GeeCON 2023: Daniel Oh - Event-driven autoscaling for Serverless Java

Kubernetes makes it possible to autoscale various business usage cases from web apps to mobile, IoT edge streaming, and AI/ML in more reliable and unchangeable ways. 1 caveat of Kubernetes autoscaling is based on hardware resource utilization (CPU, memory) through Horizontal Pod Autoscaling. This causes a fresh challenge to build an event-driven serverless Java on Kubernetes due to the fact that the event metrics from multiple event sources (e.g., Apache Kafka, AWS SQS) are more applicable than a pod's CPU usage for deciding erstwhile applications request to be scaled out and in. Fortunately, KEDA and Knative on Kubernetes are designed to solve this challenge by autoscaling both standard apps and serverless by event metrics in a separate way. This session will teach you how to redesign your Kubernetes autoscaling architecture by event-driven metrics from Apache Kafka over standard resources (CPU, Memory) with Knative and KEDA integration for serverless Java utilizing Quarkus.