How to build a self-healing system using supervision tree in Elixir

The post explains that Elixir supervision trees help build reliable, fault-tolerant, self-healing systems. It shows how supervisors restart failed workers, when they escalate restarts to a parent, and how different restart strategies change the outcome.

Reading notes

#

• Supervision trees are presented as a way to keep services available by restarting the part of the system that failed.
• The tree has two process types, workers and supervisors, and supervisors manage direct children only.
• A supervisor can restart children when they terminate and can also terminate itself and its children if failures keep happening.
• The three restart strategies are :one_for_one, :one_for_all, and :rest_for_one.
• The default restart tolerance is 3 restarts in 5 seconds.
• A simple tree is built with one supervisor and two worker processes started through Supervisor.start_link/2.
• Each child defines child_spec/1, which returns the :id and :start fields needed by the supervisor.
• Task.start_link/1 is used in the example workers to create long-running processes.
• With :one_for_one, killing one worker restarts only that worker.
• With :one_for_all, killing one worker restarts all children.
• With :rest_for_one, killing a process restarts it and the children listed after it.
• When a child exceeds the restart limit, the supervisor exits and the parent supervisor restarts it.
• In the more complex example, process C is itself a supervisor with children D and E.
• Killing process C restarts C and then its children D and E.
• Repeated crashes of D or E eventually cause C to terminate and be restarted by the parent supervisor.