Recently I had to refactor a feature that I had introduced to the old Servicemix Kernel, now Felix Karaf project. The issue was relatively minor to resolve, however testing it would require profiling memory and CPU usage over a period of time (to avoid an eventual out of memory error).
There are multiple tools out there that would allow me to do this sort of monitoring, for this particular task I chose to try out VisualVM as it was already bundled with my JDK 1.6 install. VisualVM is a visual tool that integrates several commandline JDK tools and lightweight profiling capabilities for the Java SE platform.
To begin my refactoring task I first setup Karaf with the configuration required to reproduce the out of memory condition. No special configuration or instrumentation of Karaf itself was required for using the profiler. Starting Karaf, using JDK 1.6, I was able to attach to the running process via the VisualVM interface. Once connected to the process I could view the live heap usage, threading, permgen, and loaded classes, further I could inspect the parameters fed to the JVM at start up to confirm the environment was as I configured. To view memory and CPU usage of Karaf, I selected the profiler display tab. Adjusting the profiler settings to only monitor org.apache.felix classes, I could quickly identify the offending set of method calls that contained the source of my error (high CPU usage, and multiple instances accumulating in memory).
Resolving the error didn't require much work except to ensure a few resources were being properly freed, and removal of an unneeded for-loop. To test out my modifications I redeployed Karaf, and began reviewing its run time characteristics via VisualVM. This time the trouble methods were not being called as often, and not creating memory leaks, while still providing the same functionality the original code was intended to provide.
As a simple profiler I can recommend VisualVM as a quick, convenient way to observe your Java applications. It will be interesting to see what features they integrate into the tool in future releases.