SCIENTIFIC-LINUX-USERS Archives

December 2011

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Reply To:
Date:
Wed, 7 Dec 2011 10:21:32 -0800
Content-Type:
MULTIPART/MIXED
Parts/Attachments:
TEXT/PLAIN (2101 bytes)
On Wed, 7 Dec 2011, Ryan C. England wrote:

> Everyone,
> I'm having an issue with an XFS stack overflow.  I've included the logs
> below.  After a few searches it looks like this is a problem with the xfs
> module included in 2.6.32.  Has anyone else experienced this problem?
> 
> I have found a couple of patches however, I thought this would have been
> included within patches from our repos eventually.  I have not patched
> outside of the mainstream so I'm not familiar with the process.  Any
> assistance would be greatly appreciated.
> 
> http://oss.sgi.com/archives/xfs/2011-03/msg00399.html
> http://oss.sgi.com/archives/xfs/2010-05/msg00385.html
> http://oss.sgi.com/archives/xfs/2010-06/msg00312.html
> 
>snip

Historically xfs is vulnerable to issues with stack overflow when 
many additional kernel modules are in use; example of an old discussion:
   http://marc.info/?l=linux-xfs&m=113600600704818

especially on 32-bit kernels, which should not be used with xfs -
however I see that yours is 64-bit.

You might want to look at what you are using on that server.  For this
reason we avoid lvm, for example, on our xfs-based servers.  iscsi is
a killer in some old tests we did.  We don't run nfs daemons on them
either ..

However in your output the list of modules doesn't seem large, assuming
that the kernel oops output listed them all.

There might be some kernel or nfs tuning you could do too?  My favourite
value to bump up is vm.min_free_kbytes, at least for contributed issues 
with 10 Gb network drivers.  As well, sometimes people bump up the memory on
their systems without revisiting any previously tuned values
in /etc/sysctl.conf

Unrelated, but if you aren't actually using ipv6, and aren't bonding network
interfaces, it is always nice to just turn it off, so that it doesn't hang
around for no reason, polluting the kernel module list:

$ cat /etc/modprobe.d/disable-ipv6.conf 
install ipv6 /bin/true

cheers, etc
  Denice
-- 
deatrich @ triumf.ca, Science/Atlas         PH: +1 604-222-7665
<*> This moment's fortune cookie:
Secretary's Revenge:
 	Filing almost everything under "the".

ATOM RSS1 RSS2