[ https://issues.jboss.org/browse/ISPN-5937?page=com.atlassian.jira.plugin.... ] Dan Berindei commented on ISPN-5937: ------------------------------------ I have managed to reproduce the test reliably by forcing it to run on a single core ({{taskset -c 1 mvn test -pl client/hotrod-client -Dtest=ExpiryTest}}). There seem to be 2 problems: 1. {{RemoveExpiredCommand}} only knows the key, the value, and the lifespan. Since all test methods use the same keys and values, and the default lifespan, then a delayed {{RemoveExpiredCommand}} deleting a newer value is a limitation that was known when Will started implementing cluster-wide expiration. On the other hand, HotRod uses versions, so {{RemoveExpiredCommand}} should also check the versions to make sure it's removing the correct value. 2. {{ClusterExpirationManager}} doesn't always create the {{RemoveExpiredCommand}} with the proper value, because {{ClusterExpirationManager.handleLifespanExpired}} doesn't lock the entry and the value is actually read in the async thread. This allows another thread to modify the value in the data container *and* in the entry that was passed to the async thread, before the {{RemoveExpiredCommand}} was created. -- This message was sent by Atlassian JIRA (v6.4.11#64026)