Build Anomaly Detection

Abstract

In today's build environment, the compilation logs are very long due to the number of modules included to bring the final compiled package. Basic sanity test, static code analysis etc. are performed along with compilation to identify the bugs and vulnerabilities. The process of reading these logs and finding deviations in them, when there is a change in the log pattern, is a time-consuming process for an engineer. Thus, a study is carried out to find out an algorithm solution that automatically identifies anomalies from log patterns in a normal execution and detect the anomalies when the log patterns deviate from the template under the normal execution. Present set of process is time-consuming, inefficient, and cost-intensive. The proposed model will help to eliminate this possibility by raising errors when a flag, a component, or a product is enabled or added out of ordinary. To achieve the above-mentioned goal the present project ‘Build Anomaly Detection’ is conceived, wherein the study includes three phases for development of model. First phase is concerned with developing the model using Python scripts. The second phase focuses on building machine learning capabilities using Python, so that the model can evolve based on the inputs. The third and the final phase is concerned with the testing of the model with various inputs in the form of build logs. The study was undertaking on Linux product and android product under the present set of conditions of setup-box using these two products under its operating system. An UI is developed which contains two comparator one is for Linux and another for Android for making the UI more user friendly respectively. In the first Comparator build logs are taken as input which will be compared by the template log i.e. developed in the first phase and the final report is generated which can be seen in the UI and the user will identify the compliances through report and intimate the changes to the team lead for taking necessary action before sending to that testing team. Thus, the whole process helps the developer to timely address the changes to the team leads and reduces the product time to market and its overall credentials of the program. Similarly, same concept is used for the second model with change of input by json file as because the number of config parameters are more in android products and analysis will be easy, but results are same.