Oracle中ROWNUM的使用技巧
2015-06-25 16:13
525 查看
转自：http://blog.sina.com.cn/s/blog_683b7a840100mhhs.html
参考：http://blog.csdn.net/xueshuangshuang123/article/details/8781442
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(ROWNUMmin_
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(*)
----------
361928
SQL> select count(*) from css_bl_view a
2 where rownum > 1;
COUNT(*)
----------
0
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(*)
----------
361927
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(*)
----------
21645
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 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 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 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。
3.4 性能陷阱
在之前，我们已经介绍过ROWNUM 的使用及一些要注意的问题。这里，再介绍一种由
ROWNUM 导致的性能陷阱。事实上，如果能清楚了解ROWNUM 的基本机制，那么对这个陷阱
就很容易理解。
我们有定义如下视图，
CREATE OR REPLACE VIEW MY_VIEW AS
SELECT ROWNUM AS ID,
T0.COMP_ID AS COMPANY_ID,
T0.SH_ID AS B_NUM,
'' AS L_NUM,
'' AS C_NUM,
T0.C_COMP_ID AS SP_COMP_ID,
'' AS C_TYPE,
T0.REC_UPD_DT AS LAST_MOD_DT
FROM T_COM T0
WHERE T0.DELETED_IND = 0
AND T0.VIEW_TYPE = 'BK'
在引用表T_COM 的(COMP_ID, VIEW_TYPE, DELETED_IND, SH_ID)已经建立索引，且
COMP_ID 的选择性很强，VIEW_TYPE 和DELETED_IND 选择性则很弱。
再看下面的查询
select id,
company_id,
c_num,
b_num,
l_num,
last_mod_dt,
c_type,
sp_comp_id
from MY_VIEW
where company_id = '1111111111'
嗯，优化器应该先进行查询重写（Predicate Pushing），将主查询中的company_id
= '1111111111'与视图中的T0.DELETED_IND = 0 AND T0.VIEW_TYPE = 'BK'共同作用到表
T_COM 上，命中到索引。但是，是这样吗，看看它的查询计划：
---------------------------------------------------------------------------------
---------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------------
---------
| 0 | SELECT STATEMENT | | 397K| 26M| 49582 (2)|
00:09:55 |
|* 1
| FILTER | | | | |
|
|* 2 | VIEW | MY_VIEW | 397K| 26M| 49582 (2)|
00:09:55 |
| 3
| COUNT | | | | |
|
|* 4 | TABLE ACCESS FULL| T_COM | 397K| 23M| 49582 (2)|
00:09:55 |
---------------------------------------------------------------------------------
---------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("COMPANY_ID"='111111111')
3 - filter("T0"."DELETED_IND"=0 AND "T0"."VIEW_TYPE"='BK')
并未命中索引。仔细分析视图的定义语句，其中包含了对ROWNUM 的查询。ROWNUM 是
一个虚字段，只有产生结果集时才会有值。因此，为保证逻辑结果，优化器并没有将视图
查询条件与外部主查询条件合并后再执行查询操作，而是先执行子查询部分（条件不足以
命中索引），得到结果集（计划中的第一步Full Table Scan）和对应的ROWNUM 值（计划
中的第二步COUNT）以后再根据外部条件对结果集过滤（计划中的第三步FILTER）。当我
们将ROWNUM 从视图中拿掉，执行计划就能像我们之前所预想的命中索引了。
CREATE OR REPLACE VIEW MY_VIEW AS
SELECT 1 AS ID,
T0.COMP_ID AS COMPANY_ID,
T0.SH_ID AS B_NUM,
'' AS L_NUM,
'' AS C_NUM,
T0.C_COMP_ID AS SP_COMP_ID,
'' AS C_TYPE,
T0.REC_UPD_DT AS LAST_MOD_DT
FROM T_COM T0
WHERE T0.DELETED_IND = 0
AND T0.VIEW_TYPE = 'BK'
---------------------------------------------------------------------------------
---------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------------
--------
| 0 | SELECT STATEMENT | | 299 | 20631 | 340 (1)|
00:00:05 |
| 1 | TABLE ACCESS BY INDEX ROWID| T_COM | 299 | 20631 | 340 (1)|
00:00:05 |
|* 2 | INDEX RANGE SCAN | T_COM_INDX4| 299 | | 77 (0)|
00:00:01 |
---------------------------------------------------------------------------------
---------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("T0"."COMP_ID"='111111111' AND "T0"."DELETED_IND"=0 AND
"T0"."VIEW_TYPE"='BK')
filter("T0"."DELETED_IND"=0 AND "T0"."VIEW_TYPE"='BK')
如果开发人员不深入理解ROWNUM 的机制，而简单的将其理解为一个子增长的字段的
话，就很容易掉如这样的陷阱了。
补充：看看假如在子查询或视图中存在ROWNUM 时，查询重写和不重写的逻辑结果的差
异吧。
SQL> select id, owner, object_name
2 from (select rownum as id, owner, object_name from t_test1 where object_name
like 'TEM%') v
3 where owner='DEMO';
ID OWNER OBJECT_NAME
---------- -------------------------------------------------- -------------------
-----
5 DEMO TEMP_BOOK_CMP
SQL>
SQL> select id, owner, object_name
2 from (select rownum as id, owner, object_name from t_test1 where object_name
like 'TEM%' and owner='DEMO') v;
ID OWNER OBJECT_NAME
---------- -------------------------------------------------- -------------------
-----
1 DEMO TEMP_BOOK_CMP
真实字段的内容没有不同，但是由ROWNUM 返回的数据就不同了。
--- Fuyuncat ---
参考：http://blog.csdn.net/xueshuangshuang123/article/details/8781442
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$_EVENT_QTAB_E
…
在没有ROWNUM 时要实现这样的功能会很复杂，但通过ROWNUM 我们可以轻松实现：
SQL> SELECT DECODE(ROWNUM-min_sno,0,a.owner,NULL) owner,DECODE(ROWNUMmin_
sno,0,1,rownum+1-min_sno) sno, a.name
2 FROM (SELECT *
3 FROM t_test8
4 ORDER BY owner, name ) a,
5 (SELECT owner, MIN(rownum) min_sno
6 FROM( SELECT *
7 FROM t_test8
8 ORDER BY owner, name)
9 GROUP BY owner) b
10 WHERE a.owner=b.owner;
OWNER SNO NAME
------------------------------ ---------- ------------------------------
APARKMAN 1 JOB_QUEUE
2 AQ$_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(*)
----------
361928
SQL> select count(*) from css_bl_view a
2 where rownum > 1;
COUNT(*)
----------
0
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(*)
----------
361927
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(*)
----------
21645
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 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 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 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。
3.4 性能陷阱
在之前，我们已经介绍过ROWNUM 的使用及一些要注意的问题。这里，再介绍一种由
ROWNUM 导致的性能陷阱。事实上，如果能清楚了解ROWNUM 的基本机制，那么对这个陷阱
就很容易理解。
我们有定义如下视图，
CREATE OR REPLACE VIEW MY_VIEW AS
SELECT ROWNUM AS ID,
T0.COMP_ID AS COMPANY_ID,
T0.SH_ID AS B_NUM,
'' AS L_NUM,
'' AS C_NUM,
T0.C_COMP_ID AS SP_COMP_ID,
'' AS C_TYPE,
T0.REC_UPD_DT AS LAST_MOD_DT
FROM T_COM T0
WHERE T0.DELETED_IND = 0
AND T0.VIEW_TYPE = 'BK'
在引用表T_COM 的(COMP_ID, VIEW_TYPE, DELETED_IND, SH_ID)已经建立索引，且
COMP_ID 的选择性很强，VIEW_TYPE 和DELETED_IND 选择性则很弱。
再看下面的查询
select id,
company_id,
c_num,
b_num,
l_num,
last_mod_dt,
c_type,
sp_comp_id
from MY_VIEW
where company_id = '1111111111'
嗯，优化器应该先进行查询重写（Predicate Pushing），将主查询中的company_id
= '1111111111'与视图中的T0.DELETED_IND = 0 AND T0.VIEW_TYPE = 'BK'共同作用到表
T_COM 上，命中到索引。但是，是这样吗，看看它的查询计划：
---------------------------------------------------------------------------------
---------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------------
---------
| 0 | SELECT STATEMENT | | 397K| 26M| 49582 (2)|
00:09:55 |
|* 1
| FILTER | | | | |
|
|* 2 | VIEW | MY_VIEW | 397K| 26M| 49582 (2)|
00:09:55 |
| 3
| COUNT | | | | |
|
|* 4 | TABLE ACCESS FULL| T_COM | 397K| 23M| 49582 (2)|
00:09:55 |
---------------------------------------------------------------------------------
---------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("COMPANY_ID"='111111111')
3 - filter("T0"."DELETED_IND"=0 AND "T0"."VIEW_TYPE"='BK')
并未命中索引。仔细分析视图的定义语句，其中包含了对ROWNUM 的查询。ROWNUM 是
一个虚字段，只有产生结果集时才会有值。因此，为保证逻辑结果，优化器并没有将视图
查询条件与外部主查询条件合并后再执行查询操作，而是先执行子查询部分（条件不足以
命中索引），得到结果集（计划中的第一步Full Table Scan）和对应的ROWNUM 值（计划
中的第二步COUNT）以后再根据外部条件对结果集过滤（计划中的第三步FILTER）。当我
们将ROWNUM 从视图中拿掉，执行计划就能像我们之前所预想的命中索引了。
CREATE OR REPLACE VIEW MY_VIEW AS
SELECT 1 AS ID,
T0.COMP_ID AS COMPANY_ID,
T0.SH_ID AS B_NUM,
'' AS L_NUM,
'' AS C_NUM,
T0.C_COMP_ID AS SP_COMP_ID,
'' AS C_TYPE,
T0.REC_UPD_DT AS LAST_MOD_DT
FROM T_COM T0
WHERE T0.DELETED_IND = 0
AND T0.VIEW_TYPE = 'BK'
---------------------------------------------------------------------------------
---------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
Time |
---------------------------------------------------------------------------------
--------
| 0 | SELECT STATEMENT | | 299 | 20631 | 340 (1)|
00:00:05 |
| 1 | TABLE ACCESS BY INDEX ROWID| T_COM | 299 | 20631 | 340 (1)|
00:00:05 |
|* 2 | INDEX RANGE SCAN | T_COM_INDX4| 299 | | 77 (0)|
00:00:01 |
---------------------------------------------------------------------------------
---------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("T0"."COMP_ID"='111111111' AND "T0"."DELETED_IND"=0 AND
"T0"."VIEW_TYPE"='BK')
filter("T0"."DELETED_IND"=0 AND "T0"."VIEW_TYPE"='BK')
如果开发人员不深入理解ROWNUM 的机制，而简单的将其理解为一个子增长的字段的
话，就很容易掉如这样的陷阱了。
补充：看看假如在子查询或视图中存在ROWNUM 时，查询重写和不重写的逻辑结果的差
异吧。
SQL> select id, owner, object_name
2 from (select rownum as id, owner, object_name from t_test1 where object_name
like 'TEM%') v
3 where owner='DEMO';
ID OWNER OBJECT_NAME
---------- -------------------------------------------------- -------------------
-----
5 DEMO TEMP_BOOK_CMP
SQL>
SQL> select id, owner, object_name
2 from (select rownum as id, owner, object_name from t_test1 where object_name
like 'TEM%' and owner='DEMO') v;
ID OWNER OBJECT_NAME
---------- -------------------------------------------------- -------------------
-----
1 DEMO TEMP_BOOK_CMP
真实字段的内容没有不同，但是由ROWNUM 返回的数据就不同了。
--- Fuyuncat ---
内容来自用户分享和网络整理，不保证内容的准确性，如有侵权内容，可联系管理员处理 
相关文章推荐
- plsql与tsql的语法不同
- plsql连接oracle数据库报ora 12154错误解决方法
- plsql和tsql常用函数比对
- oracle客户端PLSQL连接失败解决方法
- 解决plsql遇到乱码的问题
- plsql修改数据库连接
- Record is locked by another user --Oracle行锁解锁
- 安装Oracle数据库客户端
- 使用PLSQL 创建Oracle数据库用户
- 登陆oracle慢原因分析和解决方案
- 西南科技大学 Oracle 数据库实验1 建表
- pl/sql 创建表空间 用户并授予权限
- ORACLE INSTANT CLIENT和PLSQL的配置
- PL/SQL用户名、密码保存
- plsql 无法解析指定的连接标识符
- PL/SQL 下邮件发送程序
- win7+64位+Oracle+11g+64位下使用PLSQL+Developer+的解决办法
- Oracle中Binary_Integer与Pls_Integer 的区别
- Oracle建分区表
- Oracle执行计划详解