If I set minReplicaCount to 0 in the ScaledObject why is the HorizontalPodAutoscaler minReplicas set to 1?

[ecerulm]

My ScaledObject has idleReplicaCount and minReplicaCount set to 0

spec:
  idleReplicaCount: 0
  minReplicaCount: 0
  maxReplicaCount:3

but the HPA generated by KEDA has minReplicas: 1 , why is that? is that the expected behaviour?

I know keda is reading the values because if I set minReplicaCount: 2 the HPA gets updated to minReplicas: 2, but if change to minReplicaCount: 0 then the HPA gets minReplicas: 1

For my setup it looks like KEDA tries to scale down the deployment to 0 replicas, and immediately after it scales from 0 to 1 (I'm assuming it's because of the HPA).

[zroubalik]

HPA by default doesn’t support min = 0. Feel free to ignore, ScaledObject is the important thing here. KEDA manages 0-1 scaling. https://keda.sh/docs/2.13/concepts/#architecture

[ecerulm]

Repeating the test and monitoring the HPA during the first trigger I can see that at the very beginning

  AbleToScale    True    SucceededGetScale  the HPA controller was able to get the target's current scale
  ScalingActive  False   ScalingDisabled    scaling is disabled since the replica count of the target is zero

Then it changes to

I'm no expert (actually started trying KEDA yesterday) but it looks like it takes the external metric "s1-cron-Europe-Stockholm-12xxxx-13xxxx" that tells that it should be 1 replica.

[ecerulm]

Anyway I think I'll write an issue instead, thanks for the help @zroubalik

[zroubalik]

@ecerulm do you see this #3609 ?

[ecerulm]

I just scanned through #3609, #1759 and #2153 and if I'm not mistaken they say:

• that in their case it did not scale to 0 because they had minReplicaCount: 1 but I have minReplicaCount: 0
• it's should only be used for scale up, don't use desiredReplicas: 0 ever , specify the timewindow where you want more replicas instead and let it go to minReplicaCount outside that timewindow.

And I think that way it works, so assuming that desiredReplicas: 0 is never supported I will write a PR to change the documentation for cron scaler https://keda.sh/docs/2.13/scalers/cron/ so that this case is clear there.

And I will write an issue to ask for a validation that forbids desiredReplicas: "0" on the cron scaler. It's still bafflic that minReplicaCount: 0 and desiredReplicas: "0" together produce target replicas 1 .

Thanks

[ecerulm]

For reference

• Cron Scaler: desiredReplicas is 0 but it sets to 1 anyway #5586
• cron scaler: clarify desiredReplicas 0 keda-docs#1332

[ecerulm]

Expanding on the answer by @zroubalik, with some info that @JorTurFer provided in slack:

The HPA object itself does not support setting minReplicas: 0, KEDA controller itself handling the scaling of the Deployment from replicas: 1 to replicas: 0. When the HPA controller see that the workload (in this case a Deployment) has replicas: 0 it "disables" itself , meaning it won't try manage it and it won't try to scale to replicas: 1.