Skip to main content

Explicit control over query execution in Django applications.

Project description


Build Status pypi release

Gives you control over which parts of your code are allowed to run queries, and which aren't.

Tested against Django 2.2 - 3.1 on Python 3.6 - 3.9.


Explicit is better than implicit

(The Zen Of Python)

The greatest strength of Django's ORM is also its greatest weakness. By freeing developers from having to think about when database queries are run, the ORM encourages developers to not think about when database queries are run! This often has great benefits for quick development turnaround, but can have major performance implications in anything other than trivially simple systems.

Django's ORM makes queries implicit. The Zen of Python tells us that explicit is better than implicit, so let's be explicit about which parts of our code are allowed to run queries, and which aren't.

Check out this blog post for more background.


Imagine a pizza restaurant website with the following models:

class Topping(models.Model):
    name = models.CharField(max_length=100)

class Pizza(models.Model):
    name = models.CharField(max_length=100)
    toppings = models.ManyToManyField(Topping)

And here's the menu view:

def menu(request):
    pizzas = Pizza.objects.all()
    context = {'pizzas': pizzas}
    return render(request, 'pizzas/menu.html', context)

Finally, the template:

<h1>Pizza Menu</h1>

{% for pizza in pizzas %}
  <li>{{ }}</li>
{% endfor %}

How many queries are run here? Well, the answer is easy to see: it's just one! The query emitted by Pizza.objects.all() is all you need to get the information to show on the menu.

Now: imagine the client asks for each pizza on the menu to include a count of how many toppings are on the pizza. Easy! Just change the template:

<h1>Pizza Menu</h1>

{% for pizza in pizzas %}
  <li>{{ }} ({{ pizza.toppings.count }})</li>
{% endfor %}

But how many queries are run now? Well, this is the classic n queries problem. We now have one query to get all our pizzas, and then another query per pizza to get the toppings count. The more pizzas we have, the slower the app gets. And we probably won't discover this until the website is in production.

If you were reading a Django performance tutorial, the next step would be to tell you how to fix this problem (.annotate and Count etc). But that's not the point. The example above is just an illustration of how code in different parts of the codebase, at different levels of abstraction, even possibly (in larger projects) the responsibility of different developers, can interact to result in poor performance. Object-oriented design encourages black-box implementation hiding, but hiding the points at which queries are executed is the worst thing you can do if your aim is to build high-performance web applications. So how do we fix this without breaking all our abstractions?

There are two tricks here:

  1. Prevent developers from accidentally running queries without realising.
  2. Encourage code design that separates fetching data from rendering data.

This package provides three very simple things:

  1. A context manager to allow developers to be explicit about where queries are run.
  2. A utility to make querysets less lazy.
  3. Some tools to make it easy to use the context manager with Django templates and Django REST framework serializers.

To be absolutely clear: this package does not give you any tools to actually improve your query patterns. It just tells you when you need to do it!


To demonstrate how to use django-zen-queries, let's go back to our example. We want to make it impossible for changes to a template to trigger queries. So, we change our view as follows:

def menu(request):
    pizzas = Pizza.objects.all()
    context = {'pizzas': pizzas}
    with queries_disabled():
        return render(request, 'pizzas/menu.html', context)

The queries_disabled context manager here does one very simple thing: it stops any code inside it from running database queries. At all. If they try to run a query, the application will raise a QueriesDisabledError exception and blow up.

That's almost enough to give us what we need, but not quite. The code above will always raise a QueriesDisabledError, because the queryset (Pizza.objects.all()) is lazy. The database query doesn't actually get run until the queryset is iterated - which happens in the template! So, django-zen-queries provides a tiny helper function, fetch, which forces evaluation of a queryset:

def menu(request):
    pizzas = Pizza.objects.all()
    context = {'pizzas': fetch(pizzas)}
    with queries_disabled():
        return render(request, 'pizzas/menu.html', context)

Now we have exactly what we need: when a developer comes along and adds {{ pizza.toppings.count }} in the template, it just won't work. They will be forced to figure out how to use annotate and Count in order to get the data they need up front, rather than sometime in the future when customers are complaining that the website is getting slower and slower!


You can also use queries_disabled as a decorator to prohibit database interactions for a whole function or method:

def validate_xyz(pizzas):

This also works with Django's method_decorator utility.

Extra tools

As well as the context managers, the package provides some tools to make it easier to use in common situations:

Render shortcut

If you're using the Django render shortcut (as in the example above), to avoid having to add the context manager to every view, you can change your import from django.shortcuts import render to from zen_queries import render. All the views in that file will automatically be disallowed from running queries during template rendering.

TemplateResponse subclass

TemplateResponse (and SimpleTemplateResponse) objects are lazy, meaning that template rendering happens on the way "out" of the Django stack. zen_queries.TemplateResponse and zen_queries.SimpleTemplateResponse are subclasses of these with queries_disabled applied to the render method.

You can tell Django's class-based views to use these subclasses instead of the default TemplateResponse by setting the response_class attribute on the view to zen_queries.TemplateResponse.

Django REST framework Serializer and View mixins

Django REST framework serializers are another major source of unexpected queries. Adding a field to a serializer (perhaps deep within a tree of nested serializers) can very easily cause your application to suddenly start emitting hundreds of queries. zen_queries.rest_framework.QueriesDisabledSerializerMixin can be added to any serializer to wrap queries_disabled around the .data property, meaning that the serialization phase is not allowed to execute any queries.

You can add this mixin to an existing serializer instance with zen_queries.rest_framework.disable_serializer_queries like this: serializer = disable_serializer_queries(serializer).

If you're using REST framework generic views, you can also add a view mixin, zen_queries.rest_framework.QueriesDisabledViewMixin, which overrides get_serializer to mix the QueriesDisabledSerializerMixin into your existing serializer. This is useful because you may want to use the same serializer class between multiple views but only disable queries in some contexts, such as in a list view. Remember that Python MRO is left-right, so the mixin must come before (to the left of) any base classes that implement get_serializer. The view mixin only disables queries on GET requests, so can safely be used with ListCreateAPIView and similar.

Escape hatch

If you absolutely definitely can't avoid running a query in a part of your codebase that's being executed under a queries_disabled block, there is another context manager called queries_dangerously_enabled which allows you to temporarily re-enable database queries.

Template Tags

Block tags for Django's template system are provided which allow you to enable or disable query execution directly in your templates.

Important note: In order to use the template libary, you must add "zen_queries" to your INSTALLED_APPS setting. Then, use {% load zen_queries %} at the top of your template to load the tag library.

The {% queries_disabled} tag is most useful if you wish to apply django-zen-queries patterns to a third-party library which provides customisation via overriding templates, such as the Django admin.

{% load zen_queries %}

{% queries_disabled %}
{% for pizza in pizzas %}
  <li>{{ }}</li>
{% endfor %}
{% end_queries_disabled %}

The {% queries_dangerously_enabled %} tag is handy if you are using the render shortcut or TemplateResponse subclass (see above) but wish to allow particular parts of your templates to execute queries. This should be used with caution, and you should wrap only the smallest possible sections of your template: the precise line or lines that need to execute the queries.

{% load zen_queries %}

{% queries_dangerously_enabled %}
There are {{ pizzas.count }} pizzas.
{% end_queries_dangerously_enabled %}

Permissions gotcha

Accessing permissions in your templates (via the {{ perms }} template variable) can be a source of queries at template-render time. Fortunately, Django's permission checks are cached by the ModelBackend, which can be pre-populated by calling request.user.get_all_permissions() in the view, before rendering the template.

How does it work?

It uses the Database Instrumentation features introduced in Django 2.0.


Install from PyPI

pip install django-zen-queries

Code of conduct

For guidelines regarding the code of conduct when contributing to this repository please review

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

django-zen-queries-2.1.0.tar.gz (10.4 kB view hashes)

Uploaded source

Built Distribution

django_zen_queries-2.1.0-py3-none-any.whl (12.4 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page