Evolution of Financial‑Grade Message Queues at Ant Financial

Ant Financial's message queue has been in production for over a decade, initially using ESB‑style mechanisms and later co‑building a new system with Taobao to address message loss issues.

The financial‑grade environment imposes four critical demands: extremely high reliability (no message loss), strong consistency (transactional correctness), continuous availability (service must be up during peak events), and ultra‑high performance (handling billions of messages and millions of TPS).

To satisfy these demands, the system implements ACK mechanisms inspired by TCP, retry logic, persistent storage guarantees, two‑phase transactional messaging, and extensive availability strategies ranging from thread‑pool isolation and rate limiting to multi‑data‑center active‑active deployments and the LDC (multi‑active) architecture that provides flexible zone‑aware routing.

With the rise of big‑data (OLAP) workloads, Ant Financial introduced a pull‑mode message queue based on log semantics, deploying it on physical machines to efficiently serve analytical scenarios.

Subsequently, a compute‑storage separation strategy was adopted: first a "挂盘" (mounted) mode using distributed file systems for durability, then an API mode that moved commit logs into queues, added global fixed partitions, idempotent sending, and strong ordering, improving performance and configurability.

Finally, computation capabilities were embedded into the queue using a lightweight, non‑centralized streaming framework that supports common operators and window semantics, turning the queue into a platform that not only transports and stores data but also processes it.

Overall, the message queue has evolved from a simple data conduit to a smart transmission‑computation service, continuously expanding its role in Ant Financial's distributed architecture.