#celery

IIRC

messy for on_stop, you mean? is there a better way?

messy in general, it was only recently that I changed the signals to ensure normal cleanup

this was because posix semaphores are not released when the process is terminated

so executing finally: blocks would cause the semaphore to be released

(but not if the process is KILLED)

so in 3.1 I rewrote the pool to not use posix semaphores

it still calls finally: blocks, but that is very brittle since it will also call the parents finally:

right i saw that return value to handle that?

I guess, even with on_pool_stop it could call os._exit if that is decided agains though

if it's decided that calling finally: blocks is a bad idea I mean

the worker_process_shutdown signal should also come with a warning that it may not be called (e.g. on SIGKILL)

right

well also i didnt want to submit a pull request until i figured out some way to test it, if that's possible

i thought you might have some pointers

i had trouble in test actually creating a process pool

unittests would only test that the function is called

Hello

sure, on_stop

but not actually fork a worker process

if we used Finalize within billiard, that still wouldn't fire on kill, right

it seemed like.. not a great test but yeah simple is easy

hi Jo_

no, nothing will fire on kill, it's not possible to catch the kill signal

right thats what i thought

hey jo_

Just small question, I'm trying to use celery with python3.2 but got a lot of errors, what is the recommended version ?

Jo_: what version of celery are you using, on what platform?

celery==3.0.23 on Ubuntu 12.04 LTS

you may try using the development version as this is no longer using 2to3

pip install https://github.com/celery/py-amqp/zipball/maste... https://github.com/celery/kombu/zipball/master#... https://github.com/celery/billiard/master/zipba... https://github.com/celery/celery/zipball/master...

thx

not sure what errors you are experiencing, but some configuration options may not be as well tested as others

e.g. which broker you use

or serialization library, and so on

often the errors are easy to fix, so you can open an issue

I tried redis first

if you want to show errors here please use psatie.org

or any other paste site

so if i were to try from within billiard/pool.py: Worker.run(): Finalize(self, self.on_loop_stop, exitpriority=1), what else do I need to do to install that as a handler? just assign it to self._finalize =[<here>] ?

i just tried again to be sure and it doesnt seem to be calling it twice

mdan: right, the object must still be alive at shutdown

mdan: (to be sure of that you could make it a global)

where you goin' with that gun in your hand

oops, was scrolled up

yeah i just tried that, neither worked

i created on_stop that mocked a call to on_loop_stop and disabled my on_loop_stop call from finally: block

whether it was self._finalize or global _finalized defined within that run loop, my cleanup functions defined in celery didn't get called

i wonder if there's something i'm missing or if it's related to why atexit isn't getting called

maybe the signal is somehow going to the parent or something

anyway, it works with the finally: block :)

Jo_: I just fixed a Python3 bug in billiard, so you should reinstall the latest master

Jo_: small errors merging multiprocessing changes from python3.3 back in. I haven't tested this in 3.2 yet

Jo_: seems to work on 3.2 as well, using redis as broker and result backend

so my work day is starting to get absurdly long, and i have a conference call for my other job soon, so I'm going to have to go get dinner in a minute

the unit tests are passing on 3.2 and 3.3, but that doesn't always help uncover errors

but thanks for your help

and i'll surely have more questions later :)

ok seems to work now on 3.2 with rabbitmq will test now redis

mdan, are you in the north of england?

nope, I'm in japan

i lived briefly in cambridge a number of years ago

ah, just the timing of dinner threw me

being in a different timezone makes sense

ok works with redis on 3.2 thanks a lot asksol, will look into your commit

right

here in the north, they call the midday meal dinner

ah

asksol, before I go, any idea why celery/worker/__init__.py might hang on blueprint.join() when using processes ? the comment in bootsteps.py for Blueprint.join says it should only be called for running green, and it's wrapped in a try, but instead of falling through to the except it just hangs for me. i see there's a timeout, should maybe we pass in a timeout if not self.pool_cls.is_green ?

not a big deal, im simply not calling it now, not sure if that will cause a problem

i'll send you my email, i dont have IRC on at home

mdan: is this gevent?

because I'm aware of this bug with gevent where it doesn't shut down

I have been debugging it, going crazy over it

for step in steps: step.stop(); return

every step.stop() in the list returns ok

but it does not exit the for loop for some reason

it's whacky

with eventlet it works fine

can't seem to get a scheduled task working with a local django setup. I've got the worker running and the beat process running with the database backend configured. The task shows correctly in djcelery_periodictask but the last run field remains null

oh, ho hum

every five minutes, not five minutes every hour

oops

I'm unfamiliar with the term 'oh hum', is that another thing from the north? :)

"ho hum" is very much like a sigh.

I think Treebeard used it in the lord of the rings books :)

can someopne tell me what a node is for celery?

is it a worker?

Jo_: yes a node is a 'celery worker' instance

so there's worker instances and there's pool processes

you can say either "node" or "worker instance", node is used when speaking of clusters

when using the multiprocessing pool the worker instance manages pool processes that executes the tasks

but you cannot communicate with pool processes, you communicate with the parent which is the worker instance

should probably be in the glossary (http://docs.celeryproject.org/en/master/glossar...)

terms like node, pool and backend are so generic they're doomed to be confused

ok was asking because when I try to inspect my workers using the redis broker I always get "No nodes replied within time constraint."

Jo_: try increasing the timeout

rabbitmq broker I get the worker list but not with redis and py3.2

I tried by putting 20sec same

ok, let me try here

getting the same

it should reply within a few ms, so this is probably something stupid

Jo_: you're using the development version, right?

yes latest commits

ah, this seems to be a bytes vs str problem

there's a check when it receives a pubsub event from redis: if payload['type'] == 'message'

but 'message' here is b'message'

Jo_: https://github.com/celery/kombu/commit/7c1663bd...

I have NameError: global name 'bytes_to_string' is not defined

weird

oh

or no, I cannot find any use of 'bytes_to_string'

ok just checked master out of kombu

ok working now

thx

Hi, I still have problesm with .delay not publishing tasks to rabbitmq after a couple of minutes in both low and heavy traffic in a rather complicated setup (2 app > AWS ELB > 2 rabbitmq in cluster, and 2 workers > AWS ELB > 2 rabbitmq in cluster), I thought .delay would throw an exception when the connection fails but it seems it doesn't fail in a normal connection loss siutation. How should I go about debugging this? It's next to impossible to replicate the