search

kleinhenz / SlurmClusterManager.jl

julia package for running code on slurm clusters
46 stars 5 forks source link

n allocated processes, and n+1 spawned in julia #3

Closed matthiasbe closed 3 years ago

matthiasbe commented 3 years ago

If I use the command sbatch -n 4 script.jl with the script of your example, I see that there are 4 workers registered, plus 1 master process.

How will these 5 processes be distributed on the 4 allocated resources ? Is the master process sharing a resource with one of the worker processes ?

kleinhenz commented 3 years ago

yeah currently the master process shares resources with the 1st worker process. It might be nice to have an option to not do this for cases where you need master to stay responsive. Do you have a case where this causes problems?

matthiasbe commented 3 years ago

Ok interesting ! I wanted to know what was happening, in my use case it actually what I would expect.