Elasticsearch supports a scale-out deployment architecture, with the main priority on availability. The Elasticsearch cluster is composed of nodes playing master and data node roles. Quorum based availability drives cluster availability in general.
While the most common deployment is that of single data center deployment, which can comprise dedicated master nodes and dedicated data nodes, for larger clusters, multiple data center deployments are also required for very high availability of critical applications.
It is generally not a very good idea to deploy Elasticsearch clusters across data centers, since the Elasticsearch leader election algorithm and data node selection are based on network distance, which assumes that all the nodes are identical in terms of all resources; it is expected that in a cluster all nodes are equidistant. This can go wrong in cross data center clusters...
