Hi Metaflow team, I've noticed that in the

cli.main()

function there is a general warning suppression:

def main(flow, args=None, handle_exceptions=True, entrypoint=None):
    # Ignore warning(s) and prevent spamming the end-user.
    # TODO: This serves as a short term workaround for RuntimeWarning(s) thrown
    # in py3.8 related to log buffering (bufsize=1).
    import warnings

    warnings.filterwarnings("ignore")
    ...

It feels too aggressive, silencing everything can hide some computational issues, such as NumPy's nan_functions' handling of all-nan arrays. I would suggest at least targeting specific warnings.