#django

anybody here deploying a django app with gunicorn and an async worker? i'm giving it a try but requests are still being processes sequentially.

define async worker?

and what makes you think that they are processed sequentially

it's a gunicorn worker

also how did you make your database connection async etc…

yeah but which type of "async"

the one that isn't synchronous apollo13

duh

apollo13: i set "worker_class" to "gthread" (i've tried others, too): http://docs.gunicorn.org/en/stable/settings.htm...

stickperson: well async doesn't "just" work, you have to ensure that your code works with it too

what makes you think that you need async workers?

apollo13: with a sync worker i'm running 8 workers and can only process 8 requests at a time. thinking about handling more than than.

stickperson: did you set the number of threads?

knbk: i did not. i thought 1 would be fine with an async worker.

stickperson: 8 requests at a time at 100ms/request is already 80 requests / second, what are you trying to achieve?

gthread workers default to 1 thread

apollo13: i don't have a set number, but our db queries are slooow so i figured i'd try to work on something else while someone fixes them.

stickperson: then changing to async will not do anything I think

especially not with gthread

apollo13: hm, even if the query takes say 3s when there is 1-10 concurrent requests?

a) python has a GIL, so only one thread can ever run, that is unless the db adapter properly releases the gil

at which point a properly threaded sync worker has roughly the same throughput

and a query that takes 3s on a webpage is something you should get rid of yesterday, especially if it is a customer facing page

apollo13: ok, so like you said earlier if i did want to give async a try i'd have to write the db connection/call to be async as well

that makes sense. makes me wonder if anyone actually does use an async worker with django

they do

but you need to think about the conesequences

killed our db

async gevent worker + monkey patched psycopg will easily spawn 1000 connections to the db

ie dead

so you'd want connection pools and whatnot

yea

but the easy fix would be to fix your queries

well alright then. i'll wait until the queries are fixed and will take it from there.

thanks for the info

actually, we have a raid of ssds now for the db. maybe i should have another look..

I'd personally still stay away from async

yeah, the overhead probably isn't worth it

at least as long as possible

and the slow quries still fill up the queue

true

the gthread worker isn't really async, is it? It's just a thread pool where each thread synchronously handles requests

knbk: yeah, at least that is what I read from the docs

gaiohttp would be async