Well I'm at a loss as to what's happening. If I run "test" on my Mac or on a Linux box, all of the UnitTests passes consistently:
*<klund@Kirks-MacBook-Pro>/Users/klund/dev/geode [1405]$ ./gradlew test* *...* *BUILD SUCCESSFUL* *Total time: 1 mins 30.552 secs* If I log into a worker on AWS and run "test", LocatorLauncherTest fails because InternalDistributedSystem.getConnectedInstance() is returning something other than null. All of the tests run by "test" target are true UnitTests which do NOT create a DistributedSystem or Cache or use the file system. I'm going to commit a pre-assertion to LocatorLauncherTest that tests InternalDistributedSystem.getConnectedInstance() and asserts that it must be null. If it's non-null then things are going to fail. If any UnitTest sets a non-null value in DistributedSystem#existingSystems then it's the responsibility of that test to set it back to null during tearDown. This problem is precisely why we group (non-Distributed) tests into either UnitTest or IntegrationTest. Only the IntegrationTest should ever create a DistributedSystem and those JVMs are forked for every test class. Since it doesn't reproduce on any local machines, well that's just weird and I don't know what to do for precheckin runs on AWS. On Fri, May 19, 2017 at 3:13 PM, Kirk Lund <kl...@apache.org> wrote: > The LocatorLauncherTest (a UnitTest) is now failing but only in > precheckin. When I run it locally it passes consistently. I don't have a > fix yet but I'm looking into it. It might continue failing all weekend. > Sorry. > >
