[ https://issues.jboss.org/browse/ISPN-2300?page=com.atlassian.jira.plugin.... ] Galder Zamarreño commented on ISPN-2300: ---------------------------------------- Fix for 1 is not as simple as it looks: - In the current code base, when a versioned entry fails to update due to conditional operation not fulfilling the condition, the previous value entry is added to the context with isChanged=true (if present in the cache, as a result of marking RepeatableReadEntry.copyForUpdate), and then EntryWrappingInterceptor loops through all the entries in the context, and if "changed", it will try commit them. What happens here is that the entry is not the result of an actual update in the transaction, so its version is null and it breaks later on when ClusteredRepeatableReadEntry checks for the version. - Pedro's fix works for this particular case, but breaks other use cases, such as CacheLoaderFunctionalTest.testLoading and PassivationFunctionalTest.testRemoveAndReplace. What happens there is that if an entry is in the cache but not in the cache, and the conditional operation to try to putIfAbsent a new value fails, it should load the previous value into the cache. With Pedro's fix, the cache is no longer updated since it's marked that entry to not be "changed", and it's "changed" since you're storing it in the cache, even if it's the result of activating/loading the entry. A better fix IMO would be to avoid committing the entry if the version is null in the 1st place. This is a more controlled or isolated fix, and avoids messing up with the business of unsetting the "changed" flag which is a bit fragile. Having just tested this quickly, it breaks other things... I'll keep trying things... -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira