fix(endpoints): scrape endpoints and not services

[paologallinaharbur]

Fix #27

The idea is that when a service is detected the underlying endpoints are fetched. For backward compatibility the metric directly coming from the service is kept

Two new config options have been added ScrapeServices (default true) and ScrapeEndpoints(default false)

After merging we should update documentation and the helm chart to ask for listing/watching permission on endpoints

[paologallinaharbur]

Currently after merging this we will be sending more data than before.

We should decide to filter out some targets, es:

func (k *KubernetesTargetRetriever) GetTargets() ([]Target, error) {
	length := 0
	k.targets.Range(func(_, _ interface{}) bool {
		length++
		return true
	})
	targets := make([]Target, 0, length)
	k.targets.Range(func(_, y interface{}) bool {
		for _, t := range y.([]Target) {
			if t.Object.Kind == "service" && excludeServiceTargets {
				continue
			}
			if t.Object.Kind == "endpoints" && excludeEndpointsTargets {
				continue
			}
			targets = append(targets, t)
		}
		return true
	})

	return targets, nil
}

[gsanchezgavier]

I think we need to have these behind a feature flag keeping the old behavior by default. Or have a mayor release. I would also propose to have this feature in beta for some time.

Could you try the load test to se how is the result with the new feature?

[paologallinaharbur]

@gsanchezgavier I added two feature flags and given what we talked about offline in the past days, for now the option to scrape endpoints will be disabled by default.

I'll update the documentation and the HelmChart to further clarify this

[gsanchezgavier]

lgtm

[paologallinaharbur]

After merging and releasing we need to update documentation and expose the new config through the helm charts