One of the nifty features of the original "Docker Swarm" was that it:
turns a pool of Docker hosts into a single, virtual Docker host
allowing tools (such as the docker CLI, and docker-compose) to be agnostic about whether they were operating against a single instance of Docker Engine, or a Swarm cluster.
Docker 1.12 brings an integrated "swarm mode", which is an exciting new take on Docker orchestration. But, have we lost the "cluster as virtual Docker host" feature in the process? Using docker run against a swarm-mode master only ever seems to start containers on the master node itself. docker ps now returns no hint of which machine each container is running on. It it true that one must now use cluster-specific subcommands like docker service and docker deploy to do things across the cluster?
That's a correct interpretation. You will longer manage a Swarm as you have done with a single engine, the commands to manage and schedule in the Swarm are changing. Using docker run and docker-compose will target a single host. Using docker service is needed to schedule on the swarm. The compose is moving to bundles that you docker deploy and manage with docker stack. I'd say one huge plus of the new design is not forgetting when you're managing the swarm vs a single node because you didn't check your setting of $DOCKER_HOST.
Migration has been considered with docker-compose bundle, and there's nothing preventing you from running the old swarm container implementation on the 1.12 platform and accessing it as you always have. You just won't get the orchestration benefits of the 1.12 Swarm.