Python 3.14.0a2 should have raised exception when a socket is already in used, shouldn't it?

[rayluo]

Bug report

Bug description:

In earlier versions of Python all the way to 3.13, it would not allow a server to listen on a port that has already been used, and rightfully so.

$ python -m http.server 8000
Serving HTTP on 0.0.0.0 port 8000 (http://0.0.0.0:8000/) ...

# Trying to start the second server on the same port
$ python -m http.server 8000
Traceback (most recent call last):
  File "<frozen runpy>", line 198, in _run_module_as_main
  File "<frozen runpy>", line 88, in _run_code
  File "/usr/lib/python3.11/http/server.py", line 1309, in <module>
    test(
  File "/usr/lib/python3.11/http/server.py", line 1256, in test
    with ServerClass(addr, HandlerClass) as httpd:
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.11/socketserver.py", line 456, in __init__
    self.server_bind()
  File "/usr/lib/python3.11/http/server.py", line 1303, in server_bind
    return super().server_bind()
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.11/http/server.py", line 136, in server_bind
    socketserver.TCPServer.server_bind(self)
  File "/usr/lib/python3.11/socketserver.py", line 472, in server_bind
    self.socket.bind(self.server_address)
OSError: [Errno 98] Address already in use

But in Python 3.14.0a2, the second command would also start a server without exception. Is that expected?

CPython versions tested on:

3.14

Operating systems tested on:

Linux

[skirpichev]

I confirm this in the main. It seems, for two processes, the last started process - serves requests. If it's terminated - works the first.

At least, seems as a documentation issue for me (nothing in news).

[oskar-skog]

I believe it's caused by this commit:
192d17c

See also comment: #120485 (comment)
Maybe allow_reuse_port should default to False?

[rayluo]

I confirm this in the main. It seems, for two processes, the last started process - serves requests. If it's terminated - works the first.

At least, seems as a documentation issue for me (nothing in news).

Based on #120485 (comment) (thanks for the finding, @oskar-skog ), "the kernel may dispatch the request to different server process randomly", which is concerning.

Maybe allow_reuse_port should default to False?

+1