Indeed, though this is irrelevant to this particular use case. NATS doesn't have replication, sharding or total ordering. Consistency is a challenge for cluste

Indeed, though this is irrelevant to this particular use case. NATS doesn't have replication, sharding or total ordering. Consistency is a challenge for clustered messaging brokers that need this. With NATS, queues are effectively sharded by node. If a node dies, its messages are lost. Incoming messages to the live nodes will still go to connected subscribers, and subscribers are expected to reconnect to the pool of available nodes. Once a previously dead node rejoins, it will start receiving messages. NATS in this case replaces something like HAProxy; a simple in-memory router of requests to backends.