Retry Storm vs Jitter

Posted Aug 15, 2026 · 1 view

A service fails. Every client notices at the same moment, waits the same computed delay, and retries — together. The retry wave is synchronized, which means the recovering service is hit by its whole client base at once, fails again, and the cycle repeats.

Jittered backoff changes one thing: the delay is drawn at random below the exponential ceiling instead of being computed exactly. Same clients, same ceiling, same total number of retries — spread across the window instead of stacked on one instant.

The animation holds everything constant except comet timing. That is the entire lesson: the fix is not fewer retries, it is unsynchronized ones.