This document describes the details of the Model API. It builds on the\nmaterial presented in the model and database\nquery guides, so you’ll probably want to read and\nunderstand those documents before reading this one.
Throughout this reference we’ll use the example blog models presented in the database query guide.
\nTo create a new instance of a model, instantiate it like any other Python\nclass:
\nThe keyword arguments are the names of the fields you’ve defined on your model.\nNote that instantiating a model in no way touches your database; for that, you\nneed to save().
The from_db() method can be used to customize model instance creation\nwhen loading from the database.
The db argument contains the database alias for the database the model\nis loaded from, field_names contains the names of all loaded fields, and\nvalues contains the loaded values for each field in field_names. The\nfield_names are in the same order as the values. If all of the model’s\nfields are present, then values are guaranteed to be in the order\n__init__() expects them. That is, the instance can be created by\ncls(*values). If any fields are deferred, they won’t appear in\nfield_names. In that case, assign a value of django.db.models.DEFERRED\nto each of the missing fields.
In addition to creating the new model, the from_db() method must set the\nadding and db flags in the new instance’s _state attribute.
Below is an example showing how to record the initial values of fields that\nare loaded from the database:
\nfrom django.db.models import DEFERRED\n\n\n@classmethod\ndef from_db(cls, db, field_names, values):\n # Default implementation of from_db() (subject to change and could\n # be replaced with super()).\n if len(values) != len(cls._meta.concrete_fields):\n values = list(values)\n values.reverse()\n values = [\n values.pop() if f.attname in field_names else DEFERRED\n for f in cls._meta.concrete_fields\n ]\n instance = cls(*values)\n instance._state.adding = False\n instance._state.db = db\n # customization to store the original field values on the instance\n instance._loaded_values = dict(\n zip(field_names, (value for value in values if value is not DEFERRED))\n )\n return instance\n\n\ndef save(self, **kwargs):\n # Check how the current values differ from ._loaded_values. For example,\n # prevent changing the creator_id of the model. (This example doesn't\n # support cases where 'creator_id' is deferred).\n if not self._state.adding and (\n self.creator_id != self._loaded_values["creator_id"]\n ):\n raise ValueError("Updating the value of creator isn't allowed")\n super().save(**kwargs)\nThe example above shows a full from_db() implementation to clarify how that\nis done. In this case it would be possible to use a super() call in the\nfrom_db() method.
If you delete a field from a model instance, accessing it again reloads the\nvalue from the database:
\n>>> obj = MyModel.objects.first()\n>>> del obj.field\n>>> obj.field # Loads the field from the database\nAsynchronous version: arefresh_from_db()
If you need to reload a model’s values from the database, you can use the\nrefresh_from_db() method. When this method is called without arguments the\nfollowing is done:
All non-deferred fields of the model are updated to the values currently\npresent in the database.
Any cached relations are cleared from the reloaded instance.
Only fields of the model are reloaded from the database. Other\ndatabase-dependent values such as annotations aren’t reloaded. Any\n@cached_property attributes\naren’t cleared either.
The reloading happens from the database the instance was loaded from, or from\nthe default database if the instance wasn’t loaded from the database. The\nusing argument can be used to force the database used for reloading.
It is possible to force the set of fields to be loaded by using the fields\nargument.
For example, to test that an update() call resulted in the expected\nupdate, you could write a test similar to this:
def test_update_result(self):\n obj = MyModel.objects.create(val=1)\n MyModel.objects.filter(pk=obj.pk).update(val=F("val") + 1)\n # At this point obj.val is still 1, but the value in the database\n # was updated to 2. The object's updated value needs to be reloaded\n # from the database.\n obj.refresh_from_db()\n self.assertEqual(obj.val, 2)\nNote that when deferred fields are accessed, the loading of the deferred\nfield’s value happens through this method. Thus it is possible to customize\nthe way deferred loading happens. The example below shows how one can reload\nall of the instance’s fields when a deferred field is reloaded:
\nclass ExampleModel(models.Model):\n def refresh_from_db(self, using=None, fields=None, **kwargs):\n # fields contains the name of the deferred field to be\n # loaded.\n if fields is not None:\n fields = set(fields)\n deferred_fields = self.get_deferred_fields()\n # If any deferred field is going to be loaded\n if fields.intersection(deferred_fields):\n # then load all of them\n fields = fields.union(deferred_fields)\n super().refresh_from_db(using, fields, **kwargs)\nThe from_queryset argument allows using a different queryset than the one\ncreated from _base_manager. It gives you more\ncontrol over how the model is reloaded. For example, when your model uses soft\ndeletion you can make refresh_from_db() to take this into account:
obj.refresh_from_db(from_queryset=MyModel.active_objects.all())\nYou can cache related objects that otherwise would be cleared from the reloaded\ninstance:
\nobj.refresh_from_db(from_queryset=MyModel.objects.select_related("related_field"))\nYou can lock the row until the end of transaction before reloading a model’s\nvalues:
\nobj.refresh_from_db(from_queryset=MyModel.objects.select_for_update())\nA helper method that returns a set containing the attribute names of all those\nfields that are currently deferred for this model.
\nThere are four steps involved in validating a model:
\nValidate the model fields - Model.clean_fields()
Validate the model as a whole - Model.clean()
Validate the field uniqueness - Model.validate_unique()
Validate the constraints - Model.validate_constraints()
All four steps are performed when you call a model’s full_clean()\nmethod.
When you use a ModelForm, the call to\nis_valid() will perform these validation steps for\nall the fields that are included on the form. See the ModelForm\ndocumentation for more information. You should only\nneed to call a model’s full_clean() method if you plan to handle\nvalidation errors yourself, or if you have excluded fields from the\nModelForm that require validation.
This method calls Model.clean_fields(), Model.clean(),\nModel.validate_unique() (if validate_unique is True), and\nModel.validate_constraints() (if validate_constraints is True)\nin that order and raises a ValidationError that\nhas a message_dict attribute containing errors from all four stages.
The optional exclude argument can be used to provide a set of field\nnames that can be excluded from validation and cleaning.\nModelForm uses this argument to exclude fields that\naren’t present on your form from being validated since any errors raised could\nnot be corrected by the user.
Note that full_clean() will not be called automatically when you call\nyour model’s save() method. You’ll need to call it manually\nwhen you want to run one-step model validation for your own manually created\nmodels. For example:
from django.core.exceptions import ValidationError\n\ntry:\n article.full_clean()\nexcept ValidationError as e:\n # Do something based on the errors contained in e.message_dict.\n # Display them to a user, or handle them programmatically.\n pass\nThe first step full_clean() performs is to clean each individual field.
This method will validate all fields on your model. The optional exclude\nargument lets you provide a set of field names to exclude from validation.\nIt will raise a ValidationError if any fields\nfail validation.
The second step full_clean() performs is to call Model.clean().\nThis method should be overridden to perform custom validation on your model.
This method should be used to provide custom model validation, and to modify\nattributes on your model if desired. For instance, you could use it to\nautomatically provide a value for a field, or to do validation that requires\naccess to more than a single field:
\nimport datetime\nfrom django.core.exceptions import ValidationError\nfrom django.db import models\nfrom django.utils.translation import gettext_lazy as _\n\n\nclass Article(models.Model):\n ...\n\n def clean(self):\n # Don't allow draft entries to have a pub_date.\n if self.status == "draft" and self.pub_date is not None:\n raise ValidationError(_("Draft entries may not have a publication date."))\n # Set the pub_date for published items if it hasn't been set already.\n if self.status == "published" and self.pub_date is None:\n self.pub_date = datetime.date.today()\nNote, however, that like Model.full_clean(), a model’s clean()\nmethod is not invoked when you call your model’s save() method.
In the above example, the ValidationError\nexception raised by Model.clean() was instantiated with a string, so it\nwill be stored in a special error dictionary key,\nNON_FIELD_ERRORS. This key is used for errors\nthat are tied to the entire model instead of to a specific field:
from django.core.exceptions import NON_FIELD_ERRORS, ValidationError\n\ntry:\n article.full_clean()\nexcept ValidationError as e:\n non_field_errors = e.message_dict[NON_FIELD_ERRORS]\nTo assign exceptions to a specific field, instantiate the\nValidationError with a dictionary, where the\nkeys are the field names. We could update the previous example to assign the\nerror to the pub_date field:
class Article(models.Model):\n ...\n\n def clean(self):\n # Don't allow draft entries to have a pub_date.\n if self.status == "draft" and self.pub_date is not None:\n raise ValidationError(\n {"pub_date": _("Draft entries may not have a publication date.")}\n )\n ...\nIf you detect errors in multiple fields during Model.clean(), you can also\npass a dictionary mapping field names to errors:
raise ValidationError(\n {\n "title": ValidationError(_("Missing title."), code="required"),\n "pub_date": ValidationError(_("Invalid date."), code="invalid"),\n }\n)\nThen, full_clean() will check unique constraints on your model.
This method is similar to clean_fields(), but validates\nuniqueness constraints defined via Field.unique,\nField.unique_for_date, Field.unique_for_month,\nField.unique_for_year, or Meta.unique_together on your model instead of individual\nfield values. The optional exclude argument allows you to provide a set\nof field names to exclude from validation. It will raise a\nValidationError if any fields fail validation.
UniqueConstraints defined in the\nMeta.constraints are validated\nby Model.validate_constraints().
Note that if you provide an exclude argument to validate_unique(), any\nunique_together constraint involving one of\nthe fields you provided will not be checked.
Finally, full_clean() will check any other constraints on your model.
This method validates all constraints defined in\nMeta.constraints. The\noptional exclude argument allows you to provide a set of field names to\nexclude from validation. It will raise a\nValidationError if any constraints fail\nvalidation.
To save an object back to the database, call save():
Asynchronous version: asave()
For details on using the force_insert and force_update arguments, see\nForcing an INSERT or UPDATE. Details about the update_fields argument\ncan be found in the Specifying which fields to save section.
If you want customized saving behavior, you can override this save()\nmethod. See Overriding predefined model methods for more details.
The model save process also has some subtleties; see the sections below.
\n\nIf a model has an AutoField — an auto-incrementing\nprimary key — then that auto-incremented value will be calculated and saved as\nan attribute on your object the first time you call save():
>>> b2 = Blog(name="Cheddar Talk", tagline="Thoughts on cheese.")\n>>> b2.id # Returns None, because b2 doesn't have an ID yet.\n>>> b2.save()\n>>> b2.id # Returns the ID of your new object.\nThere’s no way to tell what the value of an ID will be before you call\nsave(), because that value is calculated by your database, not by Django.
For convenience, each model has an AutoField named\nid by default unless you explicitly specify primary_key=True on a field\nin your model. See the documentation for AutoField\nfor more details.
pk propertyLink to this heading#Regardless of whether you define a primary key field yourself, or let Django\nsupply one for you, each model will have a property called pk. It behaves\nlike a normal attribute on the model, but is actually an alias for whichever\nattribute is the primary key field for the model. You can read and set this\nvalue, just as you would for any other attribute, and it will update the\ncorrect field in the model.
If a model has an AutoField but you want to define a\nnew object’s ID explicitly when saving, define it explicitly before saving,\nrather than relying on the auto-assignment of the ID:
>>> b3 = Blog(id=3, name="Cheddar Talk", tagline="Thoughts on cheese.")\n>>> b3.id # Returns 3.\n>>> b3.save()\n>>> b3.id # Returns 3.\nIf you assign auto-primary-key values manually, make sure not to use an\nalready-existing primary-key value! If you create a new object with an explicit\nprimary-key value that already exists in the database, Django will assume you’re\nchanging the existing record rather than creating a new one.
\nGiven the above 'Cheddar Talk' blog example, this example would override the\nprevious record in the database:
b4 = Blog(id=3, name="Not Cheddar", tagline="Anything but cheese.")\nb4.save() # Overrides the previous blog with ID=3!\nSee How Django knows to UPDATE vs. INSERT, below, for the reason this\nhappens.
\nExplicitly specifying auto-primary-key values is mostly useful for bulk-saving\nobjects, when you’re confident you won’t have primary-key collision.
\nIf you’re using PostgreSQL, the sequence associated with the primary key might\nneed to be updated; see Manually-specifying values of auto-incrementing primary keys.
\nWhen you save an object, Django performs the following steps:
\nEmit a pre-save signal. The pre_save\nsignal is sent, allowing any functions listening for that signal to do\nsomething.
Preprocess the data. Each field’s\npre_save() method is called to perform any\nautomated data modification that’s needed. For example, the date/time fields\noverride pre_save() to implement\nauto_now_add and\nauto_now.
Prepare the data for the database. Each field’s\nget_db_prep_save() method is asked to provide\nits current value in a data type that can be written to the database.
Most fields don’t require data preparation. Simple data types, such as\nintegers and strings, are “ready to write” as a Python object. However, more\ncomplex data types often require some modification.
\nFor example, DateField fields use a Python\ndatetime object to store data. Databases don’t store datetime\nobjects, so the field value must be converted into an ISO-compliant date\nstring for insertion into the database.
Insert the data into the database. The preprocessed, prepared data is\ncomposed into an SQL statement for insertion into the database.
Emit a post-save signal. The post_save\nsignal is sent, allowing any functions listening for that signal to do\nsomething.
You may have noticed Django database objects use the same save() method\nfor creating and changing objects. Django abstracts the need to use INSERT\nor UPDATE SQL statements. Specifically, when you call save() and the\nobject’s primary key attribute does not define a\ndefault or\ndb_default, Django follows this algorithm:
If the object’s primary key attribute is set to anything except None,\nDjango executes an UPDATE.
If the object’s primary key attribute is not set or if the UPDATE\ndidn’t update anything (e.g. if primary key is set to a value that doesn’t\nexist in the database), Django executes an INSERT.
If the object’s primary key attribute defines a\ndefault or\ndb_default then Django executes an UPDATE\nif it is an existing model instance and primary key is set to a value that\nexists in the database. Otherwise, Django executes an INSERT.
The one gotcha here is that you should be careful not to specify a primary-key\nvalue explicitly when saving new objects, if you cannot guarantee the\nprimary-key value is unused. For more on this nuance, see Explicitly specifying\nauto-primary-key values above and Forcing an INSERT or UPDATE below.
\nIn Django 1.5 and earlier, Django did a SELECT when the primary key\nattribute was set. If the SELECT found a row, then Django did an UPDATE,\notherwise it did an INSERT. The old algorithm results in one more query in\nthe UPDATE case. There are some rare cases where the database doesn’t\nreport that a row was updated even if the database contains a row for the\nobject’s primary key value. An example is the PostgreSQL ON UPDATE trigger\nwhich returns NULL. In such cases it is possible to revert to the old\nalgorithm by setting the select_on_save\noption to True.
In some rare circumstances, it’s necessary to be able to force the\nsave() method to perform an SQL INSERT and not fall back to\ndoing an UPDATE. Or vice-versa: update, if possible, but not insert a new\nrow. In these cases you can pass the force_insert=True or\nforce_update=True parameters to the save() method.\nPassing both parameters is an error: you cannot both insert and update at the\nsame time!
When using multi-table inheritance, it’s also\npossible to provide a tuple of parent classes to force_insert in order to\nforce INSERT statements for each base. For example:
Restaurant(pk=1, name="Bob's Cafe").save(force_insert=(Place,))\n\nRestaurant(pk=1, name="Bob's Cafe", rating=4).save(force_insert=(Place, Rating))\nYou can pass force_insert=(models.Model,) to force an INSERT statement\nfor all parents. By default, force_insert=True only forces the insertion of\na new row for the current model.
It should be very rare that you’ll need to use these parameters. Django will\nalmost always do the right thing and trying to override that will lead to\nerrors that are difficult to track down. This feature is for advanced use\nonly.
\nUsing update_fields will force an update similarly to force_update.
Sometimes you’ll need to perform a simple arithmetic task on a field, such\nas incrementing or decrementing the current value. One way of achieving this is\ndoing the arithmetic in Python like:
\n>>> product = Product.objects.get(name="Venezuelan Beaver Cheese")\n>>> product.number_sold += 1\n>>> product.save()\nIf the old number_sold value retrieved from the database was 10, then\nthe value of 11 will be written back to the database.
The process can be made robust, avoiding a race condition, as well as slightly faster by expressing\nthe update relative to the original field value, rather than as an explicit\nassignment of a new value. Django provides F expressions for performing this kind of relative update. Using\nF expressions, the previous example is expressed\nas:
>>> from django.db.models import F\n>>> product = Product.objects.get(name="Venezuelan Beaver Cheese")\n>>> product.number_sold = F("number_sold") + 1\n>>> product.save()\nFor more details, see the documentation on F expressions and their use in update queries.
If save() is passed a list of field names in keyword argument\nupdate_fields, only the fields named in that list will be updated.\nThis may be desirable if you want to update just one or a few fields on\nan object. There will be a slight performance benefit from preventing\nall of the model fields from being updated in the database. For example:
product.name = "Name changed again"\nproduct.save(update_fields=["name"])\nThe update_fields argument can be any iterable containing strings. An\nempty update_fields iterable will skip the save. A value of None will\nperform an update on all fields.
Specifying update_fields will force an update.
When saving a model fetched through deferred model loading\n(only() or\ndefer()) only the fields loaded\nfrom the DB will get updated. In effect there is an automatic\nupdate_fields in this case. If you assign or change any deferred field\nvalue, the field will be added to the updated fields.
Asynchronous version: adelete()
Issues an SQL DELETE for the object. This only deletes the object in the\ndatabase; the Python instance will still exist and will still have data in\nits fields, except for the primary key set to None. This method returns the\nnumber of objects deleted and a dictionary with the number of deletions per\nobject type.
For more details, including how to delete objects in bulk, see\nDeleting objects.
\nIf you want customized deletion behavior, you can override the delete()\nmethod. See Overriding predefined model methods for more details.
Sometimes with multi-table inheritance you may\nwant to delete only a child model’s data. Specifying keep_parents=True will\nkeep the parent model’s data.
When you pickle a model, its current state is pickled. When you unpickle\nit, it’ll contain the model instance at the moment it was pickled, rather than\nthe data that’s currently in the database.
A few object methods have special purposes.
\n__str__()Link to this heading#The __str__() method is called whenever you call str() on an object.\nDjango uses str(obj) in a number of places. Most notably, to display an\nobject in the Django admin site and as the value inserted into a template when\nit displays an object. Thus, you should always return a nice, human-readable\nrepresentation of the model from the __str__() method.
For example:
\nfrom django.db import models\n\n\nclass Person(models.Model):\n first_name = models.CharField(max_length=50)\n last_name = models.CharField(max_length=50)\n\n def __str__(self):\n return f"{self.first_name} {self.last_name}"\n__eq__()Link to this heading#The equality method is defined such that instances with the same primary\nkey value and the same concrete class are considered equal, except that\ninstances with a primary key value of None aren’t equal to anything except\nthemselves. For proxy models, concrete class is defined as the model’s first\nnon-proxy parent; for all other models it’s simply the model’s class.
For example:
\nfrom django.db import models\n\n\nclass MyModel(models.Model):\n id = models.AutoField(primary_key=True)\n\n\nclass MyProxyModel(MyModel):\n class Meta:\n proxy = True\n\n\nclass MultitableInherited(MyModel):\n pass\n\n\n# Primary keys compared\nMyModel(id=1) == MyModel(id=1)\nMyModel(id=1) != MyModel(id=2)\n# Primary keys are None\nMyModel(id=None) != MyModel(id=None)\n# Same instance\ninstance = MyModel(id=None)\ninstance == instance\n# Proxy model\nMyModel(id=1) == MyProxyModel(id=1)\n# Multi-table inheritance\nMyModel(id=1) != MultitableInherited(id=1)\n__hash__()Link to this heading#The __hash__() method is based on the instance’s primary key value. It\nis effectively hash(obj.pk). If the instance doesn’t have a primary key\nvalue then a TypeError will be raised (otherwise the __hash__()\nmethod would return different values before and after the instance is\nsaved, but changing the __hash__() value of an instance is\nforbidden in Python.
get_absolute_url()Link to this heading#Define a get_absolute_url() method to tell Django how to calculate the\ncanonical URL for an object. To callers, this method should appear to return a\nstring that can be used to refer to the object over HTTP.
For example:
\ndef get_absolute_url(self):\n return "/people/%i/" % self.id\nWhile this code is correct and simple, it may not be the most portable way to\nto write this kind of method. The reverse() function is\nusually the best approach.
For example:
\ndef get_absolute_url(self):\n from django.urls import reverse\n\n return reverse("people-detail", kwargs={"pk": self.pk})\nOne place Django uses get_absolute_url() is in the admin app. If an object\ndefines this method, the object-editing page will have a «View on site» link\nthat will jump you directly to the object’s public view, as given by\nget_absolute_url().
Similarly, a couple of other bits of Django, such as the syndication feed\nframework, use get_absolute_url() when it is\ndefined. If it makes sense for your model’s instances to each have a unique\nURL, you should define get_absolute_url().
It’s good practice to use get_absolute_url() in templates, instead of\nhard-coding your objects” URLs. For example, this template code is bad:
<!-- BAD template code. Avoid! -->\n<a href="/people/{{ object.id }}/">{{ object.name }}</a>\nThis template code is much better:
\n<a href="{{ object.get_absolute_url }}">{{ object.name }}</a>\nThe logic here is that if you change the URL structure of your objects, even\nfor something small like correcting a spelling error, you don’t want to have to\ntrack down every place that the URL might be created. Specify it once, in\nget_absolute_url() and have all your other code call that one place.
In addition to save(), delete(), a model object\nmight have some of the following methods:
For every field that has choices set, the\nobject will have a get_FOO_display() method, where FOO is the name of\nthe field. This method returns the «human-readable» value of the field.
For example:
\nfrom django.db import models\n\n\nclass Person(models.Model):\n SHIRT_SIZES = {\n "S": "Small",\n "M": "Medium",\n "L": "Large",\n }\n name = models.CharField(max_length=60)\n shirt_size = models.CharField(max_length=2, choices=SHIRT_SIZES)\n>>> p = Person(name="Fred Flintstone", shirt_size="L")\n>>> p.save()\n>>> p.shirt_size\n'L'\n>>> p.get_shirt_size_display()\n'Large'\nFor every DateField and\nDateTimeField that does not have null=True, the object will have get_next_by_FOO() and\nget_previous_by_FOO() methods, where FOO is the name of the field. This\nreturns the next and previous object with respect to the date field, raising\na DoesNotExist exception when appropriate.
Both of these methods will perform their queries using the default\nmanager for the model. If you need to emulate filtering used by a\ncustom manager, or want to perform one-off custom filtering, both\nmethods also accept optional keyword arguments, which should be in the\nformat described in Field lookups.
\nNote that in the case of identical date values, these methods will use the\nprimary key as a tie-breaker. This guarantees that no records are skipped or\nduplicated. That also means you cannot use those methods on unsaved objects.
\n\n_stateLink to this heading#The _state attribute refers to a ModelState object that tracks\nthe lifecycle of the model instance.
The ModelState object has two attributes: adding, a flag which is\nTrue if the model has not been saved to the database yet, and db,\na string referring to the database alias the instance was loaded from or\nsaved to.
Newly instantiated instances have adding=True and db=None,\nsince they are yet to be saved. Instances fetched from a QuerySet\nwill have adding=False and db set to the alias of the associated\ndatabase.