How to Build a High‑Availability, Scalable E‑Commerce Backend for Mega Sales

Challenges of Promotional E‑Commerce Scenarios

E‑commerce platforms driven by flash sales such as 618 and Double 11 must handle extreme traffic spikes, requiring 99.99% availability, rapid iteration, and the ability to scale from tens of thousands to millions of orders without disrupting the architecture.

Layered Architecture Overview

Data Layer : Real‑time user behavior and transaction data are stored in NoSQL, relational databases, and big‑data platforms.

Infrastructure Layer : Includes middleware and services such as message queues (MQ), job debugging centers, SSO, distributed file storage, application monitoring, automated deployment frameworks, and A/B testing support.

Basic Service Layer : Provides common services like OMS, PMS, shipping templates, and delivery zones—core functionalities for any e‑commerce system.

Business Service Layer : Hosts user‑facing features (shopping cart, order processing, homepage) and third‑party API integrations, representing the core of the website applications.

Virtual categories act as tags (e.g., “618 Flash Sale”) to group products for display, while the topmost layer consists of client endpoints such as H5, PC, and mobile apps.

Stateless Micro‑Service Design

Legacy monolithic applications are first transformed into stateless services using SSO tokens, enabling horizontal scaling. The implementation typically relies on Spring Boot and frameworks like Dubbo, Zookeeper, Hystrix, RocketMQ, Elasticsearch, and Redis.

Monolith Splitting Strategies

Splitting can follow system boundaries (frontend vs. backend), functional domains (user service, pricing, inventory), or read/write separation (search index vs. transaction services), allowing independent deployment of each micro‑service.

Data Heterogeneity and Caching

Large order tables are sharded by ID or user dimension, with hot, warm, and cold data stored in different locations. Aggregated data is cached in Redis to form a data‑closed loop, reducing AJAX request volume.

Caching techniques span browser‑level headers, app‑side asset bundling, CDN distribution, and in‑memory caches behind soft load balancers.

Message Queue Usage

Message queues decouple services, handle failures with retries and compensation mechanisms, and achieve eventual consistency—sufficient for most e‑commerce scenarios.

High‑Availability Design

Key practices include service degradation switches in a configuration center, service grouping and isolation (e.g., dedicated services for flash‑sale modules), and rate‑limiting mechanisms.

Traffic Governance

During peak events, traffic is throttled at multiple layers, bots are limited, and pre‑sale circuit breakers are employed. Versioned data structures enable safe rollbacks.

Business Logic Considerations

Order payment flows incorporate idempotency keys or tables to prevent duplicate processing, especially for financial operations. State machines manage order lifecycle transitions, handling both forward and reverse flows.

Mobile‑Specific Architecture

Mobile clients cache static assets, perform intelligent dynamic routing, and use CDN and network‑condition detection to optimize performance in complex network environments.

Instrumentation and Monitoring

Full‑link tracing captures every user action from the app through network, service, and middleware layers, enabling rapid issue localization. Gateways enforce rate limiting and dynamic load balancing, while monitoring systems provide alerts based on predefined thresholds.