Instead of installing dependencies on the host, directly use the binary
shipped with the redis docker image. This will also prevent possible
problems when API change for example.
This will avoid issues where `dokku cleanup` removes the stopped containers (it should not do so, and instead should allow them to stay in stopped mode).
Containers will be labeled as follows:
- service: dokku=service dokku.service=SERVICE_NAME
- ambassador: dokku=ambassador dokku.ambassador=SERVICE_NAME
To destroy the containers, simply run the `service:destroy` command.
`expose` and `unexpose` commands are now fully functionnal using the
ambassador pattern. This has the advantage to be a 100% docker solution
and to not have to restart the service container. Multiple ports are
exposed when service specify them.
`start` and `stop` commands have been made more robust. When calling
`start` or `stop` on an already started (or stopped) service, an error
will be printed out instead of issuing failing commands.
Exposed services will still be exposed after a `restart`. Also the
service containers (and ambassador containers) will start automatically
when docker boot up.
note that at the moment the iptables calls complete successfully but the container isn't actually exposed. We'll probably need to use the ambassador pattern to do this properly.
