Troubleshooting Configuration Errors via Information Retrieval and Configuration Testing

Software systems introduce an increasing number of configuration options to provide flexibility and customizability for users. As configurations in mature systems become increasingly complex, configuration errors have become one of major causes of software failure. Due to the complexity of the correlation between configuration options and operating environment, misconfiguration diagnosing faces many challenges, such as low efficiency of program analysis and incomplete statistical data. In order to improve the efficiency and accuracy of diagnosing configuration errors, we propose a method to construct a database of historical configuration errors, and use information retrieval to troubleshoot configuration errors. We design and implement ConfErrShooter, which first extracts log features for known configuration errors and recommends potential misconfigurations by detecting the similarity between the new failure logs with known log features. We use manual and real-world error cases from MySQL, PostgreSQL and Httpd to evaluate the effectiveness. The result shows the accuracy reaches 86% for the manual error cases, and 73% for real-world cases.