Parity with django-celery-results

The target is django-celery-results 2.6.0. This page lists what is identical and every known difference.

Identical

• Backend: storing results with result_extended properties (task_name, task_args, task_kwargs, periodic_task_name, worker, traceback), meta with children and request.meta, date_started on STARTED, binary content base64 encoding, forget, cleanup, group save/restore/delete, chord counting and callback triggering including error propagation, retry with result_backend_always_retry.
• Models: TaskResult, GroupResult, ChordCounter with the same field names, types, verbose_name, help_text, nullability, defaults, editable, Meta.ordering, verbose_name(_plural), as_dict(), __str__() and ChordCounter.group_result().
• Managers: get_task, store_result (same signature), get_group, store_group_result, get_all_expired, delete_expired.
• Admin: TaskResultAdmin and GroupResultAdmin use the same list_display, list_filter, date_hierarchy, search_fields, fieldsets, readonly_fields and ALLOW_EDITS behaviour. The changelist, filters (including facet counts), search, sorting, pagination, "show all", date hierarchy, change form, delete view, bulk delete action, and history are Django's own code running on a Redis-backed queryset. ChordCounter is not registered, as upstream.
• Views and URLs: the four JSON views, the task_pattern converter and URL names.
• Translations: upstream message ids are reused, and upstream catalogues are shipped.

Differences

Topic	django-celery-results	This package	Reason
Primary key	integer id, task_id unique	task_id / group_id	Natural Redis key; no counter or id mapping. Admin URLs use the task id.
App label	django_celery_results	django_celery_results_redis	Both packages can be installed side by side.
Admin section title	"Celery Results"	"Celery Results (Redis)"	Distinguishable when both are installed.
ID_FIRST_URLS default	True (deprecated URLs, warning on import)	False	The URLs are deprecated upstream.
Transactions	results are written inside the caller's transaction	Redis writes are immediate and never rolled back	Redis is not a Django database.
using= arguments	select a database alias	accepted and ignored	One Redis connection.
date_started on STARTED	database clock (Now())	Python clock (timezone.now())	No database; one clock also keeps date_started <= date_done.
String comparison	database collation	Unicode code points; case-insensitive lookups use str.casefold()	Deterministic across drivers.
NULL ordering	database dependent	last in ascending, first in descending order (PostgreSQL behaviour)	Deterministic across drivers.
Unsupported query features	full ORM	no joins, annotations, extra(), raw(), set operations, select_for_update(); datetimes()/dates() return lists	Not needed by the admin or the backend. Such calls raise NotSupportedError.
Lookups	every lookup of the database	exact, iexact, contains, icontains, startswith, istartswith, endswith, iendswith, regex, iregex, in, gt, gte, lt, lte, range, isnull, and the datetime transforms year, iso_year, quarter, month, week, week_day, iso_week_day, day, hour, minute, second, date and time (chainable as date__year, time__hour, ...)	Database-specific lookups such as unaccent or trigram_similar raise FieldError; typed into an admin URL they redirect with ?e=1.
Regular expressions	the database dialect (POSIX on PostgreSQL)	Python re.search(), like Django on SQLite	No database to delegate to.
MySQL repeatable-read warning	emitted by get_task	not applicable
_delete_group of a missing group	raises on the unsaved instance	no-op	Upstream bug.

How parity is tested

• Static tests import django_celery_results and compare admin options, model field metadata, Meta options and method output.
• Differential tests load identical data into both packages and compare query results and rendered admin changelists (row ids, filter choices, facet counts, date hierarchy links, result counts) for a table of URL parameters.
• The upstream backend and model tests are ported and run against every driver.