This is much clearer thank you. From your initial example, I thought you were suggesting to just load 2 large tables into memory at once and iterate through them to find matches.

Also, at the surface, this seems like it would be a good application for using materialized views. I obviously don't know the ins and outs of what the usage here was, but this seems like a case where slowing down the insert would be a good speed sacrifice for retrieval.

I'm not going to even feign being an expert at database optimization, but that seems like it might be a more efficient way to approach it.