#celery

dhackner: you would use a django signal no ?

kennethreitz: sounds like a deadlock - are you using 3.0 ?

ionelmc: 3.1 w/ rabbitmq

what causes deadlock?

as soon as a add a new worker, jobs start running through it

well there were lots of deadlock -type issues in 3.0

mainly around low values of maxtasksperchild

ah interesting

i tried setting that to 500 today

hoping that they'd be killed after that limit it hit

*is

didn't seem to have much effect though

but do you have workers (the subprocesses) ? does the node respond to events (inspect blablabla) ?

last i checked, yes

it's the oddest behavior

hard to say

i have hit a snag myself while upgrading to 3.1 :(

maybe downgrading will fix it :)

yeah but 3.0 is weird

i have one install that leaks around 120mb in the master process once in a while

that weight is carried to the children when forking ofcourse

ionelmc: problem with using a django signal there is that i don't *believe* i'd have access to the underlying task body

dhackner: https://docs.djangoproject.com/en/1.6/ref/signa...

hmm

i get no entries in djcelery_taskstate

because I'm using rabbitmq for the broker right?

oh wait

taskstate is for events (the monitor)

ah

ok that makes sense

taskmeta is for results (if you use the django orm backend)

right, which i'm grabbing

but I dont think it contains the task data

why not use asyncresult then?

you mean task arguments or the result?

i'm processing a model in a task

so i want to attach the associated state and result to the model

you could just wrap the processing code yourself

either way, to attach to the model is either a task_id or a taskmeta entry right?

then just lookup the asyncresult based on the task_id

ok, so you want the task id?

ideally i would directly attach the taskmeta entry

the task id is easy

but requires another lookup down the road.

i see

why do you want to attach the taskmeta ? you only need to save the taskid and status

taskmeta tracks the status

the backend would do cleaning that would not work if you have foreignkeys pointing to result objects

the goal

i imagine you want task result expiration no?

that is handled elsewhere

the point of attaching the taskmeta

is that it was a clean way to look at the model along with its status

but its ok to just attach an id and do a asyncresult lookup

that object would be created after the task is done iirc

the taskmeta

yea i think you are right

you could just subclass the backend

i was HOPING it was created as a placeholder first

upon producing the task

thus my initial question

but I think I have a solution, no worries :-)

i think custom result backend would be a better solution

but if you want to attach something that indicates progress then taskmeta is not what you want

taskstate is

and for that you need to enable monitoring

no progress to track here

just successful, pending, failed

think an outbox kind of

i dont see why a custom result backend would be necessary for tracking ids and states

brb

hello, Im using celery for my django project, Ive used everytime "from celery.decorators import task" now it is not possible anymore what happens ?

and as I understand it right, I dont django-celeryd anymore ?

Hi. I am having an hard time figuring out how to connect between my application to the worker is running on another terminal session

It looks like my application doesn't know about the celery configuration, I can't find some docs about that online... any ideas ?

ledil: Do you want to import the Task class?

rons: You could duplicate the config, make sure to use the same config?

Flonka: I want just register a celery task

Flonka:where should I duplicate to? the configuration now is inside the proj/celeryconfig.py. and the connection is inside proj/celery, this is right ?

rons: just import your tasks and apply them.

rons: from proj.tasks import your_task; your_task.apply_async((args...))

ledil: Checked the user guide?