5 / 6 Stampedes, Hot Keys & Cache Failure

Every Key At Once

Two overload events on the same service. Event A: one key taking 40,000 reads per second expires. Event B: the cache tier is restarted and every key is gone.

Select every mitigation that bounds total origin load during event B, no matter how many distinct keys the traffic asks for. More than one applies.

A

Probabilistic early expiration, recomputing shortly before expiry

B

Per-key request coalescing, capping each key at one in-flight origin call

C

A warm second tier, or a replica the restarted nodes read through before the origin

D

A bounded concurrency limiter on origin calls, shedding or queueing anything above the limit

E

Restarting cache nodes one at a time so most of the tier stays warm

F

Randomised TTL jitter so keys do not expire together