Lessons learned from migrating complex stateful applications onto serverless platforms

Abstract

Serverless computing is increasingly seen as a pivot cloud computing paradigm that has great potential to simplify application development while removing the burden of operational tasks from developers. Despite these advantages, the use of serverless computing has been limited to few application scenarios exhibiting stateless and parallel executions. In addition, the significant effort and cost associated with rearchitecting existing codebase limits the range of these applications and hinder efforts to enhance serverless computing platforms to better suit the needs of current applications. In this paper, we report our experience and observations from migrating four complex and stateful microservice applications (involving 8 programming languages, 5 application frameworks, and 40 application logic services) to ApacheOpenWhisk, a widely used serverless computing platform. We highlight a number of patterns and guidelines that facilitate this migration with minimal code changes and practical performance considerations, and imply a path towards further automating this process. We hope our guidelines will help increase the applicability of serverless computing and improve serverless platforms to be more application friendly.