This is a bit disturbing.
As much as it would be nice (though painful) to switch to Suse or
something else, the RedHat line (FC3, RHEL3/4) is what the big vendors
wish to support in movie production. Alias is the most notable. None
of our vendors require XFS, but it is vastly preferable to me.
Consequently, we have hundreds of machines running RH9 with xfs, and
implementing the superior 2.6 kernel is a pain in RH9 (and breaks enough
things I wish not to support that in-house on anyone's machine but my
own). My upgrade path is now called into question on this stack
issue--I was thinking of using RHEL4, but I'm building FC3 prototypes.
Is there a light at the end of this tunnel?
Is this just a 32-bit OS issue? Because I can ride it out until we
migrate entirely to 64-bit platforms again (Indigo2 and Octane were so
far ahead of their time, huh...)
Will XFS change to conform with the OS, or must we start using a new
vendor distro when we upgrade from RH9? We hammer our machines pretty
good--using up all the memory and swap on a pretty regular basis, and
sometimes treat workstations as NFS servers (well, I don't, but users do).
I read all the other posts, but the general concensus so far is "try it,
see how it goes" or "seems fine to me." I'm worried it's gonna wreck us
sooner or later.
Russell Cattelan wrote:
Robin Humble wrote:
Under what conditions should we be worried about XFS with a kernel
compiled with 4k stacks?
XFS is probably fine but I wouldn't try to stack on top a volume manager.
The last time XFS on a md raid 5 volume the system would last
about 30 seconds before spewing all sorts of weird stack traces.
And the lovely folks doing the FC kernel not only decided to
patch out the option to turn off 4k stacks they also change some
interfaces that rely the "remove-any-option-of-sane-stack" patch.
At that point I formatted the drive and switch to box to a different
Ohh I did try a stock kernel on the box, but apparently the FC pthreads
libraries do not work with vanilla linux kernels. (That may have
changed since the last time I tried that)
As we are lazy, we would prefer to run a stock fc3 kernel on dual Xeon
32bit with 3ware hardware SATA RAID - mostly accessed via NFS. No LVM,
no ACLs, no software RAID, so nothing too fancy.
Is this still asking for trouble?
If so, then it's not hard to recompile the fc3 kernel (or a stock
kernel) for 8k stacks, just we're not too sure how risky the default
fc3 setup is.
We're also thinking of running RHEL AS4 instead of fc3 (they are very
similar), but that definitely needs a recompile as XFS isn't included in
the default AS4 kernel :-/