Skip to main content

KEDA Cron Scale-to-Zero for Dev Environments

Implemented: 2026-07-07 Applies to: platform-demo-dev, platform-demo-staging

Dev and staging environments don't need to run overnight or on weekends. A KEDA ScaledObject with a cron trigger automatically scales them to 0 replicas outside business hours and back up when the day starts — no manual intervention, no wasted compute.

How it works​

KEDA creates a Kubernetes HPA that owns spec.replicas on the target Deployment. The cron trigger defines an active window. Outside that window, KEDA drives replicas to minReplicaCount (0). Inside the window, it drives to desiredReplicas.

Mon–Fri 00:00            08:00                19:00          23:59
         │                 │                    │               │
replicas ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░░░░░░░░░░░░▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
         0                 1 (dev) / 2 (staging) 0

Sat–Sun: replicas = 0 all day

ScaledObject spec​

Dev overlay (overlays/dev/scaledobject.yaml)​

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: platform-demo-cron-scaler
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: platform-demo
  minReplicaCount: 0
  maxReplicaCount: 1
  cooldownPeriod: 60
  triggers:
    - type: cron
      metadata:
        timezone: Europe/Paris
        start: "0 8 * * 1-5"
        end: "0 19 * * 1-5"
        desiredReplicas: "1"

Staging overlay (overlays/staging/scaledobject.yaml)​

Same spec, maxReplicaCount: 2 and desiredReplicas: "2" — staging runs two replicas for anti-affinity spread testing during the day.

ArgoCD integration — the replicas conflict​

When KEDA manages a Deployment, it sets spec.replicas to 0 at night via the HPA it creates. ArgoCD compares the live cluster state against Git. The Git manifest has replicas: 1 (from patch-resources.yaml). Without additional config, ArgoCD selfHeal would restore replicas to 1 every time KEDA scales to 0 — the two would fight in a loop.

Fix: add ignoreDifferences for spec.replicas on the Deployment, and RespectIgnoreDifferences=true in syncOptions so the ignored field also affects the OutOfSync status calculation:

# apps/platform-demo-dev.yaml
syncPolicy:
  automated:
    prune: true
    selfHeal: true
  syncOptions:
    - ServerSideApply=true
    - RespectIgnoreDifferences=true
ignoreDifferences:
  - group: apps
    kind: Deployment
    name: platform-demo
    namespace: platform-demo-dev
    jqPathExpressions:
      - .spec.replicas

Without RespectIgnoreDifferences=true, ArgoCD ignores the field only during sync but still reports OutOfSync in the UI — the app permanently shows yellow. With the option, both the sync and the status calculation skip the field.

Kustomize wiring​

Each overlay's kustomization.yaml includes the ScaledObject as a resource:

resources:
  - ../../base
  - scaledobject.yaml

The ScaledObject lives in the overlay, not the base, because:

  • It sets minReplicaCount / maxReplicaCount which are environment-specific
  • Prod does not get a cron ScaledObject (always-on)
  • Each overlay can define a different schedule or replica count

Prod is excluded by design​

The overlays/prod/ directory has no scaledobject.yaml. Prod runs 3 replicas continuously. Cron scaling on prod creates the risk of a demo or customer request arriving at 19:01 and hitting a deployment that is still warming up. The cost saving does not justify the availability trade-off.

Schedule reference​

FieldValueMeaning
timezoneEurope/ParisCET/CEST — adjusts automatically for DST
start0 8 * * 1-5Scale up at 08:00 Monday–Friday
end0 19 * * 1-5Scale down at 19:00 Monday–Friday
minReplicaCount00 replicas when no trigger is active
cooldownPeriod60Seconds KEDA waits before scaling down after trigger deactivates

Nights (19:00–08:00) and full weekends → 0 replicas. Approximately 70% of wall-clock hours the dev/staging namespaces consume zero compute.

Extending to a new service​

  1. Create overlays/dev/scaledobject.yaml in your service directory, targeting your Deployment by name.
  2. Add - scaledobject.yaml to resources: in the overlay kustomization.yaml.
  3. Add ignoreDifferences for spec.replicas to the ArgoCD Application in apps/<your-service>-dev.yaml.
  4. Repeat for overlays/staging/ if staging should also scale to zero.

No changes to the base or prod overlay needed.

Verification​

# Check ScaledObject status
kubectl get scaledobject -n platform-demo-dev
kubectl get scaledobject -n platform-demo-staging

# ACTIVE=True  → inside business hours, running at desiredReplicas
# ACTIVE=False → outside window, scaled to 0

# Check the KEDA-managed HPA
kubectl get hpa -n platform-demo-dev

# Current replica count
kubectl get deployment platform-demo -n platform-demo-dev