oracle仅部分记录建立索引的方法

  今天研究了一下oracle对部分记录建立索引的方法。其实对部分记录建立的索引就是基于
函数的索引。由于部分记录函数化以后，返回了NULL值，而ORACLE索引不索引NULL值，
导致这些记录没有索引条目。
 
  这样我们就可以对自已关注的少数记录建立很小索引，提高查询速度。
一个例子，学校有10000学生，其中女同学仅有100人。我们经常关注女生，几乎不关注男生。
这样我们就可以只在女生记录上添加索引。这个很小的索引就能帮我们很快把女生找到。
做个测试：

SQL> create table students(id int , name varchar2(200),sex varchar(16))
    2    /
Table created

SQL> declare
    2     style varchar2(16);
    3     begin
    4     for i in 1..10000 loop
    5     if(i mod 100 = 0) then
    6         style:= 'F';
    7     else style := 'M';
    8     end if;
    9     insert into students values(i,rpad( 'student'||i,150, 'F'),style);
10     end loop;
11     commit;
12     end;
13    /
PL/SQL procedure successfully completed.

这里为了让女生分布均匀，加了i mod 100；

SQL> select count(1) from students where sex='F'
  2  /
 
  COUNT(1)
----------
       100
刚好100位女生.

SQL> select * from students where ( case when sex= 'F' then sex end)= 'F'
    2    /
已选择100行。
执行计划
----------------------------------------------------------
Plan hash value: 4078133427
------------------------------------------------------------------------------
| Id    | Operation                 | Name         | Rows    | Bytes | Cost (%CPU)| Time         |
------------------------------------------------------------------------------
|     0 | SELECT STATEMENT    |                    |        93 | 11625 |        56     (2)| 00:00:01 |
|*    1 |     TABLE ACCESS FULL| STUDENTS |        93 | 11625 |        56     (2)| 00:00:01 |
------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
     1 - filter( CASE "SEX" WHEN 'F' THEN "SEX" END = 'F')
Note
-----
     - dynamic sampling used for this statement

统计信息
----------------------------------------------------------
                169    recursive calls
                    0    db block gets
                337    consistent gets
                    0    physical reads
                    0    redo size
            17196    bytes sent via SQL*Net to client
                451    bytes received via SQL*Net from client
                    8    SQL*Net roundtrips to/ from client
                    4    sorts (memory)
                    0    sorts ( disk)
                100    rows processed

执行计划采用了全表扫描。
建立部分元组索引:

SQL> create index femaleIndex on students( case when sex= 'F' then sex end);
Index created
SQL> exec dbms_stats.gather_index_stats( 'study', 'femaleIndex')
    
PL/SQL procedure successfully completed
    
SQL> select num_rows,blevel from user_indexes where index_name= upper( 'femaleIndex')
    2    /
    
    NUM_ROWS         BLEVEL
---------- ----------
             100                    0

索引条目刚好100个，与女生数相等。

SQL> select * from students where ( case when sex= 'F' then sex end)= 'F'
    2    /
已选择100行。

执行计划
----------------------------------------------------------
Plan hash value: 2152294204
| Id    | Operation    | Name                | Rows    | Bytes | Cost (%CPU)| Time |
-------------------------------------------------------------------------------
|    0 | SELECT STATEMENT                        |                    | 93 | 11625 |    2     (0)|00:00:01 |
|    1 |     TABLE ACCESS BY INDEX ROWID| STUDENTS | 93 | 11625 |    2     (0)| 00:00:01 |
|* 2 |     INDEX RANGE SCAN                    | FEMALEINDEX| 38 |         |    1     (0)| 00:00:01 |
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
     2 - access( CASE "SEX" WHEN 'F' THEN "SEX" END = 'F')
Note
------ dynamic sampling used for this statement

统计信息
----------------------------------------------------------
                 11    recursive calls
                    0    db block gets
                174    consistent gets
                    0    physical reads
                    0    redo size
            17196    bytes sent via SQL*Net to client
                451    bytes received via SQL*Net from client
                    8    SQL*Net roundtrips to/ from client
                    0    sorts (memory)
                    0    sorts ( disk)
                100    rows processed

可以看到速度提高了很多。但是如果我们查询男生：

SQL> select * from students where ( case when sex= 'F' then sex end)= 'M'
    2    /
未选定行

执行计划
----------------------------------------------------------
Plan hash value: 2152294204
-------------------------------------------------------------------------------
| Id    | Operation | Name                | Rows    | Bytes | Cost (%CPU)|Time         |
--------------------------------------------------------------------------------
|     0 | SELECT STATEMENT                |                    |         3 | 375 | 2     (0)|00:00:01 |
|     1 |     TABLE ACCESS BY INDEX ROWID| STUDENTS|    3 | 375 | 2     (0)|00:00:01 |
|*    2 |     INDEX RANGE SCAN            | FEMALEINDEX |    38 |        |    1     (0)|00:00:01 |
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
     2 - access( CASE "SEX" WHEN 'F' THEN "SEX" END = 'M')
Note
-----
     - dynamic sampling used for this statement

统计信息
----------------------------------------------------------
                 11    recursive calls
                    0    db block gets
                 67    consistent gets
                    0    physical reads
                    0    redo size
                375    bytes sent via SQL*Net to client
                374    bytes received via SQL*Net from client
                    1    SQL*Net roundtrips to/ from client
                    0    sorts (memory)
                    0    sorts ( disk)
                    0    rows processed

居然没有记录返回!! 当然了因为(case when sex='F' then sex end)这个函数，我们姑且把它看成一个函数，是不会返回'M'这个值的。这也说明了如果建立了函数索引，查询条件上使用了该函数，则查询必走此索引。

以上是个简单的例子，假如我们一个项目管理系统，项目只有三种状态：投标，开发，验收。历经很多年,验收的验目很多，但正在投标，开发的项目肯定不多(IBM可能有很多)，这样我们就可以只在需要关注的投标与开发的记录上添加索引:
create index test on projects(case when status='投标' then  status when status='开发' then  status end).

部分记录建立的索引还有一个经典用法就是建立唯一索引，完成对记录的约束，比如上面的例子，如果建成唯一索引，那么全校只能有一个女生了，而男生无限制。我不喜欢这样，所以就不演示了，记得加个unique就成。