Sphinx根据排序返回不一致的结果集

Question

I'm trying to implement multilingual indexes for the web application I'm developing. At the moment, records exist in a few languages, English, Malay & Arabic (but they are not separated into different columns). Only English stemmer is currently enabled.

Only two indexes are built, for the stemmed and the non-stemmed indexes. I'm having the problem with the stemmed index, as the result set returned is not consistent, depending on the sort column.

These two queries (from the stemmed index), each returns a different number of total results, although the difference between them is only the sort order.

SELECT * FROM test1stemmed WHERE MATCH('@institution universiti') GROUP BY art_id ORDER BY art_title_ord ASC;

SELECT * FROM test1stemmed WHERE MATCH('@institution universiti') GROUP BY art_id ORDER BY art_title_ord DESC;

However, if the same queries were run on the non-stemmed index, the numbers of results are equal.

I'm also having the same problem with Sphinx PHP API:

$sp = new SphinxClient();
$sp->SetServer('localhost', 9312);
$sp->SetMatchMode(SPH_MATCH_EXTENDED);
$sp->SetGroupBy('art_id', SPH_GROUPBY_ATTR, "$sp_sort_column $sort");
$sp->SetLimits($offset, $rows_per_page, 1000);
$sp->Query("$q", 'test1stemmed');

What am I missing?

Answer 1

Something that I missed from the documentation here http://sphinxsearch.com/docs/2.0.2/clustering.html

WARNING: grouping is done in fixed memory and thus its results are only approximate; so there might be more groups reported in total_found than actually present. @count might also be underestimated. To reduce inaccuracy, one should raise max_matches. If max_matches allows to store all found groups, results will be 100% correct.

So I can workaround this by increasing the value in max_matches, but since putting a very large value is absolutely undesirable, I would fix the query instead.