A distributed cron triggers the same job on two nodes due to a network partition. What design makes this harmless?