如何Shrink Undo表空间,释放过度占用的空间

环境:
OS:Red Hat Enterprise Linux AS release 4 (Nahant)
DB:Oracle Database 10g Enterprise Edition Release 10.2.0.1.0 - Production
一台Oracle10gR2数据库报出如下错误:
ORA-1653: unable to extend table SYSMAN.MGMT_SYSTEM_ERROR_LOG by 8 in tablespace SYSAUX
ORA-1653: unable to extend table SYSMAN.MGMT_SYSTEM_ERROR_LOG by 8 in tablespace SYSAUX
ORA-1653: unable to extend table SYSMAN.MGMT_SYSTEM_ERROR_LOG by 8 in tablespace SYSAUX
ORA-1653: unable to extend table SYSMAN.MGMT_SYSTEM_ERROR_LOG by 8 in tablespace SYSAUX
ORA-1653: unable to extend table SYSMAN.MGMT_SYSTEM_ERROR_LOG by 8 in tablespace SYSAUX

登陆检查，发现是SYSAUX表空间空间用尽,不能扩展,尝试手工扩展SYSAUX表空间:
alter database datafile '+ORADG/danaly/datafile/sysaux.266.600173881' resize 800m
Tue Nov 29 23:31:38 2005
ORA-1237 signalled during: alter database datafile '+ORADG/danaly/datafile/sysaux.266.600173881' resize 800m...

出现ORA-1237错误，提示空间不足。这时候我才认识到是磁盘空间可能被用完了.

是谁"偷偷的"用了那么多空间呢(本来有几十个G的Free磁盘空间的)?
检查数据库表空间占用空间情况:
SQL> select tablespace_name,sum(bytes)/1024/1024/1024 GB
2 from dba_data_files group by tablespace_name
3 union all
4 select tablespace_name,sum(bytes)/1024/1024/1024 GB
5 from dba_temp_files group by tablespace_name order by GB;

TABLESPACE_NAME GB
------------------------------ ----------
USERS .004882813
UNDOTBS2 .09765625
SYSTEM .478515625
SYSAUX .634765625
WAPCM_TS_VISIT_DETAIL .9765625
HY_DS_DEFAULT 1
MINT_TS_DEFAULT 1
MMS_TS_DATA2 1.375
MMS_IDX_SJH 2
MMS_TS_DEFAULT 2
IVRCN_TS_DATA 2

TABLESPACE_NAME GB
------------------------------ ----------
MMS_TS_DATA1 2
CM_TS_DEFAULT 5
TEMP 20.5498047
UNDOTBS1 27.1582031

15 rows selected.
不幸的发现，UNDO表空间已经扩展至27G，而TEMP表空间也扩展至20G，这2个表空间加起来占用了47G的磁盘空间,导致了空间不足。
显然曾经有大事务占用了大量的UNDO表空间和Temp表空间，Oracle的AUM(Auto Undo Management)从出生以来就经常出现只扩展,不收缩(shrink)的情况(通常我们可以设置足够的UNDO表空间大小,然后取消其自动扩展属性).
现在我们可以采用如下步骤回收UNDO空间:

1.确认文件
SQL> select file_name,bytes/1024/1024 from dba_data_files
2 where tablespace_name like 'UNDOTBS1';

FILE_NAME
--------------------------------------------------------------------------------
BYTES/1024/1024
---------------
+ORADG/danaly/datafile/undotbs1.265.600173875
27810
2.检查UNDO Segment状态
SQL> select usn,xacts,rssize/1024/1024/1024,hwmsize/1024/1024/1024,shrinks
2 from v$rollstat order by rssize;

USN XACTS RSSIZE/1024/1024/1024 HWMSIZE/1024/1024/1024 SHRINKS
---------- ---------- --------------------- ---------------------- ----------
0 0 .000358582 .000358582 0
2 0 .071517944 .071517944 0
3 0 .13722229 .13722229 0
9 0 .236984253 .236984253 0
10 0 .625144958 .625144958 0
5 1 1.22946167 1.22946167 0
8 0 1.27175903 1.27175903 0
4 1 1.27895355 1.27895355 0
7 0 1.56770325 1.56770325 0
1 0 2.02474976 2.02474976 0
6 0 2.9671936 2.9671936 0

11 rows selected.
3.创建新的UNDO表空间
SQL> create undo tablespace undotbs2 ;
（经测试，在9i环境下后面还要加上datafile '/opt/..../undotbs2.dbf' size 1024M）
Tablespace created.
4.切换UNDO表空间为新的UNDO表空间
SQL> alter system set undo_tablespace=undotbs2 scope=both;

System altered.

创建了新的UNDO表空间以后，如果不知道系统使用的是pfile还是spfile文件，应使用参数both,会同时修改spfile文件，避免出现冲突。

5.等待原UNDO表空间所有UNDO SEGMENT OFFLINE
SQL> select usn,xacts,status,rssize/1024/1024/1024,hwmsize/1024/1024/1024,shrinks
2 from v$rollstat order by rssize;


USN XACTS STATUS RSSIZE/1024/1024/1024 HWMSIZE/1024/1024/1024 SHRINKS
---------- ---------- --------------- --------------------- ---------------------- ----------
14 0 ONLINE .000114441 .000114441 0
19 0 ONLINE .000114441 .000114441 0
11 0 ONLINE .000114441 .000114441 0
12 0 ONLINE .000114441 .000114441 0
13 0 ONLINE .000114441 .000114441 0
20 0 ONLINE .000114441 .000114441 0
15 1 ONLINE .000114441 .000114441 0
16 0 ONLINE .000114441 .000114441 0
17 0 ONLINE .000114441 .000114441 0
18 0 ONLINE .000114441 .000114441 0
0 0 ONLINE .000358582 .000358582 0

USN XACTS STATUS RSSIZE/1024/1024/1024 HWMSIZE/1024/1024/1024 SHRINKS
---------- ---------- --------------- --------------------- ---------------------- ----------
6 0 PENDING OFFLINE 2.9671936 2.9671936 0

12 rows selected.
再看:
11:32:11 SQL> /

USN XACTS STATUS RSSIZE/1024/1024/1024 HWMSIZE/1024/1024/1024 SHRINKS
---------- ---------- --------------- --------------------- ---------------------- ----------
15 1 ONLINE .000114441 .000114441 0
11 0 ONLINE .000114441 .000114441 0
12 0 ONLINE .000114441 .000114441 0
13 0 ONLINE .000114441 .000114441 0
14 0 ONLINE .000114441 .000114441 0
20 0 ONLINE .000114441 .000114441 0
16 0 ONLINE .000114441 .000114441 0
17 0 ONLINE .000114441 .000114441 0
18 0 ONLINE .000114441 .000114441 0
19 0 ONLINE .000114441 .000114441 0
0 0 ONLINE .000358582 .000358582 0

11 rows selected.

Elapsed: 00:00:00.00
6.删除原UNDO表空间

11:34:00 SQL> drop tablespace undotbs1 including contents;
（本人经测试觉得其实是否可以用drop tablespace undotbs1 including contents and datafiles;直接连硬盘里面的dbf文件件一起删除）
Tablespace dropped.

Elapsed: 00:00:03.13
7.检查空间情况
由于我使用的ASM管理,可以使用10gR2提供的信工具asmcmd来察看空间占用情况.
[oracle@danaly ~]$ export ORACLE_SID=+ASM
[oracle@danaly ~]$ asmcmd
ASMCMD> du
Used_MB Mirror_used_MB
21625 21625
ASMCMD> exit

空间已经释放。