How to use caches

All cache instances are the subtype of CacheAbstract. Users can use the following method to determine whether an unknown variable is a cache or not.

is_a_cache: bool = isinstance(value, dash_file_cache.caches.abstract.CacheAbstract)

The caches module are designed for letting users to initialize the service. In most cases, users should only use the cache like this:

import dash_file_cache as dfc

service = dfc.ServiceData(dfc.CachePlain(1))

where dfc.CachePlain can be replaced by dfc.CacheQueue or dfc.CacheFile. The usage scenarios for these caches are different:

Cache	Scenarios
CachePlain	The service is always accessed by the same process (but can be accessed by different threads). It can be used if the data can be kept in the memory and the background callback is not used.
CacheQueue	The service needs to be accessed by different processes. When the background callback is used, CachePlain will not work but this cache still works. The data is still kept in the memory.
CacheFile	The service needs to be accessed by different processes, and the cached data should be kept in the disk rather than the memory.

tip

The background callback is a feature of dash>=2.6. To find more details about this feature, please check

https://dash.plotly.com/background-callbacks

Use the cache with service

CachePlain

The usage of the cache should be like this:

service = dfc.ServiceData(dfc.CachePlain(cache_size: int))

where the size is the maximum of items that can be kept in the cache. If the cache is full, adding a new value to the cache will cause the least-recently-used (LRU) value removed.

CacheQueue

service = dfc.ServiceData(dfc.CacheQueue(cache_size: int, qobj: queue.Queue))

Different from CachePlain, this queue-based cache needs a queue object as the first argument. This queue object should be provided by a Process Data Manager. In many cases, such a manager needs to be initialized guarded by the if __name__ == "__main__" check. For example,

use_queue_with_manager.py

import multiprocessing as mproc
import dash
import dash_file_cache as dfc


def create_app():
    app = dash.Dash("demo")

    ctx = mproc.get_context("spawn")
    man = ctx.Manager()
    service = dfc.ServiceData(dfc.CacheQueue(1, man.Queue()))
    service.serve(app)
    return app


if __name__ == "__main__":
    app = create_app()
    app.run()

where we need to ensure that create_app() is used in if __name__ == "__main__". That's because create_app needs to initialize a new Manager().

CacheFile

service = dfc.ServiceData(dfc.CacheFile(cache_dir: str | None = None, chunk_size: int = 1))

The initialization of CacheFile takes two arguments.

The first argument cache_dir specifies the directory where the cache data is saved. If it is not specified, will use the system temporary folder. When the cache is not used anymore (typically, this case happens when the program exists), the folder will be removed.

The second argument is the chunk_size when saving or loading data from the cache. Its unit is MB. Users do not need to change it in most cases.

Different from CacheQueue, the initialization of CacheFile can be put anywhere.

Use the cache independently

warning

For somehow, you should not use these internal features mentioned in the following parts. Note that the best way of using CacheAbstract is to work with ServiceData.

The abstract class CacheAbstract provides the following functionalities:

Method	Description
dump	Put a value into the cache. Since some caches may be implemented by LRU scheduling, adding a new value into the cache may cause an old value removed.
load	Fetch a value from the cache. Running this method will not make the returned value deleted from the cache.
remove	Explicitly remove a value from the cache. The removed value will not be accessed anymore.
__contains__	Check whether a keyword exists in the cache.

Supposes we have a cache, the usage could be like this:

test_cache.py

import dash_file_cache as dfc

# The following codes are written by the style of Python>=3.12
def test[S, T](cache: dfc.caches.abstract.CacheAbstract[S, T], info: S, value: T):
    cache.dump("newval", info=info, value=value)
    assert "newval" in cache
    _info, _value_loader = cache.load_info("newval")
    assert info == _info

    # _value_loader is used for deferred loading for the value.
    _value = _value_loader()
    assert value == _value

The above generic function accepts three values. The first one is the cache. The second and the third values are info and value, respectively. When using the cache, both the info and value will be saved in the cache. However, when the value is loaded from the cache, the behavior of the returned values will be different:

_info, _value_loder = cache.load("newval")
_info: S
_value_loader: Callable[[], T]

where _info will be loaded instantly. However, _value_loader will be a closure that returns the value to be loaded. This mechanism allows users to put some light-weighted information in info while the large-amount data in value, and implement the conditional loading by the following steps:

1. Check whether _info satisfies some specific conditions.
2. If the condition is not fulfiled, abort _value_loader by not calling it.

Example of using the CachePlain independently

Take the following codes as an example:

CachePlain_and_conditonal_loading.py

import io
import dash_file_cache as dfc

data = io.BytesIO(" " * 1024)
len_data = len(data)

cache = dfc.CachePlain()
cache.dump("newval", info=len_data, value=data)

info, value_loader = cache.load("newval")
if isinstance(info, int) and info > 0:
    print(value_loader())

The loaded info is the length of the data. If the length is 0, will not load the value.

Example of using the CacheQueue independently

Using CacheQueue is a little bit tricky, because CacheQueue contains some data that should not be passed to the sub-processes. Therefore, we specially provide a mirror property of CacheQueue to allows the access of the cache in the subprocess.

CacheQueue_and_pool.py

import multiprocessing as mproc
from concurrent.futures import ProcessPoolExecutor
import dash_file_cache as dfc

if __name__ == "__main__":
    ctx = mproc.get_context("spawn")
    man = ctx.Manager()

    cache = dfc.CacheQueue(3, man.Queue())
    cache_m = cache.mirror  # `cache` cannot be directly sent to the process pool.

    with ProcessPoolExecutor(mp_context=ctx) as exe:
        exe.submit(cache_m.dump, "a", 1, 1).result()
        exe.submit(cache_m.dump, "b", 2, 2).result()
        exe.submit(cache_m.dump, "c", 3, 3).result()
        exe.submit(cache_m.dump, "d", 4, 4).result()

    print(dict(cache.cache.items()))
    # {'d': (4, 4), 'c': (3, 3), 'b': (2, 2)}

This example pass cache.mirror into the process pool. The values are added to the cache in the sub-processes. After all, the cached values can be accessed in the main process.

Do not use CacheFile independently

Different from CachePlain and CacheQueue which accept arbitrary types of data, CacheFile can be only used for caching the data that can be saved on the disk. In other words, CacheFile itself does not provide the file serialization or deserialization. When using this value with ServiceData, since users can only register a file path or a file-like object to the cache, the usage of CacheFile has no difference compared with CachePlain or CacheQueue. However, in most cases, CacheFile cannot be used for replacing CachePlain or CacheQueue.

CacheFile has such behaviors:

1. If a path of a file is registered to CacheFile, only the path will be registered because we assume that the registered path refers to a file that permanently saved on the disk.
2. If a file-like object (StringIO or BytesIO) are registered to CacheFile, the file-like object will be saved as a copy in the temporary storage of the cache.