Implement bounded retries with backoff and jitter

Retries can recover a transient failure or amplify an outage. Classify errors and enforce a total budget.

Generate deterministic delays

The integration lab uses two 503s followed by success.

from integration_cases import retry_case

result = retry_case()
assert result["statuses"] == [503, 503, 200]
assert result["attempts"] == 3
assert result["bounded"] is True
assert result["retryable_only"] is True
assert len(result["delays_ms"]) == 2
assert result["slept"] is False

The calculation is deterministic and does not actually sleep.

Retry timeouts, throttling and selected 5xx responses; do not retry invalid input, authentication or permanent business errors. Honour Retry-After, use exponential backoff plus randomized jitter, cap delay/attempts and stop at the request deadline.

Writes require idempotency. Propagate cancellation and record attempt/outcome. Test that many clients do not synchronize on one delay. Circuit breakers and queue backpressure can protect an unhealthy dependency.

The FDE for Freshers course connects retries to API reliability and cloud operations.

Exercise

Create a fake service returning 503, 429 and 400. Prove only transient responses retry, Retry-After is respected and the total deadline wins.

Continue learning

This article is part of the FDE integration and deployment foundations sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in Design a webhook receiver that tolerates duplicate events.
• Continue with Add idempotency to an order-processing endpoint.

Reference: AWS Builders’ Library on timeouts, retries and backoff.