It happens that when using Arquillian WFLY tests that app server could be spontaneously killed by Arquillian itself. It happens in cases when the test uses ManagedContainer.kill commands. Which is usual practice in Narayana byteman crashrecovery tests. This relates to XTS, LRA, any byteman crashrec on WFLY.

The most prominent place where this can be observed is client debugging the tests (WFLY Arq + crashrecovery). E.g. running the LRA coordinator test like: mvn clean install -Parq -Dit.test=LRACoordinatorRecovery1TestCase -Dmaven.failsafe.debug and then waiting with debugger on TestCase#recover() call makes the WildFly being stopped (killed) in a minute.

This is done by Arquillian because of wrongly used processed in async shutdown hook - ie. https://issues.redhat.com/browse/WFARQ-78
(some details about this at: https://github.com/wildfly/wildfly-arquillian/pull/164)