The Slurm scheduler sets --ncpus-per-task to a float

[pfebrer]

Describe the bug

When submitting a job defining num_cores_per_machine, the scheduler is unable to submit the job because --ncpus-per-task is set to a float, which SLURM doesn't understand. The workchain gets then paused forever because it is unsuccesful at submitting the job.

Steps to reproduce

Using a slurm scheduler in a computer with mpiprocs_per_machine defined:

submit(CalculationClass, options={"resources": {"num_machines": x ,"num_cores_per_machine": y})

Your environment

• Operating system [e.g. Linux]: Linux
• Python version [e.g. 3.7.1]: 3.6.9
• aiida-core version [e.g. 1.2.1]: 1.4.2

Other relevant software versions, e.g. Postres & RabbitMQ

Additional context

I think it's related to this part of the code:

aiida-core/aiida/schedulers/plugins/slurm.py

Lines 133 to 139 in 9ff07c1

	# In this plugin we never used num_cores_per_machine so if it is not defined it is OK.
	resources.num_cores_per_mpiproc = (resources.num_cores_per_machine / resources.num_mpiprocs_per_machine)
	if isinstance(resources.num_cores_per_mpiproc, int):
	raise ValueError(
	'`num_cores_per_machine` must be equal to `num_cores_per_mpiproc * num_mpiprocs_per_machine` and in'
	' particular it should be a multiple of `num_cores_per_mpiproc` and/or `num_mpiprocs_per_machine`'
	)

As I understand, this division should return an integer, but the opposite check is performed. I don't know, maybe I'm wrong.

[chrisjsewell]

thanks @pfebrer, I will close this by fixing the bug that you noted. But feel free to re-open if you still find an issue after