Execution and placement require two related views: where components run and connect, and how a host schedules their work. Keeping those views separate lets review distinguish changes to topology from changes to execution policy.
Engineering question. Where does work run, and which scheduling decisions may change with the host it runs on?
Representation. Use separate views for topology and host policy.
Deployment topology represents where components run and how they connect: web → queue → worker → storage.
Host execution policy maps each host profile to a scheduling policy: elastic fan-out, bounded fan-out, or serialized execution.
Figure C.3-1 shows the stable graph with the policy as a separate layer beneath it.
Figure C.3-1.Stable topology, variable execution policy. Deployment topology remains stable while host profiles select different scheduling behavior.
Property. A host's concurrency ceiling belongs to scheduling policy, not to the deployment graph.
Authority and correspondence. Deployment configuration reconciles against the topology; scheduler behavior reconciles against the host execution policy. Either finding may remain advisory or feed a deployment or scheduler gate.