Imagine the scenario:

A build slave contains a Linux system with "/" or critical directories thereof being mounted as a unionfs between a RO persistent mount and a tmpfs transient in-memory FS. Each job populates its own build environment.

In such a case a reboot of the slave would produce a pristine environment within tens of seconds ready to service another job, providing building and testing environment that is fully build-time generated and perfectly reproducible every time.

In order to achieve this EC2 plugin needs to be taught to reboot a slave after a posted Task on it completes (as long as the task is not executed by an a one-off-executor).