SYSTEM Cited by 1 source

Airbnb Knowledge Graph Infrastructure

Airbnb's internally managed, multi-tenant knowledge graph platform built on JanusGraph + DynamoDB + OpenSearch. Provides a paved-path graph data platform where each tenant operates in an isolated namespace with schema enforcement, index management, and schematized Thrift APIs.

Architecture

• Graph engine: Internal fork of JanusGraph with custom optimizations (conditional-write transactions, parallel multi-slice fetching, distributed tracing).
• Storage: DynamoDB (leverages AWS's scalability and reliability without reinventing distributed storage operations).
• Indexing: OpenSearch for property-based lookups.
• Management service: Schema enforcement, index lifecycle, Thrift API generation per tenant.
• Tenant isolation: Namespace-based; each use case (identity graph, inventory knowledge graph, data lineage, fraud detection) is a separate tenant.

Tenants

• Identity graph — 7B nodes, 11B edges, ~5M edges/day; Trust & Safety use cases (fraud detection, linked-account identification); 4–8 hop queries.
• Inventory knowledge graph — mentioned but not detailed.
• Data lineage — mentioned but not detailed.

Seen in

• sources/2026-05-19-airbnb-scaling-identity-graph-unified-knowledge-graph-infrastructure — canonical post describing the platform architecture, the identity graph migration from third-party vendor, and the optimizations that delivered 10× write QPS, significant P99 latency reduction, and elimination of periodic manual reboots.

Related

• systems/janusgraph — the graph database engine underneath
• systems/dynamodb — the storage backend
• concepts/multi-tenant-graph-platform — the architectural pattern
• concepts/identity-graph — the first and largest tenant
• concepts/pluggable-storage-backend — enabling DynamoDB as persistence