kombu.exceptions.EncodeError when trying to execute Job

[tim-fiola]

Environment

• Python version: 3.8.10
• Nautobot version: 1.1.0b1

Steps to Reproduce

1. Copy Planned Site Job from Nautobot docs to a .py file $NAUTOBOT_ROOT/jobs/
2. In Web UI, Extensibility-->Jobs
3. Run the New Branch job

Expected Behavior

Job to execute

Observed Behavior

Got the following message in the UI:

Server Error
There was a problem with your request. Please contact an administrator.

The complete exception is provided below:

<class 'kombu.exceptions.EncodeError'>

Object of type DeviceType is not JSON serializable

Python version: 3.8.10
Nautobot version: 1.1.0b1

[jathanism]

Oof this one is ugly. I am able to reliably reproduce it and attempts to get nested serialization of model objects using JSON and the NautobotKomboJsonDecoder class will alleviate issues with our objects such as DeviceType, but ultimately results the same error but with ModelState:

Object of type ModelState is not JSON serializable

This is django.db.models.base.ModelState which is the internally cached state of the model. I've never seen this leak before.

I can completely obviate this error scenario entirely by hard-coding these settings in nautobot_config.py:

CELERY_ACCEPT_CONTENT = {"pickle"}
CELERY_RESULT_ACCEPT_CONTENT = CELERY_ACCEPT_CONTENT
CELERY_TASK_SERIALIZER = "pickle"
CELERY_RESULT_SERIALIZER = "pickle"

This forces Celery to use Pickle. This isn't necessarily desirable, but I think we have two options:

1. Go with Pickle for now while we sort it out
2. Try to get to the bottom of how to do nested JSON serialize/deserialize for messaging in Celery

[lampwins]

Given the current feature set in 1.1, pickle would work fine. However the switch to json is driven by the scheduler because args and kwargs have to be stored in the DB. So in a pinch, to get 1.1 we could switch it back to pickle and then figure out what the issue is.