Use static CPU Manager for Core Pinning

[mamoutou-diarra]

Context

RLN proof generation takes long time in the vaclab cluster because of the CPU frequency and the way kubernetes allocates cores to pods in non-dedicated manner: a core allocated to a pod is also used by other pods and by the system, this is particularly harmful for single-threaded apps, since they'll be constantly preempted and need to fight for cpu cycles.

Changes

• Set kubelet cpuManagerPolicy to static: guarantee pinning for guaranteed-qos class pods, the allocated core_id will never be allocated to another pod

Other Unrelated Changes

• Update dst_dashboard backend image to v1.3 (new feature: avoid boxplot to cross yaxis upon zoom; add vaclab node resources and topology via prometheus)

Note

The static cpuManager is live only on the metal-01 node for now, I'll do it on all the nodes after a couple of days of observation