If Flight Recorder is started dynamically,we use bytecode instrumentation to add the implementation to classes that derive from jdk.jfr.Event.

This works, but we have no testing that verifies it works if there is another agent running. For example, a tools vendor could use bytecode instrumentation to modify all constructors to track allocation in an application. It would be a severe problem if JFR is not able to run together with such tools. Today we have a test that does this for the instrumentation added for socket and file classes (TestInstrumentation). We should extend that test or create a new one that just verifies the same for subclasses to jdk.jfr.Event.