Reputation: 6429
What's the best way to get the valid cpu ids from within a running job?
My idea is to do an allocation --> wrap a docker command with the limits of the allocation --> run nvidia-docker
on an remote gpu server.
To limit the docker to the allocation I need to tell it the cpu_ids
.
My job submission will look like:
sbatch -o test.txt -c2 -n 10 --mem=10GB --wrap="job that needs the cpu_ids"
Upvotes: 0
Views: 592
Reputation: 4571
In the script you launch with sbatch, you should execute the next line in order to get the CPU IDs and save them in a file:
scontrol -dd show job $SLURM_JOB_ID | grep "CPU_IDs" | sed 's/^ *//' > cpuids.out
Upvotes: 2