Re: [squid-users] Failover from direct connect to sibling

On 04.12.2012 05:41, Kim Zeitler wrote:
> Hello,
>
> we are running 2 squid servers on separate gateway machines each with
> its own internet connection.
>
> Our clients are given their configuration via WPAD.
>
> Recently we have had some trouble with the main connection and now
> would like to implement following scenario:
>
> Both Squid Servers as siblings, if the direct connection of one of
> the servers dies, it uses its sibling to provide a response.
>
> I found some older posts here, sadly only 'use direct if parent
> dies'.

That is because of the definitions of sibling and parent types. Sibling
is a cache in the same cluster/pool/group which may have a copy of the
object if we don't - they do not normally provide any upstream
connectivity. Parent is a proxy providing connectivity upstream towards
the server where the resources is hosted - they may or not provide
caching.

Siblings are also normally sharing the same upstream connectivity, so
if DIRECT has been tried and failed there is no reason to expect a
sibling to have better connection. Whereas parent is expected to be
useful as an alternative route.

The way to do this is to configure the proxies as _parent_ of each
other, with prefer_direct to make them try DIRECT access first.

If you need to share cache data between them, setup a second linkage of
type sibling for HTCP exchanges to happen on.

Amos
Received on Tue Dec 04 2012 - 00:24:49 MST