See the Slurm preemption web page here: http://www.schedmd.com/slurmdocs/preempt.html
You apparently want PreemptMode=Suspend,Gang Quoting Nan Yu <[email protected]>: > Hi, > We have some low priority jobs that runs for a long > time. We also have some high priority jobs that finish in a short > time. > Both jobs uses all CPUs on one nodes (parallelized program). > I would like to be able to run the low priority jobs > using all available CPUs on the nodes when there are no high > priority jobs waiting on the same node. When there are high priority > jobs submitted and waiting (because low priority jobs occupies all > of the CPUs), I would like the low priority jobs to be > suspended/paused and give the slots to high priority jobs. And one > high priority jobs finish, the low priority ones could resume the run. > Could anyone let me know if there is an easy way to > achieve this? > Could I just suspend the low priority jobs (and the > slot will be available for high priority jobs)? Or the slot still > will be occupied by the suspended low priority jobs (thus preventing > the high priority jobs from running)? > Or I will have to make the low priority program > being able to have some "check point" (and use kill/restart instead > of suspend/unsuspend)? > Thanks for your help! > Nan Yu > > > Please note that my email address has changed to [email protected] > > ________________________________ > *** The information contained in this communication may be > confidential, is intended only for the use of the recipient(s) named > above, and may be legally privileged. If the reader of this message > is not the intended recipient, you are hereby notified that any > dissemination, distribution, or copying of this communication, or > any of its contents, is strictly prohibited. If you have received > this communication in error, please return it to the sender > immediately and delete the original message and any copies of it. If > you have any questions concerning this message, please contact the > sender. *** >
