How Ant Group Built OceanBase: Key Lessons for Self‑Developed Databases

Background

OceanBase originated in 2010 within the Taobao ecosystem and was later incorporated into Ant Group (formerly Ant Financial). The project was driven by the need to replace licensed commercial databases with a self‑controlled solution capable of handling Ant Group’s massive transaction volume.

Key Considerations for Self‑Developed Database Technology

Controlled arena: Early‑stage technology requires a protected market or internal business to obtain sufficient iteration time before facing external competition.

Core value: Differentiated value must stem from real‑world industry practice rather than simple imitation of existing commercial or open‑source products.

Product mindset: Clear problem definition, a feasible technical path, and decisive leadership are essential to turn a technology into a stable product.

OceanBase Deployment Milestones

2014: 10 % of Ant Group’s transaction traffic switched to OceanBase for the first production test.

2015: 100 % of transaction traffic and 50 % of payment traffic migrated.

2016: Full coverage of transaction, payment, accounting, and membership services.

2017: All commercial databases in Ant Group’s core business were replaced, achieving complete autonomous control.

During Double‑11 (Nov 11) peak events, transaction throughput exceeded 40 million TPS, representing a >500× increase over the original baseline.

Unit‑Based Architecture

Ant Group introduced a “unit” as a self‑contained service stack comprising gateway, application, and database components. Each unit satisfies four properties:

Self‑containment: All data and compute needed for a logical business domain reside within the unit.

Loose coupling: Units interact only via service calls; direct database access across units is prohibited.

Fault isolation: Failure of one unit does not affect others.

Built‑in disaster recovery: Units replicate data across multiple locations using OceanBase’s multi‑region consistency protocol.

This design shifts capacity planning from adding individual servers to adding whole units, simplifying scaling and limiting the blast radius of failures.

Three‑Site Five‑Center (Three‑City Five‑Replica) Architecture

The traditional “two‑site three‑center” model (two active data centers in one city plus a disaster‑recovery site) could not meet cross‑city scalability and ultra‑high‑availability requirements. Ant Group therefore deployed OceanBase clusters across three geographically separated cities, each cluster maintaining five replicas (three primary + two extra). The architecture relies on Paxos‑based majority voting:

• Majority (≥3 of 5) replicas must be alive for the cluster to remain available.
• Any single‑city outage leaves at least three healthy replicas, guaranteeing RPO = 0 and <30 s failover.
• Data consistency is preserved automatically at the database layer, invisible to upper‑level services.

By increasing the replica count from three to five, the system achieves both zero‑data‑loss disaster recovery and sub‑30‑second switchover while supporting continuous service during city‑wide failures.

Technical Achievements

Extreme scalability: Transaction peaks grew >500×, surpassing 40 million TPS during peak shopping events.

Extreme disaster‑recovery: Zero‑data‑loss failover across three cities within 30 seconds.

Extreme cost efficiency: Per‑transaction cost kept ultra‑low to support inclusive financial services at massive scale.