Netflix Scales Observability with Ontology-Driven Knowledge Graph

As Netflix expands its global reach and diversifies its content offerings, maintaining a seamless user experience across hundreds of client platforms, microservices, and infrastructure components has become increasingly complex. To address this challenge, Netflix engineers Prasanna Vijayanathan and Renzo Sanchez-Silva have spearheaded the development of an end-to-end (E2E) observability platform built around an ontology-driven knowledge graph. This approach, presented at QCon London 2026, aims to correlate user experience with system performance, enabling faster incident resolution and proactive problem prediction.

The Challenge of End-to-End Observability

Traditional observability systems often struggle with siloed data sources, disconnected alerting, and complex triage processes. These limitations can lead to lengthy incident investigations, as evidenced by a recent Netflix incident that took four hours to resolve, requiring the involvement of nine teams and over 30 engineers QCon London 2026. The core problem lies in the difficulty of connecting the dots between user-facing issues and the underlying infrastructure components responsible.

Introducing the Knowledge Graph and Ontology

To overcome these challenges, Netflix is leveraging a knowledge graph – a representation of data as interconnected entities and relationships. Central to this approach is the concept of an ontology, defined as a formal specification of types, properties, and relationships. Essentially, an ontology provides a structured way to encode knowledge about the Netflix ecosystem.

The fundamental building block of the knowledge graph is the “Triple,” consisting of (Subject | Predicate | Object). For example:

 api-gateway | rdf:type | ops:Application api-gateway | ops:ownedBy | "Team Bedrock" INC-5377 | rdf:type | ops:Incident INC-5377 | ops:affects | api-gateway 

Netflix utilizes 12 operational namespaces – including Slack, Alerts, Metrics, Logs, Incident, and Harvest – to connect all elements of its infrastructure QCon London 2026. The ontology provides order by capturing, structuring, and preserving this information in a machine-readable format, mitigating the operational chaos caused by scattered incident knowledge.

The MELT Layer and Connectedness

Vijayanathan introduced the MELT Layer (Metrics, Events, Logs, Traces) as a unified observability layer for users, devices, and services, designed to improve incident resolution times QCon London 2026. The concept of “Connectedness” is crucial, bridging gaps and breaking down silos to enrich data, minimize duplication, and facilitate more accurate diagnostics.

The Knowledge Flywheel and Automated Adaptation

The “Knowledge Flywheel” represents a continuous cycle of learning and adaptation. It operates through three states: Observer, Enrich, and Infer. This iterative process encodes knowledge with each rotation, leading to smarter subsequent iterations.

Netflix is integrating this flywheel with developer tools like Claude, utilizing it to propose pull requests (PRs) for code changes. Human review and merging complete the cycle, demonstrating a collaborative approach to automated adaptation.

Future Directions: Automation and Self-Healing

Looking ahead, Netflix plans to leverage its ontology-driven knowledge graph to automate root cause analyses, provide auto-remediation capabilities, and ultimately create a self-healing infrastructure QCon London 2026. This represents a significant step towards proactive and intelligent observability, minimizing downtime and ensuring a consistently high-quality user experience.