Work

Cloud Infrastructure Reliability & Incident Management: A Dependency-Aware Approach to Detection and Resolution

Year: 2026
Type: article
Abstract: Aim: This study examines monitoring-related failures in cloud infrastructure and proposes a dependency-aware incident analysis framework to support systematic diagnosis, root-cause identification, remediation, and incident prevention. Methods: The study adopts a quantitative secondary-data analysis approach based on a production-incident investigation involving 306 Microsoft services in 2022. The source study identified 973 monitoring-related repair items, from which a validated analytical dataset of 579 items was constructed through manual annotation, pseudo-labelling, validation, and exclusion of items with insufficient information. Results: The analysis revealed that missing monitors or alerts accounted for 40.41% of repair items, while missing or improper signals represented 18.13%. Improper monitoring logic accounted for 12.78%, and inadequate monitor coverage accounted for 10.02%. Furthermore, 27.25% of missed-detection incidents were associated with outages. Customer-reported incidents took approximately 10.7 times longer to detect and 3.75 times longer to mitigate than monitor-detected incidents. Based on these findings, the study proposes a nine-stage dependency-aware incident analysis framework comprising symptom identification, telemetry interpretation, dependency mapping, failure-domain isolation, hypothesis formulation, root-cause confirmation, remediation, recovery verification, and post-incident review with preventive action. Conclusion: The findings demonstrate the importance of effective monitoring and systematic dependency analysis in diagnosing cloud infrastructure failures. The proposed nine-stage framework provides a structured approach to identifying failure sources, confirming root causes, implementing corrective measures, and preventing incident recurrence. Recommendation: Further studies should evaluate the proposed framework through controlled cloud failure-injection experiments to establish its diagnostic accuracy and operational effectiveness. Such evaluations should assess mean time to detect (MTTD), mean time to recover (MTTR), and incident recurrence.
Language: English
Open Access status: diamond
OpenAlex ID: W7237345669
FWCI:
Cites: 21
Cited by:
Related to: