I think that microservices are just a deployment model of the service boundary and there should not be really a distinction between whether something is deployed as a microservice or a monolith, because application should support both for the scenarios when it makes sense.

Consider the following API:

  UsersService:
   CreateUser
   GetUser

  AppCatalog:
   GetApp
   CreateApp

What if AppCatalog and UsersService implement both local version of the interface and GRPC one? Then the distinction whether it's a microservice vs a monolith goes away, it becomes a matter of whether they are deployed in a single linux process or across boundaries of processes/servers.

I have implemented this technique in teleport:

https://github.com/gravitational/teleport/tree/master/lib/se...

Integration test suite is run against RPC version and local version at the same time to make sure the contract remains the same:

https://github.com/gravitational/teleport/blob/master/lib/se...

A single teleport binary can be deployed on one server with all microservices, or multiple cluster scenarios.

where the binary is simply instantiated with different roles:

  auth_service:
    enabled: yes
  node_service:
    enabled: no

Is Teleport a monolith? Yes! Is it a micro-service app? Yes! I'm so happy that we don't have to think about this split any more.