Re: [squid-users] Cookie problem

fre 2007-06-01 klockan 09:40 +0200 skrev Ludovit Koren:

> > tor 2007-05-31 klockan 13:39 +0200 skrev Ludovit Koren:
> >
> > > just last question. Is it planned to work correctly the cache
> > > hierarchy I mentioned earlier any time soon?
> >
> > Which was? (reading some hundreds email a day...)
> >
> > Regards
> > Henrik
> >
> >
>
> my question with your answer follows:

And what you want to know in addition? It's just a matter of configuring
the hierarchy so that request forwarding is deterministic based on the
destination or client.

ICP, HTCP and round-robin is not.

sourcehash forwarding method is.

so is having a blacklist of broken sites.

Regards
Henrik

>
>
> >>>>> On Mon, 28 May 2007 22:12:10 +0200
> >>>>> henrik@henriknordstrom.net(Henrik Nordstrom) said:
> >
> >
> > --=-d9Fl78gvdQZD+JvJS9o/
> > Content-Type: text/plain; charset=utf-8
> > Content-Transfer-Encoding: quoted-printable
> >
> > m=C3=A5n 2007-05-28 klockan 14:57 +0200 skrev Ludovit Koren:
> >
> > > The caches work ok. The problem is when a client requests a page where
> > > a cookie is set. When all the pages are served using the same parent
> > > cache it works correctly. When the parent cache changes I get an error
> > > message that the cookie is not valid (it is coming from another
> > > IP). It does not matter if I am using any options on the cache_peer
> > > configuration line (I used round-robin and sourcehash).
> >
> > The only forwarding mechanism where this works is sourcehash, but you
> > must then disable everything else (ICP, Digest, HTCP, ICMP), and parent
> > cache hit ratio will suffer.
> >
> > As a mid-approach you can build a blacklist of broken sites which
> > assumes the IP will remain unchanged for a complete browsing session and
> > send these to a set of cache_peer using sourcehash.. (might be the same
> > peers but with different names..). See cache_peer_access for how to
> > select what gets sent to which peers, and the name option to cache_peer
> > for specifying the same peer multiple times with different options.
> >
> > Regards
> > Henrik
> >