Implementation and Benchmarking of Kubernetes Horizontal Pod Autoscaling Method to Event-Driven Messaging System

The rapid development of technology is accompanied by a large number of demands and users. One of the solutions is to create reliable infrastructure that can support application performance. Kubernetes is a platform which offers one of the solutions by providing container application orchestration and enabling high availability and scalability through various automatic scaling mechanisms such as Horizontal Pod Autoscaler (HPA), which dynamically scales the number of pod resources without restarting the entire system. By default, Kubernetes only monitors built-in resource metrics such as CPU and memory usage for each host machine and pod. To use Custom Metrics, external software such as Prometheus can be employed to monitor metrics as needed. On the other hand, Kubernetes Event-driven Autoscaling (KEDA) software offers a solution to simplify and facilitate autoscaling by efficiently implementing event-based automatic scaling and achieving the scale-to-zero capability. In this paper, we investigate the performance of autoscaling through various experiments to understand the behavior of each custom metrics autoscaling solution.