I setup a 3 billion plus document ES cluster using AWS Elasticsearch Service. As a whole it works really well, and it's nice not to have to manage individual EC2 instances. With that said, there are some very annoying things. Previously if you modified the access policy of the ES cluster, it would completely rebuild the entire cluster (create all new nodes and copy all data), which can take many hours.

Recently I turned off some slow query and slow index advanced logging options, again AWS completely rebuilt the entire ES cluster when these options surely should be able to be applied without copying all data.