ApacheConUs2009MeetUp

We had a "Web Crawler Developer" MeetUp at this year's ApacheCon US in Oakland.

It wound up being an UnMeetUp (MeetDown?) on Wednesday, November 4th from 11am - 1pm.

Attendees

• Andrzej Bialecki - Apache Nutch
• Thorsten Sherler - Apache Droids
• Michael Stack - Formerly with Heritrix, now HBase
• Ken Krugler - Bixo

Topics

Roadmaps

• Nutch - become more component based.
• Droids - get more people involved.

Sharable Components

• robots.txt parsing
• URL normalization
• URL filtering
• Page cleansing
  • General purpose
  • Specialized
• Sub-page parsing (portlets)
• AJAX-ish page interactions
• Document parsing (via Tika)

• HttpClient (configuration)

• Text similarity
• Mime/charset/language detection

Tika

• Needs help to become really usable
• Would benefit from large test corpus
• Could do comparison with Nutch parser
• Needs option for direct DOM querying (screen scraping tasks)

• Handles mime & charset detection now (some issues)

• Could be extended to include language detection (wrap other impl)

URL Normalization

• Includes both domain (www.x.com == x.com), path, and query portions of URL
• Often site-specific rules
  • Option to derive rules using URLs to similar documents.

AJAX-ish Page Interaction

• Not applicable for broad/general crawling
• Can be very important for specific web sites
• Use Selenium or headless Mozilla

Component API Issues

• Want to avoid using an API that's tied too closely to any implementation.
• One option is to have simple (e.g. URL param) API that takes meta-data.
  • Similar to Tika passing in of meta-data.

Hosting Options

• As part of Nutch - but easy to get lost in Nutch codebase, and can be associated too closely with Nutch.
• As part of Droids - but Droids is both a framework (queue-based) and set of components.
• New sub-project under Lucene TLP - but overhead to set up/maintain, and then confusion between it and Droids.
• Google code - seems like a good short-term solution, to judge level of interest and help shake out issues.

Next Steps

• Get input from Gordon re Heritrix. Stack to follow up with him. Ideally he'd add his comments to this page.
• Get input from Thorsten on Google code option. If OK as starting point, then Andrzej to set up.

• Make decision about build system (and then move on to code formatting debate

  • I'm going to propose ant + maven ant tasks for dependency management. I'm using this with Bixo, and so far it's been pretty good.
• Start contributing code
  • Ken will put in robots.txt parser.

Original Discussion Topic List

Below are some potential topics for discussion - feel free to add/comment.

• Potential synergies between crawler projects - e.g. sharing robots.txt processing code.
• How to avoid end-user abuse - webmasters sometimes block crawlers because users configure it to be impolite.
• Politeness vs. efficiency - various options for how to be considered polite, while still crawling quickly.
• robots.txt processing - current problems with existing implementations
• Avoiding crawler traps - link farms, honeypots, etc.

• Parsing content - home grown, Neko/!TagSoup, Tika, screen scraping

• Search infrastructure - options for serving up crawl results (Nutch, Solr, Katta, others?)
• Testing challenges - is it possible to unit test a crawler?
• Fuzzy classification - mime-type, charset, language.
• The future of Nutch, Droids, Heritrix, Bixo, etc.
• Optimizing for types of crawling - intranet, focused, whole web.