geronimo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jeff Genender <jgenen...@apache.org>
Subject Re: Clustering and Tomcat 5.5.12
Date Mon, 14 Nov 2005 17:49:50 GMT


Dave Colasurdo wrote:
> I'm interested in testing and helping to document geronimo Tomcat 
> clustering.  As a first step, I've setup a Tomcat cluster independent of 
> Geronimo.  Once I get this fully working, I plan to reuse the 
> configuration info in the geronimo environment.

Thank you for taking the initiative on this.  You will be our main test 
on this since it has been limited for me due to my time constraints.

> 
> I've managed to setup the Apache http server and mod_jk as the load 
> balancer and do see the requests get sprayed to different tomcat servers 
> (on two diferent machines).  Sticky port is enabled and I see the node 
> id in the cookie and the rewritten url..
> 
> Initially, I tried using memory replication.  Cluster membership between 
> the nodes isn't working.  I believe this is due to the fact that both 
> machines are using DHCP and that their IP addresses aren't on the same 
> subnet.  I can always use two Tomcat installs on the same machine though 
> was hoping to setup a more realistic scenario using multiple machines. I 
> guess I need to setup a shared file system or use a database.
> Any advice?

Yep...(relative to geronimo) if you want to run it on the same 
machine...you have some hacking to do on the plans.  You should be able 
to alias your IP address to support multiple IPs.  Then in the Tomcat 
plan, you need the connectors of each instance to listen on one of the 
IP addresses (of course each instance should be listening on a different 
IP address).  For stand alone tomcat, this would be a similar 
process...the connectors need to listen on their own addresses.

IMHO, I would test this on multiple machines before a single machine 
with a single instance since you will probably run into other problems 
as well with other services in Geronimo.  After you get a multi-machine 
cluster working, then we  can configure it for a single machine.

Also, The engine GBean needs to use the jvmRoute initparam and you 
should make separate names for each instance (i.e. node1, node2, etc).

> 
> Assuming this works out.. A few question concerning running clustering 
> within Geronimo.
> 
> For tomcat, I update server.xml to configure AJP Connector, Cluster 
> config, Engine jvmRoute, etc.. How will we configure silmilar type
> information in Geronimo?

It follows the same paradigm as the other Tomcat GBeans.  You can set up 
a cluster at the Engine, Host, or Context level.  The Valves and Realms 
are a great example of this.  For the host and engine clustering you 
will do it in the j2ee-tomcat-plan.xml.  For context you will do it in a 
geronimo-web.xml or web descriptor plan.

I will write up an example and email it to the lists and then you can 
give it a try.

> 
> Do the distributable applications need to create deployment plans that 
> reference the new GBeans?

Depends whether you only want to cluster at a Context level or a host level.

> 
> Any additional steps?

Yep...stay tuned...I'll write something up.  Thank you very much for 
being the guinea p^H^H^H^H^H^H^H^H test person on this ;-)

> 
> Thanks
> -Dave-
> 
> 
> 
> Jeff Genender wrote:
>> I have checked in some initial Clustering GBeans for Tomcat.  As I 
>> developed this, it appeared that the Tomcat version 5.5.9 did not 
>> match the Tomcat doc.  This led me to examine the Tomcat objects and 
>> changelog.  It would appear that there were some improvements between 
>> TC v 5.5.9 and 5.5.12, especially with regard to Clustering.  so due 
>> to this, I have upgraded our version of Tomcat to 5.5.12.
>>
>> I have tested this with some of our applications (i.e. console) and 
>> the unit tests.  All looks good to me... so far ;-)
>>
>> If there are any issues, let me know.
>>
>> Jeff
>>
>>

Mime
View raw message