OSP18 has gone forward with Galera using a clean multi-master approach, which in CI testing has led to the observation of race conditions in individual services such as keystone, nova, cinder, and others which require some combination of MySQL session tuning and/or code changes to accommodate.    to this end, services have begun hardcoding wsrep_sync_wait in their configurations.   However, it's not clear that this will solve all issues, for example nova team has concerns over data issues that may require a compare-and-swap approach to accommodate correctly, which may not yet be present in nova source code.  to this end we are considering a novel approach to restore the "single master" behavior of OSP17 under OSP18 running on openshift.    I will sketch out sub tasks that represent thinking so far.