CentOS - Apr 2007 - Re: [CSL #329283] centosplus XFS bug?

We're running a mix of ext3 and xfs file systems on CentOS 4
workstations at our site. We have benefitted greatly from the increased
performance and fast (crash recovery) reboot times that we get from
xfs. We're currently running CentOS kernel 2.6.9-42.0.8.EL and xfs
version 0.2-1.

We have one host where we've got issues. It's a (very) busy imap server
with external scsi hardware raid, dual Xeon CPUs, and 8GB of RAM (it
runs the hugemem kernel). Twice now in the last few weeks we've seen it
hang with the follow log message repeated many times in the logs at the
time of the hang:

XFS: possible memory allocation deadlock in kmem_alloc (mode:0x250)

I see this error discussed in:

http://oss.sgi.com/bugzilla/show_bug.cgi?id=410

...and I'm wondering if the fixes mentioned in that bug are present in
the 0.2-1 centosplus xfs drivers, or if it could get included in an xfs
update for CentOS 4. Can anyone comment?

Thanks for the filesystem, and the OS, it all rocks, and have a
great day.

Dave Thompson
UW-Madison

On Wed, 2007-04-25 at 10:07 -0500, thomas at cs.wisc.edu
wrote:
> We're running a mix of ext3 and xfs file systems on CentOS 4
> workstations at our site. We have benefitted greatly from the increased
> performance and fast (crash recovery) reboot times that we get from
> xfs. We're currently running CentOS kernel 2.6.9-42.0.8.EL and xfs
> version 0.2-1.
> 
> We have one host where we've got issues. It's a (very) busy imap
server
> with external scsi hardware raid, dual Xeon CPUs, and 8GB of RAM (it
> runs the hugemem kernel). Twice now in the last few weeks we've seen it
> hang with the follow log message repeated many times in the logs at the
> time of the hang:
> 
> XFS: possible memory allocation deadlock in kmem_alloc (mode:0x250)
> 
> I see this error discussed in:
> 
> http://oss.sgi.com/bugzilla/show_bug.cgi?id=410
> 
> ...and I'm wondering if the fixes mentioned in that bug are present in
> the 0.2-1 centosplus xfs drivers, or if it could get included in an xfs
> update for CentOS 4. Can anyone comment?
> 
> Thanks for the filesystem, and the OS, it all rocks, and have a
> great day.
> 
Dave,

I would guess that this is not in there ... but I will double check and
see if we can get it rolled in.

I should be in the CentOS-5 module though.

Thanks,
Johnny Hughes
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL:
<http://lists.centos.org/pipermail/centos/attachments/20070425/5de6787e/attachment-0003.sig>

Johnny Hughes wrote:
>
>Dave,
>
>I would guess that this is not in there ... but I will double check and
>see if we can get it rolled in.
>
>I should be in the CentOS-5 module though.
It's going to be awhile before we'll be ready to upgrade that host to
CentOS
5, so I'm looking at either getting a different xfs module for c4 or
switching
the partition back to ext3.

So that I understand what I'm installing for c5, will the CentOS 5 xfs
module
for the mainline ("compatible with upstream") kernels be based on the 
in-kernel source, or will it be a replacement module (as in CentOS 4)?  How 
about the centosplus kernels?

Thanks again,

Dave

On Wednesday 25 April 2007, thomas at cs.wisc.edu wrote:
> We're running a mix of ext3 and xfs file systems on CentOS 4
> workstations at our site. We have benefitted greatly from the increased
> performance and fast (crash recovery) reboot times that we get from
> xfs. We're currently running CentOS kernel 2.6.9-42.0.8.EL and xfs
> version 0.2-1.
>
> We have one host where we've got issues. It's a (very) busy imap
server
> with external scsi hardware raid, dual Xeon CPUs, and 8GB of RAM (it
> runs the hugemem kernel).
This implies that you are running i386 and not x86_64. It has been my 
experience that xfs is less that solid on c4.i386 but fine on c4.x86_64. This 
is most likely due to the fact that i386 uses 4k kernel stack while x86_64 
uses 8k.

/Peter
> Twice now in the last few weeks we've seen it 
> hang with the follow log message repeated many times in the logs at the
> time of the hang:
...
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
URL:
<http://lists.centos.org/pipermail/centos/attachments/20070426/4bb6f5ad/attachment-0003.sig>

Peter Kjellstrom wrote:
>
>This implies that you are running i386 and not x86_64. It has been my=20
>experience that xfs is less that solid on c4.i386 but fine on c4.x86_64.
Th>is=20
>is most likely due to the fact that i386 uses 4k kernel stack while
x86_64>=20
>uses 8k.
Yes, that's correct.  Going to x86_64 may be a possibility for future
hardware
incarnations of the host.

Dave