Yes this solution was already discussed on stackoverflow. Problem is that this spark functionality should be adjusted in DBR to handle everything automatically via dbfs. Problem is that it seems that it was partly adjusted but not fully.