Once you’ve created your data models, Django automatically gives you a database-abstraction API that lets you create, retrieve, update and delete objects. This document explains how to use this API. Refer to the data model reference for full details of all the various model lookup options.
Throughout this guide (and in the reference), we’ll refer to the following models, which comprise a Weblog application:
from django.db import models class Blog(models.Model): name = models.CharField(max_length=100) tagline = models.TextField() def __str__(self): # __unicode__ on Python 2 return self.name class Author(models.Model): name = models.CharField(max_length=50) email = models.EmailField() def __str__(self): # __unicode__ on Python 2 return self.name class Entry(models.Model): blog = models.ForeignKey(Blog) headline = models.CharField(max_length=255) body_text = models.TextField() pub_date = models.DateField() mod_date = models.DateField() authors = models.ManyToManyField(Author) n_comments = models.IntegerField() n_pingbacks = models.IntegerField() rating = models.IntegerField() def __str__(self): # __unicode__ on Python 2 return self.headline
To represent database-table data in Python objects, Django uses an intuitive system: A model class represents a database table, and an instance of that class represents a particular record in the database table.
To create an object, instantiate it using keyword arguments to the model class, then call save() to save it to the database.
Assuming models live in a file mysite/blog/models.py, here’s an example:
>>> from blog.models import Blog >>> b = Blog(name='Beatles Blog', tagline='All the latest Beatles news.') >>> b.save()
This performs an INSERT SQL statement behind the scenes. Django doesn’t hit the database until you explicitly call save().
The save() method has no return value.
Saving changes to objects¶
To save changes to an object that’s already in the database, use save().
Given a Blog instance b5 that has already been saved to the database, this example changes its name and updates its record in the database:
>>> b5.name = 'New name' >>> b5.save()
This performs an UPDATE SQL statement behind the scenes. Django doesn’t hit the database until you explicitly call save().
Saving ForeignKey and ManyToManyField fields¶
Updating a ForeignKey field works exactly the same way as saving a normal field – simply assign an object of the right type to the field in question. This example updates the blog attribute of an Entry instance entry, assuming appropriate instances of Entry and Blog are already saved to the database (so we can retrieve them below):
>>> from blog.models import Entry >>> entry = Entry.objects.get(pk=1) >>> cheese_blog = Blog.objects.get(name="Cheddar Talk") >>> entry.blog = cheese_blog >>> entry.save()
>>> from blog.models import Author >>> joe = Author.objects.create(name="Joe") >>> entry.authors.add(joe)
>>> john = Author.objects.create(name="John") >>> paul = Author.objects.create(name="Paul") >>> george = Author.objects.create(name="George") >>> ringo = Author.objects.create(name="Ringo") >>> entry.authors.add(john, paul, george, ringo)
Django will complain if you try to assign or add an object of the wrong type.
A QuerySet represents a collection of objects from your database. It can have zero, one or many filters. Filters narrow down the query results based on the given parameters. In SQL terms, a QuerySet equates to a SELECT statement, and a filter is a limiting clause such as WHERE or LIMIT.
>>> Blog.objects <django.db.models.manager.Manager object at ...> >>> b = Blog(name='Foo', tagline='Bar') >>> b.objects Traceback: ... AttributeError: "Manager isn't accessible via Blog instances."
Managers are accessible only via model classes, rather than from model instances, to enforce a separation between “table-level” operations and “record-level” operations.
Retrieving all objects¶
>>> all_entries = Entry.objects.all()
Retrieving specific objects with filters¶
- Returns a new QuerySet containing objects that match the given lookup parameters.
- Returns a new QuerySet containing objects that do not match the given lookup parameters.
The lookup parameters (**kwargs in the above function definitions) should be in the format described in Field lookups below.
With the default manager class, it is the same as:
>>> Entry.objects.filter( ... headline__startswith='What' ... ).exclude( ... pub_date__gte=datetime.date.today() ... ).filter( ... pub_date__gte=datetime(2005, 1, 30) ... )
This takes the initial QuerySet of all entries in the database, adds a filter, then an exclusion, then another filter. The final result is a QuerySet containing all entries with a headline that starts with “What”, that were published between January 30, 2005, and the current day.
Filtered QuerySets are unique¶
Each time you refine a QuerySet, you get a brand-new QuerySet that is in no way bound to the previous QuerySet. Each refinement creates a separate and distinct QuerySet that can be stored, used and reused.
>>> q1 = Entry.objects.filter(headline__startswith="What") >>> q2 = q1.exclude(pub_date__gte=datetime.date.today()) >>> q3 = q1.filter(pub_date__gte=datetime.date.today())
These three QuerySets are separate. The first is a base QuerySet containing all entries that contain a headline starting with “What”. The second is a subset of the first, with an additional criteria that excludes records whose pub_date is today or in the future. The third is a subset of the first, with an additional criteria that selects only the records whose pub_date is today or in the future. The initial QuerySet (q1) is unaffected by the refinement process.
QuerySets are lazy¶
QuerySets are lazy – the act of creating a QuerySet doesn’t involve any database activity. You can stack filters together all day long, and Django won’t actually run the query until the QuerySet is evaluated. Take a look at this example:
>>> q = Entry.objects.filter(headline__startswith="What") >>> q = q.filter(pub_date__lte=datetime.date.today()) >>> q = q.exclude(body_text__icontains="food") >>> print(q)
Though this looks like three database hits, in fact it hits the database only once, at the last line (print(q)). In general, the results of a QuerySet aren’t fetched from the database until you “ask” for them. When you do, the QuerySet is evaluated by accessing the database. For more details on exactly when evaluation takes place, see When QuerySets are evaluated.
Retrieving a single object with get¶
>>> one_entry = Entry.objects.get(pk=1)
Note that there is a difference between using get(), and using filter() with a slice of . If there are no results that match the query, get() will raise a DoesNotExist exception. This exception is an attribute of the model class that the query is being performed on - so in the code above, if there is no Entry object with a primary key of 1, Django will raise Entry.DoesNotExist.
Other QuerySet methods¶
Most of the time you’ll use all(), get(), filter() and exclude() when you need to look up objects from the database. However, that’s far from all there is; see the QuerySet API Reference for a complete list of all the various QuerySet methods.
Use a subset of Python’s array-slicing syntax to limit your QuerySet to a certain number of results. This is the equivalent of SQL’s LIMIT and OFFSET clauses.
For example, this returns the first 5 objects (LIMIT 5):
This returns the sixth through tenth objects (OFFSET 5 LIMIT 5):
Negative indexing (i.e. Entry.objects.all()[-1]) is not supported.
Generally, slicing a QuerySet returns a new QuerySet – it doesn’t evaluate the query. An exception is if you use the “step” parameter of Python slice syntax. For example, this would actually execute the query in order to return a list of every second object of the first 10:
To retrieve a single object rather than a list (e.g. SELECT foo FROM bar LIMIT 1), use a simple index instead of a slice. For example, this returns the first Entry in the database, after ordering entries alphabetically by headline:
This is roughly equivalent to:
Note, however, that the first of these will raise IndexError while the second will raise DoesNotExist if no objects match the given criteria. See get() for more details.
Basic lookups keyword arguments take the form field__lookuptype=value. (That’s a double-underscore). For example:
translates (roughly) into the following SQL:
SELECT * FROM blog_entry WHERE pub_date <= '2006-01-01';
How this is possible
Python has the ability to define functions that accept arbitrary name-value arguments whose names and values are evaluated at runtime. For more information, see Keyword Arguments in the official Python tutorial.
The field specified in a lookup has to be the name of a model field. There’s one exception though, in case of a ForeignKey you can specify the field name suffixed with _id. In this case, the value parameter is expected to contain the raw value of the foreign model’s primary key. For example:
If you pass an invalid keyword argument, a lookup function will raise TypeError.
The database API supports about two dozen lookup types; a complete reference can be found in the field lookup reference. To give you a taste of what’s available, here’s some of the more common lookups you’ll probably use:
An “exact” match. For example:
>>> Entry.objects.get(headline__exact="Man bites dog")
Would generate SQL along these lines:
SELECT ... WHERE headline = 'Man bites dog';
If you don’t provide a lookup type – that is, if your keyword argument doesn’t contain a double underscore – the lookup type is assumed to be exact.
For example, the following two statements are equivalent:
>>> Blog.objects.get(id__exact=14) # Explicit form >>> Blog.objects.get(id=14) # __exact is implied
This is for convenience, because exact lookups are the common case.
A case-insensitive match. So, the query:
>>> Blog.objects.get(name__iexact="beatles blog")
Would match a Blog titled "Beatles Blog", "beatles blog", or even "BeAtlES blOG".
Case-sensitive containment test. For example:
Roughly translates to this SQL:
SELECT ... WHERE headline LIKE '%Lennon%';
Note this will match the headline 'Today Lennon honored' but not 'today lennon honored'.
There’s also a case-insensitive version, icontains.
- startswith, endswith
- Starts-with and ends-with search, respectively. There are also case-insensitive versions called istartswith and iendswith.
Again, this only scratches the surface. A complete reference can be found in the field lookup reference.
Lookups that span relationships¶
Django offers a powerful and intuitive way to “follow” relationships in lookups, taking care of the SQL JOINs for you automatically, behind the scenes. To span a relationship, just use the field name of related fields across models, separated by double underscores, until you get to the field you want.
This example retrieves all Entry objects with a Blog whose name is 'Beatles Blog':
>>> Entry.objects.filter(blog__name='Beatles Blog')
This spanning can be as deep as you’d like.
It works backwards, too. To refer to a “reverse” relationship, just use the lowercase name of the model.
This example retrieves all Blog objects which have at least one Entry whose headline contains 'Lennon':
If you are filtering across multiple relationships and one of the intermediate models doesn’t have a value that meets the filter condition, Django will treat it as if there is an empty (all values are NULL), but valid, object there. All this means is that no error will be raised. For example, in this filter:
(if there was a related Author model), if there was no author associated with an entry, it would be treated as if there was also no name attached, rather than raising an error because of the missing author. Usually this is exactly what you want to have happen. The only case where it might be confusing is if you are using isnull. Thus:
will return Blog objects that have an empty name on the author and also those which have an empty author on the entry. If you don’t want those latter objects, you could write:
Spanning multi-valued relationships¶
When you are filtering an object based on a ManyToManyField or a reverse ForeignKey, there are two different sorts of filter you may be interested in. Consider the Blog/Entry relationship (Blog to Entry is a one-to-many relation). We might be interested in finding blogs that have an entry which has both “Lennon” in the headline and was published in 2008. Or we might want to find blogs that have an entry with “Lennon” in the headline as well as an entry that was published in 2008. Since there are multiple entries associated with a single Blog, both of these queries are possible and make sense in some situations.
The same type of situation arises with a ManyToManyField. For example, if an Entry has a ManyToManyField called tags, we might want to find entries linked to tags called “music” and “bands” or we might want an entry that contains a tag with a name of “music” and a status of “public”.
To handle both of these situations, Django has a consistent way of processing filter() calls. Everything inside a single filter() call is applied simultaneously to filter out items matching all those requirements. Successive filter() calls further restrict the set of objects, but for multi-valued relations, they apply to any object linked to the primary model, not necessarily those objects that were selected by an earlier filter() call.
That may sound a bit confusing, so hopefully an example will clarify. To select all blogs that contain entries with both “Lennon” in the headline and that were published in 2008 (the same entry satisfying both conditions), we would write:
To select all blogs that contain an entry with “Lennon” in the headline as well as an entry that was published in 2008, we would write:
Suppose there is only one blog that had both entries containing “Lennon” and entries from 2008, but that none of the entries from 2008 contained “Lennon”. The first query would not return any blogs, but the second query would return that one blog.
In the second example, the first filter restricts the queryset to all those blogs linked to entries with “Lennon” in the headline. The second filter restricts the set of blogs further to those that are also linked to entries that were published in 2008. The entries selected by the second filter may or may not be the same as the entries in the first filter. We are filtering the Blog items with each filter statement, not the Entry items.
The behavior of filter() for queries that span multi-value relationships, as described above, is not implemented equivalently for exclude(). Instead, the conditions in a single exclude() call will not necessarily refer to the same item.
For example, the following query would exclude blogs that contain both entries with “Lennon” in the headline and entries published in 2008:
Blog.objects.exclude( entry__headline__contains='Lennon', entry__pub_date__year=2008, )
However, unlike the behavior when using filter(), this will not limit blogs based on entries that satisfy both conditions. In order to do that, i.e. to select all blogs that do not contain entries published with “Lennon” that were published in 2008, you need to make two queries:
Blog.objects.exclude( entry=Entry.objects.filter( headline__contains='Lennon', pub_date__year=2008, ), )
Filters can reference fields on the model¶
In the examples given so far, we have constructed filters that compare the value of a model field with a constant. But what if you want to compare the value of a model field with another field on the same model?
Django provides F expressions to allow such comparisons. Instances of F() act as a reference to a model field within a query. These references can then be used in query filters to compare the values of two different fields on the same model instance.
For example, to find a list of all blog entries that have had more comments than pingbacks, we construct an F() object to reference the pingback count, and use that F() object in the query:
>>> from django.db.models import F >>> Entry.objects.filter(n_comments__gt=F('n_pingbacks'))
Django supports the use of addition, subtraction, multiplication, division, modulo, and power arithmetic with F() objects, both with constants and with other F() objects. To find all the blog entries with more than twice as many comments as pingbacks, we modify the query:
>>> Entry.objects.filter(n_comments__gt=F('n_pingbacks') * 2)
The power operator ** was added.
To find all the entries where the rating of the entry is less than the sum of the pingback count and comment count, we would issue the query:
>>> Entry.objects.filter(rating__lt=F('n_comments') + F('n_pingbacks'))
You can also use the double underscore notation to span relationships in an F() object. An F() object with a double underscore will introduce any joins needed to access the related object. For example, to retrieve all the entries where the author’s name is the same as the blog name, we could issue the query:
For date and date/time fields, you can add or subtract a timedelta object. The following would return all entries that were modified more than 3 days after they were published:
>>> from datetime import timedelta >>> Entry.objects.filter(mod_date__gt=F('pub_date') + timedelta(days=3))
The F() objects support bitwise operations by .bitand() and .bitor(), for example:
The pk lookup shortcut¶
For convenience, Django provides a pk lookup shortcut, which stands for “primary key”.
In the example Blog model, the primary key is the id field, so these three statements are equivalent:
>>> Blog.objects.get(id__exact=14) # Explicit form >>> Blog.objects.get(id=14) # __exact is implied >>> Blog.objects.get(pk=14) # pk implies id__exact
The use of pk isn’t limited to __exact queries – any query term can be combined with pk to perform a query on the primary key of a model:
# Get blogs entries with id 1, 4 and 7 >>> Blog.objects.filter(pk__in=[1,4,7]) # Get all blog entries with id > 14 >>> Blog.objects.filter(pk__gt=14)
pk lookups also work across joins. For example, these three statements are equivalent:
>>> Entry.objects.filter(blog__id__exact=3) # Explicit form >>> Entry.objects.filter(blog__id=3) # __exact is implied >>> Entry.objects.filter(blog__pk=3) # __pk implies __id__exact
Escaping percent signs and underscores in LIKE statements¶
The field lookups that equate to LIKE SQL statements (iexact, contains, icontains, startswith, istartswith, endswith and iendswith) will automatically escape the two special characters used in LIKE statements – the percent sign and the underscore. (In a LIKE statement, the percent sign signifies a multiple-character wildcard and the underscore signifies a single-character wildcard.)
This means things should work intuitively, so the abstraction doesn’t leak. For example, to retrieve all the entries that contain a percent sign, just use the percent sign as any other character:
Django takes care of the quoting for you; the resulting SQL will look something like this:
SELECT ... WHERE headline LIKE '%\%%';
Same goes for underscores. Both percentage signs and underscores are handled for you transparently.
Caching and QuerySets¶
Each QuerySet contains a cache to minimize database access. Understanding how it works will allow you to write the most efficient code.
In a newly created QuerySet, the cache is empty. The first time a QuerySet is evaluated – and, hence, a database query happens – Django saves the query results in the QuerySet’s cache and returns the results that have been explicitly requested (e.g., the next element, if the QuerySet is being iterated over). Subsequent evaluations of the QuerySet reuse the cached results.
>>> print([e.headline for e in Entry.objects.all()]) >>> print([e.pub_date for e in Entry.objects.all()])
That means the same database query will be executed twice, effectively doubling your database load. Also, there’s a possibility the two lists may not include the same database records, because an Entry may have been added or deleted in the split second between the two requests.
To avoid this problem, simply save the QuerySet and reuse it:
>>> queryset = Entry.objects.all() >>> print([p.headline for p in queryset]) # Evaluate the query set. >>> print([p.pub_date for p in queryset]) # Re-use the cache from the evaluation.
When querysets are not cached¶
Querysets do not always cache their results. When evaluating only part of the queryset, the cache is checked, but if it is not populated then the items returned by the subsequent query are not cached. Specifically, this means that limiting the queryset using an array slice or an index will not populate the cache.
For example, repeatedly getting a certain index in a queryset object will query the database each time:
>>> queryset = Entry.objects.all() >>> print queryset # Queries the database >>> print queryset # Queries the database again
However, if the entire queryset has already been evaluated, the cache will be checked instead:
>>> queryset = Entry.objects.all() >>> [entry for entry in queryset] # Queries the database >>> print queryset # Uses cache >>> print queryset # Uses cache
Here are some examples of other actions that will result in the entire queryset being evaluated and therefore populate the cache:
>>> [entry for entry in queryset] >>> bool(queryset) >>> entry in queryset >>> list(queryset)
Simply printing the queryset will not populate the cache. This is because the call to __repr__() only returns a slice of the entire queryset.
Complex lookups with Q objects¶
A Q object (django.db.models.Q) is an object used to encapsulate a collection of keyword arguments. These keyword arguments are specified as in “Field lookups” above.
For example, this Q object encapsulates a single LIKE query:
from django.db.models import Q Q(question__startswith='What')
Q objects can be combined using the & and | operators. When an operator is used on two Q objects, it yields a new Q object.
For example, this statement yields a single Q object that represents the “OR” of two "question__startswith" queries:
Q(question__startswith='Who') | Q(question__startswith='What')
This is equivalent to the following SQL WHERE clause:
WHERE question LIKE 'Who%' OR question LIKE 'What%'
You can compose statements of arbitrary complexity by combining Q objects with the & and | operators and use parenthetical grouping. Also, Q objects can be negated using the ~ operator, allowing for combined lookups that combine both a normal query and a negated (NOT) query:
Q(question__startswith='Who') | ~Q(pub_date__year=2005)
Each lookup function that takes keyword-arguments (e.g. filter(), exclude(), get()) can also be passed one or more Q objects as positional (not-named) arguments. If you provide multiple Q object arguments to a lookup function, the arguments will be “AND”ed together. For example:
Poll.objects.get( Q(question__startswith='Who'), Q(pub_date=date(2005, 5, 2)) | Q(pub_date=date(2005, 5, 6)) )
... roughly translates into the SQL:
SELECT * from polls WHERE question LIKE 'Who%' AND (pub_date = '2005-05-02' OR pub_date = '2005-05-06')
Lookup functions can mix the use of Q objects and keyword arguments. All arguments provided to a lookup function (be they keyword arguments or Q objects) are “AND”ed together. However, if a Q object is provided, it must precede the definition of any keyword arguments. For example:
Poll.objects.get( Q(pub_date=date(2005, 5, 2)) | Q(pub_date=date(2005, 5, 6)), question__startswith='Who')
... would be a valid query, equivalent to the previous example; but:
# INVALID QUERY Poll.objects.get( question__startswith='Who', Q(pub_date=date(2005, 5, 2)) | Q(pub_date=date(2005, 5, 6)))
... would not be valid.
The OR lookups examples in the Django unit tests show some possible uses of Q.
To compare two model instances, just use the standard Python comparison operator, the double equals sign: ==. Behind the scenes, that compares the primary key values of two models.
Using the Entry example above, the following two statements are equivalent:
>>> some_entry == other_entry >>> some_entry.id == other_entry.id
If a model’s primary key isn’t called id, no problem. Comparisons will always use the primary key, whatever it’s called. For example, if a model’s primary key field is called name, these two statements are equivalent:
>>> some_obj == other_obj >>> some_obj.name == other_obj.name
The delete method, conveniently, is named delete(). This method immediately deletes the object and has no return value. Example:
For example, this deletes all Entry objects with a pub_date year of 2005:
Keep in mind that this will, whenever possible, be executed purely in SQL, and so the delete() methods of individual object instances will not necessarily be called during the process. If you’ve provided a custom delete() method on a model class and want to ensure that it is called, you will need to “manually” delete instances of that model (e.g., by iterating over a QuerySet and calling delete() on each object individually) rather than using the bulk delete() method of a QuerySet.
When Django deletes an object, by default it emulates the behavior of the SQL constraint ON DELETE CASCADE – in other words, any objects which had foreign keys pointing at the object to be deleted will be deleted along with it. For example:
b = Blog.objects.get(pk=1) # This will delete the Blog and all of its Entry objects. b.delete()
Note that delete() is the only QuerySet method that is not exposed on a Manager itself. This is a safety mechanism to prevent you from accidentally requesting Entry.objects.delete(), and deleting all the entries. If you do want to delete all the objects, then you have to explicitly request a complete query set:
Copying model instances¶
Although there is no built-in method for copying model instances, it is possible to easily create new instance with all fields’ values copied. In the simplest case, you can just set pk to None. Using our blog example:
blog = Blog(name='My blog', tagline='Blogging is easy') blog.save() # blog.pk == 1 blog.pk = None blog.save() # blog.pk == 2
Things get more complicated if you use inheritance. Consider a subclass of Blog:
class ThemeBlog(Blog): theme = models.CharField(max_length=200) django_blog = ThemeBlog(name='Django', tagline='Django is easy', theme='python') django_blog.save() # django_blog.pk == 3
Due to how inheritance works, you have to set both pk and id to None:
django_blog.pk = None django_blog.id = None django_blog.save() # django_blog.pk == 4
This process does not copy related objects. If you want to copy relations, you have to write a little bit more code. In our example, Entry has a many to many field to Author:
entry = Entry.objects.all() # some previous entry old_authors = entry.authors.all() entry.pk = None entry.save() entry.authors = old_authors # saves new many2many relations
Updating multiple objects at once¶
# Update all the headlines with pub_date in 2007. Entry.objects.filter(pub_date__year=2007).update(headline='Everything is the same')
You can only set non-relation fields and ForeignKey fields using this method. To update a non-relation field, provide the new value as a constant. To update ForeignKey fields, set the new value to be the new model instance you want to point to. For example:
>>> b = Blog.objects.get(pk=1) # Change every Entry so that it belongs to this Blog. >>> Entry.objects.all().update(blog=b)
The update() method is applied instantly and returns the number of rows matched by the query (which may not be equal to the number of rows updated if some rows already have the new value). The only restriction on the QuerySet that is updated is that it can only access one database table, the model’s main table. You can filter based on related fields, but you can only update columns in the model’s main table. Example:
>>> b = Blog.objects.get(pk=1) # Update all the headlines belonging to this Blog. >>> Entry.objects.select_related().filter(blog=b).update(headline='Everything is the same')
Be aware that the update() method is converted directly to an SQL statement. It is a bulk operation for direct updates. It doesn’t run any save() methods on your models, or emit the pre_save or post_save signals (which are a consequence of calling save()), or honor the auto_now field option. If you want to save every item in a QuerySet and make sure that the save() method is called on each instance, you don’t need any special function to handle that. Just loop over them and call save():
for item in my_queryset: item.save()
Calls to update can also use F expressions to update one field based on the value of another field in the model. This is especially useful for incrementing counters based upon their current value. For example, to increment the pingback count for every entry in the blog:
>>> Entry.objects.all().update(n_pingbacks=F('n_pingbacks') + 1)
However, unlike F() objects in filter and exclude clauses, you can’t introduce joins when you use F() objects in an update – you can only reference fields local to the model being updated. If you attempt to introduce a join with an F() object, a FieldError will be raised:
# THIS WILL RAISE A FieldError >>> Entry.objects.update(headline=F('blog__name'))
Falling back to raw SQL¶
If you find yourself needing to write an SQL query that is too complex for Django’s database-mapper to handle, you can fall back on writing SQL by hand. Django has a couple of options for writing raw SQL queries; see Performing raw SQL queries.
Finally, it’s important to note that the Django database layer is merely an interface to your database. You can access your database via other tools, programming languages or database frameworks; there’s nothing Django-specific about your database.