See SOLR-4557 for the background, but here's the short form: TestCoreContainer.testReload just starts up 4 threads, each of which does a reload on collection1 then quits. Without going too far astray, there is one leaked thread for every thread testReload uses.....
Problem is that it's _not_ the thread that testReload actually fires off, I don't know where it's coming from. I've looked at the corecontainer code and nothing's jumping out at me. Anyone have any clues? I've attached a patch to SOLR-4557, without this patch the test doesn't fail regularly, although it did fail on one of the build machines which is why I started looking at it. I'm being compulsive here because I'm pretty sure this has been broken for a while, but we were being saved by throwing an exception.