I will try to explain our needs and if some does not make sense please let me know.
Today we have a system that is written in java and deployed on a 3 nodes cluster.
The cluster is master slave.
Each operation is performed on the master and replicated to the nodes.
Each node has it's own relational DB and it persists into it's own DB.
Our java application has to worry about synchronization between nodes, failover, recovery of node etc.
We want to move part of our DB into elasticsearch and use it's clustering features in order to use the replication and auto failover without our intervention (at least for the DB).
In order to do that, we thought about using the es cluster and let it do the magic for us.