Log Parsing Evaluation in the Era of Modern Software Systems

• DOI: 10.48550/arxiv.2308.09003
• Publication Date: 2023-01-01
• Authors: Stefan Petrescu; Floris den Hengst; Alexandru Uta; Jan S. Rellermeyer
• Keywords: Robustness

Due to the complexity and size of modern software systems, amount logs generated is tremendous. Hence, it infeasible manually investigate these data in a reasonable time, thereby requiring automating log analysis derive insights about functioning systems. Motivated by an industry use-case, we zoom-in on one integral part automated analysis, parsing, which prerequisite deriving any from logs. Our investigation reveals problematic aspects within parsing field, particularly its inefficiency handling heterogeneous real-world We show this assessing 14 most-recognized approaches literature using (i) nine publicly available datasets, (ii) dataset comprised combined data, (iii) infrastructure large bank. Subsequently, toward improving robustness production scenarios, propose tool, Logchimera, that enables estimating performance contexts through generating synthetic resemble contributions serve as foundation consolidate past research efforts, facilitate future advancements, establish strong link between parsing.