Doesn't play well with another collector

[zalando-incubator]

Doesn't play well with another collector

[tomaspinho]

Is there a surefire way of introducing logic to the HPA provider to only target HPAs for which the collector is configured for? Or that have any annotations beginning with metric-config?

I'm willing to open a PR, just need some guidance

[mikkeloscar]

What is the actual impact you observe? Does it just print errors because it can't enable a collector for the HPA, or does it cause actual issues?

If it's just an error we can maybe maybe it less verbose?

[tomaspinho]

As it fails to create the collector, the errors get logged and inserted as events on the HPA. We're seeing thousands of such events, it's unnecessary noise. But there's no other impact.

[mikkeloscar]

Ah ok, I see what you mean now. Yeah, that is annoying if you don't intend to use it for this.

In the general case I think the event/log is still useful if you are only running the kube-metrics-adapter in your cluster, but it would probably make sense to add a flag for disabling reporting HPAs which doesn't match any collector plugin. What do you think?

[tomaspinho]

Yeah, that's the kind of thing I'd be willing to implement. What do you think about the two ways I suggested above?

[mikkeloscar]

Yeah, that's the kind of thing I'd be willing to implement. What do you think about the two ways I suggested above?

The best way would be to check if the error here:

kube-metrics-adapter/pkg/provider/hpa.go

Line 165 in 1209500

collector, err := p.collectorFactory.NewCollector(&hpa, config, interval)

is indicating that no plugin matched the HPA and then just not log the error/emit the event if you have this disabled via a flag.

Does that make sense?

[tomaspinho]

PR for this opened

[arjunrn]

Issue fixed with #95