Autoscaling in Kubernetes: From Native Resource-Based Scaling to Event-Driven Scaling with KEDA

Containerized applications and microservice-based architectures have become the dominant paradigms for modern cloud-native systems. In such environments, efficiently adapting computational resources to workload variations is essential to guarantee both service performance and resource utilization efficiency. Kubernetes has emerged as the de facto standard platform for container orchestration, providing automated deployment, workload management and native autoscaling capabilities. This thesis investigates autoscaling mechanisms in Kubernetes environments, with a particular focus on Kubernetes Event-Driven Autoscaling (KEDA), an open-source framework that extends native Kubernetes scaling capabilities by enabling scaling decisions based on external events and application-specific metrics. After introducing the fundamental concepts of Kubernetes and its native autoscaling solutions, the architecture and operating principles of KEDA are analyzed. A set of practical demonstrations is first presented to illustrate the behavior of native Kubernetes autoscaling and selected KEDA scalers. The proposed approach is then evaluated through a more comprehensive case study based on a RabbitMQ workload and Prometheus metrics collected from a Kubernetes cluster. The experimental results highlight the trade-offs between static resource provisioning and event-driven autoscaling, showing how KEDA can dynamically scale the number of application instances in response to workload conditions while improving resource utilization and maintaining application responsiveness. These results demonstrate the effectiveness of KEDA as a flexible and efficient solution for autoscaling cloud-native applications in dynamic cloud-native environments.