How a Chinese FinTech Built a Scalable MySQL Cloud Platform for Billions of Transactions

Background

In 2015 a leading Chinese internet‑finance company, serving over 100 million registered users and processing more than 1 billion SQL statements per day, required a highly available, horizontally scalable MySQL‑based platform. All database services were built on open‑source MySQL.

Stage 1 – Private‑Cloud DBaaS Deployment

The initial design used a public‑cloud partner’s infrastructure with KVM virtualization and an IP‑based distributed storage pool across multiple x86 servers. The storage layer introduced excessive latency, causing MySQL nodes to reach full load and preventing horizontal scaling.

A Share‑Nothing multi‑machine cluster DBaaS was introduced to replace virtualized storage. Key technical features:

Multi‑path data replication and redundant nodes eliminate the need for shared storage.

Read/write separation, in‑memory caching, and load‑balancing modules enable dynamic scaling.

30 physical servers handled >1 billion daily SQL statements; user count grew from millions to tens of millions.

Stage 2 – Low‑Bandwidth Disaster Recovery

Traditional disaster‑recovery (DR) relied on high‑end storage appliances that asynchronously replicated data, consuming large bandwidth and incurring high cost. A MySQL‑native DR tool, Qreplicator , was developed with the following characteristics:

Remote binlog replication uses roughly 1/15 of the bandwidth required by disk‑array replication and about 50 % of the source MySQL binlog traffic.

The standby database remains in an OPEN state, allowing read queries, reporting, and load‑sharing during failover, thus avoiding an unstartable standby.

Stage 3 – Distributed MySQL Cluster for High‑Volume Red‑Packet Campaign

To support a promotional event requiring 1 million transactions per second, the existing read/write‑separated cluster was insufficient. A distributed MySQL cluster was built with the following properties:

Node‑level horizontal scaling without service interruption; new nodes can be added seamlessly.

Data is sharded and replicated across hundreds of MySQL servers, guaranteeing no data loss during scaling.

During the campaign the cluster sustained the peak load, and overall platform users reached 100 million.

Current operation processes several hundred million SQL statements daily across the cluster.

Stage 4 – Enterprise‑Grade PaaS Extension

Building on the DBaaS and DR foundations, the platform was expanded into a B2D (business‑to‑developer) PaaS offering that bundles ready‑to‑use components:

EDAS – Enterprise Distributed Application Service for micro‑service deployment.

RTMQ – Reliable Transmission Message Queue for asynchronous messaging.

Eagle EYE – Full‑link monitoring system for end‑to‑end performance visibility.

Log Analytics – Centralized log collection and analysis.

These services provide BAT‑class distributed processing capabilities to traditional enterprises, enabling rapid modernization of their IT stacks.

Outcome

The combined technical solutions transformed the fintech client from a private‑cloud experiment into a production‑grade, high‑availability MySQL cloud platform. The architecture now supports billions of daily transactions, offers low‑bandwidth disaster recovery, seamless horizontal scaling, and a suite of enterprise PaaS services.