[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] Condor View and flocking
- Date: Wed, 29 Jun 2005 11:33:14 -0500
- From: "David A. Kotz" <dkotz@xxxxxxxxxxxxx>
- Subject: Re: [Condor-users] Condor View and flocking
Michael Yoder wrote:
I have two internal Condor pools configured for two-way flocking, as
well as two pools external to our organization also configured for
two-way flocking with us. I'm running Condor View on both of my
The users submit jobs from desktops in one of the pools, with jobs
running on both. Condor View indicates that jobs are running on the
pool from which the jobs are submitted, and that there are only idle
jobs on the second pool.
Just to be sure, are you getting this from the "Machine" statistics or
the "Job" statistics? I think you're only looking at the 'job' side,
right? (One could infer job behavior based on machine usage...)
Yes. I'm using the User statistics of Condor View.
Is this the expected behavior? Are there any other reporting quirks
which I should be aware in regards to flocking? Should jobs flocked
from other pools show up in my Condor View stats?
This does seem strange. You're saying that pool 1 has some amounts of
running/idle jobs, and pool 2 shows just the idle jobs from pool 1? I
would expect for pool 2 to see either both idle and running, or nothing.
That's correct. Actually, it doesn't even seem to show all of the idle
jobs from the primary pool (the one with submit nodes). The peak idle
jobs number is lower on the second pool.
Thanks for taking the time for the explanation.
More details...this explanation may be a little too in-depth, but here
goes. During normal operation, a schedd sends "submittor" ads to the
collector. There is one submittor ad per schedd per
job-submitting-user. You can see these ads via 'condor_status -sub'.
If the submittor ad shows that a user has jobs to run, then the
negotiator will contact the schedd and negotiate. If the jobs can't be
matched with the normal collector, the schedd will then send submittor
ads to other collectors in $(FLOCK_TO).
For our purposes, the interesting parts of the submittor ad are
RunningJobs, IdleJobs, and FlockedJobs. It looks to me like the
submittor ads are always filled in with the number of idle jobs a user
has. If this gets sent to both collectors, both pools will show idle
jobs from that submittor. For the running jobs...I suspect that the ads
to the pool 1 collector are given
RunningJobs = the number of running jobs for that submittor (for pool 1)
FlockedJobs = 0 (?)
and the pool 2 collector gets
RunningJobs = 0
FlockedJobs = the number of running jobs for that submittor (for pool 2)
And the "FlockedJobs" value is totally ignored by the collector,
negotiator, and condor view. :-) That's just a guess, but you could do
some 'condor_status -sub -l' queries on both pools and confirm/deny.
Principal Member of Technical Staff
Direct : +1.408.321.9000
Fax : +1.408.321.9030
Mobile : +1.408.497.7597
2860 Zanker Road, Suite 201
San Jose, CA 95134
Condor-users mailing list