Hi,

This case could help to give you some ideas: https://www.prosysopc.com/cases/kruger/ .

The number of tags is very large. They wont probably fit into memory nor it might not anyway make sense to even try, if you need multiple servers operating the data at the same time. Therefore you probably need to use MyBigNodeManager-style approach (see SDK samples), where you query some datastorage (kafka or some database inbetween or something else, this is up to you) instead of having UaNodes.

Note that if you use some kind of load balancer: Normal OPC UA is Session-based, therefore one session or in practice one connection must always route to the same server instance.