Automated diagnosis of software misconfigurations based on static analysis

摘要

Software configuration settings are an effective way to customize applications. However, inconsistencies or mistakes in option values can result in a system crash and need huge time and effort to diagnose. We present a technique to identify the root causes of configuration errors. It uses static program analysis to link the misconfiguration of an application to a specific configuration option. Our technique has two prominent characteristics compared to existing approaches: it relies only on static analysis, and it does not need profiles of the application with correct configuration. Based on the proposed techniques, we developed a tool called ConfDebugger. We evaluated its effectiveness on 8 configuration errors in the Java program JChord. ConfDebugger successfully diagnosed 7 out of 8 errors. For 5 of them, root cause was exactly pinpointed without a false positive, and in total, the average number of false positives was only 0.5. This is better than two state-of-the-art methods, with average numbers of false positives of 1.7 and 5.7, respectively.