Finding reverse foreign key references in Django ORM

A forward ForeignKey only requires changing the field value itself, but a reverse reference requires first changing the records of every model that points to that object. As an example, we set up a ProductCategory-Product-Purchase-Sell structure, and split it into two cases: fixing the category of a mis-categorized product, and merging a duplicate product into one.

When fixing a forward reference, such as a typo in a Product's category, it can be handled simply with something like Product.objects.filter(...).update(category=...). On the other hand, before deleting product_2, all rows in Purchase and Sell that reference product_2 must first be moved to product_1, and a method that only checks references within the same app may miss models in other apps.

Here, the key is _meta.related_objects. A model class's _meta contains relation metadata, and by iterating through related_objects you can find the reverse relations that reference that model.

• related_object.related_model: the referencing model class
• related_object.remote_field.name: the foreign key field name
• related_object.get_accessor_name(): the reverse accessor name

In the example, Product._meta.related_objects shows relations such as Purchase, Sell, and ProductStatistics, each represented as a ManyToOneRel. By selecting only ManyToOneRel entries and processing them like model.objects.filter(**{field_name: product_2}).update(**{field_name: product_1}), you can update every ForeignKey pointing to Product without missing any.

Since _meta.related_objects can contain not only ManyToOneRel but also OneToOneRel and ManyToManyRel, the approach in this article restricts itself to ManyToOneRel only. Because the update method differs by relation type, you should apply logic appropriate to each relation rather than handling them all with the same pattern.

Finally, you can clean up the duplicate data with product_2.delete(). If needed, there is also a way to check reverse references by querying SQL metadata, and for PostgreSQL you can refer to Dataedo's documentation.