On Mon, 20 May 2002, Sidik Isani wrote:
> Hello -
>
> Thanks again for your help several months ago with xfs_growfs!
> Now we have a new problem . . .
> I was trying to resolve the slow performance issue of some of our
> RAID5+XFS (2.4.16 kernel) by upgrading to 2.4.18 and XFS-1.1, and
> reformatting with an external log, as suggested in the FAQ.
2.4.18 has a lot better RAID5 performance with an internal log as well.
I also believe that the fixes that went into the CVS tree for the
multiple block sizes support makes the raid5 support better.
I don't think anyone benchmarked this yet but the performance difference
is probably not as large anymore.
> During resyncing, one of the disks failed and the raid 5 went into
> degraded mode (no other disks had errors). After a clean reboot, still
> running in degraded mode, (shouldn't matter to XFS, but I thought I'd
> mention it) everything seemed OK until I tried to remove a directory:
>
> May 19 19:03:51 ike kernel: xfs_force_shutdown(md(9,0),0x8) called from line
> 1039 of file xfs_trans.c. Return address = 0xc01ed751
> May 19 19:03:51 ike kernel: Corruption of in-memory data detected. Shutting
> down filesystem: md(9,0)
> May 19 19:03:51 ike kernel: Please umount the filesystem, and rectify the
> problem(s)
> [System froze!]
I think something went bad the moment that the raid5 went into degraded
mode. This shouldn't happen but you are wise in not automatically
repairing the fs.
> to replay the log, and ran xfs_repair -n. The output is included below.
> I'm thinking of trying again later today, maybe with an internal
> log again (which should be usable now, with 2.4.18, right?) But the
> crash above worries me. Please let me know if there are any other
> tests I should run on the crashed filesystem before starting over.
As stated aboce the internal log raid5 should be OK. I still have a system
raid5 with an internal log that did get a bit faster with the change to
2.4.18.
Since it looks like filesystem corruption I think it is best to be careful
from here on. I think making a backup now would be a good idea.
Cheers
Seth
|