On practitioners’ concerns when adopting service mesh frameworks

On practitioners’ concerns when adopting service mesh frameworks

연구 분야: Software Development

학회: Empirical Software Engineering

The emerging service mesh architecture tries to simplify microservices by delegating crucial tasks to dedicated infrastructure. However, service mesh introduces new notions and enables complex capabilities such as sidecar proxies that inevitably bring major adoption concerns. We investigate the adoption concerns in two dominant open-source service mesh frameworks via a mixed-methods empirical investigation of the past, current and evolution of 5,497 practitioner questions posted on generic and framework-specific question-and-answer fora. We first mine the topics of questions with the help of Dynamic Topic Modeling (DTM). We identify evolution by applying topic modelling to time periods and aggregating topics into macro-topics. We conduct a qualitative analysis to understand the three major types of questions and to generalize common fix patterns for the extracted error symptoms. We consulted a service mesh domain expert to provide feedback on our findings and discuss implications. We found that about half of the questions are error-related and mined 18 topics, covering service mesh traffic, infrastructure, security, observability and application. We discovered a drastic decline in traffic-related concerns while finding persisting infrastructure-related concerns and a rise in security and observability concerns. We identified 54 error symptoms from two popular service mesh frameworks and generalized 9 common fix patterns. We found complex symptom-to-fix relationships, yet, surprisingly, minimal configuration changes were able to fix most symptoms. Providing consistent documentation and practical automation that assists customization of service mesh deployment and functionalities is crucial in the current service mesh domain, given the diversity of discovered intentions, goals and symptoms. Furthermore, there should be more work towards better container orchestration to deploy service mesh frameworks and reliable customization of security and observability service mesh features.