How to handle null columns in a Relational Database Design

Question

While mostly working with non-relational databases I need to switch gears and use a relational database as the application that I need to build will run complex queries and the join operation between tables is needed.

Before starting to create the database itself I've had to think about the architecture and I've set up an UML for Database Design:

This is how the TransactionDEpositBreakdown table may look:

id  amount  date        reference_number  batch_id   payment_processor_id  mid_id  main_dep_id
1   100     2020-10-11  900               null       1                     100     2
2   101     2020-10-11  900               null       1                     100     2
3   102     2020-10-11  900               null       1                     100     1
4   103     2020-10-11  350               null       1                     100     1
5   104     2020-10-11  350               null       1                     100     3
6   105     2020-10-11  600               null       1                     100     4
7   106     2020-10-11  null              1000       2                     201     null
8   107     2020-10-11  null              1001       2                     201     null
9   108     2020-10-11  null              1002       2                     201     null
10  109     2020-10-11  null              1003       2                     201     null

• A reference_number can be assigned to multiple transaction deposit breakdowns
• A batch_id is assigned to only one transaction deposit breakdown

There is a use case where a TransactionDepositBreakdown may have a reference number or a batch id , depending on the payment processor type (type 1 - reference number, type 2 - batch id). I'm not sure how to handle this case, but I'm thinking about the following options:

1. Add two tables TransactionDepositBatch and TransactionDepositReference which will have the transaction_deposit_id as a foreign key, batch_id on the first table and reference_number on the latter one:

2. Keep the reference_number and batch_id columns in the TransactionDepositBreakdown table and have at all times one of them null depending on the payment processor type.

Note: There might be a need of adding another column to the TransactionDepositBreakdown table, such as card_type , which will have a value assigned only when the payment processor type is 1.

Is the first option the correct way to handle this, by also taking into consideration the above note?

Also, any recommendations regarding the UML that I've built would be really useful.

Answer 1

These one-of relationships are difficult to model in relational databases. Different databases have different capabilities, so some may have extensions that can be applied to this problem (such as Postgres's support of table inheritance).

Your situation is rather simple, given just two options. Under those circumstances, I would go for the first option for one simple reason: it easily allows you to design the data model with declared foreign key relationships. The downside is that the you need space for both foreign keys, even if one of them is going to be NULL .

You can also enforce that one or the other is set, but not both using a check constraint:

constraint chk_TransactionDepositBreakdown_reference_or_batch 
    check (reference_number is null or batch_id is null);