Software Misconfiguration Troubleshooting Based on State Analysis

Misconfiguration has become one of the dominant causes of service failure and software abnormality. It will not only affect the user experience and cause economic losses, but also require a lot of manpower for troubleshooting. Therefore, an automated tool to quickly diagnose misconfiguration is essential. This paper introduces a tool called ConfDetect, which treats system as data to derive the troubleshooting clues using state analysis. ConfDetect ranks logs based on heuristics to give problem-related messages, uses NLP for Part-Of-Speech tagging to filter out suspicious words which are compared with configuration key-value pairs and environment information, or searches the knowledge base to find misconfiguration items. The knowledge base realizes self-learning by analyzing the configuration changes of the sick system after it has been healed, and predicts the configuration item resulting in the error log. Misconfiguration tests on MySQL, Ngnix and an autonomous vehicle simulation program on ROS show that ConfDetect is able to deliver error message with 91% accuracy and find specific misconfiguration item with 74% accuracy in general. Besides, the time cost of ConfDetect is in proportion to the volume of logs and it takes less than a few seconds to process thousands of lines log. Experiment results prove that ConfDetect is effective in diagnosing misconfiguration and runs fast.