On 04/29/2010 10:25 AM, Ian Chesal wrote:
We've done some work in recent versions of condor to make bulk condor_rm's go faster. One problem was that at the level of job removal, the schedd didn't know that it was working on a request to remove all jobs, so for each running job that was removed, it tried to re-match the now-available machine for one of the idle jobs, resulting in some embarrassing n^2 runtimes. If you aren't using DAGman, one quick and dirty way to work around this problem, is to remove all jobs in two steps: first remove all idle jobs:What about setting START = False on your machines so nothing runs? This buys you time to do throttled condor_rm's. - Ian
condor_rm -const 'JobStatus == 1' then remove all the rest of the jobs: condor_rm -a -greg