Skip to content

A small tool which uses the CommonCrawl URL Index to download documents with certain file types or mime-types. This is used for mass-testing of frameworks like Apache POI and Apache Tika

License

Notifications You must be signed in to change notification settings

centic9/CommonCrawlDocumentDownload

Repository files navigation

Build Status Gradle Status Release GitHub release Tag Maven Central Maven Central

This is a small tool to find matching URLs and download the corresponding binary data from the CommonCrawl indexes.

Support for the newer URL Index (http://blog.commoncrawl.org/2015/04/announcing-the-common-crawl-index/) is available, older URL Index as described at https://github.com/trivio/common_crawl_index and http://blog.commoncrawl.org/2013/01/common-crawl-url-index/ is still available in the "oldindex" package.

Please note that a full run usually finds a huge number of files and thus downloading will require a large amount of time and lots of disk-space if the data is stored locally!

NOTE This project does not implement backoff on HTTP errors about too many requests. Due to the current high rate of access by many GPT/LLM experiments, the CommonCrawl S3 bucket very often returns HTTP errors about rate exceeded. See https://github.com/tballison/commoncrawl-fetcher-lite for a newer implementation of this with more advanced functionality that work more reliably.

NOTE: CommonCrawl only stores up to 1MB per file and cuts off any bytes exceeding this length. So larger documents will be truncated and might not be valid and parsable any more. You can try to download the original file via the URL that is part of the crawl-data, but this project does not implement this due to potential "crawling" restrictions on target websites.

Getting started

Grab it

git clone https://github.com/centic9/CommonCrawlDocumentDownload.git

Build it and create the distribution files

cd CommonCrawlDocumentDownload
./gradlew check

Run it

Fetch a list of interesting documents

./gradlew lookupURLs

Reads the current Common Crawl URL index data and extracts all URLs for interesting mime-types or file extensions, stores the URLs in a file called commoncrawl-CC-MAIN-<year>-<crawl>.txt

Download documents

./gradlew downloadDocuments

Uses the URLs listed in commoncrawl-CC-MAIN-<year>-<crawl>.txt to download the documents from the Common Crawl

Deduplicate files

./gradlew deduplicate

Some files have equal content, this task will detect these based on file-size and content-hash and move all duplicates to a backup-directory to leave only unique files in place.

Deprecated: Download documents from the old-index

./gradlew downloadOldIndex

Starts downloading the URL index files from the old index and looks at each URL, downloading binary data from the common crawl archives.

The longer stuff

Change it

Run unit tests

./gradlew check jacocoTestReport

Adjust which files are found

There are a few things that you can tweak:

  • The file-extensions that are detected as download-able files are handled in the class Extensions.
  • The mime-types that are detected as download-able files isare handled in the class MimeTypes.
  • Adjust the name of the list of found files in DownloadURLIndex.COMMON_CRAWL_FILE.
  • Adjust the location where files are downloaded to in Utils.DOWNLOAD_DIR.
  • The starting file-index (of the approximately 300 cdx-files) is currently set as constant in class org.dstadler.commoncrawl.index.DownloadURLIndex, this way you can also re-start a download that was interrupted before.

Adjust which commoncrawl-index is fetched

CommonCrawl periodically runs crawls and publishes them. You can switch to newer crawls by adjusting the constant CURRENT_CRAWL in DownloadURLIndex.java to the proper <year>-<week> number of the newer crawl.

See https://commoncrawl.org/connect/blog/ for announcemnts of the latest crawls.

Ideas

  • Old Index: By adding a new implementation of BlockProcesser (likely re-using existing stuff by deriving from one of the available implementations), you can do things like streaming processing of the file instead of storing the file locally, which will avoid using too much disk-space

Estimates (based on Old Index)

  • Size of overall URL Index is 233689120776, i.e. 217GB
  • Header: 6 Bytes
  • Index-Blocks: 2644
  • Block-Size: 65536
  • => Data-Blocks: 3563169
  • Aprox. Files per Block: 2.421275
  • Resulint aprox. number of files: 8627412
  • Avg. size per file: 221613
  • Needed storage: 1911954989425 bytes = 1.7TB!

Related projects/pages

Release it

./gradlew --console=plain release && ./gradlew closeAndReleaseRepository
  • This should automatically release the new version on MavenCentral
  • Afterwards go to the Github releases page and add release-notes

Support this project

If you find this library useful and would like to support it, you can Sponsor the author

Licensing

About

A small tool which uses the CommonCrawl URL Index to download documents with certain file types or mime-types. This is used for mass-testing of frameworks like Apache POI and Apache Tika

Topics

Resources

License

Stars

Watchers

Forks

Sponsor this project

 

Packages

No packages published