[Linux-cluster] GFS2 crash

Steven Whitehouse swhiteho at redhat.com
Fri Mar 5 10:10:04 UTC 2010


Hi,

On Thu, 2010-03-04 at 08:35 -0800, Scooter Morris wrote:
> On 03/04/2010 02:25 AM, Steven Whitehouse wrote:
> > Hi,
> >
> > On Wed, 2010-03-03 at 21:23 -0800, Scooter Morris wrote:
> >    
> >> Hi all,
> >>       Just had a crash on our 3 node RedHat Enterprise Linux 5.4 cluster
> >> that looks a lot like
> >> https://bugzilla.redhat.com/show_bug.cgi?id=520720.  We're running
> >> kernel 2.6.18-164.11.1.el5.  Here is the traceback:
> >>
> >>      
> > That seems a reasonable conclusion. I assume that you were running with
> > one or more files with the journaled data flag set?
> >    
> Not intentionally.  We do have a several MySQL databases running, but we 
> haven't explicitly set the jdata flag on any files or directories.

That sounds just like the previous reporter in that case.

> > [snip]
> >    
> >> Since we're already running the latest 5.4 kernel, it's not clear what
> >> might be going on, here.  There is a note in the bug about making sure
> >> the gfs2-kmod from 5.2 isn't still around.  What version of gfs2-kmod is
> >> the old version, or should I just remove all instances of gfs2-kmod?
> >>
> >> -- scooter
> >>
> >>      
> > You can remove all versions of the kmod since they are all old. This is
> > the result of a packaging issue (which we are attempting to solve by
> > providing an empty kmod in future versions which will override the old
> > one) but in the mean time, upgrades from 5.2 or before require the old
> > gfs2 kmod to be removed manually.
> >
> > I don't see any sign of the kmod in the stack trace you sent though, so
> > I suspect its not an issue in this case. Certainly worth checking though
> > to be certain.
> >
> > Steve.
> >
> >    
> OK, I'm removing kmod-gfs2 from all nodes and will schedule a rolling 
> reboot this evening.
> 
Ok. Sounds good. If this solves the problem do let us know since we can
then close that bug. The original reporter seems to have disappeared
without telling us whether the issue was related to the old kmod which I
suspect, but we haven't confirmed yet,

Steve.





More information about the Linux-cluster mailing list