Browsers often use multiple connections with Connection: keep-alive.
If --nothreading is specified, the WSGI server cannot accept new
connections until the old connection is closed, causing hangs.
Force Connection: close when --nothreading option is used.
Ticket #25619 changed the default protocol to HTTP/1.1 but did not
properly implement keep-alive. As a "fix" keep-alive was disabled in
ticket #28440 to prevent clients from hanging (they expect the server to
send more data if the connection is not closed and there is no content
length set).
The combination of those two fixes resulted in yet another problem:
HTTP/1.1 by default allows a client to assume that keep-alive is
supported unless the server disables it via 'Connection: close' -- see
RFC2616 8.1.2.1 for details on persistent connection negotiation. Now if
the client receives a response from Django without 'Connection: close'
and immediately sends a new request (on the same tcp connection) before
our server closes the tcp connection, it will error out at some point
because the connection does get closed a few milli seconds later.
This patch fixes the mentioned issues by always sending 'Connection:
close' if we cannot determine a content length. The code is inefficient
in the sense that it does not allow for persistent connections when
chunked responses are used, but that should not really cause any
problems (Django does not generate those) and it only affects the
development server anyways.
Refs #25619, #28440.
One handler in WSGIServer, to catch the error when raised from
SocketServer.BaseServer's finish_request, and one in WSGIRequestHandler
(by creating a subclass of ServerHandler), to catch the error when
raised in wsgiref.handlers.BaseHandler's finish_response.
Filtering out static file requests in runserver has been judged
arbitrary and can hide some debugging-related activity.
Thanks Roy Smith for the report and Aymeric Augustin for the
review.
Our WSGIServer rewrapped the socket errors from server_bind into
WSGIServerExceptions, which is used later on to provide nicer
error messages in runserver and used by the liveserver to see if
the port is already in use. But wrapping server_bind isn't enough since
it only binds to the socket, socket.listen (which is called from
server_activate) could also raise "Address already in use".
Instead of overriding server_activate too I chose to just catch socket
errors, which seems to make more sense anyways and should be more robust
against changes in wsgiref.
We probably historically customized it for good reasons, but
currently, the differences with upstream Python are not
significant any longer.
Also fixes#19075 for which a test has been added.
atsuo ishimoto
Jon Dufresne
Carlton Gibson
Konstantin Alekseev
Florian Apolloner
Tim Graham
Tom
Claude Paroz
Anton Samarchyan
Nadège Michel
Vytis Banaitis
chillaranand
Simon Charette
Simon Charette
David Sanders
Flavio Curella
Marten Kenbeek
Alex Vidal
Carl Meyer
Andreas Pelme
Matthew Somerville
Anubhav Joshi
Moayad Mardini
Ramiro Morales
Berker Peksag
Aymeric Augustin
Matthew Wood
konarkmodi