I have a running Django application on Heroku with migrations auto-run on release. While in most times this works fine sometimes there is a problem when:
There are more than one migration in given release (they can be in different apps)
Some migration will fail, but not the first one
In this case manage.py migrate will fail so Heroku will not finish the release and will not deploy the new code. This means that code is in the old version and the database is in the state "somewhere between old and new".
Is there a simple way to autorun Django run reversed migrations in case of the failure of the release command on Heroku?
Transactions won't help here as there might be more than one migration (multiple apps) and Django run each migration in seperate transaction.
As I couldn't find any existing solutions I am posting a gist I have written to solve this.
https://gist.github.com/pax0r/0591855e73b9892c28d3e3cdd15f4985
The code stores the state of migrations before running the migration and in case of any exception reverts back to this state. It also checks if all migrations are reversible during the migrate step.
It's not yet well tested, but I will work forward to create a library from it for easier use by others.
Related
I pulled the latest version of the master branch (which is currently running in production) and downloaded a fresh dump of the production base. I restored the database, launched it locally - I get a message that I have 10 unapplied migrations.
At the same time, CI / CD is configured in the project, all migrations during the deployment process are applied automatically.
Where could non-applied migrations come from? I can only explain this by the fact that someone edited the table with migrations on the prod, but there may be other versions?
If I try to apply the migrations locally, then the InconsistentMigrationHistory error comes out (one of the migrations was applied earlier than the one on which it depends), but this is the next problem for me (although it is obviously related to the first one)
I have been having this issue when managing Django migrations on deployment, and I would like to know what approach should I take:
I am developing an application using Django, and I am using PythonAnywhere to deploy the web app. I am using SQLite as the database. I understand the Django migrations work like a tree or a sequence (001, 002).
Every time I make a change in a field locally, it works fine because the tree has been saved and the sequence has not changed. But when deploying the changes through GitHub (having deployed the web app and calling the migrations and migrate command, which creates another migrations files and sequence), I usually get an error indicating that the migration tree is broken. So I have to go to the app's migration folder, delete them, and call the migrations and migrate commands again.
This is causing me a lot of problems, as I do not want to mess with schema of the database and lose the information.
It is just me or anyone else is having this issue the migrations tree, not only on PythonAnywhere but on others servers?
Thank you!
Thank you guys! #Ankit Tiwari, not I did not as the Django documentation says that it is important to keep the migrations files on deployment as #caseneuve recommended. Even thought I read somewhere that on the server side is not necessary to call the makemigrations command, just the migrate as the migrations files already exist; so I try it and so far it is not giving me more errors. Thanks for the answers.
While developing a Django project, all your migrations are stored within each app folder, however, in production I don't want those migrations, I want to keep a Production database, and a Development database:
How do I handle Django migrations in a Production and Development environment?
I'm asking this question because it's been really hard to update my deployed project with new additions in the development one, my ideal scenario would be to keep each set of migrations in a folder outside my source code, just like the databases.
The best idea is to keep production and development migrations the same and while developing you clean migrations before pushing the code and you should push migrations into your Version Control System too.
In development, you might end up deleting a table and re-creating it so make sure you don't push the un-intended migrations. The thing is you should treat migrations as code, not an automated script. I have done a lot of mistakes in the past, so, I came to the conclusion of including migrations in code. and that's effective and gives more control.
Moreover you might have to do data migrations in production, how will you do if you wont push the code?
I'm having trouble with my Django-app that's been deployed.
It was working fine but I had to do a minor modification (augmented the max_length) of a Charfield of some model. I did migrations and everything was working fine in the local version.
Then I commited the changes without a problem and the mentioned field of the web version now accepts more characters, as expected, but whenever I click the save button a Server Error rises.
I assume I have to do some kind of migration/DB update for the web version but I don't seem to find how.
(I'm working with Django 1.11, postgresql 9.6, and DigitalOcean).
EDIT
I've just realized that the 'minor modification' also included a field deletion in the model.
Short answer
You have to run
python manage.py migrate
on the server, too. Before you do that, make sure all migration scripts you have locally are also present on the server.
Explanation
After changing the model, you probably locally ran
python manage.py makemigrations
This creates migration scripts that'll transform database schema accordingly. Hopefully, you've committed these newly created scripts to Git, together with the changed model. (If not, you can still do so now.)
after running makemigrations (either before or after committing, that shouldn't matter), you've probably locally ran
python manage.py migrate
This applies the migration scripts to the database that haven't been applied to it, yet. (The information which ones have already been applied is stored in the database itself.)
You probably (and hopefully) haven't checked in your local database into Git, so when you pushed your tracked changes to a remote repo and pulled them down on your server (or however else the new Git revisions got there), the changes to the server database haven't happened, yet. So you have to repeat the last local step (migrate) on the server.
Further reading
For more information, refer to the Django 1.11 documentation w.r.t. migrations. (You can e.g. limit migration creation or migration application to a single Django app, instead of the whole Django project.) To get the grip of these things, I can recomment the free Django Girls tutorial.
We have migrated the Django 1.4 application to Django 1.8 successfully. The Django 1.4 version of applicaiton is still in use in production until we go live with Django 1.8. The problem is that lots of data have been updated on production server which needs to be migrated to version 1.8. Is there any way I can migrate the data from database of 1.4 to 1.8 except manually doing that? Note that the model/database columns are different in both the version.
Can anybody suggest some good options ?
Thanks.
Required pre-reading
Django migrations:
https://docs.djangoproject.com/en/1.8/ref/django-admin/#makemigrations-app-label
Assuming you were using South:
https://docs.djangoproject.com/en/1.8/topics/migrations/#upgrading-from-south
Getting Started
Firstly dump your local database I prefer using mysql/postgres/whatever docs for this rather than using ./manage.py dumpdata.
You will also want to dump your production database as well just for safekeeping.
Next in your local environment I would dump the database and create a new database.
Then I would test that all your migrations actually work on a blank database.
These are the instructions for django 1.8
./manage.py makemirgrations
./manage.py migrate
That will help show up if any migrations are in an inconsistent state operating from a blank slate. If you encounter any errors they should be fixed first.
Given that works, now I would test that your migrations actually work against your production data.
So drop your local database, create a fresh one then load the production dump in.
If the tables are already configured correctly (i.e. they are in an up to date state in your production database) then you will need to fake all migrations.
./manage.py migrate --fake <appname>
However given that you have changed some models since upgrading to 1.8 in your local environment, then you might need to fake only some of the migrations. This could be the tricky part depending on the timing of when you upgraded and when you created the migrations.
Because django 1.7 will just create an initial migration for each app, you may need to actually break up the migration for some apps. That is, instead of 0001_initial you might need to manually break up that migration into 2 components:
1. a migration to match the current state of your production database
2. a migration to match any additional changes you have made to your model since then.
One way to do this is to checkout your first commit once you had django 1.8 working correctly locally then run
./manage.py makemigrations
then commit that
then go forward to your latest commit then run
./manage.py makemigrations
Now you should have 2+ migrations in each app that you have modified since upgrading to django 1.8.
Then you can fake initial on those apps that have 2+ new migrations for django 1.8
./manage.py migrate --fake-initial app1 app2
and the rest just
./manage.py migrate app3 app4
Now run your tests to confirm everything is working locally.
If you have changed migrations you will again want to test locally against a blank database to test that they work smoothly
Once that is working, record the 'migrate' commands that you used - then deploy your app to production and run just those migration commands once you have upgraded to django 1.8 on your server.
After Successful Completion
Take new dumps of your local and production databases
Uninstall South (assuming you had it installed before) from local and production environments
I'm sure there are a couple of holes in the above but hopefully that gives you the gist of what you need to do.