George Psarakis

@ericraymond the database result backend still uses [PickleType](https://github.com/celery/celery/blob/master/celery/backends/database/models.py#L29), which maps to a `LargeBinary` field in SQLAlchemy as far as I can tell. I suggest that you review the previous comments...

@ericraymond I think I came up with a possible solution. We can redefine the field as [LargeBinary](https://docs.sqlalchemy.org/en/latest/core/type_basics.html#sqlalchemy.types.LargeBinary) and handle the pickling part separately (before storing the result). From the documentation,...

@remeika I can't imagine this being an issue (2 extra bytes for a row of multiple KB/MB). If you would like some guidance, you can open a draft Pull Request...

So [this](http://docs.celeryproject.org/en/latest/faq.html#why-do-workers-delete-tasks-from-the-queue-if-they-re-unable-to-process-them) documentation excerpt does not apply anymore?

@VictorLeP there are some channel subscriptions opened by the worker (in case of chord or other workflows using results) and also by the client that waits for the task result....

@jackdbernier @chetandhembre I opened https://github.com/celery/celery/pull/4666 that may be resolving this issue. Please test this before deploying to production, it may have unknown consequences.

@VictorLeP @jackdbernier @chetandhembre @jetz @uday-rayala we would really appreciate your feedback on the possible fix.

@jackdbernier I understand your point, we might need to look further into this. Have you by any chance had a look at #4666 ? It might be solving the issue...

@jackdbernier please see this [PR](https://github.com/celery/celery/pull/4709) for ignored results.

@WeipingGuo should be fixed in the latest release, are you still observing this?