Make SO_REUSEPORT Optional

[martinbroadhurst]

I'm open to opinions as to whether SO_REUSEPORT should be the default or not.
Also, would appreciate some advice about unit testing this change.

[tilgovi]

I am slightly in favor of a False default. This option matters most for users seeking the performance benefits, I think. Those users may know that they want this and they are likely to read through the server options for tweaks. New users, however, can sometimes be killing and restarting their server over and over while testing out code in development and reusing the port can sometimes lead to having two instances of Gunicorn running on a development machine, serving different code. To minimize confusion, I would make it False by default. I'm open to arguments for True, though.

No strong opinion on how to test this.

[martinbroadhurst]

I agree with you about the default being False. I have changed the code accordingly.

[tilgovi]

This looks great to me. Thank you for opening a PR. I'll give others a chance to weigh in on whether we want to test this at all, otherwise I'll merge in the next day or so.

[martinbroadhurst]

Thanks. I have updated the documentation.

[berkerpeksag]

Documentation for the new option will be generated automatically from gunicorn/config.py when you run make html in docs/ directory.

[martinbroadhurst]

I'm not sure I understand your comment, @berkerpeksag. I generated this change by running make -C docs html. Am I not supposed to commit settings.rst?

[berkerpeksag]

I thought you manually edited it because I didn't see the "Set the SO_REUSEPORT flag on the listening socket." part in the generated file. Now I don't know why make html didn't get the documentation in the desc attribute.

[martinbroadhurst]

I generated it, but then edited it because running it also brought in logconfig_dict (which presumably was omitted previously and should be in its own PR), and also changed chdir, user and group. I must have got the edit wrong. Sorry about that. It's all there now.

[berkerpeksag]

Please add .. versionadded:: 19.8.

[tilgovi]

I always forget this and you always catch it. Thanks! :-D

[berkerpeksag]

The cause of build failure is prospector-dev/prospector#200. All tests passed and I don't think we need a test case for this since testing the actual change reliably would be hard.

[berkerpeksag]

Thanks!