Nebula held its second Beijing meetup in Jingdong Beichen. CEO Xiaomeng Ye led Geobase, Ant Financial’s graph database. Two technical directors—Heng Chen and Fenglin Hou (dutor)—drive storage (similar to TiKV) and the query engine (similar to TiDB), respectively.
The architecture separates storage and compute. The compute layer is SQL-like, with Go syntax for edge hops and pipelines to simplify nested subqueries. Storage is key-value with static hash partitioning by key (no dynamic partitioning yet). The hash is computed from the vertex, so all properties and in/out edges of one vertex live in one partition—friendly for edge traversal. Queries like “top 10 weighted edges from A to B” are harder and need separate indexes.
Each partition is a three-replica Raft group with RocksDB. A node may host many partitions; each partition needs a thread to push Raft logs. They use a shared thread pool to cap thread count. Groups can include learner replicas for async replication to other systems (offline scenarios, etc.). Meta servers mirror data servers except meta servers handle metadata KV and data servers handle graph-operator KV.
The product is alpha; a production-ready release is planned for the second half of the year. Financial anti-fraud and anti-money-laundering seem like strong graph-database demand: Ant Financial pioneered graph DB in China; the founding team partnered first with JD Digits where demand and Nebula’s experience complemented each other. I expect banks, finance, insurance, and securities to be core customers. Second- and third-tier internet companies beyond BAT may join the community to develop, test, fix bugs, and raise mindshare—if the query language becomes a de facto standard, switching costs rise.
Transactions are not supported yet; they plan a practical Percolator-style protocol—timestamp service limits scale, but it is simple and maps well to RocksDB row transactions and decentralized lock conflict detection.
Open source, a hot market, a strong team, and VC backing—promising outlook.
Personal note: infrastructure is both engineering and product. For users, product often matters more than pure engineering. I hope they polish a refined infra product.