makemigrations hangs forever on complex apps/models

[Seth Syberg]

I have a relatively complex django/wagtail application with a large number of apps, models, and migrations. Because this is a private project where we have full control over how migrations roll out, every 6 months or so we delete all migrations and reproduce them into a single file per app (I believe this is a fairly common practice and significantly improves development setup and CD times for us). This has worked for years until recently... Now when we run makemigrations it hangs forever after producing the initial migration for the first 6 or 7 apps.

I'm certain there is something about our setup causing a circular dependency issues or the like, which is certainly not django's problem, but the hanging/lack of error/lack of output does feel like a legitimate bug to me as we have no way to debug what is happening.

I understand this is a particularly nebulous / difficult to reproduce bug so I'm not sure what next steps could be. Perhaps this is a feature request for much more explicit output during makemigrations to facilitate debugging.

[Simon Charette]

Hello Seth,

Unfortunately as you've acknowledged yourself there is little Django can do to help with the level of details of you provided; if you can't identify whether the migration framework or your project is to blame while having to the source of Django and your project there is very little triagers can do to help without a way to reproduce.

I would suggest adding breakpoints to the source code of the makemigration command ​as well as the Autodetector and try to narrow down the origin of the issue.

TicketClosingReasons/UseSupportChannels might come handy in your debugging adventure.

[Seth Syberg]

Hmm I think it goes without saying that the migration framework is to blame right? Even if there's something funky in my models I would expect the migration framework to tell me something about it and make it possible to debug. Suggesting that I should set breakpoints directly in the code you claim is not to blame smells a bit funny to me.

[Natalia Bidart]

Replying to Seth Syberg:

Hmm I think it goes without saying that the migration framework is to blame right? Even if there's something funky in my models I would expect the migration framework to tell me something about it and make it possible to debug. Suggesting that I should set breakpoints directly in the code you claim is not to blame smells a bit funny to me.

Hello Seth!

Django is a community driven project so the framework needs everyone's help for improving and progressing. That includes reporting tickets providing as much information as possible, and for cases like this, a failing test case or a reproducer are crucial for us to action it.

Do you think you could try to provide a minimal Django test project showcasing the hanging when running makemigrations?

[Seth Syberg]

One point of clarification on the Autodetector suggestion - I'm running makemigrations on an empty database and no initial migrations at all. It's as if I just coded all of my models at once and ran makemigrations for the first time. In theory this should always work right? Especially given that this is live production code and there exists a set of migrations that can be run successfully to create this database state.

I'm open to trying to produce a minimal reproduction, but I have no idea where to even begin. This only cropped up after we'd created 7+ apps with 100s of models. I think if I knew how to create a minimal reproduction, I'd also know what to comment out to create an initial migration and wouldn't be submitting this ticket at all.

[Seth Syberg]

FWIW I was able to unblock myself here, but something still smells very odd to me, I'll add some details for posterity/anyone who ends up here. Maybe this is better suited as a feature request to (1) better handle custom user models and (2) make debug output more helpful.

As I googled around about this I found lots of info about custom user models/auth causing issues (including this now 8 year old open ticket: https://code.djangoproject.com/ticket/25313) so I had a hunch it was related to this. What finally got me over the hump though was killing the hung makemigrations (ctrl-c) and attempting to rerun makemigrations gave a little more information (on the 2nd run it oddly did not hang but errored with a circular dependency error). This resulted in a flurry of changes and moving add fields around and deleting some other field references that I was getting errors on and then running makemigrations again (I unfortunately did not record my steps as I was in deep debug mode), and eventually I did end up in a stable state. I went from 200+ migration files across 7 apps to about 12 migration files total and spinning up now is much much faster.