db.models.ManyToManyField.through_fields

ManyToManyField.through_fields

Only used when a custom intermediary model is specified. Django will normally determine which fields of the intermediary model to use in order to establish a many-to-many relationship automatically. However, consider the following models:

from django.db import models

class Person(models.Model):
    name = models.CharField(max_length=50)

class Group(models.Model):
    name = models.CharField(max_length=128)
    members = models.ManyToManyField(
        Person,
        through='Membership',
        through_fields=('group', 'person'),
    )

class Membership(models.Model):
    group = models.ForeignKey(Group, on_delete=models.CASCADE)
    person = models.ForeignKey(Person, on_delete=models.CASCADE)
    inviter = models.ForeignKey(
        Person,
        on_delete=models.CASCADE,
        related_name="membership_invites",
    )
    invite_reason = models.CharField(max_length=64)

Membership has two foreign keys to Person (person and inviter), which makes the relationship ambiguous and Django can’t know which one to use. In this case, you must explicitly specify which foreign keys Django should use using through_fields, as in the example above.

through_fields accepts a 2-tuple ('field1', 'field2'), where field1 is the name of the foreign key to the model the ManyToManyField is defined on (group in this case), and field2 the name of the foreign key to the target model (person in this case).

When you have more than one foreign key on an intermediary model to any (or even both) of the models participating in a many-to-many relationship, you must specify through_fields. This also applies to recursive relationships when an intermediary model is used and there are more than two foreign keys to the model, or you want to explicitly specify which two Django should use.

Recursive relationships using an intermediary model are always defined as non-symmetrical – that is, with symmetrical=False – therefore, there is the concept of a “source” and a “target”. In that case 'field1' will be treated as the “source” of the relationship and 'field2' as the “target”.