Primay keys: UUID vs bigint – or both?

Some tables in the project I’m working on have UUIDs as primary keys others are indexed by bigint. To have both wasn’t a conscious decision. I began by choosing UUIDs for some tables because I expected I’d need to expose those ids in the URLs. I intended to use UUIDs throughout the model but I forgot and ended up with some tables being keyed by integers.

I wondered if it was a bad idea to have primary keys of different types, and I came across this post. There’s plenty to take in and I’ll need to read it a few more times to fully understand all that the author is expressing, but my question of whether or not it’s a good idea to mix primary key types is answered by the author in the comments in which he states the rule:

“Normalize first, de-normalize only when needed.”

I don’t yet have a need to de-normalize. It might not matter for me yet – not at the scale I’m working at, at least – but I’m about to implement a model that has a polymorphic association where the foreign key might point to either a UUID or bigint. I think that will be a problem.

And then there’s the question of which I should normalize to. My initial assumption about UUIDs exposed in a URL being more secure might be misplaced. “Using a PK in a URL is generally a bad practice.”