MPI rank always returns 0

installation
1

Hello,
I am running a piece of fenics code on a cluster that has it’s own mpi installations. When I run a command like
mpirun -n 4 python fenics_code.py
I inevitably run many of the functions 4 times, even though I specify these parts within a piece of code such as
if(rank ==0):

On looking closer, I found that rank always returns 0. Can someone help me diagnose the problem, and correct it. I find the solution in [Why do I get comm.rank == 0 everytime when using mpirun?] is not relevant to me because I am not using (at least explicitly) docker.
Thanks!

EDIT:
Minimal example:

from fenics import *
mesh = UnitSquareMesh(10,10)
comm = mesh.mpi_comm()
print('comm.size =', comm.size)
print('comm.rank=', comm.rank)

The above code when launched using slurm (sbatch) returns
comm.size = 1
comm.rank= 0
comm.size = 1
comm.rank= 0
comm.size = 1
comm.rank= 0
comm.size = 1
comm.rank= 0

which is wrong. When I run this locally on my system, it gives me the correct expected output.

2

Please supply a minimal working example, such that others can try to reproduce your error and help you.

3

I’ve put it up in the original question. Hope it helps.

4

As this works for you locally (it also works for me locally using docker), it is clearly something with the installation on the cluster.
Are you sure dolfin has been installed correctly on the cluster?

5

Can you post your slurm job script. Do you request:
#SBATCH -N 1 #Nodes
#SBATCH -n 4 #numTasks

?

6

Thank you for your responses. I do use
#SBATCH -N 1 #Nodes
#SBATCH -n 4 #numTasks

My first test was to run which mpirun in the slurm jobscript. It was pointing to the correct installation of mpi installed with the fenics environment on anaconda. Nonetheless, the error I received was Error: node list format not recognized. Try using '-hosts=<hostnames>’. The solution involved using

mpirun -n 4 -hosts $SLURM_JOB_NODELIST mpi_test.py

Does this improve my performance on actual computations? Not really but that’s a whole other question, I think.