'Fail early' and 'throw exceptions in exceptional circumstances'. You need good top level error handling and logging of course. I still work with engineers that

'Fail early' and 'throw exceptions in exceptional circumstances'. You need good top level error handling and logging of course. I still work with engineers that try catch exceptions and return some sort of more 'graceful' error, for problems that can't really be handled in any other way but to let the system fail. This practice makes it clear when problems arise and where they come from.