Hi, we had a slightly similar setup on an older cluster;
- The frontend and the home directories were the same system - Home dirs were physically under /export/home - Compute nodes mounted the home dirs over NFS using the automounter at /home/$USER This sounds a bit like the situation you describe? We solved this in the (standard?) fashion, that is - In the password database, home directories were /home/$USER - On the frontend, we also ran the automounter, which bind mounted /export/home/$USER to /home/$USER So in the end, users never had to work with or even be aware of the /export/home stuff, their home directory was always at /home/$USER regardless of whether they were at the frontend or running a batch job on one of the compute nodes. -- Janne Blomqvist ________________________________ From: Kevin M. Hildebrand [ke...@umd.edu] Sent: Wednesday, September 04, 2013 15:40 To: slurm-dev Subject: [slurm-dev] SLURM vs automounters Hi, at our site we use an automounter to mount user home directories. Unfortunately this is confusing the job launch mechanism as SLURM is picking up the automount paths instead of the desired home directory paths. i.e., my desired path is /homes/kevin, but the path picked up by SLURM is the physical path /export/home/xyz/kevin, which doesn't exist on the compute nodes. I can write a simple lua job_submit plugin to fix work_dir if it has the wrong path elements, and that solves the problem nicely for sbatch, but not for srun. (srun doesn't appear to use work_dir and uses cwd instead...) I'd like to make this as easy as possible for the users and not have to make them specify extra command line arguments to make this work. Has anyone else encountered this and come up with a good solution? Thanks, Kevin -- Kevin Hildebrand University of Maryland College Park Division of IT
