I'm attempting to fix http://jira.rhq-project.org/browse/RHQ-1930. I need my latest checkin (rev 3634) peer reviewed and tested. I ran it over here and it looks like its working (even stepped thru a debugger to see it working). I basically wrap calls to the discovery components in a proxy that times out the invocations if the component takes too long (30 seconds hardcoded - yes, it needs to be configurable, that would be nice :) Do you see anything missing? There is one thing I already know about: the thread pool can grow unbounded if a discovery component is misbehaved. In other words, if a discovery component's discoverResource method consistently deadlocks or enters an infinite loop, our thread pool will grow unbounded until we run out of mem (i.e. if we run discovery every 15 minutes, then over a span of one day, 96 deadlocked threads will have been created but not terminated ... 10 days == 960 threads, etc. etc. This is why I think I need to add the ability for the agent to disable a misbehaving discovery component. Something like, "if a discovery component times out N times, it should be disabled and never invoked again". Thoughts?