Rand Jared ; Miranskyy Andriy

Abstract

Software log analysis helps to maintain the health of software solutions andensure compliance and security. Existing software systems consist ofheterogeneous components emitting logs in various formats. A typical solutionis to unify the logs using manually built parsers, which is laborious. Instead, we explore the possibility of automating the parsing task byemploying machine translation (MT). We create a tool that generates syntheticApache log records which we used to train recurrent-neural-network-based MTmodels. Models' evaluation on real-world logs shows that the models can learnApache log format and parse individual log records. The median relative editdistance between an actual real-world log record and the MT prediction is lessthan or equal to 28%. Thus, we show that log parsing using an MT approach ispromising.