How Flink Powers Real‑Time Process Operations in China Construction Bank

Company Introduction

Jianxin FinTech is the fintech subsidiary of China Construction Bank, transformed from the bank's software development center, dedicated to driving the “new finance” ecosystem and supporting the digital transformation of the bank and digital China.

Business Background and Challenges

Modern applications use front‑back separation; each transaction in the bank generates three messages: front‑end trace, HTTP request, HTTP response. Hundreds of scenarios (cash delivery, credit‑card approval, etc.) produce massive, fragmented data that is isolated across systems, making unified analysis difficult.

The goal is to ingest all logs, reconstruct business processes from a business‑centric view, and provide real‑time analytics for various roles (customers, staff, managers).

Solution Evolution and Technical Challenges

Three‑stage evolution:

Version 1.0 – sliding‑window join with Redis for state; low throughput, Redis pressure.

Version 2.0 – Flink interval‑join using RocksDB; OOM and large checkpoint size.

Version 3.0 – custom keyedProcessFunction with manual state management, reducing state by 90 % and improving stability.

Key challenges include multi‑source data, high latency tolerance (up to one hour), massive data volume (hundreds of billions per day), and the need for a one‑to‑one join of request, response, and trace.

Process Indicators

Two iterations of real‑time metrics:

1.0 – hybrid Flink‑Spark pipeline with minute‑level latency, data flow: Kafka → Flink → Kafka → Spark → GP → Oracle.

2.0 – pure Flink pipeline writing directly to Oracle, achieving second‑level latency.

Metrics cover channel, product, and institution dimensions, such as approval rates, activation rates, and average processing times.

Business Impact

Real‑time process reconstruction enables end‑to‑end visibility, risk intervention, resource optimization, and supports the bank’s digital transformation.

Future Outlook

The solution will be productized and extended to other industries, bringing financial‑grade process operations beyond the bank.