Thanks all for the comments - I will rerun my stress test on Monday and report back. I do not think my url-based parameter were being used because of pilot error: I was using a jndi.properties file (a practice I have now abandoned) and the connection url was specified in that file, not in my external properties files.
The text fixture consists of 2 separate jmeter test cases (.jmx files), so each runs in its own jvm. Producers in one JVM, consumers in the other. Each JVM has max 512m heap and max 128m permgen. I have configured the jmeter Subscriber Sampler to initiate the jms connection once at startup, but I now realize I do not know if that is once per consumer thread or one shared connection -- I'll verify that and report back. I'll see if the flow control ameliorates this issue. As I said, I am stress testing and actively wanting to see how the system fails. Learning how to tune the system to make it more resilient & learning how to monitor/diagnose problem events are my goals. Yes, the consumers are pulling lots of messages but hopefully the JVM can gc them as they are not collected AFAIK. Have a great weekend and thanks guys! -- View this message in context: http://activemq.2283324.n4.nabble.com/Stress-testing-OOME-tp4677959p4678009.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
