#celery

has anyone managed to run flower behind / under / with nginx?

I have set up a proxy_pass with nginx, but all the links in the templates point to 127.0.0.1 and there are no css and js (I assume because of the same reason)

this seems to be the function that constructs the urls: https://github.com/mher/flower/blob/master/flow...

__rodrigo__: you should fix the host header in the nginx requests

try adding proxy_set_header Host 'the-host-you-want-in-urls';

ionelmc, great!

ionelmc, it improved a lot, I get the header / navbar of the page but it says "Page not found"

__rodrigo__: can't help you with so little details

the links in the navbar are pointing ok with the --url_prefix option to "/flower"

ionelmc, ok, what do you think I should seek and tell you?

maybe is it the port?

settings, what you get and what you expect

i cannot guess what you want :)

ionelmc, ahh... ok

I want to be able to go to /flower in the server and use flower

I'm running flower in 127.0.0.1:5555, the defaults, and I want to access it through nginx with the full name

I put this on the nginx conf:

location /flower/ {

proxy_pass http://127.0.0.1:5555;

proxy_set_header Host 'staging.webcassette.com';

}

and I start flower with

FLOWERCTL="$FLOWER_CHDIR/manage.py celery flower --url_prefix=/flower

that sounds about right

what's wrong with it?

but I get this: https://staging.webcassette.com/flower/

’lo all, so I’m using amqp as the result backend but unfortunately had a lot of errant tasks which should’ve had ignore_result=True but didn’t and config CELERY_RESULT_PERSISENT was True…so now my RabbitMQ is full of result queues and I’d like a way to get rid of these messages

keeping in mind some of the results /are/ actually important :/

s/these messages/their messages/

eppa: it's more that the expiration timeout is disabled

AIUI CELERY_RESULT_PERSISENT just tells the queue to make them restart-proof

FunkyBob: ah sorry yep you’re right

(FunkyBob and I are working together)

__rodrigo__: could it be a missing slash at the end of the prefix?!

ionelmc, do you see the navbar and the error?

ionelmc, I don't know, let me try

eppa: makes me wonder how many old results are polluting the DB

it's likely some flower configuration issue

so I guess my question is: will setting CELERY_RESULT_PERSISTENT to some timedelta clean up the old results as well as any new ones?

sadly i don't use flower

eppa: I thought it was CELERY_TASK_RESULT_EXPIRES that was the issue

eppa: keeping the results?! the amqp backend deletes the results aftet they are consumed ( .get() or whatever )

ionelmc: yes, but we have nothing trying to consume the results, AFAIK

ionelmc, nope, --url_prefix=/flower/ does the same

FunkyBob: shit, paste buffer screw up sorry

ionelmc: the tasks were erroneously not marked ignore_result

ionelmc, do you know who may know?

__rodrigo__: mher is the author but he's not here

see https://github.com/mher/flower/issues/148

ionelmc: ordinarily yes, but our config also had CELERY_TASK_RESULT_EXPIRES=None which disables the automatic cleanup of results

ionelmc: because, for tasks we care about, we don’t want to run the risk of losing results

ionelmc, https://github.com/mher/flower/pull/13#issuecom... !!!!

__rodrigo__: so weird :))

i wouldn't do it like that but eh ...

ionelmc, IT WORKED!!!!!!!!!!!!!!!!

uffffff....

just add the rewrite and try to forget the horror or urlprefix-thats-not-really-a-prefix

s/or/of/

ionelmc, thanks, thanks a loooot!

so, does anyone happen to know if setting CELERY_TASK_RESULT_EXPIRES to some $seconds works retroactively on task results, or only for new task results?

i don't think so

there's no way to do it retroactively

*reliable way*

ionelmc: I see, OK

eppa: I think it depends on the backend... if it's a backend that requires celerybeat to reap old results, then I think it does, but for rabbitmq or redis no

so did you guys find the cause of the segfault ?

snewpy: OK, so I’ll have to think of something else then thanks :)

ionelmc: not directly.. just having old dj-celery setup called and using new API...

as I keep telling you

FunkyBob: well yeah, i hoped you have a more detailed explanation now :)

normal python code can't segfault - you don't have direct memory access

djcelery doesn't use ctypes or c extensions (beyond what django uses) as far as i know

yeah, I know... as soon as I saw segfault, I assumed it was a clib somewher

FunkyBob, eppa, one way would be to just purge all of rabbitmq

eciv: as eppa said, we don't want to lose results for tasks that we _do_ care about them for

just in case anyone is interested: https://github.com/mher/flower/issues/28#issuec...

FunkyBob, ah, well yup you're pretty much stuck with making a script that reads the logs for the tasks you want to purge, get their task ids, and get() those tasks

__rodrigo__, nice!

eciv, thanks! :)

ionelmc: I think our resulting fix also included completely uninstalling django-celery, celery, kombi and billiar... and then installing celery[librabbitmq] ... just to be sure

but that was before I discovered djcelery was still in the mix

I don't believe we've managed to replicate the issue in development... and we're certainly not going to set it up again in production... so, I'm afraid it's something left to the "there have been reports" pile

ionelmc: I don't suppose there's a blessed way to set ignore_result=True to be the default?

FunkyBob: I know there’s CELERY_IGNORE_RESULT config option, but can tasks override that on an individual basis I wonder?

FunkyBob: task annotations i think

ok... but still unclear is if, once I've applied this setting, I can override it on the task itself

not sure now

my impression is tthat annotations will override whatever is on tasks

yah... sort of the other end of what we're after

want to set the default, and override per task

then do everything in the annotations

i question the choice of a dict for the annotations

dicts aren't ordered ...

hrm