Skip to content
This repository has been archived by the owner on Feb 26, 2020. It is now read-only.
/ bahdit Public archive

A scalable search engine based on Accumulo.

License

Notifications You must be signed in to change notification settings

booz-allen-hamilton/bahdit

Repository files navigation

Overview

Bahdit is a search engine prototype developed at Booz Allen Hamilton during the summer of 2012 by a team of interns. The goal of Bahdit is to create a modular system of components to support online indexing of data in a scalable way that is not currently realized in other open source search solutions.

Bahdit uses Accumulo, a distributed key-value database, to store the document indices. The initial version includes a map/reduce based web crawling system that uses map/reduce primarily for parallel job control and works partially outside of usual map/reduce paradigms. This may eventually be retooled to better conform to map/reduce paradigms and use something like Cascading for job control and coordination.

Eventually a storm/message queue based restful indexing api is planned as well, which will mimic the Solr restful indexing api.

The actual index design is a hybrid term-document partitioned index, which has become popular for many indexing systems based on Accumulo. Essentially, a dictionary with term frequencies is used to find the lowest frequency term in a user query. This is used to access a term partitioned index, which itself contains a document-partitioned index.

Details

Bahdit has three main components: the map/reduce based web crawler and indexing control system, the java search api, and the web search UI.

Build

Build Bahdit by running the mvn package command. This will produce jars in the various sub-project target directories that can be used to run Bahdit.

Installation

Prerequisites

You must have Accumulo, Hadoop, and Zookeeper installed and functioning.

Configuration

See properties.conf and further details below.

Ingest

The Bahdit ingest application can, and should, be run as a user process. No real installation steps are necessary.

Web UI

The web UI can be run as a user process, as a system daemon, or in a web application container.

To run as a user process, simply copy the war to your home directory and include the properties.conf file in the classpath using the -cp option for java. The war file is runnable, so you can use java -jar to run it.

To run as a system daemon, you will need to create an init script to to start the application. Reference versions of how to do this may be found for applications such as Gerrit and Hudson which use a similar deployment strategy.

Follow the usual war file deployment instructions for your application container if you wish to deploy the application that way.

Running

Currently there are no bootstrap shell scripts for Bahdit, as a result you must construct your java command invocation by hand. Use java -jar. Further details on constructing java classpaths, for instance, are outside of the scope of this readme.

Bahdit requires a configuration file to be passed to the various utilities. This can be passed to the web UI by including a file named properties.conf in the root of the classpath. The ingest utilities accept the path to a file as their first argument.

Ingest Utilities

The ingest jar is an executable jar. Run it using java -jar. It includes built in help documentation which can be accessed using the --help option.

Web UI

The web UI is a runnable war file. You can run it using java -jar or by uploading it to your favorite JavaEE application container.

Be sure to include a properties.conf file in the root of the application classpath when running the web UI. This can be done from the command line using the -cp option for java or in an application container by putting the file in WEB-INF/classes.

A reference properties.conf is included in the source code; the options are fairly self-explainatory.

About

A scalable search engine based on Accumulo.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published