A lingualization strategy for knowledge sharing in large-scale DevOps

Abstract

DevOps has become a generally accepted practice for software projects in the last decade and approaches certain shortcomings of the agile software development and the steadily gaining popularity of cloud infrastructure. While it shifts more and more responsibilities towards software engineering teams, the prevailing opinion is to keep DevOps teams small to reduce the complexity of inter-team communication. In circumstances where products outgrow the performance capability of a single team, microservice architecture enables multiple DevOps teams to contribute to the same application and meet the increased requirements. Since DevOps teams operate typically self-sufficiently and more or less independently inside an organization, such large-scale DevOps environments are prone to knowledge-sharing barriers.

Textual Domain-Specific Languages (DSLs) are one of the cornerstones of DevOps and enable key features like automation and infrastructure provisioning. Nonetheless, most commonly accepted DSLs in the context of DevOps are cumbersome and have a steep learning curve. Thus, they fall short of their potential to truly enable cross-functional collaboration and knowledge sharing, not only between development and operation, but to the whole organization. DevOps teams require tools and DSLs, that treat knowledge sharing and reuse as a first-class citizen, in order to operate sufficiently on a large scale. However, developing DSLs is still presumed as an expensive task which can easily offset the resulting benefits.

This dissertation presents a lingualization strategy for addressing the challenge of knowledge sharing in large-scale DevOps. The basic idea is to provide custom-tailored Domain-Specific Modeling Languages (DSMLs) that target single phases of the DevOps lifecycle and ease the DevOps adoption for newly formed teams. The paradigm of Language-Driven Engineering (LDE) bridges the semantic gap between stakeholders by custom-tailored DSMLs and thus is a natural fit for knowledge sharing.

Key to a successful practice of LDE is as a new class of stakeholders. In the context of large-scale DevOps, language development can be realized by so-called Meta DevOps teams. Those teams, which themselves practice DevOps internally, manage a centralized repository of small DSMLs and offer them as a service. DevOps teams act as the customers of the Meta DevOps teams and can request new features or complete new DSMLs and provide feedback to already existing DSMLs. The presented Rig modeling environment serves as an exemplary DSML that targets the purpose of Continuous Integration and Deployment (CI/CD), one of the most important building blocks of DevOps. Rig comes with an associated code generator to fully-generate CI/CD workflows from graphical models. Those graphical models provide an executable documentation and assist knowledge-sharing between stakeholders.

The fundamental modeling concepts of the lingualization strategy are evaluated against previously published requirements by Bordeleau et al. on a DevOps modeling framework in an industrial context. In addition to that, Rig is evaluated based on results of a workshop during the 6th International School on Tool-Based Rigorous Engineering of Software Systems. Both evaluations yield encouraging results and demonstrate the potential of the lingualization strategy to break down knowledge-sharing barriers in large-scale DevOps environments.