Re: [squid-users] squid3.0.25 hoggeing the CPU, serving little

* Amos Jeffries <squid3_at_treenet.co.nz>:

> > 'http://js.ui-portal.de/logout/gmx/css/100721/logout.css'
> > 'accept-encoding="gzip,deflate", user-agent="Mozilla%2F5.0%20(X11%3B%20U%3B%20Linux%20i686%3B%20de%3B%20rv%3A1.9.0.6)%20Gecko%2F2009020409%20Firefox%2F3.0.6%20(Debian-3.0.6-1)"'
> > 2010/08/11 17:42:41| clientProcessHit: Vary object loop!
> >
> > messages.
>
> Ouch. I am Mean ouch for that website.

That' GMX, Germany's biggest webmailer
See Whois:

Organisation: 1&1 Internet AG
Address: Elgendorfer Straße 57
Pcode: 56410
City: Montabaur

> Is it your website?

No :)
 
> This looks entirely expected to me. Using the User-Agent fields as part of
> Vary is not recommended. The header is usually very long and close to
> unique for every visitor. Which means that Vary will mismatch on the
> majority of requests and caches will bloat out with N copies of things like
> that favicon.
>
> IMO, it's far more efficient to instead use unique URLs for the variants
> of real content (CSS styles I assume being the big one) and issue a small
> non-cachable 302/303 redirect with Location: based on the user-agent
> decision at the server.
> The 302 (only) needs to be non-cachable to prevent crossover between
> visitor agents, while still allowing the unique per-agent content at the
> real URL to be cached for long times.

-- 
Ralf Hildebrandt
  Geschäftsbereich IT | Abteilung Netzwerk
  Charité - Universitätsmedizin Berlin
  Campus Benjamin Franklin
  Hindenburgdamm 30 | D-12203 Berlin
  Tel. +49 30 450 570 155 | Fax: +49 30 450 570 962
  ralf.hildebrandt@charite.de | http://www.charite.de