Connection Pooling doesn't work as intended

[Ashish-Bansal]

Description

I found that connection pooling doesn't work in django-redis package as intended if you use it as Cache Backend in Django.

Even if you configure it like -

CACHES = {
    "default": {
        "BACKEND": "django_redis.cache.RedisCache",
        # ...
        "OPTIONS": {
            "CONNECTION_POOL_KWARGS": {"max_connections": 100}
        }
    }
}

you would use a single connection for django.core.cache.cache operations.

You would actually use pooling, only if you interact with the Connection Pool manually by requesting connections.

Intended behavior

If we configure django_redis as mentioned in docs, we expect it to use multiple connections to the redis. This would be especially required in the concurrent world.

Stack trace

Here's code references -

• Inject django_redis.cache.RedisCache into django
• Instantiate single instance of django_redis.client.DefaultClient and use it to implement django cache interface
• DefaultClient implements redis client operations
• Almost all operations call get_client to get low level redis client
• get_client maintains list of redis connections, one for each mentioned server.
• If connection doesn't exist, get_client would call connect on connection_factory available through ConnectionPool to get a connection from the pool. Otherwise it reuses last created connection for that server.

I think pooling behaviour would be correct in Django's official implementation but Django's Redis backend may not be mature enough yet.

EDIT: Hm, I was wrong, it seems like pooling is handled at redis.client.Redis level. So, the implementation logic is same as Django's Redis backend. django_redis's implementation is correct!! I'll dig why connection pooling isn't working properly in my case.

[Ashish-Bansal]

Implementation is correct, it probably works as intended, my bad!