Atomicity of handlers

[zesterer]

Atomicity of handlers

[zesterer]

Thanks, that clarifies things.

[thomaseizinger]

I am not too familiar with actix API but my thoughts on long-running handlers are:

• If you have a long-running task, you should just spawn that into a runtime and send a message back to the actor once finished.
• Be careful to not circumvent back-pressure this way. Spawning tasks is "unbounded" on pretty much all executors I think?
• Tasks are not owned by an actor, meaning they will keep running even if the actor is shut down. Keep that in mind when you spawn tasks from handlers.

For IO, I'd typically spawn 1(!) task that reads from the IO source and immediately delegates to an actor, waiting for the handler to finish before reading the next message.

Does this help?