Making Runtime Data Useful for Incident Diagnosis: An Experience Report

摘要

Important and critical aspects of technical debt often surface at runtime only and are difficult to measure statically. This is a particular challenge for cloud applications because of their highly distributed nature. Fortunately, mature frameworks for collecting runtime data exist but need to be integrated. In this paper, we report an experience from a project that implements a cloud application within Kubernetes on Azure. To analyze the runtime data of this software system, we instrumented our services with Zipkin for distributed tracing; with Prometheus and Grafana for analyzing metrics; and with fluentd, Elasticsearch and Kibana for collecting, storing and exploring log files. However, project team members did not utilize these runtime data until we created a unified and simple access using a chat bot. We argue that even though your project collects runtime data, this is not sufficient to guarantee its usage: In order to be useful, a simple, unified access to different data sources is required that should be integrated into tools that are commonly used by team members.