Source: mpi4py Followup-For: Bug #993214 There have been some strange mpi failures. pytest-mpy also started failing tests for no good reason.
spawning has been unreliable in OpenMPI for a long time, cf. https://github.com/jsquyres/ompi/pull/4#issuecomment-806897758 mpi4py upstream informs me that OpenMPI 5 is fixing some substantial problems, particularly with respect to spawning. Hopefully things will be calmer in openmpi space once it's released. The workaround is to skip spawning tests, done in 3.1.1 in experimental mpi4py upstream is working to fix the s390x failure in mpi4py 3.1.1. In the meantime I'll apply the workaround to unstable.
