Hi. Good question. We actually have both use cases. In the case where we require reliability and guaranteed delivery we publish to multiple nodes and standard c

Hi. Good question. We actually have both use cases. In the case where we require reliability and guaranteed delivery we publish to multiple nodes and standard channels that persist messages in the case of subscribers disconnecting or overflow in what can't be buffered in memory. In the other case we just publish once and can use ephemeral channels. We scale clusters depending on usage but start with a minimum of 3 and use nsqlookupd to find topics in the cluster. In the case of things that can't handle duplicate deliveries we offload deduping to simple things memcached.