Repository navigation
[BUG] — HierarchicalWheelTimer: non-daemon worker thread never stopped on shutdown() + unbounded single-thread task executor #6668
Description
- severity: High (leak) / Medium (bottleneck)
- files:
shenyu-common/src/main/java/org/apache/shenyu/common/timer/HierarchicalWheelTimer.java:83-86(ShenyuThreadFactory.create(executorName, false)→ non-daemon),:84-85(taskExecutor = new ThreadPoolExecutor(1,1,0L,MS,new LinkedBlockingQueue<>())),:146-148(shutdownonly stopstaskExecutor, neverworkerThread),:164-170(Worker.runwhile(true)swallowsInterruptedException) - description:
shutdown()never stops/interrupts the non-daemonworkerThread, so itswhile(true)loop keeps running and ignores interrupts → thread leak + JVM hangs on redeploy. All expired timer tasks run on a single-thread executor with an unboundedLinkedBlockingQueue→ burst of simultaneous expirations or any slow task creates head-of-line blocking and unbounded queue growth. Used byFailbackRegistryRepository(client) andFallbackShenyuClientRegisterService(admin). - impact: Leaked non-daemon thread on redeploy; retry backpressure bottleneck.
- suggested_fix: Make
workerThreada daemon (or interrupt+join inshutdown); boundtaskExecutor's queue with a real rejection handler. - confidence: High
Identified during the 2026-08-02 deep re-scan; full list in docs/scan2-2026-08-02/00-consolidated-critical-high.md.
Activity
github-actions commentedon Aug 3, 2026
👋 Hello @yuluo-yx
A new issue has been opened by @Aias00:
HierarchicalWheelTimer: non-daemon worker thread never stopped onshutdown()+ unbounded single-thread task executorHierarchicalWheelTimer: non-daemon worker thread never stopped onshutdown()+ unbounded single-thread task executor #6668Please review when you have time. Thank you! 🙏
Aias00 commentedon Aug 3, 2026
Cross-link: related to the periodic-task / worker-thread lifecycle hardening pass #6608 (tracked there as a related, non-duplicate defect — this is an idempotent-start / worker-shutdown leak, not the exception-cancellation family). Folding into the same hardening pass if its scope is widened.
sunnysabor commentedon Oct 1, 2026
I checked the current
upstream/master(09c6a528) against this report. The worker-thread shutdown part is already addressed by merged #6982 (58dd5a517):shutdown()interrupts the worker under the write lock,Worker.run()exits when interrupted, andHierarchicalWheelTimerTest#testShutdownStopsWorkerAndRejectsNewTasksjoins the worker and asserts it has terminated.The unbounded
LinkedBlockingQueueremains. That is a separate overload/backpressure question: a bounded queue needs an explicit overflow policy because rejecting an expired timer callback can leave caller state stale, while caller-runs can stall the timing-wheel worker. Could we narrow this issue to the queue/backpressure behavior and agree on the expected overload semantics, while marking the worker-leak portion resolved by #6982?wy471x commentedon Oct 4, 2026
/assign
fix: make HierarchicalWheelTimer threads daemon and bound its task qu…