Exactly-Once Semantics
A processing guarantee aimed at making the externally visible effect of an operation occur logically once even when retries or duplicate deliveries happen underneath.
In distributed systems the useful goal is rarely to prove that a packet or message physically travelled only once. What matters is that durable state appears as if the logical event was applied once, typically through transactions, idempotency keys, deduplication, or a coordinated source-to-sink protocol.
The scope of “exactly once” must be stated explicitly. A stream engine may provide the guarantee for its own managed state while an external HTTP service or independent database remains outside that transaction boundary. Safe retry design in critical systems covers this practical boundary.