If you really want to kill all of the pending, running and completed jobs, start slurmctld with the "-c" option. ________________________________________ From: [email protected] [[email protected]] On Behalf Of Jette, Moe [[email protected]] Sent: Tuesday, February 22, 2011 2:21 PM To: [email protected] Subject: RE: [slurm-dev] squeue display stale output
If you login to the compute nodes, you should the user processes and/or slurmstep processes and/or Epilog scripts running. The CG state will persist until everything is cleaned up. You can also check the SlurmdLogFile for more info. ________________________________________ From: [email protected] [[email protected]] On Behalf Of Paul Thirumalai [[email protected]] Sent: Tuesday, February 22, 2011 1:06 PM To: [email protected] Subject: [slurm-dev] squeue display stale output I was using select/cons_res but had to go back to select/linear since the distribution of utilization among all nodes was not good. All jobs were landing on a few nodes with 4 cpus. Is there a way to fix this. I also changed all partitions from Shared=NO to Shared=YES Now whats happening is that when i do an squeue it shows a number of jobs I ran earlier in CG state. It seems like slurm is not completeing any new jobs. 1. Is there a way I could flush the slurm output Thanks Paul
