Thank you for your answer. We are aware of that our approach is not the clean one as the one with explicitly defined affinity keys would be. But still for the sake of data distribution defining affinityKeys, in this example this would be cacheName/key for a given key, should be equivalent to overriding the partition method in the described way. Hence we would still get a data-distribution far from being uniform. To me this is a bit surprising as this would imply that even in the case of distinct affinity-keys one does not achieve a balanced data-distribution with RendezvousAffinityFunction.
-- Sent from: http://apache-ignite-users.70518.x6.nabble.com/
