As a user, I can use 'flush' after the migrations are applied and receive the required AccessPolicies
As a user experienced if a user does the following they will have a problem:
- User runs migration and receives the AccessPolicy via a data migration like this one
- User then later calls 'flush' which keeps migrations marked as "run" but drops all data including the AccessPolicy
- User then tries to use Pulp, but receives
AccessPolicy matching query does not exist.
Change where we recommend the AccessPolicy data is created.
- Stop doing it in migrations
- Instead do it in a post-migrate signal
- If the AccessPolicy is already in the db catch the Duplicate entry and move on
- This would need to be changed in the plugin writer docs here
- Also any existing AccessPolicy made in migrations would also have to be put here too. Currently that is only one, created in migration 42.
Use a fixtures approach instead.
Solution 1 seems more user-friendly since user doesn't need to do anything, however this will run with every django-migrate.
Solution 2 is a Django-native way to fill in the database, it's explicit and can be run on_demand. It's less user-friendly since it's one more step.
I hear a strong resistance in the community with regards to signals, so maybe solution #2 is the way.
I'm also somewhat concerned what to do if/when data model changes for the data we load in. Is any of the solutions provide a better way to handle that? and also a check that a certain migration applied before loading the data.
Note this means that if you change one of the rows created by a fixture and then run loaddata again, you’ll wipe out any changes you’ve made.
It seems like fixture way handles existing data and it overwrites it in case someone changed any row.
The overwriting nature of fixtures I think is not a good fit for this need because users can customize the AccessPolicy at any point and we can't overwrite it.
I also read the #django recommendation to not use signals, but I didn't read any evidence with it, so I put it into the fear, uncertainty, and doubt category. Solution 1 has a great user experience, so I think we should prototype solution 1 and see how reliable (or not) it is for us.
Please register to edit this issue