Cache Stampede
A failure pattern in which many clients simultaneously recompute or reload the same expired or missing cache entry, overloading the backing system.
Consistency and Failure-Model Boundary
A cache stampede is coordinated miss amplification around the same item; it is related to but distinct from a general thundering-herd wakeup pattern.
Related Distributed-Systems Concepts
Why One Hot Key Can Create Disproportionate Load
A simple capacity approximation shows why this failure can be abrupt. If one key receives λ requests per second and regeneration takes S seconds, then around expiry an unprotected system can start roughly λ×S overlapping regenerations. This is not a strict bound; it is a coarse concurrency indicator whose exact value depends on the arrival process and application behavior.
Single-flight or request coalescing can reduce regeneration of one key to one in-flight job; randomized TTLs reduce synchronized expiry across keys; stale-while-revalidate can temporarily serve an older value while regeneration happens separately. A cache stampede is therefore more specific than a high cache-miss rate and differs from the broader Thundering Herd pattern. For the problem definition and probabilistic early-expiration approach, see Vattani et al..