soravux / scoop

SCOOP (Scalable COncurrent Operations in Python)
https://github.com/soravux/scoop
GNU Lesser General Public License v3.0
634 stars 87 forks source link

Using scoop with SLURM #54

Closed anandtrex closed 7 years ago

anandtrex commented 7 years ago

Is there any documentation for how to use scoop with SLURM?

One of the main things I'm wondering about is whether to provide a hosts file to scoop or not when running it from SLURM. Does it automatically figure out the hosts and run simulations on them otherwise?

#!/bin/bash
#SBATCH --mail-user=abc@xyz.com
#SBATCH --mail-type=ALL
#SBATCH --nodes=7
#SBATCH --ntasks=72
#SBATCH --time=99:00:00
#SBATCH --mem=10G
#SBATCH --output=python_job_slurm.out

# Which one is correct?
python -m scoop --hostfile hosts.txt my-script.py
python -m scoop my-script.py

and I run it with sbatch python.slurm

joernhees commented 7 years ago

i'm currently running it on slurm... it seems the intention is to figure out the hosts itself as in https://github.com/soravux/scoop/blob/master/scoop/utils.py#L204 (so the second variant should be correct).

that said, the parsing fails for me (the code you see above is not the one of the latest release!), so at the moment i'm running manually with python -m scoop --host ... my-script.py, which seems to work

anandtrex commented 7 years ago

Yeah it seems to work without the hostfile list with the version of scoop on the master branch.