Although development nodes are configured very similar to compute nodes, sometimes it is useful to prototype scripts interactively on a compute node. This can be done by requesting a interactive job from the scheduler using the srun command.
For example, to request an interactive, three-hour, 32 GiB, four-core slot on one of the compute nodes, use:
[alice@c4-log1 ~]$ srun --export=NONE --nodes=1 --cpus-per-task=4 --mem=32G --time=03:00:00 --pty $SHELL
srun: job 366712 queued and waiting for resources
srun: job 366712 has been allocated resources
[alice@c4-n3:job=366712 ~]$ echo "Number of CPUs on machine: $(nproc --all)"
Number of CPUs on machine: 64
[alice@c4-n3:job=366712 ~]$ echo "Number of CPUs (per cgroups): $(nproc)"
Number of CPUs (per cgroups): 4
[alice@c4-n3:job=366712 ~]$ echo "Number of CPUs (per Slurm): ${SLURM_CPUS_PER_TASK:-1}"
Number of CPUs (per Slurm): 4
[alice@c4-n3:job=366712 tests-slurm]$ module load CBI r
[alice@c4-n3:job=366712 tests-slurm]$ Rscript -e 'parallelly::availableCores(which = "all")'
system nproc Slurm
64 4 4
[alice@c4-n3:job=366712 ~]$
Comment: For most users, --pty $SHELL is the same as --pty bash; using $SHELL makes sure it works for users running other types of shells.
To exit an interactive job, use exit or logout, e.g.
[alice@c4-n3:job=366712 ~]$ exit
[alice@c4-log1 ~]$
Just like for regular jobs, you can get a detailed summary of the resources your interactive job consumed using sstat -j <job_id>.