Messages are not consumed by MDB if long running processes active

[abl-dam]

in JBossEAP 6.0.1 we are sending JMS messages to start long running background processes. Message are consumed by a MDB with a max of 10 sessions:

@MessageDriven(activationConfig =
{   @ActivationConfigProperty(propertyName="destinationType", propertyValue="javax.jms.Queue"),
    @ActivationConfigProperty(propertyName="destination", propertyValue="queue/workflow"),
    @ActivationConfigProperty(propertyName="maxSession", propertyValue="10")
})
@TransactionAttribute(TransactionAttributeType.NOT_SUPPORTED)
public class WfExecutionMDB extends BaseBeanSL implements MessageListener

this background processes can last from minutes up to 2-3 hours.

most of the time this works without problems.

but from time to time we see the effect that running processes are blocking new message sent to the queue, although there are MDB sessions left. the messages are not consumed from the queue by the MDB.

further message are still consumed, but the blocked messages are consumed only until the long running jobs are finished! as an interesting fact, these messages are processed by the same thread as the long running process.

a short example (by the order sent to the queue):

1. long running job 1 (Thread-4)

2. long running job 2 (Thread-5)

3. blocked message for job 3

4. blocked message for job 4

5. job 5 (Thread-6) -> consumed and executed

--- now job 1 finishes ---

-> job 3 message consumed and executed (on Thread-4)

it "seems" there is some kind of timeout that leads to reusing the HornetQ Thread from Pool, but it's still in use and therefor the job cannot start.

I consider it is an configuration or usage issue, but I found nothing in the docs or on the web. thanks for any advice.

[ataylor]

whats your consumerwindowsize? clients buffer messages so the long running client may already have these messages in its buffer.

[abl-dam]

it's not set in standalone.xml at all, so I guess some kind of default value (1 MB?)

we are not using the HornetQ native API

[ataylor]

you need to set it on the ResourceAdapter(pooled connection factory) or as an activation property on the MDB.

[abl-dam]

this way?

                    <pooled-connection-factory name="hornetq-ra">
                        <transaction mode="xa"/>
                        <connectors>
                            <connector-ref connector-name="in-vm"/>
                        </connectors>
                        <entries>
                            <entry name="java:/JmsXA"/>
                        </entries>
                        <!-- Set the consumer window size to 0 to have *no* buffer on the client side -->
                        <consumer-window-size>0</consumer-window-size>
                    </pooled-connection-factory>

but is it the way it's intended to work? I understand that this client-side buffer is for better performance (avoid roundtrip to server for every message)

why doesn't the MDB simply use another (free) thread for consuming the message? in our case the delivery order (FIFO) is no longer guaranteed

[ataylor]

yes thats correct.

its nothing to do with the MDB using another thread, each mdb instance uses its own session/consumer.

[abl-dam]

ok, I will test with that setting and see what happens. Many thanks for your help!

@ActivationConfigProperty(propertyName="consumer-window-size", propertyValue="0")

or

@ActivationConfigProperty(propertyName="consumerWindowSize", propertyValue="0")

?

[ataylor]

consumerWindowSize