Kedify Scalers
Scalers determine how and when autoscaling should be activated or deactivated, and they can also provide custom metrics tailored to a particular event source. In a typical Kedify setup, the scaler feeds demand into KEDA and Kubernetes HPA applies the resulting replica changes.
Which Scaler Should You Start With?
Section titled “Which Scaler Should You Start With?”- HTTP Scaler - scales apps based on incoming HTTP requests incl. scale to zero
- HTTP Scaler for Inference - scales inference workloads based on incoming HTTP requests incl. scale to zero
- Envoy HTTP Scaler - scales apps based on HTTP requests by using custom Envoy proxy
- OTEL Scaler - scales apps based on OpenTelemetry metrics
- Vertical Scalers - combines PRP declarative transitions and PRA instant utilization-based autoscaling
- Built-in KEDA Scalers - Built-in open source KEDA scalers
How to Choose
Section titled “How to Choose”- Choose Built-in KEDA Scalers when a standard upstream event source already matches your workload.
- Choose HTTP Scaler when request traffic is the scaling signal and scale-to-zero behavior matters.
- Choose OTEL Scaler when you already use OpenTelemetry or want custom-metric autoscaling without a full Prometheus stack.
- Choose Predictive Scaler when reactive autoscaling is consistently too late.
- Choose Vertical Scalers when bottlenecks are caused by per-pod CPU or memory sizing rather than replica count.