RE: [squid-users] REQUEST: Turn "reply_body_max_size" into ACL - Please!!

From: Robert Collins <robert.collins@dont-contact.us>
Date: 26 Jul 2001 11:00:37 +1000

On 25 Jul 2001 13:24:07 +1000, Robert Collins wrote:
> > -----Original Message-----
> > From: Mads Rasmussen [mailto:mads@cit.com.br]
> > Sent: Tuesday, July 24, 2001 11:53 PM
> > To: squid-users@squid-cache.org
> > Cc: Robert Collins
> > Subject: Re: [squid-users] REQUEST: Turn
> > "reply_body_max_size" into ACL
> > - Please!!
> >
> >
> > On Tuesday 24 July 2001 10:09, you wrote:
> > > The problem with doing this in a "simple" way is that the
> > ACL must be
> > > tested on every network read()/write() (choose one) for the object -
> > > because not all objects have a size. (ie dynamic objects,
> > transfer encoded
> > > objects both don't).
> > >
> > > As an ACL, the generic http_reply acl handling code will
> > handle the test -
> > > but calling that code for every network transfer will incur a large
> > > overhead. So this means we need two sorts of acls for
> > replies - ones that
> > > are checked once, and ones checked continually. OR we need to ignore
> > > transfer encoded and dynamic objects with the ACL - which means it's
> > > coverage will be less than coplete.
> > >
> > > Also: What about partial object transfers - how should they
> > be handled?
> >
> > Yeah maybe it shouldn't be turned into an ACL but we need a
> > way to create
> > exceptions. Like a list of powerusers maybe?, could this be done?
>
> An exception list is quite a bit easier yes :}. I'll see if I can put
> something sensible together.
Thinking about this: what about an approach where we calculate the limit
that will apply to a transfer, and then that result gets cached and
tested against? (Adrian - is that the result you meant? If so - Doh!
Sorry!)

Rob
Received on Wed Jul 25 2001 - 18:57:48 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:01:19 MST