[slurm-users] understanding resource reservations
breedthoughts.www at gmail.com
Mon Oct 21 22:05:08 UTC 2019
I have a cluster where machines are used for both compute jobs and for
interactive research by humans - we're resource constrained so getting
machines dedicated to slurm is a tough task. What i'd like to do is,
during normal weekday work hours, take some machines entirely out of the
cluster, and on some other machines just restrict the cores allocated to
slurm. For example, I want machine A to be unavailable to slurm from
9am-5pm Monday-Friday, machine B to only have 50% of its cores
available during this time, but machine C to be 100% available at all
It sounds like resource reservations might do what i want, but after
reading the guide at https://slurm.schedmd.com/reservations.html but some
things still seem unclear to me.
1) It looks like there's a way to create a daily recurring reservation by
specifying "flags=daily" . How would I make a regular reservation for
2) When a node enters a reservation period, is it like going into the DRAIN
state where currently running jobs will be allowed to finish but no new
ones will be started, or will any currently running jobs just be killed?
And would the behavior be different if I'm reserving entire machines versus
reserving a few cores?
3) If I have jobs that I know will take a long time to run, I'd want to
keep them from being started on a machine that's about to be reserved. Are
there any flags I can set on the jobs to do this?
thanks in advance for any help.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the slurm-users