A Data-Driven Approach for Adaptive Real-Time Log Parsing in Cloud Environments

In the era of rapidly expanding cloud computing centers and large-scale services, analyzing system logs has become crucial for monitoring the quality of service. With systems generating vast amounts of logs, manual analysis is no longer feasible, necessitating automatic and precise log analysis techniques. The process begins with log parsing, a critical first step towards automating the analysis by transforming unstructured logs into structured records. However, current log parsing techniques lack adaptability. First, they struggle with software or firmware updates, as previously learned templates fail to recognize new log types. Second, they perform poorly across different services, unable to accurately parse logs from newly introduced services, further hindering effective log parsing. To address these challenges, we propose an adaptive log parsing method for largescale cloud environments called AdapLog. AdapLog leverages an online data-driven approach that efficiently processes grouped log messages without manual parameter tuning. Evaluation results indicate that our log parsing method outperforms state-of-the-art techniques across most experiments with respect to parsing accuracy (up to 4.2 x higher) and time (up to 13.4 x less per log).