Major Description
In our testing we have discovered that messages are not redelivered as expected when we are using a transactional session. The test goes as follows:-
1. Create a connection and a transactional session.
2. Create a queue consumer, add a listener and start the connection.
3. Add a message to the queue, the message will arrive on the consumer.
4. Kill the session (either dispose of the session and connection, or just kill the process)
5. Perform steps 1 & 2 again.
Note that the message is not redelivered.
We expect the message to be redelivered because the transaction was never committed. Looking at the Queues page on the web admin, you can see the queue count is still 1. So the message is still on the broker.
To diagnose whether this is an NMS issue or an ActiveMQ issue, I created a Java application to test this issue using JMS.
The resulting behaviour is as expected when using JMS. After the message is received, I can either cleanly dispose the session, or kill the application, and the message will be redelivered next time you start the app.
I was a little confused as to why the message did not get redelivered, in the example where I simply kill the process. I would expect the broker to handle NMS and JMS clients the same, as the TCP connection would simply be terminated.
I then noticed a difference between NMS and JMS, when looking in the web admin. When a message arrives on a JMS client the queue count shows 1, and clicking on the queue shows the message still on the queue. When a message arrives on an NMS client, the queue count goes to 1, but clicking on the queue shows no messages.
Clearly, there is a difference between how JMS and NMS handle an incoming message on a transactional session. Could it be that NMS is sending an ack for each message when it shouldn't?