WERCA: Workflow-guided and Evidence-constrained Root Cause Analysis for DevOps Systems
Root cause analysis (RCA) in DevOps systems is difficult because user-visible failures are often downstream symptoms of longer workflows. The actual fault may occur during event generation, forwarding, permission checking, scheduling, artifact synchronization, or event consumption. Methods based on global log analysis, anomaly detection, or unconstrained semantic reasoning therefore miss workflow-stage constraints and explicit evidence checks. This paper presents WERCA, a DevOps RCA framework guided by an expert-maintained Workflow/DAG repository. Given a failure ticket, WERCA selects a likely execution path, retrieves logs only from nodes on that path, and checks each stage against its responsibility. A candidate root cause is retained only when supported by log evidence, workflow position, and stage semantics. When later evidence contradicts the current path, WERCA rolls back and revises the diagnosis. Historical incidents provide retrieval-based priors for ranking, confidence estimation, and remediation generation. Our experimental results on real cases from DevOps systems show that WERCA can improve the ranking results compared with the baseline using only logs, despite a slight increase in time overhead. The results also indicate that Workflow/DAG can guide the RCA process of DevOps systems by tracing concrete execution paths for failure as evidence reasoning.
Sun 19 JulDisplayed time zone: Brisbane change
11:40 - 12:40 | Session 8: Software Maintenance, Repair, and DevOpsResearch Track / New Idea at Promenade Chair(s): Kaifeng Huang Tongji University | ||
11:40 15mTalk | Fixing Atomicity Violations via Task-driven Dynamic Code Abstraction and Repair Research Track | ||
11:55 15mTalk | LLMs as Continuous Learners: Improving Software Issue Reproduction by Distilling Experiences Research Track Yalan Lin Shanghai Jiao Tong University, Silin Chen Shanghai Jiao Tong University, Yingwei Ma Tongyi Lab, Alibaba, Rongyu Cao Tongyi Lab, Alibaba, China, Binhua Li Tongyi Lab, Alibaba, China, Fei Huang Tongyi Lab, Alibaba, China, Xiaodong Gu Shanghai Jiao Tong University, Yongbin Li Tongyi Lab, Alibaba, China | ||
12:10 10mTalk | When Should Dependency Updates Invoke Repair Agents? A Lightweight Routing Study New Idea Liheng Fan China Academy of Telecommunication Technology, Jialun Yin School of Vehicle and Mobility, Tsinghua University, Yuzhi Chen Intelligent Transportation System Research Center, Southeast University | ||
12:20 10mTalk | WERCA: Workflow-guided and Evidence-constrained Root Cause Analysis for DevOps Systems New Idea yi zhang , Yonghang Wu Beijing XingYun Digital Technology Co., Ltd, Xin Ma Beijing XingYun Digital Technology Co., Ltd, Xinhang Yin Beijing XingYun Digital Technology Co., Ltd, Bing Han Beijing XingYun Digital Technology Co., Ltd, Yijian Zheng Xidian University, Chenxi Zhang Xidian University, Di Cui Xidian University | ||
12:30 10mTalk | Rethinking Agent Memory for Routine Code Fixes: MarkovFix Beats Frontier Agents on Deployable LLM New Idea Zhiqing Zhong The Chinese University of Hong Kong, Shenzhen (CUHK-Shenzhen), Jonathan Kristian Tjandra The Chinese University of Hong Kong, Shenzhen, Feivel Ehren Hermanto The Chinese University of Hong Kong, Shenzhen, Ruizhe Ye Huawei Cloud Computing Technologies Co., Ltd., Jingyuan Tan Huawei Cloud Computing Technologies Co., Ltd., Yuechan Hao Huawei Cloud Computing Technologies Co., Ltd., Yuchi Ma Huawei Cloud Computing Technologies, Pinjia He Chinese University of Hong Kong, Shenzhen | ||