I agree about the cause being due to the number of requests to torque.
Various programs are doing this (maui, gip, jobmanager). Our problems
are correlated with large load which upsets torque and I was thinking
this load could be elevated by the slow gip script (among other things).
The biggest disruption occurs when users startup huge numbers of
globus-job-manager, usually when using condor to submit. Apparently this
was fixed in condor.
I wonder how your maui probs are relate to CE load?
Peter
ps. the thread I referred to is on TB-SUPPORT, not rollout
Alessandra Forti ([log in to unmask]) wrote:
> mmmhh I don't think that is the problem. If a badly written parsing
> routing can disrupt maui maybe we should change scheduler.
>
> I haven't had time to look at it but I have the feeling it is connected
> to the number of requests users send to the
> pbs /maui server. I'm dumping my IS every minute and the problem is not
> detected by gstat because it has a larger interval, but 4444 values
> appear all the time. Sometimes for long periods but sometimes for only 1
> or 2 minutes.
>
> cheers
> alessandra
>
>
>
>
> Peter Love wrote:
> >I made these changes too, looks like there is no improvement. However,
> >the gip changes may be something to look at (Mona's thread on Rollout).
> >I'll wait until the new script matures from alpha to beta :-)
> >
> >Peter
> >
> >Alessandra Forti ([log in to unmask]) wrote:
> >
> >>Hello,
> >>
> >>this might help those who are publisihng randomly 4444 peaks of waiting
> >>jobs because maui is not responding. I've just modified my maui cfg so I
> >>hope it'll work.
> >>
> >>cheers
> >>alessandra
> >>
> >>--
> >>*******************************************
> >>* Dr Alessandra Forti *
> >>* Technical Coordinator - NorthGrid Tier2 *
> >>* http://www.hep.man.ac.uk/u/aforti *
> >>*******************************************
> >>
> >>
> >
> >
> >>Date: Thu, 27 Apr 2006 17:42:09 +0200
> >>From: Jeff Templon <[log in to unmask]>
> >>Reply-To: LHC Computer Grid - Rollout
> >><[log in to unmask]>
> >>To: [log in to unmask]
> >>Subject: [LCG-ROLLOUT] on Maui freakouts
> >>User-Agent: Thunderbird 1.5.0.2 (Macintosh/20060308)
> >>
> >>Hi *
> >>
> >>I know many of us are having periodic Maui freakouts, where for some
> >>reason Maui hangs for awhile or just dies; this phenomenon is made clear
> >>now since it is reflected in the published ERT.
> >>
> >>This has been really irritating for me the last few days, as I have been
> >>trying to get fair sharing in shape on our cluster, and of course a
> >>catatonic Maui doesn't really help things.
> >>
> >>I did some googling, perhaps more agressively than usual, and came up
> >>with the following link:
> >>
> >>http://www.clusterresources.com/pipermail/mauiusers/2005-August/001669.html
> >>
> >>the error they talk about here:
> >>
> >> ERROR: cannot get node info: NULL
> >>
> >>is often what I see right before Maui hangs for 15 minutes. They claim
> >>that the problem is a disagreement between Maui and Torque about certain
> >>timing issues, which is plausible given what I see.
> >>
> >>I set the following two things in maui.cfg:
> >>
> >>RMCFG[0] TIMEOUT=90
> >>NODESYNCTIME 0:00:30
> >>
> >>and it's too early to be conclusive, but I haven't seen the Maui lockup
> >>since doing so and restarting maui. Will keep you posted, interested
> >>whether this solves others' problems as well!!
> >>
> >> J "no present like the time" T
> >>
>
> --
> *******************************************
> * Dr Alessandra Forti *
> * Technical Coordinator - NorthGrid Tier2 *
> * http://www.hep.man.ac.uk/u/aforti *
> *******************************************
|