[alhom]

... since otherwise we get just "writeGrid failed" and no diagnostics at all. Can go deeper to add these, but raise I think, has not been used here since bulk writing has been considered non-critical (... but restart writing failure will not kill the job either?).

Could be done with try-catch, of course, and with a write-errors-are-fatal switch... I would hazard a guess that it wouldn't have a prohibitive overhead in conjunction with IO.