feat: add requested resources info to servers endpoint

[lorenzo-cavazzi]

Add the requested resources info to the GET /servers API response. This is an example, the new part is in the attribute resources

{
  "servers": {
    "flights-tutorial-d991bad2": {
      "annotations": {
        [...]
      },
      "name": "flights-tutorial-d991bad2",
      "resources": {
        "cpu": "0.1",
        "memory": "1073741824"
      },
      "started": "2020-03-12T16:54:30.000000Z",
      "state": {
        "pod_name": "jupyter-lorenzo-2ecavazzi-2etech-flights-2dtutorial-2dd991bad2"
      },
      "status": {
        [...]
      },
      "url": "https://lorenzo.dev.renku.ch/jupyterhub/user/lorenzo.cavazzi.tech/flights-tutorial-d991bad2/"
    }
  }
}

UPDATE: It's possible to test the change in lorenzotest deplyoment: https://lorenzotest.dev.renku.ch/

fix #223

[rokroskar]

Thanks @lorenzo-cavazzi! I'm wondering if it would make sense to show both resource request and limit? otoh it might make the most sense to enforce both the memory and cpu requests as limits. At the moment, only the memory is enforced. Then we only need to show a single number, like you have done here.

[lorenzo-cavazzi]

We don't get any cpu value in the limits object if we don't enforce one. Currently, we only get the memory limit and it's always equal to the request counterpart.
On the other hand, do we want to enforce cpu limits? Is there any advantage? I see how allocating extra memory may be a problem when we need to reclaim that, but for the cpu I don't see the same problem

[rokroskar]

We don't get any cpu value in the limits object if we don't enforce one

yes, that was my point - maybe we should. What can happen is that one user can starve a node of CPU. CPU is a compressible resource, meaning the OS can decide to throttle it or not. If you don't enforce CPU limits, a single greedy process can take over all of the CPU on the node adversely affecting all others.

[rokroskar]

Looks good! We can decide if we need to change the CPU policy at a later point.

[lorenzo-cavazzi]

I was going to look into enforcing the CPU limit since your argument is convincing. Should I open a different issue to keep track of that?

[rokroskar]

You just need to add the CPU limit to the pod in the spawner in the same way it's done for the memory. But maybe it warrants a discussion with the others - although the decision is either: a) we hope that the current way doesn't make too many people unhappy or b) we enforce it and then document the limitations