Sorry about the examples, the general idea is that you can execute a command on the server when there are alerts for a specific problem. Upstart and daemontools can't keep your processes running. For this specific task there are tools like Monit, God, Puppet to some extend
Perhaps a better example would be automating a previously manual failover, i.e. if a service becomes unavailable on one machine, start it on the warm standby. Or in EC2, terminate an instance and start a new one.
