If you examine the travis test outputs, testing currently doesn't actually test using Celery 3.0.
django-celery>=3.0,<3.0.21 didn't have an upper cap on their celery dependency, so on a fresh dependency install it installs celery 3.1.
django-celery>=3.0.21,<3.1 has an upper cap of celery<3.1, so it uses celery 3.0.
When using celery 3.0, the broker fallback tests and progress tests start failing.
It turns out when you use with self.settings(CELERY_ALWAYS_EAGER=True): in those tests, that properly changes the app setting for the way celery 3.1 builds the app, but doesn't change the setting for celery 3.0. So the tasks all turn up PENDING instead of SUCCESS.
Either testing should be fixed to cover that celery version, or the decision should be made to stop supporting celery<3.1. Doing the latter might be preferable, since celery supports django out of the box now, without needing django-celery.
If you examine the travis test outputs, testing currently doesn't actually test using Celery 3.0.
django-celery>=3.0,<3.0.21 didn't have an upper cap on their celery dependency, so on a fresh dependency install it installs celery 3.1.
django-celery>=3.0.21,<3.1 has an upper cap of celery<3.1, so it uses celery 3.0.
When using celery 3.0, the broker fallback tests and progress tests start failing. It turns out when you use
with self.settings(CELERY_ALWAYS_EAGER=True):in those tests, that properly changes the app setting for the way celery 3.1 builds the app, but doesn't change the setting for celery 3.0. So the tasks all turn upPENDINGinstead ofSUCCESS.Either testing should be fixed to cover that celery version, or the decision should be made to stop supporting celery<3.1. Doing the latter might be preferable, since celery supports django out of the box now, without needing django-celery.