- From: Willy Tarreau <w@1wt.eu>
- Date: Thu, 5 Jun 2014 00:04:00 +0200
- To: Michael Sweet <msweet@apple.com>
- Cc: "William Chan (?????????)" <willchan@chromium.org>, HTTP Working Group <ietf-http-wg@w3.org>, Matt Menke <mmenke@chromium.org>
Hi Michael,
On Wed, Jun 04, 2014 at 05:00:52PM -0400, Michael Sweet wrote:
> Willy,
>
> I can go along with a 408 (and only a 408) response being sent by a server
> right before a connection is torn down due to inactivity. Document this as
> "the server MAY send a 408 response prior to closing a connection and a
> client SHOULD use the presence of a 408 response as an indication that the
> request should be retried" or something along those lines, with the
> discussion below distilled to the critical bits.
Concerning the second part of the sentence, I think that the existing text
already covers it (though it uses MAY instead of SHOULD as we shouldn't
dictate to the client how it needs to handle the retry I guess) :
"If the client has an outstanding request in transit, the
client MAY repeat that request on a new connection."
> We should also be clear
> that this problem is most pronounced when using small inactivity timeouts.
I think that it might be worth mentionning it, yes, so that clients decide
whether or not they want to implement it depending on their target use case.
Eg: maybe your cups client does not have any benefit in retrying in such a
case (I have not idea, just an example).
Regards,
Willy
Received on Wednesday, 4 June 2014 22:04:28 UTC