Principles
Use timestamps, queue IDs, remote responses, source IP, VMTA, sender domain and message class to establish the failure boundary. Avoid reacting to one isolated delivery attempt.
Policy design
- Fast early retries for transient network conditions
- Progressive spacing for provider throttling
- Finite lifetime matched to message value
- Separate policies for transactional and bulk traffic
Validation
Review the distribution of successful deliveries by retry attempt before shortening or extending queue lifetime.
Operational rule: collect evidence before changing policy, make one controlled change, verify the result, and retain a rollback path.