On Wed, 28 Sep 2011 16:15:33 +0200, Jan Sievers wrote:
> Hi,
>
> I was recently updating squid from version v3.0.STABLE23 to 3.1.15
> and noticed, that the handling of PUT-requests with "Expect:
> 100-continue" header changed.
>
>
> While with version v3.0.STABLE23 the proxy wrote to the log file
>
> 1316481430.883 2303 192.168.0.1 TCP_MISS/200 314 PUT
> http://www.example.com/services/submit.php? - DIRECT/192.0.2.1
> text/html
>
> what triggered the client to send a request shortly after:
>
> 1316481432.180 1294 192.168.0.1 TCP_MISS/200 314 PUT
> http://www.example.com/services/submit.php? - DIRECT/192.0.2.1
> text/html
>
> Which look to me that the proxy has forwarded at least something to
> the
> origin server. I don't know if it really reported a 200 to the client
> which sounds strange for a PUT-request having a "Expect:
> 100-continue".
> Probably the client (using libcurl 7.16.1) just reached a timeout and
> issued another PUT-request without the Expect-header and with a body.
>
>
> Using version 3.1.15 the proxy wrote to the log file
>
> 1317091304.706 0 192.168.0.1 NONE/417 6181 PUT
> http://www.example.com/services/submit.php? - NONE/- text/html
>
> No other request comes hereafter. The client user reports the
> application says "Error 417".
>
> It seems the client does not repeat the POST-request without
> "expectations".
>
> On the other hand squid seems not to ask the origin server which is
> HTTP/1.1 aware and sends a 417 on it own, which at least for a
> HTTP/1.1
> proxy would be wrong.
>
> Or am I missing something?
Squid-3.1 is HTTP/1.1 on the side communicating with servers.
> Can I do something about it?
In order of preference:
* report teh problem to teh client software developers. They have
failed to handle all the RFC 2616 requirements of chunked encoding. ie
to retry without HTTP/1.1 when 417 is received.
* ignore_expect_100
> Is setting "ignore_expect_100 on" the right thing and without
> side-effects?
It is there to resolve this problem if you wish to use it.
There are side effects. Squid now passes HTTP/1.1 to servers, relaying
the Expect: headers and discards the 100-continue if it comes back. This
could make the client wait a long timeout and die anyway.
Amos
Received on Thu Sep 29 2011 - 04:20:58 MDT
This archive was generated by hypermail 2.2.0 : Thu Sep 29 2011 - 12:00:02 MDT