Should we return error when target lost for prometheus scaler

[bamboo12366]

Report

when Prometheus target is lost, the prometheus query will return an empty result here:
https://github.com/kedacore/keda/blob/main/pkg/scalers/prometheus_scaler.go#L55

The hpa controller will get an 0 value and using the minReplicas. Maybe we should return an error when len(reslut) = 0? To let the hpa forget about this workload for a while until the prometheus is back to normal?

There are related issue regarding to this: #657. But I think we shouldn't patch the replicas when the dependency is not available. Maybe somebody can share more thoughts about that?

Expected Behavior

keda-metrics return 0 when prometheus is unavailable, hpa controller is patching the minReplicas to workload

Actual Behavior

keda-metrics should return error when prometheus is unavailable, hpa controller should maintain the same replicas

Steps to Reproduce the Problem

1. Lost prometheus target

Logs from KEDA operator

example

KEDA Version

2.7.1

Kubernetes Version

1.20

Platform

Any

Scaler Details

No response

Anything else?

No response

[zroubalik]

@bamboo12366 is this somehow related to #3065?

[bamboo12366]

@bamboo12366 is this somehow related to #3065?

seems the same thing, can I give a try if no one is working on this?

[zroubalik]

@bamboo12366 yeah that would be great, let me close this issue and continue on the other one.

Duplicate of #3065