Hi all, I'm reading the flink cdc doc
https://nightlies.apache.org/flink/flink-cdc-docs-master/docs/connectors/flink-sources/mysql-cdc/#tables-without-primary-keys
It says that if there's an update operation on the chunk key column, then flink cdc can only provide at-least-once semantic. This does not make sense to me because I don't see differences between incremental snapshot algo handling non-pk chunk column and pk column. So why would that be a problem?
To be more specific, the doc provides an example
https://nightlies.apache.org/flink/flink-cdc-docs-master/docs/connectors/flink-sources/mysql-cdc/#warning
This also confuses me, why won't the cdc converts the update into a delete and an insert, and split-0 handles delete, which deletes the chunk record id=0,pid=2, ignores the insert, while split-1 ignores the delete, and simply add an insert event ——just like a table with pk?
So at the end of the day, what's the essential difference between handling updates with pk, and without pk?