[OpenAFS] 'amd64_linux26' is being replaced by '@sys'

[Jeff Blaine]

Hmm.  Very odd.

% mv amd64_linux26 amd64_linux26_realdir
% ln -s ./amd64_linux26_realdir ./amd64_linux26
% vos release root.apps
[ examined /afs/rcf/apps to be sure the new structure appears ]

% export MATLAB=/afs/rcf/apps/Matlab/amd64_linux26/R2009b
% sh -x $MATLAB/bin/matlab
...
+ newdir=/afs/rcf/apps/Matlab/amd64_linux26/R2009b/bin/
+ '[' '!' /afs/rcf/apps/Matlab/amd64_linux26/R2009b/bin/ ']'
+ '[' 0 -ne 0 ']'
+ cd /afs/rcf/apps/Matlab/amd64_linux26/R2009b/bin/
++ /bin/pwd
+ newdir=/afs/rcf.our.org/apps/Matlab/@sys/R2009b/bin
...

However:

% cd /afs/rcf/apps/Matlab/amd64_linux26/R2009b
% /bin/pwd
/afs/rcf.our.org/apps/Matlab/amd64_linux26/R2009b
%

On 5/24/2010 12:27 PM, Derrick Brashear wrote:
> Well... hm.
> Can I suggest making the object @sys points to be a symlink to the
> real directory, instead of the real directory?
>
> That should result in the right thing happening, albeit still wretched.
>
> On Mon, May 24, 2010 at 12:20 PM, Jeff Blaine<jblaine@kickflop.net>  wrote:
>> Oh to be so lucky.
>>
>> Matlab barfs all over '@sys' refs.  I'll tell our users
>> they may get lucky :)
>>
>> On 5/24/2010 12:12 PM, Derrick Brashear wrote:
>>>
>>> On Mon, May 24, 2010 at 12:01 PM, Jeff Blaine<jblaine@kickflop.net>
>>>   wrote:
>>>>
>>>> We're confused here.
>>>>
>>>> RHELv5 amd64_linux26 + OpenAFS 1.4.6
>>>>
>>>> % pwd
>>>> /afs/rcf/apps/Matlab/amd64_linux26/current/bin
>>>> % /bin/pwd
>>>> /afs/rcf.our.org/apps/Matlab/@sys/R2009b/bin
>>>> %
>>>>
>>>> This does not happen on the following other hosts we've tested:
>>>>
>>>>     RHELv5 amd64_linux26 with OpenAFS 1.4.7
>>>>     RHELv5 i386_linux26 with OpenAFS 1.4.11
>>>>
>>>> How do we get the AFS client to stop converting 'amd64_linux26'
>>>> to '@sys' ?
>>>
>>> It's not the AFS client. It's the fact that you have more than one
>>> path to the same vnode; Someone references it via @sys, and the kernel
>>> caches the dentry with the path @sys. You get one dentry per inode.
>>> Unless we find a way to implement dentries that effectively point to
>>> alias inodes, and we haven't yet, that can happen.
>>>
>>> It's not new. It's just "luck".
>>>
>>>> Is this something changed post-1.4.6?
>>>
>>> _______________________________________________
>>> OpenAFS-info mailing list
>>> OpenAFS-info@openafs.org
>>> https://lists.openafs.org/mailman/listinfo/openafs-info
>>>
>> _______________________________________________
>> OpenAFS-info mailing list
>> OpenAFS-info@openafs.org
>> https://lists.openafs.org/mailman/listinfo/openafs-info
>>
>
>
>