I’m starting to get deeper into the subtleties of supervisors and I could use some help understanding how and when to restart failed tasks.

Given I have a risky_function() that may raise errors, I can ensure that it eventually completes by supervising a task that calls it by doing something like this (assume MyTaskSupervisor is started):

Task.Supervisor.start_child(MyTaskSupervisor, fn -> 
  risky_work() end, 
  restart: :transient)

Even if the risky_work function hits errors, it restarts and it eventually completes successfully.

This is (I think) an example of a non-awaited task. How would this need to be structured if I wanted that same risky_work to be part of an awaited task?

The following attempt doesn’t ensure that the task completes:

task = Task.Supervisor.async_nolink(MyTaskSupervisor, fn ->
  risky_work()
end)
Task.await(task) # or Task.yield(task)

The risky_work function is called immediately and any errors it raises are immediately visible. The docs show how Task.Supervisor.async_nolink/3 might get used inside of a GenServer… I can get that example working, but it still doesn’t ensure that the risky_work ever completes. I can see where I can restart the task – there’s even a comment in the example # Log and possibly restart the task..., but if I am restarting the task, I feel like I’m doing something wrong. Isn’t that what the supervisor is supposed to do?

Thanks for any clarifications! I feel like maybe I’m thinking about this the wrong way.