The fix for JBTM-3533 uses the TransactionReaper to manage LRA timeouts.

A new test was added for the work (LRATest.java#testTimeout) where the participant sleeps for longer than the LRA timeout (to verify that the reaper cancels the LRA).

However, this test was seen to fail on a CI run (on a pull request) because the reaper reported the the participant (a ReaperElement) as "wedged" (see attached test output file). The reason is as follows:

The reaper sees it as wedged because the participant is running in the same VM as the test and the participant sleeps (Thread.sleep(2000)) to allow time for the timeout to expire. So this test failure is actually a false positive because the reaper reinserts the element into the timeout queue marking it as ABORT_ONLY and since LongRunningAction extends BasicAction it handles the ActionStatus.ABORT_ONLY state correctly.

The tests needs updating to accommodate this race condition.