Brian: What you describe below should work, but there is a caveat. Condor doesn't try to match all permutations of all parts of a parallel job to every machine, so if there's exactly one machine that has database slice foo, and that machine matches both the database slice foo part of the job and the coordinator, there's the possibility that condor will successfully match the coordinator to machine foo, and then be unable to match the whole job, because the "foo slice" job can't match. So, it is best to write these requirements to be mutually exclusive of one another. -Greg
|