Nice article, nice to see some alternatives to the "insert in staging table and then do an update".

However, it's a we can't review the UPDATE statement used. The article says the UPDATE is performed using the surrogate key, so I can imagine this partitions nicely within the conditional split and you avoid locking issues on the destination table.

I can think about other scenarious though where the UPDATE statement isn't that straight forward and where multiple threads may interfere with each other.

Anyway, very nice article and thanks for the effort.