This is one of my favorite feature launches ever. Let me tell you why.
One of the best parts of EC2 was that you were no longer concerned how many racks you had in your data center, where they sat, etc. You just launched instances onto capacity that AWS managed.
Modern container orchestration, especially single tenant systems essentially regressed and made that capacity a "thing" by choosing the cluster of machines a first order construct.
You should be thinking about launching a service or Task and should not care about which maching in your cluster those are running. So the question becomes "how do you abstract capacity away again?" Obviously there's Fargate, which was an answer to that question.
But we can do even more. And this is where capacity providers come in and they take advantage of all the automation that modern container schedulers give you. Capacity Providers move the focus from cluster scheduling to task and service scheduling.
While we are starting simple, over time our goal is to make boundaries between different types of capacity super fuzzy and driven by your business or application requirements, which you can express in a specfile. Let us know how you would like us to evolve this new concept.
