Persistent, stale-free, local and cross-machine caching for Python functions.
from cachier import cachier
import datetime
@cachier(stale_after=datetime.timedelta(days=3))
def foo(arg1, arg2):
"""foo now has a persistent cache, trigerring recalculation for values stored more than 3 days."""
return {'arg1': arg1, 'arg2': arg2}
Contents
Install cachier
with:
pip install cachier
For the latest version supporting Python 2.7 please use:
pip install 'cachier==1.2.8'
- Pure Python.
- Compatible with Python 3.5+ (and Python 2.7 up until version 1.2.8).
- Supported and tested on Linux, OS X and Windows.
- A simple interface.
- Defining "shelf life" for cached values.
- Local caching using pickle files.
- Cross-machine caching using MongoDB.
- Thread-safety.
Cachier is NOT:
- Meant as a transient cache. Python's @lru_cache is better.
- Especially fast. It is meant to replace function calls that take more than... a second, say (overhead is around 1 millisecond).
- S3 core.
- Multi-core caching.
- Cache replacement policies
Cachier provides a decorator which you can wrap around your functions to give them a persistent cache. The positional and keyword arguments to the wrapped function must be hashable (i.e. Python's immutable built-in objects, not mutable containers). Also, notice that since objects which are instances of user-defined classes are hashable but all compare unequal (their hash value is their id), equal objects across different sessions will not yield identical keys.
You can add a default, pickle-based, persistent cache to your function - meaning it will last across different Python kernels calling the wrapped function - by decorating it with the cachier
decorator (notice the ()
!).
from cachier import cachier
@cachier()
def foo(arg1, arg2):
"""Your function now has a persistent cache mapped by argument values!"""
return {'arg1': arg1, 'arg2': arg2}
Class and object methods can also be cached. Cachier will automatically ignore the self parameter when determining the cache key for an object method. This means that methods will be cached across all instances of an object, which may not be what you want.
from cachier import cachier
class Foo():
@staticmethod
@cachier()
def good_static_usage(arg_1, arg_2):
return arg_1 + arg_2
# Instance method does not depend on object's internal state, so good to cache
@cachier()
def good_usage_1(self, arg_1, arg_2)
return arg_1 + arg_2
# Instance method is calling external service, probably okay to cache
@cachier()
def good_usage_2(self, arg_1, arg_2)
result = self.call_api(arg_1, arg_2)
return result
# Instance method relies on object attribute, NOT good to cache
@cachier()
def bad_usage(self, arg_1, arg_2)
return arg_1 + arg_2 + self.arg_3
The Cachier wrapper adds a clear_cache()
function to each wrapped function. To reset the cache of the wrapped function simply call this method:
foo.clear_cache()
Settings can be globally configured across all Cachier wrappers through the use of the set_default_params function. This function takes the same keyword parameters as the ones defined in the decorator, which can be passed all at once or with multiple calls. Parameters given directly to a decorator take precedence over any values set by this function.
The following parameters will only be applied to decorators defined after set_default_params is called:
- hash_func
- backend
- mongetter
- cache_dir
- pickle_reload
- separate_files
These parameters can be changed at any time and they will apply to all decorators:
- caching_enabled
- stale_after
- next_time
- wait_for_calc_timeout
The current defaults can be fetched by calling get_default_params.
To limit the number of threads Cachier is allowed to spawn, set the CACHIER_MAX_WORKERS
with the desired number. The defeault is 8, so to enable Cachier to spawn even more threads, you'll have to set a higher limit explicitly.
Caching can be turned off across all decorators by calling disable_caching, and then re-activated by calling enable_caching.
These functions are convenience wrappers around the caching_enabled default setting.
You can set any duration as the shelf life of cached return values of a function by providing a corresponding timedelta
object to the stale_after
parameter:
import datetime
@cachier(stale_after=datetime.timedelta(weeks=2))
def bar(arg1, arg2):
return {'arg1': arg1, 'arg2': arg2}
Now when a cached value matching the given arguments is found the time of its calculation is checked; if more than stale_after
time has since passed, the function will be run again for the same arguments and the new value will be cached and returned.
This is useful for lengthy calculations that depend on a dynamic data source.
Sometimes you may want your function to trigger a calculation when it encounters a stale result, but still not wait on it if it's not that critical. In that case, you can set next_time
to True
to have your function trigger a recalculation in a separate thread, but return the currently cached stale value:
@cachier(next_time=True)
Further function calls made while the calculation is being performed will not trigger redundant calculations.
As mentioned above, the positional and keyword arguments to the wrapped function must be hashable (i.e. Python's immutable built-in objects, not mutable containers). To get around this limitation the hash_func
parameter of the cachier
decorator can be provided with a callable that gets the args and kwargs from the decorated function and returns a hash key for them.
def calculate_hash(args, kwds):
key = ... # compute a hash key here based on arguments
return key
@cachier(hash_func=calculate_hash)
def calculate_super_complex_stuff(custom_obj):
# amazing code goes here
See here for an example:
Question: How to work with unhashable arguments
If you want to load a value into the cache without calling the underlying function, this can be done with the precache_value function.
@cachier()
def add(arg1, arg2):
return arg1 + arg2
add.precache_value(2, 2, value_to_cache=5)
result = add(2, 2)
print(result) # prints 5
Cachier also accepts several keyword arguments in the calls of the function it wraps rather than in the decorator call, allowing you to modify its behaviour for a specific function call.
You can have cachier
ignore any existing cache for a specific function call by passing ignore_cache=True
to the function call. The cache will neither be checked nor updated with the new return value.
@cachier()
def sum(first_num, second_num):
return first_num + second_num
def main():
print(sum(5, 3, ignore_cache=True))
You can have cachier
overwrite an existing cache entry - if one exists - for a specific function call by passing overwrite_cache=True
to the function call. The cache will not be checked but will be updated with the new return value.
You can have cachier
print out a detailed explanation of the logic of a specific call by passing verbose_cache=True
to the function call. This can be useful if you are not sure why a certain function result is, or is not, returned.
The default core for Cachier is pickle based, meaning each function will store its cache is a separate pickle file in the ~/.cachier
directory. Naturally, this kind of cache is both machine-specific and user-specific.
You can configure cachier
to use another directory by providing the cache_dir
parameter with the path to that directory:
@cachier(cache_dir='~/.temp/.cache')
You can slightly optimise pickle-based caching if you know your code will only be used in a single thread environment by setting:
@cachier(pickle_reload=False)
This will prevent reading the cache file on each cache read, speeding things up a bit, while also nullifying inter-thread functionality (the code is still thread safe, but different threads will have different versions of the cache at times, and will sometime make unnecessary function calls).
Setting the optional argument separate_files
to True
will cause the cache to be stored in several files: A file per argument set, per function. This can help if your per-function cache files become too large.
from cachier import cachier
@cachier(separate_files=True)
def foo(arg1, arg2):
"""Your function now has a persistent cache mapped by argument values, split across several files, per argument set"""
return {'arg1': arg1, 'arg2': arg2}
You can get the fully qualified path to the directory of cache files used by cachier
(~/.cachier
by default) by calling the cache_dpath()
function:
>>> foo.cache_dpath()
"/home/bigus/.cachier/"
You can set a MongoDB-based cache by assigning mongetter
with a callable that returns a pymongo.Collection
object with writing permissions:
from pymongo import MongoClient
def my_mongetter():
client = MongoClient(get_cachier_db_auth_uri())
db_obj = client['cachier_db']
if 'someapp_cachier_db' not in db_obj.list_collection_names():
db_obj.create_collection('someapp_cachier_db')
return db_obj['someapp_cachier_db']
@cachier(mongetter=my_mongetter)
This allows you to have a cross-machine, albeit slower, cache. This functionality requires that the installation of the pymongo
python package.
In certain cases the MongoDB backend might leave a deadlock behind, blocking all subsequent requests from being processed. If you encounter this issue, supply the wait_for_calc_timeout
with a reasonable number of seconds; calls will then wait at most this number of seconds before triggering a recalculation.
@cachier(mongetter=False, wait_for_calc_timeout=2)
You can set an in-memory cache by assigning the backend
parameter with 'memory'
:
@cachier(backend='memory')
Note, however, that cachier
's in-memory core is simple, and has no monitoring or cap on cache size, and can thus lead to memory errors on large return values - it is mainly intended to be used with future multi-core functionality. As a rule, Python's built-in lru_cache
is a much better stand-alone solution.
Current maintainers are Shay Palachy Affek ([email protected], @shaypal5) and Judson Neer (@lordjabez); You are more than welcome to approach them for help. Contributions are very welcomed! :)
Clone:
git clone [email protected]:python-cachier/cachier.git
Install in development mode with test dependencies:
cd cachier
pip install -e ".[test]"
To run the tests, call the pytest
command in the repository's root, or:
python -m pytest
To run only MongoDB core related tests, use:
pytest -m mongo
To run only memory core related tests, use:
pytest -m memory
To run all tests EXCEPT MongoDB core related tests, use:
pytest -m "not mongo"
To run all tests EXCEPT memory core AND MongoDB core related tests, use:
pytest -m "not (mongo or memory)"
This project is documented using the numpy docstring conventions, which were chosen as they are perhaps the most widely-spread conventions that are both supported by common tools such as Sphinx and result in human-readable docstrings (in my personal opinion, of course). When documenting code you add to this project, please follow these conventions.
Additionally, if you update this README.rst
file, use python setup.py checkdocs
to validate it compiles.
Created by Shay Palachy Affek ([email protected]).
Current lead developer/contributor: Judson Neer (@lordjabez on GitHub).
Other major contributors:
- Judson Neer - Precaching, method caching support and numerous improvements and bugfixes.
- cthoyt - Base memory core implementation.
- amarczew - The
hash_func
kwarg. - non-senses - The
wait_for_calc_timeout
kwarg. - Elad Rapapor - Multi-file Pickle core, a.k.a
separate_files
(released onv1.5.3
). - John Didion - Support for pickle-based caching for cases where two identically-named methods of different classes are defined in the same module.
Notable bugfixers:
- MichaelRazum.
- Eric Ma - The iNotify bugfix (released on
v1.5.3
). - Ofir - The iNotify bugfix (released on
v1.5.3
).