Consider this a proposal for a v2 version of the Scheduler interface. Not necessarily looking to get this merged quickly.

The more I think about it, the more I realize that our Scheduler interface is just a generic interface for running 12factor apps against a scheduling backend, like ECS, Nomad, Kubernetes, etc.

Empire's sole purpose is to take a manifest, perform some transformations on it (merging in environment variables, release version, image) and then pass it off to a scheduler. I think there's a lot of benefits to making this fact more explicit in the code base.

Goals of this PR

1. Implement a higher quality ECS implementation. Better tests (functional tests against ECS), faster performance (parallelize as much as possible).
2. Define terminology for what we consider a "12factor" application, and provide Go types that mirror that definition (see package twelvefactor).
3. Support multiple provisioning backends for the ECS scheduler (cloudformation and raw api).
4. Potentially produce an isolated binary that can run 12factor applications without all of Empire. (e.g. Here's my Procfile and .env file, run this with the ECS backend)
5. Keep things generic enough so that other 12 factor PaaS's can use it.

This could probably be merged before we actually change Empire to start consuming it.

TODO

• Set logging driver: Add a command line argument to set the logConfiguration #731
• Set correct UpdatedAt in tasks: Use CreatedAt/UpdatedAt/StoppedAt from ECS API. #683
• Update SSL cert on load balancer: Use CreatedAt/UpdatedAt/StoppedAt from ECS API. #683
• Ulimits Support setting nproc limit #761
• ELB stuff
• Don't use TaskDefinition to get Task information, pull the tasks and then retrieve the task definition for the task.