Thanks for stopping by and sharing a link to your presentation.

Yes, there are valid uses for Cartesian products – they are rare. In this example I used one to efficiently find all possible combinations of two row sources.

Other ways Cartesian joins may appear, even if the developer has written the SQL statement correctly:
* The query optimizer using transitive closure to generate additional predicates, and in the process the optimizer automatically removed the join condition between those two tables. Example 1 from the Cost-Based Oracle Fundamentals book Example 2 from that author’s website.
* The optimizer incorrectly calculates the row source cardinality of at least one of the row sources as a very small number, such as 1, and determines that a Cartesian join between two possibly unrelated row sources is safe and efficient. Randolf Geist and I included an example of this behavior in chapter 9 of the Expert Oracle Practices book. In this case, all of the correct join conditions were provided in the query.

I am not sure that I would classify your #3 and #4 items as something that would necessarily cause a Cartesian join. I think that those examples could be considered a legitimate 1 to many type join. This, of course, is just my opinion, and I could be completely wrong.

One more use for Cartesian joins: stress testing the server admins or the database admins.

I kept finding them in queries over the years. They were very slow to finish, produced way too many rows, and strange, inaccurate results.

What was odd was that while it was blantantly obvious to me, my coworkers didn’t see it as a cartesian product. They kept barking up the statistics
and index trees. But actually, those can’t speed them up at all.

So, I made a presentation on cartesian products, and presented it at the Oracle User’s Group. See it here:
http://www.lepinsky.com/cartesian/index.htm

It lists five ways to get CP in queries. If you know of any more, please let me know.

The only two real uses that I came up with for CP were
- inserting more test dummy data quickly using rownum
- stress testing your servers