Learning Collaborative and Robust Scheduling Policies for Microservice Backends under Uncertainty

Abstract

With the continuous development of cloud computing, containerized deployment, and service-oriented architecture, microservice server backends face higher demands in handling high-concurrency requests, coordinating complex service dependencies, and managing dynamic resources. Traditional scheduling methods relying on static rules or local optimization are no longer suitable for operating environments with multiple coupled services, rapid state changes, and abnormal disturbances. To address this issue, this paper proposes a reinforcement learning-based collaborative scheduling and robust optimization method for microservice server backends. Starting from the overall system operation process, it unifies information such as queue state, resource consumption, dependency strength, and latency fluctuations into a scheduling state representation. Based on this, a joint decision-making mechanism is constructed, incorporating request allocation, instance scaling, and resource reallocation into a unified action space to improve overall coordination capabilities under complex service chain conditions. Furthermore, to address uncertainties in the backend environment, such as traffic surges, node degradation, and resource jitter, this paper introduces robust optimization concepts into the policy learning process. Through disturbance modeling and constraint control, it enhances the stability and adaptability of the scheduling strategy in complex operating scenarios. To ensure the engineering usability of the method, this paper simultaneously considers service quality constraints, resource capacity limitations, and system operating costs, enabling the proposed model to balance deployment rationality and control feasibility while pursuing improved throughput and response efficiency. Related research results show that the proposed method can achieve good overall scheduling performance in microservice server backend scenarios, demonstrating strong advantages in service processing capacity, resource utilization, and operational stability. This research provides a modeling approach for intelligent scheduling in microservice environments that considers global collaboration, dynamic decision-making, and robust control, and also offers methodological references for research on cloud platform backend resource management and service governance.