#23422 closed Bug (wontfix)
Cannot add Permission to Group in data migration
Reported by: | Tomas Walch | Owned by: | nobody |
---|---|---|---|
Component: | Migrations | Version: | 1.7 |
Severity: | Normal | Keywords: | |
Cc: | bronger@…, scott.woodall@…, hv@…, TZanke, frnhr, adehnert, alan.justino@…, django@…, info@… | Triage Stage: | Unreviewed |
Has patch: | no | Needs documentation: | yes |
Needs tests: | no | Patch needs improvement: | no |
Easy pickings: | no | UI/UX: | no |
Description
As initial_data fixtures are now deprecated I want to move my things there to migration files instead. One thing I do is to create auth.Group(s) and give these some custom permissions that I have defined on the app's models. This does not seem to be possible as, from what I can tell, the permissions and content types are created in a signal AFTER all the Migrations are run. Neither custom nor standard permissions can be found.
A Group and its permissions are data and the initialization thereof should therefore be possible to achieve through the migration system since its supposed to replace the fixtures.
my_new_group.permissions.add( Permission.objects.get_by_natural_key('add_my_model', 'my_app', 'my_model') )
This gives:
django.contrib.contenttypes.models.DoesNotExist: ContentType matching query does not exist.
even though I created the model in the previous migration step. There are no ContentTypes ate all yet.
Change History (29)
comment:1 by , 10 years ago
comment:2 by , 10 years ago
Resolution: | → wontfix |
---|---|
Status: | new → closed |
This is a limitation of the signalling system we use to create ContentTypes; if we fixed this to work, there would then be issues where the ContentTypes were created too early.
I would suggest that the data migration uses ContentType.objects.get_for_model to get the ContentType before making the Permission, as this creates the ContentType if necessary and has internal caching too.
comment:3 by , 10 years ago
I understand the won't fix, but a way to solve this group permissions problem would be appreciated. I'm blocked on a project by this...
comment:4 by , 10 years ago
I figure out ContentType.objects.get_for_model won't work "as is" in datamigration, because only default model managers are available.
comment:5 by , 10 years ago
If using ContentType.objects.get_for_model
is not possible, perhaps calling django.contrib.auth.management.create_permissions(...)
in the migration would work?
comment:6 by , 10 years ago
Here's an example of calling create_permissions
directly from a data migration, as suggested by carljm:
def make_permissions(apps, schema_editor, with_create_permissions=True): Group = apps.get_model("auth", "Group") Permission = apps.get_model("auth", "Permission") try: perm = Permission.objects.get( codename='add_mymodel', content_type__app_label='my_app') except Permission.DoesNotExist: if with_create_permissions: # Manually run create_permissions from django.contrib.auth.management import create_permissions assert not getattr(apps, 'models_module', None) apps.models_module = True create_permissions(apps, verbosity=0) apps.models_module = None return make_permissions( apps, schema_editor, with_create_permissions=False) else: raise my_new_group = Group.objects.create(name="test_group") my_new_group.permissions.add(perm)
Feels a little hacky, but it solved this wontfix
for me . I'd prefer a migration configuration that says "stop migrating, emit post_migrate
signals, and restart migrations."
comment:7 by , 10 years ago
South used to have send_pending_create_signals()
. A django equivalent would be useful.
comment:8 by , 10 years ago
Cc: | added |
---|
comment:9 by , 10 years ago
One thing we could do is have a post_migration
signal that runs after _every_ migration. That way it's consistent.
comment:11 by , 9 years ago
Cc: | added |
---|
follow-up: 20 comment:12 by , 9 years ago
I am not happy with this issue being "wontfix".
A helper method create_missing_content_types_and_permissions()
would be nice.
comment:13 by , 9 years ago
Cc: | added |
---|
comment:14 by , 9 years ago
Cc: | added |
---|
comment:16 by , 9 years ago
Chiming in with another +1 on the not-wontfixing - this is stopping me from running tests.
comment:17 by , 9 years ago
Cc: | added |
---|
Same problem seems to be affecting Generic relations in migrations, since they too depend on Content type framework...
comment:18 by , 9 years ago
The workaround jwhitlock proposes with calling create_permissions
from migration fixed one of my problematic migration, but not another. The other migration is calling loaddata
command to do a one-time fixture import, and it was failing because required ContentType
objects haven't been created. So it seems to be another face of the same problem.
Here is a more dirty workaround until an official solution becomes available. I think it has better chances of fixing the problem in case solution with create_permissions
fails:
if ContentType.objects.filter(**whats_needed_for_this_migration).count() < 1: print("Running another migrate command from migration to create ContentType objects:") call_command('migrate', 'users', '0002') # this should cite migration from the same app, just prior to this one print("Finished with the migrate command.")
I'm guessing the only thing really needed from the migrate
management command in this situation is:
emit_post_migrate_signal(created_models, self.verbosity, self.interactive, connection.alias)
but not sure how to populate the arguments.
I have no idea if using call_command("migrate")
from a migration (of all places...) can be problematic, but in my particular case there don't seem to be any problems.
The lengths we have to go to... Please willfix this.
comment:19 by , 9 years ago
Cc: | added |
---|
follow-ups: 23 25 28 comment:20 by , 9 years ago
Needs documentation: | set |
---|
Replying to guettli:
I am not happy with this issue being "wontfix".
A helper method
create_missing_content_types_and_permissions()
would be nice.
Found! This helper method exists already: django.core.management.sql.emit_post_migrate_signal
from django.core.management.sql import emit_post_migrate_signal def create_group(apps, schema_editor): db_alias = schema_editor.connection.alias try: emit_post_migrate_signal(2, False, 'default', db_alias) except TypeError: # Django < 1.8 emit_post_migrate_signal([], 2, False, 'default', db_alias) Group = apps.get_model('auth', 'Group') Permission = apps.get_model('auth', 'Permission') (...)
I suggest this behavior and "how to deal with" to be documented on the Permissions page.
Is this acceptable instead of a wontfix ?
comment:21 by , 9 years ago
Cc: | added |
---|
comment:22 by , 9 years ago
Cc: | added |
---|
comment:23 by , 9 years ago
Replying to alanjds:
Is this acceptable instead of a wontfix ?
way better. it works for me, thanks!
one minor thing is that when you add group permission for the first time you need to pass it's id, for the second time Permission object is also good - strange...
m
comment:24 by , 9 years ago
I can confirm that it also works for me.
0001.py added custom permissions
0002.py assigned custom permissions to a group
Without the workaround by guettli in 0002.py: the custom permissions weren't available.
Thanks
R
comment:25 by , 9 years ago
Replying to alanjds:
I've failed with Django 1.9, even after adapting emit_post_migrate_signal to its new signature:
db_alias = schema_editor.connection.alias try: # Django 1.9 emit_post_migrate_signal(2, False, db_alias) except TypeError: # Django < 1.9 try: # Django 1.8 emit_post_migrate_signal(2, False, 'default', db_alias) except TypeError: # Django < 1.8 emit_post_migrate_signal([], 2, False, 'default', db_alias)
This led me to this most significant part of stack trace and error:
... File "<venv>/local/lib/python2.7/site-packages/django/core/management/sql.py", line 50, in emit_post_migrate_signal using=db) File "<venv>/local/lib/python2.7/site-packages/django/dispatch/dispatcher.py", line 192, in send response = receiver(signal=self, sender=sender, **named) File "<venv>/olivia/local/lib/python2.7/site-packages/django/contrib/sites/management.py", line 20, in create_default_site if not Site.objects.using(using).exists(): ... django.db.utils.OperationalError: no such table: django_site
Has anyone succeeded with 1.9?
UPD: Sorry for the panic. This could be easily fixed supplying dependencies of the migration with sites. So you'll need at least that in the dependencies to keep going:
('contenttypes', '__latest__'), ('sites', '__latest__'),
comment:26 by , 9 years ago
FWIW I started to work on this issue (which is related to #24100 to #24067) on a branch.
The plan here is to make the pre_migrate
signal dispatch its plan (#24100) and inject CreatePermision
operations after CreateContentType
operations (that are injected after CreateModel
operations just like RenameContentType
operations are meant to be injected after RenameModel
operations in order to solve #24067).
I'll try to get this into 1.10.
comment:27 by , 8 years ago
Cc: | added |
---|
comment:28 by , 5 years ago
Replying to Alan Justino da Silva:
It still works with django 2.2 with some few changes:
from django.core.management.sql import emit_post_migrate_signal def create_group(apps, schema_editor): emit_post_migrate_signal(2, False, 'default') Group = apps.get_model('auth', 'Group') Permission = apps.get_model('auth', 'Permission') (...)
Thank you, because else, I can't use migration for that, and it should be a problem!
comment:29 by , 3 years ago
The emit_post_migrate_signal
is not a complete safe way, for this might cause other post migrate in apps to be triggered as well that now expect their tables to exist while they don't yet. For instance the Sites framework will run create_default_site
in post migrate, causing this error.
Other solutions are proposed in:
- https://stackoverflow.com/questions/31735042/adding-django-admin-permissions-in-a-migration-permission-matching-query-does-n
- https://stackoverflow.com/questions/57958722/django-python-loaddata-fails-with-django-db-utils-integrityerror
Do note that these solutions required a dependency of migration ("contenttypes", "0002_remove_content_type_name")
since Django 3.2.5.
I'm a bit new to the migrations system, but I was able to verify this. I created a new app called "my_app" and a model called "MyModel". I created the first migration and then the following data migration:
If I run these separately (
python manage.py migrate my_app 0001;python manage.py migrate my_app 0002
), everything is cool. Together it breaks (python manage.py migrate my_app
). You could solve this yourself by doing a get_or_create on the content type and on the permission, but that is by no means an adequate solution.