Skip to slide
152 / 221

CH 13 · Reliability: Retries, Idempotency, and Failure Handling · 04/13

Read as one page

Retries with backoff

For transient failures (timeouts, rate limits, 5xx from a provider or external API), retry, but correctly:

  • Only retry idempotent or safe operations. A read is safe to retry; an action with side effects needs idempotency first (below).
  • Exponential backoff with jitter. Wait progressively longer between attempts, with randomization, so you don't synchronize retries into a thundering herd against a recovering service.
  • Cap attempts. A few retries, then surface a clear failure. Infinite retries just move the hang.
  • Respect rate-limit signals. If a provider returns a retry-after, honor it rather than guessing.

Distinguish retryable errors (transient) from terminal ones (bad request, auth failure); retrying a terminal error just wastes time.

Backoff with jitter
Backoff with jitter

Finished this idea?

Your learning progress stays in this browser.