Hi all,
On the topic of testing things like glexec/ARGUS... Has anyone tried
setting up ARGUS in a redundant manner (if it's even possible)?
Clearly it could easily become a single point of failure if you don't
have more than one of them...
Regards,
Simon
On Wednesday 23 March 2011 15:47:46 Daniela Bauer wrote:
> Personally I would have waited with the sermon until the beginning
> of July to see how many sites actually haven't deployed it.
>
> As for the cream-ce, my main customer is CMS who only started
> using it 6 months after I deployed it (after a 46 long email
> exchange with the cream developers, to make this 'production'
> level software work), until very recently it didn't count
> towards site availability and there was much screaming and
> shouting back then, so oddly enough, the second time around I see
> the whole thing a lot more relaxed. Boy who cried wolf anyone ?
>
> Cheers,
>
> Daniela (high on Orangina)
>
> On 23 March 2011 15:31, John Gordon <[log in to unmask]> wrote:
> > There have been presentations at the GDB every few months for
> > several years reporting on various aspects of multi-user pilot
> > jobs such as: security review of glexec; pilot jobs
> > frameworks,; site questionnaires, site tests; experiment pilots.
> > Throughout this, glexec has been the one constant. I don’t know
> > how it could have been clearer.
> >
> >
> >
> > I think sites should always be looking into potential new
> > services that they might be asked to use just to reassure
> > themselves that it won’t cause particular problems for their
> > particular circumstances. This needn’t mean installing and
> > trying it all yourself but it should include reassuring yourself
> > that at least someone else has tested your setup. This is just
> > rudimentary covering your a***, aka project management. GridPP
> > has been good in the past in delegating such testing so that
> > there has usually been someone in the UK who gets experience
> > first and can help others. This breaks down for components
> > integrated with batch systems. Since EDG days it has been the
> > rule that developers of things that interact with batch do their
> > main development for one batch system and there are volunteers
> > to deploy and test on other batch systems. CEs, info providers,
> > apel-parser all needed such support from sites running
> > particular batch systems. This is a hot topic in EMI right now
> > as they are trying to ensure coverage of all relevant batch
> > systems for several components. Some sites who did this work
> > on a best-efforts basis don’t want all the overheads of EMI
> > build/test/release infrastructure.
> >
> >
> >
> > I have no problems with sites who follow a wait-and-see strategy
> > as long as they are confident that they are agile enough to do
> > things in a hurry when they need to. What is not good are sites
> > who do nothing for a long time and then at the last minute when
> > they are set a deadline say ‘ I can’t possibly do anything for
> > the next six months as I didn’t know this was coming’. As a
> > tactic it might work once but when we hear ‘the CREAM-CE will
> > never happen, the experiments don’t want SL5, I met a man in the
> > pub who told me that Argus is buggy, I never knew about
> > glite-APEL, glexec is unclear’ then a pattern emerges. The
> > message is that most of these changes DO happen (glite-CE was
> > the exception, not the rule) and if you plan for them life is
> > easier for you than playing catchup. </sermon>
> >
> >
> >
> > Happy to buy the first round if this discussion continues in
> > Hove.
> >
> >
> >
> > John
> >
> >
> >
> > From: Testbed Support for GridPP member institutes
> > [mailto:[log in to unmask]] On Behalf Of Wahid Bhimji
> > Sent: 23 March 2011 13:04
> > To: [log in to unmask]
> > Subject: Re: glexec in London
> >
> >
> >
> >
> >
> > Hello
> >
> >
> >
> > On 23 Mar 2011, at 12:35, Stephen Burke wrote:
> >
> > Testbed Support for GridPP member institutes [mailto:TB-
> >
> > [log in to unmask]] On Behalf Of Wahid Bhimji said:
> >
> > or else there should be a (quite long) bedding
> >
> > in period where it is "required" but sites won't get their
> > income
> >
> > steam cut off if it isn't working.
> >
> > Well, glexec has been around for several years so there has been
> > plenty of time for sites to test it if they wanted to ... in
> > practice it seems that most sites don't install things until
> > they are made mandatory.
> >
> >
> >
> > well: as others have now mentioned: it is has not been clear to
> > me what stage glexec was in during that time and whether it
> > really was for use. Also it was not clear that experiments
> > wanted it to be used. Even now those are not clear. The
> > bedding-in period should start from when it is really ready.
> >
> > Also could someone point me to the documentation for glexec?
> > googling
> >
> > takes me to a nikhef twiki that looks like it might be out of
> > date.
> >
> > The NIKHEF twiki should be right; I find
> >
> > https://www.nikhef.nl/pub/projects/grid/gridwiki/index.php/GLExec
> >
> > which was last modified last June.
> >
> >
> >
> > That page refers to : The latest stable version is gLExec
> > 0.7.0-2.
> >
> > If it is felt the product is ready for mandatory installation
> > everywhere then it would fill me with more confidence if they
> > released version 1.X
> >
> >
> >
> > Wahid
> >
> >
> >
> >
> >
> >
> >
> > Stephen
|