[OpenAFS] Files getting corrupted from some clients

Alex Rolfe arolfe@MIT.EDU
Fri, 13 Feb 2009 09:48:14 -0500

Have you tried doing similar operations (eg, unpacking linux kernel
source) just on the local disk of the clients from which you see
corruption?  The one time we saw something like this turned out to be a
linux kernel bug that only showed up when certain modules were loaded (I
think it was the raid0 module)- we noticed it as corruption in AFS but
it turned out to be easy to replicate with only the local file system.
AFS was suffering because files in the client's on-disk cache were
corrupted locally and then written back to the server.


Troy Telford <ttelford.groups@gmail.com> writes:

> And I'm running into issues with files being corrupted.
> On the exact same volume, for instance, I can have a host that works fine with
> AFS -
> * I'll take the Linux Kernel source, and unpack it.
> * I then rename the directory that gets unpacked.
> * I then unpack the source again.
> * There are no differences in the two directories.
> And then I have other hosts where I can unpack the tarball, and end up with
> files that are truncated, missing sections of code, etc.
> I know you're not supposed to use AFS for database-like things, but it seems
> that if I can't even unpack a tarball without corruption, there's some sort of
> problem that needs to be resolved.
> All of my systems are using OpenAFS 1.4.8.  The "problem" systems don't seem
> to follow any particular linux distribution - or kernel.  I have had two
> systems that have the same binary kernel, the same source code, and one
> produces file corruption, and the other does not.
> I'm new to AFS - and am more accustomed to NFS; so I'm not sure if I'm just
> doing something that AFS doesn't handle well, or if AFS varies that much from
> client to client.  I want to give it a chance, but getting corrupted files
> when a client writes to disk tends to put a damper on the whole experience.
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info