@ejklock,
I managed to reproduce this issue and I tried to debug this. It’s baffling: 50% of the time the task is routed to the LMS worker, and in the other 50% the task is routed to the CMS worker. Can you open an issue on the Tutor github repo that precisely explains how to reproduce this issue, what is the expected outcome, and what is the observed result? Please include copy-pasted logs, and not screen captures. Also, add a link to this conversation.
I’ll also attempt to reproduce this in Juniper, as this might not be a problem anymore in the next release.
For the record, a possible explanation might be that the ‘edx.lms.core.default’ queue is included both in the LMS and the CMS workers.
LMS:
>>> settings.CELERY_QUEUES
{'edx.lms.core.high': {}, 'edx.lms.core.high_mem': {}, 'edx.lms.core.default': {}}
>>>
CMS:
>>> settings.CELERY_QUEUES
{'edx.cms.core.high_mem': {}, 'edx.cms.core.high': {}, 'edx.cms.core.default': {}, u'edx.lms.core.default': {}}
(note for self: it is very likely that this is linked to https://github.com/overhangio/tutor/issues/302#issuecomment-613573142. In the CMS we have settings.ALTERNATE_QUEUES = [u'edx.lms.core.default'])