Backend & systems·Resilience and failure handling
you add a random amount to every retry delay so ten thousand clients do not all come back at exactly the same second.
Jitter
Also calledrandomised backoff
Randomness added to a retry or refresh interval to decorrelate clients. Without it, a fleet that failed together retries together, arriving as a synchronised spike that reproduces the original failure. Full jitter — a uniform random value up to the backoff bound — beats adding a small perturbation and is the version worth remembering.