python-resilience

Use the tenacity library's @retry decorator with stop_after_attempt and wait_exponential_jitter. Combine retry_if_exception_type for network errors with retry_if_result to also retry on transient HTTP status codes like 429, 502, 503, and 504.

Exponential backoff increases wait time between retries so recovering services are not overwhelmed, while jitter adds randomness to prevent many clients from retrying simultaneously. Tenacity provides this via wait_exponential_jitter with configurable initial and maximum delays.

Never retry permanent errors such as ValueError, TypeError, authentication failures, or HTTP 4xx client errors other than 429. These indicate bugs or invalid requests that will fail identically on every attempt, wasting time and resources.

Wrap the coroutine with asyncio.wait_for inside a reusable decorator that accepts a seconds parameter. Apply the decorator to any async function so it raises a timeout error when execution exceeds the configured limit.

When many clients fail simultaneously and retry on the same fixed schedule, they hit the recovering service in synchronized waves. Adding jitter randomizes each client's wait time, spreading requests out and giving the service capacity to recover.