condor_rm does not remove all jobs¶

The technical background is that the scheduler is keeping track of the jobs lifecycle and in case contact to the executing instance (called the startd) is lost the sched is reluctant to remove the job from it's own records as the remote state is unknown. The forcex option is 'forcing' the sched to do so.

condor_rm followed by condor_rm -forcex does the trick

The fact that the sched lost track of the jobs in question is as you probably imagine caused by comunication problems between sched and worker due to load problems (most of the times) or the worker dying etc.