Scaling a FinTech Platform to $100B Transactions with Four Overhauls

First Generation: Rapid Launch with PHP & Java

The initial platform was built to launch quickly. It used a three‑tier architecture: a mobile client layer, a PHP API layer, and Java services for scheduling and business logic. Core components included MySQL master‑slave replication, Memcached for concurrency control, ActiveMQ for asynchronous messaging, Apache as the web server, Tomcat 6 for Java, and LVS for load balancing. This generation emphasized speed over scalability and resulted in tightly coupled business logic.

Second Generation: Distributed Architecture & Big‑Data Foundations

Rapid business growth exposed technical debt, prompting a shift to a distributed system. New vertical services (contract, points, messaging, monitoring, finance, risk, sales, external integration) were added. Data pipelines replicated MySQL binlogs to MongoDB in real time using tungsten‑relicator, enabling analytics via MongoDB MapReduce. The stack added Redis for caching, Nginx for static asset delivery, a dedicated image server, and FastDFS for file storage. MySQL master‑slave separation was retained, and the architecture began to fragment account management across services.

Third Generation: SOA Governance and Service Consolidation

To reduce inter‑service coupling, the team introduced service‑oriented architecture (SOA) governance:

Service registration and discovery with Dubbo and ZooKeeper Single sign‑on via CAS and permission control with Shiro Configuration management standardized on Disconf after evaluating Qihoo360/QConf, Spring Cloud Config, Taobao Diamond, and Baidu Disconf

Common utility libraries extracted for reuse

FastDFS cluster for file storage and a unified scheduling system for all timed tasks

Redis clusters for caching and continued MySQL master‑slave separation

The front‑end remained PHP‑based, but development environments grew more complex.

Fourth Generation: Full Micro‑service Adoption with Spring Cloud

Recognizing the stagnation of Dubbo and the comprehensive capabilities of Spring Cloud, the platform migrated to a Spring Boot + Spring Cloud stack. Key motivations were:

Active development and frequent updates of Spring Cloud versus the dormant Dubbo project

Spring Cloud’s built‑in configuration, routing, and service‑discovery features covering the full micro‑service lifecycle

Non‑intrusive integration with existing Spring projects, improving developer productivity

The migration was incremental: new projects were built directly on Spring Cloud, while legacy services were routed through a gateway to ensure uninterrupted operation. The final architecture features:

Independent databases per micro‑service

Redis clusters for distributed caching

A Spring Cloud‑based service mesh handling configuration, routing, and discovery

Gradual refactoring of existing services to Spring Boot while preserving business continuity

Open‑Source Contributions

generator‑web : a web UI for MyBatis Generator that automates mapper, entity, and DAO code generation. Repository: https://github.com/zhongxintech/generator-web favorites‑web : a cloud‑based bookmark manager built with Spring Boot, Spring Data JPA, Redis, Thymeleaf, and Gradle. Repository: https://github.com/cloudfavorites/favorites-web This case study demonstrates how systematic architectural upgrades—driven by business pressure and emerging open‑source technologies—can evolve a modest fintech platform into a high‑throughput, resilient system.