Resource requests and limits on all pharos created pods

[captncraig]

I am trying my best to enforce a policy that all pods in the cluster have reasonable requests and limits set for memory and cpu.

A number of pharos components do not have these set, or have requests and no limits:

• metrics-server
• calico-node
• kube-controller-manager
• kube-proxy
• etcd
• kube-apiserver
• kube-scheduler
• pharos-proxy

Would it be reasonable for pharos to create limits on these containers? Some things like the apiserver are pretty critical, and should have high limits, but I understand it is still preferable to have a limit set for the scheduler to deal with. I've heard of failures when the host os starts oom killing pods instead of kubernetes doing it, and things get crazy fast. Perhaps the limits should be configurable?

[captncraig]

Perhaps a clean option would be an addon that patches the appropriate resources with limits from config. I will probably work on such a thing and publish it when I'm done.