Joint Scheduling Method for Resource Orchestration Optimization in Distributed Microservices via Hierarchical Reinforcement Learning

Abstract

With the continuous development of cloud-native platforms and service-oriented software architectures, distributed microservice systems are increasingly widely used in complex business scenarios. However, the increasing number of service instances, intertwined call relationships, and rapid fluctuations in resource states pose greater challenges to resource orchestration and joint scheduling. Addressing the limitations of traditional methods, such as insufficient complex dependency modeling, limited global optimization capabilities, and single-granularity decision-making, this paper proposes a joint scheduling method for distributed microservice resource orchestration optimization, organically integrating call topology-aware representation learning and hierarchical reinforcement learning into a unified framework. First, a topology-aware state representation is constructed around the dynamic call relationships between microservices, jointly encoding service node attributes, dependency relationships, and runtime resource information to enhance the model's ability to identify key call chains, local congestion areas, and structural coupling features. Second, a representation learning mechanism is used to extract service embeddings with structural semantics, enabling the scheduling process to obtain more stable and discriminative state representations in complex dependency contexts. Subsequently, a hierarchical reinforcement learning joint scheduling framework is constructed, completing global scheduling intent planning in high-level strategies and executing fine-grained resource allocation and action control in low-level strategies, thereby achieving effective connection between system-level goals and local execution behaviors. Furthermore, this paper designs a unified optimization objective around latency control, resource utilization, and quality of service constraints, enabling the joint scheduling process to balance response efficiency, resource coordination, and operational stability. Related results show that the proposed method can more effectively adapt to the scheduling requirements of complex call dependencies and dynamic resource competition in a distributed microservice environment, demonstrating strong advantages in comprehensive orchestration capabilities, service stability maintenance, and resource collaborative optimization. This method provides a new modeling approach for intelligent microservice resource management and offers a targeted solution to the joint scheduling optimization problem in complex cloud platforms.