Should one or multiple controller_manager nodes be used for modular robots (e.g. base + arm)? [ros2_control]

edit

If you have a robot composed of several sub-robots (e.g. a mobile base platform with a manipulator attached), each of which has its own independent controllers defined, is there a preferred approach to managing the controllers for the combination?

Are there reasons for or against either:

• aggregating the controllers for both the base and arm into a single controller_manager node, or
• running separate controller_manager nodes for each sub-robot (e.g. base_controller_manager and arm_controller_manager)?

I'd imagine the former might have a bit less overhead, but the latter might make lifecycle management a bit more flexible?