[urbandw311er]

Can sometimes be scaling/monitoring loops. i.e. cluster comes up, provides some limited service, gets overloaded and drops below required performance metric, gets killed by monitoring/scaling system, repeat...