Comparison
JittervsRetry storm
Jitter
you add a random amount to every retry delay so ten thousand clients do not all come back at exactly the same second.
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.
Full entry →Retry storm
the dependency got slow, everything retried, and the extra load is now the reason it cannot recover.
A feedback loop where failures trigger retries that increase load and cause more failures. It is why an incident can outlive its original cause: the trigger passed, but the system is now generating its own load. Jitter, budgets and circuit breakers all exist to break this loop.
Full entry →