#celery

sure

describe your issue and environment

jon__: http://docs.celeryproject.org/en/latest/usergui...

Thanks! I'm setting up another ubuntu server, and using supervisor to run celery beat and celery worker.

This is for a django app.

Running the celery commands manually works fine, running them through supervisor doesn't. But my config is taken directly from another one of my servers where it works fine.

well ok, and what do you have in the logs ?

And now it's working fine.

ionelmc: I have no idea why it's working now, but thanks!

ionelmc: maybe i wasn't specific enough, but that link only covers updating a task while its running. once the task finishes it will be given a state in states.py such as success or failure

you can manually call the result backend yourself if you want

jon__: but there's a cache witht he results

there's a store_result method

cool looking

thanks

is there anyway to check if a queue has message? I don't know if this is kombu question or celery

I don't want to consume the message

I just want to see if anything is available

whyzgeek: generally it's broker -specific

ionelmc: can kombu/celery abstract it? I am currently running redis but I might migrate to rabbit. So I want something that works on both

it could

not sure if it does it now

whyzgeek: there are some methods in kombu that return the size

check queue_declare

reading this book: http://www.amazon.co.uk/Performance-Modeling-De...

asksol nice book

I do some performance modeling based on queueing theory

nice, like formally?

yeah for my PhD research @ NCSU

one problem we run into a lot is that closed form queueing models are based on Poisson traffic assumptions

and in practice that is not always true

We talked a week or two ago about me implementing a new AMQP 1.0 Kombu transport that will use Proton to use AMQP 1.0 and QMF for broker management

I notice that the virtual transports that support blocking (ie: non-polling) reads from their brokers (zmq and redis) both use a MultiChannelPoller

but that the amqp based transports that also use blocking reads (ie: non-polling) don't use MultiChannelPoller

I suspect my amqp 1.0 transport should also use MultiChannelPoller

any thoughts on that?

right, it would need to I guess but probably better to just connect it to the kombu.async event loop

the multichannelpoller in redis is used in the blocking consume api

but is also reused by the event loop

it should be possible to reuse this implemetation

optimally I don't want to have one for each transport

yeah I was considering both paths

I was thinking about registering my channels directly to the kombu.async event loop

the redis transport is blocking, but it uses select to know when to read

I think that is a better approach

I do too

brb

it's not clear to me when channels are created and destroyed, and which objects are responsible for that. If I'm going to register them, I need to manage their file descriptors too

so I can register their fds with kombu.async