Io subsystem worker pool threads cannot die even if task-keepalive is specified ------------------------------------------------------------------------------- Key: WFCORE-2149 URL: https://issues.jboss.org/browse/WFCORE-2149 Project: WildFly Core Issue Type: Bug Components: IO Reporter: Mathieu Lachance Assignee: Jason Greene The Wildlfy io subsystem should allow to set thread keep alive time as documented in wildfly_io_1.1.xsd: {code} <xs:attribute name="task-keepalive" type="xs:int" default="60"/> {code} Though when you have a look at the XNIO implementation i.e. https://github.com/xnio/xnio/blob/f1bd055381436683789c7e69a3f6455b7141ad2...: {code} taskPool = new TaskPool( threadCount, // ignore core threads setting, always fill to max threadCount, optionMap.get(Options.WORKER_TASK_KEEPALIVE, 60000), TimeUnit.MILLISECONDS, taskQueue, new WorkerThreadFactory(threadGroup, optionMap, markThreadAsDaemon), new ThreadPoolExecutor.AbortPolicy()); final class TaskPool extends ThreadPoolExecutor { TaskPool(final int corePoolSize, final int maximumPoolSize, final long keepAliveTime, final TimeUnit unit, final BlockingQueue<Runnable> workQueue, final ThreadFactory threadFactory, final RejectedExecutionHandler handler) { super(corePoolSize, maximumPoolSize, keepAliveTime, unit, workQueue, threadFactory, handler); } protected void terminated() { taskPoolTerminated(); } } {code} The created TaskPool/ThreadPoolExecutor will have both the core and max thread size equals. By setting them both equals, means that if ThreadPoolExecutor#allowCoreThreadTimeOut is not true, no thread will ever be able to die even if there's a keepAliveTime set. I think previously there's was a setting to set the actual amount of "core" thread pool but I think it was removed in Wildfly 9. But from an outsider, it is not really clear reading the code that this is really the case. That said, I think having either that setting back or setting the "allowCoreThreadTimeOut" set to true will fix the issue. Not letting thread die: # is a "waste" of resource # can lead to some ThreadLocal grows out of proportions (even if bad practice) Also, the documentation should be more precise to specify the time unit that will be used when configuring the keep alive time; from what I understand, it should state milliseconds.