incubator-cvs mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Apache Wiki <>
Subject [Incubator Wiki] Update of "LuceneConnectorFrameworkProposal" by KarlWright
Date Thu, 31 Dec 2009 14:51:00 GMT
Dear Wiki user,

You have subscribed to a wiki page or wiki category on "Incubator Wiki" for change notification.

The "LuceneConnectorFrameworkProposal" page has been changed by KarlWright.


  A connector framework is vital for search engines and other tools that need to access data
located in corporate repositories.  By abstracting the problem into a framework, applications
can code to a set of well-defined interfaces instead of having to use a different interface
for each connector.
- Key features:
-  * Connectors for many existing corporate repositories: Sharepoint, web crawling, RSS, Database,
FileNet, LiveLink, Documentum, etc.
-  * Supports incremental connections
-  * Provides awareness of security
-  * User interface for configuring the connectors
+ Connector Framework is an extendible incremental crawler, which uses a database to manage
configuration and crawl history, and provides reasonably high performance in accessing content
in multiple repositories for the main purpose of search engine indexing.  Connector Framework
also establishes a repository-specific security model which can be used to limit search user
access to repository content based on a user's identity.  Connector Framework also includes
existing connectors and authorities for:
+  * File system
+  * Windows shares
+  * JDBC-supported databases
+  * RSS feeds
+  * General websites
+  * LiveLink [from OpenText]
+  * Documentum [from EMC]
+  * SharePoint [from Microsoft]
+  * Meridio [from Meridio]
+  * Memex [from Memex]
+  * FileNet [from IBM]
+ Key design points for Connector Framework are as follows:
+  * Extendability - you can add new connectors for new repositories, and new authorities
for specific repository security models
+  * Incrementality - the ability to process only what changed between crawls, in a repository-specific
+  * Restartability - using a database with ACID properties to insure that crawls are safe
against process interruption or machine shutdown
+  * Security - establishing a model of security tokens that allows a search engine to enforce
a repository's security model
+  * Limited footprint - ability to operate reliably within a fixed amount of process memory,
regardless of configuration
+  * Performance - management of connector-specific resources to maximize overall thoughput
+  * Transparency - ability to generate reports on the activity of all crawls and repository
  == Background ==
@@ -26, +45 @@

  The Connector Framework fills an often significant gap in the Lucene experience, namely,
how to get content locked away in a content repository into Lucene/Solr/Nutch/Mahout/Tika.
 Naturally, many other tools (search engines and others) will also have this same problem.
 A Connector Framework would also be useful for someone wishing to migrate between content
repositories, too.
  = Current Status =
+ Connector Framework has been under development and in use in the field for close to five
years, deployed on a MetaCarta search engine appliance.  Almost all development of the project
has been done by Karl Wright (  Some individual connectors were developed
initially by contractors hired by MetaCarta, Inc., but maintenance and further development
is currently handled by the MetaCarta team.
+ Development of Connector Framework can therefore be viewed as core framework development,
plus development of individual connectors.  Core framework development is currently not a
terribly collaborative process, as there are no maintainers of the core functionality other
than Mr. Wright.  Development of new connectors has been done in the past in a much more collaborative
way by supplying a developer with a "development kit", and then integrating the resulting
connector (with whatever changes might have been necessary) into the source tree.
+ Reasonable efforts have been made to maintain the generality of the code base during the
time that MetaCarta has owned it.  Nevertheless, certain MetaCarta-specific changes have been
made which may require review and modification.  The following areas probably need to be addressed
in the code before graduation can occur:
+  1. Branding.  The UI brands it as a MetaCarta project.
+  2. Package names.  Package names would have to be changed.
+  3. How Connector Framework handles document delivery needs to be generalized, at least
for a single, configurable target output connector, and perhaps for multiple, independently-configurable
targets.  Simple example output connectors need to be written.
+  4. Connector Framework-specific dependent package modifications need to be addressed somehow.
For instance, the following projects that Connector Framework depends upon have been modified,
but the modifications have not been accepted upstream: commons-httpclient NTLMv2 and NTLM2
support [RSS, Web, SharePoint, Meridio, and Livelink connectors]; commons-httpclient custom
HTTPS protocol factory support [Web, SharePoint, Meridio, and Livelink connectors]; xerces
ability to handle non-legal RSS feeds [RSS and Web connectors]
+  5. MetaCarta-specific features, like document templates, are explicitly handled by the
UI and the infrastructure.  These features should be generalized so that they are controlled
by the choice of output connector.
+  6. Some specific hooks, namely support for configuration change notification, and for database
maintenance notification, may need to be made more generic.
+  7. Share Connector has a "fingerprinting" feature, which prefilters documents based on
a document type it surmises using a document inspection technique.  This feature is only viable
at the moment for very basic document types.  It should either be removed, or generalized
significantly to be much more flexible.
+  8. Documentation needs to be fleshed out, including javadoc and overall usage documents.
+  9. Tests need to be written and/or ported from MetaCarta's test suite.
+ Longer term, the project will likely grow into a more distributed crawler, where multiple
machines might well be involved in coordinated crawling activity.
  == Meritocracy ==
@@ -66, +103 @@

  == Cryptography ==
+ Connector Framework itself has no real cryptography component, although it does currently
obfuscate passwords it saves to the database or to a configuration file using a proprietary
algorithm.  The algorithm is present simply to avoid using cleartext and is not secure in
any sense other than by obscurity.
+ Various connectors, such as Share Connector, Web Connector, RSS Connector, SharePoint Connector,
LiveLink Connector, and Meridio Connector make use of cryptographic principles via secondary
libraries.  Specifically, these connectors support NTLM, NTLMv2, and NTLM2 Session authentication
via commons-httpclient and jCIFS.  The changes to commons-httpclient necessary to support
these varieties of Windows protocols have not yet been accepted upstream by the Apache httpclient
+ It is unknown at this time exactly to what degree the Oracle JDBC driver, the jtds JDBC
driver, or the Postgresql JDBC driver uses cryptography.  Also, the FileNet API class, the
Memex API classes, the OpenText LAPI api classes, and the Documentum DFC classes all may or
may not use cryptography.
  == Legal Concerns ==
@@ -90, +132 @@

  = Initial Source =
- All code is currently developed in-house at MetaCarta.
+ All initial sources will be coming from MetaCarta, Inc., with the goal of folding in changes
from others shortly thereafter.
  = Source and Intellectual Property Submission Plan =
+ Code IP grants need to be made from MetaCarta, Inc.  But, in addition, several connectors
(notably Documentum, LiveLink, Memex, and FileNet) rely directly on client API's in order
to be compiled.  Another connector (JDBC) relies on the existence of the Oracle JDBC Driver
in the classpath in order to enable crawls against Oracle databases.
+ It is unlikely that EMC, OpenText, Memex, or IBM would grant Apache-license-compatible use
of these client libraries.  Thus, the expectation is that users of these connectors obtain
the necessary client libraries from the owners prior to building or using the corresponding
connector.  An alternative would be to undertake a clean-room implementation of the client
API's, which may well yield suitable results in some cases (FileNet, Memex, FileNet), while
being out of reach in others (Documentum).  Conditional compilation, for the short term, is
thus likely to be a necessity.
+ Other external dependencies, such as jCIFS for the Share Connector, are licensed with LGPL,
and thus may need to be treated in a manner similar to the closed API's even though they are
open source.  These include the postgresql JDBC driver, and JTDS.
  TODO: Upload an tarball and a Software Grant
  = External Dependencies =
+ The project dependencies, other than on other Apache projects, are as follows:
+ The ConnectorFramework core currently uses the Bitmechanic JDBC pool driver, which is BSD
licensed, and the Postgresql JDBC driver, which is also BSD licensed.
+ The LiveLink Connector relies on LAPI, which is privately licensed by OpenText.
+ The Documentum Connector relies on DFC, which is privately licensed by EMC.
+ The Share Connector relies on jCIFS, which is LGPL.
+ The Memex Connector relies on privately licensed java libraries from Memex.
+ The FileNet Connector relies on privately licensed java libraries from IBM.
  = Required Resources =
   * Mailing lists

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message