[OpenAFS-devel] Strange hangs in openafs 1.4.1 linux 2.6.17.7

Jerry Lundström jerry.lundstrom@it.su.se
Tue, 12 Sep 2006 09:57:19 +0200


Jeffrey Altman wrote:
> Jerry Lundström wrote:
>> Jeffrey Altman wrote:
>>> This is fixed in 1.4.2 beta 1.
>> Running linux 2.6.17.11 and openafs 1.4.2rc1 now and still getting han=
gs.
>>
>> Is this problem a client only problem or has to do with the servers al=
so?
> 
> This is a server problem.

I'm very unsure about this being a server problem.

Think I need to give some background to what we running on so here goes.

We are running a small installation script that uses themis to install
new servers/clients on a netboot image that itself was created by themis
based on the same dist that we install.

Currently the dist is lunar linux (www.lunar-linux.org) 1.6 running
2.6.17.11 with gcc 3.4.6 and glibc 2.3.6 with nptl threads.

The netboot image contains all but /usr and /var basiclly (and ofc
/usr/vice) and is booted with pxelinux. The /usr and /var directory is
symlinked into afs and afs is started right after the network is up.

We have been using this way of installing some time now and we have had
no such problems with the old dist (lunar linux 1.4 kernel 2.4.29 with
openafs 1.2.13.

Now using the current netboot (2.6.17.11 openafs 1.4.2rc1) I have tried
to change the cache in any way I could.
Using -memcache with no -blocks option and with -blocks 16000, using
$SMALL/$MEDIUM/$LARGE with a filebased cache running on the ramdisk with
a size of 8meg/16meg/200meg and using a partition of the drive as cache
of 16meg.

Using memcache it hangs from time to time, using filecache on the
ramdisk it can hang within seconds but every single test when the cache
was resided on a drive it was a success... I even ran 2 installations at
the same time without problems.

Rewriting the installer to make afs use cache on a drive isnt that hard
but we use this netboot for diskless computers also and we would like
that to work also...

If you need any other information or the image itself just ask...

-- 
Jerry Lundström, System Developer
The Division of IT and media, Stockholm University, Sweden
+46 (0)8 16 19 99 / http://www.it.su.se