In a Django QuerySet, how to filter for “not exist

Question

I have two models like this:

class User(models.Model):
    email = models.EmailField()

class Report(models.Model):
    user = models.ForeignKey(User)

In reality each model has more fields which are of no consequence to this question.

I want to filter all users who have an email which starts with 'a' and have no reports. There will be more .filter() and .exclude() criteria based on other fields.

I want to approach it like this:

users = User.objects.filter(email__like = 'a%')

users = users.filter(<other filters>)

users = ???

I would like ??? to filter out users who do not have reports associated with them. How would I do this? If this is not possible as I have presented it, what is an alternate approach?

Answer 1

To filter users who do not have reports associated with them try this:

users = User.objects.exclude(id__in=[elem.user.id for elem in Report.objects.all()])

Answer 2

New in Django 1.11 you can add EXISTS subqueries:

User.objects.annotate(
    no_reports=~Exists(Reports.objects.filter(user__eq=OuterRef('pk')))
).filter(
    email__startswith='a',
    no_reports=True
)

This generates SQL something like this:

SELECT
    user.pk,
    user.email,
    NOT EXISTS (SELECT U0.pk FROM reports U0 WHERE U0.user = user.pk) AS no_reports
FROM user
WHERE email LIKE 'a%' AND NOT EXISTS (SELECT U0.pk FROM reports U0 WHERE U0.user = user.pk);

A NOT EXISTS clause is almost always the most efficient way to do a "not exists" filter.

Once #25367 is released, you'll be able to use ~Exists() directly in a .filter(), avoiding the duplicate clause.

Answer 3

The only way to get native SQL EXISTS/NOT EXISTS without extra queries or JOINs is to add it as raw SQL in the .extra() clause:

users = users.extra(where=[
    """NOT EXISTS(SELECT 1 FROM {reports} 
                  WHERE user_id={users}.id)
    """.format(reports=Report._meta.db_table, users=User._meta.db_table)
])

In fact, it's a pretty obvious and efficient solution and I sometimes wonder why it wasn't built in to Django as a lookup. Also it allows to refine the subquery to find e.g. only users with[out] a report during last week, or with[out] an unanswered/unviewed report.

Answer 4

Alasdair's answer is helpful, but I don't like using distinct(). It can sometimes be useful, but it's usually a code smell telling you that you messed up your joins.

Luckily, Django's queryset lets you filter on subqueries.

Here are a few ways to run the queries from your question:

# Tested with Django 1.9.2
import logging
import sys

import django
from django.apps import apps
from django.apps.config import AppConfig
from django.conf import settings
from django.db import connections, models, DEFAULT_DB_ALIAS
from django.db.models.base import ModelBase

NAME = 'udjango'


def main():

    setup()

    class User(models.Model):
        email = models.EmailField()

        def __repr__(self):
            return 'User({!r})'.format(self.email)

    class Report(models.Model):
        user = models.ForeignKey(User)

    syncdb(User)
    syncdb(Report)

    anne = User.objects.create(email='[e-mail]')
    User.objects.create(email='[e-mail]')
    alice = User.objects.create(email='[e-mail]')
    User.objects.create(email='[e-mail]')

    Report.objects.create(user=anne)
    Report.objects.create(user=alice)
    Report.objects.create(user=alice)

    logging.info('users without reports')
    logging.info(User.objects.filter(report__isnull=True, email__startswith='a'))

    logging.info('users with reports (allows duplicates)')
    logging.info(User.objects.filter(report__isnull=False, email__startswith='a'))

    logging.info('users with reports (no duplicates)')
    logging.info(User.objects.exclude(report__isnull=True).filter(email__startswith='a'))

    logging.info('users with reports (no duplicates, simpler SQL)')
    report_user_ids = Report.objects.values('user_id')
    logging.info(User.objects.filter(id__in=report_user_ids, email__startswith='a'))

    logging.info('Done.')


def setup():
    db_file = NAME + '.db'
    with open(db_file, 'w'):
        pass  # wipe the database
    settings.configure(
        DEBUG=True,
        DATABASES={
            DEFAULT_DB_ALIAS: {
                'ENGINE': 'django.db.backends.sqlite3',
                'NAME': db_file}},
        LOGGING={'version': 1,
                 'disable_existing_loggers': False,
                 'formatters': {
                    'debug': {
                        'format': '%(asctime)s[%(levelname)s]'
                                  '%(name)s.%(funcName)s(): %(message)s',
                        'datefmt': '%Y-%m-%d %H:%M:%S'}},
                 'handlers': {
                    'console': {
                        'level': 'DEBUG',
                        'class': 'logging.StreamHandler',
                        'formatter': 'debug'}},
                 'root': {
                    'handlers': ['console'],
                    'level': 'INFO'},
                 'loggers': {
                    "django.db": {"level": "DEBUG"}}})
    app_config = AppConfig(NAME, sys.modules['__main__'])
    apps.populate([app_config])
    django.setup()
    original_new_func = ModelBase.__new__

    # noinspection PyDecorator
    @staticmethod
    def patched_new(cls, name, bases, attrs):
        if 'Meta' not in attrs:
            class Meta:
                app_label = NAME
            attrs['Meta'] = Meta
        return original_new_func(cls, name, bases, attrs)
    ModelBase.__new__ = patched_new


def syncdb(model):
    """ Standard syncdb expects models to be in reliable locations.

    Based on https://github.com/django/django/blob/1.9.3
    /django/core/management/commands/migrate.py#L285
    """
    connection = connections[DEFAULT_DB_ALIAS]
    with connection.schema_editor() as editor:
        editor.create_model(model)

main()

If you put that into a Python file and run it, you should see something like this:

2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) PRAGMA foreign_keys; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) PRAGMA foreign_keys = 0; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.schema.execute(): CREATE TABLE "udjango_user" ("id" integer NOT NULL PRIMARY KEY AUTOINCREMENT, "email" varchar(254) NOT NULL); (params None)
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) CREATE TABLE "udjango_user" ("id" integer NOT NULL PRIMARY KEY AUTOINCREMENT, "email" varchar(254) NOT NULL); args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) PRAGMA foreign_keys = 0; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) PRAGMA foreign_keys; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) PRAGMA foreign_keys = 0; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.schema.execute(): CREATE TABLE "udjango_report" ("id" integer NOT NULL PRIMARY KEY AUTOINCREMENT, "user_id" integer NOT NULL REFERENCES "udjango_user" ("id")); (params None)
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) CREATE TABLE "udjango_report" ("id" integer NOT NULL PRIMARY KEY AUTOINCREMENT, "user_id" integer NOT NULL REFERENCES "udjango_user" ("id")); args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.schema.execute(): CREATE INDEX "udjango_report_e8701ad4" ON "udjango_report" ("user_id"); (params [])
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) CREATE INDEX "udjango_report_e8701ad4" ON "udjango_report" ("user_id"); args=[]
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) PRAGMA foreign_keys = 0; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) INSERT INTO "udjango_user" ("email") VALUES ('[e-mail]'); args=['[e-mail]']
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) INSERT INTO "udjango_user" ("email") VALUES ('[e-mail]'); args=['[e-mail]']
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) INSERT INTO "udjango_user" ("email") VALUES ('[e-mail]'); args=['[e-mail]']
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) INSERT INTO "udjango_user" ("email") VALUES ('[e-mail]'); args=['[e-mail]']
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) INSERT INTO "udjango_report" ("user_id") VALUES (1); args=[1]
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) INSERT INTO "udjango_report" ("user_id") VALUES (3); args=[3]
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) BEGIN; args=None
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) INSERT INTO "udjango_report" ("user_id") VALUES (3); args=[3]
2017-10-06 09:56:22[INFO]root.main(): users without reports
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) SELECT "udjango_user"."id", "udjango_user"."email" FROM "udjango_user" LEFT OUTER JOIN "udjango_report" ON ("udjango_user"."id" = "udjango_report"."user_id") WHERE ("udjango_report"."id" IS NULL AND "udjango_user"."email" LIKE 'a%' ESCAPE '\') LIMIT 21; args=(u'a%',)
2017-10-06 09:56:22[INFO]root.main(): [User(u'[e-mail]')]
2017-10-06 09:56:22[INFO]root.main(): users with reports (allows duplicates)
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) SELECT "udjango_user"."id", "udjango_user"."email" FROM "udjango_user" INNER JOIN "udjango_report" ON ("udjango_user"."id" = "udjango_report"."user_id") WHERE ("udjango_report"."id" IS NOT NULL AND "udjango_user"."email" LIKE 'a%' ESCAPE '\') LIMIT 21; args=(u'a%',)
2017-10-06 09:56:22[INFO]root.main(): [User(u'[e-mail]'), User(u'[e-mail]'), User(u'[e-mail]')]
2017-10-06 09:56:22[INFO]root.main(): users with reports (no duplicates)
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) SELECT "udjango_user"."id", "udjango_user"."email" FROM "udjango_user" WHERE (NOT ("udjango_user"."id" IN (SELECT U0."id" AS Col1 FROM "udjango_user" U0 LEFT OUTER JOIN "udjango_report" U1 ON (U0."id" = U1."user_id") WHERE U1."id" IS NULL)) AND "udjango_user"."email" LIKE 'a%' ESCAPE '\') LIMIT 21; args=(u'a%',)
2017-10-06 09:56:22[INFO]root.main(): [User(u'[e-mail]'), User(u'[e-mail]')]
2017-10-06 09:56:22[INFO]root.main(): users with reports (no duplicates, simpler SQL)
2017-10-06 09:56:22[DEBUG]django.db.backends.execute(): (0.000) SELECT "udjango_user"."id", "udjango_user"."email" FROM "udjango_user" WHERE ("udjango_user"."email" LIKE 'a%' ESCAPE '\' AND "udjango_user"."id" IN (SELECT U0."user_id" FROM "udjango_report" U0)) LIMIT 21; args=(u'a%',)
2017-10-06 09:56:22[INFO]root.main(): [User(u'[e-mail]'), User(u'[e-mail]')]
2017-10-06 09:56:22[INFO]root.main(): Done.

You can see that the final query uses all inner joins.

Answer 5

Use isnull.

users_without_reports = User.objects.filter(report__isnull=True)
users_with_reports = User.objects.filter(report__isnull=False).distinct()

When you use isnull=False, the distinct() is required to prevent duplicate results.