Using other management commands¶

Most other django-admin commands that interact with the database operate in the same way as migrate – they only ever operate on one database at a time, using --database to control the database used.

An exception to this rule is the makemigrations command. It validates the migration history in the databases to catch problems with the existing migration files (which could be caused by editing them) before creating new migrations. By default, it checks only the default database, but it consults the allow_migrate() method of routers if any are installed.

Database routers¶

A database Router is a class that provides up to four methods:

db_for_read(model, **hints)¶

Suggest the database that should be used for read operations for objects of type model.

If a database operation is able to provide any additional information that might assist in selecting a database, it will be provided in the hints dictionary. Details on valid hints are provided below.

Returns None if there is no suggestion.

db_for_write(model, **hints)¶

Suggest the database that should be used for writes of objects of type Model.

If a database operation is able to provide any additional information that might assist in selecting a database, it will be provided in the hints dictionary. Details on valid hints are provided below.

Returns None if there is no suggestion.

allow_relation(obj1, obj2, **hints)¶

Return True if a relation between obj1 and obj2 should be allowed, False if the relation should be prevented, or None if the router has no opinion. This is purely a validation operation, used by foreign key and many to many operations to determine if a relation should be allowed between two objects.

If no router has an opinion (i.e. all routers return None), only relations within the same database are allowed.

allow_migrate(db, app_label, model_name=None, **hints)¶

Determine if the migration operation is allowed to run on the database with alias db. Return True if the operation should run, False if it shouldn’t run, or None if the router has no opinion.

The app_label positional argument is the label of the application being migrated.

model_name is set by most migration operations to the value of model._meta.model_name (the lowercased version of the model __name__) of the model being migrated. Its value is None for the RunPython and RunSQL operations unless they provide it using hints.

hints are used by certain operations to communicate additional information to the router.

When model_name is set, hints normally contains the model class under the key 'model'. Note that it may be a historical model, and thus not have any custom attributes, methods, or managers. You should only rely on _meta.

This method can also be used to determine the availability of a model on a given database.

makemigrations always creates migrations for model changes, but if allow_migrate() returns False, any migration operations for the model_name will be silently skipped when running migrate on the db. Changing the behavior of allow_migrate() for models that already have migrations may result in broken foreign keys, extra tables, or missing tables. When makemigrations verifies the migration history, it skips databases where no app is allowed to migrate.

A router doesn’t have to provide all these methods – it may omit one or more of them. If one of the methods is omitted, Django will skip that router when performing the relevant check.

Using routers¶

Database routers are installed using the DATABASE_ROUTERS setting. This setting defines a list of class names, each specifying a router that should be used by the base router (django.db.router).

The base router is used by Django’s database operations to allocate database usage. Whenever a query needs to know which database to use, it calls the base router, providing a model and a hint (if available). The base router tries each router class in turn until one returns a database suggestion. If no routers return a suggestion, the base router tries the current instance._state.db of the hint instance. If no hint instance was provided, or instance._state.db is None, the base router will allocate the default database.

An example¶

So – what does this mean in practice? Let’s consider another sample configuration. This one will have several databases: one for the auth application, and all other apps using a primary/replica setup with two read replicas. Here are the settings specifying these databases:

DATABASES = {
    "default": {},
    "auth_db": {
        "NAME": "auth_db_name",
        "ENGINE": "django.db.backends.mysql",
        "USER": "mysql_user",
        "PASSWORD": "swordfish",
    },
    "primary": {
        "NAME": "primary_name",
        "ENGINE": "django.db.backends.mysql",
        "USER": "mysql_user",
        "PASSWORD": "spam",
    },
    "replica1": {
        "NAME": "replica1_name",
        "ENGINE": "django.db.backends.mysql",
        "USER": "mysql_user",
        "PASSWORD": "eggs",
    },
    "replica2": {
        "NAME": "replica2_name",
        "ENGINE": "django.db.backends.mysql",
        "USER": "mysql_user",
        "PASSWORD": "bacon",
    },
}

Now we’ll need to handle routing. First we want a router that knows to send queries for the auth and contenttypes apps to auth_db (auth models are linked to ContentType, so they must be stored in the same database):

class AuthRouter:
    """
    A router to control all database operations on models in the
    auth and contenttypes applications.
    """

    route_app_labels = {"auth", "contenttypes"}

    def db_for_read(self, model, **hints):
        """
        Attempts to read auth and contenttypes models go to auth_db.
        """
        if model._meta.app_label in self.route_app_labels:
            return "auth_db"
        return None

    def db_for_write(self, model, **hints):
        """
        Attempts to write auth and contenttypes models go to auth_db.
        """
        if model._meta.app_label in self.route_app_labels:
            return "auth_db"
        return None

    def allow_relation(self, obj1, obj2, **hints):
        """
        Allow relations if a model in the auth or contenttypes apps is
        involved.
        """
        if (
            obj1._meta.app_label in self.route_app_labels
            or obj2._meta.app_label in self.route_app_labels
        ):
            return True
        return None

    def allow_migrate(self, db, app_label, model_name=None, **hints):
        """
        Make sure the auth and contenttypes apps only appear in the
        'auth_db' database.
        """
        if app_label in self.route_app_labels:
            return db == "auth_db"
        return None

And we also want a router that sends all other apps to the primary/replica configuration, and randomly chooses a replica to read from:

import random


class PrimaryReplicaRouter:
    def db_for_read(self, model, **hints):
        """
        Reads go to a randomly-chosen replica.
        """
        return random.choice(["replica1", "replica2"])

    def db_for_write(self, model, **hints):
        """
        Writes always go to primary.
        """
        return "primary"

    def allow_relation(self, obj1, obj2, **hints):
        """
        Relations between objects are allowed if both objects are
        in the primary/replica pool.
        """
        db_set = {"primary", "replica1", "replica2"}
        if obj1._state.db in db_set and obj2._state.db in db_set:
            return True
        return None

    def allow_migrate(self, db, app_label, model_name=None, **hints):
        """
        All non-auth models end up in this pool.
        """
        return True

Finally, in the settings file, we add the following (substituting path.to. with the actual Python path to the module(s) where the routers are defined):

DATABASE_ROUTERS = ["path.to.AuthRouter", "path.to.PrimaryReplicaRouter"]

The order in which routers are processed is significant. Routers will be queried in the order they are listed in the DATABASE_ROUTERS setting. In this example, the AuthRouter is processed before the PrimaryReplicaRouter, and as a result, decisions concerning the models in auth are processed before any other decision is made. If the DATABASE_ROUTERS setting listed the two routers in the other order, PrimaryReplicaRouter.allow_migrate() would be processed first. The catch-all nature of the PrimaryReplicaRouter implementation would mean that all models would be available on all databases.

With this setup installed, and all databases migrated as per Synchronizing your databases, lets run some Django code:

>>> # This retrieval will be performed on the 'auth_db' database
>>> fred = User.objects.get(username="fred")
>>> fred.first_name = "Frederick"

>>> # This save will also be directed to 'auth_db'
>>> fred.save()

>>> # These retrieval will be randomly allocated to a replica database
>>> dna = Person.objects.get(name="Douglas Adams")

>>> # A new object has no database allocation when created
>>> mh = Book(title="Mostly Harmless")

>>> # This assignment will consult the router, and set mh onto
>>> # the same database as the author object
>>> mh.author = dna

>>> # This save will force the 'mh' instance onto the primary database...
>>> mh.save()

>>> # ... but if we re-retrieve the object, it will come back on a replica
>>> mh = Book.objects.get(title="Mostly Harmless")

This example defined a router to handle interaction with models from the auth app, and other routers to handle interaction with all other apps. If you left your default database empty and don’t want to define a catch-all database router to handle all apps not otherwise specified, your routers must handle the names of all apps in INSTALLED_APPS before you migrate. See Behavior of contrib apps for information about contrib apps that must be together in one database.

Manually selecting a database for a QuerySet¶

You can select the database for a QuerySet at any point in the QuerySet «chain.» Call using() on the QuerySet to get another QuerySet that uses the specified database.

using() takes a single argument: the alias of the database on which you want to run the query. For example:

>>> # This will run on the 'default' database.
>>> Author.objects.all()

>>> # So will this.
>>> Author.objects.using("default")

>>> # This will run on the 'other' database.
>>> Author.objects.using("other")

Selecting a database for save()¶

Use the using keyword to Model.save() to specify to which database the data should be saved.

For example, to save an object to the legacy_users database, you’d use this:

>>> my_object.save(using="legacy_users")

If you don’t specify using, the save() method will save into the default database allocated by the routers.

>>> p = Person(name="Fred")
>>> p.save(using="first")  # (statement 1)
>>> p.save(using="second")  # (statement 2)

>>> p = Person(name="Fred")
>>> p.save(using="first")
>>> p.pk = None  # Clear the primary key.
>>> p.save(using="second")  # Write a completely new object.

>>> p = Person(name="Fred")
>>> p.save(using="first")
>>> p.save(using="second", force_insert=True)

Selecting a database to delete from¶

By default, a call to delete an existing object will be executed on the same database that was used to retrieve the object in the first place:

>>> u = User.objects.using("legacy_users").get(username="fred")
>>> u.delete()  # will delete from the `legacy_users` database

To specify the database from which a model will be deleted, pass a using keyword argument to the Model.delete() method. This argument works just like the using keyword argument to save().

For example, if you’re migrating a user from the legacy_users database to the new_users database, you might use these commands:

>>> user_obj.save(using="new_users")
>>> user_obj.delete(using="legacy_users")

Using managers with multiple databases¶

Use the db_manager() method on managers to give managers access to a non-default database.

For example, say you have a custom manager method that touches the database – User.objects.create_user(). Because create_user() is a manager method, not a QuerySet method, you can’t do User.objects.using('new_users').create_user(). (The create_user() method is only available on User.objects, the manager, not on QuerySet objects derived from the manager.) The solution is to use db_manager(), like this:

User.objects.db_manager("new_users").create_user(...)

db_manager() returns a copy of the manager bound to the database you specify.

class MyManager(models.Manager):
    def get_queryset(self):
        qs = CustomQuerySet(self.model)
        if self._db is not None:
            qs = qs.using(self._db)
        return qs

Cross-database relations¶

Django doesn’t currently provide any support for foreign key or many-to-many relationships spanning multiple databases. If you have used a router to partition models to different databases, any foreign key and many-to-many relationships defined by those models must be internal to a single database.

This is because of referential integrity. In order to maintain a relationship between two objects, Django needs to know that the primary key of the related object is valid. If the primary key is stored on a separate database, it’s not possible to easily evaluate the validity of a primary key.

If you’re using Postgres, Oracle, or MySQL with InnoDB, this is enforced at the database integrity level – database level key constraints prevent the creation of relations that can’t be validated.

However, if you’re using SQLite or MySQL with MyISAM tables, there is no enforced referential integrity; as a result, you may be able to “fake” cross database foreign keys. However, this configuration is not officially supported by Django.

Behavior of contrib apps¶

Several contrib apps include models, and some apps depend on others. Since cross-database relationships are impossible, this creates some restrictions on how you can split these models across databases:

• each one of contenttypes.ContentType, sessions.Session and sites.Site can be stored in any database, given a suitable router.
• auth models — User, Group and Permission — are linked together and linked to ContentType, so they must be stored in the same database as ContentType.
• admin depends on auth, so its models must be in the same database as auth.
• flatpages and redirects depend on sites, so their models must be in the same database as sites.

In addition, some objects are automatically created just after migrate creates a table to hold them in a database:

• a default Site,
• a ContentType for each model (including those not stored in that database),
• the Permissions for each model (including those not stored in that database).

For common setups with multiple databases, it isn’t useful to have these objects in more than one database. Common setups include primary/replica and connecting to external databases. Therefore, it’s recommended to write a database router that allows synchronizing these three models to only one database. Use the same approach for contrib and third-party apps that don’t need their tables in multiple databases.