2 ms·
What the parent poster described isn’t what makes Kafka’s “exactly once” semantics work. It’s the use of an idempotency token associated with each publication,
by otterley 1y ago
What the parent poster described isn’t what makes Kafka’s “exactly once” semantics work. It’s the use of an idempotency token associated with each publication, which effectively turns “at-least-once” semantics into effectively “exactly once” via deduplication.
- saberience 1y agoWe used Flink and S3 compatible backends (for the Flink state) to ensure exactly once processing. I didn't say Kafka magically solves these problems for you, but it was required for the scalability we needed.