Investigate what prevent us from using docker swarm mode instead of separate docker swarm

[garagatyi]

We need a bulleted list of features that are missing from swarm mode in Docker 1.12 that prevent us from using that instead of the Docker Swarm.

[garagatyi]

Below I listed problems that I found during investigation of Docker swarm mode:

1. It doesn't support build of images, so we have to drop that capability for users or reimplement it fully. It would take a lot of time.
2. It doesn't support execs, commit, tag, push, get container logs,
3. It doesn't support privileged mode, pids, swap, disk limits, extra hosts
4. It doesn't support centralized retrieval of container events.

Apart from 1 we probably can also implement 2 ourselves. It would be something like finding node address and container id for machine and connect to target docker daemon directly to perform needed operation. It is big downgrade from what we have with Docker swarm as a separate product. It is also quite error prone approach.

Below I listed questions I haven't found answer to during this investigation:

1. Is it OK for us that it doesn't support whole syntax of Docker compose? Will users be OK with such downgrade?
2. I haven't found is it possible to create isolated overlay networks for each workspace.
3. How to know when container started (except performing some ls commands periodically)?
4. Does command+ args of service represents entrypoint + cmd of container?
5. Is it possible to use volumes from?
6. How to implement container links?

Probably we should investigate how to address 2, 3, 4, 5 and we could find some solutions (appropriate for us or not) in investigations of each specific problem.