HotRod Transaction Cache - Wrong values when a transactional cache stops in the middle of PrepareTransactionOperation/CompleteTransactionOperation -------------------------------------------------------------------------------------------------------------------------------------------------- Key: ISPN-9425 URL: https://issues.jboss.org/browse/ISPN-9425 Project: Infinispan Issue Type: Bug Reporter: Diego Lovison Assignee: Pedro Ruivo Priority: Critical Fix For: 10.0.0.Final There are 3 scenarios and I believe that they are correlated. If it is not the case, a JIRA for each item should be open. All scenarios below are multiple clients and one server. The following steps should be executed does not matter what is the scenario. * Add one breakpoint at the first line of the method `acceptResponse` at the class PrepareTransactionOperation * Add one breakpoint at the first line of the method `executeOperation` at the class PrepareTransactionOperation *Scenario 1* * Execute the client code ( Program 1 ) * Execute the client code ( Program 2 ) * Program 1 -> Resume, Resume, Resume. It will stop at `acceptResponse`. * Program 2 -> Resume, Resume. It will stop at `executeOperation`. * Wait 5 minutes or any time greater than 60 seconds ( I was not able to identify the right period of time but greater than 60 seconds should work), Resume and Resume the "Program 2". It will fail. * The issue in scenario 1 is: Why doesn't matter the amount of time since it is greater than 60 seconds the request will fail? I am expecting something like a timeout in the "Program 1" and not expire when a call in that key was called for the first time. * Kill all clients and servers *Scenario 2* * Execute the client code ( Program 1 ) * Execute the client code ( Program 2 ) * Program 1 -> Resume, Resume, Resume. It will stop at `acceptResponse`. * Program 2 -> Resume, Resume, Resume, Resume. It will fail. Here you don't need to wait. Just press the Resume. * Program 1 -> Resume. * The issue in scenario 2 is: Why the test fails due: expected value 'null' different from 'random-value-generated' * Kill all clients and servers *Scenario 3* * Execute the client code ( Program 1 ) * Execute the client code ( Program 2 ) * Program 1 -> Resume, Resume, Resume. It will stop at `acceptResponse`. * Program 2 -> Resume, Resume, Resume, Resume. It will fail. Here you don't need to wait. Just press the Resume. * Wait 60 seconds * Re-run Program 2 -> Resume, Resume, Resume, Resume. It will work. * Program 1 -> Resume, it will fail due to an assertion error * The issue in scenario 3 is: When resuming the "Program 1" I am expecting an error during the commit phase and not an assertion error. * Kill all clients and servers Program 1 means that you should run the client code. It will create a separated JVM. Program 2 means that you should run the client code. It will create a separated JVM. It will be nice if we find out an way to automate this kind of tests.