【Oracle】undo损坏，无备份非常规恢复

客户的一个测试环境，主机异常断电，启动后发现undo文件损坏，无法启动，在open阶段报错如下： Errors in file /u01/app/oracle/diag/rdbms/cdrdb/CDRDB/trace/CDRDB_ora_4109.trc: ORA-01122: database file 3 failed verification check ORA-01110: data f

客户的一个测试环境，主机异常断电，启动后发现undo文件损坏，无法启动，在open阶段报错如下：

Errors in file /u01/app/oracle/diag/rdbms/cdrdb/CDRDB/trace/CDRDB_ora_4109.trc:

ORA-01122: database file 3 failed verification check

ORA-01110: data file 3: '/u01/app/oracle/oradata/CDRDB/undotbs01.dbf'

ORA-01210: data file header is media corrupt

ORA-1122 signalled during: ALTER DATABASE OPEN...

 

由于是测试环境，没有备份，但是又需要里边的一些数据，所以我尝试使用非常规恢复方法进行了尝试。

先冷备份现有环境！！！！！！！！！！！！！！！！

 

创建pfile文件：

create pfile from spfile;

在pfile中修改这两个参数

#*.undo_tablespace='UNDOTBS1'

*.undo_management= MANUAL

之后用这个pfile启动：

SYS@CDRDB>startup force pfile='/u01/app/oracle/product/11.2.0/dbhome_1/dbs/initCDRDB.ora';

ORACLE instance started.

 

Total System Global Area  523108352 bytes

Fixed Size                  1337632 bytes

Variable Size             364906208 bytes

Database Buffers          150994944 bytes

Redo Buffers                5869568 bytes

Database mounted.

ORA-01122: database file 3 failed verification check

ORA-01110: data file 3: '/u01/app/oracle/oradata/CDRDB/undotbs01.dbf'

ORA-01210: data file header is media corrupt

仍然报错

之后尝试drop掉这个undo

SYS@CDRDB>alter database datafile 3 offline drop;

 

Database altered.

之后重新开库

SYS@CDRDB>alter database open;

 

Database altered.

之后创建新的undo表空间undotbs2

SYS@CDRDB>create undo tablespace undotbs2 datafile '/u01/app/oracle/oradata/CDRDB/undotbs02_01.dbf' size 100M;

create undo tablespace undotbs2 datafile '/u01/app/oracle/oradata/CDRDB/undotbs02_01.dbf' size 100M

*

ERROR at line 1:

ORA-00604: error occurred at recursive SQL level 1

ORA-01552: cannot use system rollback segment for non-system tablespace

'DATA_OL'

ORA-06512: at line 999

ORA-01552: cannot use system rollback segment for non-system tablespace

'DATA_OL'

产生报错如上

 

比较疑惑创建undo为什么会影响到DATA_OL表空间,所以做了个10046

SYS@CDRDB>oradebug event 10046 trace name context off

Statement processed.

SYS@CDRDB>oradebug tracefile_name

/u01/app/oracle/diag/rdbms/cdrdb/CDRDB/trace/CDRDB_ora_4279.trc

[oracle@centos-1 ~]$ tkprof /u01/app/oracle/diag/rdbms/cdrdb/CDRDB/trace/CDRDB_ora_4279.trc

output = 1.trm

查看1.trm，发现了原因：

  "OGG".DDLReplication.dbQueried IS NULL THEN

                SELECT database_role,

  open_mode

                INTO dbRole, dbOpenMode

                FROM v$database;

 

  "OGG".DDLReplication.dbQueried := TRUE;

              END IF;

                IF NOT (

^@                      (dbRole = 'PRIMARY' OR dbRole = 'LOGICAL STANDBY')

                        AND dbOpenMode =

  'READ WRITE'

                        )

                        THEN

                        -- do not write any trace even though it

  should work as this is standby

                        "OGG"

  .DDLReplication.setCtxInfo(-1,-1,-1,-1,-1);

                        RETURN; -- do not use

  trigger if not read/write and primary/logical_standby

                END IF;

        EXCEPTION

......略

 

原来是因为这个库配置过OGG的DDL同步，有DDL产生时会产生insert操作，使用DATA_OL表空间。

原因找到~  跑脚本关闭该库OGG的DDL配置即可

SYS@CDRDB>@ddl_disable.sql

 

Trigger altered.

之后再重新创建undotbs2表空间

SYS@CDRDB>create undo tablespace undotbs2 datafile '/u01/app/oracle/oradata/CDRDB/undotbs02_01.dbf' size 1000M;

 

Tablespace created.

这次很顺利的添加成功。

 

undotbs2添加完毕，接下来关闭数据库，修改pfile参数

*.undo_tablespace='UNDOTBS2'

*.undo_management=AUTO

 

重启数据库用pfile启动

SYS@CDRDB>startup pfile='/u01/app/oracle/product/11.2.0/dbhome_1/dbs/initCDRDB.ora';

ORACLE instance started.

 

Total System Global Area  523108352 bytes

Fixed Size                  1337632 bytes

Variable Size             364906208 bytes

Database Buffers          150994944 bytes

Redo Buffers                5869568 bytes

Database mounted.

Database opened.

数据库成功启动，然而在很多业务表查询时依然会报错：

ERROR at line 1:

ORA-00376: file 3 cannot be read at this time

ORA-01110: data file 3: '/u01/app/oracle/oradata/CDRDB/undotbs01.dbf'

依然需要原来的undotbs01.dbf回滚数据库崩溃时未提交的事务

 

之后我先尝试了设置event 10513，来屏蔽smon的回滚

SYS@CDRDB>alter system set events '10513 trace name context forever, level 2';

 

System altered.

发现还是会报之前的错误，看来需要自己手工去屏蔽回滚段了。

SYS@CDRDB>SELECT SEGMENT_NAME,STATUS FROM DBA_ROLLBACK_SEGS;

 

SEGMENT_NAME                   STATUS

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

SYSTEM                         ONLINE

_SYSSMU10_4131489474$          NEEDS RECOVERY

_SYSSMU9_1735643689$           NEEDS RECOVERY

_SYSSMU8_3901294357$           NEEDS RECOVERY

_SYSSMU7_3517345427$           NEEDS RECOVERY

_SYSSMU6_2897970769$           NEEDS RECOVERY

_SYSSMU5_538557934$            NEEDS RECOVERY

_SYSSMU4_1003442803$           NEEDS RECOVERY

_SYSSMU3_1204390606$           NEEDS RECOVERY

_SYSSMU2_967517682$            NEEDS RECOVERY

_SYSSMU1_592353410$            NEEDS RECOVERY

 

SEGMENT_NAME                   STATUS

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

_SYSSMU30_244658789$           ONLINE

_SYSSMU29_1020880693$          ONLINE

_SYSSMU28_2912622077$          ONLINE

_SYSSMU27_747253598$           ONLINE

_SYSSMU26_560868814$           ONLINE

_SYSSMU25_1357066082$          ONLINE

_SYSSMU24_103440716$           ONLINE

_SYSSMU23_1006903361$          ONLINE

_SYSSMU22_2808190508$          ONLINE

_SYSSMU21_39626587$            ONLINE

 

21 rows selected.

 

之后把查询到的NEEDS RECOVERY的表加到以下两个参数中，屏蔽这些回滚段

_offline_rollback_segments/_corrupted_rollback_segments参数

*._offline_rollback_segments=(_SYSSMU10_4131489474$,_SYSSMU9_1735643689$,_SYSSMU8_3901294357$,_SYSSMU7_3517345427$,_SYSSMU6_2897970769$,_SYSSMU5_538557934$,_SYSSMU4_1003442803$,_SYSSMU3_1204390606$,_SYSSMU2_967517682$,_SYSSMU1_592353410$)

*._corrupted_rollback_segments=(_SYSSMU10_4131489474$,_SYSSMU9_1735643689$,_SYSSMU8_3901294357$,_SYSSMU7_3517345427$,_SYSSMU6_2897970769$,_SYSSMU5_538557934$,_SYSSMU4_1003442803$,_SYSSMU3_1204390606$,_SYSSMU2_967517682$,_SYSSMU1_592353410$)

 

使用这个pfile启动数据库

startup force pfile='/u01/app/oracle/product/11.2.0/dbhome_1/dbs/initCDRDB.ora';

 

之后对那些测试表都查询了一下，确认可以查询，之后导出帮他们导出数据，一切OK~