W3C home > Mailing lists > Public > ietf-http-wg@w3.org > April to June 2003

Re: mistake in section 8.1.2 of RFC2616?

From: Alex Rousskov <rousskov@measurement-factory.com>
Date: Wed, 30 Apr 2003 11:11:55 -0600 (MDT)
To: Pierre Phaneuf <pphaneuf@nit.ca>
cc: ietf-http-wg@w3.org
Message-ID: <Pine.BSF.4.53.0304301058080.91392@measurement-factory.com>


On Wed, 30 Apr 2003, Pierre Phaneuf wrote:

> Alex Rousskov wrote:
>
> >> In section 8.1.2, you can read the following:
> >>
> >>> Once a close has been signaled, the client MUST NOT send any more
> >>> requests on that connection.
> >>
> >> Is it just me, or that "MUST NOT" would forbid pipelining? A
> >> pipelining client MAY send more requests, but they will be ignored
> >> by the server and it should be prepared to that possibility
> >> (section 8.1.2.2 mentions that a client MUST be prepared to the
> >> server closing the connection before it handled them all).
> >
> > This is not meant to prevent pipelining. In this context, you should
> > interpret the above RFC wording as "once the close signal has been
> > received by the client, the client MUST NOT send any more requests on
> > that connection." as opposed to your current interpretation of "once
> > the close signal has been sent by the server, the client ..."
>
> I know that this is the spirit of the document, but maybe it should
> be clarified in the errata.

Perhaps. Although if we get that picky, the errata will get longer
than the RFC itself :-).

> I also suspect that it could be fine sending more requests to the
> server in a simplistic client implementation, since they will be
> harmlessly ignored by the server, but on the other hand, this
> requires being ready to re-send those requests, so while you're at
> it, better do it right.

Depends on your definition of "fine", I guess. The client will be
wasting bandwidth and other resources. The server HTTP stack is indeed
likely to see no "extra" requests after the TCP connection is closed
(which usually happens right after the final server response is sent).
However, asynchronous I/Os, half-closed TCP connections, SSL/TLS
wrappers, and other real-world complications make the situation less
predictable.

As for re-sending pipelined requests, we already have a MUST for that:

   Clients MUST also be prepared to resend their requests if
   the server closes the connection before sending all of the
   corresponding responses.

Alex.

-- 
                            | HTTP performance - Web Polygraph benchmark
www.measurement-factory.com | HTTP compliance+ - Co-Advisor test suite
                            | all of the above - PolyBox appliance
Received on Wednesday, 30 April 2003 13:12:22 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Friday, 27 April 2012 06:49:23 GMT