settings.num_cpus = 1 does not work

I am distributing some qutip code to several cores. I want to enforce that qutip only uses one core per job with qutip.settings.num_cpus=1, but it is not working, the processes still take all the processors of my machine, so that if I have N cores, the multiprocessing ends up using N*N threads, slowing things down a lot! How can I force qutip to use only 1 core? Currently loading like this:

from qutip import create, destroy, Qobj, expect, tensor, qeye, fidelity, basis, variance, settings
from pathos.multiprocessing import ProcessingPool

settings.num_cpus = 1

[Ericgig]

This parallelism does not come from qutip but scipy: coherent use scipy.sparse.linalg.expm which run in parallel for big matrix. This can be fixed with enviroment variable like$ export OPENBLAS_NUM_THREADS=1 in bash or os.environ["OPENBLAS_NUM_THREADS"] = "1" in python, but there are a few possibilities:

OMP_NUM_THREADS: openmp,
OPENBLAS_NUM_THREADS: openblas,
MKL_NUM_THREADS: mkl,
VECLIB_MAXIMUM_THREADS: accelerate,
NUMEXPR_NUM_THREADS: numexpr

See https://stackoverflow.com/questions/30791550/limit-number-of-threads-in-numpy

[nathanshammah]

I remember that @fminga had a similar necessity while exploiting qutip.parallel on a cluster (the RAM or SWAP were clogging the cluster) and I think he solved it by enforcing the num_cpus in the parfor. I don’t know if this would work for your case.