Skip to content

Latest commit

 

History

History
106 lines (69 loc) · 4.75 KB

README.md

File metadata and controls

106 lines (69 loc) · 4.75 KB
Raw

NEWS-CRAWL

Crawler for news based on StormCrawler. Produces WARC files to be stored as part of the Common Crawl. The data is hosted as AWS Open Data Set – if you want to use the data and not the crawler software please read the announcement of the news dataset.

Prerequisites

  • Java 8
  • Install Elasticsearch 7.5.0 (ev. also Kibana)
  • Install Apache Storm 1.2.4
  • Start Elasticsearch and Storm
  • Build ES indices by running bin/ES_IndexInit.sh

Crawler Seeds

The crawler relies on RSS/Atom feeds and news sitemaps to find links to news articles on news sites. A small collection of example seeds (feeds and sitemaps) is provided in ./seeds/. Adding support for news sites which do not provide a news feed or sitemap is an open issue, see #41.

Configuration

The default configuration should work out-of-the-box. The only thing to do is to configure the user agent properties send in the HTTP request header. Open the file conf/crawler-conf.yaml in an editor and fill in the values for http.agent.name and all further properties starting with the http.agent. prefix.

Run the crawl

Generate an uberjar:

mvn clean package

And run ...

storm jar target/crawler-1.18.1.jar org.commoncrawl.stormcrawler.news.CrawlTopology -conf $PWD/conf/es-conf.yaml -conf $PWD/conf/crawler-conf.yaml $PWD/seeds/ feeds.txt

This will launch the crawl topology. It will also "inject" all URLs found in the file ./seeds/feeds.txt in the status index. The URLs point to news feeds and sitemaps from which links to news articles are extracted and fetched. The topology will create WARC files in the directory specified in the configuration under the key warc.dir. This directory must be created beforehand.

Of course, it's also possible to add (or remove) the seeds (feeds and sitemaps) using the Elasticsearch API. In this case, the can topology can be run without the last two arguments.

Alternatively, the topology can be run from the crawler.flux, please see the Storm Flux documentation. Make sure to adapt the Flux definition to your needs!

Monitor the crawl

When the topology is running you can check that URLs have been injected and news are getting fetched on [http://localhost:9200/status/_search?pretty]. Or use StormCrawler's Kibana dashboards to monitor the crawling process. Please follow the instructions to install the templates for Kibana provided as part of StormCrawler's Elasticsearch module documentation.

There is also a shell script bin/es_status to get aggregated counts from the status index, and to add, delete or force a re-fetch of URLs. E.g.,

$> bin/es_status aggregate_status
3824    DISCOVERED
34      FETCHED
5       REDIRECTION

Run Crawl from Docker Container

First, download Apache Storm 1.2.4. from the download page and place it in the directory downloads:

STORM_VERSION=1.2.4
mkdir downloads
wget -q -P downloads --timestamping https://downloads.apache.org/storm/apache-storm-$STORM_VERSION/apache-storm-$STORM_VERSION.tar.gz

Do not forget to create the uberjar (see above) which is included in the Docker image. Simply run:

mvn clean package

Then build the Docker image from the Dockerfile:

Note: the uberjar is included in the Docker image and needs to be built first (see above).

docker build -t newscrawler:1.18.1 .

To launch an interactive container:

docker run --net=host \
    -v $PWD/data/elasticsearch:/data/elasticsearch \
    -v $PWD/data/warc:/data/warc \
    --rm --name newscrawler -i -t newscrawler:1.18.1 /bin/bash

NOTE: don't forget to adapt the paths to mounted volumes used to persist data on the host. Make sure to add the user agent configuration in conf/crawler-conf.yaml.

CAVEAT: Make sure that the Elasticsearch port 9200 is not already in use or mapped by a running ES instance. Otherwise Elasticsearch commands may affect the running instance!

The crawler is launched in the running container by the script

/home/ubuntu/news-crawler/bin/run-crawler.sh

After 1-2 minutes if everything is up, connect to Elasticsearch on port 9200 or Kibana on port 5601.