On Mar 2, 2008, at 8:24 PM, Chris Chen wrote:

> Scott,
> With all the new changes to the watchdog process to make it more
> secure (which is a good thing), I have come across behavior that is
> limited:
> I have two config files, one for the web tier and one for the backend
> app tiers.  The configuration is fairly similar to how the
> documentation states.
> 1) the resin-web.conf (for the web tier) has its own set of
> configuration settings.  However, in order for the Load Balancing
> Servlet to see the app tier clusters and servers, I need to include
> minimal <cluster> config inside resin-web.conf:
> <cluster id='abc'>
>   <server id='abc' port='6800' .../>
> </cluster>
> This is the minimal that will allow the load balancing servlet to
> understand where to dispatch the requests in the rewrite section.
> 2) I then have resin.conf file that contains all the app tier config
> settings for all clusters.
> This configuration worked well even in 3.1.5 snapshots, but when I
> upgraded to 3.1.5, I started seeing "duplicate server id" exceptions
> thrown by the watchdog process.
> From what I can gather, I believe that the watchdog process is tying
> configuration filenames to the settings now.  So in this case, if I
> start up the web tier cluster on the same machine as the app tier and
> then start up one of the app tier cluster servers, watchdog will think
> that it's trying to register a duplicate server id.
> By putting all the resin-web.conf configuration info into resin.conf,
> watchdog process doesn't complain any longer.

I've filed this as http://bugs.caucho.com/view.php?id=2484

The check was written assuming it would be a mistake for a second  
resin.conf to have the same server-id as the original one.  But your  
case shows that it's a normal configuration.
> Given the new watchdog process monitoring, I'm thinking that the Load
> Balancer Servlet can be enhanced to work with the new watchdog
> process.  Here's what I'm proposing:
> 1) Watchdog process contains all the cluster names and server ids.
> 2) web tier's load balancer servlet only needs to know the cluster
> name and doesn't require any additional <cluster> configurations in
> the CONF file.  It will then query the cluster information from the
> watchdog process in some way to determine where to redirect the
> request to.

This is related to http://bugs.caucho.com/view.php?id=2314 (although I  
thought there was a dynamic cluster bug).

The issues are a little bigger, including distributed sessions, jms,  
and the load balancer, so it's a feature we're putting off until 3.2.0  
(which is right after 3.1.6)

-- Scott
> Or is there perhaps another way to configure the watchdog process so
> that it won't think that the web tier config is trying to register new
> clusters?  Or possibly I need to create three config files, one for
> web tier, one for app tier, and one shared conf file that contains all
> the cluster configurations that will be resin:import'ed into the web
> and app tier?
> -Chris
> _______________________________________________
> resin-interest mailing list
> resin-interest@caucho.com
> http://maillist.caucho.com/mailman/listinfo/resin-interest

resin-interest mailing list

Reply via email to