It looks like both tests assume that the cluster will failover leadership within a fixed sleep window (1 sec?). But this is not guaranteed. The tests should be changed to continue attempting to resend messages until they succeed, subject to some total wait timeout.