Bulk update using one query over Django ORM
Clone or download
Latest commit 399e64d Mar 15, 2018

README.md

django-bulk-update

Build Status Coverage Status

Simple bulk update over Django ORM or with helper function.

This project aims to bulk update given objects using one query over Django ORM.

Installation

pip install django-bulk-update

Usage

With manager:

import random
from django_bulk_update.manager import BulkUpdateManager
from tests.models import Person

class Person(models.Model):
    ...
    objects = BulkUpdateManager()

random_names = ['Walter', 'The Dude', 'Donny', 'Jesus']
people = Person.objects.all()
for person in people:
  person.name = random.choice(random_names)

Person.objects.bulk_update(people, update_fields=['name'])  # updates only name column
Person.objects.bulk_update(people, exclude_fields=['username'])  # updates all columns except username
Person.objects.bulk_update(people)  # updates all columns
Person.objects.bulk_update(people, batch_size=50000)  # updates all columns by 50000 sized chunks

With helper:

import random
from django_bulk_update.helper import bulk_update
from tests.models import Person

random_names = ['Walter', 'The Dude', 'Donny', 'Jesus']
people = Person.objects.all()
for person in people:
  person.name = random.choice(random_names)

bulk_update(people, update_fields=['name'])  # updates only name column
bulk_update(people, exclude_fields=['username'])  # updates all columns except username
bulk_update(people, using='someotherdb')  # updates all columns using the given db
bulk_update(people)  # updates all columns using the default db
bulk_update(people, batch_size=50000)  # updates all columns by 50000 sized chunks using the default db

Note: You can consider to use .only('name') when you only want to update name, so that Django will only retrieve name data from db.

And consider to use .defer('username') when you don't want to update username, so Django won't retrieve username from db. These optimization can improve the performance even more.

Performance Tests:

Here we test the performance of the bulk_update function vs. simply calling .save() on every object update (dmmy_update). The interesting metric is the speedup using the bulk_update function more than the actual raw times.

# Note: SQlite is unable to run the `timeit` tests
# due to the max number of sql variables
In [1]: import os
In [2]: import timeit
In [3]: import django

In [4]: os.environ['DJANGO_SETTINGS_MODULE'] = 'tests.test_settings'
In [5]: django.setup()

In [6]: from tests.fixtures import create_fixtures

In [7]: django.db.connection.creation.create_test_db()
In [8]: create_fixtures(1000)

In [9]: setup='''
import random
from django_bulk_update import helper
from tests.models import Person
random_names = ['Walter', 'The Dude', 'Donny', 'Jesus']
ids = list(Person.objects.values_list('id', flat=True)[:1000])
people = Person.objects.filter(id__in=ids)
for p in people:
    name = random.choice(random_names)
    p.name = name
    p.email = '%s@example.com' % name
bu_update = lambda: helper.bulk_update(people, update_fields=['name', 'email'])
'''

In [10]: bu_perf = min(timeit.Timer('bu_update()', setup=setup).repeat(7, 100))

In [11]: setup='''
import random
from tests.models import Person
from django.db.models import F
random_names = ['Walter', 'The Dude', 'Donny', 'Jesus']
ids = list(Person.objects.values_list('id', flat=True)[:1000])
people = Person.objects.filter(id__in=ids)
def dmmy_update():
    for p in people:
        name = random.choice(random_names)
        p.name = name
        p.email = '%s@example.com' % name
        p.save(update_fields=['name', 'email'])
'''

In [12]: dmmy_perf = min(timeit.Timer('dmmy_update()', setup=setup).repeat(7, 100))
In [13]: print 'Bulk update performance: %.2f. Dummy update performance: %.2f. Speedup: %.2f.' % (bu_perf, dmmy_perf, dmmy_perf / bu_perf)
Bulk update performance: 7.05. Dummy update performance: 373.12. Speedup: 52.90.

Requirements

  • Django 1.8+

Contributors

TODO

  • Geometry Fields support

License

django-bulk-update is released under the MIT License. See the LICENSE file for more details.