Re: Red Hat 2.1, Oracle and 25GB file block corruption on HP

  • From: Tim Gorman <tim@xxxxxxxxx>
  • To: <oracle-l@xxxxxxxxxxxxx>
  • Date: Thu, 27 Jan 2005 15:49:28 -0700


After all, EXP will drop dead on only the first corrupt block within the
high-water mark of tables;  it won't check indexes.

Some better alternatives:

1) If error messages (or the trace files accompanying the ORA-00600)
displayed actual block addresses (either FILE#/BLOCK# or DBA), then you can
use the DBMS_REPAIR.CHECK_OBJECT procedure to verify, after obtaining the
name of the segment by querying DBA_EXTENTS...

2) ALTER SYSTEM DUMP DATAFILE [ file-id | 'file-name' ] [ BLOCK block-id |
BLOCK MIN min-block-id BLOCK MAX max-block-id ] can help as well?  It
produces a trace file in the USER_DUMP_DEST, and usually if the block is
corrupted, the dump will say exactly that...

3) RMAN "backup validate" command performs a "false backup" (i.e. just reads
datafiles, doesn't write to backup media).  This is the "EXP FULL=Y
FILE=/dev/null" command done the right way...

Hope this helps...


on 1/27/05 3:04 PM, aj wells at awellsdba@xxxxxxxxx wrote:

> Okay... not that that is out of the way...
> We have a multi TB databae (one of several) that has many 25GB files
> (since it seems that is the biggest file we can make... ) and we have
> several file systems.  "we" just created two new file systems that are
> 600+ GB each across several disks (0+1).
> When we allocated new files on the first of these file systems, we got
> block corruption... (ORA-00600: INTERNAL ERROR CODE, ARGUMENTS:
> [25012], [27]) on several of the files.
> HP says not hardware, RH says either Oracle or Hardware and Oracle
> says it is OS or hardware.
> Anyone else seen anything vaguely resembling this... it is becoming a
> big giant Charlie Foxtrot and no one is of much help (although Oracle
> did have us start an export 13 hours ago on the tablespace and it is
> now nearly half done... that will tell us if the two extents in the
> only file left showing corruption is actually corrupt or only
> pretending to be... )
> thanks
> aj
> --
> //


Other related posts: