Currently, we spin clusters up in mixed modes which are know to fail, and assert that they are unable to work for some period of time. We currently wait far too long, and produce gigabytes of logs.