Friday 10 June 2016

What's new in StormCrawler 0.10?



The version 0.10 of Storm-Crawler has just been released.  It contains many improvements and bugfixes and we recommend all existing users to upgrade to it.

Here are the main changes :

Core

Apart from the usual dependency upgrades (Apache Tika 1.12, Jsoup 1.8.3) and various bugfixes (notably #280 and 293), we completely removed the configuration files (parse and URL filters) from the core module (#227) as these should be specified by the user or provided by the archetype. This fixes an old issue we were having with user files being possibly overwritten with the ones provided by the core jar when generating the big jar.

We also added a mechanism to allow custom scheduling of the URLs based on metadata (#283). This applies to fetched documents only for the time being and can be used to revisit some pages more frequently depending on their nature,  for instance news feeds (see below). 

It is now possible to configure a list of metadata to persist (metadata.persist) into the status storage but not transfer to the outlinks of a document (#293). This is useful for the _redirTo values that we now generate to track the target of redirections (#96). Until now this would have been passed on to the outlinks, which would have been wrong.

The JSoupParser now has the option of passing on documents which it can't handle to the default stream so that another bolt can try to deal with them (#266).  This can be useful e.g. to chain the JSoup parser - which generates a good DOM for XPath extraction, with the Tika one which gives rubbish DOM but can handle all sorts of file formats. JSoupParser also generates more complete DOMs, including inline Javascript and css nodes (#219).

New resources

LinkParseFilter takes Xpath expressions in the parsefilter configuration and allows to add the matching elements to the outlinks.

FeedParserBolt uses the ROME library to process news feeds and generate the outlinks accordingly.

There is a separate repository for resources to generate WARC files. These might be moved to the storm-crawler repository later on.

Elasticsearch

There has been loads of work done in this module. The main thing is that we upgraded the code to the current version 2.3.1 of Elasticsearch (#275), which should provide better performance and new functionalities. The Kibana dashboards have also been improved and can display generic metrics for Apache Storm (receive queues, memory heaps),  top hosts/domains on status dashboard (see below) and total bytes fetched on the crawl metrics dashboard.

Kibana dashboard showing the status counts and top hostnames

What's next?

There is already a branch to move the code to Storm 1.x and some related improvements (#278). We will keep improving the existing components, in particular add sampling to the AggregationSpout for Elasticsearch.

Thanks to all users and contributors who helped with this release. Remember that you can follow the project on Twitter @stormcrawlerapi.