[Midway] Modules need to be repeatedly loaded_unloaded before they work on a compute node. - antoszewski/test-wiki GitHub Wiki

Problem

This is a partial solution to the need to repeatedly load/unload modules such as PLUMED and GROMACS before they start working on compute nodes.

Part of the problem comes from how SLURM sets up a job: the environment variables (run env to see the list of current environment variables) which are not related to SLURM are copied to a newly started process. The result is that module believes that the loaded modules are already set up, while in fact they are only partially set up.

Solution

One way to prevent some of this problem from occurring is to not copy any environment variables related to module to the job. A side benefit is that submit scripts are now more reproducible since the environment is the same each time you submit.

To partially solve this:

1. Remove all traces of module from your .bashrc file.
2. Add #SBATCH --export=NONE to your submit script. What this does is instead of copying the environment variables from the shell you're submitting the job from, it starts a new shell (.bashrc  will be sourced) and copies the environment variables from that. Note that you'll have to all --export=ALL to any srun commands you call in your submit script if you want environment variables to be passed on.