Hi, A recent pull-request has raised the question of what to do with containers, specifically rkt and docker.
This PR proposes docker as a compute node driver, https://github.com/apache/libcloud/pull/654 As you can see, the implementation doesn't line up perfectly with the node driver, which was really designed for VM hypervisors. There are some alternatives here to implement a better option. Please consider the following options and vote accordingly: 1) proceed with docker as a compute driver as stated in the #654 PR 2) create a new driver type, ContainerNodeDriver, which INHERITS from NodeDriver, list_nodes would be overloaded to return ContainerNode's instead of Nodes. a ContainerNode (which inherits from Node) would share the same properties, but also have an additional property stating which Node it belongs to. There would be some other methods like pull_image, etc. That way users can use list_nodes etc. but also take advantage of the other functions. 3) create a new driver type, ContainerDriver and start from scratch, just design and implement which methods based on usage patterns in rkt and docker 4) do not implement docker and/or rkt drivers at all Regards, Anthony
