Dear All
On yesterday's agenda there was an item about ATLAS reports of problems
with information inconsistencies between VOView and the CE for some
sites. Specifically:
"Sites with these CEs should check their VOView publishing (not
consistent with CE info):
ce.epcc.ed.ac.uk
ce00.hep.ph.ic.ac.uk
epgce1.ph.bham.ac.uk
fal-pygrid-18.lancs.ac.uk
hep-ce.cx1.hpc.ic.ac.uk
heplnx207.pp.rl.ac.uk
lcgce01.phy.bris.ac.uk
mars-ce2.mars.lesc.doc.ic.ac.uk
pc90.hep.ucl.ac.uk
serv03.hep.phy.cam.ac.uk
svr016.gla.scotgrid.ac.u
kt2ce02.physics.ox.ac.uk:
Fuller explanation from ops meeting: "there is a mismatch between all
inclusive information published for a CE and information published in
the VOViews. As an example, the queue mentioned below supports only
ATLAS, therefore, the number of waiting jobs in the inclusive view
should be the same as the one for the ATLAS VoView. But it is not. The
VOView publishes all zeroes. Moreover, there are some queues where the
number of waiting jobs for all views do not add up to the total
published in the inclusive view. In total more than 130 ATLAS queues are
affected, among which almost all T1s. Since the WMS uses information in
the VOView and the latest one is generally the wrongly published one,
ATLAS is submitting jobs almost randomly with accumulation of jobs at
small sites. The issue is extremely severe."
I agreed to follow up and this is what I get in response from Jeff
Templon:
"Hi
I have been assigned the ticket. it's not a bug in the
dynamic-scheduler, it's a bug in the configuration from yaim, which is
halfway supporting VOViews with FQANs and halfway not.
I will answer the ticket, putting all relevant information there.
JT
http://savannah.cern.ch/bugs/?29922"
So please take a look at the ticket if you want immediate information on
this problem. Anyway since this is a bug with YAIM please don't spend
time trying to resolve the original request. Once a fix is known we'll
send out pointers.
Thanks,
Jeremy
|