Dynamic Analysis of Coupling in Asynchronous Flows of Microservice: An Empirical Study
Software architecture; Microservices; Modular monolith; Dynamic coupling; Distributed tracing; OpenTelemetry.
Over the last years, microservice-based architectures have become well established in industry as a response to the demands for scalability and continuous delivery. However, the physical distribution of microservices imposes an operational overhead that is not justified in every domain. Determining the point at which such a distribution ceases to be beneficial is hindered by the lack of quantitative methods that relate structural quality to operational efficiency. This difficulty is compounded in event-driven architectures, where message brokers' mediation conceals the logical dependencies between producers and consumers. This dissertation investigates an extension of the SYMBIOTE method, originally proposed for analyzing synchronous HTTP calls, to capture logical coupling in asynchronous flows. The proposed extension combines the OpenTelemetry peer. service semantic convention with the Grafana Tempo Service Graph Metrics mechanism, which enables reconstruction of direct Producer → Consumer edges while abstracting away the broker as an intermediate node, preventing it from artificially emerging as a God Service. The extended method is evaluated on Instameter, an industrial social media data collection and indexing system maintained by Buzzmonitor. The resulting structure is compared with that of a prototype version of the system, reorganized into a modular monolith. Preliminary results confirm the fidelity of the asynchronous reconstruction and characterize the ecosystem by globally low functional coupling, indicating that architectural consolidation is structurally viable.