Quoting Ask from celery-users thread where he explained difference between @task a @shared_task. Here is link to the thread
TL;DR; @shared_task will create the independent instance of the task for each app, making task reusable.
There is a difference between @task(shared=True) and @shared_task
The task decorator will share tasks between apps by default so that if you do:
app1 = Celery()
@app1.task
def test():
pass
app2 = Celery()
the test task will be registered in both apps:
assert app1.tasks[test.name]
assert app2.tasks[test.name]
However, the name ‘test’ will always refer to the instance bound to the ‘app1’
app, so it will be configured using app1’s configuration:
assert test.app is app1
The @shared_task decorator returns a proxy that always uses the task instance
in the current_app:
app1 = Celery()
@shared_task
def test():
pass
assert test.app is app1
app2 = Celery()
assert test.app is app2
This makes the @shared_task decorator useful for libraries and reusable apps,
since they will not have access to the app of the user.
In addition the default Django example project defines the app instance
as part of the Django project:
from proj.celery import app
and it makes no sense for a Django reusable app to depend on the project module,
as then it would not be reusable anymore.