Comparison
Retry budgetvsRetry storm
Retry budget
you cap retries at a small fraction of total traffic so a struggling dependency does not get three times the load it was already failing under.
A global limit on retry volume, expressed as a ratio of retries to requests. It exists because per-call retry policies compose disastrously: three layers each retrying three times is twenty-seven calls. Budgets keep retries useful for isolated failures and prevent them from amplifying a systemic one.
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 →