[slurm-dev] random segmentation fault in user app.

[Yann Sagon]

Hello,

I have a random segfault. Even a very simplist mpi program crash randomly
depending on the number of cores used.

How I launch the executable:

srun --reservation=sagon_39 --partition=shared --time=00:50:00 -n 34 a.out
srun: error: node09: task 11: Segmentation fault

If I try another number of cores, it randomly crash. For example with 64
cores, it doesn't. It seems as well that if I do first a salloc and a
mpirun, it's working.

Here is the code of the soft I try to launch:

****************************
#include <mpi.h>

int main(int argc, char* argv[]) {
    MPI::Init(argc, argv);
    MPI::Finalize();
}
****************************
how I compile the soft:

mpiCC test.cpp

srun --version :
slurm 14.11.3

mpirun --version
mpirun (Open MPI) 1.8.4

interconnect is ib.

mpiCC --version
g++ (GCC) 4.8.2 20140120 (Red Hat 4.8.2-15)


How I compiled slurm:

rpmbuild -ta slurm-14.03.1-2.tar.bz2

How I compiled openmpi:

./configure --prefix=/usr/mpi/gcc/openmpi-1.8.4 --with-verbs
--with-slurm --with-pmi --with-usnic=no

any clue greatly appreciated!