Re-spawning failed actors

[bobkocisko]

Is there a recommended pattern to concisely re-spawn failed actors? The dissertation suggests potential tools to use to notice dead actors like a retraction of an advertisement but it's unclear to me how to retry the failed operation in those cases without code duplication (especially in the case of something like during/spawn)

[howell]

There is a supervise form and a small example of its use here.

Supervision could be integrated more tightly with during/spawn; as is, you can use the optional #:on-crash argument to respond/restart the actor. (This option is actually even lightly documented!)

[bobkocisko]

How would one re-spawn the same actor when it dies using during/spawn with #:on-crash? I see the optional argument but I'm not sure how to use that to cause the previously-defined spawn to re-execute itself.

[howell]

You would have the #:on-crash spawn an actor with the same behavior. To avoid duplicating the code, you'd probably need to lift it out to a function, as in:

(assertion-struct example (x y))
(spawn
  (define (actor-body x y)
    ... endpoints in here ...)
  (during/spawn (example $x $y)
    #:on-crash (spawn (actor-body x y))
    (actor-body x y))))

It looks like this could be pretty straightforward to add as a macro if it seems useful enough.

[bobkocisko]

Ok nice, thanks! I wasn't sure if the lexical context was important with these clauses. Good to know that the actions can be pulled out into their own function. I agree that a macro would be nice!