Paul Taylor and Aurelién Mino have completed the re-write of our search server — please help us test this new code-base. Here is Paul’s write up:
This update rewrites the existing Search Indexer and Searcher with a
pure Java release whilst maintaining full compatibility with the
existing Musicbrainz Server, prototype by luks, completion by ijabz and
murdos. The aims were
1. Better performance
Search performance should be greatly improved for indexing and
especially search.
The intermediate python layer has been removed.
The webserver layer has been removed
When searching have a single Java VM , that can now be properly
configured/optimized
The latest Lucene and Java libraries are used
2. Better Search Results
The following bugs have already been fixed
Perhaps most importantly stop words are no longer and issue so when do a
search for something that exists in the database you should always get a
match, i.e artist:”the the” or track:”is this it?”
and range queries work properly
i.e
and theres a host of other bugs and enhancements that can now be
resolved quite easily
3. Maintainability
The code has been simplified, i.e no hand coding of xml or html or
having to deal with escape characters.
Now has a full set of unit tests, can also easily generate metrics such
as code coverage reports.
No dependency on Linux or Python, making it easier for new developers
to contribute, removing the burden from ruaok.
Testing Stage
Please try it out on http://musicbrainz.homeip.net both as a websearch
and as an XML Service to find any omissions in functionality and any
situations where the search results aren’t good as before.
Paul / ijabz
MetaBrainz Blog 