As described here, when a READ_FAILURE or WRITE_FAILURE is returned from a coordinator, the driver currently throws this back to the user as ReadFailureException and WriteFailureException and skips RetryPolicy completely.
We should consider enabling users to specify that these errors can be retried via RetryPolicy.onRequestError, although I believe in the general case, these should not be retried, so if we do make that change, we should add an exception in that method to not retry those by default.