RE: [squid-users] cache_peer and acl random problem

From: Jenny Lee <bodycare_5_at_live.com>
Date: Fri, 30 Jul 2010 16:44:11 +0000

> >> At least I can point to a full disabling list. Look for test-suite/build
> >> tests/layer-01-minimal.opts. Thats everything that can safely be turned
> >> off. Though you may find some of them are wanted.
> >
> > That is what I am looking for! Unfortunately, there is no such list or reference to it in any file in:
> >
> > 3.1.5
> > 3.HEAD
> >
>
> Oops, sorry I keep forgetting those are not bundled.
> You can find a copy of them here:
>
> http://build.squid-cache.org/job/3.HEAD-amd64-CentOS-5.3/ws/test-suite/buildtests/
 
Thank you! Exactly what I was looking for!
 
 
 
> They build in code that hard-codes the particular file formats. Slightly
> faster to log with and less RAM used. My plans are to replace the
> options with built-in formats like "squid" and "common" are already.
> That will means some of their code is always-present in future, but for
> now it can be built without.
>
> >
> >> underscores can be turned off. It enables a DNS violation which makes
> >> Squid accept and lookup invalid domains like broken_host.example.com
> >
> > People want that. They say, your cache is not working (when accessing these hosts). Better to keep it on than dealing with them.
> >
>
> It only works internally because the public DNS does not support them. I
> always wondered what types of networks those people ran.
 
What do you mean it works internally? Users browse websites totally unrelated to us.
 
BIND has a check-names option that would allow underscores. Everytime we turned this off, users complained that cache is broken.
 
I must admit I haven't followed this up quite a while. But we keep it on.
 

> > Is this because of an error on my configuration?
>
> I don't think so. Something else really, really weird is going on.
>
> Those log lines are built from an array of strings using a code number
> for each piece fo text. The CD_* entries are not even built unless you
> have "#define USE_CACHE_DIGESTS 1" in your autoconf.h
>
> My thoughts is perhapse something happend to build an old copy of some
> objects. See if a completely clean build still has it.
 
All my builds are clean builds, they run from a script. squid directory is removed, tarball unzipped and configure ran.
 

> > Behaviour is how I want. Only this log entries bothering me. Over the years, I came to greatly trust squid logs. They don't lie. If they point out an error, it is because of me... or Microsoft.
> >
> > On my main question would the hit go direct with this config if parent2 or parent3 down?
>
> Depends on the results of random when it was tested to find the peers.
>
> To find FIRST_UP parent random is run once for each by your
> cache_peer_access allow rule.
>
> * if the first 'fails' its 1/3 chance but second passes. the second is
> used, and direct is a backup method.
>
> * if first passes, the second is still tested and still has 1/3 chance
> of being next on the backup list if first is down. direct as a
> guaranteed backup again.
>
> * if both fail their 1/3 tests backup is used.
 

Thank you.
 
J
 
 
 
 
 
 
 
 
 
 
 
 
 
                                                
Received on Fri Jul 30 2010 - 16:44:18 MDT

This archive was generated by hypermail 2.2.0 : Sat Jul 31 2010 - 12:00:05 MDT