Creating a new ScaledObject with the “paused” annotation is not working

[Dark0096]

Report

There is an issue where, after creating the initial ScaledObject with the “paused” annotation set, when resuming, the ScaledObject correctly generates the HPA, but the "Paused" condition state becomes True again, and the HPA remains in place.

Expected Behavior

1. Init condition(Create the ScaledObject)
   Ready: True
   Active: True
   Fallback: False
   Paused: True
   HPA: None

2. After resume(Paused Annotation Removed)
   Ready: True
   Active: True
   Fallback: False
   Paused: False
   HPA: New one created by the scaled object

Actual Behavior

1. Init condition(Create the ScaledObject)
   Ready: True
   Active: Unknown
   Fallback: Unknown
   Paused: True
   HPA: None

2. After resume(Paused Annotation Removed)

2-1. Init
Ready: True
Active: Unknown
Fallback: Unknown
Paused: False
HPA: New one created by scaled object

2-2. After a few seconds
Ready: True
Active: Unknown
Fallback: False
Paused: True
HPA: New one created by scaled object

Here is a video from the Lens app demonstrating the issue

• https://github.com/user-attachments/assets/1bf899cf-2e69-4c47-8a5a-f40421534e23

Steps to Reproduce the Problem

1. Write a ScaledObject manifest including the CPU and Cron Scalers.
2. Add the paused annotation to the created manifest.
3. Apply the created manifest using kubectl apply.

Logs from KEDA operator

There are no specific logs available.

KEDA Version

2.15.1

Kubernetes Version

1.30

Platform

Amazon Web Services

Scaler Details

cpu, memory, cron

Anything else?

Hi,

I’ve been using KEDA effectively, and I have a question.
Has the pattern of initially creating a ScaledObject resource with a paused annotation by default and resuming it as needed been considered?

If there are any parts of the code you suspect might be relevant, please let me know, and I’ll try making modifications myself.

Thank you!

[vtomasr5]

We have a similar issue with pausing.

If we add the autoscaling.keda.sh/paused-replicas: "10" annotation to the ScaledObject (SO) we see those logs in the keda-operator:

...
2024-12-18T15:49:18Z	ERROR	Reconciler error	{"controller": "scaledobject", "controllerGroup": "keda.sh", "controllerKind": "ScaledObject", "ScaledObject": {"name":"querier","namespace":"mimir"}, "namespace": "mimir", "name": "querier", "reconcileID": "c931a2a2-0820-4847-9fd3-e88987b09e7e", "error": "ScaledObject paused replicas are being scaled"} 
  |   | /workspace/vendor/sigs.k8s.io/controller-runtime/pkg/internal/controller/controller.go:227 |  
  |   | sigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).Start.func2.2 |  
  |   | /workspace/vendor/sigs.k8s.io/controller-runtime/pkg/internal/controller/controller.go:266 |  
  |   | sigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).processNextWorkItem |  
  |   | /workspace/vendor/sigs.k8s.io/controller-runtime/pkg/internal/controller/controller.go:316 |  
  |   | sigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).reconcileHandler |  
  |   | /workspace/vendor/sigs.k8s.io/controller-runtime/pkg/internal/controller/controller.go:119 |  
  |   | sigs.k8s.io/controller-runtime/pkg/internal/controller.(*Controller).Reconcile |  
  |   | /workspace/controllers/keda/scaledobject_controller.go:196 |  
  |   | github.com/kedacore/keda/v2/controllers/keda.(*ScaledObjectReconciler).Reconcile
...

Which comes from this code

The Active type is Unknown, an issue for the ArgoCD syncing process that gets stuck forever.

status:
  conditions:
  - message: ScaledObject check failed
    reason: UnknownState
    status: Unknown
    type: Active

[SpiritZhou]

I believe there is a bug, but I am unable to reproduce it. Would you be able to provide the ScaledObject YAML?

[rickbrouwer]

I can only reproduce this if I specify both in a conflicting situation (both false and specifying paused-replicas). This:

annotations:
    autoscaling.keda.sh/paused-replicas: "2"
    autoscaling.keda.sh/paused: "false"

I think it would be clearer if autoscaling.keda.sh/paused will be leading? So if it is set to false, it is really paused regardless of whether paused-replicas is filled?
It am also curious if Dark0096 and vtomasr5 specify both or not. If not, I am also curious about the ScaledObject.yaml.

[Dark0096]

Hi, @SpiritZhou, @rickbrouwer.

This issue occurred when paused was used during the initial creation of a new resource.
Below is a reproducible example, including the YAML configuration.

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  annotations:
    autoscaling.keda.sh/paused: 'false'
  labels:
    scaledobject.keda.sh/name: resource-name-1
  name: resource-name-1
  namespace: deployment-dev
spec:
  advanced:
    horizontalPodAutoscalerConfig:
      behavior:
        scaleDown:
          policies:
            - periodSeconds: 60
              type: Percent
              value: 30
          selectPolicy: Min
          stabilizationWindowSeconds: 300
        scaleUp:
          policies:
            - periodSeconds: 15
              type: Pods
              value: 4
            - periodSeconds: 15
              type: Percent
              value: 100
          selectPolicy: Max
          stabilizationWindowSeconds: 0
      name: resource-name-1
    scalingModifiers: {}
  cooldownPeriod: 300
  maxReplicaCount: 3
  minReplicaCount: 1
  pollingInterval: 30
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: resource-name-1
  triggers:
    - metadata:
        desiredReplicas: '1'
        end: 30 1 * * *
        name: lateNight
        start: 0 1 * * *
        timezone: Asia/Seoul
      type: cron
    - metadata:
        desiredReplicas: '3'
        end: 30 9 * * *
        name: earlyMorning
        start: 0 9 * * *
        timezone: Asia/Seoul
      type: cron
    - metadata:
        desiredReplicas: '3'
        end: 0 16 * * *
        name: earlyEvening
        start: 30 15 * * *
        timezone: Asia/Seoul
      type: cron
    - metadata:
        value: '55'
      metricType: Utilization
      type: cpu

Please feel free to let me know if you need any additional information.

[rickbrouwer]

I use Keda 2.16.0 (with ArgoCD v2.12.8).

The only thing I can reproduce is the following:

When i initially place the above scaledObject with only a paused with value false I get the following status condition:

- status: Unknown
type: Paused

When I look at the code, an initial creation with the value false will give an Unknown status because the condition has never been true.

With a little adjustment in the reconcileScaledObject we can give a better status and message.