corrupt blocks in Oracle 10.2.0 DB server running on HP-UX 11.23 PA and VxFS 4.1

[Jdamian]

Hi

Oracle: 10g (10.2.0)
O.S: 11.23 on PA-RISC
FS: Veritas FS 4.1
mount options: largefiles,mincache=direct,delaylog,nodatainlog,convosync=direct
VxFS layout: 5

The patch PHCO_38434 is present.

The alert_log file contains corrupt block messages:

Corrupt block relative dba: 0x00400179 (file 1, block 377)
Fractured block found during buffer read

BUT the FS is not corrupted and the data files can be succesfully read by means of the dd command.

The syslog.log contains NO MESSAGE about file system corruption, PV link fails, I/O error, HBA error... NO ERROR MESSAGE found in syslog.log

any ideas?

Should mount options be changed? (buffer-cache is disabled for DB file systems but this behaviour also happened where buffer-cache was enabled)


Thanx in advance

[Analyst]

If Iam right,

The Filesystem is perfect, only issue with database.

You can use dbverify utility to find the corrupt block of your datafile ... and if you find corrupt datafile then just do one thing restore that datafile from you old backup... and recovery the database ... but for this you must have your database already running in archive log mode Smile other wise you will end up with crash database.

Note:- when you will restore your old datafile from backup the data will not be consistent that means you had crash the database... so you have to recove the database & for complete recovery in above senario what i had explained your database must be in archive log mode.

[Jdamian]

Thanx Analyst

but the corruption happened 3 times for last 3 months and data have been recovered from backup.

But this is no usual... I'm trying to find out the roots of the problem.

[Ganesan R]

Hi,

If dd is able to read then there is no possiblity of bad block at disk level. It could be the database structure corruption.

You can even run fsck with "-o full" option if you want to check file system level corruption.

[Wim Rombauts]

Ever considered an Oracle Bug ?
oracle has had multiple bugs in the past that can cause block corruption. Maybe you are hitting one.

Are you running plan 10.2 ? Or a specific PatchSet ? Maybe it is more logical toopen an SR with Oracle.

[Eric Antunes]

Hi Oscar,

Your issue must be in the buffer cache (memory): see Metalink Note 28814.1.

Best Regards,

Eric Antunes

[Jdamian]

Hi Wim.

The Oracle10g version number is:

ORACLE RDBMS Version: 10.2.0.4.0


Hi Eric.

I read the Oracle note and found nothing about buffer-cache.

I guess you point to Oracle's buffer-cache, because Veritas FS buffer-cache is bypassed by means of mount options 'mincache=direct,convosync=direct'

[Eric Antunes]

Hi Oscar,

The fisrt thing you should do is to verify the extent of the problem with dbverify. Run, as the oracle database owner (tipically oracle):

$dbv blocksize=8192 file=

Then, let us know the Original Error column (I suspect it will be a ORA-1578).

Best Regards,

Eric Antunes

[Vadim Loginov]

Hi Oscar,

Do you use user-managed online backups (scripts) to backup your database? If so, please make sure that you put your tablespaces in backup mode before taking a physical copy of your datafiles.

Check this link out: http://download.oracle.com/docs/cd/A91202_01/901_doc/server.901/a90133/glossary.htm#432645

Regards
Vadim