Hi Patrick,
We thought about that and that is a possibility but we haven't got yet
to the "failover setup".
The possibility we are studying is to cluster the application servers
"behind" the proxy and let the proxy be as simple as possible. The
failover would be in this case, transparent. Another possibility would
be to let the SSI script issue a redirect to another place, but this
would mean an external redirect (not transparent to the user). If you
want this thing to be transparent, you would have to modify the Apache
configuration to modify your proxy settings to point to the new
location. I haven't tried but I'm not sure this can be done dynamically
. That's what we do manually when we modify an application server
location but I wouldn't call this failover ;). In the end, you would be
kind of replicating, through Apache and SSI, what container-clustering
is already supposed to give you so... why?
Next step I want to try is to use the proxy also as a cache for static
content, hence improving the speed of the content that doesn't need to
be dynamically generated. I already tried but I have to find the proper
settings as it is caching more than what I need. Then we would just have
the extra trip for the dinamically generated content.
Dan
> Patrik Andersson wrote:
>
> Ok,
>
> very interesting to hear ways to setup large applications. But in
> other
> words, the SSI script that shows the "For maintenence reasons..."
> message could also be used as some kind of fail over?
>
> Patrik
>
> -----Ursprungligt meddelande-----
> Från: Daniel López [mailto:[EMAIL PROTECTED]]
> Skickat: den 30 maj 2001 15:15
> Till: Orion-Interest
> Ämne: Re: SV: proxying orion with IIS or iPlanet Web Server
>
> Hi Patrick,
>
> I agree with you, in our case we are using Apache as a proxy for other
>
> reasons. One reason is that the Apache SSL certificate that we have
> doesn't work with Orion (quite easy to fix) but the main reason is
> that
> we use several instances of Orion, one per set of related
> applications,
> and we use the proxy to "concentrate/redirect" the traffic from the
> port
> 80 to the appropriate application port. This way we have one single
> point of failure, the proxy, but orion instances are independent of
> each
> other and we can start/stop/move them without bothering the other
> applications. Besides, we also have an SSI script inside the proxy
> that,
> in case an application server is down, redirects the request to the
> appropriate "For maintenance reasons..." page. Right now the proxy and
>
> the orion instances are, some of them, in the same host, but in the
> future I guess they won't be as next step for us is clustering.
> Anyway,
> no Tomcat anywhere and no plans for it, at least as it is now.
> We are quite happy with this set up, but we are in a situation were
> our
> main worry is not speed, so we don't care about the small extra trip
> between the proxy and the orion instance, but the number of
> applications. With hot deployment you are supposed not to have to stop
>
> the orion server but sometimes...
> Just my 2ec,
> D.
> -------------------------------------------
> Daniel Lopez Janariz ([EMAIL PROTECTED])
> Web Services
> Computer Center
> Balearic Islands University
> -------------------------------------------
>
> > Patrik Andersson wrote:
> >
> > Just for the sake of asking,
> >
> > why do you have tomcat serving jsp/servlets and orion serving ejbs
> if
> > they're both running on the same machine? For me, that sounds like
> > asking for extra maintenance trouble. And another thing, from having
>
> > one "single point of failure" you now have 3. If either one if these
>
> > three applications decide to call it a day your whole application
> dies
> > and that goes for having two machines running different software
> > aswell. Why not use two or three machines all running orion and
> having
> > them split the workload by clustering them?
> >
> > regards,
> > Patrik
> >
<snipped for brevity>
