Charged Twice on Retry

A payments consumer reads charge events at-least-once and calls a card processor. After a deploy that killed pods mid-batch, a handful of customers were charged twice. Which change actually prevents this?

Raise the redelivery timeout well past the processor's slowest observed response so an in-progress charge is never handed to a second worker

Insert the event id into a dedup table with a unique constraint in the same transaction as the charge record, skip the event on conflict, and pass that id to the processor as its idempotency key

Turn on producer idempotence so the broker stamps each batch with a sequence number and rejects any duplicate the publisher retries after a timeout

Have the consumer acknowledge each message immediately after the processor returns, so the window between the charge and the acknowledgement is as small as possible