Discovering correctness constraints for self-management of system configuration

Abstract

Managing the configuration of computer systems is a difficult task. Too easily, a computer user or administrator can make a simple mistake and misconfigure a system, causing instabilities, unexpected behavior, and general unreliability. Bugs in software that changes these configurations, such as installers, only worsen the situation. A self-managing configuration system should be continuously monitoring itself for invalid settings, preventing the bugs from harming the system. Unfortunately, while there are many constraints which can differentiate between valid and invalid settings, few are explicitly written down, much less written down in a form usable by an automatic monitor. We propose an approach to automatically infer these correctness constraints based on samples of known good configurations. In this paper we present Glean, a system for analyzing the structure of configurations and automatically inferring four types of correctness constraints on that structure.