CMS, wikis, text files … modern companies store important data in many different places, and that data must be accessible down to the tiniest detail through a single search. Commercial software vendors such as Google [1] offer tools that will index the data and store the index on an external server. But many organizations prefer to keep control of the search capabilities – for security and privacy reasons, but also to add flexibility and promote innovation and customization.

A handy constellation of open source tools from the Apache project will help you build your own search index for the assorted documents and data on your network: Nutch, Solr, Apache, and Lucene.

Nutch [2] is a powerful web crawler, and Apache Solr [3] is a search engine based on Apache Lucene [4]. You can combine Nutch with Solr to create a complete search engine – a miniature Google, if you like.

The Nutch crawler uses HTTP and FTP to discover information. If you want Nutch to inspect your local files, you need to store the files on an HTTP or FTP server and point to the directories you want Nutch to crawl. Nutch fetches data that is then searched and indexed by Solr. Solr depends on the Apache Lucene search libraries and is written in Java, and it requires a Java Servlet container server. The Jetty Java Servlet container tool is installed by default, but many users prefer a more robust solution such as Apache Tomcat. (See "A Note of Caution" box for more info.)

This workshop shows how to build your own search engine using on an Ubuntu 14.04.2 LTS system.

Installing the Components

On Canonical's Enterprise Linux, Solr is available from the package sources; you only need to install Nutch manually (Listing 1, lines 1-4). Then back up Solr's default XML schema and replace it with the file supplied by Nutch (Lines 6 and 7).

apt-get install solr-tomcat
wget http://www.eu.apache.org/dist/nutch/1.9/apache-nutch-1.9-bin.tar.gz
tar vfx apache-nutch-1.9-bin.tar.gz
mv apache-nutch-1.9 /opt/nutch
mv /etc/solr/conf/schema.xml /etc/solr/conf/schema.xml.orig
cp /opt/nutch/conf/schema.xml /etc/solr/conf/schema.xml

By default, the server does not save the content of pages or documents it finds. When it re-indexes, it transfers all the contents again. If you want to enable caching, you can do so in the /etc/solr/conf/schema.xml configuration file by changing the stored="false" entry in the following line:

<field name="content" type="text" stored="false" indexed="true"/>

line to "true"; then restart Tomcat by typing service tomcat6 restart.

Configuring the Nutch Crawler

Although you can control the crawler's default behavior with the /opt/nutch/conf/nutch-default.xml file, it makes more sense to customize the /opt/nutch/conf/nutch-site.xml file with site-specific details.

The example in Listing 2 shows how you can configure the name of the HTTP agent. This name will appear in the web server's logfiles.

01 <?xml version="1.0"?>
02 <?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
03 <!-- Put site-specific property overrides in this file. -->
04   <configuration>
05     <property>
06        <name>http.agent.name</name>
07        <value>Company Search Agent</value>
08     </property>
09   </configuration>

The nutch-default.xml file contains various settings that control the crawler's behavior. In nutch-site.xml, you need to do this:

<property>
  <name>file.content.ignored</name>
  <value>false</value>
</property>

Additionally, you need Nutch to remove any documents that the users have deleted in the meantime from the search engine's database:

<property>
  <name>db.update.purge.404</name>
  <value>true</value>
</property>

On a local network, with few servers and clients compared with the Internet, the five-second default setting between two requests to the same server leads to an unnecessarily large number of inactive threads, which slows down the search engine. The fetcher.server.delay is useful for ensuring that the search engine will not overload a server with requests:

<property>
  <name>fetcher.server.delay</name>
  <value>0.0</value>
</property>

It makes sense to disable this value and only re-enable it if problems occur.

Large Documents

On the Internet, it is sometimes useful to index large documents, but you need to be careful not to let the crawler get hung up on a gigantic tome with no useful information. Nutch lets you define the content.limit class parameters that define the maximum size of the content that crawler processes (Listing 3). You can also define the length of the document title, say, to achieve a more informative view in the search results – the value is in characters not in bytes:

<property>
<name>indexer.max.title.length</name>
  <value>150</value>
</property>

Another useful variable, fetcher.threads.fetch, defines the number of concurrent threads reading content. The http.timeout reduces the time the thread needs to wait for a request to time out.

01 <property>
02   <name>file.content.limit</name>
03   <value>131072</value>
04 </property>
05 <property>
06   <name>http.content.limit</name>
07   <value>131072</value>
08 </property>
09 <property>
10   <name>ftp.content.limit</name>
11   <value>131072</value>
12 </property>