HyperDBDiag: Multi-Root-Cause Identification and Operator-Level Localization for Database Performance Anomalies
Hu Wang, Zhengjin Yan, Hong Zhu, Qiong Zuo
Abstract
Database performance anomalies in modern production environments are not always triggered by a single cause. Instead, they often arise from the concurrent interaction of multiple causes within the same time window, posing substantial challenges to accurate operator-level root-cause analysis. When multiple root causes interact through shared resources, metric anomalies may be amplified, attenuated, or masked. Consequently, salient local anomalies may no longer correspond directly to the actual root causes, making conventional single-path diagnosis and shallow correlation-based methods prone to misdiagnosis. To address this challenge, we propose HyperDBDiag, a method for multi-root-cause identification and operator-level localization of database performance anomalies. HyperDBDiag connects root causes, performance metrics, SQL templates, and execution operators in a knowledge graph, captures joint anomaly patterns across metrics with a metric hypergraph, and uses a constrained LLM to determine the final root-cause set and provide remediation suggestions. When finer-grained localization is possible, HyperDBDiag also identifies the associated anomalous SQL statements and execution operators. We evaluate HyperDBDiag on a database performance anomaly dataset constructed using DB-MAGS. Experimental results show that HyperDBDiag effectively reduces graph complexity while improving root-cause localization accuracy.
Create a lesson
Related papers
Bridging the Discovery Gap in Solid Dataspaces with a Semantic Data Catalog
Florian Hölken, Alexander Paulus, Tobias Meisen et al.
GenSPARQL: A Generative Extension of SPARQL for Incomplete Knowledge Graphs
Jingcheng Wu, Yi Wang, Hongkuan Zhou et al.
Standardized Tolerance for Adaptive Reuse and Repair: A Method for HNSW Migration Under Embedding Model Change Scenarios
Hengzhao Ma, Chenyang Zhu, Bin Wang et al.
The American Medical Directory Panel: Comparing AI-Assisted and LLM Transcription with Manual Data Entry at Scale
Christian M. Dahl, Sophie Li, Rasmus Emil Nyholm et al.
CRAFT: Constrained Route planning with Adaptive Framework and Template-based code generation
Shahd Elmahallawy, Ahmed Eldawy
ProvCards: Artifact-Oriented Provenance for Heterogeneous Scientific Workflows
Tanja Auge, Marinus Holitschke, Meike Klettke