ADDITIONAL SYSTEM INFORMATION :
openjdk version "25" 2025-09-16
OpenJDK Runtime Environment (build 25+36-3489)
OpenJDK 64-Bit Server VM (build 25+36-3489, mixed mode, sharing)

Tested also on JVM 21, 25-temurin, 25-zulu

OS: Darwin Kernel Version 25.0.0: Wed Sep 17 21:41:45 PDT 2025; root:xnu-12377.1.9~141/RELEASE_ARM64_T6000 arm64

A DESCRIPTION OF THE PROBLEM :
The issue occurs rarely when running a test suite for a library making heave use of Virtual Threads. The specific tests repeatedly creates a large number of threads, interrupts & joins them.

I shared a the test (*), on a branch which additionally logs when the problematic thread is started, finished and joined. The `StressTest.testMultipleOperationsDirect` test is run in a loop, sometimes it hangs after a couple of minutes, and sometimes it just goes on without problems. If it hangs, it always hangs in the same place. There are no exceptions logged or caught. I tried simplifying the code and creating a smaller reproducible test-case, but unfortunately removing any of the code (assertions, branches etc.) causes the test to stop hanging.

When the test hangs, `jcmd [pid] Thread.dump_to_file [file]` produces a stack trace with the following virtual threads running:

#49394 "" virtual WAITING 2025-10-02T08:14:38.821216Z
    at java.base/java.lang.VirtualThread.park(VirtualThread.java:738)
    - parking to wait for <java.util.concurrent.CompletableFuture$Signaller@38994429>
    at java.base/java.lang.System$1.parkVirtualThread(System.java:2284)
    at java.base/java.util.concurrent.locks.LockSupport.park(LockSupport.java:221)
    at java.base/java.util.concurrent.CompletableFuture$Signaller.block(CompletableFuture.java:1885)
    at java.base/java.util.concurrent.ForkJoinPool.unmanagedBlock(ForkJoinPool.java:4364)
    at java.base/java.util.concurrent.ForkJoinPool.managedBlock(ForkJoinPool.java:4310)
    at java.base/java.util.concurrent.CompletableFuture.waitingGet(CompletableFuture.java:1919)
    at java.base/java.util.concurrent.CompletableFuture.get(CompletableFuture.java:2093)
    at com.softwaremill.jox/com.softwaremill.jox.StressTest.lambda$testAndVerify$0(StressTest.java:97)
    at com.softwaremill.jox/com.softwaremill.jox.TestUtil.lambda$scoped$0(TestUtil.java:17)
    at java.base/java.lang.VirtualThread.run(VirtualThread.java:456)

#49401 "" virtual WAITING 2025-10-02T08:14:38.821305Z
    at java.base/java.lang.VirtualThread.park(VirtualThread.java:738)
    - parking to wait for <java.util.concurrent.CountDownLatch$Sync@6cfa0497>
    at java.base/java.lang.System$1.parkVirtualThread(System.java:2284)
    at java.base/java.util.concurrent.locks.LockSupport.park(LockSupport.java:221)
    at java.base/java.util.concurrent.locks.AbstractQueuedSynchronizer.acquire(AbstractQueuedSynchronizer.java:790)
    at java.base/java.util.concurrent.locks.AbstractQueuedSynchronizer.acquireSharedInterruptibly(AbstractQueuedSynchronizer.java:1139)
    at java.base/java.util.concurrent.CountDownLatch.await(CountDownLatch.java:230)
    at java.base/java.lang.VirtualThread.joinNanos(VirtualThread.java:1002)
    at java.base/java.lang.Thread.join(Thread.java:1870)
    at java.base/java.lang.Thread.join(Thread.java:1963)
    at com.softwaremill.jox/com.softwaremill.jox.TestUtil$1.cancel(TestUtil.java:101)
    at com.softwaremill.jox/com.softwaremill.jox.StressTest.stressTestIteration(StressTest.java:234)
    at com.softwaremill.jox/com.softwaremill.jox.StressTest.lambda$testAndVerify$1(StressTest.java:79)
    at com.softwaremill.jox/com.softwaremill.jox.TestUtil.lambda$fork$0(TestUtil.java:36)
    at java.base/java.lang.VirtualThread.run(VirtualThread.java:456)

The first is waiting for the test completion, the second is interesting: here it's thread number 49401 waiting on a `Thread.join()` called from `TestUtil$1.cancel(TestUtil.java:101)`.

If you take a look at the source (+), cancelling amounts to an interrupt and join. There's also a `Thread.yield()` before calling cancel.

From the thread dump, we know that it's thread 49401 that is hanging. Looking at the system output (the debug messages), that thread has been joining on a number of threads, each has a "joining"-"joined" pair of messages (as printed out in the source), except the last one:

Starting 50097 <- a worker thread
(...)
Joining on thread: 50097 from 49401 <- hanging thread
(...)
Exit 50097 <- printed in finally of the worker thread

However, despite thread 50097 exiting, the `join()` never completes. In various runs, sometimes the thread seems to finish before the join is even invoked (but this might timing of the `System.out.println`). Sometimes it's far apart (with many other messages in-between), sometimes the three messages are clustered together. The thread dump does not contain thread 50097.