Test org.infinispan.api.mvcc.repeatable_read.WriteSkewTest.testWriteSkewWithOnlyPut fails intermittently with ER3 and Infinispan 5.2.5.Final as well.
Pedro Ruivo <pedroruivo2> updated the status of jira ISPN-2771 to Coding In Progress
Pedro Ruivo <pedroruivo2> updated the status of jira ISPN-2771 to Reopened
Pedro Ruivo <pedroruivo2> made a comment on jira ISPN-2771 This test is still failing because it performs the local write skew check during the copyForUpdate(). Solutions: easy: don't perform the local write skew check in the copyForUpdate because it will be performed in the prepare phase anyway. complex: the correct behavior here is to perform the local write skew check when * the key has previous read || the command does not ignore the return value try to do a test case for this race condition...
Dan Berindei <dberinde> made a comment on jira ISPN-2771 [~pruivo] why is the second approach more correct than the first? Shouldn't one write skew check be enough?
Pedro Ruivo <pedroruivo2> made a comment on jira ISPN-2771 [~dan.berindei] don't miss understand me. I think the most correct is only to perform the write skew check at prepare time (as it is done in clustered mode).
Already fixed upstream.