Note it is not the full SAM tests that are being run in the job wrapper,
just a subset. The aim being that these tests are run as every VO so
they find instances where something works as ops but not as atlas.
When we raised this issue originally Piotr published details via the
weekly operations meeting of how to switch off the tests until the
problem was addressed. It wasn't planned that these test took 15 mins
and my recollection is that this was fixed within days.
John
> -----Original Message-----
> From: Testbed Support for GridPP member institutes
> [mailto:[log in to unmask]] On Behalf Of Coles, J (Jeremy)
> Sent: 18 May 2007 10:27
> To: [log in to unmask]
> Subject: Re: SAME tests run in job wrapper
>
> Hi Stephen
>
> Graeme's response gives most of the details. (Note that the
> "E" of SAME was dropped some time ago so we are talking about
> wrapping the SAM tests with each job. I know there are some
> on this list who think doing this is madness).
>
> The one thing missing in the response is whether there was
> input from sites and I'm a little uncertain where it was
> first discussed. I do not remember gathering any feedback. I
> could not find much in the ops meeting lists but it
> apparently came up in the 19th December ROC manager's meeting
> via a talk by Piotr:
> https://edms.cern.ch/document/808821. As Steve Fisher has
> pointed out though, the PPS testing did not reveal any
> problem which is why it went into production.
>
> The need to deploy new releases on only a subset of
> production sites before widely adopting them is becoming
> increasingly apparent (as some of you will know if you've
> followed Yves's recent ticket
> https://gus.fzk.de/pages/ticket_details.php?ticket=21815)
>
> Jeremy
>
>
>
>
>
> > -----Original Message-----
> > From: Testbed Support for GridPP member institutes [mailto:TB-
> > [log in to unmask]] On Behalf Of Graeme Stewart
> > Sent: 18 May 2007 09:54
> > To: [log in to unmask]
> > Subject: Re: SAME tests run in job wrapper
> >
> > On 18 May 2007, at 09:40, Stephen Childs wrote:
> >
> > > Graeme Stewart wrote:
> > >> Yes, I've noticed this at Glasgow - and in particular R-GMA is
> > >> adding 5-15 minutes of wallclock time to every job, which is a
> > >> terrible waste of resources (particularly for our GRAM GT2 user
> > >> groups, who run some pretty short jobs).
> > >> Something I really need to look into...
> > >
> > > I've just remembered about this (as I'm trying to do simple tests
> > > using globus-job-run which aren't simple any more).
> Graeme, did you
> > > ever find out where this was being done in order to disable it?
> >
> > Yes. Have a look at Alessandra's GGUS ticket:
> >
> > https://gus.fzk.de/ws/ticket_info.php?ticket=20364&from=search
> >
> > and
> >
> > http://scotgrid.blogspot.com/2007/04/disabling-samer-gma-in-job-
> > wrapper.html
> >
> > Should be a cinch to do in Quattor as well.
> >
> > >
> > > Jeremy, could you comment on when this was introduced and whether
> > > there was input from sites on it?
> >
> > There was some discussion and feedback involving Steve
> Fisher in PMB
> > Minutes 254 (16th April 2007):
> >
> > http://www.gridpp.ac.uk/pmb/minutes/070416.txt
> >
> > Cheers
> >
> > Graeme
> >
> > --
> > Dr Graeme Stewart -
> http://wiki.gridpp.ac.uk/wiki/User:Graeme_stewart
> > ScotGrid - http://www.scotgrid.ac.uk/ http://scotgrid.blogspot.com/
>
|