Does `timeout` config can be applied for gthread worker`s Threads?

[gcavalcante8808]

Hi Folks,

We are striving to simulate timeouts for the gthread worker type. Previously, with sync, we could just put an sleep(gunicorn_timeout_in_secs+1) to verify this behaviour and it worked flawlessly.

Using gthread workers even with 1 worker and 1 thread (test purposes) it doesn`t work.

We`re trying with the following command:

gunicorn app:api -t 3 -k gthread -w 1 --threads 1 -b 0.0.0.0:8000 --log-level DEBUG --error-logfile -

The following code illustrate the problem as well:

import falcon

import time


class QuoteResource:
    def on_get(self, req, resp):
        """Handles GET requests"""
        time.sleep(5)
        quote = {
            'quote': (
                "I've always been more interested in "
                "the future than in the past."
            ),
            'author': 'Grace Hopper'
        }

        resp.media = quote

api = falcon.API()
api.add_route('/', QuoteResource())

With sync worker, gunicorn have 1 timeout per request; with gthread worker (1 worker, 1 thread) it just finish to process the request.

Components

We are using gunicorn==19.9.0.

Thanks in Advance.

[tilgovi]

It is the case that the timeout setting is not a request timeout. It functions this way for the sync worker, but not for other workers. I have noted this inconsistency before: #1974 (comment)

[gcavalcante8808]

Hi @tilgovi,

I just managed to implement a request-timeout for gthread worker by using the stopit library (which I presume that takes a function or a callable and execute it in separate timed thread) with the following changes on https://github.com/benoitc/gunicorn/blob/19.9.0/gunicorn/workers/gthread.py#L328:

            with stopit.ThreadingTimeout(6):
                respiter = self.wsgi(environ, resp.start_response)

What do you think?

[tilgovi]

There is a known issue in stopit related to Gunicorn / Gevent: glenfant/stopit#13

If it works for you and you need a quick fix, I think it makes sense for you to put this inside your own request handling code inside your application.

For gevent itself, I would like to make some decision about what the timeout setting should do, since it behaves differently on the different workers. It does have a behavior on non-sync workers, just not the same behavior as the sync worker. Before we make any change to the core, I want to understand the behavior that we want and then implement that consistently, either for the timeout setting or a new setting that is explicitly a request timeout and not a frozen / stuck worker timeout.

[gcavalcante8808]

Nice to hear!

Anyway, the more polite version of the Gthreadwork with timeout support is here:

class GThreadWorkerWithTimeoutsSupport(ThreadWorker):
    def handle_request(self, req, conn):
        with stopit.ThreadingTimeout(4):
            return super(ThreadWorkerWithTimeoutsSupport, self).handle_request(req, conn)

Do you guys have any irc or slack that we can talk? I`d like to contribute with a more featureful (async/gevent wise) version in the future.

[tilgovi]

@gcavalcante8808 there is an IRC on Freenode. I'm not there much these days, though.

We have been experimenting with using GitHub for discussion. You can label issues as "mailing list". https://github.com/benoitc/gunicorn/wiki/Project-management

[tilgovi]

For real-time communication, we can use IRC. If you want to talk, you can also e-mail directly and I would gladly join IRC to chat about something. I'll try to make a habit of being online more there, anyway.