[Question] How to stop an Actor?

[fermuch]

Suppose I have a system where I spawn an actor to manage each item in a list. Once the item is removed from the list, how can I stop the Actor?

• tarant version: 3.1.1

[gpicron]

any answer ?

[fermuch]

Sadly not. I had to create a special message requesting the Actor to "die". Once the actor receives that message, it kills itself.

I would love to have an official way to kill actors, tho'.

[kmruiz]

Hi @fermuch,

apologies for the late response.

You are totally right, there is no specific way to stop a single actor right now, as there would be different strategies to do it and we (@kanekotic and I) were designing which would be the best way. Can I ask you what would you consider the most efficient way to stop an actor to gather your feedback?

Questions we were trying to answer:

• What happens with the mailbox when an actor is stopped? Should we process pending messages? What happens when a pending message is then failing?
• What happens when an actor is stopping and receives new messages? Should they go to a dead letter? Or notify the sender that the actor does not exist anymore?

Thanks a lot and sorry for the late response.

[fermuch]

What happens with the mailbox when an actor is stopped? Should we process pending messages? What happens when a pending message is then failing?

I think there should be two ways to stopping an actor: request and force.

REQUESTING

• The actor should decide if it must stop (a callback with a boolean response?)
• The actor should stop getting new messages (trying to append to a stopped actor should raise an error).
• The actor should process every message it had right until the stop message.
• A last message should be sent, so the actor can clean its resources.

FORCING

• The actor is stopped immediately.
• All pending messages in the mailbox are discarded.

Note: in both cases, the requester (sender) of the stop message should have the possibility to wait for the actor to be fully stopped, so a new actor can be spawned with the same details.

---

By separating into these two, I think all use cases are solved. By requesting, you give the actor time to process its pending messages. And, if you need, you can kill it immediately.