=head1 NAME

Full text indexing in RT

=head1 LIMITATIONS

While all of the below solutions can search for Unicode characters, they
are not otherwise Unicode aware, and do no case folding, normalization,
or the like.  That is, a string that contains C<U+0065 LATIN SMALL
LETTER E> followed by C<U+0301 COMBINING ACUTE ACCENT> will not match a
search for C<U+00E9 LATIN SMALL LETTER E WITH ACUTE>.  They also only
know how to tokenize C<latin-1>-ish languages where words are separated
by whitespace or similar characters; as such, support for searching for
Japanese and Chinese content is extremely limited.

=head1 POSTGRES

=head2 Creating and configuring the index

Postgres 8.3 and above support full-text searching natively; to set up
the required C<ts_vector> column, and create either a C<GiN> or C<GiST>
index on it, run:

    rt-setup-fulltext-index-5

If you have a non-standard database administrator username or password,
you may need to pass the C<--dba> or C<--dba-password> options:

    rt-setup-fulltext-index-5 --dba postgres --dba-password secret

This will then tokenize and index all existing attachments in your
database; it may take quite a while if your database already has a large
number of tickets in it.

Finally, it will output an appropriate C<%FullTextSearch> configuration
to add to your F<RT_SiteConfig.pm>; you will need to restart your
webserver after making these changes.


=head2 Updating the index

To keep the index up-to-date, you will need to run:

    rt-fulltext-indexer-5

...at regular intervals.  By default, this will only tokenize up to 200
tickets at a time; you can adjust this upwards by passing C<--limit
500>.  Larger batch sizes will take longer and consume more memory.

If there is already an instance of C<rt-fulltext-indexer> running, new
ones will exit with the message "rt-fulltext-indexer is already running".
You can suppress this message and other output to C<STDERR> using the
C<--quiet> option; this is particularly useful when running the command via
C<cron>:

    rt-fulltext-indexer-5 --quiet

=head1 MYSQL and MariaDB

On MySQL, full-text search can either be done using native support
(which may use MyISAM tables on pre-5.6 versions of MySQL), or RT can
integrate with the external Sphinx full-text search engine.

RT supports native full-text search on MariaDB versions 10.0 and
greater. It does not support Sphinx integration on MariaDB.

=head2 Native MySQL and MariaDB

As RT marks attachment data as C<BINARY>, MySQL and MariaDB cannot
index this content without creating an additional table.  To create
the required table (which is InnoDB on versions of MySQL and MariaDB
which support it), run:

    rt-setup-fulltext-index-5

If you have a non-standard database administrator username or password,
you may need to pass the C<--dba> or C<--dba-password> options:

    rt-setup-fulltext-index-5 --dba root --dba-password secret

This will then tokenize and index all existing attachments in your
database; it may take quite a while if your database already has a large
number of tickets in it.

Finally, it will output an appropriate C<%FullTextSearch> configuration
to add to your F<RT_SiteConfig.pm>; you will need to restart your
webserver after making these changes.


=head3 Updating the index

To keep the index up-to-date, you will need to run:

    rt-fulltext-indexer-5

...at regular intervals.  By default, this will only tokenize up to 200
tickets at a time; you can adjust this upwards by passing C<--limit
500>.  Larger batch sizes will take longer and consume more memory.

If there is already an instance of C<rt-fulltext-indexer> running, new
ones will exit with the message "rt-fulltext-indexer is already running".
You can suppress this message and other output to C<STDERR> using the
C<--quiet> option; this is particularly useful when running the command via
C<cron>:

    rt-fulltext-indexer-5 --quiet

=head3 Caveats

Searching is done in "boolean mode."  As such, the TicketSQL query
C<Content LIKE 'winter 2014'> will return tickets with transactions that
contain I<either> word.  To find transactions which contain both (but
not necessarily adjacent), use C<Content LIKE '+winter +2014'>.  To find
transactions containing the precise phrase, use C<Content LIKE '"winter
2014">.

See the mysql documentation, at
L<http://dev.mysql.com/doc/refman/5.6/en/fulltext-boolean.html>, for a
list of the full capabilities.


=head2 MySQL with Sphinx

RT can also integrate with the external Sphinx engine, available from
L<http://sphinxsearch.com>.  Unfortunately, Sphinx integration (using
SphinxSE) does require that you recompile MySQL from source.  Most
distribution-provided packages for MySQL do not include SphinxSE
integration, merely the external Sphinx tools; these are not sufficient
for RT's needs.

RT does not support MariaDB with Sphinx.

=head3 Compiling MySQL and SphinxSE

MySQL 5.1 supports adding pluggable storage engines; after compiling
against the appropriate version of MySQL, the F<ha_sphinx.so> file is
the only that needs to be installed in production, generally into
C</usr/lib/mysql/plugin/>.  It can then be enabled via:

    INSTALL PLUGIN Sphinx SONAME "ha_sphinx.so"

Sphinx versions 0.9.x and 2.0.x are known-working versions, but later
versions may work as well.  Complete compilation and installation
instructions for MySQL with SphinxSE can be found at
L<http://sphinxsearch.com/docs/current.html#sphinxse-mysql51>.

=head3 Creating and configuring the index

Once MySQL has been recompiled with SphinxSE, and Sphinx itself is
installed, you may create the required SphinxSE communication table via:

    rt-setup-fulltext-index-5

If you have a non-standard database administrator username or password,
you may need to pass the C<--dba> or C<--dba-password> options:

    rt-setup-fulltext-index-5 --dba root --dba-password secret

This will also provide you with the appropriate C<%FullTextSearch>
configuration to add to your F<RT_SiteConfig.pm>; you will need to
restart your webserver after making these changes.  It will also print a
sample Sphinx configuration, which should be placed in
F</etc/sphinx.conf>, or equivalent.

To fill the index, you will need to run the C<indexer> command-line tool
provided by Sphinx:

    indexer rt

Finally, start the Sphinx search daemon:

    searchd

=head3 Updating the index

To keep the index up-to-date, you will need to run:

    indexer rt --rotate

...at regular intervals in order to pick up new and updated attachments
from RT's database.  Failure to do so will result in stale data.

=head3 Caveats

RT's integration with Sphinx relies on the use of a special index; there
exist queries where the MySQL optimizer elects to I<not> use that index,
instead electing to scan the table, which causes no results to be
returned.  However, this is rare, and generally only occurs on complex
queries.

Sphinx also only returns a finite number of matches to any query; this
number is controlled by C<max_matches> in F</etc/sphinx.conf> and
C<%FullTextSearch>'s C<MaxMatches> in C<RT_SiteConfig.pm>, which must be
kept in sync.  The default, set during C<rt-setup-fulltext-index>, is
10000.  This limit may lead to false negatives in search results if the
maximum number of matches is reached but the results returned do not
match RT's other criteria.  However, a too-large value will notably
degrade performance, as it adds memory allocation overhead to every
query.

Take, for example, the instance where Sphinx is configured to return a
maximum of three results, and tickets 1, 2, 3, 4, and 5 contain the
string "target", but only ticket 5 is in status "Open".  A search for
C<Content LIKE 'target' AND Status = 'Open'> may return no results,
despite ticket 5 matching those criteria, as Sphinx will only return
tickets 1, 2, and 3 as possible matches.

After index creation, altering C<MaxMatches> in C<RT_SiteConfig.pm> is
insufficient to adjust this limit; both C<max_matches> in
F</etc/sphinx.conf> and C<%FullTextSearch>'s C<MaxMatches> in
C<RT_SiteConfig.pm> must be updated.

=head1 ORACLE

=head2 Creating and configuring the index

Oracle supports full-text indexing natively using the Oracle Text
package.  Once Oracle Text is installed and configured, run:

    rt-setup-fulltext-index-5

If you have a non-standard database administrator username or password,
you may need to pass the C<--dba> or C<--dba-password> options:

    rt-setup-fulltext-index-5 --dba sysdba --dba-password secret

This will create an Oracle CONTEXT index on the Content column in the
Attachments table, as well as several preferences, functions and
triggers to support this index.  The script will also output an
appropriate C<%FullTextSearch> configuration to add to your
F<RT_SiteConfig>.

=head2 Updating the index

To update the index, you will need to run the following at regular
intervals:

    rt-fulltext-indexer-5

This, in effect, simply runs:

    begin
    ctx_ddl.sync_index('rt_fts_index', '2M');
    end;

The amount of memory used for the sync can be controlled with the
C<--memory> option:

    rt-fulltext-indexer-5 --memory 10M

If there is already an instance of C<rt-fulltext-indexer> running, new
ones will exit with the message "rt-fulltext-indexer is already running".
You can suppress this message and other output to C<STDERR> using the
C<--quiet> option; this is particularly useful when running the command via
C<cron>:

    rt-fulltext-indexer-5 --quiet

Instead of being run via C<cron>, this may instead be run via a
DBMS_JOB; read the B<Managing DML Operations for a CONTEXT Index>
chapter of Oracle's B<Text Application Developer's Guide> for details
how to keep the index optimized, perform garbage collection, and other
tasks.

=head1 UNINDEXED SEARCH

It is also possible to enable full-text search without database indexing
support, simply by setting the C<Enable> key of the C<%FullTextSearch>
option to 1, while leaving C<Indexed> set to 0:

    Set(%FullTextSearch,
        Enable  => 1,
        Indexed => 0,
    );

This is not generally suggested, as unindexed full-text searching can
cause severe performance problems.

=head1 LIMIT ATTACHMENT SIZE

On some systems, very large attachments can cause memory and other
performance issues for the indexer making it unable to complete
indexing. See L<RT_Config.pm/"$MaxFulltextAttachmentSize"> for
details on setting a maximum attachment size to index.

=cut