Comment by p1necone

Imo microservices should be a logical boundary only. Whether they're in the same process/server/cluster etc ideally should be purely based on throughput/latency/scale concerns. Even more ideally the services should be dynamically shuffled around hosts based on load at runtime.