Beyond HPA, VPA, and KEDA: Proactive Kubernetes autoscaling using deep Q-networks

Cloud native architecture is about building and running scalable microservice applications to take full advantage of cloud environments. Managed Kubernetes orchestrates cloud native applications with elastic scaling. However, traditional Kubernetes autoscalers are reactive, meaning scaling controllers adjust resources only after detecting demand within the cluster and do not incorporate predictive measures. This can lead to overprovisioning and increased costs or underprovisioning and performance degradation. In this talk, Indrajith discusses the possibility of using a reinforcement learning agent to provide proactive autoscaling. This approach is compared with built-in scaling controllers such as Horizontal Pod Autoscaler and the event-driven autoscaler KEDA. Experimental results demonstrate how a proactive framework can improve performance and cost efficiency compared with existing reactive methods.