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

Re: [Condor-users] flocking problem



On Thu, 2005-01-06 at 16:49 -0600, Derek Wright wrote:
> On Thu, 06 Jan 2005 15:58:09 -0600  "David A. Kotz" wrote:
> 
> > Mere seconds after I submitted my question, the jobs inexplicably
> > decided to flock.  Prior to sending, I had watched at least three
> > consecutive negotiator cycles say that there were no matches for my
> > test jobs.
> 
> keep in mind that with the flocking setup you described, you'll have 3
> *different* negotiators at work.  flocking is entirely handled by your
> condor_schedd.  if it's told: "you're in pool A, but you can flock to
> pools B and C", the schedd will initially just send classads to the
> collector for pool A.  eventually, the negotiator for pool A will
> contact the schedd and attempt to find machines that match the
> schedd's jobs.  if the schedd finishes talking to negotiator A and
> still has idle jobs, it'll start sending classads to both collector A
> *and* collector B.  eventually, negotiator B will notice this schedd
> appeared, and will contact it to see if it can give it any matches.
> if, after talking to both negotiator A and negotiator B, the schedd
> *still* has idle jobs, it'll start to advertise itself in pool C, and
> the process will continue.
> 
> so, if you were only looking at the negotiator log for pool A, you'll
> *ALWAYS* see it say "can't find any matches for the job that wants to
> be in pool C".  make sense?
> 
> -derek


What you're saying makes sense, and it's basically what I was expecting,
but I tailed the Negotiator logs for both Pool A (submitter pool) and
Pool B (requested pool) through 3 or 4 negotiation cycles, and I saw no
signs that the job was going to run.  Then I clicked 'Send'.  :-)

I had run that experiment to make sure I had the details straight before
posting.  The day before I had left two small job clusters (using the
same submit file) in the queue overnight with no success, so I really
didn't expect these jobs to flock.  Perhaps lack of faith is the key to
successful flocking.

One question that remains for me is how long is "eventually"?  At what
point will the schedd begin shopping its jobs around to other pools?


-- 
David A. Kotz <dkotz@xxxxxxxxxxxxx>