An issue I had with AWS was that even after they launched VPC ES clusters, they still hid the node IPs in the API output(I guess they patched ES?!). Our use case as a data lake meant that we had severely degraded performance with the hadoop/spark drivers than one would expect form it.