I am designing a database which needs to store some fairly complex relationships among many different types of entities.

In the simplest example, let us say I have three types of entities - "Student", "Tutor", and "Course", each with their own tables (student, tutor, course). First of all, I need to represent simple many-to-many relationships between each possible pair of entity types:

  • student_course - Each student can be enrolled in zero or more courses, and each course can have zero or more students. (ex: "Alex is enrolled in Biology").
  • student_tutor - Each student can be working with zero or more tutors, and each tutor can be assigned to zero or more students. (ex: "Emma is assigned to Alex").
  • course_tutor - Each tutor can be approved to teach zero or more courses, and each course can have zero or more approved tutors. (ex: "Emma knows how to tutor Physics").

Easy enough. However, I also need to represent some more advanced relationships, like:

  • "Tutor John is working with students Alex and Maria on course Chemistry"
  • "Tutor John is working with student Alex on course Algebra"

Notice that these are not captured by the binary many-to-many relationships I described earlier - John can be tutoring Maria, and Maria can be enrolled in History, but that doesn't mean that John is tutoring Maria in History.

The way I see it, I have two options:

Option 1 - A single table containing triplets

Table student_course_tutor:

| student_id | course_id | tutor_id |
| 1          | 1         | 1        |
| 1          | 1         | 2        |
| 1          | 2         | 5        |

This is nice because it explicitly lays out the full relationship in a single row. However, I now need to maintain this table in addition to my simple relationship tables.

Option 2 - Meta-relationships

Table student_course:

| id | student_id | course_id |
| 1  | 1          | 1         |
| 2  | 1          | 2         |

Table student_course__tutor:

| student_course_id | tutor_id  |
| 1                 | 1         |
| 1                 | 2         |
| 2                 | 5         |

Since I was going to have a student_course table anyway, adding this "meta-relationship" table is less redundant than representing the full triplet. However, it now becomes a lot more abstract and difficult to understand.

Which way is more correct?

  • 1
    The correct one is the one you understand better, and follows good design principles. There's no perfect design, and you have to use this database daily. Go with your gut. You know what you're doing. Aug 14, 2016 at 22:34

3 Answers 3


The two examples are not equivalent and interchangable implementations; they embody different semantics.

In the first, the three-way table imposes no conditions on the participants. The example shows this. Alex is enrolled in Biology, Emma is assigned to Alex and Emma knows how to tutor Physics. The tutor's subject is Physics but has a student who's enrolled in Biology and there's nothing in the model to stop them discussing, say, Economics.

In the second, the tutor_id can only be associated with a pre-existing student_course_id. So Alex has to be enrolled in Biology before Emma can become his tutor. Since there's only one course_id (through the associated student_course) we can assume this is what they will discuss in their tutorials. However, there is still no way of asserting that Emma is part of Alex's faculty.

If you're still working out the logical data model I would suggest you skip the surrogate keys and use only the natural keys for now i.e. drop the various "id" columns and only use course.name, student.name and tutor.name. Surrogate keys are great performance enhancers in real DBMS implementation but are not required when understanding and documenting the problem. They can be substituted in later when you're confident you have solved your problems.

Next you need to understand the constraints on the data and the questions the DB has to answer. For example a constraint may be that a tutor can only work in the subject which she is employed. A question may be "who's available to tutor a Chemistry student?" Once you have these, the tables to enforce and answer them will emerge, as will the foreign key constraints.

You may be content with any tutor teaching any subject to any student. That's OK if it is so, I can't tell form the question. Or you may choose to enforce this outside of the database - in the application, say, or through a written policy enforced by management. The important thing is to understand the rules as they are, and how they're implemented, so that when they change the appropriate adjustments can be made.

Do not be afraid of having lots of tables with a few rows if this is what the data demands. Would you bolt on a new clickstream half way through an existing web site just because you "were going to have it anyway"?

  • Thanks, you've made it clear that I haven't completely thought through my requirements. In my application layer, I'll probably end up automatically adding the student-course and tutor-course relations to support a particular student-tutor-course tuple. However, as I mentioned in another comment, it is possible that Emma will tutor Alex in Chemistry this semester, but then realize that actually she hates tutoring Chemistry next semester. Thus I'd want to remove her from her assignment to Chemistry without erasing her historical assignments. I guess soft deletes could support that too.
    – alexw
    Aug 17, 2016 at 17:17

I like option 1 better. My first thought is always maintaining data integrity so my question is, "Could it be possible to relate a student to a tutor for a course the tutor is not approved for?" With table student_course_tutor, the field student_id would, as you say the student needn't be attending the course, be a FK to the Students table and the fields (course_id, tutor_id) would be a FK to the course_tutor table. If the requirement that the student must be enrolled in the course as well as the tutor must be approved for the course, that would require only that the fields (student_id, course_id) be a FK to the student_course table.

So you cover the existing specifications and you have an easy solution for a not unreasonable future requirement.


Option 2 is correct, Option 1 - not, because in 1-st option you denormalize you data. It's bring problem to you in future. For example if you update tutor in course you will be update all tutor id in triplet.

  • Could you elaborate? I can imagine a scenario where a tutor used to teach a particular course, but is no longer teaching that course now.
    – alexw
    Aug 15, 2016 at 16:00
  • Yes, i can. It's mean you system have time dimension in model. In this case everything is more complex. You need more data structures to keep info about cases. Aug 15, 2016 at 17:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.