JMS detection of failed network connection

Question

I have a JMS topic that has a number of consumers, each of which is on a different machine. Messages are being sent on the topic at a low enough rate that the consumers are usually able to handle them immediately. If I look at the monitor screen (on admin console) for JMS, I see that I never have more than one or 2 messages on the server; they are consumed just about immediately, and the high water mark for the topic is 1 or 2.
Part of our system robustness testing is disconnecting the network from one of the machines hosting a consumer. When I do that and monitor the number of JMS messages on the server, I see the number of messages climb. My assumption here is that the server keeps track of its consumers, and it is holding on to messages until it can deliver them to the consumer. At some point, usually, the number of consumers connected to the topic is reduced by 1 and the number of messages on the server is reduced down to a low number. I am assuming that the server has detected that the consumer is unreachable and it no longer holds onto the messages for delivery to that consumer.
First, are my assumptions accurate?
Second, if so, how can I configure how quickly the server concludes that the client is unreachable and stops holding messages? What I want to do is limit the number of messages in a topic using a quota, and I need to know how to set a reasonable size, given a message production rate. For instance, is there a way to say how long the server should try to deliver messages to a consumer before giving up on it?

Answer 1

Hi,
I'm assuming the subscribers are non-durable, in which case I think your assumptions are correct - it takes some time to detect a network disconnect.
Its possible to tune how quickly WebLogic tears down stuck network connections, but this is generally not recommended (I think the configurables are on the "Network Channel").
Some options:
You can programmatically set a message expiration time using the standard JMS API (javax.jms.MessageProducer.setTimeToLive() are as a parameter to publish()), or, more conveniently, you can configure an expiration time override on the WebLogic topic itself.
You can configure a quota on the destination, but this would cause new publish attempts to fail with a resource allocation exception once the quota condition is reached (rather than cause the oldest messages to be dropped). That said, it is recommended best practice to always configure quotas in order to prevent the possibility of out-of-memory errors, and there are some flow-control options to help (see the JMS section of the Performance & Tuning guide in the edocs). You might also wish to tune JMS paging.
Regards,
Tom