Keep a CDC consumer from doubling a row after it restarts

Your consumer crashes right before it commits its Kafka offset, and every row it just wrote lands in the table twice.

IntermediateRead 6 minDoing: a few hours6 steps

▶ Start

What you'll have at the end

A consumer that writes to the sink and commits its offset in the same step, tested by killing it mid-batch

Before you start

You need

A consumer already reading a CDC topic's change messages and writing each one into a sink table, with no protection yet against reading the same message twice.

Not covered

What happens when the change is a delete, or when two genuinely different updates to the same row arrive out of order; both need their own handling this recipe doesn't build.

Leans on

6 steps

Where to go next

Make a deleted row vanish downstream with a tombstone

go there when the change is a delete instead of an insert or update.

Commit Kafka consumer offsets so a crash doesn't skip or repeat messages

go there for the same crash-safety problem in a plain Kafka consumer with no CDC sink write involved.

Checked 26 Aug 2026

Part of the Change Data Capture (CDC) cookbook