Cast AI's Workload Autoscaler is fully compatible with KEDA (Kubernetes Event-driven Autoscaling) when KEDA is configured to scale standard Kubernetes workloads using KEDA ScaledObjects. Our system recognizes and works seamlessly with it, the same as with native Kubernetes HPAs.

How KEDA and Workload Autoscaler work together

KEDA creates HPA objects based on your ScaledObject configurations. Cast AI's Workload Autoscaler integrates with these KEDA-generated HPAs and applies different optimization strategies depending on which metrics are configured in your ScaledObject.

Compatibility

Cast AI's Workload Autoscaler works with KEDA ScaledObjects. KEDA ScaledJobs are not supported.

KEDA scaling optimization strategies

The level of optimization applied by Workload Autoscaler depends on which metric types are configured in your KEDA ScaledObject.

KEDA with only CPU or memory metrics

When your KEDA ScaledObject uses only CPU or memory metrics:

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: cpu-only-scaler
  namespace: default
spec:
  scaleTargetRef:
    name: my-deployment
  triggers:
  - type: cpu
    metadata:
      type: Utilization
      value: "70"

Workload Autoscaler will:

• Run the full simulation-based optimization, evaluating CPU candidates across cost, stability, and replica behavior factors
• Calibrate the CPU recommendation away from the raw percentile when a nearby value produces better workload behavior

KEDA with CPU/memory combined with custom metrics

When your KEDA ScaledObject combines CPU or memory metrics with custom metrics:

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: combined-metrics-scaler
  namespace: default
spec:
  scaleTargetRef:
    name: my-deployment
  triggers:
  - type: cpu
    metadata:
      type: Utilization
      value: "70"
  - type: prometheus
    metadata:
      serverAddress: http://<prometheus-host>:9090
      threshold: '100'
      query: sum(rate(http_requests_total{deployment="my-deployment"}[2m]))

Workload Autoscaler will:

• Run the simulation and apply vertical scaling optimization (CPU/memory resource requests)
• Limit how far the recommendation moves away from the originally detected percentile
• Respect KEDA's custom metric scaling decisions

Because custom metrics primarily drive scaling decisions for these workloads, CPU-based corrections may not reflect actual throughput. The simulation still runs, but calibration precision is reduced compared to CPU- or memory-only configurations.

KEDA with only custom metrics

When your KEDA ScaledObject uses only custom metrics (no CPU or memory):

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: prometheus-scaledobject
  namespace: default
spec:
  scaleTargetRef:
    name: my-deployment
  triggers:
  - type: prometheus
    metadata:
      serverAddress: http://<prometheus-host>:9090
      threshold: '100'
      query: sum(rate(http_requests_total{deployment="my-deployment"}[2m]))

Workload Autoscaler will:

• Continue vertical scaling optimization (CPU/memory resource requests)
• Not apply the intelligent vertical and horizontal scaling optimization algorithms
• Respect KEDA's custom metric scaling decisions