[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Processes stuck in D state..



On Thu, 2003-08-07 at 01:13, Gordon Henderson wrote:
> On Wed, 6 Aug 2003, Keith Owens wrote:
> 
> > Do not blame XFS for this problem, it is almost certainly an NFS
> > problem.
> 
> Maybe my original email was unclear, but this is an NFS server, not a
> client. Why would an NFS exported filesystem hang xfsdump? (or even du -k
> ?)
> ie.
> 
>  1149 ?        D      2:21 xfsdump -F -J -l 1 - /dev/md4
> 
> This xfsdump is trying to dump /dev/md4 which is obviously local to the
> server and it's stuck. Why is it stuck? /dev/md4 itself doesn't appear to
> be stuck, as I can access files on it both locally and via NFS mounts on
> other Linux boxes (and samba mounts on Win boxes)
> 
> BTW: It's a Debian server (3.0 with security updates, I downloaded and
> compiled xfsprogs and xfsdump rather than use the supplied Debian ones)
> It's running a stock 2.4.21 kernel patched with Alan Coxes ac4 patches
> which include XFS. I've heard that these patches are "old" though. Right
> now I have lots more processes stuck in D state so I'll have to reboot it
> in the morning.

I'm running two BIG NFS Servers with xfs only (Debian 3.0, too + recent
xfsprogs...). I'm using them since 2.4.19-cvs, and never ever had stuck
D states on the Server. But I've been bitten by stuck D's on client side
and workstations even w/o nfs with older kernels (Until March's cvs). 
Now even those seem to be gone.

These Servers have hardware RAID rather than Linux MD, so maybe your
probs are related to MD + any scsi/ide drivers?!?

Give 2.4.21+ xfs-1.3pre a try. (To fix security issues found in 2.4.21,
you might want to try debian unstable's kernel-patch-debian (2.4.21-4),
but backout the O_DIRECT changes they made in this Release - they won't
play nicely with XFS)

Christian