Exactly-Once Semantics

ELI5 - The Vibe Check

In a distributed system, messages can get lost, or worse, delivered twice. Exactly-once semantics is the promise that no matter how many retries or crashes happen along the way, each event still ends up processed exactly once, not zero times, not five times. It's one of the hardest promises in distributed computing to actually keep, and most systems that claim it are really faking it with idempotency tricks underneath.

Real Talk

Exactly-once semantics guarantees that a message or event is processed and its effects applied precisely one time, even in the presence of retries, failures, or duplicate delivery. True exactly-once delivery is generally considered impossible across an unreliable network, so systems like Kafka achieve effectively-once processing by combining at-least-once delivery with idempotent producers or transactional commits.

Show Me The Code

// Kafka producer config for effectively-once semantics
const producer = kafka.producer({
  idempotent: true,
  transactionalId: 'orders-producer-1'
});

When You'll Hear This

"Kafka's transactional producer gives us exactly-once semantics end to end." / "We faked exactly-once with an idempotency key instead of fighting the framework."