fractal-analytics-platform / fractal-server

Fractal backend
https://fractal-analytics-platform.github.io/fractal-server/
BSD 3-Clause "New" or "Revised" License
11 stars 3 forks source link

SlurmConfig does not read options that have hyphens in their name #348

Closed jluethi closed 1 year ago

jluethi commented 1 year ago

In the current setup, all our executors only use a single CPU core, even the CPU-high (supposed to take 16) only takes one.

tcompa commented 1 year ago

For the record, at UZH this submission script

#!/bin/bash

#SBATCH --job-name=test
#SBATCH --partition=main
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=4

echo Start
sleep 5
echo End

leads to

JOBID            USER    ACCOUNT                           NAME    STATE       TIME TIME_LIMIT CPUS MIN_MEMORY NODELIST(REASON) DEPENDENCY
9430760       fractal       root                           test  RUNNING       0:01 7-00:00:00    4     62533M pelkmanslab-slurm-worker-014 

Which looks correct. Then we should just check that the fractal submission scripts look like this one.

Do you observe the same behavior on the cluster you are using?

jluethi commented 1 year ago

Will check on the UZH server tomorrow, current tests were on FMI servers. But I thought I also saw it on the UZH cluster.

tcompa commented 1 year ago

There is a bug in the way some options are read from the configuration JSON - will report about is soon.

TL;DR It seems that options which have an hyphen (like ntasks-per-node) are not parsed correctly.