Re: [squid-users] Squid Cluster

Serge Fonville wrote:
> Hi,
>
> I am in the process of setting up a two node cluster. Since I am
> setting up Squid on the private side. I am looking into how data is
> stored. Basically I am trying to detemine what data can be shared
> between instances of Squid. Say a host opens a connection to a remote
> location and collects all kinds of state information and that squid
> goes down. will that mean that all information is kept? Can a download
> that is broken (due to the squid going down) halfway, still continue
> while happily using the other (which has the same IP)

The download-in-progress will be interrupted. If a range request is
made, asking for the "rest" of the object, the remaining Squid will
honor that request, but the object fragment will not be cached
(depending on your range_offset_limit).

> Are there any specific things I need to look in to?
>
> I intend to set up a cluster consisting of the following
> GlassFish
> PostgreSQL
> Nagios
> Postfix
> Squid
> Heartbeat
> Subversion
> Named
> DHCPd
> TFTPd
> Apache HTTPd
> DRBD (dual primary with either GFS2 or OCFS2)
> ldirectord or keepalived (all traffic is being balanced between the
> two real servers and both nodes should be active)
>
> It will probably run on either Gentoo or Centos x64
>
> What are the important thing in regard to squid that I need to take
> special attention to.
> Things I can imagine
> IPaddress sharing
>

Works fine.

> Storage (cache) sharing
>

Not supported.

> Synchronization of configuration
>

To a point. "cache_peer" and either "visible_hostname" or
"unique_hostname" will be different. But you can use an "include" to
achieve that.

> Sharing of logon data to squid (I intend to use form based
> authentication for squid)
>

This will have to be accomplished by your external_acl_type helper.

> I am aware of the fact that the majority of this setup is in no way
> relevant to squid, but it may impact it (I can not yet determine that)
> I am especially interested in anything that is relevant to impact in
> regard to availability, performance and load balancing
>
> Any help is greatly appreciated!!
>
> Regards,
>
> Serge Fonville
>

Chris
Received on Thu Jul 02 2009 - 21:28:24 MDT