savepoint_rollback() after an IntegrityError gives TransactionManagementError

[ketanbhatt]

Hey,

I noticed that if my model throws an IntegrityError while saving, and we do a savepoint_rollback() after that, it throws TransactionManagementError.
I investigated a little and found that in an atomic() block, Integrity Error causes automatic rollbacks, so calling savepoint_rollback() raises error.

[sergei-maertens]

+1, running into the same issue at the moment

[bmihelac]

reference: #377

[bmihelac]

This should be fixed in branch 412-drop-django-15-support

https://github.com/django-import-export/django-import-export/tree/412-drop-django-15-support

Can you please test if this works for your case?

[int-ua]

I'm getting this on 0.6.0. The IntegrityError is raised somewhere on init_instance (still trying to find the exact point).

My case is having a "legal person" model that have O2O field to a Person model and a code field that is preferred for determining the right legal person. I'm trying to determine a Person by multiple fields with my custom widget and today there was two legal person rows with different codes. So it failed since I'm returning a second legal person connected to the same Person. But it doesn't raise the IntegrityError exception in my widget code for some reason. Will continue investigating it tomorrow.

[int-ua]

AFAIU the reason is that Resource.import_row doesn't account for IntegrityError possibility and this example from Django docs cannot be used:

try:
    with transaction.atomic():
        generate_relationships()
except IntegrityError:
    handle_exception()

So we need to prevent any savepoint_rollback call if there was an IntegrityError since it has been already rolled back. Is that right?

[rhunwicks]

@bmihelac In dd1c085 you removed the exception handling from the call to before_import.

Did you mean to remove the raise() as well as the rollback?

Because what happens now is that if you call import_data with raise_errors=True and before_import has an Exception, then it logs the exception but doesn't raise it. So the Resource then carries on and tries to import all the rows in the dataset, which isn't what we want. The after_import call does raise errors if requested.

I will prepare an MR to add

            if raise_errors:
                raise

to the before_import call.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.