From CDN static caching to queue-based peak shaving, we break down the architecture practices that supported a single flash sale with 500,000 users.
A chain brand's anniversary flash sale peaked at 500,000 users online simultaneously. All four layers of protection are essential:
Layer 1: CDN static caching, so 95% of product detail page requests never reach the origin server; Layer 2: rate limiting and degradation, with token bucket rate limiting at the gateway; Layer 3: queue-based peak shaving, with order requests made asynchronous and inventory deduction serialized; Layer 4: data fallback, with Redis pre-deducted inventory plus eventual consistency in the database.
The most easily overlooked issue is the "post-flash-sale" traffic: if the inventory replenishment logic for released unpaid orders is handled poorly, it will cause overselling or inventory lockup.