Jan Kundrát wrote:
> Jan Kundrát wrote:
>> Will I have the XFS filesystem after the upgrade?
> 
> Quick answer: yes, it's integrated into the kernel RPM. Other issues
> (yum willing to remove bunch of X11-related packages if the "xfs" RPM
> was installed before) still stand.
> 
> Have fun,
> Jan
> 

Hi Jan,
Hopefully Jon Peatfield's answer already answered your question, but 
just incase it didn't.

Our original design of getting the xfs file system onto a machine was 
flawed, because of the package "xfs".  At the time, we didn't know it, 
but xorg-x11-xfs "provided" xfs.  So there was a conflict, but it wasn't 
easy to see until yum started doing crazy things, like what you saw.
So we changed xfs to be xfs-filesystem (as Jon said) and this solved the 
problem for newly installed machines.
If people already had the xfs package installed, then they had to 
manually update to xfs-filesystem by hand, or just remove xfs the 
package, they still have this xfs versus xorg-x11-xfs problem.  But the 
problem sat dormant as long as nothing changed.

Then RedHat started putting xfs into the x86_64 kernel in SL 5.4.  So 
that there was no longer a kernel-module-xfs rpm.  Also, yum has the 
"installonlyn" kernel feature, that removes older kernels after you've 
reached 'n'.  See the installonly_limit line in /etc/yum.conf if you 
want to change that.  The default limit is set to 4.
So as you update your kernel's, yum is removing your older kernels along 
with their kernel-module's related to that kernel.  And eventually, all 
of your kernel-module-xfs packages are gone.

But xfs (and xfs-filesystem) depend on kernel-module-xfs.  They are 
designed this way so that in order to install xfs, all you have to do is 
install xfs-filesystem (or originally xfs) and yum will automatically 
pull in the correct kernel-module-xfs.

Since yum is pulling off the last kernel-module-xfs, and xfs depends on 
that, yum starts pulling off the xfs package.  But since other packages 
think that xfs is the "x font server" they start getting removed also, 
and then ... well, you saw what happens.

I know that is sortof a long winded description, but hopefully it 
answers all your questions.  It can be summed up with "Our original try 
at XFS in Scientific Linux was flawed.  We apologize for that."

Troy
-- 
__________________________________________________
Troy Dawson  [log in to unmask]  (630)840-6468
Fermilab  ComputingDivision/LSCS/CSI/USS Group
__________________________________________________