An Approach to Identifying Error Patterns for Infrastructure as Code

Abstract

Infrastructure as Code (IaC), which specifies system configurations in an imperative or declarative way, automates environment set up, system deployment and configuration. Despite wide adoption, developing and maintaining high-quality IaC artifacts is still challenging. This paper proposes an approach to handling the fine-grained and frequently occurring IaC code errors. The approach extracts code changes from historical commits and clusters them into groups, by constructing a feature model of code changes and employing an unsupervised machine learning algorithm. It identifies error patterns from the clusters and proposes a set of inspection rules to check the potential IaC code errors. In practice, we take Puppet code artifacts as subject objects and perform a comprehensive study on 14 popular Puppet artifacts. In our experiment, we get 41 cross-artifact error patterns, covering 42% crawled code changes. Based on these patterns, 30 rules are proposed, covering 60% identified error patterns, to proactively check IaC artifacts. The approach would be helpful in improving code quality of IaC artifacts.