When Failure is (Not) an Option: Reliability Models for Microservices Architectures

Modern application development and deployment is rapidly evolving to microservices based architectures, in which thousands of microservices communicate with one another and can be independently scaled and updated. While these architectures enable flexibility of deployment and frequency of upgrades, the naive use of thousands of communicating and frequently updated microservices can significantly impact the reliability of applications. To address these challenges, service meshes are used to rapidly detect and respond to microservices failures without necessitating changes to the microservices themselves. However, there are inherent tradeoffs that service meshes must make with regards to how quickly they assume a microservice has failed and the subsequent impact on overall application reliability. We present in this paper a modeling framework for microservices and service mesh reliability that takes these tradeoffs into account.