Arbres de supervision complexes

In Erlang, supervisors are special processes that oversee other processes, called children. If a child process crashes, the supervisor can restart it, ensuring fault tolerance.

A supervision tree is formed when a supervisor itself becomes a child of another supervisor. This creates a hierarchy, much like an organizational chart.

As applications grow, a single supervisor isn't enough. Complex supervision trees allow us to:

• Manage dependencies: Group related processes so they start and stop together.
• Isolate failures: A crash in one part of the tree won't necessarily bring down unrelated parts.
• Improve modularity: Each supervisor can be responsible for a specific subsystem, making the application easier to understand and maintain.

Every process a supervisor manages is defined by a child specification. A child spec tells the supervisor how to start, restart, and shut down the child process.

Crucially, a child can be of two types:

• worker: A regular process (like a gen_server) that performs application logic.
• supervisor: Another supervisor process, forming a nested level in the tree.

Let's start with a simple worker process. This gen_server will be the leaf node in our supervision tree. It just prints messages when it starts or receives calls.

This supervisor will manage our my_worker_module processes. It defines two workers, 'WorkerA' and 'WorkerB', each with slightly different restart strategies.

Notice how its child_specs define type => worker.

This is the root of our complex tree. It supervises my_nested_sup. Notice that its child_spec for my_nested_sup has type => supervisor. This is how you build nested trees!

To see our tree in action, compile all three modules (my_worker_module.erl, my_nested_sup.erl, my_app_sup.erl) and then start the top-level supervisor. You'll see the workers start up!

You can use supervisor:which_children(PidOrName) to inspect the tree.

Our complex supervision tree looks like this:

• my_app_sup (top-level supervisor)
  • supervises my_nested_sup (a child supervisor)
    • supervises worker_A (a worker process)
    • supervises worker_B (a worker process)

This structure ensures that if worker_A crashes, only my_nested_sup handles it. If my_nested_sup itself crashes, my_app_sup will restart it, bringing worker_A and worker_B back to life.

Complex supervision trees are a cornerstone of building robust Erlang applications. They provide:

• Fault Isolation: Failures are contained to specific branches.
• Logical Grouping: Components with related functions are supervised together.
• Clear Responsibilities: Each supervisor has a well-defined set of processes it's responsible for.
• Scalability: Easier to add or remove subsystems without disrupting the entire application.

Today, we explored complex supervision trees in Erlang. We learned that supervisors can manage other supervisors, creating nested hierarchies. This powerful pattern enables robust fault tolerance by isolating failures, managing dependencies, and improving the modularity of your applications. By defining child specs with type => supervisor, you can build intricate and resilient Erlang systems.