There should be a way to set a restart policy for a task in a similar to Docker way: `never`, `on-failure`, `always`.