## The Latency Pyramid in Web Applications Every database query carries network overhead. When an article trends on social media and receives thousands of requests per second, querying MongoDB or PostgreSQL on every request quickly degrades responsiveness. A multi-tier caching strategy guarantees sub-10ms response times globally. ### 1. Cache-Aside Pattern with Automatic Circuit Breakers In VELO, our caching layer is built to be resilient: if the Redis cluster ever becomes unreachable, queries fail open and transparently query the primary database: ```typescript export async function runCached<T>( keySegments: (string | number)[], tags: string[], ttlSeconds: number, fetcher: () => Promise<T> ): Promise<T> { const cacheKey = keySegments.join(":"); try { const cached = await redis.get<T>(cacheKey); if (cached !== null) return cached; } catch (err) { console.warn("Redis read failed, falling back to database:", err); } const freshData = await fetcher(); try { await redis.set(cacheKey, freshData, { ex: ttlSeconds }); } catch (err) { // Non-blocking write failure logging } return freshData; } ``` ### 2. Tag-Based Invalidation Instead of guessing raw keys to clear, tag-based invalidation allows flushing all posts belonging to an author or topic with a single tag broadcast.