Error Handling / Supervision

[christoph-fricke]

Elixir's and Akka actor system follows a "let it crash" mentality and use supervisors to create actors trees, which are partly restarted if an actor in the tree crashes¹². XState is currently not really concerned with proper error handling and it is handled rather inconsistently, although errors in the system are inevitable. There have been a few discussions on the Discord³ and in the RFCs⁴, which can inspire a solution.

We might try to tackle the problem in XSystem first, before forces changes in XState, since XSystem is smaller and easier to change. A solution probably requires our own spawning implementation and potential extensions to the Behavior and ActorRef interface.

An initial thought is to wrap any behavior transition in a spawn function in a try/catch block and "restart" the behavior in case the catch block is reached.

I am open for ideas and suggestions regarding this topic. 😄

Footnotes

1. https://elixir-lang.org/getting-started/mix-otp/supervisor-and-application.html ↩

2. https://doc.akka.io/docs/akka/current/typed/fault-tolerance.html ↩

3. https://discord.com/channels/795785288994652170/877478623202906132/895041466538266704 ↩

4. https://github.com/statelyai/rfcs/blob/0001-errors/text/0001-errors.md ↩