django-adv-cache-tag

Django advanced cache template tag:

• versioning
• compress
• partial caching
• easily extendable/customizable

Introduction

First, notice that the arguments of the {% cache %} templatetag provided by django-adv-cache-tag are the same as for the default cache templatetag included in django, so it's very easy to use this new one.

With django-adv-cache-tag you can :

• add a version number (int, string, date or whatever, it will be stringified) to your templatetag : the version will be compared to the cached one, and the exact same cache key will be used for the new cached template, avoiding keeping old unused keys in your cache, allowing you to cache forever.
• avoid to be afraid of an incompatible update in our algorithm, because we also use an internal version number, updated only when the internal algorithm changes
• define your own cache keys (or simply, just add the primary key (or what you want, it's a templatetag parameter) to this cache key
• compress the data to be cached, to reduce memory consumption in your cache backend, and network latency (but it will use more time and cpu to compress/decompress, your choice)
• choose which cache backend will be used
• define {% nocache %}...{% endnocache %} blocks inside your cached template, that will only be rendered when asked (for these parts, the content of the template is cached, not the rendered result)
• easily define your own algorithm, as we provide a single class (with short methods) you can inherit from, and simply change options or whatever behavior you want, and define your own tags for them
• use a variable for the name of your cache fragment

Installation

django-adv-cache-tag is available on PyPI:

pip install django-adv-cache-tag

Or you can find it on Github: https://github.com/twidi/django-adv-cache-tag

When installed, just add adv_cache_tag to your INSTALLED_APPS in the settings.py file of your django project.

See examples in the next sections to see how it works (basically the same way as the default django cache templatetag)

Features

Versioning

Description

With the default django cache templatetag, you can add as many arguments as you want, including a version, or date, and then the cache key will change if this version change. So your cache is updated, as expected.

But the older key is not deleted and if you have a long expire time, it will stay there for a very long time, consuming your precious memory.

django-adv-cache-tag provides a way to avoid this, while still regenerating the cache when needed. For this, when activated, we use the last argument passed to your templatetag as a "version number", and remove it for the arguments used to generate the cache key.

This version will be used in the content of the cached template, instead of the key, and when the cache exists and is loaded, the cached version will be compared to the wanted one, and if the two match, the cache is valid and returned, else it will be regenerated.

So if you like the principle of a unique key for a given template for a given object/user or whatever, be sure to always use the same arguments, except the last one, and activate the ADV_CACHE_VERSIONING setting.

Note that we also manage an internal version number, which will always be compared to the cached one. This internal version number is only updated when the internal algorithm of django-adv-cache-tag changes. But you can update it to invalidate all cached templates by adding a ADV_CACHE_VERSION to your settings (our internal version and the value from this settings will be concatenated to get the internal version really used)

Settings

ADV_CACHE_VERSIONING, default to False

ADV_CACHE_VERSION, default to ""

Example

In the following template, if ADV_CACHE_VERSIONING is set to True, the key will always be the same, based on the string "myobj_main_template" and the value of obj.pk, but the cached value will be regenerated each time the obj.date_last_updated will change.

So we set a expire_time of 0, to always keep the template cached, because we now we won't have many copies (old ones and current one) of it.

The value to set to have no expiry may depend of your cache backend (it's not always 0).

{% load adv_cache %}
{% cache 0 myobj_main_template obj.pk obj.date_last_updated %}
  {{ obj }}
{% endcache %}

Primary key

Description

In the default django cache templatetag, the cache keys are like this one :

:1:template.cache.your_fragment_name.64223ccf70bbb65a3a4aceac37e21016

You may want to have more explicit cache keys, so with django-adv-cache-tag you can add a "primary key" that will be added between the fragment name and the hash :

:1:template.cache.your_fragment_name.your_pk.64223ccf70bbb65a3a4aceac37e21016

Although the main use of this primary key is to have one cached fragment per object, so we can use the object primary key, you can use whatever you want, an id, a string...

To add a primary key, simply set the ADV_CACHE_INCLUDE_PK setting to True, and the first argument (after the fragment's name) will be used as a pk.

If you want this only for a part of your cache templatetags, read the Extending the default cache tag part later in this readme (it's easy, really).

Unlike the version, the primary key will be kept as an argument to generate the cache key hash.

Settings

ADV_CACHE_INCLUDE_PK, default to False

Example

A common use of django-adv-cache-tag is to only use a primary key and a version:

{% cache 0 myobj_main_template obj.pk obj.date_last_updated %}

Compression

Description

The default django cache templatetag simply saves the generated html in the cache. Depending of your template, if may be a lot of html and your cache memory will grow very quickly. Not to mention that we can have a lot of spaces because of indentation in templates (two ways i know to remove them without django-adv-cache-tag: the {% spaceless %} templatetag, provided by django, and django-template-preprocessor).

django-adv-cache-tag can do this for you. It is able to remove duplicate spaces (including newlines, tabs) by replacing them by a simple space (to keep the space behavior in html), and to compress the html to be cached, via the zlib (and pickle) module.

Of course, this cost some time and CPU cycles, but you can save a lot of memory in your cache backend, and a lot of bandwidth, especially if your backend is on a distant place. I haven't done any test for this, but for some templates, the saved data can be reduced from 2 ko to less than one.

To activate these feature, simply set to True one or both of the settings defined below.

WARNING : If the cache backend used use pickle and its default protocol, compression is useless because binary is not really well handled and the final size stored in the cache will be largely bigger than the compressed one. So check for this before activating this option. It's ok for the default django backends (at least in 1.4), but not for django-redis-cache, waiting for my pull-request, but you can check my own version: https://github.com/twidi/django-redis-cache/tree/pickle\_version

Settings

ADV_CACHE_COMPRESS, default to False, to activate the compression via zlib

ADV_CACHE_COMPRESS_LEVEL, default to -1, to set le compression level for zlib (actually the default is zlib.Z_DEFAULT_COMPRESSION, which is -1, that will be in fact 6 as the actual default defined in zlib)

ADV_CACHE_COMPRESS_SPACES, default to False, to activate the reduction of blank characters.

Example

No example since you don't have to change anything to your templatetag call to use this, just set the settings.

Choose your cache backend

Description

In django, you can define many cache backends. But with the default cache templatetag, you cannot say which one use, it will automatically be the default one.

django-adv-cache-tag can do this for your by providing a setting, ADV_CACHE_BACKEND which will take the name of a cache backend defined in your settings. And by extending the provided CacheTag object, you can even define many backends to be used by many templatetags, say one for heavily accessed templates, one for the others... as you want. Read the Extending the default cache tag part to know more about this (it's easy, really, but i already told you...)

Settings

ADV_CACHE_BACKEND, default to "default"

Example

No example since, like for the compression, you don't have to change anything to your templatetag to use this, just set the setting.

Partial caching

With the default django cache templatetag, your templates are cached and you can't update them before display, so you can't cache big parts of html with a little dynamic fragment in it, for the user name, the current date or whatever. You can cheat and save two templates surrounding your dynamic part, but you will have more accesses to your cache backend.

django-adv-cache-tag allow the use of one or many {% nocache %} blocks (closed by {% endnocache %}) to put in your {% cache %} blocks. These {% nocache %} block will be saved "as is" in the cache, while the rest of the block will be rendered to html. It's only when the template is finally displayed that the no-cached parts will be rendered.

You can have as many of these blocks you want.

Settings

There is no settings for this feature, which is automatically activated.

Example

{% cache 0 myobj_main_template obj.pk obj.date_last_updated %}
    <p>This is the cached part of the template for {{ obj }}, evaluated at {% now "r" %}.</p>
    {% nocache %}
        <p>This part will be evaluated each time : {% now "r" %}</p>
    {% endnocache %}
    <p>This is another cached part</p>
{% endcache %}

The fragment name

Description

The fragment name is the name to use as a base to create the cache key, and is defined just after the expiry time.

The Django documentation states The name will be taken as is, do not use a variable.

In django-adv-cache-tag, by setting ADV_CACHE_RESOLVE_NAME to True, a fragment name that is not quoted will be resolved as a variable that should be in the context.

Settings

ADV_CACHE_RESOLVE_NAME, default to False

Example

With ADV_CACHE_RESOLVE_NAME set to True, you can do this if you have a variable named fragment_name in your context:

{% cache 0 fragment_name obj.pk obj.date_last_updated %}

And if you want to pass a name, you have to surround it by quotes:

{% cache 0 "myobj_main_template" obj.pk obj.date_last_updated %}

With ADV_CACHE_RESOLVE_NAME set to False, the default, the name is always seen as a string, but if surrounded by quotes, they are removed.

In the following example, you see double-quotes, but it would be the same with single quotes, or no quotes at all:

{% cache 0 "myobj_main_template" obj.pk obj.date_last_updated %}

Extending the default cache tag

If the five settings explained in the previous sections are not enough for you, or if you want to have a templatetag with a different behavior as the default provided ones, you will be happy to know that django-adv-cache-tag was written with easily extending in mind.

It provides a class, CacheTag (in adv_cache_tag.tag), which has a lot of short and simple methods, and even a Meta class (idea stolen from the django models :D ). So it's easy to override a simple part.

All options defined in the Meta class are accessible in the class via self.options.some_field

Below we will show many ways of extending this class.

Basic override

Imagine you don't want to change the default settings (all to False, and using the default backend) but want a templatetag with versioning activated :

Create a new templatetag file (myapp/templatetags/my_cache_tags.py) with this:

from adv_cache_tag.tag import CacheTag

class VersionedCacheTag(CacheTag):
    class Meta(CacheTag.Meta):
        versioning = True

from django import template
register = template.Library()

VersionedCacheTag.register(register, 'ver_cache')

With these simple lines, you now have a new templatetag to use when you want versioning:

{% load my_cache_tags %}
{% ver_cache 0 myobj_main_template obj.pk obj.date_last_updated %}
    obj
{% endver_cache %}

As you see, just replace {% load adv_cache %} (or the django default {% load cache %}) by {% load my_cache_tags %} (your templatetag module), and the {% cache %} templatetag by your new defined one, {% ver_cache ... %}. Don't forget to replace the closing tag too: {% endver_cache %}. But the {% nocache %} will stay the same, except if you want a new one. For this, just add a parameter to the register method:

MyCacheTag.register(register, 'ver_cache', 'ver_nocache')

{% ver_cache ... %}
    cached
    {% ver_nocache %}not cached{% endver_nocache %}
{% endver_cache %}

Note that you can keep the name cache for your tag if you know that you will not load in your template another templatetag module providing a cache tag. To do so, the simplest way is:

MyCacheTag.register(register)  # 'cache' and 'nocache' are the default values

All the django-adv-cache-tag settings have a matching variable in the Meta class, so you can override one or many of them in your own classes. See the "Settings" part to see them.

Internal version

When your template file is updated, the only way to invalidate all cached versions of this template is to update the fragment name or the arguments passed to the templatetag.

With django-adv-cache-tag you can do this with versioning, by managing your own version as the last argument to the templatetag. But if you want to use the power of the versioning system of django-adv-cache-tag, it can be too verbose:

{% load adv_cache %}
{% with template_version=obj.date_last_updated|stringformat:"s"|add:"v1" %}
    {% cache 0 myobj_main_template obj.pk template_version %}
    ...
    {% endcache %}
{% endwith %}

django-adv-cache-tag provides a way to do this easily, with the ADV_CACHE_VERSION setting. But by updating it, all cached versions will be invalidated, not only those you updated.

To do this, simply create your own tag with a specific internal version:

class MyCacheTag(CacheTag):
    class Meta(CacheTag.Meta):
       internal_version = "v1"

MyCacheTag.register('my_cache')

And then in your template, you can simply do

{% load my_cache_tags %}
{% my_cache 0 myobj_main_template obj.pk obj.date_last_updated %}
...
{% endmy_cache %}

Each time you update the content of your template and want invalidation, simply change the internal_version in your MyCacheTag class (or you can use a settings for this).

Change the cache backend

If you want to change the cache backend for one templatetag, it's easy:

class MyCacheTag(CacheTag):
    class Meta:
        cache_backend = 'templates'

But you can also to this by overriding a method:

from django.core.cache import get_cache

class MyCacheTag(CacheTag):
    def get_cache_object(self):
        return get_cache('templates')

And if you want a cache backend for old objects, and another, faster, for recent ones:

from django.core.cache import get_cache

class MyCacheTag(CacheTag):
    class Meta:
        cache_backend = 'fast_templates'

    def get_cache_object(self):
        cache_backend = self.options.cache_backend
        if self.get_pk() < 1000:
            cache_backend = 'slow_templates'
        return get_cache(cache_backend)

The value returned by the get_cache_object should be a cache backend object, but as we only use the set and get methods on this object, it can be what you want if it provides these two methods. And even more, you can override the cache_set and cache_get methods of the CacheTag class if you don't want to use the default set and get methods of the cache backend object.

Note that we also support the django way of changing the cache backend in the template-tag, using the using argument, to be set at the last parameter (without any space between using and the name of the cache backend).

{% cache 0 myobj_main_template obj.pk obj.date_last_updated using=foo %}

Change the cache key

The CacheTag class provides three classes to create the cache key:

• get_base_cache_key, which returns a formatable string ("template.%(nodename)s.%(name)s.%(pk)s.%(hash)s" by default if include_pk is True or "template.%(nodename)s.%(name)s.%(hash)s" if False
• get_cache_key_args, which returns the arguments to use in the previous string
• get_cache_key, which combine the two

The arguments are:

• nodename parameter is the name of the templatetag: it's "my_cache" in {% my_cache ... %}
• name is the "fragment name" of your templatetag, the value after the expire-time
• pk is used only if self.options.include_pk is True, and is returned by this.get_pk()
• hash is the hash of all arguments after the fragment name, excluding the last one which is the version number (this exclusion occurs only if self.options.versioning is True)

If you want to remove the "template." part at the start of the cache key (useless if you have a cache backend dedicated to template caching), you can do this:

class MyCacheTag(CacheTag):
    def get_base_cache_key(self):
        cache_key = super(MyCacheTag, self).get_base_cache_key()
        return cache_key[len('template:'):]  # or [9:]

Add an argument to the templatetag

By default, the templatetag provided by CacheTag takes the same arguments as the default django cache templatetag.

If you want to add an argument, it's easy as the class provides a get_template_node_arguments method, which will work as for normal django templatetags, taking a list of tokens, and returning ones that will be passed to the real templatetag, a Node class tied to the CacheTag.

Say you want to add a foo argument between the expire time and the fragment name:

from django import template

from adv_cache_tag.tag import CacheTag, Node

class MyNode(Node):
    def __init__(self, nodename, nodelist, expire_time, foo, fragment_name, vary_on):
        """ Save the foo variable in the node (not resolved yet) """
        super(MyNode, self).__init__(self, nodename, nodelist, expire_time, fragment_name, vary_on)
        self.foo = foo


class MyCacheTag(CacheTag):

    Node = MyNode

    def prepare_params(self):
        """ Resolve the foo variable to it's real content """
        super(MyCacheTag, self).prepare_params()
        self.foo = template.Variable(self.node.foo).resolve(self.context)

    @classmethod
    def get_template_node_arguments(cls, tokens):
        """ Check validity of tokens and return them as ready to be passed to the Node class """
        if len(tokens) < 4:
            raise template.TemplateSyntaxError(u"'%r' tag requires at least 3 arguments." % tokens[0])
        return (tokens[1], tokens[2], tokens[3], tokens[4:])

Prepare caching of templates

This one is not about overriding the class, but it can be useful. When an object is updated, it can be better to regenerate the cached template at this moment rather than when we need to display it.

It's easy. You can do this by catching the post_save signal of your model, or just by overriding its save method. For this example we will use this last solution.

The only special thing is to know the path of the template where your templatetag is. In my case, i have a template just for this (included in other ones for general use), so it's easier to find it and regenerate it as in this example.

As we are not in a request, we don't have the Request object here, so context processors are not working, we must create a context object that will be used to render the template, with all needed variables.

from django.template import loader, Context

class MyModel(models.Model):
    # your fields

    def save(self, *args, **kwargs):
        super(MyModel, self.save(*args, **kwargs)

        template = 'path/to/my_template_file_with_my_cache_block.html'

        context = Context({
            'obj': self,

            # as you have no request, we have to add stuff from context processors manually if we need them
            'STATIC_URL': settings.STATIC_URL,

            # the line below indicates that we force regenerating the cache, even if it exists
            '__regenerate__': True,

            # the line below indicates if we only want html, without parsing the nocache parts
            '__partial__': True,

        })

        loader.get_template(template).render(context)

Load data from database before rendering

This is a special case. Say you want to display a list of objects but you have only ids and versions retrieved from redis (with ZSET, with id as value and updated date (which is used as a version) as score , for example)

If you know you always have a valid version of your template in cache, because they are regenerated every time they are saved, as seen above, it's fine, just add the object's primary key as the pk in your templatetag arguments, and the cached template will be loaded.

But if it's not the case, you will have a problem: when django will render the template, the only part of the object present in the context is the primary key, so if you need the name or whatever field to render the cached template, it won't work.

With django-adv-cache-tag it's easy to resolve this, as we can load the object from the database and adding it to the context.

View

def my_view(request):
    objects = [
        dict(
            pk=val[0],
            date_last_updated=val[1]
        )
        for val in
            redis.zrevrange('my_objects', 0, 19, withscores=True)
    ]
    return render(request, "my_results.html", dict(objects=objects))

Template "my_results.html"

{% for obj in objects %}
    {% include "my_result.html" %}
{% endfor %}

Template "my_result.html"

{% load my_cache_tags %}
{% my_cache 0 myobj_main_template obj.pk obj.date_last_updated %}
    {{ obj }}
{% endmy_cache %}

Templatetag

In myapp/templatetags/my_cache_tags

from my_app.models import MyModel

class MyCacheTag(CacheTag):

    class Meta(CacheTag.Meta):
        """ Force options """
        include_pk = True
        versioning = True

    def create_content(self):
        """ If the object in context is not a real model, load it from db """
        if not isinstance(context['obj'], MyObject):
            context['obj'] = MyModel.objects.get(id=self.get_pk())
        super(MyCacheTag, self).create_content()

MyCacheTag.register('my_cache')

Careful with this, it generates as database requests as objects to be loaded.

And more...

If you want to do more, feel free to look at the source code of the CacheTag class (in tag.py), all methods are documented.

Settings

django-adv-cache-tag provide 5 settings you can change. Here is the list, with descriptions, default values, and corresponding fields in the Meta class (accessible via self.options.some_field in the CacheTag object)

• ADV_CACHE_VERSIONING to activate versioning, default to False (versioning in the Meta class)
• ADV_CACHE_COMPRESS to activate compression, default to False (compress in the Meta class)
• ADV_CACHE_COMPRESS_LEVEL to set the compression level (from 1 (min compression) to 9 (max compression), default to -1 (equivalent to 6) (compress_level in the Meta class)
• ADV_CACHE_COMPRESS_SPACES to activate spaces compression, default to False (compress_spaces in the Meta class)
• ADV_CACHE_INCLUDE_PK to activate the "primary key" feature, default to False (include_pk in the Meta class)
• ADV_CACHE_BACKEND to choose the cache backend to use, default to "default" (cache_backend in the Meta class)
• ADV_CACHE_VERSION to create your own internal version (will be concatenated to the real internal version of django-adv-cache-tag), default to "" (internal_version in the Meta class)

How it works

Here is a quick overview on how things work in django-adv-cache-tag

Partial caching

Your template :

{% load adv_cache %}
{% cache ... %}
    foo
    {% nocache %}
        bar
    {% endnocache %}
    baz
{% endcache %}

Cached version (we ignore versioning and compress here, just to see how it works):

foo
{% endRAW_xyz %}
    bar
{% RAW_xyz %}
baz

When cached version is loaded, we parse :

{% RAW_xyz %}
foo
{% endRAW_xyz %}
    bar
{% RAW_xyz %}
baz
{% endRAW_xyz %}

The first {% RAW_xyz %} and the last {% endRAW_xyz %} are not included in the cached version and added before parsing, only to save some bytes.

Parts between {% RAW_xyz %} and {% endRAW_xyz %} are not parsed at all (seen as a TextNode by django)

The xyz part of the RAW and endRAW templatetags depends on the SECRET_KEY and so is unique for a given site.

It allows to avoid at max the possible collisions with parsed content in the cached version.

We could have used {% nocache %} and {% endnocache %} instead of {% RAW_xyz %} and {% endRAW_xyz %} but in the parsed template, stored in the cache, if the html includes one of these strings, our final template would be broken, so we use long ones with a hash (but we can not be sure at 100% these strings could not be in the cached html, but for common usages it should suffice)

License

django-adv-cache-tag is published under the MIT License (see the LICENSE file)

Running tests

poetry install
poetry run black --check adv_cache_tag
poetry run ruff adv_cache_tag
export DJANGO_SETTINGS_MODULE=adv_cache_tag.tests.testproject.settings
poetry run pytest

Supported versions

Django version	Python version	Library version
3.2	3.8 to 3.10	>=1.1.5
4.1	3.8 to 3.11	>=1.1.5
4.2	3.8 to 3.11	>=1.1.5