Add a cache=NUM_SECONDS argument to QuerySet

[Adrian Holovaty]

It'd be convenient for the lookup API to have caching baked in.

[wojtek@…]

We're working on a caching framework which would also cache listings and handle automatic invalidation when objects are changed. The results will be posted within a few days.

[Gary Wilson <gary.wilson@…>]

This could get tricky if you've got the same query in different places and don't want them cached the same.

[tbarta@…]

It would be nice to be able to plug Django into something like ​Memcache and turn on caching, but I would like to see it above the database layer rather than below it.

try:
    polls = memcache.get('polls_blah')
except CacheNotFound:
    polls = Polls.objects.all()
    memcache.set('polls_blah', list(polls), memcache.FIFTEEN_MINUTES)

Putting caching into the database layer IMHO adds too much magic, and could produce weird results when cache behavior is unexpected. Forcing the app developers to use a cache explicitly keeps them "in the know" about what they're doing. There could even be a Django shortcut like

polls = get_cached('polls_blah', lambda: Polls.objects.all(), cache.FIFTEEN_MINUTES)

[James Bennett]

Replying to tbarta@gmail.com:

You've mostly described what Django's cache API already does -- you choose a cache backend (memcached is an option), and then a standardized API lets you talk to it via cache.get and cache.set. See ​the cache documentation for more information.

[anonymous]

Thanks for the pointer, ubernostrum. That cache documentation describes exactly what I was suggesting in my first example.

Given that Django has fairly complete caching middleware already available, I'd recommend that this either be a shortcut (something like my get_cached example above) or just be skipped. The convenience of embedding caching into the database layer is (IMHO) outweighed by the unnecessary coupling.

If an application wants to use the admin interface but still support model caching, invalidation could simply be put in the model's save method, right?

[Jacob]

This has languished long enough without any real interest. I'm marking wontfix given that this would only be sugar for the existing cache api.