Retry with Exponential Backoff

Back to Distributed System Patterns

Retrying failed operations with exponentially increasing delays (1s, 2s, 4s, 8s…) and random jitter to prevent thundering herd. Must set max retries and total timeout. Only retry on transient errors (5xx, timeouts), not permanent failures (4xx). Operations must be idempotent for safe retries.

Key Properties


distributed-systems patterns resilience