Duplicates DownstreamKafka · medium · ~12 minFinance found 1,842 duplicated orders in the fulfillment database over the past month. Each duplicate cluster timestamps within seconds of a consumer deploy or rebalance. The pipeline: orders topic → consumer → INSERT into Postgres.The team wants to 'turn on exactly-once'. Decide what's actually happening and what actually fixes it.Evidence

Evidence

Config

Checkpoints

Where do the duplicates come from?

Why doesn't 'turn on exactly-once' (Kafka transactions) fix this pipeline?

The robust fix for this sink?

Explain at-least-once vs at-most-once vs exactly-once for THIS pipeline in interview language — where each guarantee lives and what you chose.