Filesystem Indexing with libferris

The libferris Project began in early 2001 in order to create a virtual filesystem operating as a shared library. Many tree-like structures are presented through a single filesystem interface by libferris. Operating in the user address space instead of in the kernel allows libferris to expose a large number of tree-like sources. These sources would be difficult to access from the Linux kernel. All filesystems are accessible through the root:// URI in libferris and include kernel file:// URLs, relational databases, XML files and databases, network-accessible resources like HTTP/FTP servers and other composite files like db4, tarballs and RDF, as well as standard kernel filesystems like ext3 and XFS.

The notion of files and directories is merged in libferris into a single abstraction. This allows things like tar archives to be mounted as a filesystem implicitly by libferris. In this case, the tarball is both a file and a directory at the same time.

The extended attribute (EA) interface presents data from many sources, including the kernel's listxattr(2) interface, RDF/bdb repositories and dynamically extracted values. An example of a dynamic EA is the width of an image. When an image's width EA is read, libferris uses a plugin to determine the width of the image file. Another example is the sample rate for audio files.

More EAs and their descriptions are listed on the libferris Web site (see the on-line Resources).

This leads to two distinct types of indexes that libferris can create and query—full-text and EA. Full-text indexes allow you to find files based on the words that they contain. EA indexes allow you to find files based on the metadata for the file. The indexing structures needed to resolve queries against full text or EA are significantly different. For example, full-text indexes may store the list of documents containing each word (an inverted file) to resolve queries such as “find all documents containing the word libferris”. EA indexes need to be able to handle range queries such as “find all files modified last month”.

libferris uses plugins to handle the implementation of these indexes. For full-text indexes, you can use any or all of these: an internal format based on inverted files, Apache Lucene compiled with gcj, an ODBC supporting relational database, Xapian or the TSearch2 module in PostgreSQL. For EA indexes you can choose an internal design based on sorted inverted files, LDAP, Apache Lucene compiled with gcj, an ODBC supporting relational database or native PostgreSQL using some PGSQL. The recommended choices for general use are Xapian or TSearch2 for full text, and PostgreSQL or ODBC for EA indexes.

The PostgreSQL modules are similar to those of the ODBC, but they use PGSQL and other PostgreSQL-specific functionality. Using the PostgreSQL TSearch2 plugin for full text requires a template database to be set up on your PostgreSQL server. See the on-line Resources for details.

All indexes exist in their own directory in libferris. Default full-text and EA indexes are in ~/.ferris in full-text-index and ea-index, respectively. Indexes are created with either fcreate or gfcreate from the ferriscreate package. Like many tools in libferris, the gf prefixed tool does much the same thing as the f prefixed tool, but it offers a GTK+2 interface. The following sections describe creation, population and querying of both index types.