TREAT: Automated Construction and Maintenance of Probabilistic Knowledge Bases from Logs (Extended Abstract)

Zhu, Ruiqi; Li, Xue; Wang, Fangrong; Bundy, Alan; Pan, Jeff Z.; Nuamah, Kwabena; Mauceri, Stefano; Xu, Lei

doi:10.1007/978-3-031-25599-1_24

Cited by 1 publication

(1 citation statement)

References 11 publications

(13 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, we will discuss how ABC finds and repairs the root cause. The comparison with other RCA systems that do not addressing MI is a future work, which will be given as a part of the evaluation of the TREAT project [22].…”

Section: Discussionmentioning

confidence: 99%

ABC in Root Cause Analysis: Discovering Missing Information and Repairing System Failures

Li¹,

Bundy²,

Zhu³

et al. 2023

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Root-cause analysis (RCA) is a crucial task in software system maintenance, where system logs play an essential role in capturing system behaviours and describing failures. Automatic RCA approaches are desired, which face the challenge that the knowledge model (KM) extracted from system logs can be faulty when logs are not correctly representing some information. When unrepresented information is required for successful RCA, it is called missing information (MI). Although much work has focused on automatically finding root causes of system failures based on the given logs, automated RCA with MI remains under-explored. This paper proposes using the Abduction, Belief Revision and Conceptual Change (ABC) system to automate RCA after repairing the system's KM to contain MI. First, we show how ABC can be used to discover MI and repair the KM. Then we demonstrate how ABC automatically finds and repairs root causes. Based on automated reasoning, ABC considers the effect of changing a cause when repairing a system failure: the root cause is the one whose change leaves the fewest failures. Although ABC outputs multiple possible solutions for experts to choose from, it hugely reduces manual work in discovering MI and analysing root causes, especially in large-scale system management, where any reduction in manual work is very beneficial. This is the first application of an automatic theory repair system to RCA tasks: KM is not only used, it will be improved because our approach can guide engineers to produce KM/higher-quality logs that contain the spotted MI, thus improving the maintenance of complex software systems.

show abstract

Section: Discussionmentioning

confidence: 99%