ("CUST_STATE_PROVINCE","COUNTRY_ID")

SQL>
--查看distinct数和柱状图使用情况
SQL> select e.extension col_group, t.num_distinct, t.histogram from user_stat_extensions e, user_tab_col_statistics t where e.extension_name = t.column_name and e.table_name = t.table_name and t.table_name = 'CUSTOMERS';

COL_GROUP                                                                        NUM_DISTINCT HISTOGRAM
-------------------------------------------------------------------------------- ------------ ---------------
("CUST_STATE_PROVINCE","COUNTRY_ID")                                                       FREQUENCY

SQL> 

实验：
1）当不使用多列统计信息时，真实结果是3341，执行计划是1132.

SQL> exec dbms_stats.drop_extended_stats('sh','customers','(cust_state_province,country_id)');

PL/SQL procedure successfully completed.

SQL> select count(*) from sh.customers where CUST_STATE_PROVINCE = 'CA' and country_id=;

  COUNT(*)
----------

Execution Plan
----------------------------------------------------------
Plan hash value: 

--------------------------------------------------------------------------------
| Id  | Operation          | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------
|    | SELECT STATEMENT   |           |      |     |      ()| :: |
|    |  SORT AGGREGATE    |           |      |     |            |          |
|*   |   TABLE ACCESS FULL| CUSTOMERS |   |  |      ()| :: |
--------------------------------------------------------------------------------

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

    - filter("CUST_STATE_PROVINCE"='CA' AND "COUNTRY_ID"=)

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

2）当使用多列统计信息时，真实结果是3341，执行计划是3437.

SQL> EXEC DBMS_STATS.GATHER_TABLE_STATS('SH','CUSTOMERS',METHOD_OPT =>'FOR ALL COLUMNS SIZE SKEWONLY FOR COLUMNS (CUST_STATE_PROVINCE,COUNTRY_ID) SIZE SKEWONLY');

PL/SQL procedure successfully completed.

SQL>  select count(*) from sh.customers where CUST_STATE_PROVINCE = 'CA' and country_id=;

  COUNT(*)
----------

Execution Plan
----------------------------------------------------------
Plan hash value: 

--------------------------------------------------------------------------------
| Id  | Operation          | Name      | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------
|    | SELECT STATEMENT   |           |      |     |      ()| :: |
|    |  SORT AGGREGATE    |           |      |     |            |          |
|*   |   TABLE ACCESS FULL| CUSTOMERS |   |  |      ()| :: |
--------------------------------------------------------------------------------

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

    - filter("CUST_STATE_PROVINCE"='CA' AND "COUNTRY_ID"=)

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

3）即以上情况，使用多列统计信息能让优化器得到更准确的判断！