ADAPTIVE TRACING METHODS FOR DISTRIBUTED APPLICATIONS BASED ON SELF-LEARNING SAMPLING

Author(s): Zubkov M.V., Makeev P.S.

Rubric: Information technology

DOI: 10.21777/2500-2112-2026-1-113-121

Release: 2026-1 (54)

Pages: 113-121

Keywords: microservices, observability, distributed tracing, sampling, telemetry, self-learning algorithm, adaptive monitoring, machine learning

Annotation: Modern microservice applications generate a significant amount of telemetry data necessary to ensure the observability and rapid diagnosis of incidents. However, full tracing of all requests becomes economically impractical due to the overhead of collecting, storing, and processing information. The article discusses methods of adaptive sampling of distributed tracing, which allow dynamically adjusting the amount of data collected depending on the current state of the system and query characteristics. The architecture of the tracing system is proposed, which includes a compo- nent of self-learning selection of traceable queries based on the analysis of metrics and query features. An algorithm is presented that combines system and behavioral criteria for evaluating the significance of queries. Experimental modeling has shown that the proposed approach can cover up to 90 % of abnormal cases while reducing the volume of telemetry by more than six times compared to the full collection. The results demonstrate the promise of adaptive sampling for increasing the efficiency of observability in microservice systems without increasing infrastructure load.

Bibliography: Zubkov M.V., Makeev P.S. ADAPTIVE TRACING METHODS FOR DISTRIBUTED APPLICATIONS BASED ON SELF-LEARNING SAMPLING // Education Resources and Technologies. – 2026. – № 1 (54). – С. 113-121. doi: 10.21777/2500-2112-2026-1-113-121