Scrapy S3 Pipeline

Scrapy pipeline to store items into Amazon S3 or Google Cloud Storage (GCS) bucket. Unlike built-in FeedExporter, the pipeline has the following features:

• The pipeline upload items to S3/GCS by chunk while crawler is running.
  • From Scrapy 2.3, built-in FEED_EXPORT_BATCH_ITEM_COUNT does almost the same thing.
• Support GZip compression.

The pipeline aims to run crawler and scraper in different processes, e.g. run crawler process with Scrapy in AWS Fargate and run scraper process with lxml in AWS Lambda.

Requirements

• Python 3.6+ (Tested in 3.9)
• Scrapy 1.1+ (Tested in 2.4)
• boto3 or google-cloud-storage

Install

For S3 users:

$ pip3 install scrapy-s3pipeline[s3]

For GCS users:

$ pip3 install scrapy-s3pipeline[gcs]

Getting started

1. Install Scrapy S3 Pipeline with pip.

   $ pip3 install scrapy-s3pipeline[s3]
   

   or

   $ pip3 install scrapy-s3pipeline[gcs]
   

2. Add 's3pipeline.S3Pipeline' to ITEM_PIPELINES setting in your Scrapy project.

   ITEM_PIPELINES = {
       's3pipeline.S3Pipeline': 100,  # Add this line.
   }
   

3. Add S3PIPELINE_URL setting. You need to change my-bucket to your bucket name.

   # For S3 users
   S3PIPELINE_URL = 's3://my-bucket/{name}/{time}/items.{chunk:07d}.jl.gz'
   
   # For GCS users
   S3PIPELINE_URL = 'gs://my-bucket/{name}/{time}/items.{chunk:07d}.jl.gz'
   GCS_PROJECT_ID = 'my-project' # Change to your project id
   

4. Setup AWS/GCP credentials.

   For S3 users:

   Setup AWS credentials via aws configure command or environment variables. Alternatively, use Scrapy's settings AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY.

   For GCS users:

   Setup GCP credentials via gcloud auth application-default login command or environment variable GOOGLE_APPLICATION_CREDENTIALS. Alternatively, you can set json string of service account's key file to GOOGLE_APPLICATION_CREDENTIALS_JSON settings.

5. Run your spider. You will see items in your bucket after 100 items are crawled or the spider is closed.

Settings

S3PIPELINE_URL (Required)

S3/GCS Bucket URL to store items.

e.g.:

• S3: s3://my-bucket/{name}/{time}/items.{chunk:07d}.jl.gz
• GCS: gs://my-bucket/{name}/{time}/items.{chunk:07d}.jl.gz

The following replacement fields are supported in S3PIPELINE_URL.

• {chunk} - gets replaced by a start index of items in current chunk, e.g. '0', '100', '200',....
• {time} - gets replaced by a timestamp when the spider is started.

You can also use other spider fields, e.g. {name}. You can use format string syntax here, e.g. {chunk:07d}.

File format is determined by a file extension in the URL. For example, if S3PIPELINE_URL ends with .json or .json.gz, JSON format is used. See Scrapy's built-in FEED_EXPORTERS settings for supported formats. If the file extension is not available in FEED_EXPORTERS, JSONLines format is used by default.

S3PIPELINE_MAX_CHUNK_SIZE (Optional)

Default: 100

Max count of items in a single chunk.

S3PIPELINE_MAX_WAIT_UPLOAD_TIME (Optional)

Default: 30.0

When no new item is processed in more than S3PIPELINE_MAX_WAIT_UPLOAD_TIME seconds, it will be forced to upload a chunk.

S3PIPELINE_GZIP (Optional)

Default: True if S3PIPELINE_URL ends with .gz; otherwise False.

If True, uploaded files will be compressed with Gzip.

Page item

For convinience, Scrapy S3 Pipeline provides s3pipeline.Page item class to store entire HTTP body. It has url, body and crawled_at fields.

This make it easy to store entire HTTP body and run scraper in other process. It's friendly to server-less architecture which run scraper in AWS Lambda.

Example usage of Page:

from datetime import datetime, timezone

import scrapy
from s3pipeline import Page

# ...

class YourSpider(scrapy.Spider):

    # ...

    def parse(self, response):
        # You can create Page instance just one line.
        yield Page.from_response(response)

        # Or, you can fill item fields manually.
        item = Page()
        item['url'] = response.url
        item['body'] = response.text
        item['crawled_at'] = datetime.now(timezone.utc).replace(microsecond=0).isoformat()
        yield item

Note: Page's body is omitted when printed to logs to improve readbility of logs.

Development

Test

$ python3 setup.py test

Release

$ pip install twine wheel

$ python3 setup.py bdist_wheel sdist
$ twine upload dist/*