Re: [squid-users] How to determine why an object isn't being cached?

From: Denis Haskin <dwhaskin@dont-contact.us>
Date: Wed, 11 Apr 2001 16:23:01 -0400

Henrik Nordstrom wrote:

> That said, whan you describe above can be done in HTTP by combining Vary
> and "cache-control: public" or a similar directive. Say that the reply
> varies based on Authorization: but is public. This way the shared cache
> can be abused to implement private caches.
>
> Vary support in Squid is available from
> http://squid.sourceforge.net/vary/

I've applied Henrik's vary-2_4.patch and it appears that things are working
as I expect, at least after preliminary testing. That is, I've constructed
origin response headers with "Cache-control: public" and "Vary:
Authorization", and Squid appears to be correctly serving cached content on a
per-credential basis. For my application that's okay, because our
credentials don't really represent individual users, but fairly large groups
of users, so caching per credential is still useful (and in fact required).

I need to fool with the headers a bit more and test in more detail,
particularly to examine what happens when the cache entries go stale.

One surprising thing I encountered is that Netscape 4.76 (Win32) appeared to
be sending a "Pragma: no-cache" on its initial request for the content. I
need to look at this more, but I'm sort of perplexed as to why it would do
that.

Also, I made the foolish mistake of also applying some 2.4 patches ("Invalid
object timestamp calc" and "Kill parent squid...") at the same time as the
Vary patch, and the resulting binary was dumping core on startup. I haven't
yet isolated why or checked the bug database (was failing while parsing the
config file), but I will try to do so. When I backed out those patches and
stuck with the Vary patch, all was hunky-dory.

Thanks to all for suggestions, pointers, tips, etc so far.

dwh
Received on Wed Apr 11 2001 - 14:23:48 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:59:17 MST