[Camel-users] New JMS connection being created for every message

The behaviour is caused by spring jms and is normal. The spring jms code
opens and closes the connection for every call. This is the expected
behaviour in JEE environments.
To improve the performance you need connection pooling.

Spring provides the
http://static.springsource.org/spring/docs/2.5.x/api/org/springframework/jms/connection/CachingConnectionFactory.html
for that.
You can wrap your original connection factory with it.

Alternatively you can check if MQ provides a pooling connection factory.

Christian


Am 13.04.2012 11:11, schrieb weberj:

Hi,

OK, as you suggested I configured the
org.springframework.jms.connection.CachingConnectionFactory. It deploys but
on initializing there is

com.ibm.msg.client.jms.DetailedIllegalStateException: MQJCA1031: The method
can only be called in the application client container. The application was
not running in the application client container when this method was called.
Ensure that the application runs in the application client container, or
modify the application to avoid this method call.

This seems to be similar to
https://community.jboss.org/thread/156545

An MQ Trace showed that the caller was
org.springframework.jms.connection.SingleConnectionFactory.prepareConnection(SingleConnectionFactory.java:364)

So I hacked
prepareConnection(Connection) und commented out
con.setExceptionListener(listenerToUse);

But now there is
MQJCA1018: Only one session per connection is allowed. The application
attempted to create more than one JMS session on the same JMS connection.
This exception occurs only if the application is running in a managed
environment. Modify the application so that it creates only one JMS session
on a JMS connection.

at
com.ibm.mq.connector.outbound.ConnectionWrapper.createSession(ConnectionWrapper.java:113)
at
org.springframework.jms.connection.SingleConnectionFactory.createSession(SingleConnectionFactory.java:406)

Has anybody an idea what that means?

I guess I should try the MQ connection pool instead of Spring's.

Thanks,
Juergen

Never saw that before how does your init code look like? Or do you
do it in spring?

Where do you run your application?

Christian

Am 25.04.2012 16:19, schrieb weberj:

Yes, the pool is defined in Spring XML. Camel should run in JBoss 4, MQ is in
the resource adapter (therefore the "application client container"
complaint).
Guess I'll have to remote debug the code..

The connection and route definitions cannot be totally wrong, as one or two
messages are delivered until the error comes up.

Juergen


<bean id="jmswsmq" class="org.apache.camel.component.jms.JmsComponent">
<property name="connectionFactory" ref="mqConnectionFactoryWrapper" />
<property name="acknowledgementModeName" value="AUTO_ACKNOWLEDGE" />
</bean>

<bean id="mqConnectionFactoryWrapper"
class="org.springframework.jms.connection.CachingConnectionFactory">
<property name="targetConnectionFactory" ref="mqJmsConnectionFactoryJNDI"
/>
</bean>

<jee:jndi-lookup id="mqJmsConnectionFactoryJNDI" jndi-name="java:wmqcf" />

In a container like jboss it should not be necessary to use the
CachingConnectionFactory. Instead you should configure the resource
adapter to support pooling.

Then you can simply lookup the ConnectionFactory on jndi and use it as
it is.

The CachingConnectionFactory is typically used when you are not running
in a container.

Christian

Am 26.04.2012 10:47, schrieb weberj:

This was my original configuration that pooling was in the resource adapter
(the MQ RA has a default of 10 maxConnections).
But when Camel closes the connection this somehow seems to close the RA's
underlying connection, too.

This is then probably an issue with the resource adapter. As far a I
understood a Java EE application is expected to close the connection
after each method call and the
pool should then handle that this still is performant.

Christian

Am 26.04.2012 11:40, schrieb weberj:

No, the resource adapter is fine. I wrote a simple MDB that uses the RA and
just echoes back a message to the reply Q. In the onMessage() call I create
and close the connection. This MDB can handle 20 messages/s, whereas Camel
handles only one.

Just to recap what we found today. The reason for the bad performance
seems to be that camel polls for replies in the producer with a default
of 1000ms. Setting receiveTimeout=10 on the producer seems to speed up
the route a lot.

After we found that your colleague also found that this is actually
documented in the camel-jms documentation :-)
http://camel.apache.org/jms.html
I added the relevant section below.

I am not really sure why the receiveTimeout really controls the polling
but it works. In any case the option name looks a bit misleading. I
guess pollingIntervall would
be more acurate.

Christian

-----


Request-reply over JMS and using a shared fixed reply queue

If you use a fixed reply queue when doing Request Reply
<http://camel.apache.org/request-reply.html> over JMS as shown in the
example below, then pay attention.

from(xxx)
.inOut().to("activemq:queue:foo?replyTo=bar")
.to(yyy)

In this example the fixed reply queue named "bar" is used. By default
Camel assumes the queue is shared when using fixed reply queues, and
therefore it uses a JMSSelector to only pickup the expected reply
messages (eg based on the JMSCorrelationID). See next section for
exclusive fixed reply queues. That means its not as fast as temporary
queues. You can speedup how often Camel will pull for reply messages
using the receiveTimeout option. By default its 1000 millis. So to make
it faster you can set it to 250 millis to pull 4 times per second as shown:

from(xxx)
.inOut().to("activemq:queue:foo?replyTo=bar&receiveTimeout=250")
.to(yyy)

Notice this will cause the Camel to send pull requests to the message
broker more frequent, and thus require more network traffic.
It is generally recommended to use temporary queues if possible.



Am 30.04.2012 15:52, schrieb weberj:

Its the name of the option the spring guys gave it.

Use a temporary or exclusive queue for the reply message is faster, as
JMS message selectors i not needed, and then the receiveTimeout doen't
matter.

I would recommend using WMQ's internal pooling.
Just set connectionPooling to true on the connectionfactory.

2012/4/25 weberj <juergen@jwi.de>: