Tool
KEDA (Kubernetes Event-driven Autoscaling) is a lightweight Kubernetes component for scaling containerized workloads according to event-driven signals. It works alongside the Kubernetes Horizontal Pod Autoscaler and uses ScaledObjects and scalers to map applications to event sources, including queues, databases, metrics systems, cloud services, and HTTP traffic. KEDA supports deployments, jobs, and custom resources with a /scale sub-resource, including scaling workloads down to zero and back up when events require processing. Its HTTP add-on provides an interceptor that can receive or hold requests while an application has no active pods and help trigger scale-up. The project provides built-in scalers, supports custom and community-maintained scalers, and is vendor-agnostic across cloud providers and products.
1 use taken from transcripts — each links to the moment in the video.
Uses an HTTP interceptor and scaled object to count requests, hold or reject requests while the model is cold, and scale the inference deployment down to zero or back up.
1 in the library.