On Sat, Sep 20, 2008 at 11:41 PM, Gordon, JC (John) <[log in to unmask]> wrote:
> This is an obvious thing to do if you have a central GGUS that is
> supposed to be the only helpdesk in the world. This isn't the case. The
> EGEE model is (for all/most other countries) that GGUS is a switch that
> passes tickets on to ROCs who then handle them. It is one of the few
> things that EGEE currently asks a ROC to do and we failed. Actually the
> automated handling of COD tickets automatically to pass tickets on to
> sites was originally developed by the UK with the COD. This stopped
> working when the mapping between GGUS and Footprints broke. It needs
> fixing.
Ticket title: "Problem with TGIC user's jobs at UKI-SCOTGRID-GLASGOW".
Assign that one automatically, please, without knowing if it's the
site contacting the VO or vice versa. The person who raise the ticket
knows, the TPM might have managed to guess, but a system based on
parsing the title of the ticket is not helpful in this case.
Multi-step ticketing systems are a complex disaster, if you ask me.
>
> GGUS may implement something to let the user select a site but they may
> not have enough spare effort to fix things for the UK.
If GGUS worked properly the UK ROC would need less fixing.
>
> It is one thing for an expert like you Graham to say the fault is with
> the site but do sites want thousands of users deciding 'my job failed,
> it ran at Glasgow, the fault is obviously Glasgow's, assign the ticket
> to them'?
The thing which is broken, John, is not what happens to the naive
user's ticket, but what currently happens to the experts' tickets.
These frequently get held up for hours or days in the system -
misassignment by TPMs is one issue, "lost in the ROC" is another one.
(I note in passing that GGUS is used mainly by experts. There were 88
tickets in the last week, 59 were raised by ATLAS shifters, 25 by COD.
Why are we trying to solve problems which do not exist, instead of
ones that do?)
It's a constant battle inside ATLAS to encourage people to use GGUS[1]
and not to email their favourite site admin (though often I put them
in CC). The reason that we persist with GGUS is mainly that there is a
ticket with an audit trail, though even this is quite broken because
we cannot raise team tickets against T2s, only against the T1s. So by
the time a site thinks the issue is resolved there may be someone else
on shift. That's why we're trying hard to get GGUS fixed by being able
to raise team tickets, and automatically assign them to all GOC sites.
(Another aside - we have a very good relationship with ATLAS sites and
many T2s are great at responding to and fixing problems out of hours;
they do far better than the MoU _if_ they know there is a problem,
which with a GGUS ticket they by and large don't :-).
> The TPM is supposed to introduce this quality. If they are
> failing it is worth raising this at EGEE08 this week.
84/88 tickets last week did not need the TPM. Then add the time that
EGEE, Jeremy, T2Cs, you, I, etc. spend debating GGUS and we're taking
days and days of human time, which is very expensive, we're wasting
it.
cheers
Graeme
PS. US ATLAS support assigning tickets to sites directly using RT
queues, which I imagine one of our RAL T1 people, like Derek, could
set up in a day. Might not be perfect, but it does cover the principal
use cases almost perfectly. The perennial problem, with the grid, is
the insistence on developing, from scratch, baroque solutions because
current software is inadequate. Then we wait for 2 years and get
something that's imperfect, in an almost orthogonal way, to the
original software tools we could have used for almost nothing. Sorry
for being grumpy but the first cup of coffee has not yet quite reached
the blood stream...
[1] Believe me, I have put a lot of effort into getting ADC shifters
to use GGUS.
--
Dr Graeme Stewart http://www.physics.gla.ac.uk/~graeme/
Department of Physics and Astronomy, University of Glasgow, Scotland
|