Re: [squid-users] Clustering/Redundancy with reverse proxy accelerators

From: Christopher Arnold <[email protected]>
Date: Fri, 17 Oct 2003 12:09:46 -0700

Henrik Nordstrom wrote:

>On Thu, 16 Oct 2003, Christopher Arnold wrote:
>
>
>
>>and likewise if a request comes in to Accelerator-B, contact
>>Accelerator-A in the same fashion. I thought that a dual-sibling
>>relationship would accomplish this but from what I am observing there is
>>no ICP communication between two siblings in a parent-less environment.
>>
>>
>
>Then you probably have made some kind of configuration error..
>
>how have you configured the peers?
>
>is there any error messages in cache.log?
>
>

I've tried several different methods:

The first I tried was to configure Accelerator-A and Accelerator-B as
siblings with no parent.

#####BOTH
http_port 80
icp_port 3130
httpd_accel_host cache.os.pictage.com
httpd_accel_with_proxy on
httpd_accel_port 80
http_access allow all
icp_access allow all

#####Accel-A
cache_peer accel-B.pictage.com sibling 80 3130

#####Accel-B
cache_peer accel-A.pictage.com sibling 80 3130

In this configuration I never see ICP traffic or communication between
peers. If a request is sent to
Accel-A it immediately goes DIRECT unless A has it cached. If a request
is sent to Accel-B it also
immediately goes DIRECT unless it has it cached. netstat shows that
each peer is listening on 3130.
A tcpdump and iptables LOG rule show that the two peers never contact
each other.

The second method I tried was simply to change peer type from sibling to
parent:

#####Accel-A
cache_peer accel-B.pictage.com parent 80 3130

#####Accel-B
cache_peer accel-A.pictage.com parent 80 3130

In this configuration if a request is sent to Accel-A then ICP traffic
occurs and Accel-B's cache gets
checked. If it doesn't have it then instead of sending a response back
to Accel-A it goes to the origin
server, caches the url, then sends the data to Accel-A.

Which configuration should I use to get an accelerator to retreive the
file itself if the other cache servers don't
have it but do get it from a cache server if it *is* cached among one of
the siblings?

I also tried setting the never_direct setting when I had two siblings
and it explained that
this results in an error without a parent.

>Regards
>Henrik
>
>
>

-- 
Christopher Arnold
System Administrator
Pictage, Inc.
Received on Fri Oct 17 2003 - 13:09:47 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:20:32 MST