Exactly-Once Semantics

in distributed-systems, we want nodes to process every message that needs processing exactly once, not too few and not too many times.
if nodes do not retry when their message isn’t acknowledged, they could potentially miss out on completing the delivery of a message.
if nodes retry too much, they might send the same message too many times.
one way to solve this issue is to have the application code on the receiving node do deduplication to ensure that even if it receives multiples of the same message, it only shows one to the user.
- so, we can have sent messages be at-least-once and receiving nodes to accept at-most-once.
exactly-once processing is possible in distributed-systems, but exactly-once delivery is not.

table without id file.inlinks as Backlinks
where file.name = this.file.name

References.