Case Study: KE Holdings achieves millisecond graph queries for 48 billion triples with Dgraph

Powering 48 Billion Triples In Production with Dgraph at KE Holdings

KE Holdings, China’s leading integrated online and offline housing transactions platform, needed a graph database to power its rapidly growing property knowledge graph of 48 billion ordered triples and 10 billion nodes. Its teams had to support complex, millisecond-level queries across listings, clients, brokers, communities, schools, transit, and more, while avoiding the high latency and heavy join costs of traditional databases and search systems.

KE Holdings chose Dgraph as its universal graph database, using Dgraph’s distributed cluster, BadgerDB storage, and native query/indexing capabilities deployed with Docker and Kubernetes. The result was a system that could import 48 billion triples in 15 hours after optimization, handle 1,000 concurrent queries with about 50 ms response times, and sustain roughly 15,000 QPS, giving KE Holdings a fast, lower-maintenance foundation for graph search and recommendations.

KE Holdings

Pan Gao

Chief Search Architect