Overview of Alipay System Architecture and the Metamorphosis Distributed Message Middleware

Alipay System Architecture Overview

The diagram above illustrates the overall architecture of Alipay's payment system.

Typical Processing Flow

This flow shows the default processing steps for a payment transaction.

Fund Handling Platform

The platform manages the movement of funds between accounts.

Financial Accounting

Accounting modules record all financial activities.

Payment Settlement

Settlement processes finalize the transfer of money to the intended recipients.

Accounting Center

The accounting center consolidates and reconciles all transaction data.

Transaction

Core transaction processing ensures correctness and durability.

Flexible Transactions

Various images illustrate the flexible transaction mechanisms employed by Alipay.

Metamorphosis (MetaQ) – Alibaba's Open‑Source Distributed Message Middleware

Metamorphosis (MetaQ) is a high‑performance, highly available, and scalable distributed messaging system similar to LinkedIn's Kafka. It provides sequential message storage, high throughput, support for both local and XA transactions, and is widely used in both Taobao and Alipay.

Metamorphosis originated from the author's study of LinkedIn's open‑source MQ and Apache Kafka. Unlike the typical push model, MetaQ uses a pull mechanism and heavily relies on ZooKeeper for service discovery and offset storage.

Reasons for Re‑implementing Meta (Compared to Kafka)

Kafka is written in Scala, which the author is not familiar with, and its community development is perceived as slow.

Kafka lacks several features required by the author, such as full transaction support, multiple offset storage options, and advanced high‑availability (HA) solutions.

Meta offers unique capabilities not present in Kafka.

Transparent text‑based protocol, similar to memcached stats, for broker monitoring.

Pure Java implementation covering communication, storage, client, and server.

Transaction support, including local and XA distributed transactions.

HA replication with both asynchronous and synchronous modes to ensure message reliability.

Asynchronous message sending.

Local recovery on consumer failure.

Multiple offset storage back‑ends (database, disk, ZooKeeper) with customizable implementations and group commit support.

Message broadcast mode.

Complementary projects such as a Python client, a Twitter Storm spout, and Tail4j.

Meta processes an enormous volume of messages daily—approximately 12 billion messages routed through Meta in Alipay and hundreds of millions in Taobao.

Suitable Applications for Meta

High‑throughput log transport.

Message broadcasting, e.g., cache invalidation notifications.

Ordered data synchronization, such as MySQL binlog replication.

Message routing in fully clustered broker/producer/consumer environments where order and reliability are critical.

General purpose MQ use cases.

© The content is sourced from the web; all rights belong to the original author. If any infringement is identified, please notify us for prompt removal.