As of this writing, some of the k8s control plane components, in particular the k8s apiserver, eats a good chunk of the resources of the virtual machine they run on.
For example, from the k9s screenshot above, the apiserver is using:
- 7.2GB of RAM on control-7
- 3.6GB of RAM on control-8
- 4.5GB of RAM on control-9
As of this writing, the VM have 15GB RAM total, which may give little room for the system to survive load spikes and other unexpected scenarios.
This ticket is to evaluate if we should scale the k8s control plane, either or both:
- horizontally (more nodes)
- vertically (bigger nodes)