use topology and affinity to ensure ha services are distributed evenly across zones and hosts

in this way we avoid single points of failure or minimise the impact of noisy neighbouring services on a nodegroup by evenly spreading them across zones (and more finely hosts themselves)