Deep Neural Network Based Log Analysis

Today, the logs of the services running on virtual and real servers provide us with data from which statistical inferences can be made about the service itself and the ecosystem it is in, if it is analyzed. The usage purposes of the tools used in DevOps, the content of the logs they produce, the way they are stored and the frequency of reproduction may vary. Creating metrics by analyzing logs and taking necessary actions on metrics can potentially increase productivity when the hurdles such as fragmentation of unstructured data, processing of data and processing of data are overcome. In this study, it is shown that the logs can be modeled as a natural language sequence in order that the necessary actions be taken, and it is possible to classify the process logs of Jenkins, which is one of the services used in DevOps, by using the Bidirectional Long Short-Term Memory method, with high accuracy and Cohen's Kappa score.