Kubernetes Event-Driven Autoscaler

Kubernetes Event-Driven Autoscaler (KEDA) is a component that can enhance a Kubernetes deployment by enabling scaling based on a variety of triggers. KEDA, originally created by Microsoft and Red Hat in 2019, is hosted at the Cloud Native Computing Foundation (CNCF) as an “incubating project.”

KEDA employs the concept of “scalers”—adapter-like code elements that monitor specific conditions and trigger scaling events, including scaling to zero instances. For example, when CPU utilization reaches a certain threshold, KEDA can invoke the Kubernetes Horizontal Pod Autoscaler (a built-in Kubernetes component) to create new instances or remove them if utilization decreases. Similarly, a scaler might detect a message bus becoming heavily loaded and scale the cluster in response. Scaling rules (provided in JSON format) define the thresholds as well as maximum and minimum numbers of pod instances (replicas).

Scalers are available for hardware utilization metrics such as CPU or memory, as well as for numerous software packages, such as Apache Kafka, Azure Service Bus, HTTP, MongoDB, Redis, and many others.