Add client attribute to ValkeyCache backend class to be used with `cached` decorator

[rjnocelli]

cached decorator requires backend class to implement client property.

I've implemented a quick wrapper (see example) to make it useful for a FastAPI application I am currently working on but it would be nice to have it implemented in the current client. I am happy to take over this task a create a PR. Excuse me if there is an issue/PR already opened regarding this.

from glide import GlideClientConfiguration, NodeAddress, GlideClient
from aiocache import ValkeyCache as AiocacheValkeyCache
from aiocache.serializers import PickleSerializer

class ValkeyCache(AiocacheValkeyCache):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.client = None

    async def connect(self):
        self.client = await self._init_async_client()
        logger.info("Connected to Valkey.")

    async def _init_async_client(self):
        client = await GlideClient.create(self.config)
        return client

    async def disconnect(self):
        if self.client:
            logger.info("Disconnecting from Valkey...")
            await self.client.close()
            logger.info("Disconnected from Valkey.")
            self.client = None


valkey_cache = ValkeyCache(
    config=valkey_config, namespace="main", serializer=PickleSerializer()
)

[Dreamsorcerer]

cached decorator requires backend class to implement client property.

I'm not clear what you mean here..

[rjnocelli]

@Dreamsorcerer sorry for not being very clear. The ValkeyCache class only sets the client attribute when the connection is created using a context manager but the cached decorator expects a cache session opened and set on the client attribute as you can see on the error below.

https://github.com/aio-libs/aiocache/blob/master/aiocache/backends/valkey.py#L66

backends/valkey.py", line 66, in _get
    value = await self.client.get(key)
                  ^^^^^^^^^^^
AttributeError: 'ValkeyCache' object has no attribute 'client'

We need something similar to what MemCached client already does.

EDIT:

Just to add some more iformation about: the previous RedisCache which was replaced by ValkeyCache on this PR did implement the client attribute

Also, based on the comments on that PR it seems that the feature was half implemented and very likely not tested with the @cached decorator so that might explain the reason why this attribute is missing.

[Dreamsorcerer]

We can improve the error, but you need to initiate the cache before you use it (i.e. before calling any decorated functions).

This is shown in the example/docs:

aiocache/examples/cached_decorator.py

Lines 23 to 24 in cff38a3

	async with cache:
	await cached_call()

I've also opened #1066 to handle it correctly in the aiohttp demo too.

[rjnocelli]

@Dreamsorcerer thanks for this example shared. The context manager works but I think it only makes sense when the context of the program is rather small. In a big web applications I think it makes way more sense to be able to open and close connections manually. For example in FastAPI, a good practice is creating connections using lifespan events which require to manualy create and destroy connections in different function calls. Anyways, I just insist for the sake of making this client as frictionless as possible for the most amount of use cases.

[Dreamsorcerer]

For example in FastAPI, a good practice is creating connections using lifespan events which require to manualy create and destroy connections in different function calls.

As mentioned, see #1066. Lifespan is likely based on aiohttp.