I am running an application which is a Java process that is deployed on 100s of nodes. I am planning on deploying a Hazelcast instance along with each node in the same VM, as well as enabling near cache on each member. Each app process will use the Hazelcast client library to talk to the local member.
In such a deployment model, as long as I provide localhost and the local port as the seed nodes to the client library, each app process will only talk to the local member's near cache and will never query a member running on another machine right ? (except occasionally for reading change updates made)
If you use the Hazeclast Sidecar deployment, then technically you use the client/server topology, so the data is "cached" in your Hazelcast client. That means that for the related read operations, you don't even contact any member, but just use the cached client entries.
Please check the details at Hazelcast Reference Manual: Near Cache.