REACH: Reinforcement Learning for Adaptive Microservice Rescheduling in the Cloud–Edge Continuum

Abstract

Cloud computing, despite its scalability advantages, may not fully satisfy the low-latency demands of emerging latency-sensitive applications. The cloud–edge continuum addresses this limitation by integrating the responsiveness of edge resources with cloud scalability. Microservice architecture (MSA), characterized by modular, loosely coupled services, aligns effectively with this continuum. However, heterogeneous and dynamic computing resources pose significant challenges to optimal microservice placement. Most existing approaches focus on generating one-time scheduling plans, which are ill-suited to dynamic environments where frequent and lightweight rescheduling actions are required in response to changing system conditions. We propose REACH, a reinforcement learning-based microservice rescheduling framework that enables a sim-to-real deployment pipeline for adapting microservice placement under fluctuating resource availability and performance variations across distributed infrastructures. REACH is integrated with a real Kubernetes-based cloud–edge continuum testbed, with open-source artifacts released for reproducibility.