querysets.txt

==========================
``QuerySet`` API reference
==========================

.. currentmodule:: django.db.models.query

This document describes the details of the ``QuerySet`` API. It builds on the
material presented in the :doc:`model </topics/db/models>` and :doc:`database
query </topics/db/queries>` guides, so you'll probably want to read and
understand those documents before reading this one.

Throughout this reference we'll use the :ref:`example Weblog models
<queryset-model-example>` presented in the :doc:`database query guide
</topics/db/queries>`.

.. _when-querysets-are-evaluated:

When ``QuerySet``\s are evaluated
=================================

Internally, a ``QuerySet`` can be constructed, filtered, sliced, and generally
passed around without actually hitting the database. No database activity
actually occurs until you do something to evaluate the queryset.

You can evaluate a ``QuerySet`` in the following ways:

* **Iteration.** A ``QuerySet`` is iterable, and it executes its database
  query the first time you iterate over it. For example, this will print
  the headline of all entries in the database::

      for e in Entry.objects.all():
          print(e.headline)

  Note: Don't use this if all you want to do is determine if at least one
  result exists. It's more efficient to use :meth:`~QuerySet.exists`.

* **Slicing.** As explained in :ref:`limiting-querysets`, a ``QuerySet`` can
  be sliced, using Python's array-slicing syntax. Slicing an unevaluated
  ``QuerySet`` usually returns another unevaluated ``QuerySet``, but Django
  will execute the database query if you use the "step" parameter of slice
  syntax, and will return a list. Slicing a ``QuerySet`` that has been
  evaluated also returns a list.

  Also note that even though slicing an unevaluated ``QuerySet`` returns
  another unevaluated ``QuerySet``, modifying it further (e.g., adding
  more filters, or modifying ordering) is not allowed, since that does not
  translate well into SQL and it would not have a clear meaning either.

* **Pickling/Caching.** See the following section for details of what
  is involved when `pickling QuerySets`_. The important thing for the
  purposes of this section is that the results are read from the database.

* **repr().** A ``QuerySet`` is evaluated when you call ``repr()`` on it.
  This is for convenience in the Python interactive interpreter, so you can
  immediately see your results when using the API interactively.

* **len().** A ``QuerySet`` is evaluated when you call ``len()`` on it.
  This, as you might expect, returns the length of the result list.

  Note: If you only need to determine the number of records in the set (and
  don't need the actual objects), it's much more efficient to handle a count
  at the database level using SQL's ``SELECT COUNT(*)``. Django provides a
  :meth:`~QuerySet.count` method for precisely this reason.

* **list().** Force evaluation of a ``QuerySet`` by calling ``list()`` on
  it. For example::

      entry_list = list(Entry.objects.all())

* **bool().** Testing a ``QuerySet`` in a boolean context, such as using
  ``bool()``, ``or``, ``and`` or an ``if`` statement, will cause the query
  to be executed. If there is at least one result, the ``QuerySet`` is
  ``True``, otherwise ``False``. For example::

      if Entry.objects.filter(headline="Test"):
         print("There is at least one Entry with the headline Test")

  Note: If you only want to determine if at least one result exists (and don't
  need the actual objects), it's more efficient to use :meth:`~QuerySet.exists`.

.. _pickling QuerySets:

Pickling ``QuerySet``\s
-----------------------

If you :mod:`pickle` a ``QuerySet``, this will force all the results to be loaded
into memory prior to pickling. Pickling is usually used as a precursor to
caching and when the cached queryset is reloaded, you want the results to
already be present and ready for use (reading from the database can take some
time, defeating the purpose of caching). This means that when you unpickle a
``QuerySet``, it contains the results at the moment it was pickled, rather
than the results that are currently in the database.

If you only want to pickle the necessary information to recreate the
``QuerySet`` from the database at a later time, pickle the ``query`` attribute
of the ``QuerySet``. You can then recreate the original ``QuerySet`` (without
any results loaded) using some code like this::

    >>> import pickle
    >>> query = pickle.loads(s)     # Assuming 's' is the pickled string.
    >>> qs = MyModel.objects.all()
    >>> qs.query = query            # Restore the original 'query'.

The ``query`` attribute is an opaque object. It represents the internals of
the query construction and is not part of the public API. However, it is safe
(and fully supported) to pickle and unpickle the attribute's contents as
described here.

.. admonition:: Restrictions on ``QuerySet.values_list()``

    If you recreate :meth:`QuerySet.values_list` using the pickled ``query``
    attribute, it will be converted to :meth:`QuerySet.values`::

        >>> import pickle
        >>> qs = Blog.objects.values_list('id', 'name')
        >>> qs
        <QuerySet [(1, 'Beatles Blog')]>
        >>> reloaded_qs = Blog.objects.all()
        >>> reloaded_qs.query = pickle.loads(pickle.dumps(qs.query))
        >>> reloaded_qs
        <QuerySet [{'id': 1, 'name': 'Beatles Blog'}]>

.. admonition:: You can't share pickles between versions

    Pickles of ``QuerySets`` are only valid for the version of Django that
    was used to generate them. If you generate a pickle using Django
    version N, there is no guarantee that pickle will be readable with
    Django version N+1. Pickles should not be used as part of a long-term
    archival strategy.

    Since pickle compatibility errors can be difficult to diagnose, such as
    silently corrupted objects, a ``RuntimeWarning`` is raised when you try to
    unpickle a queryset in a Django version that is different than the one in
    which it was pickled.

.. _queryset-api:

``QuerySet`` API
================

Here's the formal declaration of a ``QuerySet``:

.. class:: QuerySet(model=None, query=None, using=None, hints=None)

    Usually when you'll interact with a ``QuerySet`` you'll use it by
    :ref:`chaining filters <chaining-filters>`. To make this work, most
    ``QuerySet`` methods return new querysets. These methods are covered in
    detail later in this section.

    The ``QuerySet`` class has two public attributes you can use for
    introspection:

    .. attribute:: ordered

        ``True`` if the ``QuerySet`` is ordered — i.e. has an
        :meth:`order_by()` clause or a default ordering on the model.
        ``False`` otherwise.

    .. attribute:: db

        The database that will be used if this query is executed now.

    .. note::

        The ``query`` parameter to :class:`QuerySet` exists so that specialized
        query subclasses can reconstruct internal query state. The value of the
        parameter is an opaque representation of that query state and is not
        part of a public API.

.. currentmodule:: django.db.models.query.QuerySet

Methods that return new ``QuerySet``\s
--------------------------------------

Django provides a range of ``QuerySet`` refinement methods that modify either
the types of results returned by the ``QuerySet`` or the way its SQL query is
executed.

``filter()``
~~~~~~~~~~~~

.. method:: filter(**kwargs)

Returns a new ``QuerySet`` containing objects that match the given lookup
parameters.

The lookup parameters (``**kwargs``) should be in the format described in
`Field lookups`_ below. Multiple parameters are joined via ``AND`` in the
underlying SQL statement.

If you need to execute more complex queries (for example, queries with ``OR`` statements),
you can use :class:`Q objects <django.db.models.Q>`.

``exclude()``
~~~~~~~~~~~~~

.. method:: exclude(**kwargs)

Returns a new ``QuerySet`` containing objects that do *not* match the given
lookup parameters.

The lookup parameters (``**kwargs``) should be in the format described in
`Field lookups`_ below. Multiple parameters are joined via ``AND`` in the
underlying SQL statement, and the whole thing is enclosed in a ``NOT()``.

This example excludes all entries whose ``pub_date`` is later than 2005-1-3
AND whose ``headline`` is "Hello"::

    Entry.objects.exclude(pub_date__gt=datetime.date(2005, 1, 3), headline='Hello')

In SQL terms, that evaluates to:

.. code-block:: sql

    SELECT ...
    WHERE NOT (pub_date > '2005-1-3' AND headline = 'Hello')

This example excludes all entries whose ``pub_date`` is later than 2005-1-3
OR whose headline is "Hello"::

    Entry.objects.exclude(pub_date__gt=datetime.date(2005, 1, 3)).exclude(headline='Hello')

In SQL terms, that evaluates to:

.. code-block:: sql

    SELECT ...
    WHERE NOT pub_date > '2005-1-3'
    AND NOT headline = 'Hello'

Note the second example is more restrictive.

If you need to execute more complex queries (for example, queries with ``OR`` statements),
you can use :class:`Q objects <django.db.models.Q>`.

``annotate()``
~~~~~~~~~~~~~~

.. method:: annotate(*args, **kwargs)

Annotates each object in the ``QuerySet`` with the provided list of :doc:`query
expressions </ref/models/expressions>`. An expression may be a simple value, a
reference to a field on the model (or any related models), or an aggregate
expression (averages, sums, etc.) that has been computed over the objects that
are related to the objects in the ``QuerySet``.

Each argument to ``annotate()`` is an annotation that will be added
to each object in the ``QuerySet`` that is returned.

The aggregation functions that are provided by Django are described
in `Aggregation Functions`_ below.

Annotations specified using keyword arguments will use the keyword as
the alias for the annotation. Anonymous arguments will have an alias
generated for them based upon the name of the aggregate function and
the model field that is being aggregated. Only aggregate expressions
that reference a single field can be anonymous arguments. Everything
else must be a keyword argument.

For example, if you were manipulating a list of blogs, you may want
to determine how many entries have been made in each blog::

    >>> from django.db.models import Count
    >>> q = Blog.objects.annotate(Count('entry'))
    # The name of the first blog
    >>> q[0].name
    'Blogasaurus'
    # The number of entries on the first blog
    >>> q[0].entry__count
    42

The ``Blog`` model doesn't define an ``entry__count`` attribute by itself,
but by using a keyword argument to specify the aggregate function, you can
control the name of the annotation::

    >>> q = Blog.objects.annotate(number_of_entries=Count('entry'))
    # The number of entries on the first blog, using the name provided
    >>> q[0].number_of_entries
    42

For an in-depth discussion of aggregation, see :doc:`the topic guide on
Aggregation </topics/db/aggregation>`.

``alias()``
~~~~~~~~~~~

.. method:: alias(*args, **kwargs)

.. versionadded:: 3.2

Same as :meth:`annotate`, but instead of annotating objects in the
``QuerySet``, saves the expression for later reuse with other ``QuerySet``
methods. This is useful when the result of the expression itself is not needed
but it is used for filtering, ordering, or as a part of a complex expression.
Not selecting the unused value removes redundant work from the database which
should result in better performance.

For example, if you want to find blogs with more than 5 entries, but are not
interested in the exact number of entries, you could do this::

    >>> from django.db.models import Count
    >>> blogs = Blog.objects.alias(entries=Count('entry')).filter(entries__gt=5)

``alias()`` can be used in conjunction with :meth:`annotate`, :meth:`exclude`,
:meth:`filter`, :meth:`order_by`, and :meth:`update`. To use aliased expression
with other methods (e.g. :meth:`aggregate`), you must promote it to an
annotation::

    Blog.objects.alias(entries=Count('entry')).annotate(
        entries=F('entries'),
    ).aggregate(Sum('entries'))

:meth:`filter` and :meth:`order_by` can take expressions directly, but
expression construction and usage often does not happen in the same place (for
example, ``QuerySet`` method creates expressions, for later use in views).
``alias()`` allows building complex expressions incrementally, possibly
spanning multiple methods and modules, refer to the expression parts by their
aliases and only use :meth:`annotate` for the final result.

``order_by()``
~~~~~~~~~~~~~~

.. method:: order_by(*fields)

By default, results returned by a ``QuerySet`` are ordered by the ordering
tuple given by the ``ordering`` option in the model's ``Meta``. You can
override this on a per-``QuerySet`` basis by using the ``order_by`` method.

Example::

    Entry.objects.filter(pub_date__year=2005).order_by('-pub_date', 'headline')

The result above will be ordered by ``pub_date`` descending, then by
``headline`` ascending. The negative sign in front of ``"-pub_date"`` indicates
*descending* order. Ascending order is implied. To order randomly, use ``"?"``,
like so::

    Entry.objects.order_by('?')

Note: ``order_by('?')`` queries may be expensive and slow, depending on the
database backend you're using.

To order by a field in a different model, use the same syntax as when you are
querying across model relations. That is, the name of the field, followed by a
double underscore (``__``), followed by the name of the field in the new model,
and so on for as many models as you want to join. For example::

    Entry.objects.order_by('blog__name', 'headline')

If you try to order by a field that is a relation to another model, Django will
use the default ordering on the related model, or order by the related model's
primary key if there is no :attr:`Meta.ordering
<django.db.models.Options.ordering>` specified. For example, since the ``Blog``
model has no default ordering specified::

    Entry.objects.order_by('blog')

...is identical to::

    Entry.objects.order_by('blog__id')

If ``Blog`` had ``ordering = ['name']``, then the first queryset would be
identical to::

    Entry.objects.order_by('blog__name')

You can also order by :doc:`query expressions </ref/models/expressions>` by
calling :meth:`~.Expression.asc` or :meth:`~.Expression.desc` on the
expression::

    Entry.objects.order_by(Coalesce('summary', 'headline').desc())

:meth:`~.Expression.asc` and :meth:`~.Expression.desc` have arguments
(``nulls_first`` and ``nulls_last``) that control how null values are sorted.

Be cautious when ordering by fields in related models if you are also using
:meth:`distinct()`. See the note in :meth:`distinct` for an explanation of how
related model ordering can change the expected results.

.. note::
    It is permissible to specify a multi-valued field to order the results by
    (for example, a :class:`~django.db.models.ManyToManyField` field, or the
    reverse relation of a :class:`~django.db.models.ForeignKey` field).

    Consider this case::

         class Event(Model):
            parent = models.ForeignKey(
                'self',
                on_delete=models.CASCADE,
                related_name='children',
            )
            date = models.DateField()

         Event.objects.order_by('children__date')

    Here, there could potentially be multiple ordering data for each ``Event``;
    each ``Event`` with multiple ``children`` will be returned multiple times
    into the new ``QuerySet`` that ``order_by()`` creates. In other words,
    using ``order_by()`` on the ``QuerySet`` could return more items than you
    were working on to begin with - which is probably neither expected nor
    useful.

    Thus, take care when using multi-valued field to order the results. **If**
    you can be sure that there will only be one ordering piece of data for each
    of the items you're ordering, this approach should not present problems. If
    not, make sure the results are what you expect.

There's no way to specify whether ordering should be case sensitive. With
respect to case-sensitivity, Django will order results however your database
backend normally orders them.

You can order by a field converted to lowercase with
:class:`~django.db.models.functions.Lower` which will achieve case-consistent
ordering::

    Entry.objects.order_by(Lower('headline').desc())

If you don't want any ordering to be applied to a query, not even the default
ordering, call :meth:`order_by()` with no parameters.

You can tell if a query is ordered or not by checking the
:attr:`.QuerySet.ordered` attribute, which will be ``True`` if the
``QuerySet`` has been ordered in any way.

Each ``order_by()`` call will clear any previous ordering. For example, this
query will be ordered by ``pub_date`` and not ``headline``::

    Entry.objects.order_by('headline').order_by('pub_date')

.. warning::

    Ordering is not a free operation. Each field you add to the ordering
    incurs a cost to your database. Each foreign key you add will
    implicitly include all of its default orderings as well.

    If a query doesn't have an ordering specified, results are returned from
    the database in an unspecified order. A particular ordering is guaranteed
    only when ordering by a set of fields that uniquely identify each object in
    the results. For example, if a ``name`` field isn't unique, ordering by it
    won't guarantee objects with the same name always appear in the same order.

``reverse()``
~~~~~~~~~~~~~

.. method:: reverse()

Use the ``reverse()`` method to reverse the order in which a queryset's
elements are returned. Calling ``reverse()`` a second time restores the
ordering back to the normal direction.

To retrieve the "last" five items in a queryset, you could do this::

    my_queryset.reverse()[:5]

Note that this is not quite the same as slicing from the end of a sequence in
Python. The above example will return the last item first, then the
penultimate item and so on. If we had a Python sequence and looked at
``seq[-5:]``, we would see the fifth-last item first. Django doesn't support
that mode of access (slicing from the end), because it's not possible to do it
efficiently in SQL.

Also, note that ``reverse()`` should generally only be called on a ``QuerySet``
which has a defined ordering (e.g., when querying against a model which defines
a default ordering, or when using :meth:`order_by()`). If no such ordering is
defined for a given ``QuerySet``, calling ``reverse()`` on it has no real
effect (the ordering was undefined prior to calling ``reverse()``, and will
remain undefined afterward).

``distinct()``
~~~~~~~~~~~~~~

.. method:: distinct(*fields)

Returns a new ``QuerySet`` that uses ``SELECT DISTINCT`` in its SQL query. This
eliminates duplicate rows from the query results.

By default, a ``QuerySet`` will not eliminate duplicate rows. In practice, this
is rarely a problem, because simple queries such as ``Blog.objects.all()``
don't introduce the possibility of duplicate result rows. However, if your
query spans multiple tables, it's possible to get duplicate results when a
``QuerySet`` is evaluated. That's when you'd use ``distinct()``.

.. note::
    Any fields used in an :meth:`order_by` call are included in the SQL
    ``SELECT`` columns. This can sometimes lead to unexpected results when used
    in conjunction with ``distinct()``. If you order by fields from a related
    model, those fields will be added to the selected columns and they may make
    otherwise duplicate rows appear to be distinct. Since the extra columns
    don't appear in the returned results (they are only there to support
    ordering), it sometimes looks like non-distinct results are being returned.

    Similarly, if you use a :meth:`values()` query to restrict the columns
    selected, the columns used in any :meth:`order_by()` (or default model
    ordering) will still be involved and may affect uniqueness of the results.

    The moral here is that if you are using ``distinct()`` be careful about
    ordering by related models. Similarly, when using ``distinct()`` and
    :meth:`values()` together, be careful when ordering by fields not in the
    :meth:`values()` call.

On PostgreSQL only, you can pass positional arguments (``*fields``) in order to
specify the names of fields to which the ``DISTINCT`` should apply. This
translates to a ``SELECT DISTINCT ON`` SQL query. Here's the difference. For a
normal ``distinct()`` call, the database compares *each* field in each row when
determining which rows are distinct. For a ``distinct()`` call with specified
field names, the database will only compare the specified field names.

.. note::
    When you specify field names, you *must* provide an ``order_by()`` in the
    ``QuerySet``, and the fields in ``order_by()`` must start with the fields in
    ``distinct()``, in the same order.

    For example, ``SELECT DISTINCT ON (a)`` gives you the first row for each
    value in column ``a``. If you don't specify an order, you'll get some
    arbitrary row.

Examples (those after the first will only work on PostgreSQL)::

    >>> Author.objects.distinct()
    [...]

    >>> Entry.objects.order_by('pub_date').distinct('pub_date')
    [...]

    >>> Entry.objects.order_by('blog').distinct('blog')
    [...]

    >>> Entry.objects.order_by('author', 'pub_date').distinct('author', 'pub_date')
    [...]

    >>> Entry.objects.order_by('blog__name', 'mod_date').distinct('blog__name', 'mod_date')
    [...]

    >>> Entry.objects.order_by('author', 'pub_date').distinct('author')
    [...]

.. note::
    Keep in mind that :meth:`order_by` uses any default related model ordering
    that has been defined. You might have to explicitly order by the relation
    ``_id`` or referenced field to make sure the ``DISTINCT ON`` expressions
    match those at the beginning of the ``ORDER BY`` clause. For example, if
    the ``Blog`` model defined an :attr:`~django.db.models.Options.ordering` by
    ``name``::

        Entry.objects.order_by('blog').distinct('blog')

    ...wouldn't work because the query would be ordered by ``blog__name`` thus
    mismatching the ``DISTINCT ON`` expression. You'd have to explicitly order
    by the relation ``_id`` field (``blog_id`` in this case) or the referenced
    one (``blog__pk``) to make sure both expressions match.

``values()``
~~~~~~~~~~~~

.. method:: values(*fields, **expressions)

Returns a ``QuerySet`` that returns dictionaries, rather than model instances,
when used as an iterable.

Each of those dictionaries represents an object, with the keys corresponding to
the attribute names of model objects.

This example compares the dictionaries of ``values()`` with the normal model
objects::

    # This list contains a Blog object.
    >>> Blog.objects.filter(name__startswith='Beatles')
    <QuerySet [<Blog: Beatles Blog>]>

    # This list contains a dictionary.
    >>> Blog.objects.filter(name__startswith='Beatles').values()
    <QuerySet [{'id': 1, 'name': 'Beatles Blog', 'tagline': 'All the latest Beatles news.'}]>

The ``values()`` method takes optional positional arguments, ``*fields``, which
specify field names to which the ``SELECT`` should be limited. If you specify
the fields, each dictionary will contain only the field keys/values for the
fields you specify. If you don't specify the fields, each dictionary will
contain a key and value for every field in the database table.

Example::

    >>> Blog.objects.values()
    <QuerySet [{'id': 1, 'name': 'Beatles Blog', 'tagline': 'All the latest Beatles news.'}]>
    >>> Blog.objects.values('id', 'name')
    <QuerySet [{'id': 1, 'name': 'Beatles Blog'}]>

The ``values()`` method also takes optional keyword arguments,
``**expressions``, which are passed through to :meth:`annotate`::

    >>> from django.db.models.functions import Lower
    >>> Blog.objects.values(lower_name=Lower('name'))
    <QuerySet [{'lower_name': 'beatles blog'}]>

You can use built-in and :doc:`custom lookups </howto/custom-lookups>` in
ordering. For example::

    >>> from django.db.models import CharField
    >>> from django.db.models.functions import Lower
    >>> CharField.register_lookup(Lower)
    >>> Blog.objects.values('name__lower')
    <QuerySet [{'name__lower': 'beatles blog'}]>

An aggregate within a ``values()`` clause is applied before other arguments
within the same ``values()`` clause. If you need to group by another value,
add it to an earlier ``values()`` clause instead. For example::

    >>> from django.db.models import Count
    >>> Blog.objects.values('entry__authors', entries=Count('entry'))
    <QuerySet [{'entry__authors': 1, 'entries': 20}, {'entry__authors': 1, 'entries': 13}]>
    >>> Blog.objects.values('entry__authors').annotate(entries=Count('entry'))
    <QuerySet [{'entry__authors': 1, 'entries': 33}]>

A few subtleties that are worth mentioning:

* If you have a field called ``foo`` that is a
  :class:`~django.db.models.ForeignKey`, the default ``values()`` call
  will return a dictionary key called ``foo_id``, since this is the name
  of the hidden model attribute that stores the actual value (the ``foo``
  attribute refers to the related model). When you are calling
  ``values()`` and passing in field names, you can pass in either ``foo``
  or ``foo_id`` and you will get back the same thing (the dictionary key
  will match the field name you passed in).

  For example::

    >>> Entry.objects.values()
    <QuerySet [{'blog_id': 1, 'headline': 'First Entry', ...}, ...]>

    >>> Entry.objects.values('blog')
    <QuerySet [{'blog': 1}, ...]>

    >>> Entry.objects.values('blog_id')
    <QuerySet [{'blog_id': 1}, ...]>

* When using ``values()`` together with :meth:`distinct()`, be aware that
  ordering can affect the results. See the note in :meth:`distinct` for
  details.

* If you use a ``values()`` clause after an :meth:`extra()` call,
  any fields defined by a ``select`` argument in the :meth:`extra()` must
  be explicitly included in the ``values()`` call. Any :meth:`extra()` call
  made after a ``values()`` call will have its extra selected fields
  ignored.

* Calling :meth:`only()` and :meth:`defer()` after ``values()`` doesn't make
  sense, so doing so will raise a ``TypeError``.

* Combining transforms and aggregates requires the use of two :meth:`annotate`
  calls, either explicitly or as keyword arguments to :meth:`values`. As above,
  if the transform has been registered on the relevant field type the first
  :meth:`annotate` can be omitted, thus the following examples are equivalent::

    >>> from django.db.models import CharField, Count
    >>> from django.db.models.functions import Lower
    >>> CharField.register_lookup(Lower)
    >>> Blog.objects.values('entry__authors__name__lower').annotate(entries=Count('entry'))
    <QuerySet [{'entry__authors__name__lower': 'test author', 'entries': 33}]>
    >>> Blog.objects.values(
    ...     entry__authors__name__lower=Lower('entry__authors__name')
    ... ).annotate(entries=Count('entry'))
    <QuerySet [{'entry__authors__name__lower': 'test author', 'entries': 33}]>
    >>> Blog.objects.annotate(
    ...     entry__authors__name__lower=Lower('entry__authors__name')
    ... ).values('entry__authors__name__lower').annotate(entries=Count('entry'))
    <QuerySet [{'entry__authors__name__lower': 'test author', 'entries': 33}]>

It is useful when you know you're only going to need values from a small number
of the available fields and you won't need the functionality of a model
instance object. It's more efficient to select only the fields you need to use.

Finally, note that you can call ``filter()``, ``order_by()``, etc. after the
``values()`` call, that means that these two calls are identical::

    Blog.objects.values().order_by('id')
    Blog.objects.order_by('id').values()

The people who made Django prefer to put all the SQL-affecting methods first,
followed (optionally) by any output-affecting methods (such as ``values()``),
but it doesn't really matter. This is your chance to really flaunt your
individualism.

You can also refer to fields on related models with reverse relations through
``OneToOneField``, ``ForeignKey`` and ``ManyToManyField`` attributes::

    >>> Blog.objects.values('name', 'entry__headline')
    <QuerySet [{'name': 'My blog', 'entry__headline': 'An entry'},
         {'name': 'My blog', 'entry__headline': 'Another entry'}, ...]>

.. warning::

   Because :class:`~django.db.models.ManyToManyField` attributes and reverse
   relations can have multiple related rows, including these can have a
   multiplier effect on the size of your result set. This will be especially
   pronounced if you include multiple such fields in your ``values()`` query,
   in which case all possible combinations will be returned.

``values_list()``
~~~~~~~~~~~~~~~~~

.. method:: values_list(*fields, flat=False, named=False)

This is similar to ``values()`` except that instead of returning dictionaries,
it returns tuples when iterated over. Each tuple contains the value from the
respective field or expression passed into the ``values_list()`` call — so the
first item is the first field, etc. For example::

    >>> Entry.objects.values_list('id', 'headline')
    <QuerySet [(1, 'First entry'), ...]>
    >>> from django.db.models.functions import Lower
    >>> Entry.objects.values_list('id', Lower('headline'))
    <QuerySet [(1, 'first entry'), ...]>

If you only pass in a single field, you can also pass in the ``flat``
parameter. If ``True``, this will mean the returned results are single values,
rather than one-tuples. An example should make the difference clearer::

    >>> Entry.objects.values_list('id').order_by('id')
    <QuerySet[(1,), (2,), (3,), ...]>

    >>> Entry.objects.values_list('id', flat=True).order_by('id')
    <QuerySet [1, 2, 3, ...]>

It is an error to pass in ``flat`` when there is more than one field.

You can pass ``named=True`` to get results as a
:func:`~python:collections.namedtuple`::

    >>> Entry.objects.values_list('id', 'headline', named=True)
    <QuerySet [Row(id=1, headline='First entry'), ...]>

Using a named tuple may make use of the results more readable, at the expense
of a small performance penalty for transforming the results into a named tuple.

If you don't pass any values to ``values_list()``, it will return all the
fields in the model, in the order they were declared.

A common need is to get a specific field value of a certain model instance. To
achieve that, use ``values_list()`` followed by a ``get()`` call::

    >>> Entry.objects.values_list('headline', flat=True).get(pk=1)
    'First entry'

``values()`` and ``values_list()`` are both intended as optimizations for a
specific use case: retrieving a subset of data without the overhead of creating
a model instance. This metaphor falls apart when dealing with many-to-many and
other multivalued relations (such as the one-to-many relation of a reverse
foreign key) because the "one row, one object" assumption doesn't hold.

For example, notice the behavior when querying across a
:class:`~django.db.models.ManyToManyField`::

    >>> Author.objects.values_list('name', 'entry__headline')
    <QuerySet [('Noam Chomsky', 'Impressions of Gaza'),
     ('George Orwell', 'Why Socialists Do Not Believe in Fun'),
     ('George Orwell', 'In Defence of English Cooking'),
     ('Don Quixote', None)]>

Authors with multiple entries appear multiple times and authors without any
entries have ``None`` for the entry headline.

Similarly, when querying a reverse foreign key, ``None`` appears for entries
not having any author::

    >>> Entry.objects.values_list('authors')
    <QuerySet [('Noam Chomsky',), ('George Orwell',), (None,)]>

``dates()``
~~~~~~~~~~~

.. method:: dates(field, kind, order='ASC')

Returns a ``QuerySet`` that evaluates to a list of :class:`datetime.date`
objects representing all available dates of a particular kind within the
contents of the ``QuerySet``.

``field`` should be the name of a ``DateField`` of your model.
``kind`` should be either ``"year"``, ``"month"``, ``"week"``, or ``"day"``.
Each :class:`datetime.date` object in the result list is "truncated" to the
given ``type``.

* ``"year"`` returns a list of all distinct year values for the field.
* ``"month"`` returns a list of all distinct year/month values for the
  field.
* ``"week"`` returns a list of all distinct year/week values for the field. All
  dates will be a Monday.
* ``"day"`` returns a list of all distinct year/month/day values for the
  field.

``order``, which defaults to ``'ASC'``, should be either ``'ASC'`` or
``'DESC'``. This specifies how to order the results.

Examples::

    >>> Entry.objects.dates('pub_date', 'year')
    [datetime.date(2005, 1, 1)]
    >>> Entry.objects.dates('pub_date', 'month')
    [datetime.date(2005, 2, 1), datetime.date(2005, 3, 1)]
    >>> Entry.objects.dates('pub_date', 'week')
    [datetime.date(2005, 2, 14), datetime.date(2005, 3, 14)]
    >>> Entry.objects.dates('pub_date', 'day')
    [datetime.date(2005, 2, 20), datetime.date(2005, 3, 20)]
    >>> Entry.objects.dates('pub_date', 'day', order='DESC')
    [datetime.date(2005, 3, 20), datetime.date(2005, 2, 20)]
    >>> Entry.objects.filter(headline__contains='Lennon').dates('pub_date', 'day')
    [datetime.date(2005, 3, 20)]

``datetimes()``
~~~~~~~~~~~~~~~

.. method:: datetimes(field_name, kind, order='ASC', tzinfo=None, is_dst=None)

Returns a ``QuerySet`` that evaluates to a list of :class:`datetime.datetime`
objects representing all available dates of a particular kind within the
contents of the ``QuerySet``.

``field_name`` should be the name of a ``DateTimeField`` of your model.

``kind`` should be either ``"year"``, ``"month"``, ``"week"``, ``"day"``,
``"hour"``, ``"minute"``, or ``"second"``. Each :class:`datetime.datetime`
object in the result list is "truncated" to the given ``type``.

``order``, which defaults to ``'ASC'``, should be either ``'ASC'`` or
``'DESC'``. This specifies how to order the results.

``tzinfo`` defines the time zone to which datetimes are converted prior to
truncation. Indeed, a given datetime has different representations depending
on the time zone in use. This parameter must be a :class:`datetime.tzinfo`
object. If it's ``None``, Django uses the :ref:`current time zone
<default-current-time-zone>`. It has no effect when :setting:`USE_TZ` is
``False``.

``is_dst`` indicates whether or not ``pytz`` should interpret nonexistent and
ambiguous datetimes in daylight saving time. By default (when ``is_dst=None``),
``pytz`` raises an exception for such datetimes.

.. versionadded:: 3.1

    The ``is_dst`` parameter was added.

.. _database-time-zone-definitions:

.. note::

    This function performs time zone conversions directly in the database.
    As a consequence, your database must be able to interpret the value of
    ``tzinfo.tzname(None)``. This translates into the following requirements:

    - SQLite: no requirements. Conversions are performed in Python with pytz_
      (installed when you install Django).
    - PostgreSQL: no requirements (see `Time Zones`_).
    - Oracle: no requirements (see `Choosing a Time Zone File`_).
    - MySQL: load the time zone tables with `mysql_tzinfo_to_sql`_.

    .. _pytz: http://pytz.sourceforge.net/
    .. _Time Zones: https://www.postgresql.org/docs/current/datatype-datetime.html#DATATYPE-TIMEZONES
    .. _Choosing a Time Zone File: https://docs.oracle.com/en/database/oracle/
       oracle-database/18/nlspg/datetime-data-types-and-time-zone-support.html
       #GUID-805AB986-DE12-4FEA-AF56-5AABCD2132DF
    .. _mysql_tzinfo_to_sql: https://dev.mysql.com/doc/refman/en/mysql-tzinfo-to-sql.html

``none()``
~~~~~~~~~~

.. method:: none()

Calling none() will create a queryset that never returns any objects and no
query will be executed when accessing the results. A qs.none() queryset
is an instance of ``EmptyQuerySet``.

Examples::

    >>> Entry.objects.none()
    <QuerySet []>
    >>> from django.db.models.query import EmptyQuerySet
    >>> isinstance(Entry.objects.none(), EmptyQuerySet)
    True

``all()``
~~~~~~~~~

.. method:: all()

Returns a *copy* of the current ``QuerySet`` (or ``QuerySet`` subclass).  This
can be useful in situations where you might want to pass in either a model
manager or a ``QuerySet`` and do further filtering on the result. After calling
``all()`` on either object, you'll definitely have a ``QuerySet`` to work with.

When a ``QuerySet`` is :ref:`evaluated <when-querysets-are-evaluated>`, it
typically caches its results. If the data in the database might have changed
since a ``QuerySet`` was evaluated, you can get updated results for the same
query by calling ``all()`` on a previously evaluated ``QuerySet``.

``union()``
~~~~~~~~~~~

.. method:: union(*other_qs, all=False)

Uses SQL's ``UNION`` operator to combine the results of two or more
``QuerySet``\s. For example:

    >>> qs1.union(qs2, qs3)

The ``UNION`` operator selects only distinct values by default. To allow
duplicate values, use the ``all=True`` argument.

``union()``, ``intersection()``, and ``difference()`` return model instances
of the type of the first ``QuerySet`` even if the arguments are ``QuerySet``\s
of other models. Passing different models works as long as the ``SELECT`` list
is the same in all ``QuerySet``\s (at least the types, the names don't matter
as long as the types are in the same order). In such cases, you must use the
column names from the first ``QuerySet`` in ``QuerySet`` methods applied to the
resulting ``QuerySet``. For example::

    >>> qs1 = Author.objects.values_list('name')
    >>> qs2 = Entry.objects.values_list('headline')
    >>> qs1.union(qs2).order_by('name')

In addition, only ``LIMIT``, ``OFFSET``, ``COUNT(*)``, ``ORDER BY``, and
specifying columns (i.e. slicing, :meth:`count`, :meth:`order_by`, and
:meth:`values()`/:meth:`values_list()`) are allowed on the resulting
``QuerySet``. Further, databases place restrictions on what operations are
allowed in the combined queries. For example, most databases don't allow
``LIMIT`` or ``OFFSET`` in the combined queries.

``intersection()``
~~~~~~~~~~~~~~~~~~

.. method:: intersection(*other_qs)

Uses SQL's ``INTERSECT`` operator to return the shared elements of two or more
``QuerySet``\s. For example:

    >>> qs1.intersection(qs2, qs3)

See :meth:`union` for some restrictions.

``difference()``
~~~~~~~~~~~~~~~~

.. method:: difference(*other_qs)

Uses SQL's ``EXCEPT`` operator to keep only elements present in the
``QuerySet`` but not in some other ``QuerySet``\s. For example::

    >>> qs1.difference(qs2, qs3)

See :meth:`union` for some restrictions.

``select_related()``
~~~~~~~~~~~~~~~~~~~~

.. method:: select_related(*fields)

Returns a ``QuerySet`` that will "follow" foreign-key relationships, selecting
additional related-object data when it executes its query. This is a
performance booster which results in a single more complex query but means
later use of foreign-key relationships won't require database queries.

The following examples illustrate the difference between plain lookups and
``select_related()`` lookups. Here's standard lookup::

    # Hits the database.
    e = Entry.objects.get(id=5)

    # Hits the database again to get the related Blog object.
    b = e.blog

And here's ``select_related`` lookup::

    # Hits the database.
    e = Entry.objects.select_related('blog').get(id=5)

    # Doesn't hit the database, because e.blog has been prepopulated
    # in the previous query.
    b = e.blog

You can use ``select_related()`` with any queryset of objects::

    from django.utils import timezone

    # Find all the blogs with entries scheduled to be published in the future.
    blogs = set()

    for e in Entry.objects.filter(pub_date__gt=timezone.now()).select_related('blog'):
        # Without select_related(), this would make a database query for each
        # loop iteration in order to fetch the related blog for each entry.
        blogs.add(e.blog)

The order of ``filter()`` and ``select_related()`` chaining isn't important.
These querysets are equivalent::

    Entry.objects.filter(pub_date__gt=timezone.now()).select_related('blog')
    Entry.objects.select_related('blog').filter(pub_date__gt=timezone.now())

You can follow foreign keys in a similar way to querying them. If you have the
following models::

    from django.db import models

    class City(models.Model):
        # ...
        pass

    class Person(models.Model):
        # ...
        hometown = models.ForeignKey(
            City,
            on_delete=models.SET_NULL,
            blank=True,
            null=True,
        )

    class Book(models.Model):
        # ...
        author = models.ForeignKey(Person, on_delete=models.CASCADE)

... then a call to ``Book.objects.select_related('author__hometown').get(id=4)``
will cache the related ``Person`` *and* the related ``City``::

    # Hits the database with joins to the author and hometown tables.
    b = Book.objects.select_related('author__hometown').get(id=4)
    p = b.author         # Doesn't hit the database.
    c = p.hometown       # Doesn't hit the database.

    # Without select_related()...
    b = Book.objects.get(id=4)  # Hits the database.
    p = b.author         # Hits the database.
    c = p.hometown       # Hits the database.

You can refer to any :class:`~django.db.models.ForeignKey` or
:class:`~django.db.models.OneToOneField` relation in the list of fields
passed to ``select_related()``.

You can also refer to the reverse direction of a
:class:`~django.db.models.OneToOneField` in the list of fields passed to
``select_related`` — that is, you can traverse a
:class:`~django.db.models.OneToOneField` back to the object on which the field
is defined. Instead of specifying the field name, use the :attr:`related_name
<django.db.models.ForeignKey.related_name>` for the field on the related object.

There may be some situations where you wish to call ``select_related()`` with a
lot of related objects, or where you don't know all of the relations. In these
cases it is possible to call ``select_related()`` with no arguments. This will
follow all non-null foreign keys it can find - nullable foreign keys must be
specified. This is not recommended in most cases as it is likely to make the
underlying query more complex, and return more data, than is actually needed.

If you need to clear the list of related fields added by past calls of
``select_related`` on a ``QuerySet``, you can pass ``None`` as a parameter::

   >>> without_relations = queryset.select_related(None)

Chaining ``select_related`` calls works in a similar way to other methods -
that is that ``select_related('foo', 'bar')`` is equivalent to
``select_related('foo').select_related('bar')``.

``prefetch_related()``
~~~~~~~~~~~~~~~~~~~~~~

.. method:: prefetch_related(*lookups)

Returns a ``QuerySet`` that will automatically retrieve, in a single batch,
related objects for each of the specified lookups.

This has a similar purpose to ``select_related``, in that both are designed to
stop the deluge of database queries that is caused by accessing related objects,
but the strategy is quite different.

``select_related`` works by creating an SQL join and including the fields of the
related object in the ``SELECT`` statement. For this reason, ``select_related``
gets the related objects in the same database query. However, to avoid the much
larger result set that would result from joining across a 'many' relationship,
``select_related`` is limited to single-valued relationships - foreign key and
one-to-one.

``prefetch_related``, on the other hand, does a separate lookup for each
relationship, and does the 'joining' in Python. This allows it to prefetch
many-to-many and many-to-one objects, which cannot be done using
``select_related``, in addition to the foreign key and one-to-one relationships
that are supported by ``select_related``. It also supports prefetching of
:class:`~django.contrib.contenttypes.fields.GenericRelation` and
:class:`~django.contrib.contenttypes.fields.GenericForeignKey`, however, it
must be restricted to a homogeneous set of results. For example, prefetching
objects referenced by a ``GenericForeignKey`` is only supported if the query
is restricted to one ``ContentType``.

For example, suppose you have these models::

    from django.db import models

    class Topping(models.Model):
        name = models.CharField(max_length=30)

    class Pizza(models.Model):
        name = models.CharField(max_length=50)
        toppings = models.ManyToManyField(Topping)

        def __str__(self):
            return "%s (%s)" % (
                self.name,
                ", ".join(topping.name for topping in self.toppings.all()),
            )

and run::

    >>> Pizza.objects.all()
    ["Hawaiian (ham, pineapple)", "Seafood (prawns, smoked salmon)"...

The problem with this is that every time ``Pizza.__str__()`` asks for
``self.toppings.all()`` it has to query the database, so
``Pizza.objects.all()`` will run a query on the Toppings table for **every**
item in the Pizza ``QuerySet``.

We can reduce to just two queries using ``prefetch_related``:

    >>> Pizza.objects.all().prefetch_related('toppings')

This implies a ``self.toppings.all()`` for each ``Pizza``; now each time
``self.toppings.all()`` is called, instead of having to go to the database for
the items, it will find them in a prefetched ``QuerySet`` cache that was
populated in a single query.

That is, all the relevant toppings will have been fetched in a single query,
and used to make ``QuerySets`` that have a pre-filled cache of the relevant
results; these ``QuerySets`` are then used in the ``self.toppings.all()`` calls.

The additional queries in ``prefetch_related()`` are executed after the
``QuerySet`` has begun to be evaluated and the primary query has been executed.

If you have an iterable of model instances, you can prefetch related attributes
on those instances using the :func:`~django.db.models.prefetch_related_objects`
function.

Note that the result cache of the primary ``QuerySet`` and all specified related
objects will then be fully loaded into memory. This changes the typical
behavior of ``QuerySets``, which normally try to avoid loading all objects into
memory before they are needed, even after a query has been executed in the
database.

.. note::

    Remember that, as always with ``QuerySets``, any subsequent chained methods
    which imply a different database query will ignore previously cached
    results, and retrieve data using a fresh database query. So, if you write
    the following:

        >>> pizzas = Pizza.objects.prefetch_related('toppings')
        >>> [list(pizza.toppings.filter(spicy=True)) for pizza in pizzas]

    ...then the fact that ``pizza.toppings.all()`` has been prefetched will not
    help you. The ``prefetch_related('toppings')`` implied
    ``pizza.toppings.all()``, but ``pizza.toppings.filter()`` is a new and
    different query. The prefetched cache can't help here; in fact it hurts
    performance, since you have done a database query that you haven't used. So
    use this feature with caution!

    Also, if you call the database-altering methods
    :meth:`~django.db.models.fields.related.RelatedManager.add`,
    :meth:`~django.db.models.fields.related.RelatedManager.remove`,
    :meth:`~django.db.models.fields.related.RelatedManager.clear` or
    :meth:`~django.db.models.fields.related.RelatedManager.set`, on
    :class:`related managers<django.db.models.fields.related.RelatedManager>`,
    any prefetched cache for the relation will be cleared.

You can also use the normal join syntax to do related fields of related
fields. Suppose we have an additional model to the example above::

    class Restaurant(models.Model):
        pizzas = models.ManyToManyField(Pizza, related_name='restaurants')
        best_pizza = models.ForeignKey(Pizza, related_name='championed_by', on_delete=models.CASCADE)

The following are all legal:

    >>> Restaurant.objects.prefetch_related('pizzas__toppings')

This will prefetch all pizzas belonging to restaurants, and all toppings
belonging to those pizzas. This will result in a total of 3 database queries -
one for the restaurants, one for the pizzas, and one for the toppings.

    >>> Restaurant.objects.prefetch_related('best_pizza__toppings')

This will fetch the best pizza and all the toppings for the best pizza for each
restaurant. This will be done in 3 database queries - one for the restaurants,
one for the 'best pizzas', and one for the toppings.

The ``best_pizza`` relationship could also be fetched using ``select_related``
to reduce the query count to 2::

    >>> Restaurant.objects.select_related('best_pizza').prefetch_related('best_pizza__toppings')

Since the prefetch is executed after the main query (which includes the joins
needed by ``select_related``), it is able to detect that the ``best_pizza``
objects have already been fetched, and it will skip fetching them again.

Chaining ``prefetch_related`` calls will accumulate the lookups that are
prefetched. To clear any ``prefetch_related`` behavior, pass ``None`` as a
parameter:

   >>> non_prefetched = qs.prefetch_related(None)

One difference to note when using ``prefetch_related`` is that objects created
by a query can be shared between the different objects that they are related to
i.e. a single Python model instance can appear at more than one point in the
tree of objects that are returned. This will normally happen with foreign key
relationships. Typically this behavior will not be a problem, and will in fact
save both memory and CPU time.

While ``prefetch_related`` supports prefetching ``GenericForeignKey``
relationships, the number of queries will depend on the data. Since a
``GenericForeignKey`` can reference data in multiple tables, one query per table
referenced is needed, rather than one query for all the items. There could be
additional queries on the ``ContentType`` table if the relevant rows have not
already been fetched.

``prefetch_related`` in most cases will be implemented using an SQL query that
uses the 'IN' operator. This means that for a large ``QuerySet`` a large 'IN' clause
could be generated, which, depending on the database, might have performance
problems of its own when it comes to parsing or executing the SQL query. Always
profile for your use case!

Note that if you use ``iterator()`` to run the query, ``prefetch_related()``
calls will be ignored since these two optimizations do not make sense together.

You can use the :class:`~django.db.models.Prefetch` object to further control
the prefetch operation.

In its simplest form ``Prefetch`` is equivalent to the traditional string based
lookups:

    >>> from django.db.models import Prefetch
    >>> Restaurant.objects.prefetch_related(Prefetch('pizzas__toppings'))

You can provide a custom queryset with the optional ``queryset`` argument.
This can be used to change the default ordering of the queryset:

    >>> Restaurant.objects.prefetch_related(
    ...     Prefetch('pizzas__toppings', queryset=Toppings.objects.order_by('name')))

Or to call :meth:`~django.db.models.query.QuerySet.select_related()` when
applicable to reduce the number of queries even further:

    >>> Pizza.objects.prefetch_related(
    ...     Prefetch('restaurants', queryset=Restaurant.objects.select_related('best_pizza')))

You can also assign the prefetched result to a custom attribute with the optional
``to_attr`` argument. The result will be stored directly in a list.

This allows prefetching the same relation multiple times with a different
``QuerySet``; for instance:

    >>> vegetarian_pizzas = Pizza.objects.filter(vegetarian=True)
    >>> Restaurant.objects.prefetch_related(
    ...     Prefetch('pizzas', to_attr='menu'),
    ...     Prefetch('pizzas', queryset=vegetarian_pizzas, to_attr='vegetarian_menu'))

Lookups created with custom ``to_attr`` can still be traversed as usual by other
lookups:

    >>> vegetarian_pizzas = Pizza.objects.filter(vegetarian=True)
    >>> Restaurant.objects.prefetch_related(
    ...     Prefetch('pizzas', queryset=vegetarian_pizzas, to_attr='vegetarian_menu'),
    ...     'vegetarian_menu__toppings')

Using ``to_attr`` is recommended when filtering down the prefetch result as it is
less ambiguous than storing a filtered result in the related manager's cache:

    >>> queryset = Pizza.objects.filter(vegetarian=True)
    >>>
    >>> # Recommended:
    >>> restaurants = Restaurant.objects.prefetch_related(
    ...     Prefetch('pizzas', queryset=queryset, to_attr='vegetarian_pizzas'))
    >>> vegetarian_pizzas = restaurants[0].vegetarian_pizzas
    >>>
    >>> # Not recommended:
    >>> restaurants = Restaurant.objects.prefetch_related(
    ...     Prefetch('pizzas', queryset=queryset))
    >>> vegetarian_pizzas = restaurants[0].pizzas.all()

Custom prefetching also works with single related relations like
forward ``ForeignKey`` or ``OneToOneField``. Generally you'll want to use
:meth:`select_related()` for these relations, but there are a number of cases
where prefetching with a custom ``QuerySet`` is useful:

* You want to use a ``QuerySet`` that performs further prefetching
  on related models.

* You want to prefetch only a subset of the related objects.

* You want to use performance optimization techniques like
  :meth:`deferred fields <defer()>`:

    >>> queryset = Pizza.objects.only('name')
    >>>
    >>> restaurants = Restaurant.objects.prefetch_related(
    ...     Prefetch('best_pizza', queryset=queryset))

.. note::

    The ordering of lookups matters.

    Take the following examples:

       >>> prefetch_related('pizzas__toppings', 'pizzas')

    This works even though it's unordered because ``'pizzas__toppings'``
    already contains all the needed information, therefore the second argument
    ``'pizzas'`` is actually redundant.

        >>> prefetch_related('pizzas__toppings', Prefetch('pizzas', queryset=Pizza.objects.all()))

    This will raise a ``ValueError`` because of the attempt to redefine the
    queryset of a previously seen lookup. Note that an implicit queryset was
    created to traverse ``'pizzas'`` as part of the ``'pizzas__toppings'``
    lookup.

        >>> prefetch_related('pizza_list__toppings', Prefetch('pizzas', to_attr='pizza_list'))

    This will trigger an ``AttributeError`` because ``'pizza_list'`` doesn't exist yet
    when ``'pizza_list__toppings'`` is being processed.

    This consideration is not limited to the use of ``Prefetch`` objects. Some
    advanced techniques may require that the lookups be performed in a
    specific order to avoid creating extra queries; therefore it's recommended
    to always carefully order ``prefetch_related`` arguments.

``extra()``
~~~~~~~~~~~

.. method:: extra(select=None, where=None, params=None, tables=None, order_by=None, select_params=None)

Sometimes, the Django query syntax by itself can't easily express a complex
``WHERE`` clause. For these edge cases, Django provides the ``extra()``
``QuerySet`` modifier — a hook for injecting specific clauses into the SQL
generated by a ``QuerySet``.

.. admonition:: Use this method as a last resort

    This is an old API that we aim to deprecate at some point in the future.
    Use it only if you cannot express your query using other queryset methods.
    If you do need to use it, please `file a ticket
    <https://code.djangoproject.com/newticket>`_ using the `QuerySet.extra
    keyword <https://code.djangoproject.com/query?status=assigned&status=new&keywords=~QuerySet.extra>`_
    with your use case (please check the list of existing tickets first) so
    that we can enhance the QuerySet API to allow removing ``extra()``. We are
    no longer improving or fixing bugs for this method.

    For example, this use of ``extra()``::

        >>> qs.extra(
        ...     select={'val': "select col from sometable where othercol = %s"},
        ...     select_params=(someparam,),
        ... )

    is equivalent to::

        >>> qs.annotate(val=RawSQL("select col from sometable where othercol = %s", (someparam,)))

    The main benefit of using :class:`~django.db.models.expressions.RawSQL` is
    that you can set ``output_field`` if needed. The main downside is that if
    you refer to some table alias of the queryset in the raw SQL, then it is
    possible that Django might change that alias (for example, when the
    queryset is used as a subquery in yet another query).

.. warning::

    You should be very careful whenever you use ``extra()``. Every time you use
    it, you should escape any parameters that the user can control by using
    ``params`` in order to protect against SQL injection attacks.

    You also must not quote placeholders in the SQL string. This example is
    vulnerable to SQL injection because of the quotes around ``%s``:

    .. code-block:: sql

        SELECT col FROM sometable WHERE othercol = '%s'  # unsafe!

    You can read more about how Django's :ref:`SQL injection protection
    <sql-injection-protection>` works.

By definition, these extra lookups may not be portable to different database
engines (because you're explicitly writing SQL code) and violate the DRY
principle, so you should avoid them if possible.

Specify one or more of ``params``, ``select``, ``where`` or ``tables``. None
of the arguments is required, but you should use at least one of them.

* ``select``

  The ``select`` argument lets you put extra fields in the ``SELECT``
  clause.  It should be a dictionary mapping attribute names to SQL
  clauses to use to calculate that attribute.

  Example::

      Entry.objects.extra(select={'is_recent': "pub_date > '2006-01-01'"})

  As a result, each ``Entry`` object will have an extra attribute,
  ``is_recent``, a boolean representing whether the entry's ``pub_date``
  is greater than Jan. 1, 2006.

  Django inserts the given SQL snippet directly into the ``SELECT``
  statement, so the resulting SQL of the above example would be something like:

  .. code-block:: sql

      SELECT blog_entry.*, (pub_date > '2006-01-01') AS is_recent
      FROM blog_entry;


  The next example is more advanced; it does a subquery to give each
  resulting ``Blog`` object an ``entry_count`` attribute, an integer count
  of associated ``Entry`` objects::

      Blog.objects.extra(
          select={
              'entry_count': 'SELECT COUNT(*) FROM blog_entry WHERE blog_entry.blog_id = blog_blog.id'
          },
      )

  In this particular case, we're exploiting the fact that the query will
  already contain the ``blog_blog`` table in its ``FROM`` clause.

  The resulting SQL of the above example would be:

  .. code-block:: sql

      SELECT blog_blog.*, (SELECT COUNT(*) FROM blog_entry WHERE blog_entry.blog_id = blog_blog.id) AS entry_count
      FROM blog_blog;

  Note that the parentheses required by most database engines around
  subqueries are not required in Django's ``select`` clauses. Also note
  that some database backends, such as some MySQL versions, don't support
  subqueries.

  In some rare cases, you might wish to pass parameters to the SQL
  fragments in ``extra(select=...)``. For this purpose, use the
  ``select_params`` parameter.

  This will work, for example::

      Blog.objects.extra(
          select={'a': '%s', 'b': '%s'},
          select_params=('one', 'two'),
      )

  If you need to use a literal ``%s`` inside your select string, use
  the sequence ``%%s``.

* ``where`` / ``tables``

  You can define explicit SQL ``WHERE`` clauses — perhaps to perform
  non-explicit joins — by using ``where``. You can manually add tables to
  the SQL ``FROM`` clause by using ``tables``.

  ``where`` and ``tables`` both take a list of strings. All ``where``
  parameters are "AND"ed to any other search criteria.

  Example::

      Entry.objects.extra(where=["foo='a' OR bar = 'a'", "baz = 'a'"])

  ...translates (roughly) into the following SQL:

  .. code-block:: sql

      SELECT * FROM blog_entry WHERE (foo='a' OR bar='a') AND (baz='a')

  Be careful when using the ``tables`` parameter if you're specifying
  tables that are already used in the query. When you add extra tables
  via the ``tables`` parameter, Django assumes you want that table
  included an extra time, if it is already included. That creates a
  problem, since the table name will then be given an alias. If a table
  appears multiple times in an SQL statement, the second and subsequent
  occurrences must use aliases so the database can tell them apart. If
  you're referring to the extra table you added in the extra ``where``
  parameter this is going to cause errors.

  Normally you'll only be adding extra tables that don't already appear
  in the query. However, if the case outlined above does occur, there are
  a few solutions. First, see if you can get by without including the
  extra table and use the one already in the query. If that isn't
  possible, put your ``extra()`` call at the front of the queryset
  construction so that your table is the first use of that table.
  Finally, if all else fails, look at the query produced and rewrite your
  ``where`` addition to use the alias given to your extra table. The
  alias will be the same each time you construct the queryset in the same
  way, so you can rely upon the alias name to not change.

* ``order_by``

  If you need to order the resulting queryset using some of the new
  fields or tables you have included via ``extra()`` use the ``order_by``
  parameter to ``extra()`` and pass in a sequence of strings. These
  strings should either be model fields (as in the normal
  :meth:`order_by()` method on querysets), of the form
  ``table_name.column_name`` or an alias for a column that you specified
  in the ``select`` parameter to ``extra()``.

  For example::

      q = Entry.objects.extra(select={'is_recent': "pub_date > '2006-01-01'"})
      q = q.extra(order_by = ['-is_recent'])

  This would sort all the items for which ``is_recent`` is true to the
  front of the result set (``True`` sorts before ``False`` in a
  descending ordering).

  This shows, by the way, that you can make multiple calls to ``extra()``
  and it will behave as you expect (adding new constraints each time).

* ``params``

  The ``where`` parameter described above may use standard Python
  database string placeholders — ``'%s'`` to indicate parameters the
  database engine should automatically quote. The ``params`` argument is
  a list of any extra parameters to be substituted.

  Example::

      Entry.objects.extra(where=['headline=%s'], params=['Lennon'])

  Always use ``params`` instead of embedding values directly into
  ``where`` because ``params`` will ensure values are quoted correctly
  according to your particular backend. For example, quotes will be
  escaped correctly.

  Bad::

      Entry.objects.extra(where=["headline='Lennon'"])

  Good::

      Entry.objects.extra(where=['headline=%s'], params=['Lennon'])

.. warning::

    If you are performing queries on MySQL, note that MySQL's silent type coercion
    may cause unexpected results when mixing types. If you query on a string
    type column, but with an integer value, MySQL will coerce the types of all values
    in the table to an integer before performing the comparison. For example, if your
    table contains the values ``'abc'``, ``'def'`` and you query for ``WHERE mycolumn=0``,
    both rows will match. To prevent this, perform the correct typecasting
    before using the value in a query.

``defer()``
~~~~~~~~~~~

.. method:: defer(*fields)

In some complex data-modeling situations, your models might contain a lot of
fields, some of which could contain a lot of data (for example, text fields),
or require expensive processing to convert them to Python objects. If you are
using the results of a queryset in some situation where you don't know
if you need those particular fields when you initially fetch the data, you can
tell Django not to retrieve them from the database.

This is done by passing the names of the fields to not load to ``defer()``::

    Entry.objects.defer("headline", "body")

A queryset that has deferred fields will still return model instances. Each
deferred field will be retrieved from the database if you access that field
(one at a time, not all the deferred fields at once).

You can make multiple calls to ``defer()``. Each call adds new fields to the
deferred set::

    # Defers both the body and headline fields.
    Entry.objects.defer("body").filter(rating=5).defer("headline")

The order in which fields are added to the deferred set does not matter.
Calling ``defer()`` with a field name that has already been deferred is
harmless (the field will still be deferred).

You can defer loading of fields in related models (if the related models are
loading via :meth:`select_related()`) by using the standard double-underscore
notation to separate related fields::

    Blog.objects.select_related().defer("entry__headline", "entry__body")

If you want to clear the set of deferred fields, pass ``None`` as a parameter
to ``defer()``::

    # Load all fields immediately.
    my_queryset.defer(None)

Some fields in a model won't be deferred, even if you ask for them. You can
never defer the loading of the primary key. If you are using
:meth:`select_related()` to retrieve related models, you shouldn't defer the
loading of the field that connects from the primary model to the related
one, doing so will result in an error.

.. note::

    The ``defer()`` method (and its cousin, :meth:`only()`, below) are only for
    advanced use-cases. They provide an optimization for when you have analyzed
    your queries closely and understand *exactly* what information you need and
    have measured that the difference between returning the fields you need and
    the full set of fields for the model will be significant.

    Even if you think you are in the advanced use-case situation, **only use
    defer() when you cannot, at queryset load time, determine if you will need
    the extra fields or not**. If you are frequently loading and using a
    particular subset of your data, the best choice you can make is to
    normalize your models and put the non-loaded data into a separate model
    (and database table). If the columns *must* stay in the one table for some
    reason, create a model with ``Meta.managed = False`` (see the
    :attr:`managed attribute <django.db.models.Options.managed>` documentation)
    containing just the fields you normally need to load and use that where you
    might otherwise call ``defer()``. This makes your code more explicit to the
    reader, is slightly faster and consumes a little less memory in the Python
    process.

    For example, both of these models use the same underlying database table::

        class CommonlyUsedModel(models.Model):
            f1 = models.CharField(max_length=10)

            class Meta:
                managed = False
                db_table = 'app_largetable'

        class ManagedModel(models.Model):
            f1 = models.CharField(max_length=10)
            f2 = models.CharField(max_length=10)

            class Meta:
                db_table = 'app_largetable'

        # Two equivalent QuerySets:
        CommonlyUsedModel.objects.all()
        ManagedModel.objects.all().defer('f2')

    If many fields need to be duplicated in the unmanaged model, it may be best
    to create an abstract model with the shared fields and then have the
    unmanaged and managed models inherit from the abstract model.

.. note::

    When calling :meth:`~django.db.models.Model.save()` for instances with
    deferred fields, only the loaded fields will be saved. See
    :meth:`~django.db.models.Model.save()` for more details.

``only()``
~~~~~~~~~~

.. method:: only(*fields)

The ``only()`` method is more or less the opposite of :meth:`defer()`. You call
it with the fields that should *not* be deferred when retrieving a model.  If
you have a model where almost all the fields need to be deferred, using
``only()`` to specify the complementary set of fields can result in simpler
code.

Suppose you have a model with fields ``name``, ``age`` and ``biography``. The
following two querysets are the same, in terms of deferred fields::

    Person.objects.defer("age", "biography")
    Person.objects.only("name")

Whenever you call ``only()`` it *replaces* the set of fields to load
immediately. The method's name is mnemonic: **only** those fields are loaded
immediately; the remainder are deferred. Thus, successive calls to ``only()``
result in only the final fields being considered::

    # This will defer all fields except the headline.
    Entry.objects.only("body", "rating").only("headline")

Since ``defer()`` acts incrementally (adding fields to the deferred list), you
can combine calls to ``only()`` and ``defer()`` and things will behave
logically::

    # Final result is that everything except "headline" is deferred.
    Entry.objects.only("headline", "body").defer("body")

    # Final result loads headline and body immediately (only() replaces any
    # existing set of fields).
    Entry.objects.defer("body").only("headline", "body")

All of the cautions in the note for the :meth:`defer` documentation apply to
``only()`` as well. Use it cautiously and only after exhausting your other
options.

Using :meth:`only` and omitting a field requested using :meth:`select_related`
is an error as well.

.. note::

    When calling :meth:`~django.db.models.Model.save()` for instances with
    deferred fields, only the loaded fields will be saved. See
    :meth:`~django.db.models.Model.save()` for more details.

``using()``
~~~~~~~~~~~

.. method:: using(alias)

This method is for controlling which database the ``QuerySet`` will be
evaluated against if you are using more than one database.  The only argument
this method takes is the alias of a database, as defined in
:setting:`DATABASES`.

For example::

    # queries the database with the 'default' alias.
    >>> Entry.objects.all()

    # queries the database with the 'backup' alias
    >>> Entry.objects.using('backup')

``select_for_update()``
~~~~~~~~~~~~~~~~~~~~~~~

.. method:: select_for_update(nowait=False, skip_locked=False, of=(), no_key=False)

Returns a queryset that will lock rows until the end of the transaction,
generating a ``SELECT ... FOR UPDATE`` SQL statement on supported databases.

For example::

    from django.db import transaction

    entries = Entry.objects.select_for_update().filter(author=request.user)
    with transaction.atomic():
        for entry in entries:
            ...

When the queryset is evaluated (``for entry in entries`` in this case), all
matched entries will be locked until the end of the transaction block, meaning
that other transactions will be prevented from changing or acquiring locks on
them.

Usually, if another transaction has already acquired a lock on one of the
selected rows, the query will block until the lock is released. If this is
not the behavior you want, call ``select_for_update(nowait=True)``. This will
make the call non-blocking. If a conflicting lock is already acquired by
another transaction, :exc:`~django.db.DatabaseError` will be raised when the
queryset is evaluated. You can also ignore locked rows by using
``select_for_update(skip_locked=True)`` instead. The ``nowait`` and
``skip_locked`` are mutually exclusive and attempts to call
``select_for_update()`` with both options enabled will result in a
:exc:`ValueError`.

By default, ``select_for_update()`` locks all rows that are selected by the
query. For example, rows of related objects specified in :meth:`select_related`
are locked in addition to rows of the queryset's model. If this isn't desired,
specify the related objects you want to lock in ``select_for_update(of=(...))``
using the same fields syntax as :meth:`select_related`. Use the value ``'self'``
to refer to the queryset's model.

.. admonition:: Lock parents models in ``select_for_update(of=(...))``

    If you want to lock parents models when using :ref:`multi-table inheritance
    <multi-table-inheritance>`, you must specify parent link fields (by default
    ``<parent_model_name>_ptr``) in the ``of`` argument. For example::

        Restaurant.objects.select_for_update(of=('self', 'place_ptr'))

On PostgreSQL only, you can pass ``no_key=True`` in order to acquire a weaker
lock, that still allows creating rows that merely reference locked rows
(through a foreign key, for example) while the lock is in place. The
PostgreSQL documentation has more details about `row-level lock modes
<https://www.postgresql.org/docs/current/explicit-locking.html#LOCKING-ROWS>`_.

You can't use ``select_for_update()`` on nullable relations::

    >>> Person.objects.select_related('hometown').select_for_update()
    Traceback (most recent call last):
    ...
    django.db.utils.NotSupportedError: FOR UPDATE cannot be applied to the nullable side of an outer join

To avoid that restriction, you can exclude null objects if you don't care about
them::

    >>> Person.objects.select_related('hometown').select_for_update().exclude(hometown=None)
    <QuerySet [<Person: ...)>, ...]>

Currently, the ``postgresql``, ``oracle``, and ``mysql`` database
backends support ``select_for_update()``. However, MariaDB 10.3+ supports only
the ``nowait`` argument and MySQL 8.0.1+ supports the ``nowait``,
``skip_locked``, and ``of`` arguments. The ``no_key`` argument is supported
only on PostgreSQL.

Passing ``nowait=True``, ``skip_locked=True``, ``no_key=True``, or ``of`` to
``select_for_update()`` using database backends that do not support these
options, such as MySQL, raises a :exc:`~django.db.NotSupportedError`. This
prevents code from unexpectedly blocking.

Evaluating a queryset with ``select_for_update()`` in autocommit mode on
backends which support ``SELECT ... FOR UPDATE`` is a
:exc:`~django.db.transaction.TransactionManagementError` error because the
rows are not locked in that case. If allowed, this would facilitate data
corruption and could easily be caused by calling code that expects to be run in
a transaction outside of one.

Using ``select_for_update()`` on backends which do not support
``SELECT ... FOR UPDATE`` (such as SQLite) will have no effect.
``SELECT ... FOR UPDATE`` will not be added to the query, and an error isn't
raised if ``select_for_update()`` is used in autocommit mode.

.. warning::

    Although ``select_for_update()`` normally fails in autocommit mode, since
    :class:`~django.test.TestCase` automatically wraps each test in a
    transaction, calling ``select_for_update()`` in a ``TestCase`` even outside
    an :func:`~django.db.transaction.atomic()` block will (perhaps unexpectedly)
    pass without raising a ``TransactionManagementError``. To properly test
    ``select_for_update()`` you should use
    :class:`~django.test.TransactionTestCase`.

.. admonition:: Certain expressions may not be supported

    PostgreSQL doesn't support ``select_for_update()`` with
    :class:`~django.db.models.expressions.Window` expressions.

.. versionchanged:: 3.2

    The ``no_key`` argument was added.

    The ``of`` argument was allowed on MySQL 8.0.1+.

``raw()``
~~~~~~~~~

.. method:: raw(raw_query, params=None, translations=None)

Takes a raw SQL query, executes it, and returns a
``django.db.models.query.RawQuerySet`` instance. This ``RawQuerySet`` instance
can be iterated over just like a normal ``QuerySet`` to provide object
instances.

See the :doc:`/topics/db/sql` for more information.

.. warning::

  ``raw()`` always triggers a new query and doesn't account for previous
  filtering. As such, it should generally be called from the ``Manager`` or
  from a fresh ``QuerySet`` instance.

Operators that return new ``QuerySet``\s
----------------------------------------

Combined querysets must use the same model.

AND (``&``)
~~~~~~~~~~~

Combines two ``QuerySet``\s using the SQL ``AND`` operator.

The following are equivalent::

    Model.objects.filter(x=1) & Model.objects.filter(y=2)
    Model.objects.filter(x=1, y=2)
    from django.db.models import Q
    Model.objects.filter(Q(x=1) & Q(y=2))

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE x=1 AND y=2

OR (``|``)
~~~~~~~~~~

Combines two ``QuerySet``\s using the SQL ``OR`` operator.

The following are equivalent::

    Model.objects.filter(x=1) | Model.objects.filter(y=2)
    from django.db.models import Q
    Model.objects.filter(Q(x=1) | Q(y=2))

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE x=1 OR y=2

Methods that do not return ``QuerySet``\s
-----------------------------------------

The following ``QuerySet`` methods evaluate the ``QuerySet`` and return
something *other than* a ``QuerySet``.

These methods do not use a cache (see :ref:`caching-and-querysets`). Rather,
they query the database each time they're called.

``get()``
~~~~~~~~~

.. method:: get(**kwargs)

Returns the object matching the given lookup parameters, which should be in
the format described in `Field lookups`_. You should use lookups that are
guaranteed unique, such as the primary key or fields in a unique constraint.
For example::

    Entry.objects.get(id=1)
    Entry.objects.get(blog=blog, entry_number=1)

If you expect a queryset to already return one row, you can use ``get()``
without any arguments to return the object for that row::

    Entry.objects.filter(pk=1).get()

If ``get()`` doesn't find any object, it raises a :exc:`Model.DoesNotExist
<django.db.models.Model.DoesNotExist>` exception::

    Entry.objects.get(id=-999) # raises Entry.DoesNotExist

If ``get()`` finds more than one object, it raises a
:exc:`Model.MultipleObjectsReturned
<django.db.models.Model.MultipleObjectsReturned>` exception::

    Entry.objects.get(name='A Duplicated Name') # raises Entry.MultipleObjectsReturned

Both these exception classes are attributes of the model class, and specific to
that model. If you want to handle such exceptions from several ``get()`` calls
for different models, you can use their generic base classes. For example, you
can use :exc:`django.core.exceptions.ObjectDoesNotExist`  to handle
:exc:`~django.db.models.Model.DoesNotExist` exceptions from multiple models::

    from django.core.exceptions import ObjectDoesNotExist

    try:
        blog = Blog.objects.get(id=1)
        entry = Entry.objects.get(blog=blog, entry_number=1)
    except ObjectDoesNotExist:
        print("Either the blog or entry doesn't exist.")

``create()``
~~~~~~~~~~~~

.. method:: create(**kwargs)

A convenience method for creating an object and saving it all in one step.  Thus::

    p = Person.objects.create(first_name="Bruce", last_name="Springsteen")

and::

    p = Person(first_name="Bruce", last_name="Springsteen")
    p.save(force_insert=True)

are equivalent.

The :ref:`force_insert <ref-models-force-insert>` parameter is documented
elsewhere, but all it means is that a new object will always be created.
Normally you won't need to worry about this. However, if your model contains a
manual primary key value that you set and if that value already exists in the
database, a call to ``create()`` will fail with an
:exc:`~django.db.IntegrityError` since primary keys must be unique. Be
prepared to handle the exception if you are using manual primary keys.

``get_or_create()``
~~~~~~~~~~~~~~~~~~~

.. method:: get_or_create(defaults=None, **kwargs)

A convenience method for looking up an object with the given ``kwargs`` (may be
empty if your model has defaults for all fields), creating one if necessary.

Returns a tuple of ``(object, created)``, where ``object`` is the retrieved or
created object and ``created`` is a boolean specifying whether a new object was
created.

This is meant to prevent duplicate objects from being created when requests are
made in parallel, and as a shortcut to boilerplatish code. For example::

    try:
        obj = Person.objects.get(first_name='John', last_name='Lennon')
    except Person.DoesNotExist:
        obj = Person(first_name='John', last_name='Lennon', birthday=date(1940, 10, 9))
        obj.save()

Here, with concurrent requests, multiple attempts to save a ``Person`` with
the same parameters may be made. To avoid this race condition, the above
example can be rewritten using ``get_or_create()`` like so::

    obj, created = Person.objects.get_or_create(
        first_name='John',
        last_name='Lennon',
        defaults={'birthday': date(1940, 10, 9)},
    )

Any keyword arguments passed to ``get_or_create()`` — *except* an optional one
called ``defaults`` — will be used in a :meth:`get()` call. If an object is
found, ``get_or_create()`` returns a tuple of that object and ``False``.

.. warning::

    This method is atomic assuming that the database enforces uniqueness of the
    keyword arguments (see :attr:`~django.db.models.Field.unique` or
    :attr:`~django.db.models.Options.unique_together`). If the fields used in the
    keyword arguments do not have a uniqueness constraint, concurrent calls to
    this method may result in multiple rows with the same parameters being
    inserted.

You can specify more complex conditions for the retrieved object by chaining
``get_or_create()`` with ``filter()`` and using :class:`Q objects
<django.db.models.Q>`. For example, to retrieve Robert or Bob Marley if either
exists, and create the latter otherwise::

    from django.db.models import Q

    obj, created = Person.objects.filter(
        Q(first_name='Bob') | Q(first_name='Robert'),
    ).get_or_create(last_name='Marley', defaults={'first_name': 'Bob'})

If multiple objects are found, ``get_or_create()`` raises
:exc:`~django.core.exceptions.MultipleObjectsReturned`. If an object is *not*
found, ``get_or_create()`` will instantiate and save a new object, returning a
tuple of the new object and ``True``. The new object will be created roughly
according to this algorithm::

    params = {k: v for k, v in kwargs.items() if '__' not in k}
    params.update({k: v() if callable(v) else v for k, v in defaults.items()})
    obj = self.model(**params)
    obj.save()

In English, that means start with any non-``'defaults'`` keyword argument that
doesn't contain a double underscore (which would indicate a non-exact lookup).
Then add the contents of ``defaults``, overriding any keys if necessary, and
use the result as the keyword arguments to the model class. If there are any
callables in ``defaults``, evaluate them. As hinted at above, this is a
simplification of the algorithm that is used, but it contains all the pertinent
details. The internal implementation has some more error-checking than this and
handles some extra edge-conditions; if you're interested, read the code.

If you have a field named ``defaults`` and want to use it as an exact lookup in
``get_or_create()``, use ``'defaults__exact'``, like so::

    Foo.objects.get_or_create(defaults__exact='bar', defaults={'defaults': 'baz'})

The ``get_or_create()`` method has similar error behavior to :meth:`create()`
when you're using manually specified primary keys. If an object needs to be
created and the key already exists in the database, an
:exc:`~django.db.IntegrityError` will be raised.

Finally, a word on using ``get_or_create()`` in Django views. Please make sure
to use it only in ``POST`` requests unless you have a good reason not to.
``GET`` requests shouldn't have any effect on data. Instead, use ``POST``
whenever a request to a page has a side effect on your data. For more, see
:rfc:`Safe methods <7231#section-4.2.1>` in the HTTP spec.

.. warning::

  You can use ``get_or_create()`` through :class:`~django.db.models.ManyToManyField`
  attributes and reverse relations. In that case you will restrict the queries
  inside the context of that relation. That could lead you to some integrity
  problems if you don't use it consistently.

  Being the following models::

      class Chapter(models.Model):
          title = models.CharField(max_length=255, unique=True)

      class Book(models.Model):
          title = models.CharField(max_length=256)
          chapters = models.ManyToManyField(Chapter)

  You can use ``get_or_create()`` through Book's chapters field, but it only
  fetches inside the context of that book::

      >>> book = Book.objects.create(title="Ulysses")
      >>> book.chapters.get_or_create(title="Telemachus")
      (<Chapter: Telemachus>, True)
      >>> book.chapters.get_or_create(title="Telemachus")
      (<Chapter: Telemachus>, False)
      >>> Chapter.objects.create(title="Chapter 1")
      <Chapter: Chapter 1>
      >>> book.chapters.get_or_create(title="Chapter 1")
      # Raises IntegrityError

  This is happening because it's trying to get or create "Chapter 1" through the
  book "Ulysses", but it can't do any of them: the relation can't fetch that
  chapter because it isn't related to that book, but it can't create it either
  because ``title`` field should be unique.

``update_or_create()``
~~~~~~~~~~~~~~~~~~~~~~

.. method:: update_or_create(defaults=None, **kwargs)

A convenience method for updating an object with the given ``kwargs``, creating
a new one if necessary. The ``defaults`` is a dictionary of (field, value)
pairs used to update the object. The values in ``defaults`` can be callables.

Returns a tuple of ``(object, created)``, where ``object`` is the created or
updated object and ``created`` is a boolean specifying whether a new object was
created.

The ``update_or_create`` method tries to fetch an object from database based on
the given ``kwargs``. If a match is found, it updates the fields passed in the
``defaults`` dictionary.

This is meant as a shortcut to boilerplatish code. For example::

    defaults = {'first_name': 'Bob'}
    try:
        obj = Person.objects.get(first_name='John', last_name='Lennon')
        for key, value in defaults.items():
            setattr(obj, key, value)
        obj.save()
    except Person.DoesNotExist:
        new_values = {'first_name': 'John', 'last_name': 'Lennon'}
        new_values.update(defaults)
        obj = Person(**new_values)
        obj.save()

This pattern gets quite unwieldy as the number of fields in a model goes up.
The above example can be rewritten using ``update_or_create()`` like so::

    obj, created = Person.objects.update_or_create(
        first_name='John', last_name='Lennon',
        defaults={'first_name': 'Bob'},
    )

For detailed description how names passed in ``kwargs`` are resolved see
:meth:`get_or_create`.

As described above in :meth:`get_or_create`, this method is prone to a
race-condition which can result in multiple rows being inserted simultaneously
if uniqueness is not enforced at the database level.

Like :meth:`get_or_create` and :meth:`create`, if you're using manually
specified primary keys and an object needs to be created but the key already
exists in the database, an :exc:`~django.db.IntegrityError` is raised.

``bulk_create()``
~~~~~~~~~~~~~~~~~

.. method:: bulk_create(objs, batch_size=None, ignore_conflicts=False)

This method inserts the provided list of objects into the database in an
efficient manner (generally only 1 query, no matter how many objects there
are)::

    >>> Entry.objects.bulk_create([
    ...     Entry(headline='This is a test'),
    ...     Entry(headline='This is only a test'),
    ... ])

This has a number of caveats though:

* The model's ``save()`` method will not be called, and the ``pre_save`` and
  ``post_save`` signals will not be sent.
* It does not work with child models in a multi-table inheritance scenario.
* If the model's primary key is an :class:`~django.db.models.AutoField`, the
  primary key attribute can only be retrieved on certain databases (currently
  PostgreSQL and MariaDB 10.5+). On other databases, it will not be set.
* It does not work with many-to-many relationships.
* It casts ``objs`` to a list, which fully evaluates ``objs`` if it's a
  generator. The cast allows inspecting all objects so that any objects with a
  manually set primary key can be inserted first. If you want to insert objects
  in batches without evaluating the entire generator at once, you can use this
  technique as long as the objects don't have any manually set primary keys::

    from itertools import islice

    batch_size = 100
    objs = (Entry(headline='Test %s' % i) for i in range(1000))
    while True:
        batch = list(islice(objs, batch_size))
        if not batch:
            break
        Entry.objects.bulk_create(batch, batch_size)

The ``batch_size`` parameter controls how many objects are created in a single
query. The default is to create all objects in one batch, except for SQLite
where the default is such that at most 999 variables per query are used.

On databases that support it (all but Oracle), setting the ``ignore_conflicts``
parameter to ``True`` tells the database to ignore failure to insert any rows
that fail constraints such as duplicate unique values. Enabling this parameter
disables setting the primary key on each model instance (if the database
normally supports it).

.. warning::

    On MySQL and MariaDB, setting the ``ignore_conflicts`` parameter to
    ``True`` turns certain types of errors, other than duplicate key, into
    warnings. Even with Strict Mode. For example: invalid values or
    non-nullable violations. See the `MySQL documentation`_ and
    `MariaDB documentation`_ for more details.

.. _MySQL documentation: https://dev.mysql.com/doc/refman/en/sql-mode.html#ignore-strict-comparison
.. _MariaDB documentation: https://mariadb.com/kb/en/ignore/

Returns ``objs`` as cast to a list, in the same order as provided.

.. versionchanged:: 3.1

    Support for the fetching primary key attributes on MariaDB 10.5+ was added.

``bulk_update()``
~~~~~~~~~~~~~~~~~

.. method:: bulk_update(objs, fields, batch_size=None)

This method efficiently updates the given fields on the provided model
instances, generally with one query::

    >>> objs = [
    ...    Entry.objects.create(headline='Entry 1'),
    ...    Entry.objects.create(headline='Entry 2'),
    ... ]
    >>> objs[0].headline = 'This is entry 1'
    >>> objs[1].headline = 'This is entry 2'
    >>> Entry.objects.bulk_update(objs, ['headline'])

:meth:`.QuerySet.update` is used to save the changes, so this is more efficient
than iterating through the list of models and calling ``save()`` on each of
them, but it has a few caveats:

* You cannot update the model's primary key.
* Each model's ``save()`` method isn't called, and the
  :attr:`~django.db.models.signals.pre_save` and
  :attr:`~django.db.models.signals.post_save` signals aren't sent.
* If updating a large number of columns in a large number of rows, the SQL
  generated can be very large. Avoid this by specifying a suitable
  ``batch_size``.
* Updating fields defined on multi-table inheritance ancestors will incur an
  extra query per ancestor.
* If ``objs`` contains duplicates, only the first one is updated.

The ``batch_size`` parameter controls how many objects are saved in a single
query. The default is to update all objects in one batch, except for SQLite
and Oracle which have restrictions on the number of variables used in a query.

``count()``
~~~~~~~~~~~

.. method:: count()

Returns an integer representing the number of objects in the database matching
the ``QuerySet``.

Example::

    # Returns the total number of entries in the database.
    Entry.objects.count()

    # Returns the number of entries whose headline contains 'Lennon'
    Entry.objects.filter(headline__contains='Lennon').count()

A ``count()`` call performs a ``SELECT COUNT(*)`` behind the scenes, so you
should always use ``count()`` rather than loading all of the record into Python
objects and calling ``len()`` on the result (unless you need to load the
objects into memory anyway, in which case ``len()`` will be faster).

Note that if you want the number of items in a ``QuerySet`` and are also
retrieving model instances from it (for example, by iterating over it), it's
probably more efficient to use ``len(queryset)`` which won't cause an extra
database query like ``count()`` would.

``in_bulk()``
~~~~~~~~~~~~~

.. method:: in_bulk(id_list=None, field_name='pk')

Takes a list of field values (``id_list``) and the ``field_name`` for those
values, and returns a dictionary mapping each value to an instance of the
object with the given field value. If ``id_list`` isn't provided, all objects
in the queryset are returned. ``field_name`` must be a unique field or a
distinct field (if there's only one field specified in :meth:`distinct`).
``field_name`` defaults to the primary key.

Example::

    >>> Blog.objects.in_bulk([1])
    {1: <Blog: Beatles Blog>}
    >>> Blog.objects.in_bulk([1, 2])
    {1: <Blog: Beatles Blog>, 2: <Blog: Cheddar Talk>}
    >>> Blog.objects.in_bulk([])
    {}
    >>> Blog.objects.in_bulk()
    {1: <Blog: Beatles Blog>, 2: <Blog: Cheddar Talk>, 3: <Blog: Django Weblog>}
    >>> Blog.objects.in_bulk(['beatles_blog'], field_name='slug')
    {'beatles_blog': <Blog: Beatles Blog>}
    >>> Blog.objects.distinct('name').in_bulk(field_name='name')
    {'Beatles Blog': <Blog: Beatles Blog>, 'Cheddar Talk': <Blog: Cheddar Talk>, 'Django Weblog': <Blog: Django Weblog>}

If you pass ``in_bulk()`` an empty list, you'll get an empty dictionary.

.. versionchanged:: 3.2

    Using a distinct field was allowed.

``iterator()``
~~~~~~~~~~~~~~

.. method:: iterator(chunk_size=2000)

Evaluates the ``QuerySet`` (by performing the query) and returns an iterator
(see :pep:`234`) over the results. A ``QuerySet`` typically caches its results
internally so that repeated evaluations do not result in additional queries. In
contrast, ``iterator()`` will read results directly, without doing any caching
at the ``QuerySet`` level (internally, the default iterator calls ``iterator()``
and caches the return value). For a ``QuerySet`` which returns a large number of
objects that you only need to access once, this can result in better
performance and a significant reduction in memory.

Note that using ``iterator()`` on a ``QuerySet`` which has already been
evaluated will force it to evaluate again, repeating the query.

Also, use of ``iterator()`` causes previous ``prefetch_related()`` calls to be
ignored since these two optimizations do not make sense together.

Depending on the database backend, query results will either be loaded all at
once or streamed from the database using server-side cursors.

With server-side cursors
^^^^^^^^^^^^^^^^^^^^^^^^

Oracle and :ref:`PostgreSQL <postgresql-server-side-cursors>` use server-side
cursors to stream results from the database without loading the entire result
set into memory.

The Oracle database driver always uses server-side cursors.

With server-side cursors, the ``chunk_size`` parameter specifies the number of
results to cache at the database driver level. Fetching bigger chunks
diminishes the number of round trips between the database driver and the
database, at the expense of memory.

On PostgreSQL, server-side cursors will only be used when the
:setting:`DISABLE_SERVER_SIDE_CURSORS <DATABASE-DISABLE_SERVER_SIDE_CURSORS>`
setting is ``False``. Read :ref:`transaction-pooling-server-side-cursors` if
you're using a connection pooler configured in transaction pooling mode. When
server-side cursors are disabled, the behavior is the same as databases that
don't support server-side cursors.

Without server-side cursors
^^^^^^^^^^^^^^^^^^^^^^^^^^^

MySQL doesn't support streaming results, hence the Python database driver loads
the entire result set into memory. The result set is then transformed into
Python row objects by the database adapter using the ``fetchmany()`` method
defined in :pep:`249`.

SQLite can fetch results in batches using ``fetchmany()``, but since SQLite
doesn't provide isolation between queries within a connection, be careful when
writing to the table being iterated over. See :ref:`sqlite-isolation` for
more information.

The ``chunk_size`` parameter controls the size of batches Django retrieves from
the database driver. Larger batches decrease the overhead of communicating with
the database driver at the expense of a slight increase in memory consumption.

The default value of ``chunk_size``, 2000, comes from `a calculation on the
psycopg mailing list <https://www.postgresql.org/message-id/4D2F2C71.8080805%40dndg.it>`_:

    Assuming rows of 10-20 columns with a mix of textual and numeric data, 2000
    is going to fetch less than 100KB of data, which seems a good compromise
    between the number of rows transferred and the data discarded if the loop
    is exited early.

``latest()``
~~~~~~~~~~~~

.. method:: latest(*fields)

Returns the latest object in the table based on the given field(s).

This example returns the latest ``Entry`` in the table, according to the
``pub_date`` field::

    Entry.objects.latest('pub_date')

You can also choose the latest based on several fields. For example, to select
the ``Entry`` with the earliest ``expire_date`` when two entries have the same
``pub_date``::

    Entry.objects.latest('pub_date', '-expire_date')

The negative sign in ``'-expire_date'`` means to sort ``expire_date`` in
*descending* order. Since ``latest()`` gets the last result, the ``Entry`` with
the earliest ``expire_date`` is selected.

If your model's :ref:`Meta <meta-options>` specifies
:attr:`~django.db.models.Options.get_latest_by`, you can omit any arguments to
``earliest()`` or ``latest()``. The fields specified in
:attr:`~django.db.models.Options.get_latest_by` will be used by default.

Like :meth:`get()`, ``earliest()`` and ``latest()`` raise
:exc:`~django.db.models.Model.DoesNotExist` if there is no object with the
given parameters.

Note that ``earliest()`` and ``latest()`` exist purely for convenience and
readability.

.. admonition:: ``earliest()`` and ``latest()`` may return instances with null dates.

    Since ordering is delegated to the database, results on fields that allow
    null values may be ordered differently if you use different databases. For
    example, PostgreSQL and MySQL sort null values as if they are higher than
    non-null values, while SQLite does the opposite.

    You may want to filter out null values::

        Entry.objects.filter(pub_date__isnull=False).latest('pub_date')

``earliest()``
~~~~~~~~~~~~~~

.. method:: earliest(*fields)

Works otherwise like :meth:`~django.db.models.query.QuerySet.latest` except
the direction is changed.

``first()``
~~~~~~~~~~~

.. method:: first()

Returns the first object matched by the queryset, or ``None`` if there
is no matching object. If the ``QuerySet`` has no ordering defined, then the
queryset is automatically ordered by the primary key. This can affect
aggregation results as described in :ref:`aggregation-ordering-interaction`.

Example::

    p = Article.objects.order_by('title', 'pub_date').first()

Note that ``first()`` is a convenience method, the following code sample is
equivalent to the above example::

    try:
        p = Article.objects.order_by('title', 'pub_date')[0]
    except IndexError:
        p = None

``last()``
~~~~~~~~~~

.. method:: last()

Works like  :meth:`first()`, but returns the last object in the queryset.

``aggregate()``
~~~~~~~~~~~~~~~

.. method:: aggregate(*args, **kwargs)

Returns a dictionary of aggregate values (averages, sums, etc.) calculated over
the ``QuerySet``. Each argument to ``aggregate()`` specifies a value that will
be included in the dictionary that is returned.

The aggregation functions that are provided by Django are described in
`Aggregation Functions`_ below. Since aggregates are also :doc:`query
expressions </ref/models/expressions>`, you may combine aggregates with other
aggregates or values to create complex aggregates.

Aggregates specified using keyword arguments will use the keyword as the name
for the annotation. Anonymous arguments will have a name generated for them
based upon the name of the aggregate function and the model field that is being
aggregated. Complex aggregates cannot use anonymous arguments and must specify
a keyword argument as an alias.

For example, when you are working with blog entries, you may want to know the
number of authors that have contributed blog entries::

    >>> from django.db.models import Count
    >>> q = Blog.objects.aggregate(Count('entry'))
    {'entry__count': 16}

By using a keyword argument to specify the aggregate function, you can
control the name of the aggregation value that is returned::

    >>> q = Blog.objects.aggregate(number_of_entries=Count('entry'))
    {'number_of_entries': 16}

For an in-depth discussion of aggregation, see :doc:`the topic guide on
Aggregation </topics/db/aggregation>`.

``exists()``
~~~~~~~~~~~~

.. method:: exists()

Returns ``True`` if the :class:`.QuerySet` contains any results, and ``False``
if not. This tries to perform the query in the simplest and fastest way
possible, but it *does* execute nearly the same query as a normal
:class:`.QuerySet` query.

:meth:`~.QuerySet.exists` is useful for searches relating to both
object membership in a :class:`.QuerySet` and to the existence of any objects in
a :class:`.QuerySet`, particularly in the context of a large :class:`.QuerySet`.

The most efficient method of finding whether a model with a unique field
(e.g. ``primary_key``) is a member of a :class:`.QuerySet` is::

    entry = Entry.objects.get(pk=123)
    if some_queryset.filter(pk=entry.pk).exists():
        print("Entry contained in queryset")

Which will be faster than the following which requires evaluating and iterating
through the entire queryset::

    if entry in some_queryset:
       print("Entry contained in QuerySet")

And to find whether a queryset contains any items::

    if some_queryset.exists():
        print("There is at least one object in some_queryset")

Which will be faster than::

    if some_queryset:
        print("There is at least one object in some_queryset")

... but not by a large degree (hence needing a large queryset for efficiency
gains).

Additionally, if a ``some_queryset`` has not yet been evaluated, but you know
that it will be at some point, then using ``some_queryset.exists()`` will do
more overall work (one query for the existence check plus an extra one to later
retrieve the results) than using ``bool(some_queryset)``, which retrieves the
results and then checks if any were returned.

``update()``
~~~~~~~~~~~~

.. method:: update(**kwargs)

Performs an SQL update query for the specified fields, and returns
the number of rows matched (which may not be equal to the number of rows
updated if some rows already have the new value).

For example, to turn comments off for all blog entries published in 2010,
you could do this::

    >>> Entry.objects.filter(pub_date__year=2010).update(comments_on=False)

(This assumes your ``Entry`` model has fields ``pub_date`` and ``comments_on``.)

You can update multiple fields — there's no limit on how many. For example,
here we update the ``comments_on`` and ``headline`` fields::

    >>> Entry.objects.filter(pub_date__year=2010).update(comments_on=False, headline='This is old')

The ``update()`` method is applied instantly, and the only restriction on the
:class:`.QuerySet` that is updated is that it can only update columns in the
model's main table, not on related models. You can't do this, for example::

    >>> Entry.objects.update(blog__name='foo') # Won't work!

Filtering based on related fields is still possible, though::

    >>> Entry.objects.filter(blog__id=1).update(comments_on=True)

You cannot call ``update()`` on a :class:`.QuerySet` that has had a slice taken
or can otherwise no longer be filtered.

The ``update()`` method returns the number of affected rows::

    >>> Entry.objects.filter(id=64).update(comments_on=True)
    1

    >>> Entry.objects.filter(slug='nonexistent-slug').update(comments_on=True)
    0

    >>> Entry.objects.filter(pub_date__year=2010).update(comments_on=False)
    132

If you're just updating a record and don't need to do anything with the model
object, the most efficient approach is to call ``update()``, rather than
loading the model object into memory. For example, instead of doing this::

    e = Entry.objects.get(id=10)
    e.comments_on = False
    e.save()

...do this::

    Entry.objects.filter(id=10).update(comments_on=False)

Using ``update()`` also prevents a race condition wherein something might
change in your database in the short period of time between loading the object
and calling ``save()``.

Finally, realize that ``update()`` does an update at the SQL level and, thus,
does not call any ``save()`` methods on your models, nor does it emit the
:attr:`~django.db.models.signals.pre_save` or
:attr:`~django.db.models.signals.post_save` signals (which are a consequence of
calling :meth:`Model.save() <django.db.models.Model.save>`). If you want to
update a bunch of records for a model that has a custom
:meth:`~django.db.models.Model.save()` method, loop over them and call
:meth:`~django.db.models.Model.save()`, like this::

    for e in Entry.objects.filter(pub_date__year=2010):
        e.comments_on = False
        e.save()

Ordered queryset
^^^^^^^^^^^^^^^^

.. versionadded:: 3.2

Chaining ``order_by()`` with ``update()`` is supported only on MariaDB and
MySQL, and is ignored for different databases. This is useful for updating a
unique field in the order that is specified without conflicts. For example::

    Entry.objects.order_by('-number').update(number=F('number') + 1)

.. note::

    If the ``order_by()`` clause contains annotations, it will be ignored.

``delete()``
~~~~~~~~~~~~

.. method:: delete()

Performs an SQL delete query on all rows in the :class:`.QuerySet` and
returns the number of objects deleted and a dictionary with the number of
deletions per object type.

The ``delete()`` is applied instantly. You cannot call ``delete()`` on a
:class:`.QuerySet` that has had a slice taken or can otherwise no longer be
filtered.

For example, to delete all the entries in a particular blog::

    >>> b = Blog.objects.get(pk=1)

    # Delete all the entries belonging to this Blog.
    >>> Entry.objects.filter(blog=b).delete()
    (4, {'weblog.Entry': 2, 'weblog.Entry_authors': 2})

By default, Django's :class:`~django.db.models.ForeignKey` emulates the SQL
constraint ``ON DELETE CASCADE`` — in other words, any objects with foreign
keys pointing at the objects to be deleted will be deleted along with them.
For example::

    >>> blogs = Blog.objects.all()

    # This will delete all Blogs and all of their Entry objects.
    >>> blogs.delete()
    (5, {'weblog.Blog': 1, 'weblog.Entry': 2, 'weblog.Entry_authors': 2})

This cascade behavior is customizable via the
:attr:`~django.db.models.ForeignKey.on_delete` argument to the
:class:`~django.db.models.ForeignKey`.

The ``delete()`` method does a bulk delete and does not call any ``delete()``
methods on your models. It does, however, emit the
:data:`~django.db.models.signals.pre_delete` and
:data:`~django.db.models.signals.post_delete` signals for all deleted objects
(including cascaded deletions).

Django needs to fetch objects into memory to send signals and handle cascades.
However, if there are no cascades and no signals, then Django may take a
fast-path and delete objects without fetching into memory. For large
deletes this can result in significantly reduced memory usage. The amount of
executed queries can be reduced, too.

ForeignKeys which are set to :attr:`~django.db.models.ForeignKey.on_delete`
``DO_NOTHING`` do not prevent taking the fast-path in deletion.

Note that the queries generated in object deletion is an implementation
detail subject to change.

``as_manager()``
~~~~~~~~~~~~~~~~

.. classmethod:: as_manager()

Class method that returns an instance of :class:`~django.db.models.Manager`
with a copy of the ``QuerySet``’s methods. See
:ref:`create-manager-with-queryset-methods` for more details.

``explain()``
~~~~~~~~~~~~~

.. method:: explain(format=None, **options)

Returns a string of the ``QuerySet``’s execution plan, which details how the
database would execute the query, including any indexes or joins that would be
used. Knowing these details may help you improve the performance of slow
queries.

For example, when using PostgreSQL::

    >>> print(Blog.objects.filter(title='My Blog').explain())
    Seq Scan on blog  (cost=0.00..35.50 rows=10 width=12)
      Filter: (title = 'My Blog'::bpchar)

The output differs significantly between databases.

``explain()`` is supported by all built-in database backends except Oracle
because an implementation there isn't straightforward.

The ``format`` parameter changes the output format from the databases's
default, which is usually text-based. PostgreSQL supports ``'TEXT'``,
``'JSON'``, ``'YAML'``, and ``'XML'`` formats. MariaDB and MySQL support
``'TEXT'`` (also called ``'TRADITIONAL'``) and ``'JSON'`` formats. MySQL
8.0.16+ also supports an improved ``'TREE'`` format, which is similar to
PostgreSQL's ``'TEXT'`` output and is used by default, if supported.

Some databases accept flags that can return more information about the query.
Pass these flags as keyword arguments. For example, when using PostgreSQL::

    >>> print(Blog.objects.filter(title='My Blog').explain(verbose=True))
    Seq Scan on public.blog  (cost=0.00..35.50 rows=10 width=12) (actual time=0.004..0.004 rows=10 loops=1)
      Output: id, title
      Filter: (blog.title = 'My Blog'::bpchar)
    Planning time: 0.064 ms
    Execution time: 0.058 ms

On some databases, flags may cause the query to be executed which could have
adverse effects on your database. For example, the ``ANALYZE`` flag supported
by MariaDB, MySQL 8.0.18+, and PostgreSQL could result in changes to data if
there are triggers or if a function is called, even for a ``SELECT`` query.

.. versionchanged:: 3.1

    Support for the ``'TREE'`` format on MySQL 8.0.16+ and ``analyze`` option
    on MariaDB and MySQL 8.0.18+ were added.

.. _field-lookups:

``Field`` lookups
-----------------

Field lookups are how you specify the meat of an SQL ``WHERE`` clause. They're
specified as keyword arguments to the ``QuerySet`` methods :meth:`filter()`,
:meth:`exclude()` and :meth:`get()`.

For an introduction, see :ref:`models and database queries documentation
<field-lookups-intro>`.

Django's built-in lookups are listed below. It is also possible to write
:doc:`custom lookups </howto/custom-lookups>` for model fields.

As a convenience when no lookup type is provided (like in
``Entry.objects.get(id=14)``) the lookup type is assumed to be :lookup:`exact`.

.. fieldlookup:: exact

``exact``
~~~~~~~~~

Exact match. If the value provided for comparison is ``None``, it will be
interpreted as an SQL ``NULL`` (see :lookup:`isnull` for more details).

Examples::

    Entry.objects.get(id__exact=14)
    Entry.objects.get(id__exact=None)

SQL equivalents:

.. code-block:: sql

    SELECT ... WHERE id = 14;
    SELECT ... WHERE id IS NULL;

.. admonition:: MySQL comparisons

    In MySQL, a database table's "collation" setting determines whether
    ``exact`` comparisons are case-sensitive. This is a database setting, *not*
    a Django setting. It's possible to configure your MySQL tables to use
    case-sensitive comparisons, but some trade-offs are involved. For more
    information about this, see the :ref:`collation section <mysql-collation>`
    in the :doc:`databases </ref/databases>` documentation.

.. fieldlookup:: iexact

``iexact``
~~~~~~~~~~

Case-insensitive exact match. If the value provided for comparison is ``None``,
it will be interpreted as an SQL ``NULL`` (see :lookup:`isnull` for more
details).

Example::

    Blog.objects.get(name__iexact='beatles blog')
    Blog.objects.get(name__iexact=None)

SQL equivalents:

.. code-block:: sql

    SELECT ... WHERE name ILIKE 'beatles blog';
    SELECT ... WHERE name IS NULL;

Note the first query will match ``'Beatles Blog'``, ``'beatles blog'``,
``'BeAtLes BLoG'``, etc.

.. admonition:: SQLite users

    When using the SQLite backend and non-ASCII strings, bear in mind the
    :ref:`database note <sqlite-string-matching>` about string comparisons.
    SQLite does not do case-insensitive matching for non-ASCII strings.

.. fieldlookup:: contains

``contains``
~~~~~~~~~~~~

Case-sensitive containment test.

Example::

    Entry.objects.get(headline__contains='Lennon')

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE headline LIKE '%Lennon%';

Note this will match the headline ``'Lennon honored today'`` but not ``'lennon
honored today'``.

.. admonition:: SQLite users

    SQLite doesn't support case-sensitive ``LIKE`` statements; ``contains``
    acts like ``icontains`` for SQLite. See the :ref:`database note
    <sqlite-string-matching>` for more information.


.. fieldlookup:: icontains

``icontains``
~~~~~~~~~~~~~

Case-insensitive containment test.

Example::

    Entry.objects.get(headline__icontains='Lennon')

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE headline ILIKE '%Lennon%';

.. admonition:: SQLite users

    When using the SQLite backend and non-ASCII strings, bear in mind the
    :ref:`database note <sqlite-string-matching>` about string comparisons.

.. fieldlookup:: in

``in``
~~~~~~

In a given iterable; often a list, tuple, or queryset. It's not a common use
case, but strings (being iterables) are accepted.

Examples::

    Entry.objects.filter(id__in=[1, 3, 4])
    Entry.objects.filter(headline__in='abc')

SQL equivalents:

.. code-block:: sql

    SELECT ... WHERE id IN (1, 3, 4);
    SELECT ... WHERE headline IN ('a', 'b', 'c');

You can also use a queryset to dynamically evaluate the list of values
instead of providing a list of literal values::

    inner_qs = Blog.objects.filter(name__contains='Cheddar')
    entries = Entry.objects.filter(blog__in=inner_qs)

This queryset will be evaluated as subselect statement:

.. code-block:: sql

    SELECT ... WHERE blog.id IN (SELECT id FROM ... WHERE NAME LIKE '%Cheddar%')

If you pass in a ``QuerySet`` resulting from ``values()`` or ``values_list()``
as the value to an ``__in`` lookup, you need to ensure you are only extracting
one field in the result. For example, this will work (filtering on the blog
names)::

    inner_qs = Blog.objects.filter(name__contains='Ch').values('name')
    entries = Entry.objects.filter(blog__name__in=inner_qs)

This example will raise an exception, since the inner query is trying to
extract two field values, where only one is expected::

    # Bad code! Will raise a TypeError.
    inner_qs = Blog.objects.filter(name__contains='Ch').values('name', 'id')
    entries = Entry.objects.filter(blog__name__in=inner_qs)

.. _nested-queries-performance:

.. admonition:: Performance considerations

    Be cautious about using nested queries and understand your database
    server's performance characteristics (if in doubt, benchmark!). Some
    database backends, most notably MySQL, don't optimize nested queries very
    well. It is more efficient, in those cases, to extract a list of values
    and then pass that into the second query. That is, execute two queries
    instead of one::

        values = Blog.objects.filter(
                name__contains='Cheddar').values_list('pk', flat=True)
        entries = Entry.objects.filter(blog__in=list(values))

    Note the ``list()`` call around the Blog ``QuerySet`` to force execution of
    the first query. Without it, a nested query would be executed, because
    :ref:`querysets-are-lazy`.

.. fieldlookup:: gt

``gt``
~~~~~~

Greater than.

Example::

    Entry.objects.filter(id__gt=4)

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE id > 4;

.. fieldlookup:: gte

``gte``
~~~~~~~

Greater than or equal to.

.. fieldlookup:: lt

``lt``
~~~~~~

Less than.

.. fieldlookup:: lte

``lte``
~~~~~~~

Less than or equal to.

.. fieldlookup:: startswith

``startswith``
~~~~~~~~~~~~~~

Case-sensitive starts-with.

Example::

    Entry.objects.filter(headline__startswith='Lennon')

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE headline LIKE 'Lennon%';

SQLite doesn't support case-sensitive ``LIKE`` statements; ``startswith`` acts
like ``istartswith`` for SQLite.

.. fieldlookup:: istartswith

``istartswith``
~~~~~~~~~~~~~~~

Case-insensitive starts-with.

Example::

    Entry.objects.filter(headline__istartswith='Lennon')

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE headline ILIKE 'Lennon%';

.. admonition:: SQLite users

    When using the SQLite backend and non-ASCII strings, bear in mind the
    :ref:`database note <sqlite-string-matching>` about string comparisons.

.. fieldlookup:: endswith

``endswith``
~~~~~~~~~~~~

Case-sensitive ends-with.

Example::

    Entry.objects.filter(headline__endswith='Lennon')

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE headline LIKE '%Lennon';

.. admonition:: SQLite users

    SQLite doesn't support case-sensitive ``LIKE`` statements; ``endswith``
    acts like ``iendswith`` for SQLite. Refer to the :ref:`database note
    <sqlite-string-matching>` documentation for more.

.. fieldlookup:: iendswith

``iendswith``
~~~~~~~~~~~~~

Case-insensitive ends-with.

Example::

    Entry.objects.filter(headline__iendswith='Lennon')

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE headline ILIKE '%Lennon'

.. admonition:: SQLite users

    When using the SQLite backend and non-ASCII strings, bear in mind the
    :ref:`database note <sqlite-string-matching>` about string comparisons.

.. fieldlookup:: range

``range``
~~~~~~~~~

Range test (inclusive).

Example::

    import datetime
    start_date = datetime.date(2005, 1, 1)
    end_date = datetime.date(2005, 3, 31)
    Entry.objects.filter(pub_date__range=(start_date, end_date))

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE pub_date BETWEEN '2005-01-01' and '2005-03-31';

You can use ``range`` anywhere you can use ``BETWEEN`` in SQL — for dates,
numbers and even characters.

.. warning::

    Filtering a ``DateTimeField`` with dates won't include items on the last
    day, because the bounds are interpreted as "0am on the given date". If
    ``pub_date`` was a ``DateTimeField``, the above expression would be turned
    into this SQL:

    .. code-block:: sql

        SELECT ... WHERE pub_date BETWEEN '2005-01-01 00:00:00' and '2005-03-31 00:00:00';

    Generally speaking, you can't mix dates and datetimes.

.. fieldlookup:: date

``date``
~~~~~~~~

For datetime fields, casts the value as date. Allows chaining additional field
lookups. Takes a date value.

Example::

    Entry.objects.filter(pub_date__date=datetime.date(2005, 1, 1))
    Entry.objects.filter(pub_date__date__gt=datetime.date(2005, 1, 1))

(No equivalent SQL code fragment is included for this lookup because
implementation of the relevant query varies among different database engines.)

When :setting:`USE_TZ` is ``True``, fields are converted to the current time
zone before filtering. This requires :ref:`time zone definitions in the
database <database-time-zone-definitions>`.

.. fieldlookup:: year

``year``
~~~~~~~~

For date and datetime fields, an exact year match. Allows chaining additional
field lookups. Takes an integer year.

Example::

    Entry.objects.filter(pub_date__year=2005)
    Entry.objects.filter(pub_date__year__gte=2005)

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE pub_date BETWEEN '2005-01-01' AND '2005-12-31';
    SELECT ... WHERE pub_date >= '2005-01-01';

(The exact SQL syntax varies for each database engine.)

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: iso_year

``iso_year``
~~~~~~~~~~~~

For date and datetime fields, an exact ISO 8601 week-numbering year match.
Allows chaining additional field lookups. Takes an integer year.

Example::

    Entry.objects.filter(pub_date__iso_year=2005)
    Entry.objects.filter(pub_date__iso_year__gte=2005)

(The exact SQL syntax varies for each database engine.)

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: month

``month``
~~~~~~~~~

For date and datetime fields, an exact month match. Allows chaining additional
field lookups. Takes an integer 1 (January) through 12 (December).

Example::

    Entry.objects.filter(pub_date__month=12)
    Entry.objects.filter(pub_date__month__gte=6)

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE EXTRACT('month' FROM pub_date) = '12';
    SELECT ... WHERE EXTRACT('month' FROM pub_date) >= '6';

(The exact SQL syntax varies for each database engine.)

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: day

``day``
~~~~~~~

For date and datetime fields, an exact day match. Allows chaining additional
field lookups. Takes an integer day.

Example::

    Entry.objects.filter(pub_date__day=3)
    Entry.objects.filter(pub_date__day__gte=3)

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE EXTRACT('day' FROM pub_date) = '3';
    SELECT ... WHERE EXTRACT('day' FROM pub_date) >= '3';

(The exact SQL syntax varies for each database engine.)

Note this will match any record with a pub_date on the third day of the month,
such as January 3, July 3, etc.

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: week

``week``
~~~~~~~~

For date and datetime fields, return the week number (1-52 or 53) according
to `ISO-8601 <https://en.wikipedia.org/wiki/ISO-8601>`_, i.e., weeks start
on a Monday and the first week contains the year's first Thursday.

Example::

    Entry.objects.filter(pub_date__week=52)
    Entry.objects.filter(pub_date__week__gte=32, pub_date__week__lte=38)

(No equivalent SQL code fragment is included for this lookup because
implementation of the relevant query varies among different database engines.)

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: week_day

``week_day``
~~~~~~~~~~~~

For date and datetime fields, a 'day of the week' match. Allows chaining
additional field lookups.

Takes an integer value representing the day of week from 1 (Sunday) to 7
(Saturday).

Example::

    Entry.objects.filter(pub_date__week_day=2)
    Entry.objects.filter(pub_date__week_day__gte=2)

(No equivalent SQL code fragment is included for this lookup because
implementation of the relevant query varies among different database engines.)

Note this will match any record with a ``pub_date`` that falls on a Monday (day
2 of the week), regardless of the month or year in which it occurs. Week days
are indexed with day 1 being Sunday and day 7 being Saturday.

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: iso_week_day

``iso_week_day``
~~~~~~~~~~~~~~~~

.. versionadded:: 3.1

For date and datetime fields, an exact ISO 8601 day of the week match. Allows
chaining additional field lookups.

Takes an integer value representing the day of the week from 1 (Monday) to 7
(Sunday).

Example::

    Entry.objects.filter(pub_date__iso_week_day=1)
    Entry.objects.filter(pub_date__iso_week_day__gte=1)

(No equivalent SQL code fragment is included for this lookup because
implementation of the relevant query varies among different database engines.)

Note this will match any record with a ``pub_date`` that falls on a Monday (day
1 of the week), regardless of the month or year in which it occurs. Week days
are indexed with day 1 being Monday and day 7 being Sunday.

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: quarter

``quarter``
~~~~~~~~~~~

For date and datetime fields, a 'quarter of the year' match. Allows chaining
additional field lookups. Takes an integer value between 1 and 4 representing
the quarter of the year.

Example to retrieve entries in the second quarter (April 1 to June 30)::

    Entry.objects.filter(pub_date__quarter=2)

(No equivalent SQL code fragment is included for this lookup because
implementation of the relevant query varies among different database engines.)

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: time

``time``
~~~~~~~~

For datetime fields, casts the value as time. Allows chaining additional field
lookups. Takes a :class:`datetime.time` value.

Example::

    Entry.objects.filter(pub_date__time=datetime.time(14, 30))
    Entry.objects.filter(pub_date__time__range=(datetime.time(8), datetime.time(17)))

(No equivalent SQL code fragment is included for this lookup because
implementation of the relevant query varies among different database engines.)

When :setting:`USE_TZ` is ``True``, fields are converted to the current time
zone before filtering. This requires :ref:`time zone definitions in the
database <database-time-zone-definitions>`.

.. fieldlookup:: hour

``hour``
~~~~~~~~

For datetime and time fields, an exact hour match. Allows chaining additional
field lookups. Takes an integer between 0 and 23.

Example::

    Event.objects.filter(timestamp__hour=23)
    Event.objects.filter(time__hour=5)
    Event.objects.filter(timestamp__hour__gte=12)

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE EXTRACT('hour' FROM timestamp) = '23';
    SELECT ... WHERE EXTRACT('hour' FROM time) = '5';
    SELECT ... WHERE EXTRACT('hour' FROM timestamp) >= '12';

(The exact SQL syntax varies for each database engine.)

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: minute

``minute``
~~~~~~~~~~

For datetime and time fields, an exact minute match. Allows chaining additional
field lookups. Takes an integer between 0 and 59.

Example::

    Event.objects.filter(timestamp__minute=29)
    Event.objects.filter(time__minute=46)
    Event.objects.filter(timestamp__minute__gte=29)

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE EXTRACT('minute' FROM timestamp) = '29';
    SELECT ... WHERE EXTRACT('minute' FROM time) = '46';
    SELECT ... WHERE EXTRACT('minute' FROM timestamp) >= '29';

(The exact SQL syntax varies for each database engine.)

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: second

``second``
~~~~~~~~~~

For datetime and time fields, an exact second match. Allows chaining additional
field lookups. Takes an integer between 0 and 59.

Example::

    Event.objects.filter(timestamp__second=31)
    Event.objects.filter(time__second=2)
    Event.objects.filter(timestamp__second__gte=31)

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE EXTRACT('second' FROM timestamp) = '31';
    SELECT ... WHERE EXTRACT('second' FROM time) = '2';
    SELECT ... WHERE EXTRACT('second' FROM timestamp) >= '31';

(The exact SQL syntax varies for each database engine.)

When :setting:`USE_TZ` is ``True``, datetime fields are converted to the
current time zone before filtering. This requires :ref:`time zone definitions
in the database <database-time-zone-definitions>`.

.. fieldlookup:: isnull

``isnull``
~~~~~~~~~~

Takes either ``True`` or ``False``, which correspond to SQL queries of
``IS NULL`` and ``IS NOT NULL``, respectively.

Example::

    Entry.objects.filter(pub_date__isnull=True)

SQL equivalent:

.. code-block:: sql

    SELECT ... WHERE pub_date IS NULL;

.. deprecated:: 3.1

    Using non-boolean values as the right-hand side is deprecated, use ``True``
    or ``False`` instead. In Django 4.0, the exception will be raised.

.. fieldlookup:: regex

``regex``
~~~~~~~~~

Case-sensitive regular expression match.

The regular expression syntax is that of the database backend in use.
In the case of SQLite, which has no built in regular expression support,
this feature is provided by a (Python) user-defined REGEXP function, and
the regular expression syntax is therefore that of Python's ``re`` module.

Example::

    Entry.objects.get(title__regex=r'^(An?|The) +')

SQL equivalents:

.. code-block:: sql

    SELECT ... WHERE title REGEXP BINARY '^(An?|The) +'; -- MySQL

    SELECT ... WHERE REGEXP_LIKE(title, '^(An?|The) +', 'c'); -- Oracle

    SELECT ... WHERE title ~ '^(An?|The) +'; -- PostgreSQL

    SELECT ... WHERE title REGEXP '^(An?|The) +'; -- SQLite

Using raw strings (e.g., ``r'foo'`` instead of ``'foo'``) for passing in the
regular expression syntax is recommended.

.. fieldlookup:: iregex

``iregex``
~~~~~~~~~~

Case-insensitive regular expression match.

Example::

    Entry.objects.get(title__iregex=r'^(an?|the) +')

SQL equivalents:

.. code-block:: sql

    SELECT ... WHERE title REGEXP '^(an?|the) +'; -- MySQL

    SELECT ... WHERE REGEXP_LIKE(title, '^(an?|the) +', 'i'); -- Oracle

    SELECT ... WHERE title ~* '^(an?|the) +'; -- PostgreSQL

    SELECT ... WHERE title REGEXP '(?i)^(an?|the) +'; -- SQLite

.. _aggregation-functions:

Aggregation functions
---------------------

.. currentmodule:: django.db.models

Django provides the following aggregation functions in the
``django.db.models`` module. For details on how to use these
aggregate functions, see :doc:`the topic guide on aggregation
</topics/db/aggregation>`. See the :class:`~django.db.models.Aggregate`
documentation to learn how to create your aggregates.

.. warning::

    SQLite can't handle aggregation on date/time fields out of the box.
    This is because there are no native date/time fields in SQLite and Django
    currently emulates these features using a text field. Attempts to use
    aggregation on date/time fields in SQLite will raise ``NotSupportedError``.

.. admonition:: Note

    Aggregation functions return ``None`` when used with an empty
    ``QuerySet``. For example, the ``Sum`` aggregation function returns ``None``
    instead of ``0`` if the ``QuerySet`` contains no entries. An exception is
    ``Count``, which does return ``0`` if the ``QuerySet`` is empty.

All aggregates have the following parameters in common:

``expressions``
~~~~~~~~~~~~~~~

Strings that reference fields on the model, or :doc:`query expressions
</ref/models/expressions>`.

``output_field``
~~~~~~~~~~~~~~~~

An optional argument that represents the :doc:`model field </ref/models/fields>`
of the return value

.. note::

    When combining multiple field types, Django can only determine the
    ``output_field`` if all fields are of the same type. Otherwise, you
    must provide the ``output_field`` yourself.

.. _aggregate-filter:

``filter``
~~~~~~~~~~

An optional :class:`Q object <django.db.models.Q>` that's used to filter the
rows that are aggregated.

See :ref:`conditional-aggregation` and :ref:`filtering-on-annotations` for
example usage.

``**extra``
~~~~~~~~~~~

Keyword arguments that can provide extra context for the SQL generated
by the aggregate.

``Avg``
~~~~~~~

.. class:: Avg(expression, output_field=None, distinct=False, filter=None, **extra)

    Returns the mean value of the given expression, which must be numeric
    unless you specify a different ``output_field``.

    * Default alias: ``<field>__avg``
    * Return type: ``float`` if input is ``int``, otherwise same as input
      field, or ``output_field`` if supplied

    Has one optional argument:

    .. attribute:: distinct

        If ``distinct=True``, ``Avg`` returns the mean value of unique values.
        This is the SQL equivalent of ``AVG(DISTINCT <field>)``. The default
        value is ``False``.

``Count``
~~~~~~~~~

.. class:: Count(expression, distinct=False, filter=None, **extra)

    Returns the number of objects that are related through the provided
    expression.

    * Default alias: ``<field>__count``
    * Return type: ``int``

    Has one optional argument:

    .. attribute:: distinct

        If ``distinct=True``, the count will only include unique instances.
        This is the SQL equivalent of ``COUNT(DISTINCT <field>)``. The default
        value is ``False``.

``Max``
~~~~~~~

.. class:: Max(expression, output_field=None, filter=None, **extra)

    Returns the maximum value of the given expression.

    * Default alias: ``<field>__max``
    * Return type: same as input field, or ``output_field`` if supplied

``Min``
~~~~~~~

.. class:: Min(expression, output_field=None, filter=None, **extra)

    Returns the minimum value of the given expression.

    * Default alias: ``<field>__min``
    * Return type: same as input field, or ``output_field`` if supplied

``StdDev``
~~~~~~~~~~

.. class:: StdDev(expression, output_field=None, sample=False, filter=None, **extra)

    Returns the standard deviation of the data in the provided expression.

    * Default alias: ``<field>__stddev``
    * Return type: ``float`` if input is ``int``, otherwise same as input
      field, or ``output_field`` if supplied

    Has one optional argument:

    .. attribute:: sample

        By default, ``StdDev`` returns the population standard deviation. However,
        if ``sample=True``, the return value will be the sample standard deviation.

``Sum``
~~~~~~~

.. class:: Sum(expression, output_field=None, distinct=False, filter=None, **extra)

    Computes the sum of all values of the given expression.

    * Default alias: ``<field>__sum``
    * Return type: same as input field, or ``output_field`` if supplied

    Has one optional argument:

    .. attribute:: distinct

        If ``distinct=True``, ``Sum`` returns the sum of unique values. This is
        the SQL equivalent of ``SUM(DISTINCT <field>)``. The default value is
        ``False``.

``Variance``
~~~~~~~~~~~~

.. class:: Variance(expression, output_field=None, sample=False, filter=None, **extra)

    Returns the variance of the data in the provided expression.

    * Default alias: ``<field>__variance``
    * Return type: ``float`` if input is ``int``, otherwise same as input
      field, or ``output_field`` if supplied

    Has one optional argument:

    .. attribute:: sample

        By default, ``Variance`` returns the population variance. However,
        if ``sample=True``, the return value will be the sample variance.

Query-related tools
===================

This section provides reference material for query-related tools not documented
elsewhere.

``Q()`` objects
---------------

.. class:: Q

A ``Q()`` object represents an SQL condition that can be used in
database-related operations. It's similar to how an
:class:`F() <django.db.models.F>` object represents the value of a model field
or annotation. They make it possible to define and reuse conditions, and
combine them using operators such as ``|`` (``OR``) and ``&`` (``AND``). See
:ref:`complex-lookups-with-q`.

``Prefetch()`` objects
----------------------

.. class:: Prefetch(lookup, queryset=None, to_attr=None)

The ``Prefetch()`` object can be used to control the operation of
:meth:`~django.db.models.query.QuerySet.prefetch_related()`.

The ``lookup`` argument describes the relations to follow and works the same
as the string based lookups passed to
:meth:`~django.db.models.query.QuerySet.prefetch_related()`. For example:

    >>> from django.db.models import Prefetch
    >>> Question.objects.prefetch_related(Prefetch('choice_set')).get().choice_set.all()
    <QuerySet [<Choice: Not much>, <Choice: The sky>, <Choice: Just hacking again>]>
    # This will only execute two queries regardless of the number of Question
    # and Choice objects.
    >>> Question.objects.prefetch_related(Prefetch('choice_set')).all()
    <QuerySet [<Question: What's up?>]>

The ``queryset`` argument supplies a base ``QuerySet`` for the given lookup.
This is useful to further filter down the prefetch operation, or to call
:meth:`~django.db.models.query.QuerySet.select_related()` from the prefetched
relation, hence reducing the number of queries even further:

    >>> voted_choices = Choice.objects.filter(votes__gt=0)
    >>> voted_choices
    <QuerySet [<Choice: The sky>]>
    >>> prefetch = Prefetch('choice_set', queryset=voted_choices)
    >>> Question.objects.prefetch_related(prefetch).get().choice_set.all()
    <QuerySet [<Choice: The sky>]>

The ``to_attr`` argument sets the result of the prefetch operation to a custom
attribute:

    >>> prefetch = Prefetch('choice_set', queryset=voted_choices, to_attr='voted_choices')
    >>> Question.objects.prefetch_related(prefetch).get().voted_choices
    [<Choice: The sky>]
    >>> Question.objects.prefetch_related(prefetch).get().choice_set.all()
    <QuerySet [<Choice: Not much>, <Choice: The sky>, <Choice: Just hacking again>]>

.. note::

    When using ``to_attr`` the prefetched result is stored in a list. This can
    provide a significant speed improvement over traditional
    ``prefetch_related`` calls which store the cached result within a
    ``QuerySet`` instance.

``prefetch_related_objects()``
------------------------------

.. function:: prefetch_related_objects(model_instances, *related_lookups)

Prefetches the given lookups on an iterable of model instances. This is useful
in code that receives a list of model instances as opposed to a ``QuerySet``;
for example, when fetching models from a cache or instantiating them manually.

Pass an iterable of model instances (must all be of the same class) and the
lookups or :class:`Prefetch` objects you want to prefetch for. For example::

    >>> from django.db.models import prefetch_related_objects
    >>> restaurants = fetch_top_restaurants_from_cache()  # A list of Restaurants
    >>> prefetch_related_objects(restaurants, 'pizzas__toppings')

``FilteredRelation()`` objects
------------------------------

.. class:: FilteredRelation(relation_name, *, condition=Q())

    .. attribute:: FilteredRelation.relation_name

        The name of the field on which you'd like to filter the relation.

    .. attribute:: FilteredRelation.condition

        A :class:`~django.db.models.Q` object to control the filtering.

``FilteredRelation`` is used with :meth:`~.QuerySet.annotate()` to create an
``ON`` clause when a ``JOIN`` is performed. It doesn't act on the default
relationship but on the annotation name (``pizzas_vegetarian`` in example
below).

For example, to find restaurants that have vegetarian pizzas with
``'mozzarella'`` in the name::

    >>> from django.db.models import FilteredRelation, Q
    >>> Restaurant.objects.annotate(
    ...    pizzas_vegetarian=FilteredRelation(
    ...        'pizzas', condition=Q(pizzas__vegetarian=True),
    ...    ),
    ... ).filter(pizzas_vegetarian__name__icontains='mozzarella')

If there are a large number of pizzas, this queryset performs better than::

    >>> Restaurant.objects.filter(
    ...     pizzas__vegetarian=True,
    ...     pizzas__name__icontains='mozzarella',
    ... )

because the filtering in the ``WHERE`` clause of the first queryset will only
operate on vegetarian pizzas.

``FilteredRelation`` doesn't support:

* :meth:`.QuerySet.only` and :meth:`~.QuerySet.prefetch_related`.
* A :class:`~django.contrib.contenttypes.fields.GenericForeignKey`
  inherited from a parent model.

.. versionchanged:: 3.2

    Support for nested relations was added.