RFC: update resource management in Kata

[egernst]

Upate Resource Management in Kata

There are gaps and issues around our cgroup / resource management utilization in Kata. This is a meta issue to cover some of the work items.

There are a few PRs in flight around cgroup fixing, and I'd like to get the design and work tracked. To this end, I'm creating a PR with a design.md, to facilitate comments/discussion around design. We can discuss in an AC call aftr folks have enough time to review. Once completed, we can move the resulting doc to the documentation repository.

I ceated and am using the GitHub Milestone "cgroup-sprint" for this work. I'd really like to see this resolved for 1.7-rc0 release, which is in 3 weeks. Let's see what we can do.

Work Items:

• - agree to design proposal. See resource management: document current status, suggested fixes documentation#428
• - don't confuse CAdvisor by creating cgroups on host with container ID in the kubepod. See virtcontainers: prepend a kata specific string to host cgroups path #1518
• - guarantee guaranteed performance when using guaranteed QoS
• - Better handling of unbound burstable pods. See Incorrect implementation of burstable and best effort pods #1534