[Metrics] Kubernetes with a kubectl proxy and Prometheus do not work together

[jacekwachowiak]

As in https://github.com/coreos/prometheus-operator/issues/2392 and https://github.com/prometheus/prometheus/issues/1583#issuecomment-214661546 and https://github.com/coreos/prometheus-operator/issues/287 if we run Clipper and use the kubectl proxy, Prometheus will be available at the expected address but no variables or graphs can be created using the web GUI. I will check if I can fix that on my cloud with the extra argument that is mentioned in the threads.

[jacekwachowiak]

So good news is that using a LoadBalancer works, but I have not been successful in crating graphs on the NodePort metrics - adding the external url argument did not work for me.

[rkooo567]

@jacekwachowiak Did you figure out anything else? Seems like it should be a bug.

[jacekwachowiak]

It seems like a Prometheus bug. The problem is when you have to use the kubectl proxy and you don't have direct acccess to the master node - NodePort service is useless then, the LoadBalancer type is the solution. Anyway, it seems that this is a Prometheus design problem. I am running the default Prometheus and an external Grafana and everything is fine too.

[rkooo567]

Hmm that's weird. Let's keep tracking this issue and see if we can fix this later.