Executed a multi-test with 9 scenarios on a single OPL configured for a maximum of 10 engines. The Engine Health report shows that the Memory reached 100% on all engines. The Request Stats report does not contain labels from all of the scenarios. The test did not stop by itself and even after manually terminating the test, it took almost 25 minutes for the test to stop and the heartbeat to resume for the OPL.
The JVM XMS and XMX settings for the OPL were both set to 7800 MB, but the Linux Ubuntu system only has 32G of available memory. The 9 scenario Multi-test kicked off 9 engines each requiring 7800 MB of memory for each engine (for a total need of 70,200 MB of memory for the test), well over the memory available on the Linux system.
Component : BLAZEMETER PERFORMANCE TESTING
Three options to address this issue:
1. Add enough memory to the existing Linux Ubuntu system to be able to handle 10 engines running on the OPL with some memory left over for the operating system (78,000 MB plus some extra memory)
2. Create new OPLs with enough memory resources that will allow the option to spread the scenarios of a multi-test across multiple OPLs.
3. Reconfigure the JVM XMS and XMX values to much smaller values on the OPL (remembering not to configure the JVM XMX to a value that would end up using all 32G of memory if all 10 engines are used in a test - see Using Private Locations article). For this option, it would be necessary to verify that each individual test configured in the scenarios of the multi-test will run successfully with the new JVM memory limits (see the Calibrating a JMeter Test article).