Thundering Herd
A large number of clients acting simultaneously because they were synchronised by a shared event, producing a spike that the steady-state design never sized for.
The synchronising event is usually one of four: a cache key expiring, a service restarting and every client reconnecting at once, a cron schedule that fires on the hour across every instance, or a failure that puts every client into the same retry cycle.
The general fix is the same in all four cases — add randomness. Jittered TTLs, jittered reconnect backoff, jittered schedule offsets, jittered retry delays. Systems that back off perfectly in unison have not solved the problem, they have made it periodic.
Where randomness is not enough, coalesce: single-flight so concurrent identical requests result in one downstream call, or a leader that refreshes on behalf of the others.