Recovery messages are targeted at specific NIs. If the recovery message fails then we shouldn't attempt to resend via the normal health resend logic. Instead, we just let the monitor thread resend it on the next interval.