Distributed Reliability Decision Examples

Positive example — idempotent ingress

• Context: a sender may repeat a command after losing the acknowledgement.
• Mode: idempotent-ingress.
• Decision: use a stable operation key and atomically record the accepted effect and result.
• Why it fits: redelivery cannot duplicate the consequential effect.

Negative example — unbounded retry

• Context: a remote operation has unknown duplicate semantics and persistent failures.
• Mode: synchronous-transactional.
• Bad decision: retry forever with increasing queues.
• Better decision: bound attempts, expose failure, and establish effect semantics before retrying.