What to look for
- 01
Count retries across the whole call stack, not one client in isolation.
- 02
Check whether the timeout includes connection setup and whether retrying side effects is safe.
Keep in mind
Library examples illustrate Amazon’s systems. Retry budgets and timeout values must be chosen for your workload and failure modes.
Try this next
Trace one request across dependencies and calculate the maximum attempts when every layer retries.
Get the reliability templateNotes and exercises by Awesome CTO. The original work belongs to Marc Brooker, Amazon Builders’ Library.
