The legacy search allowed certain nodes to be clients whereas other nodes peformed the actual indexing and searching work. ES supports this very same concept. As discussed here:
If we simply want Sakai to only be a client to an external ES cluster, not store any data, but simply make remote calls for all ES work, that is as simple as just some extra config. This will still mean the Sakai cluster owns the indexing work to load things into ES, it just won't have to do the actual ES work in terms of storage or responding to search queries. In a large organization, that might be using ES for other purposes this is reasonable architecture. In addition, it takes the burden off the Sakai nodes to do searches and lowers the risk of something going bad with ES bringing down Sakai nodes.
This also gives smaller installations a way have a single sakai node serve search requests, or in the case of Sakai hosting provider, they might have one ES cluster that services multiple Sakai installations.
To tell any one single Sakai node to not store data simply do this:
To set the cluster name you would do:
unicast, multicast and all the other normal ES settings are not affected by this change. Those would still need to be configured to point to where ever the ES cluster lives.