Set the correct mpirun_command for the marvel.aiida role

[sphuber]

Fixes #177

The role is configured to setup the localhost using the SLURM scheduler
in which case the mpirun_command should be set to srun.

Note this is blocked by an update of the marvel.aiida ansible role: marvel-nccr/ansible-role-aiida#69

[sphuber]

pinging @chrisjsewell

[chrisjsewell]

have you tested this works with the common-workflows?

[sphuber]

Not yet, will try this now. I will do it in the docker container though, that ok? Should not make a difference with the OVA should it?

[chrisjsewell]

Theoretically no 😬

[sphuber]

Ok, I just tested it by hacking the mpirun_command attribute in the database and the calculation runs, except there seems to be a problem with the parallelization. Two separate parallel instances are being run. So for now, this seems to be a bad change, but it should be good. I will look further into it.

[chrisjsewell]

Two separate parallel instances are being run. So for now, this seems to be a bad change, but it should be good. I will look further into it.

Yeh that was my worry, I can never seem to get srun to work properly

[sphuber]

Just for reference. This was suggested in this issue: aiidateam/aiida-common-workflows#180

[chrisjsewell]

yep it certainly makes sense, but obviously I am a bit hesistant to introduce such a wide-reaching change without testing.

As you mention there, you will have a look into the SLURM code to see what it says, which will be great cheers 👍

[sphuber]

No I agree, let's put this on hold for now.