We're experienced with managing a big scale elastic cluster, but never managed the ML jobs, so knowing about this limitation is definitely useful!