Hi there,
I've found a few problems in the way we deal with out-of-space
conditions in XFS at the moment - probably the problem you are
hitting can be fixed with the attached patch. It resolves one
class of deadlock, I'm still looking at one more which is alot
more difficult to trip though (as it requires multiple parallel
writers under low free space conditions).
Let me know if this helps, please. If not, that kdb output I
asked for earlier would still be useful.
thanks.
On Tue, Oct 21, 2003 at 12:26:53PM -0500, Austin Gonyou wrote:
> I'm kind of curious if the disk is 100% or just near, i.e. 1mb to go, or
> even 10K free, etc. Look back in this list. I have a message around here
> about the same thing, from several revisions ago.
>
> On Tue, 2003-10-21 at 12:11, Libor Vanek wrote:
> > Hi,
> > we're using RH9 with XFS fs with all kinds of kernel (2.4.20 + XFS 1.2
> > and also 2.4.22 + XFS 1.3).
> >
> > When ANY mount point using XFS (/, /raid etc...) gets full, complete
> > system halt and needs to be rebooted - is this bug or feature? :)
> >
> > Best regards,
> > Libor
> --
> Austin Gonyou <austin@xxxxxxxxxxxxxxx>
> Coremetrics, Inc.
>
>
--
Nathan
iomap_rejigged.patch
Description: Text document
|