Re: [squid-users] Odd Corruption

From: Henrik Nordstrom <[email protected]>
Date: Tue, 16 Oct 2007 14:01:28 +0200

On tis, 2007-10-16 at 10:39 +0100, Alex Smith wrote:
> Hi Hendrik,
>
> We do appear to be using mod_deflate on our mediapool. Is it better to
> exclude the file from this altogether?

It's hard to say what's best, short of having mod_deflate fixed..
whatever else you do there is tradeoffs.

The mod_deflate bug is not at all isolated to any specific kind of
resource, and may be seen on any cachable and mod_deflate compressable
content.

The broken_vary_encoding directive in squid.conf is quite effective for
working around the issue, but only in your cache. Other caches handling
ETag will still break down when given incorrect input from the server
unless they have similar rules in place.

Disabling mod_deflate will slow down page loading and increase bandwidth
usage.

Simply removing the ETag header from responses will for no apparent
reason make some clients not cache the response, slowing down page views
and increasing bandwidth usage..

Applying the current Apache "patch" works around the problem, but
clients can no longer perform cache validation of compressed responses
which slows down revisits and increases bandwidth usage.. and if they
fix that with a "quick patch" adressing why validation no longer works
the problem is most likely back at square one again... but at least
there is some discussion going on on how to fix this and other design
problems with conditional requests in Apache.

Regards
Henrik

Received on Tue Oct 16 2007 - 06:01:34 MDT

This archive was generated by hypermail pre-2.1.9 : Thu Nov 01 2007 - 13:00:01 MDT