Allow multiple keys for cache_on_save

[chemary]

I use django-cacheops for managing MySQL replication lag, so I set cache_on_save to avoid fast reads after write to get stale data, I think it will be handy to be able to specify multiple keys.

An example:
I usually get User model by id or by username so both queries are cached pointing to the same object:

"conj:ftmanager_user:username=5feb24948664d6be7e36aed63e453e0cdc9be81d"
"conj:ftmanager_user:id=108547"

If I set cache_on_save to one of them for instance username, when User is modified the other cache is deleted so next User.objects.get(id=108547) will hit the database.

My suggested implementation:

class ManagerMixin(object):
    ...
    def _post_save(self, sender, instance, **kwargs):
            ...
            keys = ['pk'] if cache_on_save is True else cache_on_save
            for key in keys:
                cond = {key: getattr(instance, key)}
                qs = sender.objects.inplace().filter(**cond).order_by()
                if MAX_GET_RESULTS:
                    qs = qs[:MAX_GET_RESULTS + 1]
                qs._cache_results(qs._cache_key(), [instance])

[Suor]

Are you willing to make a pull request?

[chemary]

I have not made the pull request because I don't know if other find there is a need for this feature or if the solution is correct.

I have noticed for instance that having two keys, stores the same object two times in Redis

"conj:ftmanager_user:username=5feb24948664d6be7e36aed63e453e0cdc9be81d"
"conj:ftmanager_user:id=108547"

point to two objects that contains the same data, they should point to the same result?

q:9be3411703144230d7baae7c1d84cb8d
q:47d86696fdab13139c2bd19c77bcf22b

[Suor]

Key is determined by queryset, there is no way to know that 2 different querysets result in same thing until you execute them. So, yes, same object is stored twice under 2 different keys. This is normal behavior and your implementation is correct regarding this issue. You should, however, handle cache_on_save=<string> case.

The idea of somehow compressing it, e.g. using content addressed values in cache with queryset keys refering to single stored object is interesting, but probably won't be worth the hustle.

[Suor]

P.S. You can probably rewrite every place you use username to use id and thus solve this for your case. Do you really need 2 unique fields in a model?

[chemary]

I really only need to cache by username that is why is mostly used by our API, but ManagerMixin._pre_save() is reloading the object by pk, in this case id and this is why I also cached the object by id.

(I don't find clean to make username the pk of the object, we use ids as pks by design.)

[Suor]

You can just disconnect pre_save signal as a quick fix.