Skip to content

SYSTEM Cited by 1 source

JanusGraph

JanusGraph is a distributed, open-source graph database built on the Apache TinkerPop framework. It uses the labeled property graph model and supports the Gremlin traversal query language. Its distinguishing architectural feature is pluggable storage backends — the graph logic layer is decoupled from persistence, allowing operators to choose DynamoDB, Cassandra, HBase, or other stores underneath.

Key properties

  • Labeled property graph model — nodes and edges carry labels and arbitrary key-value properties; strong schema support.
  • Gremlin query language — expressive traversal language from Apache TinkerPop; supports multi-hop path queries, filtering, aggregation.
  • Pluggable storage — graph logic runs independently of the storage engine; backends include Cassandra, HBase, DynamoDB, and others.
  • Pluggable indexing — supports Elasticsearch, OpenSearch, Solr for full-text / geo / range indexing of properties.
  • Open-source codebase — allows deep customization (Airbnb forked it to add custom transaction strategies, parallel fetching, and distributed tracing).

Seen in

  • sources/2026-05-19-airbnb-scaling-identity-graph-unified-knowledge-graph-infrastructure — Airbnb chose JanusGraph as the core of their internal knowledge graph infrastructure, running it with DynamoDB as the storage backend and OpenSearch for indexing. Key customizations to the internal fork: (1) custom transaction strategy using DynamoDB conditional writes (replacing JanusGraph's default heavy locking); (2) parallel getMultiSlices interface for high-fanout queries; (3) distributed tracing integration. The platform serves a 7B-node / 11B-edge identity graph at 4–8 hop query depths for Trust & Safety use cases.
Last updated · 542 distilled / 1,571 read