LogAttention – Assessing Software Release with Attention-based Log Anomaly Detection

A Software Engineering Manager (EM) has to cater to the demand for higher reliability and resilience in Production while simultaneously addressing the evolution of software architecture from monolithic applications to multi-cloud distributed microservices. Pre-release functional testing is no longer sufficient to eliminate faults as more and more issues are generated at runtime, which is challenging to diagnose due to complex inter-service dependencies and dynamic late binding of services. Bugs in Production are known to propagate across software components and become critical as they go undetected. This praxis introduces LogAttention, a methodology based on analysis of runtime logs that provides actionable insights to the EM to identify faults and preempt failure in Production. LogAttention is a Log Anomaly Detection (LAD) technique that uses Attention-based Transformer Models to identify Anomalous Log Messages. LogAttention assigns a quality score to the software release in Production and presents remarkable logs to the EM to analyze, predict, and preempt failure. This praxis presents empirical evidence showing that LogAttention outperforms existing LAD techniques to identify anomalous log messages and ensure that the detected log anomalies are reliable indicators of the health of a software release.

Author

• Munir, Sohail

Language

• en

Keyword

• Log Anomaly Detection
• Transformer Models
• Attention Model
• Software Engineering
• AIOps
• Log Attention

Date created

• 2021

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Engineering Management

Degree

• D.Engr.

Advisor

• Fossaceca, John
• Etemadi, Amir

Committee Member(s)

• Sarkani, Shahryar

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/fq977v561