[Documentation and/or web.run_app] Default host is 0.0.0.0

[coxley]

Long story short

• Warnings for moving away from make_handler suggest Application Runners as the replacement

• make_handler returned a protocol factory. This factory could be provided to loop.create_server

• loop.create_server when host=None assumes all interfaces and will return multiple sockets (usually one for each address family)

• Both web.run_app and web.TCPSite when host=None assume 0.0.0.0

  • Confusingly, web.run_app states:

The server will listen on any host or Unix domain socket path you supply. If no hosts or paths are supplied, or only a port is supplied, a TCP server listening on 0.0.0.0 (all hosts) will be launched.

• 0.0.0.0 is not inclusive of "all hosts" unless in an IPv4-only environment

Expected behaviour

Two sockets created per port/TCPSite, matching (or at least getting close) the event loop behavior. Using host=localhost does the correct thing in my environments

I1217 165618 base_events.py:1068] <Server sockets=[<socket.socket fd=25, family=AddressFamily.AF_INET, type=2049, proto=6, laddr=('127.0.0.1', 8888)>, <socket.socket fd=27, family=AddressFamily.AF_INET6, type=2049, proto=6, laddr=('::1', 8888, 0, 0)>]> is serving

Actual behaviour

One socket is created per TCPSite for 0.0.0.0

Steps to reproduce

import asyncio
import logging
from aiohttp import web


logger = logging.getLogger()
logger.setLevel(logging.DEBUG)


async def sup():
    return web.Response(text="Sup")


async def setup():
    app = web.Application()
    app.add_routes([web.get("/", sup)])

    logging.debug("setting up test servers")
    server_fs = []
    single = web.AppRunner(app)
    await single.setup()
    server_fs.append(web.TCPSite(single, port=8080).start())
    logging.debug("AF_INET to start on: 8080")

    double = web.AppRunner(app)
    await double.setup()
    server_fs.append(web.TCPSite(double, host="localhost", port=8081).start())
    logging.debug("AF_INET + AF_INET6 to start on: 8081")
    return asyncio.ensure_future(asyncio.gather(*server_fs))


if __name__ == "__main__":
    loop = asyncio.get_event_loop()
    loop.set_debug(True)
    srv = loop.run_until_complete(setup())
    loop.run_forever()

Output:

$ python  /tmp/whatever.py
DEBUG:root:setting up test servers
DEBUG:root:AF_INET to start on: 8080
DEBUG:root:AF_INET + AF_INET6 to start on: 8081
DEBUG:asyncio:Get address info localhost:8081, type=<SocketKind.SOCK_STREAM: 1>, flags=<AddressInfo.AI_PASSIVE: 1>
INFO:asyncio:<Server sockets=[<socket.socket fd=7, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=6, laddr=('0.0.0.0', 8080)>]> is serving
DEBUG:asyncio:Getting address info localhost:8081, type=<SocketKind.SOCK_STREAM: 1>, flags=<AddressInfo.AI_PASSIVE: 1> took 1.300ms: [(<AddressFamily.AF_INET6: 10>, <SocketKind.SOCK_STREAM: 1>, 6, '', ('::1', 8081, 0, 0)), (<AddressFamily.AF_INET: 2>, <SocketKind.SOCK_STREAM: 1>, 6, '', ('127.0.0.1', 8081))]
DEBUG:asyncio:poll took 0.033 ms: 1 events
INFO:asyncio:<Server sockets=[<socket.socket fd=6, family=AddressFamily.AF_INET6, type=SocketKind.SOCK_STREAM, proto=6, laddr=('::1', 8081, 0, 0)>, <socket.socket fd=8, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=6, laddr=('127.0.0.1', 8081)>]> is serving

Your environment

Linux, 4.11

[aio-libs-bot]

GitMate.io thinks the contributor most likely able to help you is @asvetlov.

Possibly related issues are #120 (Documentation), #2591 (Documentation for server BaseRequest.host() is wrong), #2739 (Accept a coroutine in web.run_app), #1940 (Content length is 0 by default in aiohttp.web.Response), and #3243 (Enable stdout logging in run_app by default).

[asvetlov]

I believe one socket per one runner's site is a good idea.
If you want to enable IPv6 please make a second TCPSite for serving ::.
I hope it helps.

[coxley]

I'd at least document that change of behavior, while subtle, where suggesting to use Application Runners instead. Figured that'd be the want though

[asvetlov]

@coxley pull request for docs update is welcome!

[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.