I am working on an existing Django project that contains migration code like:
someFluffyModel.objects.all().delete()
or
someModel = SomeModel(....)
someModel.save()
Now wherever there is such code and newer migrations change the schema that reflect the current version of the model, there is an issue in applying migrations from scratch.
As I understand the reason is that the model used in the migration doesn't reflect the model used in the migration at that point in time. As I have found fixtures can help in loading data but how about deletion?
Is the preferred way to manually delete data from the database?
Sorry forgot to answer my own old question it's been years now but just in case anybody is curious in your migrations you should always be using historical models. The reason is that you get the model at a point in time(reconstructed incrementally base on schema migrations).
def forwards_func(apps, schema_editor):
# We get the model from the versioned app registry;
# if we directly import it, it'll be the wrong version
Country = apps.get_model("myapp", "Country")
This way if you try to re-run your migrations at some point in the future with model schema x'' old migrations will be run using x' and x depending on their dependencies.
Avoid importing models to your migrations directly at any cost.
It is always important to be able to rer-run you migrations since that allows you to migrate forwards backwards between your environments roll back faulty deployments and above all run your tests. If you are doing it wrong your tests will fail at some point when you'll get to schema x' and you'll have to follow the correct approach mentioned above to fix them.
Thanks to #Daniel Roseman for giving me some pointers that day.
Related
I have a big Django application, currently trying to upgrade from 1.6 to 1.7 (there's been attempts to upgrade straight to 1.11, but it was too much trouble, so my plan is to do it one minor at a time).
I'm following the Upgrade from South instructions, and deleted all previous migrations, but I can't get makemigrations to work. The current problem is that the auth.User model has been patched to include two new fields:
User.add_to_class('profile',
models.ForeignKey('access.Profile', null=True, blank=True,
related_name='user_foreignkey'))
User.add_to_class('profiles',
models.ManyToManyField('access.Profile', null=True,
blank=True))
This patch was made in a separate app. If I just leave it where it is, I get the following error when running python manage.py makemigrations:
ValueError: Lookup failed for model referenced by field auth.User.profiles:
access.Profiles
I tried moving the add_to_class calls to the same file where Profile is defined, (after the definition), but got the same error. I also tried changing the syntax from 'access.Profile' to Profile, to no effect. Is there something else that could make this work?
If not, since I'm adding fields to the model, I figure the correct approach would be extend the AbstractUser model, as suggested by this guide. The problem with this is that the new initial migration will create a table access_user instead of using the existing auth_user. Would it be safe to simply rename auth_user to access_user and fake the migration?
Any other suggestions on how to overcome this with the least refactoring possible (management always thinks there are more urgent things than the upgrade) are welcome.
tl, dr; I removed auth migrations and ran makemigrations again, that seems to have solved the problem.
I had decided to give option 2 a try. I removed the add_to_class patches, created a new class User(AbstractUser) with the new fields, removed all the existing migrations and ran makemigrations again. I got the exact same error, which made no sense because there was no longer any code relating auth.User to Profile, so I decided to investigate auth migrations. I ran
$ python migrate --list settings=settings.mysettings
and it showed a suspicious migration for auth, with yesterday's date, probably when I first tried to make migrations.
Tutorials on "how to reset all migrations" do not mention third-party installed apps, but thanks to this thread I knew where Django stored the auth migrations, so I removed both of them (the initial and the suspicious one) and reordered the INSTALLED_APPS so that access is now before django.contrib.auth. Then ran makemigrations again, and it got through access and auth migrations successfully.
I'm now getting other problems with migrations in other apps, but they don't seem to be related with this.
It might be nice to have a more intuitive way to reset all existing migrations, not only those from your repo. But then again, I don't think the add_to_class patches are the proper way to do what we wanted.
If one is using Django, what happens with changes made directly to the database (in my case postgres) through either pgadmin or psql?
How are such changes handled by migrations? Do they take precedence over what the ORM thinks the state of affairs is, or does Django override them and impose it's own sense of change history?
Finally, how are any of these issues effected, or avoided, by git, if at all?
Thanks.
You can exclude a model completely from the django migrations, and then you are responsible to adjust the schema to the django code (or the django code to the existing schema):
class SomeModel(models.Model):
class Meta:
managed = False
db_table = "some_table_name"
name = models.Fields....
Note that you can't have it both ways, so migrations are preferred when possible. You can always define a custom SQL migration, that will save the need for external changes. However, sometimes you do need to handle the schema elsewhere instead of migrations, and then use managed=False
The migrations system does not look at your current schema at all. It builds up its picture from the graph of previous migrations and the current state of models.py. That means that if you make changes to the schema from outside this system, it will be out of sync; if you then make the equivalent change in models.py and create migrations, when you run them you will probably get an error.
For that reason, you should avoid doing this. If it's done already, you could apply the conflicting migration in fake mode, which simply marks it as done without actually running the code against the database. But it's simpler to do everything via migrations in the first place.
git has no impact on this at all, other than to reiterate that migrations are code, and should be added to your git repo.
I am working with some other developers on a website (using Django) and as it is a new site and the database schema is constantly changing. We are using South for schema migration but are running into the issue where each of us have our own branch checked out and are working on our own separate part of the project. When the code is merged back in very often there have been multiple migrations generated for the same model (there might be three migrations with id 003 for model A).
It seems like South is pretty good about managing conflicts when they relate to different models:
http://south.readthedocs.org/en/latest/tutorial/part5.html#team-workflow
But when dealing with the same model it gets a little messy manually creating the merged migration.
I was wondering if there was any better way of doing this or a better tool than South.
One idea I have is for no one to check in the migration files and just check in the model changes directly to git. That way every person would just run their own migrations and won't have to worry about conflicts. Not sure if that is a good practice...
I am curious how other people are managing these situations.
Thanks:)
From the South documentation:
Make sure your team know who is working on what, so they don’t write migrations that affect the same parts of the DB at the same time.
So that's the thing you shouldn't do: don't create migrations that affect the same model unless you're sure you're able to merge them later into the main development trunk. A tool can't figure out what the right database schema needs to be when two developers modify the same field of a model.
In that case you'll need to manually fix migrations and get the model fields in order. This could mean migrating back and constructing a new migration before merging the branch with the trunk.
I am writing a django south migration that depends on the model in another app, so I have included --freeze OTHERAPPNAME when I ran python manage.py datamigration …
However in the forwards() function I can access the other model fine (with orm['otherappname.MyModelName']), however in the backwards() step (which also depends on the model in otherappname), if I try to access orm['otherappname.MyModelName'], I get an error
*** KeyError: "The model 'mymodelname' from the app 'otherappname' is not available in this migration."
I can see the frozen model details in the bottom of the file. Why can't I access it?
NB: This model is created in another migration inside otherappname of which this datamigration depends on.
The backwards migration actually uses the frozen ORM of the previous migration. It's logical when you think about it, but it's admittedly not very intuitive.
Moral: freeze the models you'll need in your datamigration in the schemamigration that proceeds it.
I've recently begun using South for migrations in my Django project. All was going well until recently when I ran into a peculiar issue.
I have two apps in my project, say, App-A and App-B. A model in App-A has a foreign key to a model in App-B. When I've been trying to build my system, I ran syndb which created all the auth_ and the south_ tables. Then I ran migrate which threw up errors. When it tried creating the model from App-A, which referenced a model from App-B, the model App-B wasn't migrated/created as yet and therefore the error.
In order to resolve this, I had to manually migrate App-B first and then App-A. Am i doing something wrong here? How is South supposed to know the migration order across apps?
Thanks.
This explained it https://south.readthedocs.io/en/latest/dependencies.html.
Migrations for apps are nice ‘n all, but when you start writing a
large project, with a lot of apps, you realise you have foreign key
relationships between apps and working out what order migrations would
need to be applied in for each app is just painful.
Luckily, we also had this problem, so South has a dependency system.
Inside a migration, you can declare that it depends on having another
app having run a certain migration first; for example, if my app
“forum” depends on the “accounts” app having created its user profile
table, we can do:
# forum/migrations/0002_post.py class Migration:
depends_on = (
("accounts", "0003_add_user_profile"),
)
def forwards(self):
Then, if you try and migrate to or beyond 0002_post in the forum app, it will first make sure accounts is migrated at least
up to 0003_add_user_profile, and if not will migrate it for you.
Dependencies also work in reverse; South knows not to undo that
0003_add_user_profile migration until it has undone the 0002_post
migration.
You can have multiple dependencies, and all sorts of wacky structures;
there are, however, two rules:
No circular dependencies (two or more migrations depending on each
other) No upwards dependencies in the same app (so you can’t make
0002_post in the forum app depend on 0003_room in the same app, either
directly or through a dependency chain.
South migrates apps in the order they appear in the INSTALLED_APPS tuple in settings.py. So just make sure App-B comes before App-A in your settings.py, and it should work :)