OMERO search
============
Beginning with 3.0-Beta3, the OMERO server will use
`Â Lucene <http://lucene.apache.org>`_ to index all string and timestamp
information in the database, as well as all ``OriginalFiles`` which can
be parsed to simple text (see :doc:`/developers/Search/FileParsers` for
more information). The index is stored under /OMERO/FullText (or the
``FullText`` subdirectory of your ${omero.data.dir}, and can be searched
with Google-like queries.
Field names
-----------
Each row in the database becomes a single Lucene ``Document`` parsed
into the several ``Fields``. A field is referenced by prefixing a search
term with the field name followed by a colon. For example,
`name:myImage` searches for myImage anywhere in the name field.
.. tabularcolumns:: |p{3.5cm}|p{12cm}|
.. csv-table::
:widths: 20 80
:header-rows: 1
:file: searchfieldnames.tsv
:delim: tab
Queries
-------
Search queries are very similar to Google searches. When search terms
are entered without a prefix ("name:"), then the default field will be
used which combines all available fields. Otherwise, a prefix can be
added to restrict the search.
Indexing
--------
Successful searching depends on understanding how the text is indexed.
The default analyzer used is :source:`the
FullTextAnalyzer <components/server/src/ome/services/fulltext/FullTextAnalyzer.java>`.
::
1. Desktop/image_GFP-H2B_1.dv ---> "desktop", "image", "gfp", "h2b", "1", "dv"
2. Desktop/image_GFP-H2B_2.dv ---> "desktop", "image", "gfp", "h2b", "2", "dv
3. Desktop/image_GFP_01-H2B.dv ---> "desktop", "image", "gfp", "01", "h2b", "dv"
4. Desktop/image_GFP-CSFV_a.dv ---> "desktop", "image", "gfp", "csfv", "a", "dv"
Assuming these entries above for Image.name:
- searching for **GFP-H2B** returns 1 and 2.
- searching for **"GFP H2B"** also returns 1 and 2.
- searching for **GFP H2B** returns 1, 2, and 3, since the two terms
are joined by an **OR**.
Information for developers and system administrators
----------------------------------------------------
Scheduling indexing
~~~~~~~~~~~~~~~~~~~
Indexing is not driven by the user, but happens automatically in the
background. Automatic indexing occurs at the frequency defined in
etc/omero.properties:
::
omero.search.cron=0,30 * * * * ?
omero.search.batch=100
which implies every thirty seconds of every hour, day, month, year, etc.
During each iteration, 100 ``EventLogs`` will be loaded from the
database and processed. Upon successful completion, the persistent count
in the ``configuration`` table, will be incremented.
::
omero3=# select value from configuration where name = 'PersistentEventLogLoader.current_id';
value
-------
30983
(1 row)
If you have more than one ``PersistentEventLogLoader.*`` value in your
database, then you have run indexing with multiple versions of the
server. This is fine. To allow a new server version to force an update,
the configuration key may be changed. For example,
::
PersistentEventLogLoader.currend_id
became
::
PersistentEventLogLoader.v2.current_id
in r2460.
Once an entity is indexed, it is possible to start writing querying
against the server via ``IQuery.findAllByFullText()``. Use
``new Parameters(new Filter().owner())`` and ``.group()`` to restrict
your search. Or alternatively use the ``oma.api.Search`` interface
(below).
.. _search-reindexing:
Re-indexing
^^^^^^^^^^^
If you need to re-index your database, stop your server, and:
- (Optionally) Delete the ``/OMERO/FullText`` directory
- Delete or set to 0 the entry from the ``configuration`` table:
``update configuration set value = 0 where name like 'PersistentEventLogLoader%';``
- If it is necessary to force re-indexing, use:
::
cd $OMERO_PREFIX
CLASSPATH=etc:`find lib/server | xargs | sed 's/ /:/g'`
java -Dlog4j.configuration=log4j-cli.properties -Xmx512M ome.services.fulltext.Main full
or alternatively for particular types, ...
::
java -Dlog4j.configuration=log4j-cli.properties -Xmx512M ome.services.fulltext.Main reindex ome.model.core.Image
This functionality is still being tested, but will be made more
available in future versions.
ome.api.IQuery
~~~~~~~~~~~~~~
The current IQuery implementation restricts searches to a single class
at a time.
- ``findAllByFullText(Image.class, "metaphase")`` -- Images which
contain or are annotated with "metaphase"
- ``findAllByFullText(Image.class, "annotation:metaphase")`` -- Images
which are annotated with "metaphase"
- ``findAllByFullText(Image.class, "tag:metaphase")`` -- Images which
are tagged with "metaphase" (specialization of the previous)
- ``findAllByFullText(Image.class, "file.contents:metaphase")`` --
Images which have files attached containing "metaphase"
- ``findAllByFullText(OriginalFile.class, "file.contents:metaphase")``
-- File containing "metaphase"
ome.api.Search
~~~~~~~~~~~~~~
The Search API offers a number of different queries along with various
filters and settings which are all maintained on the server.
The matrix below show which combinations of parameters and queries are
supported (S), will throw an exception (X), and which will simply silently be
ignored (I).
+--------------------------+---------------------------+---------------------------------+-------------------+
| **Query Method** --> | byFullText/SomeMustNone | byGroupForTags/byTagsForGroup | byAnnotatedWith |
+--------------------------+---------------------------+---------------------------------+-------------------+
| **Parameters** | | | |
+--------------------------+---------------------------+---------------------------------+-------------------+
| annotated between | S | S | S |
+--------------------------+---------------------------+---------------------------------+-------------------+
| annotated by | S | S | S |
+--------------------------+---------------------------+---------------------------------+-------------------+
| annotated with | S | I | I |
+--------------------------+---------------------------+---------------------------------+-------------------+
| created between | S | S | S |
+--------------------------+---------------------------+---------------------------------+-------------------+
| modified between | S | I (Immutable) | S |
+--------------------------+---------------------------+---------------------------------+-------------------+
| owned by | S | S | S |
+--------------------------+---------------------------+---------------------------------+-------------------+
| all types | X | I | X |
+--------------------------+---------------------------+---------------------------------+-------------------+
| 1 type | S | I | S |
+--------------------------+---------------------------+---------------------------------+-------------------+
| N types | X | I | X |
+--------------------------+---------------------------+---------------------------------+-------------------+
| only ids | S | I | S |
+--------------------------+---------------------------+---------------------------------+-------------------+
| **Ordering / Fetches** | | | |
+--------------------------+---------------------------+---------------------------------+-------------------+
| orderBy | S | I | S |
+--------------------------+---------------------------+---------------------------------+-------------------+
| fetchAnnotations | [1]_ | I | [2]_ |
+--------------------------+---------------------------+---------------------------------+-------------------+
| **Other** | | | |
+--------------------------+---------------------------+---------------------------------+-------------------+
| setProjections [3]_ | X | X | X |
+--------------------------+---------------------------+---------------------------------+-------------------+
| current\*Metdata [4]_ | X | X | X |
+--------------------------+---------------------------+---------------------------------+-------------------+
| setProjections [3]_ | X | X | X |
+--------------------------+---------------------------+---------------------------------+-------------------+
.. rubric:: Footnotes
.. [1] Any fetchAnnotation() argument to byFullText() or related queries,
returns **all** annotations.
.. [2] byAnnotatedWith() does not accept a fetchAnnotation() argument of
``Annotation.class``.
.. [3] setProjects may need to be removed if Lucene cannot handle OMERO's
security requirements.
.. [4] Not yet implemented.
Leading wildcard searches
^^^^^^^^^^^^^^^^^^^^^^^^^
Leading wildcard searches are disallowed by default. "?omething" or
"\*hatever", for example, would both throw exceptions. They can be run by
using:
::
Search search = serviceFactory.createSearchService();
search.setAllowLeadingWildcards(true);
There is a performance penalty, however. In addition,
wildcard searches get expanded on the server to boolean queries. For
example, assuming "ACELL", "BCELL", and "CCELL" are all terms in your
index, then the query:
::
*CELL
gets expanded to:
::
ACELL OR BCELL OR CCELL
If there are more than "omero.search.maxclause" terms in the expansion
(default is 4096), then an exception will be thrown. This requires the
user to enter a more refined search, but not because there are too many
results, only because there is not enough room in memory to search on
all terms at once.
Extension points
~~~~~~~~~~~~~~~~
Two extension points are currently available for searching. The first
are the :doc:`/developers/Search/FileParsers` mentioned above. By
configuring the map of Formats (roughly mime-types) of files to parser
instances, extracting information from attached binary files can be made
quick and straightforward.
Similarly, :doc:`/developers/Modules/Search/Bridges` provide a mechanism
for parsing all metadata entering the system. One built in bridge (the
:source:`FullTextBridge <components/server/src/ome/services/fulltext/FullTextBridge.java>`)
parses out the fields mentioned above, but by creating your own bridge
it is possible to extract more information specific to your site.
.. seealso::
:doc:`/developers/Modules/StructuredAnnotations`,
:doc:`/developers/Modules/Search/Bridges`,
:doc:`/developers/Search/FileParsers`,
`Query Parser Syntax <http://lucene.apache.org/core/3_6_0/queryparsersyntax.html>`_,
`Â Luke <http://www.getopt.org/luke/>`_
a Java application which you can download and point at your ``/OMERO/FullText`` directory to get a better feeling for Lucene queries.