Tracing and Metrics Design Patterns for Monitoring Cloud-native Applications

TL;DR

The paper extends a catalog of observability design patterns for cloud-native applications by detailing three core patterns: Distributed Tracing, Application Metrics, and Infrastructure Metrics. It argues that end-to-end visibility, structured application metrics, and infrastructure awareness are essential for diagnosing faults and optimizing performance in distributed systems. Each pattern is described with context, problem, forces, solution, consequences, examples, known uses, and related patterns, drawing on industry practice and prior work. The work aims to provide practitioners with actionable design patterns and trade-offs for scalable, low-friction observability in dynamic cloud environments, while noting the need for empirical validation and ongoing refinement.

Abstract

Observability helps ensure the reliability and maintainability of cloud-native applications. As software architectures become increasingly distributed and subject to change, it becomes a greater challenge to diagnose system issues effectively, often having to deal with fragmented observability and more difficult root cause analysis. This paper builds upon our previous work and introduces three design patterns that address key challenges in monitoring cloud-native applications. Distributed Tracing improves visibility into request flows across services, aiding in latency analysis and root cause detection, Application Metrics provides a structured approach to instrumenting applications with meaningful performance indicators, enabling real-time monitoring and anomaly detection, and Infrastructure Metrics focuses on monitoring the environment in which the system is operated, helping teams assess resource utilization, scalability, and operational health. These patterns are derived from industry practices and observability frameworks and aim to offer guidance for software practitioners.

Figures (5)

• Figure 1: Overview of the monitoring design pattern candidates proposed by the authors. The patterns highlighted with a dashed line are the ones explored in this paper.
• Figure 2: Overview of the structure for the Distributed Tracing pattern. The request is assigned a unique ID as it enters the system and the generated spans carry that ID.
• Figure 3: Overview of the structure for the Application Metrics pattern. The solutions usually only follows the pull or the push strategy. We represented both for completeness.
• Figure 4: Real-world example of the Application Metrics pattern Campuzano-2019.
• Figure 5: Overview of the structure for the Infrastructure Metrics pattern. The solutions usually only follow the pull or the push strategy. We represented both for completeness.