Saving data
A data extraction job would not be complete without saving the data for later use and processing. You've come to the final and most difficult part of this tutorial so make sure to pay attention very carefully!
Save data to the dataset
Crawlee provides a Dataset class, which acts as an abstraction over tabular storage, making it useful for storing scraping results. First, add the following import to the top of your file:
from crawlee.playwright_crawler import PlaywrightCrawler, PlaywrightCrawlingContext
from crawlee.storages.dataset import Dataset
# ...
Next, under the section where you create an instance of your crawler, create an instance of the dataset using the asynchronous constructor open():
# ...
async def main() -> None:
crawler = PlaywrightCrawler()
dataset = await Dataset.open()
# ...
Finally, instead of logging the extracted data to stdout, we can export them to the dataset:
# ...
@crawler.router.default_handler
async def request_handler(context: PlaywrightCrawlingContext) -> None:
# ...
data = {
'manufacturer': manufacturer,
'title': title,
'sku': sku,
'price': price,
'in_stock': in_stock,
}
# Push the data to the dataset.
await dataset.push_data(data)
# ...
Using a context helper
Instead of importing a new class and manually creating an instance of the dataset, you can use the context helper push_data. Remove the dataset import and instantiation, and replace dataset.push_data with the following:
# ...
@crawler.router.default_handler
async def request_handler(context: PlaywrightCrawlingContext) -> None:
# ...
data = {
'manufacturer': manufacturer,
'title': title,
'sku': sku,
'price': price,
'in_stock': in_stock,
}
# Push the data to the dataset.
await context.push_data(data)
# ...
Final code
And that's it. Unlike earlier, we are being serious now. That's it, you're done. The final code looks like this:
import asyncio
from crawlee.playwright_crawler import PlaywrightCrawler, PlaywrightCrawlingContext
async def main() -> None:
crawler = PlaywrightCrawler(
# Let's limit our crawls to make our tests shorter and safer.
max_requests_per_crawl=50,
)
@crawler.router.default_handler
async def request_handler(context: PlaywrightCrawlingContext) -> None:
context.log.info(f'Processing {context.request.url}')
# We're not processing detail pages yet, so we just pass.
if context.request.label == 'DETAIL':
# Split the URL and get the last part to extract the manufacturer.
url_part = context.request.url.split('/').pop()
manufacturer = url_part.split('-')[0]
# Extract the title using the combined selector.
title = await context.page.locator('.product-meta h1').text_content()
# Extract the SKU using its selector.
sku = await context.page.locator('span.product-meta__sku-number').text_content()
# Locate the price element that contains the '$' sign and filter out the visually hidden elements.
price_element = context.page.locator('span.price', has_text='$').first
current_price_string = await price_element.text_content() or ''
raw_price = current_price_string.split('$')[1]
price = float(raw_price.replace(',', ''))
# Locate the element that contains the text 'In stock' and filter out other elements.
in_stock_element = context.page.locator(
selector='span.product-form__inventory',
has_text='In stock',
).first
in_stock = await in_stock_element.count() > 0
# Put it all together in a dictionary.
data = {
'manufacturer': manufacturer,
'title': title,
'sku': sku,
'price': price,
'in_stock': in_stock,
}
# Push the data to the dataset.
await context.push_data(data)
# We are now on a category page. We can use this to paginate through and
# enqueue all products, as well as any subsequent pages we find.
elif context.request.label == 'CATEGORY':
# Wait for the product items to render.
await context.page.wait_for_selector('.product-item > a')
# Enqueue links found within elements matching the provided selector.
# These links will be added to the crawling queue with the label DETAIL.
await context.enqueue_links(
selector='.product-item > a',
label='DETAIL',
)
# Find the "Next" button to paginate through the category pages.
next_button = await context.page.query_selector('a.pagination__next')
# If a "Next" button is found, enqueue the next page of results.
if next_button:
await context.enqueue_links(
selector='a.pagination__next',
label='CATEGORY',
)
# This indicates we're on the start page with no specific label.
# On the start page, we want to enqueue all the category pages.
else:
# Wait for the collection cards to render.
await context.page.wait_for_selector('.collection-block-item')
# Enqueue links found within elements matching the provided selector.
# These links will be added to the crawling queue with the label CATEGORY.
await context.enqueue_links(
selector='.collection-block-item',
label='CATEGORY',
)
await crawler.run(['https://warehouse-theme-metal.myshopify.com/collections'])
if __name__ == '__main__':
asyncio.run(main())
What push_data does?
A helper context.push_data() saves data to the default dataset. You can provide additional arguments there like id or name to open a different dataset. Dataset is a storage designed to hold data in a format similar to a table. Each time you call context.push_data() or direct Dataset.push_data() a new row in the table is created, with the property names serving as column titles. In the default configuration, the rows are represented as JSON files saved on your file system, but other backend storage systems can be plugged into Crawlee as well. More on that later.
Each time you start Crawlee a default Dataset is automatically created, so there's no need to initialize it or create an instance first. You can create as many datasets as you want and even give them names. For more details see the Dataset.open() function.
Finding saved data
Unless you changed the configuration that Crawlee uses locally, which would suggest that you knew what you were doing, and you didn't need this tutorial anyway, you'll find your data in the storage directory that Crawlee creates in the working directory of the running script:
{PROJECT_FOLDER}/storage/datasets/default/
The above folder will hold all your saved data in numbered files, as they were pushed into the dataset. Each file represents one invocation of Dataset.push_data() or one table row.
Next steps
Next, you'll see some improvements that you can add to your crawler code that will make it more readable and maintainable in the long run.