ROWNUM是一种伪列，它会根据返回记录生成一个序列化的数字。利用ROWNUM，我们可以生产一些原先难以实现的结果输出，但因为它是伪列的这个特殊性，我们在使用时也需要注意一些事项，不要掉入“陷阱”。下面就介绍一下它的使用技巧及注意事项。

1 特殊结果输出

利用ROWNUM，我们可以做到一些特殊方式的输出。

1.1 Top N结果输出

我们如果希望取输出结果的前面几条数据，通过ROWNUM可以轻松实现：

sql> select * from t_test4

  2  where rownum <= 5;

USERNAME                          USER_ID CREATED

------------------------------ ---------- ---------

WOW                                    71 26-APR-07

CS2                                    70 15-JAN-07

3                                      69 01-NOV-06

DMP                                    68 12-OCT-06

PROFILER                               67 05-SEP-06

但是，如果你希望对一个排序结果取Top N数据的话，使用ROWNUM存在一些“陷阱”，我们后面部分会介绍这些“陷阱”并且说明如何避免。

1.2 分页查询

利用ROWNUM对结果进行分页，下面返回结果中的第6到第10条记录：

sql> select * from

  2  (

  3  select a.*, rownum as rn from css_bl_view a

  4  where capture_phone_num = '(1) 925-4604800'

  5  ) b

  6  where b.rn between 6 and 10;

6 rows selected.

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=2770 Card=2183 Bytes =7166789)

   1    0   VIEW (Cost=2770 Card=2183 Bytes=7166789)

   2    1     COUNT

   3    2       table ACCESS (FULL) OF 'CSS_BL_VIEW' (Cost=2770 Card=2183 Bytes=1305434)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

      29346  consistent gets

      29190  physical reads

          0  redo size

       7328  bytes sent via sql*Net to client

        234  bytes received via sql*Net from client

          4  sql*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

          5  rows processed

另外一种实现方式：

sql> select * from css_bl_view a

  2  where capture_phone_num = '(1) 925-4604800'

  3  and rownum <= 10

  4  minus

  5  select * from css_bl_view a

  6  where capture_phone_num = '(1) 925-4604800'

  7  and rownum <= 5

  8  ;

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=5920 Card=10 Bytes=8970)

   1    0   MINUS

   2    1     SORT (UNIQUE) (Cost=2960 Card=10 Bytes=5980)

   3    2       COUNT (STOPKEY)

   4    3         TABLE ACCESS (FULL) OF 'CSS_BL_VIEW' (Cost=2770 Card=2183 Bytes=1305434)

   5    1     SORT (UNIQUE) (Cost=2960 Card=5 Bytes=2990)

   6    5       COUNT (STOPKEY)

   7    6         table ACCESS (FULL) OF 'CSS_BL_VIEW' (Cost=2770 Card=2183 Bytes=1305434)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

         62  consistent gets

         50  physical reads

          0  redo size

       7232  bytes sent via sql*Net to client

        234  bytes received via sql*Net from client

          4  sql*Net roundtrips to/from client

          2  sorts (memory)

          0  sorts (disk)

          5  rows processed

第三种实现方式：

sql> select * from

  2  (

  3  select a.*, rownum as rn from css_bl_view a

  4  where capture_phone_num = '(1) 925-4604800'

  5  and rownum <= 10

  6  ) b

  7  where b.rn > 5;

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=2770 Card=10 Bytes=32830)

   1    0   VIEW (Cost=2770 Card=10 Bytes=32830)

   2    1     COUNT (STOPKEY)

   3    2       table ACCESS (FULL) OF 'CSS_BL_VIEW' (Cost=2770 Card=2183 Bytes=1305434)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

         35  consistent gets

         30  physical reads

          0  redo size

       7271  bytes sent via sql*Net to client

        234  bytes received via sql*Net from client

          4  sql*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

          5  rows processed

这里特地将三种实现方式的查询计划及统计数据打印出来，大家可以比较一下3中方式的性能。

1.3 利用ROWNUM做分组子排序

对于以下表T_TEST4的内容：

OWNER                                   NAME

------------------------------------------------------

STRMADMIN                               STREAMS_QUEUE

APARKMAN                                JOB_QUEUE

SYS                                     AQ$_AQ_SRVNTFN_TABLE_E

SYS                                     AQ$_KUPC$DATAPUMP_QUETAB_E

APARKMAN                                AQ$_JMS_TEXT_E

STRMADMIN                               AQ$_STREAMS_QUEUE_TABLE_E

SYS                                     AQ$_SCHEDULER$_EVENT_QTAB_E

…

如果我们希望结果按照OWNER进行分组后，再对每组中成员进行编号，结果类似如下：

OWNER                                   NO NAME

------------------------------------------------------

APARKMAN                                1 JOB_QUEUE

                                        2 AQ$_JMS_TEXT_E

STRMADMIN                               1 STREAMS_QUEUE

                                        2 AQ$_STREAMS_QUEUE_TABLE_E

SYS                                     1 AQ$_AQ_SRVNTFN_TABLE_E

                                        2 AQ$_KUPC$DATAPUMP_QUETAB_E

                                        3 AQ$_SCHEDULER$_EVENT_QTAB_E

…

在没有ROWNUM时要实现这样的功能会很复杂，但通过ROWNUM我们可以轻松实现：

sql> SELECT DECODE(ROWNUM-min_sno,0,a.owner,NULL) owner,DECODE(ROWNUM-min_sno,0,1,rownum+1-min_sno) sno, a.name

  2  FROM (SELECT *

  3        FROM t_test8

  4        ORDER BY owner, name ) a,

  5       (SELECT owner, MIN(rownum) min_sno

  6        FROM( SELECT *

  7              FROM t_test8

  8              ORDER BY owner, name)

  9        GROUP BY owner) b

 10  WHERE a.owner=b.owner;

OWNER                                 SNO NAME

------------------------------ ---------- ------------------------------

APARKMAN                                1 JOB_QUEUE

                                        2 AQ$_JMS_TEXT_E

STRMADMIN                               1 STREAMS_QUEUE

                                        2 AQ$_STREAMS_QUEUE_TABLE_E

SYS                                     1 AQ$_AQ_SRVNTFN_TABLE_E

                                        2 AQ$_KUPC$DATAPUMP_QUETAB_E

                                        3 AQ$_SCHEDULER$_EVENT_QTAB_E

                                        4 AQ$_SCHEDULER$_JOBQTAB_E

                                        5 AQ$_STREAMS_QUEUE_TABLE_E

                                        6 AQ$_SYS$SERVICE_METRICS_TAB_E

                                        7 AQ$_AQ_EVENT_TABLE_E

                                        8 AQ$_AQ$_MEM_MC_E

                                        9 AQ$_ALERT_QT_E

                                       10 ALERT_QUE

                                       11 AQ_EVENT_TABLE_Q

                                       12 SYS$SERVICE_METRICS

                                       13 STREAMS_QUEUE

                                       14 SRVQUEUE

                                       15 SCHEDULER$_JOBQ

                                       16 SCHEDULER$_EVENT_QUEUE

                                       17 AQ_SRVNTFN_TABLE_Q

SYSMAN                                  1 AQ$_MGMT_NOTIFY_QTABLE_E

                                        2 MGMT_NOTIFY_Q

system                                  1 DEF$_AQERROR

                                        2 DEF$_AQCALL

                                        3 AQ$_DEF$_AQERROR_E

                                        4 AQ$_DEF$_AQCALL_E

WMSYS                                   1 AQ$_WM$EVENT_QUEUE_TABLE_E

                                        2 WM$EVENT_QUEUE

29 rows selected.

2 性能

我们很多程序员在确认某个表中是否有相应数据时，喜欢加上ROWNUM=1，其思路就是只要存在一条数据就说明有相应数据，查询就可以直接返回了，这样就能提高性能了。但是在10G之前，使用ROWNUM=1是不能达到预期的性能效果的，而是需要通过<2或<=1作为过滤条件才能达到预期效果，看以下查询计划：

sql> select * from t_test1

  2  where object_id <100

  3  and rownum = 1;

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=37 Card=1 Bytes=86)

   1    0   COUNT (STOPKEY)

   2    1     table ACCESS (BY INDEX ROWID) OF 'T_TEST1' (Cost=37 Card=89 Bytes=7654)

   3    2       INDEX (RANGE SCAN) OF 'T_TEST1_PK' (UNIQUE) (Cost=2 Card=89)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

         62  consistent gets

          0  physical reads

          0  redo size

        654  bytes sent via sql*Net to client

        234  bytes received via sql*Net from client

          4  sql*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

          1  rows processed

sql> select * from t_test1

  2  where object_id <100

  3  and rownum <= 1;

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=37 Card=1 Bytes=86)

   1    0   COUNT (STOPKEY)

   2    1     table ACCESS (BY INDEX ROWID) OF 'T_TEST1' (Cost=37 Card=89 Bytes=7654)

   3    2       INDEX (RANGE SCAN) OF 'T_TEST1_PK' (UNIQUE) (Cost=2 Card=89)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

          3  consistent gets

          0  physical reads

          0  redo size

        654  bytes sent via sql*Net to client

        234  bytes received via sql*Net from client

          4  sql*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

          1  rows processed

sql> /

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=37 Card=1 Bytes=86)

   1    0   COUNT (STOPKEY)

   2    1     table ACCESS (BY INDEX ROWID) OF 'T_TEST1' (Cost=37 Card=89 Bytes=7654)

   3    2       INDEX (RANGE SCAN) OF 'T_TEST1_PK' (UNIQUE) (Cost=2 Card=89)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

          3  consistent gets

          0  physical reads

          0  redo size

        654  bytes sent via sql*Net to client

        234  bytes received via sql*Net from client

          4  sql*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

          1  rows processed

10G以后，这个问题就被修正了：

sql> select * from t_test1

  2  where rownum = 1;

Execution Plan

----------------------------------------------------------

Plan hash value: 536364188

------------------------------------------------------------------------------

| Id  | Operation          | Name    | Rows  | Bytes | Cost (%CPU)| Time     |

------------------------------------------------------------------------------

|   0 | SELECT STATEMENT   |         |     1 |    86 |     2   (0)| 00:00:01 |

|*  1 |  COUNT STOPKEY     |         |       |       |            |          |

|   2 |   table ACCESS FULL| T_TEST1 |     1 |    86 |     2   (0)| 00:00:01 |

------------------------------------------------------------------------------

Predicate Information (identified by operation id):

---------------------------------------------------

   1 - filter(ROWNUM=1)

Statistics

----------------------------------------------------------

          1  recursive calls

          0  db block gets

          4  consistent gets

          1  physical reads

          0  redo size

       1201  bytes sent via sql*Net to client

        385  bytes received via sql*Net from client

          2  sql*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

          1  rows processed

sql> select * from t_test1

  2  where rownum <= 1;

Execution Plan

----------------------------------------------------------

Plan hash value: 536364188

------------------------------------------------------------------------------

| Id  | Operation          | Name    | Rows  | Bytes | Cost (%CPU)| Time     |

------------------------------------------------------------------------------

|   0 | SELECT STATEMENT   |         |     1 |    86 |     2   (0)| 00:00:01 |

|*  1 |  COUNT STOPKEY     |         |       |       |            |          |

|   2 |   table ACCESS FULL| T_TEST1 |     1 |    86 |     2   (0)| 00:00:01 |

------------------------------------------------------------------------------

Predicate Information (identified by operation id):

---------------------------------------------------

   1 - filter(ROWNUM<=1)

Statistics

----------------------------------------------------------

          0  recursive calls

          0  db block gets

          4  consistent gets

          0  physical reads

          0  redo size

       1201  bytes sent via sql*Net to client

        385  bytes received via sql*Net from client

          2  sql*Net roundtrips to/from client

          0  sorts (memory)

          0  sorts (disk)

          1  rows processed

3 ROWNUM的使用“陷阱”

由于ROWNUM是一个伪列，只有有结果记录时，ROWNUM才有相应数据，因此对它的使用不能向普通列那样使用，否则就会陷入一些“陷阱”当中。

3.1 对ROWNUM进行>、>=、=操作

不能对ROWNUM使用>（大于1的数值）、>=（大于或等于1的数值）、=（大于或等于1的数值），否则无结果

sql> select count(*) from css_bl_view a where rownum>0;

  COUNT(*)

----------

sql> select count(*) from css_bl_view a

  2  where rownum > 1;

  COUNT(*)

----------

这是因为：

1、ROWNUM是伪列，必须要要有返回结果后，每条返回记录就会对应产生一个ROWNUM数值；

2、返回结果记录的ROWNUM是从1开始排序的，因此第一条始终是1;

这样，当查询到第一条记录时，该记录的ROWNUM为1，但条件要求ROWNUM>1，因此不符合，继续查询下一条；因为前面没有符合要求的记录，因此下一条记录过来后，其ROWNUM还是为1，如此循环，就不会产生结果。上述查询可以通过子查询来替代：

sql> select count(*)

  2  from

  3  (select BL_REF_CDE, rownum rn from css_bl_view)

  4  where rn > 1;

  COUNT(*)

----------

我们可以通过以下方式来实现对ROWNUM的>、=的查询：

查询ROWNUM=5的数据：

sql> select object_id,object_name

  2  from (select object_id,object_name, rownum as rn from t_test1)

  3  where rn = 5;

 OBJECT_ID OBJECT_NAME

---------- ------------------------------

        29 C_COBJ#

查询ROWNUM > 25的数据：

sql> select * from t_test4

  2  minus

  3  select * from t_test4

  4  where rownum <= 25;

USERNAME                          USER_ID CREATED

------------------------------ ---------- ---------

DIP                                    19 21-NOV-05

OUTLN                                  11 21-NOV-05

PUBLIC                              99999 18-JUL-07

SYS                                     0 21-NOV-05

SYSMAN                                 32 21-NOV-05

system                                  5 21-NOV-05

6 rows selected.

3.2 ROWNUM和Order BY

要注意的是：在使用ROWNUM时，只有当Order By的字段是主键时，查询结果才会先排序再计算ROWNUM，下面OBJECT_ID是表T_TEST1的主键字段：

sql> select object_id,object_name from t_test1

  2  where rownum <= 5

  3  order by object_id;

 OBJECT_ID OBJECT_NAME

---------- ------------------------------

         2 C_OBJ#

         3 I_OBJ#

         4 TAB$

         5 CLU$

         6 C_TS#

但是，对非主键字段OBJECT_NAME进行排序时，结果就混乱了：

sql> select object_id,object_name from t_test1

  2  where rownum <= 5

  3  order by object_name;

 OBJECT_ID OBJECT_NAME

---------- ------------------------------

        28 CON$

        29 C_COBJ#

        20 ICOL$

        44 I_USER1

        15 UNDO$

sql> select count(*) from t_test1

  2  where object_name < 'CON$';

  COUNT(*)

----------

出现这种混乱的原因是：Oracle先按物理存储位置（rowid）顺序取出满足rownum条件的记录，即物理位置上的前5条数据，然后在对这些数据按照Order By的字段进行排序，而不是我们所期望的先排序、再取特定记录数。

如果需要对非主键字段排序再去前n条数据，我们可以以以下方式实现：

sql> select object_id,object_name

  2  from (select object_id,object_name from t_test1

  3        order by object_name)

  4  where rownum <= 5;

 OBJECT_ID OBJECT_NAME

---------- ------------------------------

     35489 /1000e8d1_LinkedHashMapValueIt

     35490 /1000e8d1_LinkedHashMapValueIt

     21801 /1005bd30_LnkdConstant

     21802 /1005bd30_LnkdConstant

     17205 /10076b23_OraCustomDatumClosur

3.3 排序分页

当对存在重复值的字段排序后再分页输出，我们很容易会陷入到另外一个“陷阱”。

请看以下例子，我们希望对T_TEST1的OWNER字段排序后，以每页输出10个结果的方式分页输出：

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name from t_test1 order by owner) a

  4  where rownum <= 10)

  5  where rn >= 1;

OWNER                          OBJECT_NAME

------------------------------ ------------------------------

AFWOWNER                       AFWADAPTER

AFWOWNER                       AFWADAPTERCONFIGURATION

AFWOWNER                       AFWADAPTERCONFIGURATION_IDX1

AFWOWNER                       AFWADAPTERFQN_PK

AFWOWNER                       AFWADAPTERCONFIGURATION_PK

AFWOWNER                       AFWADAPTERCONFIGURATION_IDX2

AFWOWNER                       AFWSERVERCODE_PK

AFWOWNER                       AFWSERVER

AFWOWNER                       AFWADAPTERLOOKUP_IDX1

AFWOWNER                       AFWADAPTERLOOKUP

10 rows selected.

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name from t_test1 order by owner) a

  4  where rownum <= 20)

  5  where rn >= 11;

OWNER                          OBJECT_NAME

------------------------------ ------------------------------

AFWOWNER                       AFWTOKENSTATUSCODE_PK

AFWOWNER                       AFWTOKENSTATUS

AFWOWNER                       AFWTOKENADMIN_IDX1

AFWOWNER                       AFWTOKENADMINCODE_PK

AFWOWNER                       AFWTOKENADMIN

AFWOWNER                       AFWTOKEN

AFWOWNER                       AFWSERVERCONFIGURATION_PK

AFWOWNER                       AFWSERVERCONFIGURATION

AFWOWNER                       AFWSERVER

AFWOWNER                       AFWADAPTERLOOKUP

10 rows selected.

仔细比较结果，你会发现“AFWSERVER”、“AFWADAPTERLOOKUP”在两次分页结果中都出现了。但是OBJECT_NAME在每个OWNER中的值是唯一的，说明这个输出结果是错误的，我们又陷入了一个“陷阱”。这是怎么回事呢，请先看下上述语句的查询计划：

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name from t_test1 order by owner) a

  4  where rownum <= 20)

  5  where rn >= 11;

10 rows selected.

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=205 Card=20 Bytes=94

0)

   1    0   VIEW (Cost=205 Card=20 Bytes=940)

   2    1     COUNT (STOPKEY)

   3    2       VIEW (Cost=205 Card=30670 Bytes=1042780)

   4    3         SORT (ORDER BY STOPKEY) (Cost=205 Card=30670 Bytes=858760)

   5    4           table ACCESS (FULL) OF 'T_TEST1' (Cost=42 Card=30670 Bytes=858760)

看下这个“陷阱”是怎么形成的。从查询计划上，我们可以注意到，对于子查询

select a.*, rownum as rn from

     (select owner, object_name from t_test1 order by owner) a

where rownum <= 20

优化器采用了“SORT (ORDER BY STOPKEY)”。

“SORT (ORDER BY STOPKEY)”不需要对所有数据进行排序，而是只要找出结果集中的按特定顺序的最前N条记录，一旦找出了这N条记录，就无需再对剩下的数据进行排序，而直接返回结果。这种算法我们可以视为是“快速排序”算法的变种。快速排序算法的基本思想是：先将数据分2组集合，保证第一集合中的每个数据都大于第二个集合中每个数据，然后再按这个原则对每个集合进行递归分组，直到集合的单位最小。在进行“SORT (ORDER BY STOPKEY)”时，首先找出N条数据（这些数据并没有做排序）放在第一组，保证第一组的数据都大于第二组的数据，然后只对第一组数据进行递归。

可以看到，基于这样的算法基础上，如果N的数值不同，数据的分组也不同（如N=20时，第一次分组比例为12:8，然后继续递归；当N=10时，第一次分组比例为3:7 … …），这样，在数据的排序字段值都相等时，输出结果的顺序就会因为N值不同而不同。

知道原因后，我们可以通过以下几种方法来避免这个“陷阱”。

1、让查询计划避免“SORT (ORDER BY STOPKEY)”，采用“SORT (ORDER BY)”，使数据排序不受ROWNUM的影响。但这样会使所有数据都做排序：

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name, rowid from t_test1 order by owner) a)

  4  where rn <= 10

  5  and rn >= 1;

OWNER                          OBJECT_NAME

------------------------------ ------------------------------

AFWOWNER                       AFWADAPTER

AFWOWNER                       AFWADAPTERCONFIGURATION

AFWOWNER                       AFWADAPTERCONFIGURATION_IDX2

AFWOWNER                       AFWADAPTERCONFIGURATION_PK

AFWOWNER                       AFWADAPTERCONFIGURATION_IDX1

AFWOWNER                       AFWADAPTERFQN_PK

AFWOWNER                       AFWADAPTERLOOKUP_IDX1

AFWOWNER                       AFWSERVERCODE_PK

AFWOWNER                       AFWSERVERCONFIGURATION_IDX1

AFWOWNER                       AFWTOKENTYPECODE_PK

10 rows selected.

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name, rowid from t_test1 order by owner) a)

  4  where rn <= 20

  5  and rn >= 11;

OWNER                          OBJECT_NAME

------------------------------ ------------------------------

AFWOWNER                       AFWTOKENTYPE

AFWOWNER                       AFWTOKENSTATUSCODE_PK

AFWOWNER                       AFWTOKENSTATUS

AFWOWNER                       AFWTOKENADMIN_IDX1

AFWOWNER                       AFWTOKENADMINCODE_PK

AFWOWNER                       AFWTOKENADMIN

AFWOWNER                       AFWTOKEN

AFWOWNER                       AFWSERVERCONFIGURATION_PK

AFWOWNER                       AFWTOKEN_PK

AFWOWNER                       AFWTOKEN_IDX6

10 rows selected.

sql> set autot trace

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name, rowid from t_test1 order by owner) a)

  4  where rn <= 20

  5  and rn >= 11;

10 rows selected.

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=237 Card=30670 Bytes=1441490)

   1    0   VIEW (Cost=237 Card=30670 Bytes=1441490)

   2    1     COUNT

   3    2       VIEW (Cost=237 Card=30670 Bytes=1042780)

   4    3         SORT (ORDER BY) (Cost=237 Card=30670 Bytes=1073450)

   5    4           table ACCESS (FULL) OF 'T_TEST1' (Cost=42 Card=30670 Bytes=1073450)

2、在排序时，加上一个或多个字段（如主键字段、ROWID），使排序结果具有唯一性：

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name, rowid from t_test1 order by owner, object_id) a

  4  where rownum <= 10)

  5  where rn >= 1;

OWNER                          OBJECT_NAME

------------------------------ ------------------------------

AFWOWNER                       AFWADAPTER

AFWOWNER                       AFWADAPTERFQN_PK

AFWOWNER                       AFWADAPTERCONFIGURATION

AFWOWNER                       AFWADAPTERCONFIGURATION_PK

AFWOWNER                       AFWADAPTERCONFIGURATION_IDX1

AFWOWNER                       AFWADAPTERCONFIGURATION_IDX2

AFWOWNER                       AFWADAPTERLOOKUP

AFWOWNER                       AFWADAPTERLOOKUP_IDX1

AFWOWNER                       AFWSERVER

AFWOWNER                       AFWSERVERCODE_PK

10 rows selected.

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name, rowid from t_test1 order by owner, object_id) a

  4  where rownum <= 20)

  5  where rn >= 11;

OWNER                          OBJECT_NAME

------------------------------ ------------------------------

AFWOWNER                       AFWSERVERCONFIGURATION

AFWOWNER                       AFWSERVERCONFIGURATION_PK

AFWOWNER                       AFWSERVERCONFIGURATION_IDX1

AFWOWNER                       AFWTOKEN

AFWOWNER                       AFWTOKEN_PK

AFWOWNER                       AFWTOKEN_IDX1

AFWOWNER                       AFWTOKEN_IDX2

AFWOWNER                       AFWTOKEN_IDX3

AFWOWNER                       AFWTOKEN_IDX4

AFWOWNER                       AFWTOKEN_IDX5

10 rows selected.

sql> set autot trace

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select owner, object_name, rowid from t_test1 order by owner, object_id) a

  4  where rownum <= 20)

  5  where rn >= 11;

10 rows selected.

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=253 Card=20 Bytes=940)

   1    0   VIEW (Cost=253 Card=20 Bytes=940)

   2    1     COUNT (STOPKEY)

   3    2       VIEW (Cost=253 Card=30670 Bytes=1042780)

   4    3         SORT (ORDER BY STOPKEY) (Cost=253 Card=30670 Bytes=1196130)

   5    4           table ACCESS (FULL) OF 'T_TEST1' (Cost=42 Card=306

          70 Bytes=1196130)

3、对排序字段建立索引，并强制使用索引。这样就能利用索引已经建立好的排序结果：

sql> create index t_test1_idx1 on t_test1(owner);

Index created.

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select /*+index(t T_TEST1_IDX1)*/owner, object_name from t_test1 t order by owner) a

  4  where rownum <= 10)

  5  where rn >= 1

  6  ;

OWNER                          OBJECT_NAME

------------------------------ ------------------------------

AFWOWNER                       AFWADAPTER

AFWOWNER                       AFWADAPTERCONFIGURATION

AFWOWNER                       AFWADAPTERCONFIGURATION_IDX1

AFWOWNER                       AFWADAPTERCONFIGURATION_IDX2

AFWOWNER                       AFWADAPTERCONFIGURATION_PK

AFWOWNER                       AFWADAPTERFQN_PK

AFWOWNER                       AFWADAPTERLOOKUP

AFWOWNER                       AFWADAPTERLOOKUP_IDX1

AFWOWNER                       AFWSERVER

AFWOWNER                       AFWSERVERCODE_PK

10 rows selected.

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select /*+index(t T_TEST1_IDX1)*/owner, object_name from t_test1 t order by owner) a

  4  where rownum <= 20)

  5  where rn >= 11;

OWNER                          OBJECT_NAME

------------------------------ ------------------------------

AFWOWNER                       AFWSERVERCONFIGURATION

AFWOWNER                       AFWSERVERCONFIGURATION_IDX1

AFWOWNER                       AFWSERVERCONFIGURATION_PK

AFWOWNER                       AFWTOKEN

AFWOWNER                       AFWTOKENADMIN

AFWOWNER                       AFWTOKENADMINCODE_PK

AFWOWNER                       AFWTOKENADMIN_IDX1

AFWOWNER                       AFWTOKENSTATUS

AFWOWNER                       AFWTOKENSTATUSCODE_PK

AFWOWNER                       AFWTOKENTYPE

10 rows selected.

sql> set autot trace

sql> select owner, object_name from

  2  (select a.*, rownum as rn from

  3  (select /*+index(t T_TEST1_IDX1)*/owner, object_name from t_test1 t order by owner) a

  4  where rownum <= 20)

  5  where rn >= 11;

10 rows selected.

Execution Plan

----------------------------------------------------------

   0      SELECT STATEMENT Optimizer=CHOOSE (Cost=414 Card=20 Bytes=940)

   1    0   VIEW (Cost=414 Card=20 Bytes=940)

   2    1     COUNT (STOPKEY)

   3    2       VIEW (Cost=414 Card=30670 Bytes=1042780)

   4    3         table ACCESS (BY INDEX ROWID) OF 'T_TEST1' (Cost=414Card=30670 Bytes=858760)

   5    4           INDEX (FULL SCAN) OF 'T_TEST1_IDX1' (NON-UNIQUE) (

          Cost=26 Card=30670)

以上就是ROWNUM的使用技巧及其注意事项，希望编程成员正确使用ROWNUM，也希望DBA遇到相关问题能迅速定位。

转载至：http://blog.csdn.net/nestashen/article/details/3117729

Oracle中ROWNUM的使用技巧的更多相关文章

ORACLE 中ROWNUM用法总结(转)
ORACLE 中ROWNUM用法总结! 对于 Oracle 的 rownum 问题,很多资料都说不支持>,>=,=,between...and,只能用以上符号(<.<=.!=) ...
oracle 中 rownum 和 row_number()
简单的介绍下oracle 中rownum 和 row_number() 使用,实例演示. 参照:http://www.cnblogs.com/zjrstar/archive/2006/08/31/49 ...
oracle中rownum和rowid的区别
rownum和rowid的区别总括: rownum和rowid都是伪列,但是两者的根本是不同的. rownum是根据sql查询出的结果给每行分配一个逻辑编号,所以你的sql不同也就会导致最终rownu ...
ORACLE 中ROWNUM用法总结!
ORACLE 中ROWNUM用法总结! 对于 Oracle 的 rownum 问题,很多资料都说不支持>,>=,=,between...and,只能用以上符号(<.<=.!=) ...
Oracle中rownum的基本用法
Oracle中rownum的基本用法对于rownum来说它是oracle系统顺序分配为从查询返回的行的编号,返回的第一行分配的是1,第二行是2,依此类推,这个伪字段可以用于限制查询返回的总行数,且r ...
ORACLE 中ROWNUM
ORACLE 中ROWNUM用法总结! 对于 Oracle 的 rownum 问题,很多资料都说不支持>,>=,=,between...and,只能用以上符号(<.<=.!=) ...
Oracle中rownum和rowid的理解
rownum,rowid都叫伪列. 但是,rownum是逻辑上的编号,且其值总是从1开始,每行的rounum不是固定的.而rowid是“物理”编号.若数据库文件没有移动,则每行的 rowid一般是固定 ...
oracle中rownum和rowid的区别【转】
rownum和rowid的区别总括: rownum和rowid都是伪列,但是两者的根本是不同的. rownum是根据sql查询出的结果给每行分配一个逻辑编号,所以你的sql不同也就会导致最终rownu ...
转：ORACLE 中ROWNUM用法总结!
oracle 分页查询语句:select * from (select u.*,rownum r from (select * from userifno) u where rownum<大值) ...

随机推荐

6-Ubuntu与Windows不能相互复制
在安装Ubuntu系统后发现与Windows系统的文件不能相互复制,网上查了很多教程,发现都是不能用的,能实现的方法如下所示: 第一步: sudo apt-get autoremove open-vm ...
mysql查询赋值、修改拼接字符串
sql中修改字符串类型的字段可以这么拼接:update tbName set UserName='abc'+UserName; 但mysql中就不行了,需要这样:update tbName set U ...
第四章 Windows的图形设备接口及Windows绘图 P83 4-6
实验(实习)名称图形程序设计及其应用实验(实习)日期一.实验目的 1．熟悉图形设备接口的概念及其应用. 二.实验内容及步骤实验任务 1.熟悉基于图形应用的程序设计: 2.掌握刷新技术及其 ...
python高性能编程方法一-乾颐堂
阅读 Zen of Python,在Python解析器中输入 import this. 一个犀利的Python新手可能会注意到"解析"一词, 认为Python不过是另一门脚本语言. ...
WCF技术剖析(卷1)WCF全面解析文摘
第1章 wcf简介 soa体现的是一种对关注点进行分解的思想,与技术无关 soa的基本思想: a.服务自治独立部署,不依赖其他 b.依赖于开放的标准采用xml,xsd,及wsdl作为服务描述的 ...
VS2010+SVN
小乌龟版本用1.6,用1.8时老报错SVN是2.06, SVN Server是2.1.9
Chrome Command Line API 参考
VueX-状态管理器
一.VueX功能与解决的问题 1.中央状态管理器的功能: 1.1.可以管理共享状态1.2.提供一个可修改状态的方法1.3.提供状态获取的方法1.4.状态更改后,有通知机制 2.中央状态管理器解决的问 ...
RocketMQ 运维指令
1.1. 控制台使用 RocketMQ 提供有控制台及一系列控制台命令,用于管理员对主题,集群,broker 等信息的管理登录控制台首先进入RocketMQ 工程,进入/RocketMQ/bin ...
NLog日志框架简写用法
本文转载:http://www.blogjava.net/qiyadeng/archive/2013/02/27/395799.html 在.net中也有非常多的日志工具,今天介绍下NLog.NLog ...

Oracle中ROWNUM的使用技巧