How to tame the thundering herd problem

The thundering herd problem occurs when many clients simultaneously request the same resource, overwhelming backends during cache misses, traffic spikes, or coordinated events. Key causes include synchronized cache expiration, lock contention, and auto-scaling cold starts. Solutions using Redis include TTL jitter to stagger expirations, request coalescing via distributed locks, rate limiting with atomic counters, load shedding, and queue-based processing with Redis Streams. Bloom filters help prevent cache penetration. The post also compares Redis Cloud against Amazon ElastiCache and Google Memorystore, noting that both competitors are frozen at Redis 7.2 (Valkey fork) and lack features like Active-Active geo-distribution and native vector search.