Failover cluster for webserver with dynamic content ?

[Feczak Szabolcs]

This happend on the freebsd-questions list, but I got no 
answer for days. Joshua Tinnin advised me to make a try
here. I hope some of you can answer this:

On k, 2005-04-19 at 17:21 -0400, Paul Mather wrote:
> to what you describe using geom_gate for the remote component.  See
> ggated(8) and ggatec(8) for how to set up an use a geom_gate provider.

> Note that the geom_mirror + geom_gate synchronisation would be
one-way.

Bad luck, I would like to have something that creates a layer over the
two volume of the machines, and when this higher layer accessed both
execute the requested operation.

Anyway one step further, my question is 
How can I create a failover cluster with two machines 
for a freebsd webserver with dynamic content 
runing  apache with php, and postgresql.

I read about CARP, but more experienced people advised me to use
DNS-LB since its more reliable with service type pings (HTTP GET) 
than simple is the machine answers for TCP SYN. They made a point with
that to me. 

Im trying to syncronize the postgresql database with Slony, no luck
yet, all the examples I found describing master and slave on the
same machine. I got slony communicate between the two, but on updates
nothing happens on the slave. I access the master on unix socket,
maybe other type of access needed .. hm I will see 

If on failure the switching is done with DNS-LB and the SQL is in sync
Im nearly OK, but since I have file uploads on the webserver as well, 
I need a shared volume which available to both of them and after
one is out the other still has access to the data.

Maybe Im complettly wrong I have no clear ideas about what happens
when this ... and what happens whan that ... scenarios

All I want is a higher availability with two machines than one
and without messing up the consistency of the data of course.
Im not after chasing nearly 100% ... the policy/expectation is 
if one fails the other should automatically continue the 
serving data (nearly there) where the other stopped.

If anyone did something like that, and aware of some solution without
buying expensive HA hardware, please share us.
Hope this is possible at all.