Cached aiohttp.web.Response objects cause requests to hang

[lpedrosa]

Long story short

If you use a cached aiohttp.web.Response in a middleware function it will cause any subsequent requests to hang.

Expected behaviour

After thinking for a bit, I suggest that we either:

• Improve documentation, by showing an example of what to do if you wish to cache the response body, or
• Make the response object re-usable between requests. It can still remain unchangeable after prepare() has been called, preserving its semantics. However, this is probably not feasible with the current architecture.

Actual behaviour

As far as I understand, aiohttp.web.Response is a Finite State Machine. Once a Response
object is dispatched by the server, it must be emptying some internal buffer. This renders it unusable for subsequent dispatches.

The current solution for avoiding this behaviour seems to be having the middleware generate a new Response object, everytime it is called.

Steps to reproduce

Here is a simple application that reproduces the behaviour:

from aiohttp import web


def my_middleware_factory():

    # this could have been passed in as a default arg, or exist as a global
    state = web.Response(status=200, text='this is cached')

    @web.middleware
    async def middleware(request, handler):
        res = await handler(request)
        return state

    return middleware


async def handler(request):
    return web.Response(status=200, text='Hello World!')


def main():
    app = web.Application()
    app.router.add_get('/', handler)
    app.middlewares.append(my_middleware_factory())

    web.run_app(app)


if __name__ == '__main__':
    main()

• When you hit http://localhost:8080 twice
• Then the second request should hang your http client

Your environment

Python 3.6
aiohttp=3.2.1
Ubuntu 16.04
'server'

[lpedrosa]

I'm glad to help augmenting the current middleware factory usage examples for this. Your framework is really helpful!

I have run into this behaviour while building a generic Exception to Response middleware factory. My current workaround can be found here

The to_response() function generates a new web.Response everytime it is called.

[asvetlov]

Yes, web response cannot be used multiple times to answer different requests.
You should recreate it every time, the response object cannot be cached.
The rule of thumb: one request -- one response.

If you want to make a documentation pull request -- you are welcome. I doubt if we need to blow up the main docs but FAQ sounds like a good place for such things.

[lpedrosa]

The FAQ sounds like a great place to put it. Let me try to figure out how to do it.

The rule of thumb: one request -- one response.

Thanks for clarifying this!

[asvetlov]

@lpedrosa ping

[lpedrosa]

Hi @asvetlov. Sorry about going MIA. I'll create a PR in a couple of minutes, containing the basic draft of the new FAQ entry.

[asvetlov]

Fixed by #3129

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a [new issue] for related bugs.
If you feel like there's important points made in this discussion, please include those exceprts into that [new issue].
[new issue]: https://github.com/aio-libs/aiohttp/issues/new