Handle parallel execution without mpi4py. #117
Description
perun crashes when run on an MPI context without having mpi4py installed. This happens when writing the hdf5 and other output files, as multiple perun processes try to write to the same path.
What needs to happen:
- perun needs to catch the error and provide a usefull error or warning message to the user indicating about the missing mpi4py library.
- Ideally, each perun process will write the report with a different file name to avoid write conflicts. The file name could include the hostname and a process identifier (PID, MPI rank from env variables, ...) to ensure this. Optionally, just check what already exist and add a "_n" suffix to the filename.