Kubeadm: RFE: Audit default limits and priorities

[timothysc]

We need to update limit defaults and make certain we are using sane values for priorities. We have legacy cruft that also needs to be audited.

xref: https://github.com/kubernetes/kubernetes/pull/69734

[fabriziopandini]

This is spreading, even if with small variants:

https://github.com/kubernetes/kubernetes/pull/72424 (control plane components- resouce request & QoS/PriorityClass)
https://github.com/kubernetes/kubeadm/issues/1095 (control plane components - resouce request)
https://github.com/kubernetes/kubeadm/issues/1337 (core DNS - QoS/PriorityClass)

[yagonobre]

I can work on this
/lifecycle active

[yagonobre]

/assign

[fabriziopandini]

@yagonobre fyi there is also https://github.com/kubernetes/kubernetes/pull/73649

[timothysc]

needs
priorityClassName: system-cluster-critical

and remove older field (TBD)

[ereslibre]

@yagonobre as per our discussion today, are you OK with me having a look at this?

[ereslibre]

Certainly, we were using scheduler.alpha.kubernetes.io/critical-pod annotation. priorityClassName is already set to system-cluster-critical on all control plane components. I will audit everything to remove the old annotation.

[yagonobre]

@ereslibre go ahead :)
/unassign

[yagonobre]

IMO we should add request/limit too.