emord / quickcache

caching has never been easier

quickcache: caching has never been easier

Usage

quickcache has an easy default for Django, but it's just as easy to bring your own backend.

BYOB:

from quickcache import get_quickcache

quickcache = get_quickcache(cache=my_cache_backend)

Django:

from quickcache.django_quickcache import get_django_quickcache

quickcache = get_django_quickcache(memoize_timeout=10, timeout=5 * 60)

...and then lets you cache any function based on the values of some or all of its params

@quickcache(['name'])
def take_expensive_action(name):
    ...

...or any method based on values of the self object

class Person(object):
    ...
    @quickcache(['self.id'])
    def look_up_friends(self):
        ...

... and when you know you just made the cache stale, you can clear it

person.add_friend()
person.look_up_friends.clear(person)

Examples

• cache a singleton function, refresh every 5 minutes

  @quickcache([], timeout=5 * 60)
  def get_config_from_db():
      # ...

• vary on the arguments of a function

  @quickcache(['request.couch_user._rev'], timeout=24 * 60 * 60)
  def domains_for_user(request):
      return [Domain.get_by_name(domain)
              for domain in request.couch_user.domains]

  now as soon as request.couch_user._rev has changed, the function will be recomputed

• skip the cache based on the value of a particular arg

  @quickcache(['name'], skip_arg='force')
  def get_by_name(name, force=False):
      # ...

• skip_arg can also be a function and will receive the same arguments as the function:

  def skip_fn(name, address):
      return name == 'Ben' and 'Chicago' not in address
  
  @quickcache(['name'], skip_arg=skip_fn)
  def get_by_name_and_address(name, address):
      # ...

Features

• If you're using the Django default, then in addition to caching in the default shared cache, quickcache caches in memory for memoize_timeout seconds, (suggested 10 seconds, conceptually the length of a single request).

• In addition to varying on the arguments and the name of the function, quickcache will also make sure to vary on the source code of your function. That way if you change the behavior of the function, there won't be any stale cache when you deploy.

• Can vary on any number of the function's parameters

• Does not by default vary on all function parameters. This is because it is not in general obvious what it means to vary on an object, for example.

• Allows you to vary on an attribute of an object, multiple attrs of the same object, attrs of attrs of an object, etc

• Allows you to pass in a function as the vary_on arg which will get called with the same args and kwargs as the function. It should return a list of simple values to be used for generating the cache key.

A note on backends

The Django default uses a two-tier caching backend that caches in memory as well as in the default shared cache. This is optimized for multi-worker web code so that within a single request the object is cached in memory, and across requests the object is cached in the shared cache.

If you are using your own backend, you can achieve the same effect using quickcache's cache helpers, TieredCache and CacheWithTimeout. If you are rolling your own, you may want to work off the source code in quickcache/django_quickcache.py.

In the examples above, the keyword arguments timeout and memoize_timeout are only available on the Django default; when you bring your own backend, and want to override your system-wide default timeout, the equivalent will be to override the cache with the cache keyword argument. For example,

@quickcache(..., cache=get_my_cache_backend_with_timeout(timeout=TIMEOUT))

where get_my_cache_backend_with_timeout is a function you define.

Note on unicode and strings in vary_on

When strings and unicode values are used as vary on parameters they will result in the same cache key if and only if the string values are UTF-8 or ascii encoded. e.g. u'namé' and 'nam\xc3\xa9' (UTF-8 encoding) will result in the same cache key BUT u'namé' and 'nam\xe9' (latin-1 encoding) will NOT result in the same cache key

Building and deployment

Following instructions for packaging and distributing universal wheels

python setup.py bdist_wheel
pip install twine
twine upload dist/*