Re: appsession fix

Hi all ;-),

On Don 14.02.2008 22:09, Willy Tarreau wrote:
>
>On Thu, Feb 14, 2008 at 01:05:17PM -0700, Ryan Warnick wrote:
>> We would like to be able to use any query parameter that we want
>> (without out it having to start with a ;).
>
>OK I see now. IIRC, the semi-colon was selected because of common use
>on existing applications (Aleks correct me if I'm wrong). The
>semi-colon was not the start of the parameter name, but used as a
>separator between previous parameter and the one we care about. I think
>that we should just check that the parameter is either at the beginning
>of the query string, or preceeded by a non-alphanum (nor any of '_',
>'-' and a few other ones).

You are absolute right ;-)

As described in http://gbiv.com/protocols/uri/rfc/rfc3986.html the ; is a valid value in the uri.

The most 'current' application servers use the '...;$PART?' $PART to add a session id if the user have disabled cookies.

That is also the reason for the following code:

---haproxy-1.3.14.2/src/proto_http.c:get_srv_from_appsession() .
.
(request_line = memchr(begin, ';', len)) == NULL .
.

        /* skip ';' */
        request_line++;
.

---

Why do you don't use the url_param methode?!

--- from haproxy-1.3.14.2/doc/haproxy-en.txt
Version 1.3.14 introduced the "balance url_param" method.
---

Maybe it will be a good idea to add a hook for the appsession into to
url_param methode, if it possible?!

Which application is this, a self developed one or a commercial, yust
for my curiosity, if you allowed to say.



>> Also, thanks for your critique of our proposed change.  You are

>> absolutely right, we should use something like memmem instead.

>

>I'm just realizing that since the cookie name is fix and checked for

>every request, it may make sense to use a boyer-moore search method,

>and construct the table at configuration time. That way, the longer the

>cookie, the faster the scan. And with an average size around 10 chars

>(JSESSIONID, ...), it would make parsing 10 times faster.


As http://www.stedee.id.au/awffull does ;-)

---from ChangeLog
27-Jan-2006     Steve McInerney <spm#stedee.id.au>
awffull-3.3.1-beta1.tar.gz
.
.
.
         * Implement Boyer-Moore-Horspool for pattern matching (vs strstr).
             Approx. 50% faster overall from 3.02, depending on the GroupXXX
             complexity!
.
.
.
---

Steve and I talked about some algorithms and we agreed this was the
'best fit' algorithms for us ;-)

As you remember Willy, some times ago we talk about search algorithms
and http-parsing ;-)

Cheers

Aleks