Hardware Load Balancer problem

[Eden Li]

It sounds like your load balancer is configured with a default tcp
ping health check.  Since varnish will respond to tcp packets even if
apache goes down, you should either take varnish down before taking
apache down, or you should configure some sort of health check that
takes this into account.  For example, you can set up some url that
only apache knows how to serve and prevent varnish from caching it.
This will let the health check fail successfully if apache is no
longer running.

On Fri, Oct 31, 2008 at 3:05 PM, Scott Persinger <scott at vodpod.com> wrote:
> We recently switched from serving directly from Apache to using Varnish in
> front as a cache.
>
> So we have:
>
>   varnish -> apache -> mongrel (rails)
>
> and in front of varnish we have a hardware load-balancer (as a service of
> our hosting provider, softlayer) which
> balances traffic to 3 servers.
>
> In the past, as soon as we took down apache on one of our servers, the load
> balancer would detect it and route
> all traffic to the other servers.
>
> However, since we started running varnish, the load balancer doesn't seem to
> detect that a server is down, and so
> it keeps sending traffic even after varnish has quit.
>
> Has anybody run into any similar problems, or have any ideas why we would
> see different behavior running Varnish
> as opposed to simply Apache?
>
> Thanks,
>
> --Scott Persinger
>
> _______________________________________________
> varnish-dev mailing list
> varnish-dev at projects.linpro.no
> http://projects.linpro.no/mailman/listinfo/varnish-dev
>
>