[OpenAFS] Crash testing OpenAFS

ted creedon tcreedon@easystreet.com
Mon, 22 Aug 2005 10:56:41 -0700

Using kernel (SuSE 9.3 + 1.3.87) - something was hanging Linux
in the middle of an rx packet.

Its working better now - I changed the cache from ext2 to ext3. fsck on the
cache was not allowing a clean reboot and am not using /vicepb., So far
moving about 100GB is going smoothly.
I'm running X and everything else imagineable.
Using a disk cache 10GB. Both dynroot and not.
KRB5 with aklog (the SuSE aklog needs to be replaced since it doesn't use
Transace paths).
Apparently few users have SuSE installed.

Am compiling 1.4.0-rc1 and will re-test on another server.


-----Original Message-----
From: openafs-info-admin@openafs.org [mailto:openafs-info-admin@openafs.org]
On Behalf Of Blake Atkins
Sent: Monday, August 22, 2005 8:43 AM
To: openafs-info@openafs.org
Subject: Re: [OpenAFS] Crash testing OpenAFS


Are you having troubles with the OpenAFS linux client?  Are you using a
memory cache?  Are you using kernel 2.6?  I have been unable to reliably use
OpenAFS client 1.3.X + Kernel 2.6 (SuSE SLES and Pro 9 - 9.3) + Memory
Cache.  I assumed I was the only person with these troubles as I would
expect more people to be complaining if it was widespread.  When using the
above combination, I would see cache corruption and occasional crashes on
the client workstations.  Switching to a disk cache eliminated all issues.
I would be interested in knowing if anyone else is seeing the same problem.



On Friday 12 August 2005 12:07 am, ted creedon wrote:
> It seems that AFS can be made to crash pretty much at will with fairly 
> simple exercises.
> Being a file system, one would expect rock solid performance.Not the case.
> A simple cp -rpv /afs/.bigcell/foo /afs/.home-ted-doris.fam/bar hangs 
> the system so badly Linux won't even "halt".
> If there's a problem with windows file names, (Jeff has stated so) it 
> needs to be documented. And fixed.
> Those interested in stress testing (or even regression testing) AFS 
> contact me directly.
> There needs to be a set of tests run on every release before its 
> thrown over the threshold.
> No flames please.
> tedc

OpenAFS-info mailing list