Hello
I’ve got some issues when running pvserver on HPC(Slurm) by singularity. Looking for some help sincerely.
Version
ParaView-5.11.0-MPI-Linux-Python3.9-x86_64.tar.gz (download from this link)
Run Env
HPC Cluster managed by Slurm (CentOS 7) without any mpi
Question
I packet paraview to singularity SIF.
Dockerfile like
FROM centos:7
COPY paraview /usr/local/paraview
RUN yum install -y libXcursor mesa-libGL libgomp
run docker build to image
docker build . -t paraview-dev:test01
Then turn docker image to singularity sif
singularity build paraview.sif paraview.def
paraview.def
# paraview-5-11.def
BootStrap: docker
From: paraview-dev:test01
%labels
Version ParaView 5.11
%setup
%files
%environment
%runscript
exec $@
%post
sbatch script1
#!/bin/bash
#SBATCH -N 2
singularity run -e /data/wanghao/paraview.sif /usr/local/paraview/bin/mpiexec --hosts c1,c2 -np 2 /usr/local/paraview/bin/pvserver
sbatch script1 output calls error like below
bash: /usr/local/paraview/bin/hydra_pmi_proxy: No such file or directory
maybe I should not singularity run mpiexec -n 2 pvserver
? Then I try to edit sbatch script2 like mpiexec singularity -n 2 pvserver
sbatch script2
#!/bin/bash
#SBATCH -N 2
# /data/wanghao/paraview/bin/mpiexec is the same with mpiexec in singularity sif file (/usr/local/paraview/bin/mpiexec)
/data/wanghao/paraview/bin/mpiexec --hosts c1,c2 -np 2 singularity run -e /data/wanghao/paraview.sif /usr/local/paraview/bin/pvserver
The Slurm output is
[root@c1 data]# cat slurm-37.out
Waiting for client...
Connection URL: cs://c1:11111
Accepting connection(s): c1:11111
Waiting for client...
Connection URL: cs://c2:11111
Accepting connection(s): c2:11111
It seems launch two rank 0 mpi process by pvserver?
What I want
I want run pvserver in singularity and run it by mpi and slurm to do a multiple node job.
Hope to get a reply, thank you again!