Connection pool for ClusteredConnectionFactory, how to ?

[roman.mandeleil1]

Hi Fellows,

I am trying to gather some information of connection pooling for JMS.

Let's say I have the following architecture:


FrontNode ---> Node1 <---> Node2


FrontNode gets HTTP trafic and forwards it to Node1, Node2 - cluster over JMS protocol.
The system works great but the question is regarding pooling connectionFactory.createConnection():

Basicaly that code runs on FrontNode:

Queue queue = (Queue) initialContext.lookup("/queue/MyQueue");
ConnectionFactory cf = (ConnectionFactory) initialContext
.lookup("/MyClusteredConnectionFactory");

connection = cf.createConnection();

Obviously, the create connection is very heavy opperation because it creates the actual socket for the
message. Two questions:

1) How to configure the pool for the create connection operation ?
2) The connection factory is been accessed as a stub because it been
recieved over JNDI so where the configuration for the pool should be applied on the Node1 or on FronNode ?

*) I am using JBoss 5.1 GA

[gaohoward]

Q1:
JBM doesn't provide connection pooling. You need to pool the connection yourself or using some 3rd party lib (if any).

Q2:
Because JBM doesn't pool connections, so there is no place for such configurations.

Howard

[roman.mandeleil1]

"gaohoward" wrote:

Q1:
JBM doesn't provide connection pooling. You need to pool the connection yourself or using some 3rd party lib (if any).

Q2:
Because JBM doesn't pool connections, so there is no place for such configurations.

Howard

Thanks, I think I can implement this with stateless bean which will keep the connection, and will reload it if it is down.

[timfox]

If you're using a stateless session bean that means you're running on a server so you can just use JCA pooling.

There are various wiki pages on this.

[roman.mandeleil1]

"timfox" wrote:
If you're using a stateless session bean that means you're running on a server so you can just use JCA pooling.

There are various wiki pages on this.

You mean: to pool JMS stub with JCA it interesting , have you tried it ?

[timfox]

This is actually mandated by the JEE spec.

Any JMS connections you create from a SLSB must be created (and pooled) by a JCA managed connection factory.

[bershath27]

"timfox" wrote:
If you're using a stateless session bean that means you're running on a server so you can just use JCA pooling.

There are various wiki pages on this.

You mean: to pool JMS stub with JCA it interesting , have you tried it ?

Tim is referring to the JMS session pool. In other words, obtaining connections from the factory bound at java:/JmsXA. Like Tim has clearly stated,there are n+1 examples + wiki's on the web..forum references etc etc

[timfox]

http://www.jboss.org/community/wiki/ShouldIcacheJMSconnectionsandJMSsessions

[atijms]

Tim Fox wrote:

http://www.jboss.org/community/wiki/ShouldIcacheJMSconnectionsandJMSsessions

Something that is not entirely clear to me from that wiki page is how to use JMS connection pooling in combination with a remote queue.

In some examples I see that the JMS connection factory is obtained from the JNDI of the remote server, after which the topic or queue is also retrieved from that remote JNDI, and then the message is sent. I.e. this example: http://blog.emmanuelbernard.com/2008/08/remotely-send-and-consume-messages-with.html

However, other code snippets I see obtain a connection from the local connection factory, and then use this local connection with the queue obtained from the remote JNDI. That method is used in this post: http://community.jboss.org/thread/49409.

Intuitively it seems awkward to use local connections to reach a remote queue. With this mechanism, does the same pool then contains both local and remote connections? Is it possible at all to create different pools and size these differently? By default there is of course only one java:JmsXA.

Help or pointers would be greatly appreciated.

[atijms]

Anyone?