[tracker1]

The bigger issue to me, is (at least the last time I looked) you can't use the cluster module with node combined with gRPC, so the only real way to take advantage of extra CPU capacity, if available is workers or external processes that are self-managed vs. cluster integration.

[jrockway]

I think you can just run one node.js per core (or whatever the optimal balance is) and tell your load balancer that there are instances of your service available at hostname:8080, hostname:8081, etc. A lot of people are going to get this "for free" when they tell their container orchestrator that they want 8 replicas that each request 1 cpu, and the scheduler finds a node with 8 free cpus.