fix: allow calling Actor.reboot() from migrating handler, align reboot behavior with JS SDK

[fnesveda]

This fixes several issues with the Actor.reboot() behavior:

• Actor.reboot() waits for all event handlers to finish, but if itself it was called in an event handler, it would be waiting for itself, getting into a deadlock
• Actor.reboot() in the JS SDK triggers event handlers for the migrating and persistState events, but in the Python SDK it was triggering only the persistState handlers

This aligns the behavior to work like the JS SDK, and prevents reboot getting into an infinite loop by allowing it to be called only once.

Related PR in JS SDK: apify/apify-sdk-js#345

[janbuchar]

I might be missing something, but the chain.from_iterable doesn't seem to have a purpose here.

[fnesveda]

It does, _listeners_to_wrappers is a dict of dicts of lists of wrappers (because technically you can have one listener wrapped multiple times, even though it doesn't make practical sense).

Imagine this:

def my_event_listener(event_data):
    print(event_data)

Actor.on(Event.MIGRATING, my_event_listener)
Actor.on(Event.MIGRATING, my_event_listener)

Then _listeners_to_wrappers looks like this:

{
    Event.MIGRATING: {
        event_listener: [
            wrapper_for_my_event_listener_1,
            wrapper_for_my_event_listener_2,
        ],
    }
}

So _listeners_to_wrappers[Event.PERSIST_STATE]).values() is a list of lists, and with chain.from_iterable() I flatten it to a single-level list.

[janbuchar]

Huh, thanks. SDK already depends on Crawlee and that depends on more_itertools which contain flatten - could you add that dependency to SDK as well and use that instead? I know it serves no practical purpose other than readability.

[fnesveda]

Sure, will do.

[janbuchar]

LGTM