[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Is this a negotiator bug or what???



Hi

Erik Erlandson wrote:
Hi Joe,

what version are you running?

I'm running:

$CondorVersion: 7.4.3 Aug  4 2010 BuildID: 261829 $
$CondorPlatform: X86_64-LINUX_RHEL5 $


It would be interesting to see which groups your jobs are being assigned
to (recent versions will output a warning if the job is defaulting to a
different group than implied by the submitter name)

I don't see it saying anything like that but the debug output is rather large so I could be missing it.


I've been fighting with this for a long time. Occasionally one of our groups will manage to suck up all our slots even though they're over quota. Most of the time they appear to work.

This can happen if a bunch of jobs land on an otherwise-empty system,
and the group in question has autoregroup enabled.  The negotiator will
observe that nobody else is using their quota, and give the surplus to
the incoming jobs.   Once those jobs start running, they may or may not
be pre-empted depending on priorities and configuration settings.

The system stays full. We do have autoregroup enabled. The way it seems to normally run is that if a group is over quota, you get a message saying the group is over quota and that it's skipping it. Then at a later stage because of autoregroup it will go ahead and run some of those jobs if there are slots free. What's happening now is that it thinks the group is using 0.00 slots which isn't true so it goes ahead and runs a bunch more for that group because it hasn't hit it's quota.

I guess I'll try changing the group name to lower case and see if that fixes it.

joe


Erik


_______________________________________________
Condor-users mailing list
To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/condor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/condor-users/