Practices for Handling High Traffic on JD.com Product Detail Pages

JD.com product detail pages must support diverse templates (general, flash sale, global purchase) and a multitude of services such as pricing, promotion, inventory, advertising, reviews, and recommendations, which makes static page generation impractical for billions of items due to long rebuild times and inflexibility.

To overcome these limits, JD migrated from .NET static pages to a dynamic architecture: data is stored in a KV store, while the front‑end is served by Nginx with embedded Lua scripts, enabling fully dynamic template rendering and rapid response (average 30‑40 ms, peak under 50 ms).

A unified service system was built to funnel all page‑level requests through a single entry point, providing monitoring, real‑time degradation switches, and data heterogeneity handling; services are categorized by dimensions (product, category, merchant, brand) and stored separately to avoid cross‑service impact.

Key design principles include heterogeneous data storage, closed‑loop service management, and dimensional storage, allowing independent scaling and quick isolation of problematic services through level‑based prioritization (e.g., inventory as level 0, warranty as level 1).

Isolation is applied at multiple layers: servlet‑level asynchronous pools, deployment‑level grouping, and application‑level separation, preventing failures in one service from cascading to others.

Caching strategies combine local, distributed (Redis/Memcached), and consistent‑hash based Nginx caches, with selective caching for high‑traffic data and automatic degradation for non‑critical services when latency thresholds are exceeded.

Comprehensive monitoring, logging, and automated testing (including real‑traffic replay and A/B experiments via Lua) ensure system health, while mechanisms such as 304 responses, rate limiting, and CDN versioning further protect against spikes and crawler load.

The overall approach enables JD to handle massive traffic surges, maintain low latency, and quickly adapt page layouts or new services without full static regeneration.