在NULL值与索引(一)中讲述了null值与索引的一些基本情况。其主要的内容为,基于允许存在null值的索引列,其索引值不会被存储;其次 是由于这个特性导致了我们在使用is null时索引失效的情形;最后则是描述的通过为null值列添加not null约束来使得is null走索引。尽管我 们可以通过添加not null来解决is null走索引,当现实中的情况是仍然很多列根本是无法确定的,而必须保持其null特性。对于此种情形该如 何解决呢? 一、通过基于函数的索引来使得is null使用索引
-->演示环境
scott@ORCL> select * from v$version where rownum<2;
BANNER
----------------------------------------------------------------
Oracle Database 10g Enterprise Edition Release 10.2.0.1.0 - Prod
-->创建测试表t2
scott@ORCL> create table t2(obj_id,obj_name) as select object_id,object_name from dba_objects;
Table created.
-->演示表t2上不存在not null约束
scott@ORCL> desc t2
Name Null? Type
----------------------------- -------- --------------------
OBJ_ID NUMBER
OBJ_NAME VARCHAR2(128)
-->为表t2创建一个普通的B树索引
scott@ORCL> create index i_t2_obj_id on t2(obj_id);
Index created.
-->将表t2列obj_id<=100的obj_id置空
-->注:在Oracle 10g中空字符串等同于null值
scott@ORCL> update t2 set obj_id='' where obj_id<=100;
99 rows updated.
-->下面的查询亦表明在此时空字符串等同于null值
scott@ORCL> set null unknown
scott@ORCL> select * from t2 where obj_id is null and rownum<3;
OBJ_ID OBJ_NAME
---------- ------------------------------
unknown ICOL$
unknown I_USER1
-->收集统计信息
scott@ORCL> exec dbms_stats.gather_table_stats('SCOTT','T2',cascade=>true);
PL/SQL procedure successfully completed.
-->基于null值上使用not null会使用索引扫描,等同于前面 null值与索引(一) 中的描述
scott@ORCL> select count(*) from t2 where obj_id is not null;
Execution Plan
----------------------------------------------------------
Plan hash value: 3840858596
-------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 5 | 7 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 5 | | |
|* 2 | INDEX FAST FULL SCAN| I_T2_OBJ_ID | 11719 | 58595 | 7 (0)| 00:00:01 |
-------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - filter("OBJ_ID" IS NOT NULL)
-->列obj_id is null走全表扫描
scott@ORCL> select count(*) from t2 where obj_id is null;
Execution Plan
----------------------------------------------------------
Plan hash value: 3321871023
---------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 5 | 13 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 5 | | |
|* 2 | TABLE ACCESS FULL| T2 | 1 | 5 | 13 (0)| 00:00:01 |
---------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - filter("OBJ_ID" IS NULL)
-->创建基于函数的索引来使得is null走索引
-->下面使用了nvl函数来创建函数索引,即当obj_id为null值时,存储-1
scott@ORCL> create index i_fn_t2_obj_id on t2(nvl(obj_id,-1));
Index created.
-->收集索引信息
scott@ORCL> exec dbms_stats.gather_index_stats('SCOTT','I_FN_T2_OBJ_ID');
PL/SQL procedure successfully completed.
-->可以看到下面的执行计划中刚刚创建的函数索引已经生效I_FN_T2_OBJ_ID
scott@ORCL> select count(*) from t2 where nvl(obj_id,-1) = -1;
Execution Plan
----------------------------------------------------------
Plan hash value: 3983750858
------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 5 | 1 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 5 | | |
|* 2 | INDEX RANGE SCAN| I_FN_T2_OBJ_ID | 100 | 500 | 1 (0)| 00:00:01 |
------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access(NVL("OBJ_ID",(-1))=(-1))
二、使用伪列创建基于函数的索引来使得is null使用索引
-->下面通过添加一个值为-1(可取任意值)的伪列来创建索引
scott@ORCL> create index i_new_t2_obj_id on t2(obj_id,-1);
Index created.
-->收集索引信息
scott@ORCL> exec dbms_stats.gather_index_stats('SCOTT','I_NEW_T2_OBJ_ID');
PL/SQL procedure successfully completed.
-->从下面的查询可以看出obj_id is null使用了刚刚创建的索引
scott@ORCL> select count(*) from t2 where obj_id is null;
Execution Plan
----------------------------------------------------------
Plan hash value: 801885198
-------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 5 | 2 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 5 | | |
|* 2 | INDEX RANGE SCAN| I_NEW_T2_OBJ_ID | 99 | 495 | 2 (0)| 00:00:01 |
-------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJ_ID" IS NULL)
-->查看刚刚创建的所有索引的相关统计信息
scott@ORCL> select index_name,index_type,blevel,leaf_blocks,num_rows,status,distinct_keys
2 from user_indexes where table_name='T2';
INDEX_NAME INDEX_TYPE BLEVEL LEAF_BLOCKS NUM_ROWS STATUS DISTINCT_KEYS
--------------- ------------------------------ ---------- ----------- ---------- -------- -------------
I_FN_T2_OBJ_ID FUNCTION-BASED NORMAL 1 26 11719 VALID 11621
I_NEW_T2_OBJ_ID FUNCTION-BASED NORMAL 1 32 11719 VALID 11621
I_T2_OBJ_ID NORMAL 1 25 11620 VALID 11620
-->从上面的结果可知:
-->普通的B索引(I_T2_OBJ_ID)使用的索引块最小,因为null值没有被存储,NUM_ROWS与DISTINCT_KEYS即是佐证
-->使用NVL函数创建的索引I_FN_T2_OBJ_ID中如实的反应了null值,即11620 + null值 = 11621
-->使用伪列创建的索引依然属于函数索引,其耗用的叶节点块数最多,因为多出了一个值(-1)来存储
-->尽管使用NVL创建的函数占用的磁盘空间小于使用伪列创建的索引,当在书写谓词时需要带上NVL函数,而伪列索引中谓词直接使用is null。
三、NULL值与索引衍生特性
-->由前面的种种事例再次说明NULL值不会被存储到索引中,因此基于这个特性可以使用decode函数来压缩索引列。
-->在实际应用的多数情形中,如表上有打印状态列is_printed通常为两种情形,已打印或未打印,假定1表示已打印,而0表示未打印。
-->通常情况下90%以上的单据都处于已打印状态,而仅有10%左右的处于未打印。而经常要使用的情形是查询未打印的单据并重新打印。
-->基于上述情况,可以使用位图索引来解决,但此处我们讨论的是B树索引,故不考虑该情形(或者说你使用了非企业版Oracle,不支持位图索引)
-->此处对于这类情形我们可以使用decode函数来解决这个问题
-->更新表上的列,使之obj_id为1的行占绝大多数
scott@ORCL> update t2 set obj_id=1 where obj_id is not null;
11620 rows updated.
-->更新表,使之obj_id为0的行占少部分
scott@ORCL> update t2 set obj_id = 0 where obj_id is null;
99 rows updated.
scott@ORCL> commit;
-->收集统计信息
scott@ORCL> exec dbms_stats.gather_table_stats('SCOTT','T2',cascade=>true);
PL/SQL procedure successfully completed.
-->表t2上obj_id列的最终分布
scott@ORCL> select obj_id,count(*) from t2 group by obj_id;
OBJ_ID COUNT(*)
---------- ----------
1 11620
0 99
-->使用decode函数创建索引
-->注意此处decode的使用,当obj_id非0值时,其值被赋予为null值,由于该null值不会存储到索引,因此大部分obj_id列值为1的不会被索引
scott@ORCL> create index i_fn2_t2_obj_id on t2(decode(obj_id,0,0,null));
Index created.
-->收集索引上的统计信息
scott@ORCL> exec dbms_stats.gather_index_stats('SCOTT','I_FN2_T2_OBJ_ID');
PL/SQL procedure successfully completed.
-->查看新索引的执行计划
scott@ORCL> set autot trace exp;
scott@ORCL> select count(*) from t2 where decode(obj_id,0,0,null) = 0;
Execution Plan
----------------------------------------------------------
Plan hash value: 1461308992
-------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 3 | 1 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 3 | | |
|* 2 | INDEX RANGE SCAN| I_FN2_T2_OBJ_ID | 98 | 294 | 1 (0)| 00:00:01 |
-------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access(DECODE("OBJ_ID",0,0,NULL)=0)
-->当直接使用obj_id = 0来查询时使用的是普通的B树索引
scott@ORCL> select count(*) from t2 where obj_id = 0;
Execution Plan
----------------------------------------------------------
Plan hash value: 1804118247
---------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
---------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 3 | 1 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 3 | | |
|* 2 | INDEX RANGE SCAN| I_T2_OBJ_ID | 99 | 297 | 1 (0)| 00:00:01 |
---------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("OBJ_ID"=0)
-->当使用obj_id = 1来查询时走全表扫描,因为obj_id = 1占据表90%以上,由CBO特性决定了走全表扫描
scott@ORCL> select * from t2 where obj_id = 1;
Execution Plan
----------------------------------------------------------
Plan hash value: 1513984157
--------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
--------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 11620 | 249K| 14 (8)| 00:00:01 |
|* 1 | TABLE ACCESS FULL| T2 | 11620 | 249K| 14 (8)| 00:00:01 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("OBJ_ID"=1)
-->表t2上所有索引的统计信息
scott@ORCL> select index_name,index_type,blevel,leaf_blocks,num_rows,status,distinct_keys
2 from user_indexes where table_name='T2';
INDEX_NAME INDEX_TYPE BLEVEL LEAF_BLOCKS NUM_ROWS STATUS DISTINCT_KEYS
--------------- ------------------------------ ---------- ----------- ---------- -------- -------------
I_FN_T2_OBJ_ID FUNCTION-BASED NORMAL 1 40 11719 VALID 2
I_NEW_T2_OBJ_ID FUNCTION-BASED NORMAL 1 52 11719 VALID 2
I_FN2_T2_OBJ_ID FUNCTION-BASED NORMAL 0 1 99 VALID 1
I_T2_OBJ_ID NORMAL 1 40 11719 VALID 2
-->从上面的结果可知,索引I_FN2_T2_OBJ_ID仅仅存储了99跳记录,且DISTINCT_KEYS值为1个,因为所有非0值的全部被置NULL。
-->以上方法实现了索引压缩,避免了较大索引维护所需的开销,同时也提高了查询性能。
-->Author : Robinson Cheng
-->Blog : http://blog.csdn.net/robinson_0612
四、总结 1、对于用于连接或经常被谓词使用到的列应尽可能避免NULL值属性,因为它容易导致索引失效。 2、为需要使用NULL值的列添加缺省值(alter table tb modify(col default 'Y'))。 3、如果NULL值不可避免也不能使用缺省值,应考虑为该常用列使用nvl函数创建索引,或使用伪列来创建索引以提高查询性能。 4、对于复合索引应保证索引中至少有一列不为NULL值,还是因为全部列为NULL时不被索引存储,以保证使用is null是可以使用索引。 5、对于复合索引应保证索引列应使用数据类型长度最小的列来添加not null约束应节省磁盘空间。
扫码关注腾讯云开发者
领取腾讯云代金券
Copyright © 2013 - 2025 Tencent Cloud. All Rights Reserved. 腾讯云 版权所有
深圳市腾讯计算机系统有限公司 ICP备案/许可证号:粤B2-20090059 深公网安备号 44030502008569
腾讯云计算(北京)有限责任公司 京ICP证150476号 | 京ICP备11018762号 | 京公网安备号11010802020287
Copyright © 2013 - 2025 Tencent Cloud.
All Rights Reserved. 腾讯云 版权所有