DevHeads.net

OpenMPI is flaky

Hi,

I was trying to bump the espresso package to v4.1.1 and came across
multiple issues in the %check phase of the builds on Rawhide, F31 and
EPEL8.

On Rawhide and F31, some tests seem to receive an error in the MPI_INIT phase.
Error:
*** An error occurred in MPI_Init
*** on a NULL communicator
This happened with OpenMPI, but not MPICH.
I was able to workaround these failures by retriggering the builds a
couple of times, which isn't a great solution.

Upstream issues on that topic:
Rawhide: <a href="https://github.com/espressomd/espresso/issues/3306" title="https://github.com/espressomd/espresso/issues/3306">https://github.com/espressomd/espresso/issues/3306</a>
F31: <a href="https://github.com/espressomd/espresso/issues/3305" title="https://github.com/espressomd/espresso/issues/3305">https://github.com/espressomd/espresso/issues/3305</a>

On EPEL8 however, things are much worse - multiple tests fail on each
arch in MPI_INIT incl. a couple of segfaults:
<a href="https://github.com/espressomd/espresso/issues/3307" title="https://github.com/espressomd/espresso/issues/3307">https://github.com/espressomd/espresso/issues/3307</a>
I wasn't able to reproduce these issues in a CentOs8 container in docker.

Any ideas and help would be appreciated.

Unrelated, but still annoying I saw a
Fatal Python error: init_sys_streams: can't initialize sys standard streams
on s390x (F31), which got triggered by setting the somewhat unrelated
CMAKE_SKIP_RPATH=ON option.
Any suggestions on that would be more than welcome.
Upstream issue: <a href="https://github.com/espressomd/espresso/issues/3316" title="https://github.com/espressomd/espresso/issues/3316">https://github.com/espressomd/espresso/issues/3316</a>

Christoph