Hi Dave,
Thanks for the response.
No I/O errors reported in the message log or on the RAID box.
It's an Infortrend SATA RAID5 array, with a fibre channel connection to
the server.
The filesystem is build on an LVM volume.
Kernel is 2.6.13-15-smp running on an x86_64 dual CPU Xeon server with
hyper-threading enabled.
The most significant feature of the load is that it is part of an HPC
cluster, and has a large number of nodes NFS mounting the filesystem
across Gigabit ethernet.
I did notice that in the first incident, a user had a directory with
700000 files in it, and xfs_repair found fault with that directory. The
user has revised their workflow since and removed the files.
Very difficult to spot common traits in the workload between the 2
incidents.
Cheers,
Simon
> -----Original Message-----
> From: David Chinner [mailto:dgc@xxxxxxx]
> Sent: 19 April 2007 15:18
> To: Burbidge, Simon A
> Cc: xfs@xxxxxxxxxxx
> Subject: Re: XFS internal error XFS_WANT_CORRUPTED_GOTO
>
> On Thu, Apr 19, 2007 at 02:28:24PM +0100, Burbidge, Simon A wrote:
> >
> > Hi,
> >
> > We've had a couple of occurrnces of xfs shutdowns on one of our
> > fileservers.
> > The latest had the message:
> >
> > Apr 19 10:35:00 fs3 kernel: XFS internal error
> XFS_WANT_CORRUPTED_GOTO
> > at line 1745 of file fs/xfs/xfs_alloc.c. Caller 0xffffffff8819bc7c
> > Apr 19 10:35:00 fs3 kernel:
> > Apr 19 10:35:00 fs3 kernel: Call
> > Trace:<ffffffff8819a399>{:xfs:xfs_free_ag_extent+1449}
> > <ffffffff8819bc7c>{:xfs:xfs_free_extent+188}
>
> So you've got a corrupted freespace btree. What is the filesystem
> hosted on - a normal block device, iscsi, nbd? What kernel?
>
> Are there any I/O errors in the log?
>
> What we you running at the time of the shutdowns? Anything
> common between the occurrences?
>
> Cheers,
>
> Dave.
> --
> Dave Chinner
> Principal Engineer
> SGI Australian Software Group
>
|