Coroutines

Added in version 2.0.

Scrapy supports the coroutine syntax (i.e. async def).

Supported callables

The following callables may be defined as coroutines using async def, and hence use coroutine syntax (e.g. await, async for, async with):

Using Deferred-based APIs

In addition to native coroutine APIs Scrapy has some APIs that return a Deferred object or take a user-supplied function that returns a Deferred object. These APIs are also asynchronous but don’t yet support native async def syntax. In the future we plan to add support for the async def syntax to these APIs or replace them with other APIs where changing the existing ones is possible.

The following Scrapy methods return Deferred objects (this list is not complete as it only includes methods that we think may be useful for user code):

The following user-supplied methods can return Deferred objects (the methods that can also return coroutines are listed in Supported callables):

In most cases you can use these APIs in code that otherwise uses coroutines, by wrapping a Deferred object into a Future object or vice versa. See Integrating Deferred code and asyncio code for more information about this.

For example:

  • The ExecutionEngine.download() method returns a Deferred object that fires with the downloaded response. You can use this object directly in Deferred-based code or convert it into a Future object with maybe_deferred_to_future().

  • A custom download handler needs to define a download_request() method that returns a Deferred object. You can write a method that works with Deferreds and returns one directly, or you can write a coroutine and convert it into a function that returns a Deferred with deferred_f_from_coro_f().

General usage

There are several use cases for coroutines in Scrapy.

Code that would return Deferreds when written for previous Scrapy versions, such as downloader middlewares and signal handlers, can be rewritten to be shorter and cleaner:

from itemadapter import ItemAdapter


class DbPipeline:
    def _update_item(self, data, item):
        adapter = ItemAdapter(item)
        adapter["field"] = data
        return item

    def process_item(self, item, spider):
        adapter = ItemAdapter(item)
        dfd = db.get_some_data(adapter["id"])
        dfd.addCallback(self._update_item, item)
        return dfd

becomes:

from itemadapter import ItemAdapter


class DbPipeline:
    async def process_item(self, item, spider):
        adapter = ItemAdapter(item)
        adapter["field"] = await db.get_some_data(adapter["id"])
        return item

Coroutines may be used to call asynchronous code. This includes other coroutines, functions that return Deferreds and functions that return awaitable objects such as Future. This means you can use many useful Python libraries providing such code:

class MySpiderDeferred(Spider):
    # ...
    async def parse(self, response):
        additional_response = await treq.get("https://additional.url")
        additional_data = await treq.content(additional_response)
        # ... use response and additional_data to yield items and requests


class MySpiderAsyncio(Spider):
    # ...
    async def parse(self, response):
        async with aiohttp.ClientSession() as session:
            async with session.get("https://additional.url") as additional_response:
                additional_data = await additional_response.text()
        # ... use response and additional_data to yield items and requests

Note

Many libraries that use coroutines, such as aio-libs, require the asyncio loop and to use them you need to enable asyncio support in Scrapy.

Note

If you want to await on Deferreds while using the asyncio reactor, you need to wrap them.

Common use cases for asynchronous code include:

  • requesting data from websites, databases and other services (in start(), callbacks, pipelines and middlewares);

  • storing data in databases (in pipelines and middlewares);

  • delaying the spider initialization until some external event (in the spider_opened handler);

  • calling asynchronous Scrapy methods like ExecutionEngine.download() (see the screenshot pipeline example).

Inline requests

The spider below shows how to send a request and await its response all from within a spider callback:

from scrapy import Spider, Request
from scrapy.utils.defer import maybe_deferred_to_future


class SingleRequestSpider(Spider):
    name = "single"
    start_urls = ["https://example.org/product"]

    async def parse(self, response, **kwargs):
        additional_request = Request("https://example.org/price")
        deferred = self.crawler.engine.download(additional_request)
        additional_response = await maybe_deferred_to_future(deferred)
        yield {
            "h1": response.css("h1").get(),
            "price": additional_response.css("#price").get(),
        }

You can also send multiple requests in parallel:

from scrapy import Spider, Request
from scrapy.utils.defer import maybe_deferred_to_future
from twisted.internet.defer import DeferredList


class MultipleRequestsSpider(Spider):
    name = "multiple"
    start_urls = ["https://example.com/product"]

    async def parse(self, response, **kwargs):
        additional_requests = [
            Request("https://example.com/price"),
            Request("https://example.com/color"),
        ]
        deferreds = []
        for r in additional_requests:
            deferred = self.crawler.engine.download(r)
            deferreds.append(deferred)
        responses = await maybe_deferred_to_future(DeferredList(deferreds))
        yield {
            "h1": response.css("h1::text").get(),
            "price": responses[0][1].css(".price::text").get(),
            "price2": responses[1][1].css(".color::text").get(),
        }

Mixing synchronous and asynchronous spider middlewares

Added in version 2.7.

The output of a Request callback is passed as the result parameter to the process_spider_output() method of the first spider middleware from the list of active spider middlewares. Then the output of that process_spider_output method is passed to the process_spider_output method of the next spider middleware, and so on for every active spider middleware.

Scrapy supports mixing coroutine methods and synchronous methods in this chain of calls.

However, if any of the process_spider_output methods is defined as a synchronous method, and the previous Request callback or process_spider_output method is a coroutine, there are some drawbacks to the asynchronous-to-synchronous conversion that Scrapy does so that the synchronous process_spider_output method gets a synchronous iterable as its result parameter:

  • The whole output of the previous Request callback or process_spider_output method is awaited at this point.

  • If an exception raises while awaiting the output of the previous Request callback or process_spider_output method, none of that output will be processed.

    This contrasts with the regular behavior, where all items yielded before an exception raises are processed.

Asynchronous-to-synchronous conversions are supported for backward compatibility, but they are deprecated and will stop working in a future version of Scrapy.

To avoid asynchronous-to-synchronous conversions, when defining Request callbacks as coroutine methods or when using spider middlewares whose process_spider_output method is an asynchronous generator, all active spider middlewares must either have their process_spider_output method defined as an asynchronous generator or define a process_spider_output_async method.

For middleware users

If you have asynchronous callbacks or use asynchronous-only spider middlewares you should make sure the asynchronous-to-synchronous conversions described above don’t happen. To do this, make sure all spider middlewares you use support asynchronous spider output. Even if you don’t have asynchronous callbacks and don’t use asynchronous-only spider middlewares in your project, it’s still a good idea to make sure all middlewares you use support asynchronous spider output, so that it will be easy to start using asynchronous callbacks in the future. Because of this, Scrapy logs a warning when it detects a synchronous-only spider middleware.

If you want to update middlewares you wrote, see the following section. If you have 3rd-party middlewares that aren’t yet updated by their authors, you can subclass them to make them universal and use the subclasses in your projects.

For middleware authors

If you have a spider middleware that defines a synchronous process_spider_output method, you should update it to support asynchronous spider output for better compatibility, even if you don’t yet use it with asynchronous callbacks, especially if you publish this middleware for other people to use. You have two options for this:

  1. Make the middleware asynchronous, by making the process_spider_output method an asynchronous generator.

  2. Make the middleware universal, as described in the next section.

If your middleware won’t be used in projects with synchronous-only middlewares, e.g. because it’s an internal middleware and you know that all other middlewares in your projects are already updated, it’s safe to choose the first option. Otherwise, it’s better to choose the second option.

Universal spider middlewares

Added in version 2.7.

To allow writing a spider middleware that supports asynchronous execution of its process_spider_output method in Scrapy 2.7 and later (avoiding asynchronous-to-synchronous conversions) while maintaining support for older Scrapy versions, you may define process_spider_output as a synchronous method and define an asynchronous generator version of that method with an alternative name: process_spider_output_async.

For example:

class UniversalSpiderMiddleware:
    def process_spider_output(self, response, result, spider):
        for r in result:
            # ... do something with r
            yield r

    async def process_spider_output_async(self, response, result, spider):
        async for r in result:
            # ... do something with r
            yield r

Note

This is an interim measure to allow, for a time, to write code that works in Scrapy 2.7 and later without requiring asynchronous-to-synchronous conversions, and works in earlier Scrapy versions as well.

In some future version of Scrapy, however, this feature will be deprecated and, eventually, in a later version of Scrapy, this feature will be removed, and all spider middlewares will be expected to define their process_spider_output method as an asynchronous generator.

Since 2.13.0, Scrapy provides a base class, BaseSpiderMiddleware, which implements the process_spider_output() and process_spider_output_async() methods, so instead of duplicating the processing code you can override the get_processed_request() and/or the get_processed_item() method.