Identifying vs Non-Identifying Relationships (Again!!!)

Question

So, I\'ve read a whole lot of answers here on stackoverflow, but I\'m still confused about the whole concept thereof. Specifically, I\'ve gone over this article (including a

Answer 1

You are over-thinking the linkage between optionality and identity. Until the whole thing comes more naturally to you, it's best to think of them as being completely unrelated.

About optionality, it is important to remember that the optionality is directional. To use your example of employee_equipment: Sure, employees don't need equipment. The one-to-many relationship from employee to employee_equipment is optional. At the same time, looking at it from the opposite perspective, the relationship is mandatory. You can't have a record in employee_equipment unless there is an employee to associate it with.

Identity has nothing to do with optionality, except coincidentally an identifying relationship is mandatory from the child to the parent. Whether it is also mandatory from the parent to the child is neither here nor there as far as identity is concerned.

What makes a relationship identifying is that you have to know what parent you are talking about (as well as some other things) in order to know what child you are talking about. That is, the primary key of the child must include a foreign key to the parent.

Pure intersection tables (e.g. employee_equipment) are good examples of this. The primary key of a pure intersection is the combination of the foreign keys to both parent tables. Note that some people may also add a surrogate key to these kinds of tables. It doesn't matter so much from an identity perspective if there are multiple candidate keys. What is important in determining identity is whether the foreign key is part of a candidate key, whether or not that candidate key happens to be the primary key.

Another good example would be something like a database's metadata catalog, where a column is identified by the table to which it belongs, just as the table is identified by the schema it is in, and so on. Knowing that a column is called NAME doesn't tell you which column it is. Knowing that it is the NAME column in the CUSTOMER table helps. (You'll also have to know which schema CUSTOMER is in, and so forth).

Answer 2

Joel has provided a good answer (+1 to him), let me just offer a small mental shortcut that you can use when thinking about identifying relationships... just ask yourself:

Can I achieve uniqueness only with the attributes of the child entity?

If no, and you need to include the attributes migrated from the parent into the child key to make it unique, then you have an identifying relationship¹. It's about identification-dependence, not existence-dependence²!

You might be interested in this post for some more musings on the topic.

---

¹ And the child entity is "weak" or "dependent".

² Although identification-dependence usually implies existence-dependence.