When socketserver.BaseServer.verify_request() return False then we do not call shutdown_request. If we will take the TCPServer as example we will call get_request thus calling socket.accept() and creating a new socket but we will not call shutdown_request to close the unused socket.

Had a small mistake in the previous patch (did not notice process_request) call shutdown_request.

fixed the patch

The patch looks good to me. Are you interested in writing a unit test for it? Having said that, it might be a tricky test to write.

I added a test to check the specific bug.

There seems to be no testing at all on the verify_request feature so I will create some in different issue in the future.

Thanks for the test, but it is a bit confusing to have the shutdown_request() method call finish_request() and actually do normal request handling. Maybe it would be better to not handle the request (and not test that a response is received), and just check that shutdown_request() is called or that the client socket is explicitly closed.

I changed the test to just check that shutdown_request is called. Hope this is more clear then the previous test.

Yes this patch looks pretty good, thanks

New changeset 651a6d47bc78 by Martin Panter in branch '3.5':
Issue #26309: Shut down socketserver request if verify_request() is false
https://hg.python.org/cpython/rev/651a6d47bc78

New changeset 0768edf5878d by Martin Panter in branch 'default':
Issue #26309: Merge socketserver fix from 3.5
https://hg.python.org/cpython/rev/0768edf5878d

New changeset e0fbd25f0b36 by Martin Panter in branch '2.7':
Issue #26309: Shut down SocketServer request if verify_request() is false
https://hg.python.org/cpython/rev/e0fbd25f0b36

For the Python 2 version I had to make some small changes to the test. I used indirection instead of “nonlocal”, and replaced the super() call because the classes are apparently the wrong kind for super().

Some of the buildbots failed (e.g. <http://buildbot.python.org/all/builders/ARMv7%20Ubuntu%202.7/builds/1145/steps/test/logs/stdio>). I think the server is run in a separate thread, and the problem is a race between shutdown_request() being called and run_server returning.

New changeset cba717fa8e10 by Martin Panter in branch '2.7':
Issue #26309: Rewrite test in main thread and avoid race condition
https://hg.python.org/cpython/rev/cba717fa8e10

New changeset 537608bafa5a by Martin Panter in branch '3.5':
Issue #26309: Rewrite test in main thread and avoid race condition
https://hg.python.org/cpython/rev/537608bafa5a

New changeset c791d57c8168 by Martin Panter in branch 'default':
Issue #26309: Merge socketserver fix from 3.5
https://hg.python.org/cpython/rev/c791d57c8168

The race depended on how whether serve_forever had a chance to hande the first request before the main thread told it to stop.