Hi David,
mmhh the problem is not the standard, the points I was trying to make were:
a) glite WMS job submission is not compatible with lcg CEs (and it was a
requirement) and requires to setup glite variables and scripts on the
WN. I've already found 3 glite setenv scripts on my WNs... 2 partially
ovverriding each other. This is an additional one. EDG variables
together with LCG ones appear in one of the two. So I'm not sure why in
a global glite_setenv.sh they can appear and not in the job wrapper.
Possibly because removing them wuold break the system? IMO that variable
should have been kept working as long as a mixed system is around. I
haven't seen torque people renaming pbs_mom to torque_mom not moving
/var/spool/pbs to /var/spool/torque by default.
b) For this reason I thought maybe it is better to try to avoid the
middleware setup of glite_local_customization_dir and use cd $TMPDIR in
the pbs_mom prologue but it doesn't work as I explained in the last
email I sent yesterday. My fault for talking before trying with pbs
only. IC people told me that SGE can (lucky them) but torque can't.
b') I will replace EDG_WL_SCRATCH with TMPDIR in cp_1.sh because it has
it's advantages.
c) Nobody told anything to site admins about this change of behaviour
and they found themselves with /home filled by WMS submitted jobs,
despite the fact they were correctly setup.
I hope I made it more clear.
cheers
alessandra
David Groep wrote:
> Hi all,
>
> If I have understood the thread correctly, and having read the comments
> of Marco Cecchi in the report for bug 18325
> https://savannah.cern.ch/bugs/?func=detailitem&item_id=18325
>
> the gLite WMS job wrapper is correct and compliant with all standards
> and regulations by *NOT SETTING* the old EDG_WL_SCRATCH environment
> variable in the default job environment.
> So, for this time, gLite *is* compliant with all relevant standards and
> guidance, and is not to blaim in any way.
>
> If there are some specific applications out there that rely on obsolete
> behaviour, it is an application problem and the application should
> handle the presence of absence of any particular environment variable.
> Let's place the problem in the proper place, and not mess with a
> now-compliant gLite software stack that is *rightly* agnostic to any
> EDG_WL_SCRATCH stuff.
>
> PS: and don't ship a distribution with any site customizations that
> re-introduce such obsolete behaviour as a default... So NO default
> RPM in the set with the cp_1.sh script inside it!
>
> Cheers,
> DavidG.
>
>
> Jeff Templon wrote:
>> I suppose we could always do
>>
>> export EDG_WL_SCRATCH=$TMPDIR
>>
>> I am disappointed that a project like gLite that claims to be
>> interested in standards, invents its own variable name (or rather,
>> steals the name of a related but deceased project) for something that
>> already exists as a POSIX standard, and as a HEPiX proposed standard,
>> and for which support already exists in Torque, the reference LRMS
>> platform for the project.
>>
>> JT
>>
>> Roberto Santinelli wrote:
>>
>>> Hi all,
>>> you might find at http://santinel.home.cern.ch/santinel/glite.xml
>>>
>>> some preliminary results from a second round of tests over the grid
>>> for testing how the variable EDG_WL_SCRATCH is taken into account if
>>> the job gets submitted through gLite WMS. After my first broadcast
>>> some sites applied in their own the suggestion I pointed out and
>>> since then, they seem to behave as expected. This is a proof that the
>>> recipe works!
>>> So far the following sites passed this trivial test (as you can see).
>>> vr016.gla.scotgrid.ac.uk
>>> ce1-egee.srce.hr
>>> g03n02.pdc.kth.se
>>> grid6.mif.vu.lt
>>>
>>> I hope that soon all of involved sites will be able handle correctly
>>> this EDG_WL_SCRATCH variable.
>>>
>>> LHCb will restart soon to submit through gLite WMS.
>>>
>>> Thanks for your comprehension
>>> R.
>>>
>>> ************************************************ Roberto Santinelli
>>> - Ph.D
>>> EUROPEAN LABORATORY FOR PARTICLE PHYSICS IT Division - PSS-ED/EIS
>>> Team Building: 28 Office: R-019
>>> Phone: +41 22 767 1925 Mobile: +41 76 487 0443 (int. 16-0433)
>>> Fax1: +41 22 766 9214
>>> Fax2: +41 22 767 4900
>>> Email: [log in to unmask]
>>> ICQ: 257963601
>>> Web: http://santinel.web.cern.ch/santinel/
>>> ************************************************
>
>
--
Alessandra Forti
NorthGrid Technical Coordinator
University of Manchester
|