Constructor `_ActorType.__init__` is being executed twice

[vdusek]

Description

The _ActorType.__init__ constructor is executed twice when creating an instance of Actor.

Reproduction

I added simple prints to the _ActorType.__init__, _ActorType.__aenter__ and _ActorType.init methods and prepared a simple Actor:

from apify import Actor

async def main() -> None:
    async with Actor() as actor:
        actor.log.info(f'Hello! Actor={Actor}, actor={actor}.')

Execution:

$ apify run
Run: /home/vdusek/Projects/apify-sdk-python/.venv/bin/python3 -m src
Hello from _ActorType.__init__
Hello from _ActorType.__init__
Hello from _ActorType.__aenter__
Hello from _ActorType.init
[apify] INFO  Initializing Actor...
[apify] INFO  System info ({"apify_sdk_version": "2.2.2", "apify_client_version": "1.8.1", "crawlee_version": "0.5.3", "python_version": "3.13.2", "os": "linux"})
[apify] INFO  Hello! Actor=<apify.Actor>, actor=<apify._actor._ActorType object at 0x7ff7f757fed0>.
[apify] INFO  Exiting Actor ({"exit_code": 0})

This does not happen when using the Actor class without an instance, as shown below:

from apify import Actor

async def main() -> None:
    async with Actor:
        Actor.log.info(f'Hello! Actor={Actor}.')

Output

$ apify run
Run: /home/vdusek/Projects/apify-sdk-python/.venv/bin/python3 -m src
Hello from _ActorType.__init__
Hello from _ActorType.__aenter__
Hello from _ActorType.init
[apify] INFO  Initializing Actor...
[apify] INFO  System info ({"apify_sdk_version": "2.2.2", "apify_client_version": "1.8.1", "crawlee_version": "0.5.3", "python_version": "3.13.2", "os": "linux"})
[apify] INFO  Hello! Actor=<apify.Actor>.
[apify] INFO  Exiting Actor ({"exit_code": 0})

[vdusek]

I will assign @janbuchar as he wrote the Actor refactor to use the lazy object proxy.

[janbuchar]

I'm aware of this. Could you explain how does it become a problem? Until the instance is initialized, all should be fine, right?

[vdusek]

I find the fact that _ActorType.__init__ is executed twice problematic itself. E.g. when I was integrating the service locator recently, this behavior became a limiting factor.

However, even if we want to take a more pragmatic approach, and decide to ignore it until it causes a more serious issue, I suspect that this is already affecting Scrapy-Actors, preventing them from running Actor instance altogether.

When I modify the Actor from Scrapy guide like this:

async def main() -> None:
    """Apify Actor main coroutine for executing the Scrapy spider."""
    async with Actor() as actor:
        # Retrieve and process Actor input.
        actor_input = await actor.get_input() or {}
        start_urls = [url['url'] for url in actor_input.get('startUrls', [])]
        allowed_domains = actor_input.get('allowedDomains')
        proxy_config = actor_input.get('proxyConfiguration')

        # Apply Apify settings, which will override the Scrapy project settings.
        settings = apply_apify_settings(proxy_config=proxy_config)

        # Create CrawlerRunner and execute the Scrapy spider.
        crawler_runner = CrawlerRunner(settings)
        crawl_deferred = crawler_runner.crawl(
            Spider,
            start_urls=start_urls,
            allowed_domains=allowed_domains,
        )
        await deferred_to_future(crawl_deferred)

I get the following error output:

[title_spider] INFO  TitleSpider is parsing <200 https://discord.apify.com/>... ({"spider": "<TitleSpider 'title_spider' at 0x7ff0ec3201a0>"})
[scrapy.core.scraper] ERROR Error processing {'title': 'Apify Community', 'url': 'https://discord.apify.com/'} ({"spider": "<TitleSpider 'title_spider' at 0x7ff0ec3201a0>"})
      Traceback (most recent call last):
        File "/home/vdusek/Projects/apify-sdk-python/.venv/lib/python3.13/site-packages/twisted/internet/defer.py", line 1253, in adapt
          extracted: _SelfResultT | Failure = result.result()
                                              ~~~~~~~~~~~~~^^
        File "/home/vdusek/Projects/apify-sdk-python/src/apify/scrapy/pipelines/actor_dataset_push.py", line 30, in process_item
          await Actor.push_data(item_dict)
        File "/home/vdusek/Projects/apify-sdk-python/src/apify/_actor.py", line 454, in push_data
          self._raise_if_not_initialized()
          ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
        File "/home/vdusek/Projects/apify-sdk-python/src/apify/_actor.py", line 177, in _raise_if_not_initialized
          raise RuntimeError('The Actor was not initialized!')
      RuntimeError: The Actor was not initialized!

I suspect that the double initialization, or something related to that, might be responsible for this issue (?).

[janbuchar]

I suspect that the double initialization, or something related to that, might be responsible for this issue (?).

Almost - the problem is that a new instance of _ActorType is made (and stored as actor), but then the scrapy pipeline calls Actor.push_data, and Actor means "the default instance". And I believe this pushes us towards what we actually need to fix:

async with Actor(dunno="dontcare"):
    await Actor.push_data({"hello": "world"})

⬆ this here should work, not throw a runtime error (current behavior). Am I right?

[vdusek]

Well, I tried exactly that, and it failed as well.

[janbuchar]

Well, I tried exactly that, and it failed as well.

So we agree that it shouldn't, right? 😁

[vdusek]

Yep, the Actor identifier, which is a lazy proxy, always returns the default instance, even if a new one is created. This means the issue isn't caused by a double __init__ call, but rather by the way the lazy proxy object is being used. I'm not sure whether this has an easy solution.

[janbuchar]

I'll see what I can come up with 🙂

[vdusek]

Technically, the behavior from the issue title still persists, but for now, there are no known issues caused by it, so I'm okay with closing it.