Opened 16 years ago

Closed 13 years ago

#8907 closed (duplicate)

ORM-level Caching

Reported by: bradcater Owned by: anonymous
Component: Core (Cache system) Version: 1.0
Severity: Keywords:
Cc: Triage Stage: Design decision needed
Has patch: yes Needs documentation: no
Needs tests: yes Patch needs improvement: no
Easy pickings: no UI/UX: no

Description

MOTIVATION:

  • Suppose the following is very expensive:

MyStuff.objects.filter(<conditions>)

We commonly use caching to avoid doing this operation often, which can
leave our code littered with the following:

    cached = cache.get(<cachekey>)
    if not cached:
        cached = MyStuff.objects.filter(<conditions>)
        cache.set(<cachekey>,cached,<time>)
    return cached

We seek to unify the Django ORM and caching in a natural way.

USE CASES:

  • To cache QuerySets, append .cache(<time>) to the end of the Django

ORM call. This will cache the given object for <time> seconds provided
that the cachekey is not already in the cache.

  • Example: MyStuff.objects.filter(extra_type=4).cache(3600)
  • This should return all MyStuff objects with an id greater than 10 and should store the result to the cache for 3600 seconds (1 hour).
  • To force caching (even if the cachekey already exists), provide the

force argument (e.g., .cache(<time>,force=True)).

  • Example: MyStuff.objects.filter(mass=20).cache(3600,force=True)
  • This should return all MyStuff objects with a mass of 20 and should cache the result for 3600 seconds even if it would over-write a previously-cached result.
  • To ensure that a cached value is not used, use .nocache() before any

specific call.

  • Example: MyStuff.objects.nocache().filter(user__id=7)
  • This should return all MyStuff objects with a User whose id is 7 and should not use a previously-cached value even if it exists.

NOTES:

  • Cachekeys are generated as the md5 hexdigest of the generated SQL.

Ultimately, this detail should not matter and is subject to change.

  • It is a subtle but salient point that you can call, e.g., MyStuff.objects.nocache().filter(contact__id=2).cache()

thus using cache() and nocache() together. It is desirable to do this
if you wish to get the latest results from the db (ignoring any data
currently in the cache) and caching them for later use.

DJANGO CHANGES:

  • .../django/db/models/query.py :
    • cache(self,time,force=False) added for QuerySet objects
    • nocache(self) added for QuerySet objects
    • _cachekey(self) added for QuerySet objects
    • _clone(self, klass=None, setup=False, kwargs) updated for QuerySet objects

Attachments (1)

query.py.diff (1.3 KB ) - added by bradcater 16 years ago.
Diff adding ORM-level caching support.

Download all attachments as: .zip

Change History (7)

by bradcater, 16 years ago

Attachment: query.py.diff added

Diff adding ORM-level caching support.

comment:1 by anonymous, 16 years ago

Needs tests: set
Owner: changed from bradcater to anonymous
Status: newassigned

comment:2 by Philippe Raoult, 16 years ago

patch is broken ? can't see from here

comment:3 by Karen Tracey, 16 years ago

Trac isn't displaying it properly, probably because it is a simple diff of a single file instead of an svn diff from the root of the source tree. You can still see if if you follow the link to "download in original format", it's just harder.

in reply to:  1 comment:4 by rays, 15 years ago

Replying to anonymous:

has there been any progress here? the diff is unviewable and now out of date. I'm really interested in what this does. Can the poster please attach a new diff or full file?

comment:5 by (none), 15 years ago

milestone: post-1.0

Milestone post-1.0 deleted

comment:6 by anonymous, 13 years ago

Resolution: duplicate
Status: assignedclosed

The outcome of #17 should resolve in a fix for this.

Note: See TracTickets for help on using tickets.
Back to Top