Zookeeper

• supervisory process
  • Zookeeper is designed to “fail fast,” meaning it will shut down if an error occurs that it can’t recover from.
  • Because of this, we must have a supervisory process manage the Zookeeper instances so that if a Zookeeper instance does go down, the cluster as a whole can continue handling requests
  • The supervisory process will handle restarting any failed individual Zookeeper server, allowing the Zookeeper cluster to be self-healing
• log management
  • Because Zookeeper is a long-running process, its transaction logs can get quite large
  • This will eventually result in Zookeeper running out of disk space
  • Therefore, it’s critical to set up some sort of process to compact (and even archive) the data produced in these logs