>
> Hi Gabe -
>
> A couple things - would it be possible to share a code snippet that
> represents what you are doing on the client side? It's always easier to
> track these things down if we can recreate it locally.
>
> Also, in what manner are the files corrupted?
>
> Finally, what other posts have you seen referring to this? That might
> help pull it all together...
>
> Thanks,
>
> -Eric
>
Also what type of machines are the clients? Does the problem still occur if
you mount with sync option? When you say you work around the problem by
reading all the buffer into memory and writing it at once, is this done on
the client? If so, I think it sounds more like a cache consistency problem
on the clients.
Thanks,
Tad
> On Tue, 2001-10-16 at 16:42, Gabe E. Nydick wrote:
> > I run a 2.4.5 XFS 1.0.1 kernel with all of the patches and updates user-land
> > programs from SGI and I am experiencing a problem with NFS. My web servers
> > are nfs clients of my XFS 1.0.1 file server and when at least mildly loaded,
> > the web servers write uploaded files corruptly. These files are between 512K
> > and 1MB in size and being uploaded through CGI where they are read 4k at a
> > time and written. Since the web servers are at least slightly loaded, the
> > reading of the buffer and writing of the file are not continuous and I get
> > corrupt files. If I change my scheme to read all of the buffer into memory
> > and then write all of the file at once, I do not get corruption. This leads
> > me to believe that XFS over NFS is having issues. I have also seen posts in
> > the internet with similar problems. I can recreate this corruption at will.
>
> --
> Eric Sandeen XFS for Linux http://oss.sgi.com/projects/xfs
> sandeen@xxxxxxx SGI, Inc.
>
|