How a Mini‑Program Scaled to Tens of Thousands QPS with Alibaba Cloud Serverless

Shenzhen Yichuan Technology, a provider of precise marketing and internet ecosystem services, built its core business on an Alipay mini‑program that quickly grew to over one million daily active users. Rapid traffic spikes during promotions caused server overloads, low resource utilization, and high operational complexity.

To address these challenges, the company adopted Alibaba Cloud Function Compute, a serverless platform offering sub‑second elasticity, automatic scaling, and a pay‑as‑you‑go model. Over three years they migrated new, legacy, internal, and external applications to Function Compute, achieving stable operation at tens of thousands of QPS and thousands of concurrent functions.

Best Practices

1. Development Process

The team primarily uses PHP and frameworks such as Laravel. They split projects into one or more files, each deployed as an independent function. Serverless Devs tooling enables near‑zero code changes for migration, and function layers are used to share common code (e.g., red‑packet logic) across services.

2. Pipeline and Gray Release

Code is stored in SVN; commits trigger actions that call Function Compute APIs. Deployments use Function Compute versions and aliases: a new version is published, the release alias points to it for production traffic, and a separate test alias allows staged testing. Switching aliases enables one‑click rollback and clear separation of production and test environments.

3. Operations Management

Function Compute integrates with Log Service (SLS), generating a log entry per request for easy error filtering. Built‑in monitoring provides metrics such as request count, error count, concurrency, and execution latency, with alarm rules for errors, allowing developers to handle both development and operations efficiently.

Results

Stability Boost : No need to manage backend servers; Function Compute automatically scales to handle traffic spikes, maintaining smooth performance during large promotions.

Fast Onboarding : New developers can start coding functions without learning complex server management, reducing learning curves.

Cost Efficiency : Running a mini‑program with over 500,000 daily active users costs roughly ¥200 per day, considered inexpensive for the provided elasticity and support.

Future Suggestions

Support fixed IP addresses for function entry points to satisfy regulatory IP whitelist requirements.

Enable granular version publishing per individual function to improve precise gray‑release capabilities.