How to work around lack of support for foreign keys across databases in Django

Question

I know Django does not support foreign keys across multiple databases (originally Django 1.3 docs)

But I\'m looking for a workaround.

What doesn\'t work

Answer 1

Ran into a similar problem of needing to reference (mostly) static data across multiple (5) databases. Made a slight update to the ReversedSingleRelatedObjectDescriptor to allow setting the related model. It doesn't implement the reverse relationship atm.

class ReverseSingleRelatedObjectDescriptor(object):
"""
This class provides the functionality that makes the related-object managers available as attributes on a model
class, for fields that have a single "remote" value, on the class that defines the related field. Used with
LinkedField.
"""
def __init__(self, field_with_rel):
    self.field = field_with_rel
    self.cache_name = self.field.get_cache_name()

def __get__(self, instance, instance_type=None):
    if instance is None:
        return self

    try:
        return getattr(instance, self.cache_name)
    except AttributeError:
        val = getattr(instance, self.field.attname)
        if val is None:
            # If NULL is an allowed value, return it
            if self.field.null:
                return None
            raise self.field.rel.to.DoesNotExist
        other_field = self.field.rel.get_related_field()
        if other_field.rel:
            params = {'%s__pk' % self.field.rel.field_name: val}
        else:
            params = {'%s__exact' % self.field.rel.field_name: val}

        # If the related manager indicates that it should be used for related fields, respect that.
        rel_mgr = self.field.rel.to._default_manager
        db = router.db_for_read(self.field.rel.to, instance=instance)
        if getattr(rel_mgr, 'forced_using', False):
            db = rel_mgr.forced_using
            rel_obj = rel_mgr.using(db).get(**params)
        elif getattr(rel_mgr, 'use_for_related_fields', False):
            rel_obj = rel_mgr.using(db).get(**params)
        else:
            rel_obj = QuerySet(self.field.rel.to).using(db).get(**params)
        setattr(instance, self.cache_name, rel_obj)
        return rel_obj

def __set__(self, instance, value):
    if instance is None:
        raise AttributeError("%s must be accessed via instance" % self.field.name)

    # If null=True, we can assign null here, but otherwise the value needs to be an instance of the related class.
    if value is None and self.field.null is False:
        raise ValueError('Cannot assign None: "%s.%s" does not allow null values.' %
                         (instance._meta.object_name, self.field.names))
    elif value is not None and not isinstance(value, self.field.rel.to):
        raise ValueError('Cannot assign "%r": "%s.%s" must be a "%s" instance.' %
                         (value, instance._meta.object_name, self.field.name, self.field.rel.to._meta.object_name))
    elif value is not None:
        # Only check the instance state db, LinkedField implies that the value is on a different database
        if instance._state.db is None:
            instance._state.db = router.db_for_write(instance.__class__, instance=value)

    # Is not used by OneToOneField, no extra measures to take here

    # Set the value of the related field
    try:
        val = getattr(value, self.field.rel.get_related_field().attname)
    except AttributeError:
        val = None
    setattr(instance, self.field.attname, val)

    # Since we already know what the related object is, seed the related object caches now, too. This avoids another
    # db hit if you get the object you just set
    setattr(instance, self.cache_name, value)
    if value is not None and not self.field.rel.multiple:
        setattr(value, self.field.related.get_cache_name(), instance)

and

class LinkedField(models.ForeignKey):
"""
Field class used to link models across databases. Does not ensure referrential integraty like ForeignKey
"""
def _description(self):
    return "Linked Field (type determined by related field)"

def contribute_to_class(self, cls, name):
    models.ForeignKey.contribute_to_class(self, cls, name)
    setattr(cls, self.name, ReverseSingleRelatedObjectDescriptor(self))
    if isinstance(self.rel.to, basestring):
        target = self.rel.to
    else:
        target = self.rel.to._meta.db_table
    cls._meta.duplicate_targets[self.column] = (target, "o2m")

def validate(self, value, model_instance):
    pass

Answer 2

As to the ForeignKeyAcrossDb part, couldn't you possibly make some adjustments to your class inside __init__? Check if the appropriate field is Integer if not, load it from the database, or do anything else that is required. Python __class__es can be changed at runtime without much problem.

Answer 3

Inspired by @Frans ' comment. My workaround is to do this in business layer. In the example given this question. I would set fruit to an IntegerField on Article, as "not to do integrity check in data layer".

class Fruit(models.Model):
    name = models.CharField()

class Article(models.Model):
    fruit = models.IntegerField()
    intro = models.TextField()

Then honor reference relation in application code (business layer). Take Django admin for example, in order to display fruit as a choice in Article's add page, you populate a list of choices for fruit manually.

# admin.py in App article
class ArticleAdmin(admin.ModelAdmin):
    class ArticleForm(forms.ModelForm):
        fields = ['fruit', 'intro']

        # populate choices for fruit
        choices = [(obj.id, obj.name) for obj in Fruit.objects.all()]
        widgets = {
            'fruit': forms.Select(choices=choices)}

    form = ArticleForm
    list_diaplay = ['fruit', 'intro']

Of course you may need to take care of form field validation (integrity check).