KEDA doesn't validate empty array of triggers

[SpiritZhou]

Report

When triggers is set with an empty array in ScaledObject, KEDA doesn't validate this empty value and will create a default HPA with a default 80% averageUtilization CPU resource. This can mislead users.

Here is the yaml example:

---
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: kafka-scaledobject
spec:
  scaleTargetRef:
    name: sut
  pollingInterval: 30
  minReplicaCount: 3
  maxReplicaCount: 5
  triggers: []

BTW, KEDA will block creation if the value of triggers is empty.

Expected Behavior

Block this creation of ScaledObject if triggers is set with an empty array.

Actual Behavior

ScaledObject is created normally and a default HPA will also be created.

Steps to Reproduce the Problem

1. Create an ScaledObject with an empy-array triggers.
2. A CPU-resource HPA will be created soon.

KEDA Version

2.13.0

Kubernetes Version

1.27

Platform

Other

Scaler Details

No response

Anything else?

No response

[vinod827]

When triggers is set with an empty array in ScaledObject, KEDA doesn't validate this empty value and will create a default HPA with a default 80% averageUtilization CPU resource. This can mislead users.

hmmm....that's interesting. From where this 80% of averageUtilization for CPU is coming in the HPA?

[SpiritZhou]

When triggers is set with an empty array in ScaledObject, KEDA doesn't validate this empty value and will create a default HPA with a default 80% averageUtilization CPU resource. This can mislead users.

hmmm....that's interesting. From where this 80% of averageUtilization for CPU is coming in the HPA?

It's K8s default behavior. When I create an HPA directly without metrics spec, k8s will create an HPA with 80% of averageUtilization for CPU resource.

HPA example:

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: test
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: sut
  minReplicas: 1
  maxReplicas: 10

[vinod827]

So that means HPA should not be defaulted to as a scaling option when triggers are empty in the KEDA's ScaledObject

[JorTurFer]

I think that we should validate it as part of admission hooks and also raise an error in the operator. Adding default value is quite weird as it's totally unrelated with the given triggers. This doesn't happen just with empty triggers but also when there are some issues in the creation loop

[zroubalik]

I agree, we should validate both on the Operator and validation webhook side. This is an ideal "good first issue " :)

[joebowbeer]

It's K8s default behavior. When I create an HPA directly without metrics spec, k8s will create an HPA with 80% of averageUtilization for CPU resource.

Yes, this is k8s default behavior, as described in both k8s and KEDA documentation, and we have been relying on it.

I think this new behavior is a breaking change and a regression.

Can this be reverted? Or at least, can an option to configure it be added?