What concerns should I have for a multiple hour long @transaction.atomic django background process?

I have a celery background task that runs and can take 5+ hours to complete. It is primarily creating new objects and it will generate around 100k or so new objects. If something were to go wrong during this I would like to roll it all back as if it never happened. My strategy for ensuring this is to wrap the background task in a @transaction.atomic. Since this is such a long running transaction is there anything I should be particularly concerned about? If I happen to fail after 4 hours do I need to worry about my Postgres database freezing up in production? Is there another way I should be doing this or is transaction.atomic the best strategy?

I currently have it running in a @transaction.atomic block. It currently works fine in development but I have concerns that it may cause issues in production.

There are a variety of problems it might cause, but whether it does cause any of them depends on a lot of details we don't have.

If it rolls back, it won't cause PostgreSQL to simply freeze up. Any particular piece of data will be rolled back incrementally when it is encountered--either by other "natural" queries, or by VACUUM. A query needing to inspect one piece of data won't freeze waiting for 99,999 other objects to also get cleaned up.

But another transaction wanting to insert or update a tuple which would cause a constraint violation against something the large transaction has done will block until the large transaction finishes.