ElasticSearch Cluster - Best Practices
Three node ElasticSearch cluster
For High Availability and Failover, a three (3) node ElasticSearch cluster is the recommended solution.
Server 1: Data node - master eligible
Server 2: Data node - master eligible
Server 3: Non-data, non-master eligible - only for quorum
With this setup, any single node can be lost and the other two nodes will continue to function. This would allow for high availability.
Server 3 doesn't need many resources, as ElasticSearch wouldn't need many resources on that node since it would only be a quorum witness. Ideally, you would/could use the server (or one of the servers) that will be querying ElasticSearch, which would allow to send the queries through the local instance so that it could automatically route/load-balance them across the data nodes.