In the CommonManagedDeployableContainer we eagerly destroy the process if a java.util.concurrent.CancellationException is thrown. This can be known to be thrown when a shutdown operation is invoked. Instead of destroying the process, we should log the error or possibly change it a warning, but not destroy the process. We can then wait for the process and destroy if the process does not exit after the timeout.

16:58:44,898 INFO  [org.jboss.arquillian.container.test.impl.client.container.ClientContainerController] (main) Manual stopping of a server instance
16:58:44,972 ERROR [org.jboss.as.arquillian.container.managed.ManagedDeployableContainer] (main) Failed to shutdown the server: java.io.IOException: java.util.concurrent.CancellationException: JBTHR00004: Operation was cancelled
16:58:44,975 WARN  [org.jboss.as.protocol.connection] (XNIO-1 task-3) WFLYPRT0018: No such request (54) associated with channel Channel ID bf2e513f (outbound) of Remoting connection 07af1118 to /[0:0:0:0:0:0:0:1]:10090 of endpoint "management-client" <67744663>

The error occurred on the org.jboss.as.test.clustering.single.provider.ServiceProviderRegistrationTestCase in WildFly. This is a clustering test which shutdowns a server and then checks a count which is invalid because the process was destroyed instead of gracefully waiting.