32个字节限制Oracle直方图优化

Oracle直方图的限制往往让不少初学者头痛mdash;mdash;超过32字节不显示，以及前32字节相同产生错误。由于OBJECT_TYPE列上的DIS

Oracle直方图的限制往往让不少初学者头痛——超过32字节不显示，以及前32字节相同产生错误。

由于OBJECT_TYPE列上的DISTINCT值的个数小于254,ORACLE将会在此列上建立频率直方图，优化器将会准确的估算出CONTENTS='TABLE'的查询返回37条记录。

下面看看如下的情况：

SQL> TRUNCATE TABLE T;

Table truncated.

SQL> SET AUTOT OFF
SQL> INSERT INTO T
2 SELECT ROWNUM,OBJECT_NAME,RPAD('*',32,'*')||OBJECT_TYPE
3 FROM ALL_OBJECTS WHERE ROWNUM<=10000;

10000 rows created.

SQL> COMMIT;

Commit complete.

SQL> SELECT COUNT(1) FROM T;

COUNT(1)
----------
10000

SQL> SELECT CONTENTS ,COUNT(1) FROM T GROUP BY CONTENTS ;

CONTENTS COUNT(1)
-------------------------------------------------- ----------
********************************SEQUENCE 1
********************************LIBRARY 3
********************************WINDOW GROUP 1
********************************INDEX PARTITION 347
********************************PACKAGE 164
********************************SCHEDULE 1
********************************TABLE PARTITION 25
********************************VIEW 1150
********************************TABLE 37
********************************PROCEDURE 11
********************************CONSUMER GROUP 2
********************************INDEX SUBPARTITION 3328
********************************OPERATOR 15
********************************WINDOW 2
********************************INDEX 34
********************************FUNCTION 60
********************************SYNONYM 2552
********************************TABLE SUBPARTITION 1714
********************************TYPE 538
********************************JOB CLASS 1
********************************PACKAGE BODY 13
********************************EVALUATION CONTEXT 1

22 rows selected.

SQL> EXEC DBMS_STATS.GATHER_TABLE_STATS(USER,'T',method_opt=>'FOR COLUMNS CONTENTS SIZE 254');

PL/SQL procedure successfully completed.

SQL> SELECT * FROM T WHERE CONTENTS='********************************TABLE';

Execution Plan
----------------------------------------------------------
Plan hash value: 2153619298

--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 10000 | 654K| 24 (0)| 00:00:01 |
|* 1 | TABLE ACCESS FULL| T | 10000 | 654K| 24 (0)| 00:00:01 |
--------------------------------------------------------------------------

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

1 - filter("CONTENTS"='********************************TABLE')

由于CONTENTS列的前32位都一样，ORACLE在收集统计信息直方图的时候只考虑前32位，这会导致ORACLE认为
所有的记录的列CONTENTS都是相同的，因此优化器估算返回的行数为10000。

SQL> UPDATE T SET CONTENTS=SUBSTR(CONTENTS,2);

10000 rows updated.

SQL> COMMIT;

Commit complete.

SQL> EXEC DBMS_STATS.GATHER_TABLE_STATS(USER,'T',method_opt=>'FOR COLUMNS CONTENTS SIZE 254');

PL/SQL procedure successfully completed.

SQL> SELECT CONTENTS ,COUNT(1) FROM T GROUP BY CONTENTS ORDER BY 1;

CONTENTS COUNT(1)
-------------------------------------------------- ----------
*******************************CONSUMER GROUP 2
*******************************EVALUATION CONTEXT 1
*******************************FUNCTION 60
*******************************INDEX 34
*******************************INDEX PARTITION 347
*******************************INDEX SUBPARTITION 3328
*******************************JOB CLASS 1
*******************************LIBRARY 3
*******************************OPERATOR 15
*******************************PACKAGE 164
*******************************PACKAGE BODY 13
*******************************PROCEDURE 11
*******************************SCHEDULE 1
*******************************SEQUENCE 1
*******************************SYNONYM 2552
*******************************TABLE 37
*******************************TABLE PARTITION 25
*******************************TABLE SUBPARTITION 1714
*******************************TYPE 538
*******************************VIEW 1150
*******************************WINDOW 2
*******************************WINDOW GROUP 1

SQL> SELECT SUBSTR(CONTENTS,1,32),COUNT(1) FROM T GROUP BY SUBSTR(CONTENTS,1,32);

SUBSTR(CONTENTS,1,32) COUNT(1)
---------------------------------------- ----------
*******************************E 1
*******************************J 1
*******************************P 188
*******************************C 2
*******************************S 2554
*******************************T 2314
*******************************F 60
*******************************O 15
*******************************L 3
*******************************W 3
*******************************I 3709
*******************************V 1150

12 rows selected.

SQL> SET AUTOT TRACEONLY EXP
SQL> SELECT * FROM T WHERE CONTENTS='*******************************TABLE';

Execution Plan
----------------------------------------------------------
Plan hash value: 2153619298

--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 2314 | 149K| 24 (0)| 00:00:01 |
|* 1 | TABLE ACCESS FULL| T | 2314 | 149K| 24 (0)| 00:00:01 |
--------------------------------------------------------------------------

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

1 - filter("CONTENTS"='*******************************TABLE')

SQL> SELECT * FROM T WHERE CONTENTS='*******************************TYPE';

Execution Plan
----------------------------------------------------------
Plan hash value: 2153619298

--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 2314 | 149K| 24 (0)| 00:00:01 |
|* 1 | TABLE ACCESS FULL| T | 2314 | 149K| 24 (0)| 00:00:01 |
--------------------------------------------------------------------------

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

1 - filter("CONTENTS"='*******************************TYPE')

从上面可以看出,ORACLE只考虑前32个字节，下面每一个值查询的估算行数都是等于3者实际记录数之和。

*******************************TABLE 37
*******************************TABLE PARTITION 25
*******************************TABLE SUBPARTITION 1714
*******************************TYPE 538

更多Oracle相关信息见Oracle 专题页面 ?tid=12