> A foreign key doesn't necessarily imply an index. If you are using postgres, you would have to add an index in addition to the foreign key, if you want one.

Sure, I meant that an index in which the key is primary.

> On delete cascade, depending on how many rows it cascades to, can be problematic because it's a very long running blocking operation. That's something one might want to do as a background operation and in batches. Although that won't make it faster.

That can definitely be a problem (just like destructor deallocation stampedes in C++ or Rust). Still less risky than cascading manually and asynchronously, I suspect.