Leveraging the Build System to Support Modern Software Release Practices

点击翻译按钮获取中文摘要

Context: Modern software is developed at a rapid pace. Indeed, software organizations like Google, LinkedIn, and Facebook release several times daily. To cope with the rapid rate at which modern software changes, software organizations dedicate personnel to the task of developing and maintaining tools and infrastructure that automate the software release process. This so-called release pipeline, which processes changes to the source code of a software system as input, is typically composed of three phases: (1) Integration: Code changes are assessed for risk, since riskier changes may introduce defects that can slow or halt product sales, impact the reputation of an organization, or even expose an organization to potential litigation. Risky code changes require additional effort from quality assurance teams before they can be confidently released to users or customers. (2) Build: Code changes are processed by the build system, i.e., the system that invokes software tools (e.g., compilers, interpreters) in order to update system deliverables (e.g., executables) and test them for regression in system behaviour. (3) Deployment: Official releases of software systems are made available for users or customers to interact with or acquire. Overarching goal and specific objectives: The rapid release cycle of modern software systems introduces new challenges for the teams that develop and maintain release pipelines within software organizations. The long term goal of this research program is to improve the efficiency and robustness of release pipelines by gleaning actionable insights from build systems. Through analysis of build system data, we aim to improve each of the three common phases of the release process. (1) Integration: More accurate assessment of the risk of code change integration through analysis of customer exposure, i.e., the proportion of the customer base that may be impacted by a given code change. (2) Build: Optimizing the use of dedicated build machines through build duration forecasts. (3) Deployment: An understanding of the evolution of the code that specifies how the deployment phase is executed, and its co-evolution with other release pipeline specifications. Outcomes and impact: Release engineers increase the market potential of software organizations. Investment in release engineering has allowed the relatively small Mozilla organization (the developers of the Firefox web browser) to compete with software giants like Microsoft, Google, and Apple. This research program will produce 10 HQP (2 PhD, 3 MEng, 5 BEng) through hands-on training in large-scale empirical analyses and release engineering. The HQP will be well-equipped for a research or private sector career in the field of release engineering, increasing Canada's innovation potential in this high impact area of software engineering.