Duplicate connections

[mbartz]

We have found rare instances of duplicated connections in the relatedobject table.

The scenario in which we are using generic M2M is to keep a list of attached Reference Values on an XML object. We iterate through the list and attach the Reference Value objects to the original XML object. In the rare cases where the duplication has occurred, the date created field of the table shows that they were created approximately 2 seconds apart from each other.

The error occurs when we attempt another processing run of the XML object. When we attempt the connection the "second" time, the get_or_create in the connection model fails to catch the MultipleObjectsReturned error.

We cannot reliably reproduce the problem. In our database we have approximately 700k connections and have only had about a dozen errors.

We are now trapping the errors and have considered adding a unique together on the related object table.

Sorry I don't have a reproducible scenario.

[coleifer]

Thanks for the note, I'd be happy to look into this. So the get_or_create() isn't behaving correctly?

[mbartz]

In the get_or_create of the RelatedManager:

            kwargs.update(core_filters)
            return super(RelatedManager, self).get_or_create(**kwargs)

in the base get_or_create it only handles the ObjectDoesNotExist exception.

I think the implicit assumption is that the 4-tuple can only exist once.

I think the easiest thing to do is to add a unique_together(parent_id, parent_type, object_type, object_id).

I know it's redundant but it would fix the problem. If we see any patterns, I repost the data.