Towards a lightweight distributed telemetry for microservices
Manuel Otero, José María García, Pablo Fernández · 2024
Microservice architectures have become the standard framework for developing scalable distributed systems, offering significant advantages in managing the integration and evolution of complex applications. Despite their benefits, these architectures face challenges, particularly in effectively diagnosing and resolving performance and reliability issues. Traditional centralized telemetry models, such as those implemented by Prometheus, ELK, and various cloud-based platforms like Datadog or NewRelic, often require complex and costly configurations and are not inherently tailored to the unique requirements of RESTful microservices. While the OpenAPI Specification (OAS) has established itself as a key standard for describing microservice APIs, current centralized tools do not leverage this standard to enhance service analysis effectively. This paper introduces a novel, lightweight, and distributed approach to telemetry that capitalizes on the API information provided by the OAS. Our proposed model aims to simplify the diagnostic process by offering an automated, configuration-free system that provides developers and operations teams access to root cause analysis tools. Additionally, it allows for effective management of telemetry without impacting service performance, thanks to the ability to toggle telemetry features on or off as needed; in doing so, our approach is specially suitable to the dynamic nature of Distributed Computing Continuum (DCC) systems where resources and services may fluctuate and scale. As validation, we provide a first proof of concept consisting of a ready-touse package for the NodeJS ecosystem that has been tested to demonstrate that it can operate with minimal impact on system performance and resource usage.