209 openmpi path issue with khiops 1022 on debian like os #218
Conversation
popescu-v
force-pushed
the
209-openmpi-path-issue-with-khiops-1022-on-debian-like-os
branch
2 times, most recently
from
49c3e6b to
7c14b07
Compare
| # an executable name under a /bin directory | ||
| # Note: this executable name can be different, depending on the MPI | ||
| # backend and OS; for instance, "orterun" for OpenMPI on Ubuntu Linux, but | ||
| # "mpiexec" for OpenMPI on Rocky Linux | ||
| kh-status | grep "MPI command" | grep -Ewq "(/.+?)/bin/.+" |
There was a problem hiding this comment.
This check is weaker than whaty is stated.
Maybe we should check grep -Ewq "orterun|mpirun|mpiexec" ?
There was a problem hiding this comment.
Well, I am not sure: what if the executable is called something else (e.g. mpiexec.hydra for MPICH, whence we would also add this, or rather mpiexec\.?.? etc)?
It soon becomes brittle IMHO. In the current state, we just check that the MPI command points to something that is inside a bin directory. The tests that mpiexec is found are elsewhere (in the unit-tests workflow).
So, I would keep it this way; the comments contain "for instance" and this is key IMHO.
There was a problem hiding this comment.
Ok. I think though the first part should be (.+), and not (.+?)
There was a problem hiding this comment.
I wanted to accommodate the case when the path starts with /bin (this is not excluded in principle I guess).
Khiops now supports it, following PR KhiopsML/khiops#241 which fixes issue KhiopsML/khiops#235.
Thus, mpiexec can be /bin/mpiexec through symlinks related_to #209
Otherwise, in some situations, we get to /bin/mpiexec (which is a symlink to the real path) which makes OpenMPI fail; see: open-mpi/ompi#5613 closes #209
popescu-v
force-pushed
the
209-openmpi-path-issue-with-khiops-1022-on-debian-like-os
branch
from
7c14b07 to
33586e0
Compare
popescu-v
deleted the
209-openmpi-path-issue-with-khiops-1022-on-debian-like-os
branch
closes #209