Jobs are not stopped by a celery kill

[Daniele Viganò]

When you kill celery on the workers the job on the masters doesn't exit and keep running waiting for data. This behavior probably depends on the switch to iterator_native() which is not polling the workers anymore (the old iterator did)

[Daniele Viganò (daniele-vigano)]

We found two solutions:

- Modify celery to add a callback to drain_events() in backends/amqp.py [iter_native() -> get_many() -> drain_events()]
- Create a supervisor as a forked thread/process which monitor the workers status

[Daniele Viganò (daniele-vigano)]

The above patches are just a mock-up fo the solution 1) "Modify celery to add a callback to drain_events() in backends/amqp.py [iter_native() -> get_many() -> drain_events()]"

[Michele Simionato (michele-simionato)]

Solution two implemented here: https://github.com/gem/oq-engine/pull/1403