Alistair, my first impression is that it is disappointing that when there is work on improving the BDII that ATLAS is spinning off its own IS.
But on the issue I asked about, it sounds like you and others have an overview of the issue for the UK but each sysadmin is having to investigate whether they actually have a problem. My thought was that you could save the sysadmins a lot of work but I realise that it is difficult to know in advance which sort of tickets fall into this category. A good metric might be that if ATLAS (or anyone else) thinks a large number of sites have got something wrong then they should be very careful before sending out tickets that they are right.
In this case it seems surprising that the WN environment on a particular WN should be different for different CEs as prima facie the WN is set up once under the LRMS. I guess if different CEs point to different clusters then they can be different. Or does the CE, rather than the LRMS affect how the initial shell is called?
Thanks,
John
> -----Original Message-----
> From: Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of Alastair Dewhurst
> Sent: 09 December 2011 01:03
> To: [log in to unmask]
> Subject: Regarding GGUS tickets: "SITE_NAME env variable on WNs not defined
> for all the CEs"
>
> Hi
>
> John; I asked the question of what to do with multiple similar tickets
> before. Sites said that they would prefer tickets directly as its easier
> for them to track. Each site only gets one tickets, as long as the ticket
> is clear it shouldn't be necessary to cross reference other tickets.
> Certainly in this case the problem is minor.
>
> I will admit the wording of the message wasn't very good. I had asked
> Florentin to message me to check the wording before submitting them but he
> was in a rush to get things moving. The other problem seems to have been
> our test to identify which sites were broken has brought up some false
> positives. Making a sweeping generalisation these appear to be the less
> reliable sites, so possibly our test was failing for other reasons.
>
> To give a little bit more explanation as to what ATLAS want:
> ATLAS are moving to using AGIS (ATLAS Grid Information System). To use
> this a job only needs to know what site it is at. It was decided that the
> best thing to use was the SITE_NAME variable which should be defined on all
> WN. atlassgm jobs are then submitted via WMS. They will use the SITE_NAME
> to query AGIS and configure the local site settings. The first system we
> are moving over to using AGIS in full production is the Frontier
> configuration which we hope to do in late January (Which is why I am
> involved in this as I am ATLAS Frontier coordinator). The tickets should
> all have an indication of what jobs fail the test. In alot of case it was
> submissions to only one particular CE that failed.
>
> Alastair
>
>
>
> On 8 Dec 2011, at 23:12, John Gordon wrote:
>
> > Is everyone happy that ATLAS floods lots of sites with tickets like this?
> Or would it be better to ask them to send a list to the NGI and then you
> could handle them as a team. I suspect that if one person got to the bottom
> of the issue there would be far less effort expended in total.
> >
> > John
> >
> >
> ***************************************************************************
> ******
> > This is an automated mail. When replying don't change the subject line!
> > Type your text above this box and S T R I P P R E V I O U S M A I L S
> please!!
> >
> ***************************************************************************
> ******
> >
> > GGUS ticket #77186 was updated.
> >
> > REFERENCE LINK: https://ggus.eu/ws/ticket_info.php?ticket=77186
> > SUBJECT: SITE_NAME env variable on WNs not defined for all the CEs
> >
> > =====================
> > LATEST MODIFICATIONS:
> >
> > LAST MODIFIER -> Samuel Skipsey
> >
> >
> > PUBLIC DIARY ->
> > (I should clarify: I didn't fix anything, I just checked them... and they
> are all okay.)
|