High-Concurrency Architecture and Optimization Strategies in Cloud Development

Introduction

During the COVID‑19 pandemic, the Yunnan Provincial Public Security Department launched a "Yunnan Anti‑Epidemic" mini‑program to register the flow of people in public places. Each venue obtained entry and exit QR codes, and users scanned them when entering or leaving. Within a month, the program registered over 1.03 million venues and recorded 3.65 billion scans, with peak request rates exceeding 100 k requests/min for QR‑code generation and 800 k requests/min for scan registration.

These numbers raised two critical questions: how to handle such short‑term traffic spikes and how to keep the mini‑program responsive under high concurrency.

2. High Concurrency – A Significant Technical Challenge

In traditional development, engineers must predict traffic peaks, provision sufficient resources in advance, and design the architecture to handle the load, which often leads to over‑provisioning or performance bottlenecks and incurs high costs.

Cloud Development, a serverless, multi‑end solution, simplifies this problem. Its lightweight, elastic architecture can automatically scale horizontally to handle massive concurrent requests. Developers benefit from a NoOps approach: the platform automatically expands or contracts capacity, eliminating the need for manual resource provisioning and dramatically reducing operational costs.

Cloud Development also provides SDKs for WeChat Mini‑Programs, QQ Mini‑Programs, web, and mobile apps, allowing developers to focus on business logic without worrying about backend infrastructure. Daily calls exceed 700 million, with some customers reaching over 100 million requests.

3. High‑Performance, High‑Availability Architecture Design

The service ensures two aspects of high availability: (1) the Cloud Development platform itself remains continuously available, and (2) it offers customers the ability to sustain high‑traffic workloads.

All functional modules are deployed across multiple clusters and data centers, achieving >99.99% availability and ultra‑low latency. The overall architecture consists of two main layers: the "Data Pipeline" and the "Underlying Resources".

Data Pipeline

Requests from end‑side SDKs traverse a series of services (authentication, routing, security, caching, etc.) before reaching the resource layer. The pipeline is designed to be simple, stateless, degradable, multi‑level cached, and automatically fails over. Additional optimizations include multi‑cluster deployment, retry strategies, network optimizations, high‑performance implementations, and fast/slow request separation.

Underlying Resources

Key components are Cloud Functions and Cloud Databases.

Cloud Function Optimizations

Cloud Functions are deployed in multi‑cluster, cross‑region configurations with automatic fault detection and removal. A large resource pool and buffer allow rapid scaling. Cold‑start latency has been reduced to ~200 ms using a lightweight MicroVM, and hot‑starts are kept fast. Optimizations include code‑download caching, network deployment improvements, container‑startup acceleration, real‑time concurrency prediction, version aliasing with rolling updates, and instance reservation tuning. A single function can handle 1 000 concurrent executions; with 100 ms execution time, this yields 10 000 QPS per function, and 50 functions can reach 500 k QPS.

Cloud Database Optimizations

The database layer features a multi‑layer access tier with horizontal scaling, multi‑cluster deployment, automatic fault isolation, daily full backups to object storage, online hot migration, and automatic indexing. The system also supports automatic index creation based on slow‑query analysis, reducing the need for manual performance tuning.

4. Optimizing High‑Concurrency Business with Cloud Development

For gradually increasing traffic, Cloud Development handles the load out‑of‑the‑box. For bursty traffic (e.g., flash sales), additional strategies are needed:

Estimate peak QPS and concurrency in advance.

Pre‑warm the activity by triggering functions without business logic.

Adjust the billing plan (pay‑as‑you‑go or subscription) to ensure sufficient quota.

Technical optimizations focus on reducing latency:

Minimize calls to dependent services and merge requests.

Parallelize dependent service calls.

Consolidate or split functions based on business logic.

Trim code size and dependencies to lower cold‑start time.

Database optimizations include creating efficient indexes, reducing query volume, using selective query conditions, limiting returned fields, avoiding unnecessary transactions, caching intermediate results (e.g., leaderboards), and sharding large collections.

Additional recommendations: conduct load testing, isolate activity‑specific services, and monitor slow‑query logs.

5. Author Introduction

Yan Jie, senior front‑end engineer at Tencent, core member of the Cloud Development team, focuses on backend system development and architecture design.

Promotional Links

Mini‑program giveaway: 5 books + Tencent Video VIP month card

For more technical resources, see the following articles:

IDEA 2020.2 Crack Tutorial (valid until 2089)

The History of Kubernetes

What Kind of Architect Do You Have?