Question

Our Java application starts a worker thread (using Thread.start()). Shortly thereafter it calls Thread.join() on the worker thread. The worker thread does some stuff and terminates. The first thread exits the call to join() and goes on its merry way. Standard stuff:

Thread t = new WorkerThread();
t.start();

// Blah blah

t.join();

class WorkerThread extends Thread {
    public void run() {
        // Do some stuff
    }
}

At least that’s how it’s supposed to work, and how it does work in any case we can reproduce. We have one customer, however, who is persistently running into trouble.

Looking at the threads using PsiProbe, they see that the worker thread is created. It runs for awhile, but after some time disappears from the list of threads. This happens at an unexpected time (based on timing of other events related to the worker thread). The main thread never gets out of the join() call.

This would seem to break join()‘s contract, and implies to me some sort of JVM-level error. Has anybody witnessed behavior like this, or have any idea what could cause it?

EDIT 3-3-11:

I’m still waiting for conclusive data from the customer, but it seems likely that I didn’t really know what I thought I knew: the main thread is likely not blocking in join() at all, but at a point prior to it.

Thanks to all for the ideas, anyway.

Answer 1

A quick search of the Java Bugs Database didn’t throw up anything that matches your symptoms. But it would be worth your while doing a more extensive search.

---

However, it is worth noting that a JVM bug is only one of many possible theories (see the comments), and there is very little evidence for any theory at this stage. If I were you, I would:

1. figured out what the customer’s platform is
2. do a search of the Java bugs database for any known bugs that seem to match the observed symptoms (such as they are).
3. if you find a bug that looks like a “hit”, evaluate it further, try to confirm that it is really the customer’s problem, and see if there are any ways to mitigate it.
4. in addition …
   • look at the other possible theories; e.g. see above
   • try and figure out what is different about the customer’s installation compared with others who don’t have this problem
   • add more monitoring, and try to get the customer to use that version.

(On the last bullet point, it may be necessary to point out the obvious to the customer. If they want the bug fixed, they may need to do extra stuff to help you track down the bug, like installing running an “experimental” version of the application for a bit.)