#celery

Are there any good examples of CELERY_QUEUES and CELERY_ROUTES in a config_from_dict() kind of config. I seem to recall that it can be done using only primitives in the config, without having to instantiate Queue object within the config dict, for example, but I can't find anything recent, and the old stuff doesn't seem right anymore.

sorry, not config_from_dict, but rather celery.config.update()

This is a config_from_object kind of setup, but the contained dicts don't actually instantiate kombu objects or anything. That's what I want to do, but these examples don't match the current docs.

E.g. CELERY_QUEUES is a dict, but docs (http://docs.celeryproject.org/en/latest/usergui...) say it's a tuple

Well, it still seems to work. Can't find anything on this simpler usage in the official docs. Maybe it's considered too limiting for reason I haven't run into yet. I dunno.

ok, I think I got that all figured out

New question: is there a convenient way, within a task definition, to dispatch further subtasks to the same queue/routing_key as the current task, or do I just need to get it from context.delivery_info myself and specify it in my apply_async? Background: I have some common tasks that are shared between two high-level jobs that need to be prioritized separately (so neither can starve the other), call the jobs A and B. When that shared task is running

as part of A, I need it to get sent to A's queue, but ditto for B. I'd rather not hard-code queue names in my code. Right now, all of those details can be specified in the config, and I don't want to break that by hard-coding things.

More general question: In the context of celery, when should I use a topic exchange vs. a direct exchange?

hi all

it seems CELERY_ACKS_LATE = True doesnt work for me

I use rabbitmq as backend

whenever a worker raises an exception, it gets removed from the related queue

how can I prevent this?

Hey is it possible to change priority of existing tasks already in the queue?

emre: use retries

acks_late is meant for situations where something much worse happens - eg, process crashes with segfault, or oom killer does its nasty business