Mining the Limits of Granularity for Microservice Annotations

Microservice architecture style advocates the design and coupling of highly independent services. Various granularity dimensions of the constituent services have been proposed to measure the complexity and refinement levels of the service provision. Moreover, attaching annotations to operations adds granularity to the services while adding features and facilitating the implementation of applications. Microservice applications with inadequate granularity affect the system quality of service (e.g., performance), introduce issues for management, and increase the diagnosing and debugging time of microservices to days or even weeks. In this paper, we propose a semantics-driven learning approach to mining the granularity limits of operations with their annotations according to the developer community. The learning process pursues to build a vector space for clustering similar operations with their annotations that facilitate the identification of granularity. The evaluation shows that clustering annotations by operations similarity achieves significantly high accuracy when classifying unseen operations (89%).