Sunday, 21 August 2011

Omega overview

Omega overview

If you just want a very quick overview, you might prefer to read the quick-start guide.

Omega operates on a set of databases. Each database is created and updated separately using either omindex or scriptindex. You can search these databases (or any other Xapian database with suitable contents) via a web front-end provided by omega, a CGI application. A search can also be done over more than one database at once.

There are separate documents covering CGI parameters, the Term Prefixes which are conventionally used, and OmegaScript, the language used to define omega's web interface. Omega ships with several OmegaScript templates and you can use these, modify them, or just write your own. See the "Supplied Templates" section below for details of the supplied templates.

Omega parses queries using the Xapian::QueryParser class - for the supported syntax, see queryparser.html in the xapian-core documentation - available online at: http://xapian.org/docs/queryparser.html

Term construction

Documents within an omega database are stored with two types of terms: those used for probabilistic searching (the CGI parameter 'P'), and those used for boolean filtering (the CGI parameter 'B'). Boolean terms start with an initial capital letter denoting the 'group' of the term (e.g. 'M' for MIME type), while probabilistic terms are all lower-case, and are also stemmed before adding to the database.

The "english" stemmer is used by default - you can configure this for omindex and scriptindex with --stemmer=LANGUAGE (use --stemmer=none to disable stemming, see omindex --help for the list of accepted language names). At search time you can configure the stemmer by adding $set{stemmer,LANGUAGE} to the top of your OmegaScript template.

The two term types are used as follows when building the query: B(oolean) terms with the same prefix are ORed together, with all the different prefix groups being ANDed together. This is then FILTERed against the P(robabilistic) terms. This will look something like:

[ FILTER ] /      \ /        \ P-terms      [     AND     ] /     | ... \ / [    OR    ] /      | ... \ B(F,1) B(F,2)...B(F,n)

Where B(F,1) is the first boolean term with prefix F, and so on.

The intent here is to allow filtering on arbitrary (and, typically, orthogonal) characteristics of the document. For instance, by adding boolean terms "Ttext/html", "Ttext/plain" and "P/press" you would be filtering the probabilistic search for only documents that are both in the "/press" site and which are either of MIME type text/html or text/plain. (See below for more information about sites.)

If there is no probabilistic query, the boolean filter is promoted to be the query, and the weighting scheme is set to boolean. This has the effect of applying the boolean filter to the whole database.

In order to add more boolean prefixes, you will need to alter the index_file() function in omindex.cc. Currently omindex adds several useful ones, detailed below.

Probabilistic terms are constructed from the title, body and keywords of a document. (Not all document types support all three areas of text.) Title terms are stored with position data starting at 0, body terms starting 100 beyond title terms, and keyword terms starting 100 beyond body terms. This allows queries using positional data without causing false matches across the different types of term.

Sites

Within a database, Omega supports multiple sites. These are recorded using boolean terms (see 'Term construction', above) to allow filtering on them.

Sites work by having all documents within them having a common base URL. For instance, you might have two sites, one for your press area and one for your product descriptions:

You could index all documents within http://example.com/press/ using a site of '/press', and all within http://example.com/products/ using '/products'.

Sites are also useful because omindex indexes documents through the file system, not by fetching from the web server. If you don't have a URL to file system mapping which puts all documents under one hierarchy, you'll need to index each separate section as a site.

An obvious example of this is the way that many web servers map URLs of the form <http://example.com/~<username>/> to a directory within that user's home directory (such as ~<username>/pub on a Unix system). In this case, you can index each user's home page separately, as a site of the form '/~<username>'. You can then use boolean filters to allow people to search only a specific home page (or a group of them), or omit such terms to search everyone's pages.

Note that the site specified when you index is used to build the complete URL that the results page links to. Thus while sites will typically want to be relative to the hostname part of the URL (e.g. '/site' rather than 'http://example.com/site'), you can use them to have a single search across several different hostnames. This will still work if you actually store each distinct hostname in a different database.

omindex operation

omindex is fairly simple to use, for example:

omindex --db default --url http://example.com/ /var/www/example.com

For a full list of command line options supported, see man omindex or omindex --help.

You must specify the database to index into (it's created if it doesn't exist, but parent directories must exist). You will often also want to specify the base URL (which is used as the site, and can be relative to the hostname - starts '/' - or absolute - starts with a scheme, e.g. 'http://example.com/products/'). If not specified, the base URL defaults to /.

You also need to tell omindex which directory to index. This should be either a single directory (in which case it is taken to be the directory base of the entire site being indexed), or as two arguments, the first being the directory base of the site being indexed, and the second being a relative directory within that to index.

For instance, in the example above, if you separate your products by size, you might end up with:

If the entire website is stored in the file system under the directory /www/example, then you would probably index the site in two passes, one for the '/press' site and one for the '/products' site. You might use the following commands:

$ omindex -p --db /var/lib/omega/data/default --url /press /www/example/press $ omindex -p --db /var/lib/omega/data/default --url /products /www/example/products

If you add a new large products, but don't want to reindex the whole of the products section, you could do:

$ omindex -p --db /var/lib/omega/data/default --url /products /www/example/products large

and just the large products will be reindexed. You need to do it like that, and not as:

$ omindex -p --db /var/lib/omega/data/default --url /products/large /www/example/products/large

because that would make the large products part of a new site, '/products/large', which is unlikely to be what you want, as large products would no longer come up in a search of the products site. (Note that the --depth-limit option may come in handy if you have sites '/products' and '/products/large', or similar.)

omindex has built-in support for indexing HTML, PHP, text files, CSV (Comma-Separated Values) files, and AbiWord documents. It can also index a number of other formats using external programs. Filter programs are run with CPU and memory limits to prevent a runaway filter from blocking indexing of other files.

The following formats are supported as standard (you can tell omindex to use other filters too - see below):

  • HTML (.html, .htm, .shtml)
  • PHP (.php) - our HTML parser knows to ignore PHP code
  • text files (.txt, .text)
  • SVG (.svg)
  • CSV (Comma-Separated Values) files (.csv)
  • PDF (.pdf) if pdftotext is available (comes with xpdf)
  • PostScript (.ps, .eps, .ai) if ps2pdf (from ghostscript) and pdftotext (comes with xpdf) are available
  • OpenOffice/StarOffice documents (.sxc, .stc, .sxd, .std, .sxi, .sti, .sxm, .sxw, .sxg, .stw) if unzip is available
  • OpenDocument format documents (.odt, .ods, .odp, .odg, .odc, .odf, .odb, .odi, .odm, .ott, .ots, .otp, .otg, .otc, .otf, .oti, .oth) if unzip is available
  • MS Word documents (.doc, .dot) if antiword is available
  • MS Excel documents (.xls, .xlb, .xlt, .xlr) if xls2csv is available (comes with catdoc)
  • MS Powerpoint documents (.ppt, .pps) if catppt is available (comes with catdoc)
  • MS Office 2007 documents (.docx, .docm, .dotx, .dotm, .xlsx, .xlsm, .xltx, .xltm, .pptx, .pptm, .potx, .potm, .ppsx, .ppsm) if unzip is available
  • Wordperfect documents (.wpd) if wpd2text is available (comes with libwpd)
  • MS Works documents (.wps, .wpt) if wps2text is available (comes with libwps)
  • MS Outlook message (.msg) if perl with Email::Outlook::Message and HTML::Parser modules is available
  • AbiWord documents (.abw)
  • Compressed AbiWord documents (.zabw) if gzip is available
  • Rich Text Format documents (.rtf) if unrtf is available
  • Perl POD documentation (.pl, .pm, .pod) if pod2text is available
  • TeX DVI files (.dvi) if catdvi is available
  • DjVu files (.djv, .djvu) if djvutxt is available
  • XPS files (.xps) if unzip is available
  • Debian packages (.deb, .udeb) if dpkg-deb is available
  • RPM packages (.rpm) if rpm is available

If you have additional extensions that represent one of these types, you can add an additional MIME mapping using the --mime-type option. For instance:

$ omindex --db /var/lib/omega/data/default --url /press /www/example/press --mime-type doc:application/postscript

The syntax of --mime-type is 'ext:type', where ext is the extension of a file of that type (everything after the last '.'), and type is one of:

  • text/csv
  • text/html
  • text/plain
  • text/rtf
  • text/x-perl
  • application/msword
  • application/pdf
  • application/postscript
  • application/vnd.ms-excel
  • application/vnd.ms-outlook
  • application/vnd.ms-powerpoint
  • application/vnd.ms-works
  • application/vnd.ms-xpsdocument
  • application/vnd.oasis.opendocument.text
  • application/vnd.oasis.opendocument.spreadsheet
  • application/vnd.oasis.opendocument.presentation
  • application/vnd.oasis.opendocument.graphics
  • application/vnd.oasis.opendocument.chart
  • application/vnd.oasis.opendocument.formula
  • application/vnd.oasis.opendocument.database
  • application/vnd.oasis.opendocument.image
  • application/vnd.oasis.opendocument.text-master
  • application/vnd.oasis.opendocument.text-template
  • application/vnd.oasis.opendocument.spreadsheet-template
  • application/vnd.oasis.opendocument.presentation-template
  • application/vnd.oasis.opendocument.graphics-template
  • application/vnd.oasis.opendocument.chart-template
  • application/vnd.oasis.opendocument.formula-template
  • application/vnd.oasis.opendocument.image-template
  • application/vnd.oasis.opendocument.text-web
  • application/vnd.openxmlformats-officedocument.wordprocessingml.document
  • application/vnd.openxmlformats-officedocument.wordprocessingml.template
  • application/vnd.openxmlformats-officedocument.spreadsheetml.sheet
  • application/vnd.openxmlformats-officedocument.spreadsheetml.template
  • application/vnd.openxmlformats-officedocument.presentationml.presentation
  • application/vnd.openxmlformats-officedocument.presentationml.slideshow
  • application/vnd.openxmlformats-officedocument.presentationml.template
  • application/vnd.sun.xml.calc
  • application/vnd.sun.xml.calc.template
  • application/vnd.sun.xml.draw
  • application/vnd.sun.xml.draw.template
  • application/vnd.sun.xml.impress
  • application/vnd.sun.xml.impress.template
  • application/vnd.sun.xml.math
  • application/vnd.sun.xml.writer
  • application/vnd.sun.xml.writer.global
  • application/vnd.sun.xml.writer.template
  • application/vnd.wordperfect
  • application/x-abiword
  • application/x-abiword-compressed
  • application/x-debian-package
  • application/x-dvi
  • application/x-redhat-package-manager
  • image/svg+xml
  • image/vnd.djvu
  • ignore (magic token to tell omindex to quietly ignore such files)

By default, files with the following extensions are marked as 'ignore':

- a - css - dll - dylib - exe - js - lib - o - obj - so

If you wish to remove a MIME mapping, you can do this by omitting the type - for example to not index .doc files, use: --mime-type=doc:

The lookup of extensions in the MIME mappings is case sensitive, but if an extension isn't found and includes upper case ASCII letters, they're converted to lower case and the lookup is repeated, so you effectively get case insensitive lookup for mappings specified with a lower-case extension, but you can set different handling for differently cased variants if you need to.

You can add support for additional MIME content types (or override existing ones) using the --filter option - for example, if you wanted to handle files of MIME type application/octet-stream by running them through strings -n8, you can pass the option --filter=application/octet-stream:'strings -n8'. The filename of the file to be extracted will be appended to this command, separated by a space. Commands run via --filter need to produce UTF-8 text output on stdout.

A more complex example of the use of --filter makes use of OpenOffice, via the unoconv script, to extract text from various formats. First you need to start a listening instance (if you don't, unoconv will start up OpenOffice for every file, which is rather inefficient) - the & just tell the shell to run it in the background:

unoconv --listener &

Then run omindex with options such as --filter=application/msword:'unoconv --stdout -f text' (you'll want one for each format which you want to extract text from with OpenOffice).

If you know of a reliable filter which can extract text from a file format which might be of interest to others, please let us know so we can consider including it as a standard filter.

The --duplicates option controls how omindex handles documents which map to a URL which is already in the database. The default (which can be explicitly set with --duplicates=replace) is to reindex if the last modified time of the file is newer than that recorded in the database. The alternative is --duplicates=ignore, which will never reindex an existing document. If you only add documents, this avoids the overhead of checking the last modified time. It also allows you to prioritise adding completely new documents to the database over updating existing ones.

By default, omindex will remove any document in the database which has a URL that doesn't correspond to a file seen on disk - in other words, it will clear out everything that doesn't exist any more. However if you are building up an omega database with several runs of omindex, this is not appropriate (as each run would delete the data from the previous run), so you should use the --no-delete option. Note that if you choose to work like this, it is impossible to prune old documents from the database using omindex. If this is a problem for you, an alternative is to index each subsite into a different database, and merge all the databases together when searching.

--depth-limit allows you to prevent omindex from descending more than a certain number of directories. Specifying --depth-limit=0 means no limit is imposed on recursion; --depth-limit=1 means don't descend into any subdirectories of the start directory.

HTML Parsing

The document <title> tag is used as the document title, the 'description' META tag (if present) is used for the document snippet, and the 'keywords' META tag (if present) is indexed as extra document text.

The HTML parser will look for the 'robots' META tag, and won't index pages which are marked as noindex or none, for example any of the following:

<meta name="robots" content="noindex,nofollow"> <meta name="robots" content="noindex"> <meta name="robots" content="none">

Sometimes it is useful to be able to exclude just part of a page from being indexed (for example you may not want to index navigation links, or a footer which appears on every page). To allow this, the parser also understands ht://dig-style comments to mark sections of the document to not index:

Index this bit <!--htdig_noindex-->but <b>not</b> this<!--/htdig_noindex-->

Boolean terms

omindex will create the following boolean terms when it indexes a document:

T
MIME type
H
hostname of site (if supplied - this term won't exist if you index a site with base URL '/press', for instance)
P
path of site (i.e. the rest of the site base URL)
U
full URL of indexed document - if the resulting term would be > 240 characters, a hashing scheme is used to prevent omindex overflowing the Xapian term length limit.
D
date (numeric format: YYYYMMDD) date can also have the magical form "latest" - a document indexed by the term Dlatest matches any date-range without an end date. You can index dynamic documents which are always up to date with Dlatest and they'll match as expected. (If you use sort by date, you'll probably also want to set the value containing the timestamp to a "max" value so dynamic documents match a date in the far future).
M
month (numeric format: YYYYMM)
Y
year (four digits)

omega configuration

Most of the omega CGI configuration is dynamic, by setting CGI parameters. However some things must be configured using a configuration file. The configuration file is searched for in various locations:

  • Firstly, if the "OMEGA_CONFIG_FILE" environment variable is set, its value is used as the full path to a configuration file to read.
  • Next (if the environment variable is not set, or the file pointed to is not present), the file "omega.conf" in the same directory as the Omega CGI is used.
  • Next (if neither of the previous steps found a file), the file "${sysconfdir}/omega.conf" (e.g. /etc/omega.conf on Linux systems) is used.
  • Finally, if no configuration file is found, default values are used.

The format of the file is very simple: a line per option, with the option name followed by its value, separated by a whitespace. Blank lines are ignored. If the first non-whitespace character on a line is a '#', omega treats the line as a comment and ignores it.

The current options are 'database_dir' (the directory containing all the Omega databases), 'template_dir' (the directory containing the OmegaScript templates), and 'log_dir' (the directory which the OmegaScript $log command writes log files to).

The default values (used if no configuration file is found) are:

database_dir /var/lib/omega/data template_dir /var/lib/omega/templates log_dir /var/log/omega

Note that, with apache, environment variables may be set using mod_env, and with apache 1.3.7 or later this may be used inside a .htaccess file. This makes it reasonably easy to share a single system installed copy of Omega between multiple users.

Supplied Templates

The OmegaScript templates supplied with Omega are:

  • query - This is the default template, providing a typical Web search interface.
  • topterms - This is just like query, but provides a "top terms" feature which suggests terms the user might want to add to their query to obtain better results.
  • godmode - Allows you to inspect a database showing which terms index each document, and which documents are indexed by each term.
  • opensearch - Provides results in OpenSearch format (for more details see http://www.opensearch.org/).
  • xml - Provides results in a custom XML format.
  • emptydocs - Shows a list of documents with zero length. If CGI parameter TERM is set to a non-empty value, then only documents indexed by that given term are shown (e.g. TERM=Tapplication/pdf to show PDF files with no text); otherwise all zero length documents are shown.

There are also "helper fragments" used by the templates above:

  • inc/anyalldropbox - Provides a choice of matching "any" or "all" terms by default as a drop down box.
  • inc/anyallradio - Provides a choice of matching "any" or "all" terms by default as radio buttons.
  • toptermsjs - Provides some JavaScript used by the topterms template.

Document data construction

This is only useful if you need to inject your own documents into the database independently of omindex, such as if you are indexing dynamically-generated documents that are served using a server-side system such as PHP or ASP, but which you can determine the contents of in some way, such as documents generated from reasonably static database contents.

The document data field stores some summary information about the document, in the following (sample) format:

url=<baseurl> sample=<sample> caption=<title> type=<mimetype>

Further fields may be added (although omindex doesn't currently add any others), and may be looked up from OmegaScript using the $field{} command.

As of Omega 0.9.3, you can alternatively add something like this near the start of your OmegaScript template:

$set{fieldnames,$split{caption sample url}}

Then you need only give the field values in the document data, which can save a lot of space in a large database. With the setting of fieldnames above, the first line of document data can be accessed with $field{caption}, the second with $field{sample}, and the third with $field{url}.

via xapian.org

Posted by at

Flickr - projectbrainsaver

www.flickr.com
projectbrainsaver's A Point of View photoset projectbrainsaver's A Point of View photoset