Skip to content

Bahus/easy_cache

Repository files navigation

Easy caching decorators

This package is intended to simplify caching and invalidation process in python-based (primarily) web applications. It's possible to cache execution results of functions; instance, class and static methods; properties. Cache keys may be constructed in various different ways and may depend on any number of parameters.

The package supports tag-based cache invalidation and better works with Django, however any other frameworks can be used – see examples below.

The main idea of this package: you don't need to touch any existing function code to cache its execution results.

Requirements

Library was tested in the following environments:

  • Python 3.7, 3.8, 3.9, 3.10
  • Django >=2.0.0

Feel free to try it in yours, but it's not guaranteed it will work. Submit an issue if you think it should.

Installation

pip install easy_cache

Introduction

Different ways to cache something

Imagine you have a time consuming function and you need to cache an execution results, the classic way to achieve this is the next one:

# classic way
from django.core.cache import cache

def time_consuming_operation(n):
    """Calculate sum of number from 1 to provided n"""
    cache_key = 'time_consuming_operation_{}'.format(n)
    result = cache.get(cache_key, None)

    if result is None:
        # not found in cache
        result = sum(range(n + 1))
        # cache result for one hour
        cache.set(cache_key, result, 3600)

    return result

def invalidate_cache(n):
    cache.delete('time_consuming_operation_{}'.format(n))

Well, we had to add annoying boilerplate code to achieve this. Now let's take a look how easy_cache can avoid the problem and simplify the code:

# easy way
from easy_cache import ecached

@ecached('time_consuming_operation_{n}', 3600)
def time_consuming_operation(n):
    return sum(range(n + 1))

def invalidate_cache(n):
    time_consuming_operation.invalidate_cache_by_key(n)

As we can see the function code left clear. Heart of the package is two decorators with the similar parameters:

ecached

Should be used to decorate any callable and cache returned result.

Parameters:

  • cache_key – cache key generator, default value is None so the key will be composed automatically based on a function name, namespace and passed parameters. Also the following types are supported:
    • string – may contain Python advanced string formatting syntax, a given value will be formatted with a dict of parameters passed to decorated function, see examples below.
    • sequence of strings – each string must be function parameter name.
    • callable – is used to generate cache key, decorated function parameters will be passed to this callable and returned value will be used as a cache key. Also one additional signature is available: callable(meta), where meta is a dict-like object with some additional attributes – see below.
  • timeout – value will be cached with provided timeout, basically it should be number of seconds, however it depends on cache backend type. Default value is DEFAULT_VALUE – internal constant means that actually no value is provided to cache backend and thus backend should decide what timeout to use. Callable is also supported.
  • tags – sequence of strings or callable. Should provide or return list of tags added to cached value so cache may be invalidated later with any tag name. Tag may support advanced string formatting syntax. See cache_key docs and examples for more details.
  • prefix – this parameter works both: as regular tag and also as cache key prefix, as usual advanced string formatting and callable are supported here.
  • cache_alias – cache backend alias name, it can also be Django cache backend alias name.
  • cache_instance – cache backend instance may be provided directly via this parameter.

ecached_property

Should be used to create so-called cached properties, has signature exactly the same as for ecached.

Simple examples

Code examples is the best way to show the power of this package.

Decorators can be simply used with default parameters only

from easy_cache import ecached, create_cache_key

# default parameters
# cache key will be generated automatically:
#
# <__module__>.<__class__>.<function name> + function parameters converted to strings,
#
# so be careful when using complex objects, it's
# better to write custom cache key generator in such cases.
#
# timeout will be default for specified cache backend
# "default" cache backend will be used if you use Django
@ecached()
def time_consuming_operation(*args, **kwargs):
    pass

# simple static cache key and cache timeout 100 seconds
@ecached('time_consuming_operation', 100)
def time_consuming_operation():
    pass

# cache key with advanced string formatting syntax
@ecached('my_key:{b}:{d}:{c}')
def time_consuming_operation(a, b, c=100, d='foo'):
    pass

# or
@ecached('key:{kwargs[param1]}:{kwargs[param2]}:{args[0]}')
def time_consuming_operation(*args, **kwargs):
    pass

# use specific cache alias, see "caches framework" below
from functools import partial

memcached = partial(ecached, cache_alias='memcached')

# equivalent to cache_key='{a}:{b}'
@memcached(['a', 'b'], timeout=600)
def time_consuming_operation(a, b, c='default'):
    pass

Using custom cache key generators

# working with parameters provided to cached function
# cache key generator must have the same signature as decorated function
from easy_cache import create_cache_key

def custom_cache_key(self, a, b, c, d):
    return create_cache_key(self.id, a, d)

# working with `meta` object
def custom_cache_key_meta(meta):
    return '{}:{}:{}'.format(meta['self'].id, meta['a'], meta['d'])

# or equivalent
from easy_cache import meta_accepted

@meta_accepted
def custom_cache_key_meta(parameter_with_any_name):
    meta = parameter_with_any_name
    return '{}:{}:{}'.format(meta['self'].id, meta['a'], meta['d'])


class A(object):
    id = 1

    @ecached(custom_cache_key)
    def time_consuming_operation(self, a, b, c=10, d=20):
        ...

    @ecached(custom_cache_key_meta)
    def time_consuming_operation(self, a, b, c=10, d=20):
        ...

How to cache staticmethod and classmethod correctly

# ecached decorator always comes topmost
class B(object):

    # cache only for each different year
    @ecached(lambda start_date: 'get_list:{}'.format(start_date.year))
    @staticmethod
    def get_list_by_date(start_date):
        ...

    CONST = 'abc'

    @ecached('info_cache:{cls.CONST}', 3600, cache_alias='redis_cache')
    @classmethod
    def get_info(cls):
        ...

MetaCallable object description

Meta object has the following parameters:

  • args – tuple with positional arguments provided to decorated function
  • kwargs – dictionary with keyword arguments provided to decorated function
  • returned_value – value returned from decorated function, available only when meta object is handled in tags or prefix generators. You have to check has_returned_value property before using this parameter:
def generate_cache_key(meta):
    if meta.has_returned_value:
        # ... do something with meta.returned_value ...
  • call_args - dictionary with all positional and keyword arguments provided to decorated function, you may also access them via __getitem__ dict interface, e. g. meta['param1'].
  • function - decorated callable
  • scope - object to which decorated callable is attached, None otherwise. Usually it's an instance or a class.

Tags invalidation, refresh and cached properties

Tags-based cache invalidation allows you to invalidate several cache keys at once.

Imagine you created a web-based book store and your users can mark a book as liked, so you need to maintain a list of liked books for every user but, an information about a book may contain a lot of different data, e.g. authors names, rating, availability in stock, some data from external services and so on.

Some of this information can be calculated on runtime only so you decided to cache the list of liked books.

But what if a book title was updated and we have to find all cache keys where this book is stored and invalidate them. Such task may be pretty complex to complete, however if you tagged all the necessary cache keys with a specific tag you will just need to invalidate the tag only and related cache keys will be invalidated "automatically".

Here are more complex examples introducing Django models and effective tags usage. Check code comments and doc-strings for detailed description.

from django.db import models
from easy_cache import ecached, ecached_property, create_cache_key


class Book(models.Model):
    title = models.CharField(max_length=250)

    def __unicode__(self):
        return self.title


class User(models.Model):
    name = models.CharField(max_length=100)
    state = models.CharField(
        max_length=15,
        choices=(('active', 'active'), ('deleted', 'deleted')),
    )
    friends = models.ManyToManyField('self', symmetrical=True)
    favorite_books = models.ManyToManyField('Book')

    def __unicode__(self):
        return self.name

    @ecached('users_by_state:{state}', 60, tags=['users_by_states'])
    @classmethod
    def get_users_by_state(cls, state):
        """
        Caches user list by provided state parameter: there will be separate
        cached value for every different state parameter, so we are having 2 different
        cache keys:

        users_by_state:active – cached list of active users
        users_by_state:deleted – cached list of deleted users

        Note that `ecached` decorator always comes topmost.

        To invalidate concrete cached state call the following method
        with the required `state`, e.g.:
        >>> User.get_users_by_state.invalidate_cache_by_key('active')
        ... removes `users_by_state:active` cache key
        or
        >>> User.get_users_by_state.invalidate_cache_by_key(state='deleted')
        ... removes `users_by_state:deleted` cache key

        If you'd like to invalidate all caches for all states call:
        >>> User.get_users_by_state.invalidate_cache_by_tags('users_by_states')
        ... removes both keys, since `users_by_states` tag attached to all of them,

        `invalidate_cache_by_tags` supports both string and list parameter types:
        >>> invalidate_cache_by_tags(['tag1', 'tag2', 'tag3'])

        To refresh concrete cached state call the following method
        with required `state`, e.g:
        >>> User.get_users_by_state.refresh_cache('active')
        ... calls `get_users_by_state('active')` and saves returned value to cache
        or
        >>> User.get_users_by_state.refresh_cache(state='deleted')

        """
        return list(cls.objects.filter(state=state))

    @ecached_property('user_friends_count:{self.id}', timeout=3600)
    def friends_count(self):
        """
        Caches friends count of each user for 1 hour.

        To access cache invalidation functions for a property you
        have to use class object instead of instance.

        Call the following method, to invalidate cache:
        >>> User.friends_count.invalidate_cache_by_key(user)
        ... removes cache key `user_friends_count:{user.id}`
        or
        >>> type(self).friends_count.invalidate_cache_by_key(user)
        or
        >>> self.__class__.friends_count.invalidate_cache_by_key(user)

        Where `user` is desired User instance to invalidate friends count for.

        Call the following method, to refresh cached data:
        >>> User.friends_count.refresh_cache(user)
        ... Updates `user.friends_count` in a cache.
        or
        >>> type(self).friends_count.refresh_cache(user)
        or
        >>> self.__class__.friends_count.refresh_cache(user)
        """
        return self.friends.count()

    @staticmethod
    def get_books_tags(meta):
        """
        Add one tag for every book in list of favorite books.
        So we will add a list of tags to cached favorite books list.
        """
        if not meta.has_returned_value:
            return []

        favorite_books = meta.returned_value
        # yes, it may occupy a lot of cache keys
        return [create_cache_key('book', book.pk) for book in favorite_books]

    @ecached('user_favorite_books:{self.id}', 600, get_books_tags)
    def get_favorite_books(self):
        """
        Caches list of related books by user id. So in code you will use:

        >>> favorite_books = request.user.get_favorite_books() # cached for user

        You may want to invalidate this cache in two cases:

        1. User added new book to favorites:

        >>> User.get_favorite_books.invalidate_cache_by_key(user)
        or
        >>> User.get_favorite_books.invalidate_cache_by_key(self=user)
        or
        >>> from easy_cache import invalidate_cache_key, create_cache_key
        >>> invalidate_cache_key(create_cache_key('user_favorite_books', user.id))
        or
        >>> invalidate_cache_key('user_favorite_books:{}'.format(user.id))

        2. Some information about favorite book was changed, e.g. its title:
        >>> from easy_cache import invalidate_cache_tags, create_tag_cache_key
        >>> tag_cache_key = create_tag_cache_key('book', changed_book_id)
        >>> User.get_favorite_books.invalidate_cache_by_tags(tag_cache_key)
        or
        >>> invalidate_cache_tags(tag_cache_key)

        To refresh cached values use the following patterns:
        >>> User.get_favorite_books.refresh_cache(user)
        or
        >>> User.get_favorite_books.refresh_cache(self=user)
        """
        return self.favorite_books.filter(user=self)

Prefix usage

Commonly prefix is used to invalidate all cache-keys in one namespace, e. g.:

from functools import partial

class Shop(models.Model):
    single_shop_cache = partial(ecached, prefix='shop:{self.id}')

    @single_shop_cache('goods_list')
    def get_all_goods_list(self):
        return [...]

    @single_shop_cache('prices_list')
    def get_all_prices_list(self):
        return [...]

# if you have `shop` object you are able to use the following invalidation
# strategies:

# Invalidate cached list of goods for concrete shop
Shop.get_all_goods_list.invalidate_cache_by_key(shop)

# Refresh cached list of goods for concrete shop
Shop.get_all_goods_list.refresh_cache(shop)

# Invalidate cached list of prices for concrete shop
Shop.get_all_prices_list.invalidate_cache_by_key(shop)

# Refresh cached list of prices for concrete shop
Shop.get_all_prices_list.refresh_cache(shop)

# Invalidate all cached items for concrete shop
Shop.get_all_goods_list.invalidate_cache_by_prefix(shop)
# or
Shop.get_all_prices_list.invalidate_cache_by_prefix(shop)
# or
from easy_cache import invalidate_cache_prefix
invalidate_cache_prefix('shop:{self.id}'.format(self=shop))

Invalidation summary

There are two ways to invalidate cache objects: use invalidation methods bound to decorated function and separate functions-invalidators.

<decorated>.invalidate_cache_by_key(*args, **kwargs)
<decorated>.invalidate_cache_by_tags(tags=(), *args, **kwargs)
<decorated>.invalidate_cache_by_prefix(*args, **kwargs)

# <decorated> should be used with a class instance if it is used in a class namespace:
class A:
    id = 1
    
    @ecached()
    def method(self):
        pass

    @ecached_property()
    def obj_property(self):
        pass
        
    @ecached_property('{self.id}:hello')
    def world(self):
        return '<timeconsuming>'

A.method.invalidate_cache_by_key()
# or
A().method.invalidate_cache_by_key()
# only one variant is possible for a properties
A.obj_property.invalidate_cache_by_key()
# and
item = A()
A.world.invalidate_cache_by_key(item)

# and
from easy_cache import (
    invalidate_cache_key,
    invalidate_cache_tags,
    invalidate_cache_prefix,
    create_cache_key,
)

# Note that `cache_instance` and `cache_alias` may be passed
# to the following invalidators
invalidate_cache_key(cache_key)
invalidate_cache_tags(tags)
invalidate_cache_prefix(prefix)

Here tags can be a string (single tag) or a list of tags. Bound methods should be provided with parameters if they are used in cache key/tag/prefix:

@ecached('key:{a}:value:{c}', tags=['tag:{a}'], prefix='pre:{b}', cache_alias='memcached')
def time_consuming_operation(a, b, c=100):
    pass

time_consuming_operation.invalidate_cache_by_key(a=1, c=11)
time_consuming_operation.invalidate_cache_by_tags(a=10)
time_consuming_operation.invalidate_cache_by_prefix(b=2)

# or using `create_cache_key` helper
invalidate_cache_key(
    create_cache_key('key', 1, 'value', 11), cache_alias='memcached'
)
invalidate_cache_tags(create_cache_key('tag', 10), cache_alias='memcached')
invalidate_cache_prefix('pre:{}'.format(2), cache_alias='memcached')

Refresh summary

There is one way to refresh cache objects: use refresh methods bound to decorated function.

<decorated>.refresh_cache(*args, **kwargs)

# <decorated> should be used with class instance if it is used in class namespace:
class A:
    @ecached()
    def method(self):
        pass

    @ecached_property()
    def obj_property(self):
        pass

A.method.refresh_cache()
A.obj_property.refresh_cache()

Internal caches framework

Be aware: internal cache framework instance is single threaded, so if you add new cache instance in a one thread it won't appear in another.

Easy-cache uses build-in Django cache framework by default, so you can choose what cache storage to use on every decorated function, e.g.:

# Django settings
CACHES={
    'local_memory': {
        'BACKEND': 'django.core.cache.backends.locmem.LocMemCache',
        'LOCATION': 'locmem',
        'KEY_PREFIX': 'custom_prefix',
    },
    'memcached': {
        'BACKEND': 'django.core.cache.backends.memcached.PyMemcacheCache',
        'LOCATION': '127.0.0.1:11211',
        'KEY_PREFIX': 'memcached',
    }
}

# then in somewhere code
@ecached(..., cache_alias='memcached')
# or
@ecached(..., cache_alias='local_memory')
# or even
from django.core.cache import caches
another_cache = caches['another_cache']
@ecached(..., cache_instance=another_cache)

However if you don't use Django, there is cache framework built into easy-cache package, it can be used in the same fashion as Django caches:

# Custom cache instance class must implement AbstractCacheInstance interface:
from easy_cache.abc import AbstractCacheInstance
from easy_cache.core import DEFAULT_TIMEOUT, NOT_FOUND

class CustomCache(AbstractCacheInstance):

    def get(self, key, default=NOT_FOUND):
        ...

    def get_many(self, keys):
        ...

    def set(self, key, value, timeout=DEFAULT_TIMEOUT):
        ...

    def set_many(self, data_dict, timeout=DEFAULT_TIMEOUT):
        ...

    def delete(self, key):
        ...

from easy_cache import caches

custom_cache = CustomCache()
caches['new_cache'] = custom_cache
caches.set_default(CustomCacheDefault())

# and then
@ecached(..., cache_alias='new_cache')
# or
@ecached(..., cache_instance=custom_cache)
# will use `default` alias
@ecached(...)

There is already implemented redis cache instance class, based on redis-py client:

from redis import StrictRedis
from easy_cache.contrib.redis_cache import RedisCacheInstance
from easy_cache import caches

redis_cache = RedisCacheInstance(StrictRedis(host='...', port='...'))
caches.set_default(redis_cache)

# will use `default` alias
@ecached(...)

Dynamic timeout example

You may need to provide cache timeout dynamically depending on function parameters:

def dynamic_timeout(group):
    if group == 'admins':
        timeout = 10
    else:
        timeout = 100
    return timeout

@ecached('key:{group}', timeout=dynamic_timeout)
def get_users_by_group(group):
    ...

Development and contribution

Live instances of Redis and Memcached are required for few tests to pass, so it's recommended to use docker/docker-compose to setup the necessary environment:

docker-compose up -d

# to enable debug logs
# export EASY_CACHE_DEBUG="yes"

# install package locally
pip install -e .[tests]

# run tests with pytest or tox
pytest
tox

Performance and overhead

Benchmarking may be executed with tox command and it shows that decorators give about 4% of overhead in worst case and about 1-2% overhead on the average.

If you don't use tags or prefix you will get one cache request for get and one request for set if result not found in cache, otherwise two consecutive requests will be made: get and get_many to receive actual value from cache and validate its tags (prefix). Then one set_many request will be performed to save a data to cache storage.