Package: libopenmpi3 Version: 4.0.4-2 Followup-For: Bug #965352 Control: affects -1 petsc4py
petsc4py fails tests also, but the way in which it fails might give a hint to whats going on. This failure is not explicitly UCX. The petsc4py failure is in the swig test. Can reproduce manually (runtime with python3-petsc4py and python-petsc4py-doc installed): cp -r /usr/share/doc/python-petsc4py-doc/examples/demo test-demos gunzip -r test-demos cd test-demos/wrap-swig make PETSC_DIR=/usr/lib/petsc PYTHON=python3 F2PY=f2py3 MPIEXEC="mpiexec -n 2" run ... the test hangs (not a UCX segfault or recv Bad address). But on the other hand make PETSC_DIR=/usr/lib/petsc PYTHON=python3 F2PY=f2py3 MPIEXEC="mpiexec -n 3" run passes cleanly. So with petsc4py, 2 processes hangs, 3 processes passes. Sadly it's not as simple as that for mpi4py. With mpi4py, both mpirun -n 2 python3 -m pytest test/test_rma.py -v and mpirun -n 3 python3 -m pytest test/test_rma.py -v fail the same way.
