Leveraging Ops Data: Knowledge Graphs, Auto‑Fault Assessment & Unattended Changes

Scope and Challenges of Ops Data

From a narrow perspective, operational data mainly covers system stability and resource management, which can be divided into three categories: resource metadata, system state data, and event data. From a broader perspective, the coverage expands to include business data, operational data, engineering efficiency data, and user experience data, introducing more complex usage scenarios, processing rules, and association relationships.

Advanced Ops Data Scenarios

Ops data output follows timeliness rules (offline, near‑line, real‑time) and acquisition methods (periodic pull, periodic push, real‑time fetch). Offline data suits metric statistics, near‑line supports intelligent monitoring and fault prediction, while real‑time enables self‑healing, unattended, and auto‑scheduling scenarios.

Knowledge Graphs

Knowledge graphs originated from search engines and have become essential in ops for aggregating massive data, establishing granular relationships, and supporting business continuity. They extend CMDB capabilities by linking infrastructure, system, and business layers, allowing impact analysis such as identifying which services are affected when a host restarts.

Automated Fault Assessment

Automated fault assessment helps quickly evaluate business impact during emergencies or drills. By enumerating host failures and analyzing dependency graphs, it identifies critical nodes whose outage would disrupt service chains. It also emphasizes breaking data silos across infrastructure, application, and business layers.

Unattended Change Management

Unattended changes aim to reduce manual intervention while ensuring safe rollouts. Required conditions include ordered release sequences, automated switch‑off policies, comprehensive validation (package, log, metric checks), automatic rollback on failures, and post‑change monitoring of key business indicators.

Dynamic Thresholds

Dynamic thresholds address alert storms in high‑traffic e‑commerce scenarios. They adapt to three data deviation patterns: periodic fluctuations, sudden spikes, and noise (spikes). Machine‑learning or statistical methods adjust thresholds in real time, and dependency matching based on knowledge‑graph relationships helps suppress unnecessary alerts.

Conclusion

Operational data has evolved from basic monitoring to a strategic asset that drives automation, intelligence, and cost reduction. By monetizing data and integrating it into DevOps workflows, ops moves from backstage to a front‑line role in business success.