this can isolate an error for a certain probe, so that other probes can go
ahead and take their readings during each client run. that way only the bad
one is marked as "error" status by the server
note that a retry is *not* attempted within a given "take readings" run.
rather, client will consider that full readings take to have failed, if any
part of it fails.
but then we keep track of type/amount of some (database connection) failures,
and will suppress logging the full error for first 3 attempts. in practice
this lets us recover from simple database restarts, and if database becomes
truly unavailable we'll hear about it shortly.
any other type of error is immediately logged on first failure.