RE: [SQU] The no_cache command is troublesome for already-cacheobjects

From: Robert Collins <robert.collins@dont-contact.us>
Date: Fri, 8 Dec 2000 14:42:08 +1100

> -----Original Message-----
> From: Chris Conn [mailto:cconn@abacom.com]
> Sent: Friday, 8 December 2000 2:18 PM
> To: Robert Collins
> Subject: Re: [SQU] The no_cache command is troublesome for
> already-cacheobjects
>
>
> On Fri, 8 Dec 2000, Robert Collins wrote:
>
> > You are correct in that no_cache doesn't flush the cache of
> current hot or warm cache objects. Search the list for 'purge' and you
> > will see discussions on how to do that. That should solve
> your problem for existing cached objects.
>
> Hello,
>
> I appreciate the time you took to answer me. I have looked
> up the PURGE
> item and had already experimented with the ./client program to purge
> objects, however I have to purge them one by one. Is there
> not a way to
> have squid delete them as it finds them in client requests?
>
> Maybe I can ask my question another way; in the case a user contacts a
> cache administrator due to a site handling cached requests
> badly, how does
> one make sure requests bypass the caching and are always
> directed to the
> origin server? I am looking for a feature similar to the
> "bypass" feature
> of the cacheflow system, better yet, one that allows the client's IP
> address to be forwarded to the origin server and not the cache's ip
> address for sites that require IP address validation. Is
> this possible?
>
> I am using wccp and ipchains on a Linux box to provide my
> caching, and my
> only solution so far is to have a bunch of RETURN rules for the IP
> addresses of cache-unfriendly sites previous to the
> REDIRECT rule, and configure my router to not send to the
> cache packets
> that are inbound on the cache's interface to the router.
> This has forced
> me to also have an interface in the router dedicated to
> squid. This way,
> squid never actually interferes in the TCP exchanges between
> the clients
> and the origin servers, however I suspect it is quite IO
> intensive on the
> server running squid, and probably the router (and this is not taking
> added latency into account due to the less than direct
> routing path taken
> by the packets).
>
> I thank you again for your time,
>
> Chris

Hi Chris, Please keep replies at least cc:'d to the list.

You are using intercepting caching by virtue of WCCP. That is almost
certainly causing your cache friendlynesss problems due to older
browsers (IE Pre 5.5 sp 1, netscape may not be fixed) not sending
cache-control headers when no cache is configured in the browser. The
sites may be perfectly ok, but your browsers are showing the problem.

There are known problems in the http specification with intercepting
(also incorrectly known as transparent) caching.

To see if it is the site or the browser, configure a browser to use the
cache without WCCP, and see if the site is really broken. If it is then
the no_cache directive is appropriate.

How do I solve it when admining?

* I use an interception tool (WCCP/route-maps/whatever) to direct www
requests to a web server that shows a page on how to configure the users
browser.
* I use wpad & .pac files for the actual configuration to minimise user
error.

So the users cannot browse without configuring their browser proplery.

And I have _never_ run into sites that are 'cache unfriendly' with squid
using that system.

Rob

--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html
Received on Thu Dec 07 2000 - 20:47:06 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:56:53 MST