That is news to me. Can you provide a citation for this?
I can see the confusion in what I said. To clarify, I really meant that in order to retrieve that data that you need in really old WAL segments, you need to keep your replication slot in a position that will hold that WAL in place. And that is what will prevent old rows from being vacuumed away. There is no way to create a replication slot in the past, so the only way to actually use the old WAL for logical replication is to have an old slot in place with an LSN position from that long time ago.
The bit about rows not being removed is found here:
Thanks,
Jeremy
