I just identified ARQ-1964 while debugging a clustered issue, and I think this affects the reliability of many tests in Infinispan.

The essence is that while you might think you're testing things on containers {A,B}, and so doing able to verify the cross-node synchronisations, you're actually always testing couples {A,A} or {B,B}. It's possible that some things actually don't work as expected even if the tests are passing.