Multi-hosts contexts

Currently docker context create points to only a single host. It's great but not perfect when dealing with Swarm clusters. Although Swarm clusters could be seen conceptually as one big computer where services are deployed, there's still a loophole: when it comes to automatically deploy a stack (eg. using a shell script/Makefile), stack deploy has to be run on a predefined manager node. If that node is actually down for some reason, that script will fail. That's particularly problematic when the given script is run by a CI.

Docker contexts could alleviate this problem by allowing users to define multiple hosts for a single context. When running a command using this sort of contexts, the CLI would check if that host is available and would switch to another host until it finds an available node before running the command.

docker context create supports providing multiple --docker flags but only the last one is saved in the context:

$ docker context create prod \
    --docker "host=ssh://prod.node1" \
    --docker "host=ssh://prod.node2" \
    --docker "host=ssh://prod.node3"

$ docker context inspect prod
[
    {
        ...
        "Endpoints": {
            "docker": {
                "Host": "ssh://prod.node3",
                "SkipTLSVerify": false
            }
        },
        ...
    }
]

I'm willing to implement it if you're ok with this feature 🙂

Start with the docker context create and docker context inspect behavior described in the issue, then trace how stack deploy selects its Docker endpoint. Define how multiple --docker hosts are stored and tried, and verify that commands can move to an available host when the first node is down.